Documente Academic
Documente Profesional
Documente Cultură
Version 6.4
ABAQUS
THEORY MANUAL
Version 6.4
The information in this document is subject to change without notice and should not be construed as a commitment by ABAQUS, Inc. ABAQUS, Inc., assumes no responsibility for any errors that may appear in this document. The software described in this document is furnished under license and may be used or copied only in accordance with the terms of such license. No part of this document may be reproduced in any form or distributed in any way without prior written agreement with ABAQUS, Inc. ABAQUS, Inc., 2003. Printed in U.S.A. All Rights Reserved. ABAQUS is a registered trademark of ABAQUS, Inc. The following are trademarks of ABAQUS, Inc.: ABAQUS/Aqua; ABAQUS/CAE; ABAQUS/Design; ABAQUS/Explicit; ABAQUS/Foundation; ABAQUS/Standard; ABAQUS/Viewer; ABAQUS Interface for MOLDFLOW; ABAQUS Interface for MSC.ADAMS; and the ABAQUS, Inc., logo. This release of ABAQUS may contain capabilities licensed under U.S. Patents 5,920,491 and 6,044,210. ABAQUS, Inc., may also have other patents or pending patent applications, trademarks, copyrights, or other intellectual property rights covering subject matter in this document. The furnishing of this document does not give you any license to the patents, trademarks, copyrights, or other intellectual property rights except as expressly provided in any written license agreement from ABAQUS, Inc. ADAMS/Flex, ADAMS/View, MSC.ADAMS, and MSC.Patran are trademarks or registered trademarks of MSC.Software Corporation or its subsidiaries in the United States and/or other countries. Autodesk Inventor is a trademark and Autodesk Mechanical Desktop is a registered trademark of Autodesk Inc. CADKEY is a registered trademark of CADKEY Corporation. CATIA is a registered trademark of Dassault Systmes. Compaq Alpha is registered in the U.S. Patent and Trademark Ofce. DIGITAL Visual FORTRAN is a trademark of Compaq. Elysium is a pending trademark of Elysium Co., Ltd. and Elysium Inc. FEMAP, I-DEAS, Parasolid, Solid Edge, and Unigraphics are registered trademarks of Electronic Data Systems Corporation or its subsidiaries in the United States and in other countries. FE-SAFE is a trademark of Safe Technology, Ltd. FLEXlm is a registered trademark of GLOBEtrotter Software, Inc. Hewlett-Packard, HP-GL, HP-GL/2, and HP-UX are registered trademarks of Hewlett-Packard Co. IBM RS6000 is a trademark of IBM. Intel is a registered trademark of the Intel Corporation. MOLDFLOW, MOLDFLOW PLASTICS INSIGHT, and MPI are trademarks or registered trademarks of Moldow Corporation and its worldwide subsidiaries. NASTRAN is a registered trademark of the National Aeronautics and Space Administration. PostScript is a registered trademark of Adobe Systems, Inc. Pro/ENGINEER is a registered trademark of Parametric Technology Corporation or its subsidiaries in the U.S. and in other countries. Silicon Graphics and OpenGL are registered trademarks of Silicon Graphics, Inc. SolidDesigner is a trademark of CoCreate Software Inc. SolidWorks is a registered trademark of SolidWorks Corporation. SUN is a registered trademark of Sun Microsystems, Inc. UNIX and Motif are registered trademarks and X Window System is a trademark of The Open Group in the U.S. and other countries. Windows and Microsoft Visual C++ are registered trademarks of the Microsoft Corporation. ABAQUS/CAE incorporates portions of the ACIS software by SPATIAL TECHNOLOGY INC. ACIS is a registered trademark of SPATIAL TECHNOLOGY INC. This release of ABAQUS includes the gzip program obtained from the Free Software Foundation. This release of ABAQUS on Windows includes the diff program obtained from the Free Software Foundation. You may freely distribute the gzip and diff programs and/or modify them under the terms of the GNU Library General Public License as published by the Free Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA. This release of ABAQUS/CAE includes lp_solve, a simplex-based code for linear and integer programming problems by Michel Berkelaar of Eindhoven University of Technology, Eindhoven, the Netherlands. Python, copyright 19911995 by Stichting Mathematisch Centrum, Amsterdam, The Netherlands. All Rights Reserved. Permission to use, copy, modify, and distribute the Python software and its documentation for any purpose and without fee is hereby granted, provided that the above copyright notice appear in all copies and that both that copyright notice and this permission notice appear in supporting documentation, and that the names of Stichting Mathematisch Centrum or CWI or Corporation for National Research Initiatives or CNRI not be used in advertising or publicity pertaining to distribution of the software without specic, written prior permission. This software is provided with Restricted Rights for procurements governed by DFARS Part 227.4. Use, duplication, or disclosure by the U.S. Government or any of its agencies is subject to restrictions as set forth in subparagraphs (c)(1)(ii) of the Rights in Technical Data and Computer Software clause, DFARS 252.2277013 (October 1988). All other brand or product names are trademarks or registered trademarks of their respective companies or organizations.
ABAQUS, Inc.
ABAQUS, Inc. 1080 Main Street Pawtucket, RI 02860-4847 Tel: +1 401 727 4200 Fax: +1 401 727 4208 E-mail: support@Abaqus.com http://www.abaqus.com ABAQUS Europe BV Gaetano Martinolaan 95 P. O. Box 1637 6201 BP Maastricht The Netherlands Tel: +31 43 356 6906 Fax: +31 43 356 6908 E-mail: info.europe@abaqus.com Sales, Support, and Services UNITED STATES ABAQUS Central, Inc. 1440 Innovation Place West Lafayette, IN 47906-1000 Tel: +1 765 497 1373 Fax: +1 765 497 4444 E-mail: support@AbaqusCentral.com ABAQUS Erie, Inc. 3601 Green Road, Suite 316 Beachwood, OH 44122 Tel: +1 216 378 1070 Fax: +1 216 378 1072 E-mail: support@AbaqusErie.com ABAQUS South, Inc. 3700 Forums Drive, Suite 101 Flower Mound, TX 75028 Tel: +1 214 513 1600 Fax: +1 214 513 1700 E-mail: support@AbaqusSouth.com ARGENTINA KB Engineering S. R. L. Florida 274 - Ocina 35 1005 Buenos Aires Argentina Tel: +54 11 4326 9176/7542 Fax: +54 11 4326 2424 E-mail: sanchezsarmiento@arnet.com.ar ABAQUS East, LLC 300 Centerville Road, Suite 209W Warwick, RI 02886-0201 Tel: +1 401 739 3637 Fax: +1 401 739 3302 E-mail: support@AbaqusEast.com ABAQUS Great Lakes, Inc. 14500 Sheldon Road, Suite 160 Plymouth, MI 48170-2408 Tel: +1 734 451 0217 Fax: +1 734 451 0458 E-mail: support@AbaqusGreatLakes.com ABAQUS West, Inc. 39221 Paseo Padre Parkway, Suite F Fremont, CA 94538-1611 Tel: +1 510 794 5891 Fax: +1 510 794 1194 E-mail: support@AbaqusWest.com AUSTRALIA Worley Advanced Analysis Level 17, 300 Flinders Street Melbourne, Vic 3000 Tel: +61 3 9280 2834 Fax: +61 3 9205 0573 E-mail: abaqus@worley.com.au
AUSTRIA ABAQUS Austria GmbH Zinckgasse 20-22/2/13 A-1150 Vienna Austria Tel: +43 1 929 16 25-0 Fax: +43 1 929 16 25-20 E-mail: support@abaqus.at CHINA ABAQUS China Beijing Representative Ofce Room 716, Tower B, COFCO Plaza No. 8, Jiangguomennei Dajie Dong Cheng District Beijing, 100005 P. R. China Tel: +86 01 85110566/85110567 Fax: +86 01 85110568 E-mail: abaqus@abaqus.com.cn FRANCE ABAQUS France SAS 7 rue Jean Mermoz 78000 Versailles Tel: +33 01 39 24 15 40 Fax: +33 01 39 24 15 45 E-mail: support@abaqus.fr INDIA (Chennai) ABAQUS Engineering Analysis Solutions (Pvt.) Ltd. 3M, Prince Arcade 22-A Cathedral Road Chennai, 600 086 Tel: +91 44 28114624 Fax: +91 44 28115087 E-mail: abaqus@abaqus.co.in
BENELUX ABAQUS Benelux BV Huizermaatweg 576 1276 LN Huizen The Netherlands Tel: +31 35 52 58 424 Fax: +31 35 52 44 257 E-mail: support@abaqus.nl CZECH REPUBLIC Synerma s. r. o. Huntirov 58 468 22 Skuhrov Czech Republic Tel: +420 2 603 145 769 Fax: +420 2 603 181 944 E-mail: abaqus@synerma.cz
GERMANY (Aachen) ABAQUS Deutschland GmbH Theaterstrae 30-32 D-52062 Aachen Tel: +49 241 474010 Fax: +49 241 4090963 E-mail: info@abaqus.de ITALY ABAQUS Italia s. r. l. Via Domodossola, 17 20145 Milano (MI) Tel: +39 02 39211211 Fax: +39 02 39211210 E-mail: info@abaqus.it
JAPAN (Tokyo) ABAQUS, Inc. 3rd Floor, Akasaka Nihon Building 5-24, Akasaka 9-chome, Minato-ku Tokyo, 107-0052 Tel: +81 3 5474 5817 Fax: +81 3 5474 5818 E-mail: tokyo@abaqus.jp KOREA ABAQUS Korea, Inc. Suite 306, Sambo Building 13-2 Yoido-Dong, Youngdeungpo-ku Seoul, 150-010 Tel: +82 2 785 6707 Fax: +82 2 785 6709 E-mail: info@abaqus.co.kr NEW ZEALAND Matrix Applied Computing Ltd. P. O. Box 56-316, Auckland Courier: Unit 2-5, 72 Dominion Road, Mt Eden, Auckland Tel: +64 9 623 1223 Fax: +64 9 623 1134 E-mail: hks-support@matrix.co.nz RUSSIA, BELARUS & UKRAINE TESIS Ltd. Ofce 701-703, 18, Unnatov Str. 127083 Moscow, Russia Tel: +7 095 212-44-22 Fax: +7 095 212-42-62 E-mail: info@tesis.com.ru
JAPAN (Osaka) ABAQUS, Inc. 9th Floor, Higobashi Watanabe Building 6-10, Edobori 1-chome, Nishi-ku Osaka, 550-0002 Tel: +81 6 4803 5020 Fax: +81 6 4803 5021 E-mail: osaka@abaqus.jp MALAYSIA Worley Advanced Analysis 13th Floor, Empire Tower City Square Centre 182 Jalan Tun Razak 50400 Kuala Lumpur Tel: +60 3 2163 4275 Fax: +60 3 2163 0524 E-mail: abaqus@ranhill-worley.com.my POLAND BudSoft Sp. z o.o. 61-807 Poznan Sw. Marcin 58/64 Tel: +48 61 8508 466 Fax: +48 61 8508 467 E-mail: budsoft@budsoft.com.pl SINGAPORE Worley Advanced Analysis 491B River Valley Road #09-01 Valley Point Singapore, 248373 Tel: +65 6735 8444 Fax: +65 6735 7444 E-mail: abaqus@worley.com.sg
SOUTH AFRICA Finite Element Analysis Services (Pty) Ltd. Unit 4, The Waverley Wyecroft Road Mowbray 7700 Tel: +27 21 448 7608 Fax: +27 21 448 7679 E-mail: feas@feas.co.za SWEDEN ABAQUS Scandinavia AB Pilgatan 8c SE-721 30 Vsters Tel: +46 21 12 64 10 Fax: +46 21 18 12 44 E-mail: femtech@femtech.se TURKEY A-Ztech Ltd. PERDEMSAC Business Center, Technology House 17 Gulbahar Str., Bayar Road Kozyatagi 34742 Istanbul Tel: +90 216 361 8850 Fax: +90 216 361 8851 E-mail: zeki.erman@a-ztech.com.tr
SPAIN Principia Ingenieros Consultores, S.A. Velzquez, 94 E-28006 Madrid Tel: +34 91 209 1482 Fax: +34 91 575 1026 E-mail: abaqus@principia.es TAIWAN APIC 7F-2, No. 131 SungChiang Road Taipei, 10428 Tel: +886 02 25083066 Fax: +886 02 25077185 E-mail: apic@apic.com.tw UNITED KINGDOM (Cheshire) ABAQUS UK Ltd. The Genesis Centre Science Park South, Birchwood Warrington, Cheshire WA3 7BH Tel: +44 1 925 810166 Fax: +44 1 925 810178 E-mail: hotline@abaqus.co.uk
Sales Only UNITED STATES ABAQUS East, LLC, Mid-Atlantic Ofce 114 Zachary Court Forest Hill, MD 21050 Tel: +1 410 420 8587 Fax: +1 410 420 8908 E-mail: support@AbaqusEast.com ABAQUS West, Inc., Southern CA and AZ Ofce 1100 Irvine Boulevard #248 Tustin, CA 92780 Tel: +1 714 731 5895 Fax: +1 714 731 5895 E-mail: support@AbaqusWest.com FINLAND ABAQUS Finland Oy Tekniikantie 12 FIN-02150 Espoo Tel: +358 9 2517 2973 Fax: +358 9 2517 2200 E-mail: nland@femtech.se INDIA (Pune) ABAQUS Engineering Analysis Solutions (Pvt.) Ltd. C-9, 3rd Floor Bramha Estate, Kondwa Road Pune-411040 Tel: +91 20 31037511 E-mail: abaqus@abaqus.co.in ABAQUS South, Inc., Southeast Ofce 484 Broadstone Way Acworth, GA 30101 Tel: +1 770 795 0960 Fax: +1 770 795 7614 E-mail: support@AbaqusSouth.com ABAQUS West, Inc., Rocky Mountains Ofce 2894 Hughs Drive Erie, CO 80516 Tel: +1 303 664 5444 Fax: +1 303 664 5445 E-mail: support@AbaqusWest.com GERMANY (Munich) ABAQUS Deutschland GmbH Sendlinger-Tor-Platz 8 D-80336 Mnchen Tel: +49 89 5999 1768 Fax: +49 89 5999 1767 E-mail: info@abaqus.de UNITED KINGDOM (Kent) ABAQUS UK Ltd. Great Hollanden Business Centre, Unit A Mill Lane, Underriver Nr. Sevenoaks, Kent TN15 OSQ Tel: +44 1 732 834930 Fax: +44 1 732 834720 E-mail: hotline@abaqus.co.uk
Preface
This section lists various resources that are available for help with using ABAQUS, including technical engineering and systems support, training seminars, and documentation.
Support
ABAQUS, Inc., offers both technical engineering support and systems support for ABAQUS. Technical engineering and systems support are provided through the nearest local support ofce. You can contact our ofces by telephone, fax, electronic mail, or regular mail. Information on how to contact each ofce is listed in the front of each ABAQUS manual. Support is also available on the World Wide Web for your convenience. The ABAQUS Online Support System (AOSS) is accessible through the MY ABAQUS section of the ABAQUS Home Page (www.abaqus.com). When contacting your local support ofce, please specify whether you would like technical engineering support (you have encountered problems performing an ABAQUS analysis or creating a model in ABAQUS) or systems support (ABAQUS will not install correctly, licensing does not work correctly, or other hardware-related issues have arisen). The ABAQUS Online Support System has a knowledge database of ABAQUS Answers. The ABAQUS Answers are solutions to questions that we have had to answer or guidelines on how to use ABAQUS. We welcome any suggestions for improvements to the support program or documentation. We will ensure that any enhancement requests you make are considered for future releases. If you wish to le a complaint about the service or products provided by ABAQUS, refer to the ABAQUS Home Page.
Technical engineering support
ABAQUS technical support engineers can assist in clarifying ABAQUS features and checking errors by giving both general information on using ABAQUS and information on its application to specic analyses. If you have concerns about an analysis, we suggest that you contact us at an early stage, since it is usually easier to solve problems at the beginning of a project rather than trying to correct an analysis at the end. Please have the following information ready before calling the technical engineering support hotline, and include it in any written contacts: Your site identier, which can be obtained by typing abaqus whereami at your system prompt (or by selecting Help On Version from the main menu bar in ABAQUS/CAE or ABAQUS/Viewer). The version of ABAQUS that are you using.
The version numbers for ABAQUS/Standard and ABAQUS/Explicit are given at the top of the data (.dat) le. The version numbers for ABAQUS/CAE and ABAQUS/Viewer can be found by selecting Help On Version from the main menu bar. The version numbers for the ABAQUS Interface for MOLDFLOW and the ABAQUS Interface for MSC.ADAMS are output to the screen.
The type of computer on which you are running ABAQUS. The symptoms of any problems, including the exact error messages, if any. Workarounds or tests that you have already tried.
When calling for support about a specic problem, any available ABAQUS output les may be helpful in answering questions that the support engineer may ask you. The support engineer will try to diagnose your problem from the model description and a description of the difculties you are having. Frequently, the support engineer will need model sketches, which can be faxed or sent in the mail. Plots of the nal results or the results near the point that the analysis terminated may also be needed to understand what may have caused the problem. If the support engineer cannot diagnose your problem from this information, you may be asked to supply the input data. The data can be attached to a support incident in the ABAQUS Online Support System. It may also be sent by means of e-mail, tape, disk, or ftp. Please check the ABAQUS Home Page (http://www.abaqus.com) for the media formats that are currently accepted. All support incidents are tracked in the ABAQUS Online Support System. This enables you (as well as the support engineer) to monitor the progress of a particular problem and to check that we are resolving support issues efciently. To use the ABAQUS Online Support System, you need to register with the system. Visit the MY ABAQUS section of the ABAQUS Home Page for instructions on how to register. If you are contacting us by means outside the AOSS to discuss an existing support problem and you know the incident number, please mention it so that we can consult the database to see what the latest action has been and, thus, avoid duplication of effort. In addition, please give the receptionist the support engineers name or include it at the top of any e-mail correspondence.
Systems support
ABAQUS systems support engineers can help you resolve issues related to the installation and running of ABAQUS, including licensing difculties, that are not covered by technical engineering support. You should install ABAQUS by carefully following the instructions in the ABAQUS Installation and Licensing Guide. If you are able to complete the installation, please make sure that the product verication procedure was run successfully at the end of the installation procedure. Successful verication for licensed products would indicate that you can run these products on your computer; unsuccessful verication for licensed products indicates problems with the installation or licensing (or both). If you encounter problems with the installation, licensing, or verication, rst review the instructions in the ABAQUS Installation and Licensing Guide to ensure that they have been followed correctly. If this does not resolve the problems, consult the ABAQUS Answers database in the ABAQUS Online Support System for information about known installation problems. If this does not address your situation, please create an incident in the AOSS and describe your problem, including the output from abaqus info=support. If you call, mail, e-mail, or fax us about a problem (instead of using the AOSS), please provide the output from abaqus info=support. It is important that you provide as much information as possible about your problem: error messages from an aborted analysis, output from the abaqus info=support command, etc.
ABAQUS Web server
For users connected to the Internet, many questions can be answered by visiting the ABAQUS Home Page on the World Wide Web at http://www.abaqus.com The information available on the ABAQUS Home Page includes:
ii
Frequently asked questions ABAQUS systems information and computer requirements ABAQUS performance data Error status reports ABAQUS documentation price list Training seminar schedule Newsletters
Anonymous ftp site
For users connected to the Internet, ABAQUS maintains useful documents on an anonymous ftp account on the computer ftp.abaqus.com. Simply ftp to ftp.abaqus.com. Login as user anonymous, and type your e-mail address as your password. Directions will come up automatically upon login.
Writing to technical support
Address of ABAQUS Headquarters: ABAQUS, Inc. 1080 Main Street Pawtucket, RI 02860-4847, USA Attention: Technical Support Addresses for other ofces and representatives are listed in the front of each manual.
Support for academic institutions
Under the terms of the Academic License Agreement we do not provide support to users at academic institutions. Academic users can purchase technical support on an hourly basis. For more information, please see the ABAQUS Home Page or contact your local ABAQUS support ofce.
Training
All ABAQUS ofces offer regularly scheduled public training classes. The Introduction to ABAQUS/Standard and ABAQUS/Explicit seminar covers basic usage and nonlinear applications, such as large deformation, plasticity, contact, and dynamics. Workshops provide as much practical experience with ABAQUS as possible. The Introduction to ABAQUS/CAE seminar discusses modeling, managing simulations, and viewing results with ABAQUS/CAE. Hands-on workshops are complemented by lectures. Advanced seminars cover topics of interest to customers with experience using ABAQUS, such as engine analysis, metal forming, fracture mechanics, and heat transfer. We also provide training seminars at customer sites. On-site training seminars can be one or more days in duration, depending on customer requirements. The training topics can include a combination of
iii
material from our introductory and advanced seminars. Workshops allow customers to exercise ABAQUS on their own computers. For a schedule of seminars, see the ABAQUS Home Page or call ABAQUS, Inc., or your local ABAQUS representative.
Documentation
The following documentation and publications are available from ABAQUS, unless otherwise specied, in printed form and through the ABAQUS online documentation. For more information on accessing the online books, refer to the discussion of execution procedures in the ABAQUS Analysis Users Manual.
Modeling and Visualization ABAQUS/CAE Users Manual: This reference document for ABAQUS/CAE includes three
comprehensive tutorials as well as detailed descriptions of how to use ABAQUS/CAE for model generation, analysis, and results evaluation and visualization. ABAQUS/Viewer users should refer to the information on the Visualization module in this manual.
Analysis ABAQUS Analysis Users Manual: This volume contains a complete description of the elements,
ABAQUS Example Problems Manual: This volume contains more than 75 detailed examples
designed to illustrate the approaches and decisions needed to perform meaningful linear and nonlinear analysis. Typical cases are large motion of an elastic-plastic pipe hitting a rigid wall; inelastic buckling collapse of a thin-walled elbow; explosive loading of an elastic, viscoplastic thin ring; consolidation under a footing; buckling of a composite shell with a hole; and deep drawing of a metal sheet. It is generally useful to look for relevant examples in this manual and to review them when embarking on a new class of problem.
ABAQUS Benchmarks Manual: This online-only volume contains over 200 benchmark problems
and standard analyses used to evaluate the performance of ABAQUS; the tests are multiple element tests of simple geometries or simplied versions of real problems. The NAFEMS benchmark problems are included in this manual.
Training Getting Started with ABAQUS:
This document is a self-paced tutorial designed to help new users become familiar with using ABAQUS/CAE to create solid, shell, and framework models and ABAQUS/Standard or ABAQUS/Explicit to perform static, quasi-static, and dynamic stress analysis simulations. It contains a number of fully worked examples that provide practical guidelines for performing structural analyses with ABAQUS.
iv
designed to help new users become familiar with the ABAQUS/Standard input le syntax for static and dynamic stress analysis simulations. The ABAQUS/Standard keyword interface is used to model examples similar to those included in Getting Started with ABAQUS.
Getting Started with ABAQUS/Explicit: Keywords Version: This online-only document is designed to help new users become familiar with the ABAQUS/Explicit input le syntax for quasistatic and dynamic stress analysis simulations. The ABAQUS/Explicit keyword interface is used to model examples similar to those included in Getting Started with ABAQUS.
Lecture Notes: These notes are available on many topics to which ABAQUS is applied. They are
used in the technical seminars that ABAQUS, Inc., presents to help users improve their understanding and usage of ABAQUS (see the Training section above for more information about these seminars). While not intended as stand-alone tutorial material, they are sufciently comprehensive that they can usually be used in that mode. The list of available lecture notes is included in the Documentation Price List.
Documentation Information Using ABAQUS Online Documentation: This online-only manual contains instructions for
theoretical aspects of ABAQUS. It is written to be understood by users with an engineering background. ABAQUS Verication Manual: This online-only volume contains more than 5000 basic test cases, providing verication of each individual program feature (procedures, output options, MPCs, etc.) against exact calculations and other published results. It may be useful to run these problems when learning to use a new capability. In addition, the supplied input data les provide good starting points to check the behavior of elements, materials, etc. Quality Assurance Plan: This document describes the QA procedures followed by ABAQUS. It is a controlled document, provided to customers who subscribe to either the Nuclear QA Program or the Quality Monitoring Service.
Update Information ABAQUS Release Notes: This document contains brief descriptions of the new features available
ABAQUS Scripting Interface. The manual describes how commands can be used to create and analyze
ABAQUS/CAE models, to view the results of the analysis, and to automate repetitive tasks. It also contains information on using the ABAQUS Scripting Interface or C++ as an application programming interface (API) to the output database.
ABAQUS Scripting Reference Manual: This online-only manual provides a command reference
that lists the syntax of each command in the ABAQUS Scripting Interface.
ABAQUS GUI Toolkit Users Manual: This online-only manual provides a description of the
ABAQUS GUI Toolkit. The manual describes the components and organization of the ABAQUS GUI. It also describes how you can customize the ABAQUS GUI to build a particular application.
ABAQUS GUI Toolkit Reference Manual: This online-only manual provides a command reference
that lists the syntax of each command in the ABAQUS GUI Toolkit.
Interfaces ABAQUS Interface for MSC.ADAMS Users Manual:
This document describes how to use the ABAQUS Interface for MSC.ADAMS, which creates ABAQUS models of MSC.ADAMS components and converts the ABAQUS results into an MSC.ADAMS modal neutral le that can be used by the ADAMS/Flex program. It is the basic reference document for the ABAQUS Interface for MSC.ADAMS. ABAQUS Interface for MOLDFLOW, which creates a partial ABAQUS input le by translating results from a MOLDFLOW polymer processing simulation. It is the basic reference document for the ABAQUS Interface for MOLDFLOW.
ABAQUS Interface for MOLDFLOW Users Manual: This document describes how to use the
Installation and Licensing ABAQUS Installation and Licensing Guide: This document describes how to install ABAQUS
and how to congure the installation for particular circumstances. Some of this information, of most relevance to users, is also provided in the ABAQUS Analysis Users Manual.
vi
CONTENTS
CONTENTS
Introduction: general
Notation
1.1.1 1.2.1 1.3.1 1.4.1 1.4.2 1.4.3 1.4.4 1.5.1 1.5.2 1.5.3 1.5.4 1.5.5
Notation
Finite rotations
Rotation variables
Deformation, strain, and strain rates
Deformation Strain measures Rate of deformation and strain increment The additive strain rate decomposition
Equilibrium, stress, and state storage
Equilibrium and virtual work Stress measures Stress rates State storage Energy balance
2. Procedures Overview
Nonlinear solution methods in ABAQUS/Standard Quasi-Newton solution technique Direct cyclic algorithm
Buckling and postbuckling
vii
CONTENTS
2.4.4 2.4.5 2.5.1 2.5.2 2.5.3 2.5.4 2.5.5 2.5.6 2.5.7 2.5.8 2.5.9 2.6.1 2.6.2 2.7.1 2.8.1 2.8.2 2.8.3 2.8.4 2.8.5 2.9.1 2.10.1 2.11.1 2.11.2 2.11.3 2.11.4 2.11.5 2.12.1 2.13.1
Eigenvalue extraction Variables associated with the natural modes of a model Linear dynamic analysis using modal superposition Damping options for modal dynamics Modal dynamic analysis Response spectrum analysis Steady-state linear dynamic analysis Random response analysis Base motions in modal-based procedures
Complex harmonic oscillations
Effective stress principle for porous media Discretized equilibrium statement for a porous medium Constitutive behavior in a porous medium Continuity statement for the wetting liquid phase in a porous medium Solution strategy for coupled diffusion/deformation
Coupled uid-solid analysis
Piezoelectric analysis
Heat transfer
Uncoupled heat transfer analysis Shell heat conduction Convection/diffusion Cavity radiation Viewfactor calculation
Coupled thermal-electrical analysis
viii
CONTENTS
Substructuring
Submodeling analysis
Fracture mechanics
J -integral evaluation
Stress intensity factor extraction T -stress extraction Prediction of the direction of crack propagation
Stress linearization
Stress linearization
Design sensitivity analysis
3.1.1 3.2.1 3.2.2 3.2.3 3.2.4 3.2.5 3.2.6 3.2.7 3.2.8 3.2.9 3.3.1 3.3.2 3.4.1 3.4.2 3.4.3 3.5.1
Solid element overview Solid element formulation Hybrid incompressible solid element formulation Solid isoparametric quadrilaterals and hexahedra Continuum elements with incompatible modes Triangular, tetrahedral, and wedge elements Generalized plane strain elements Axisymmetric elements Axisymmetric elements allowing nonlinear bending
Innite elements
ix
CONTENTS
Beam element formulation Euler-Bernoulli beam elements Hybrid beam elements Mass and inertia for Timoshenko beams Meshed beam cross-sections
Shell elements
3.5.2 3.5.3 3.5.4 3.5.5 3.5.6 3.6.1 3.6.2 3.6.3 3.6.4 3.6.5 3.6.6 3.6.7 3.6.8 3.6.9 3.7.1 3.7.2 3.7.3 3.8.1 3.9.1 3.9.2 3.9.3 3.9.4 3.9.5 3.9.6 3.9.7 3.9.8
Shell element overview Axisymmetric shell elements Shear exible small-strain shell elements Triangular facet shell elements Finite-strain shell element formulation Small-strain shell elements in ABAQUS/Explicit Axisymmetric shell element allowing asymmetric loading Transverse shear stiffness in composite shells and offsets from the midsurface Rotary inertia for 5 degree of freedom shell elements
Rebar
Rebar modeling in two dimensions Rebar modeling in three dimensions Rebar modeling in shell, membrane, and surface elements
Hydrostatic uid elements
Elbow elements Frame elements with lumped plasticity Buckling strut response for frame elements Tube support elements Line spring elements Flexible joint element Rotary inertia element Distributing coupling elements
4. Mechanical Constitutive Theories Overview
CONTENTS
Isotropic elasto-plasticity Stress potentials for anisotropic metal plasticity Rate-dependent metal plasticity (creep) Models for metals subjected to cyclic loading Porous metal plasticity Cast iron plasticity ORNL constitutive theory Deformation plasticity Heat generation caused by plastic straining
Plasticity for non-metals
4.3.2 4.3.3 4.3.4 4.3.5 4.3.6 4.3.7 4.3.8 4.3.9 4.3.10 4.4.1 4.4.2 4.4.3 4.4.4 4.4.5 4.4.6 4.5.1 4.5.2 4.5.3 4.5.4 4.6.1 4.6.2 4.7.1 4.8.1 4.8.2 4.8.3 4.9.1
Porous elasticity Models for granular or polymer behavior Critical state models Drucker-Prager/Cap model for geological materials Mohr-Coulomb model Models for crushable foams
Other inelastic models
An inelastic constitutive model for concrete Damaged plasticity model for concrete and other quasi-brittle materials A cracking model for concrete and other brittle materials Constitutive model for jointed materials
Large-strain elasticity
Mullins effect
Viscoelasticity
Hysteresis
5. Interface Modeling Contact modeling
Small-sliding interaction between bodies Finite-sliding interaction between deformable bodies Finite-sliding interaction between a deformable and a rigid body
xi
CONTENTS
Surface interactions
Contact pressure denition Pressure and uid ow in pore pressure contact Coulomb friction Thermal interface denition Heat generation caused by frictional sliding Heat generation caused by electrical current Surface-based acoustic-structural medium interaction
6. Loading and Constraints Dynamic loading
6.1.1 6.1.2 6.2.1 6.2.2 6.2.3 6.3.1 6.4.1 6.5.1 6.5.2 6.5.3 6.6.1 6.6.2 6.6.3 6.6.4 6.6.5 6.6.6
Drag, inertia, and buoyancy loading Airy wave theory Stokes wave theory
Incident wave loading
Pressure load stiffness Load stiffness for beam elements Pressure loadings on elbow elements
Multi-point constraints
Sliding constraint Shell to solid constraint Revolute joint Universal joint Local velocity constraint Kinematic coupling
7. References
References
7.1.1
xii
INTRODUCTION
1.1.1
INTRODUCTION: GENERAL
The ABAQUS system includes ABAQUS/Standard, a general-purpose nite element program; ABAQUS/Explicit, an explicit dynamics nite element program; and the Visualization module, an interactive postprocessing program that provides displays and output lists from output database les written by ABAQUS/Standard and ABAQUS/Explicit. This manual describes the theories used in ABAQUS. Many sections in this manual apply to both ABAQUS/Standard and ABAQUS/Explicit. Certain sections obviously apply only to either ABAQUS/Standard or ABAQUS/Explicit; for example, all sections in the chapter on procedures apply to ABAQUS/Standard, except the section discussing the explicit dynamic integration procedure, which applies to ABAQUS/Explicit. If it is not obvious to which program a section applies, it is clearly indicated. ABAQUS/Standard includes three added-cost options. The ABAQUS/Aqua option includes features specically designed for the analysis of beam-like structures installed underwater and subject to loading by water currents and wave action. The ABAQUS/Design option enables the user to parametrize input le quantities and write Python scripts to perform parametric studies. The ABAQUS/Foundation option offers more efcient access to the linear static and dynamic analysis functionality in ABAQUS/Standard. Certain aspects of the theory behind these options are described in this manual. The options are available only if the users license includes them. The objective of this manual is to dene the theories used in ABAQUS that are generally not available in the standard textbooks on mechanics, structures, and nite elements but are well known to the engineer who uses ABAQUS. The manual is intended as a reference document that denes what is available in the code. Nevertheless, it is written in such a way that it can also be used as a tutorial document by a reader who needs to obtain some background in an unfamiliar area. The material is presented in a way that should make it accessible to any user with an engineering background. Some of the theories may be relatively unfamiliar to such a user; for example, few engineering curricula provide extensive background in plasticity, shell theory, nite deformations of solids, or the analysis of porous media. Yet ABAQUS contains capabilities for all of these models and many others. The manual is far from comprehensive in its coverage of such topics: in this sense it is only a reference volume. The user is strongly encouraged to pursue topics of interest through texts and papers. Chapter 7, References, at the end of this manual lists references that should provide a starting point for obtaining such information. (ABAQUS does not supply copies of papers that have appeared in publications other than those of ABAQUS. EPRI reports can be obtained from Research Reports Center (RRC), Box 50490, Palo Alto, CA 94303.) Chapter 1, Introduction and Basic Equations, discusses the notation used in the manual, some basic concepts of kinematics and mechanicssuch as rotations, stress, and equilibriumas well as the basic equations of nonlinear nite element analysis. Chapter 2, Procedures, describes the various analysis procedures (nonlinear static stress analysis, dynamics, eigenvalue extraction, etc.) that are available in ABAQUS. Chapter 3, Elements, describes the element formulations. Chapter 4, Mechanical Constitutive Theories, describes the mechanical constitutive theories. Chapter 5, Interface Modeling, discusses the most important aspects of the contact/interaction formulation in ABAQUS/Standard. Chapter 6, Loading and Constraints, describes the formulation of some of the more complicated load types and multi-point constraints.
1.1.11
NOTATION
1.2.1
NOTATION
Notation is often a serious obstacle that prevents an engineer from using advanced textbooks; for example, general curvilinear tensor analysis and functional analysis are both necessary in some of the theories used in ABAQUS, but the unfamiliar notations commonly used in these areas often discourage the user from pursuing their study. The notation used in most of this manual (direct matrix notation) may be unfamiliar to some readers; but it is not difcult or time consuming to gain enough familiarity with the notation for it to be useful, and it is denitely worthwhile. This notation is commonly used in the modern engineering literatureit is a shorthand version of the familiar matrix notation used in many older engineering textbooks. The notation is appealingonce it is understoodbecause it allows the equations to be developed concisely, and the physical ideas can be perceived without the distraction of the complexities that arise from the choice of the particular basis system that will eventually be used to express the same concepts in component form. Because the notation has become so standard in the literature, the user who wishes or needs to read textbooks and papers that are related to the use of ABAQUS will nd that familiarity with this notation is desirable. Both direct matrix notation and component form notation are used in the manual. Both notations are described in this section. Direct matrix notation is used whenever possible. However, vectors, matrices, and the higher-order tensors used in the theories must eventually be written in component form to store them as a set of numbers on the computer. Thus, both ways of writing these quantities will be needed in the manual.
Basic quantities
The quantities needed to formulate the theory are scalars, vectors, second-order tensors (matrices), andoccasionallyfourth-order tensors (for example, the stress-strain transformation for linear elasticity). In direct matrix notation these are written as: a scalar value a vector with the transpose a second-order tensor or matrix with the transpose and a fourth-order tensor
a or bac aT or fag a or a aT or a T
[ ] [ ]
A a
Vectors and second-order tensors (matrices) are written in the same way: they are distinguished by the context. In direct matrix notation there is generally no need to indicate that a vector must be transposed. The context determines whether a vector is to be used as a column vector or as a row vector T . In this case the transpose superscript is only used to improve the readability of an
1.2.11
NOTATION
expression. On the other hand, for second-order nonsymmetric tensors the addition of a transpose superscript will change the meaning of an expression. This representation of vectors and tensors is very general and convenient for developing the theory so that the equations can be understood easily in terms of their physical meaning. However, in actual computations we have to work with individual numbers, so vectors and tensors must be expressed in terms of their components. These components are associated with an axis system that denes a set of base vectors at each point in space. The simplest axis system is rectangular Cartesian, because the base vectors are orthogonal unit vectors in the same direction at all points. Unfortunately, we need more generality than this because we will be dealing with shells and beams, where stress, strain, etc. are most conveniently described in terms of directions on the surface of the shell (or associated with the axis of the beam), and these usually change as we move around on the surface. To retain this necessary generality and express vectors and matrices in component form, we introduce a general set of base vectors, , = 1; 2; 3, which are not necessarily orthogonal or of unit length but are sufcient to dene the components of a vector (for this purpose they must not be parallel or have zero length). A vector can then be written
of the ABAQUS Analysis Users Manual, for a description of how base vectors are chosen for surface elements in ABAQUS), and then the a are obtained. To save writing, we adopt the usual summation convention that a repeated index is summedin this case over the range 1 to 3so that the above equation is written
a = a 1 e 1 + a2 e 2 + a 3 e 3 ; where the numbers a1, a2 , and a3 are the components of a associated with e1 , e2 , and e3. In actual cases the e are chosen for convenience (for example, see Conventions, Section 1.2.2
a = a e :
Likewise, the component form of a matrix will be
a = ea e = a ee ;
or, written out,
While we will need such completely general base vectors for describing the stresses and strains on shells and beams, in many cases it is convenient to use rectangular Cartesian components so that the are orthogonal unit vectors. To distinguish this particular case, we will use Latin indices instead of Greek indices. Thus, are a set of general base vectors; while i are rectangular Cartesian base
1.2.12
NOTATION
vectors; and a is the component of the vector a along a general base vector, while ai , i = 1; 2; 3, is the component of a along the ith Cartesian direction. Vector and tensor concepts and their representation are discussed in many textbookssee Flugge (1972), for example.
Basic operations
The usual matrix and vector operators are indicated in this manual as follows: Dot product of two vectors: a = b1c (The dot symbol denes this operation completely, regardless of whether b or c is transposed i.e., b 1 c = bT 1 c:) Cross product of two vectors: a = b2c Matrix multiplication:
a =b1c
(It is implicitly assumed that b and c are dimensioned correctly, as needed for the operation to make sense; in addition, if b is a nonsymmetric tensor, bT 1 c 6= b 1 c:) Scalar product of two matrices: a=b:c This operation means that corresponding conjugate components of the two matrices are multiplied as pairs and the products summed. Thus, for instance, if b is the stress matrix, , and c the conjugate " " rate of strain matrix, d", then : d" would give the rate of internal work per volume, dW I . It is also necessary to dene the dyadic product of two vectors:
a = bc
or
a = bcT
This operation creates a second-order tensor (or dyad) out of two vectors. In component notation this notation is equivalent to aij = bicj . A matrix of derivatives,
means
da =
@a @b
1
@a ; @b
db :
Throughout this manual it will be assumed implicitly that, when a derivative is taken with respect to time, we mean the material time derivative; that is, the change in a variable with respect to time whilst looking at a particular material particle. When this is not the case for a particular equation, it will be stated explicitly when the equation appears.
1.2.13
NOTATION
Provided that we are careful about interpreting (@ a=@ b) in the manner illustrated above, standard concepts of elementary calculus clearly hold; for example, if a is a vector-valued function of the vectorvalued function b, which in turn is a vector-valued function of c, that is a = a(b(c)), then
da =
or, if a(b; c):
@a @b 1 1 dc; @b @c
da =
@a @a 1 db + 1 dc: @b @c
Due to these properties many useful results can be obtained quickly and expressed in a compact, easily understood, form.
Components of a vector or a matrix in a coordinate system
In the previous section we introduced the idea that a vector a or a matrix a can be written in terms of components associated with some conveniently chosen set of base vectors, e . We now show how the components a (or a ) are obtained. We can do so using the dot product. For each of the three base vectors, e, we dene a conjugate base vector e, as follows. Choose e1 as normal to e2 and e3 , such that the dot product e1 1 e1 = 1. Similarly, choose e2 normal to e3 and e1 , such that e2 1 e2 = 1; and e3 normal to e1 and e2, such that e3 1 e3 = 1. Thus,
e1 1 e1 = 1 e2 1 e1 = 0 e3 1 e1 = 0
; ; ;
e1 1 e2 = 0 e2 1 e2 = 1 e3 1 e2 = 0
; ; ;
e1 1 e3 = 0 e2 1 e3 = 0 e3 1 e3 = 1
where = 1 if = , and = 0, is called the Kronecker delta.) In matrix is the unit matrix I: we can also write the above equation dening e1 , e2 , and e3 in notation matrix form as
; otherwise. (
e
1e
8 < :
be c
1 2 be c 3 be c
9 = ;:
=I
so that, if one set of base vectorsei , sayis known, the others are easily obtained. With this additional set of base vectors, we can immediately obtain the components of a vector or a matrix as follows. Consider a vector a. Then a 1 e = a e 1 e (writing a in component form, using the basis vectors e ), and since e 1 e = = 1, only if = ,
a 1 e =
a e 1 e = a = a :
1.2.14
NOTATION
a = a 1 e
by expressing a as components associated with the e base vectors, a = ae . Similarly, for a matrix,
a
and
= (e )
T 1 a 1 e
e 1 a 1 e ; e 1 a 1 e :
a = (e )T
a 1 e
These component denitions are particularly convenient for calculating the dot product of two vectors, for we can write
a 1 b = ( a e ) 1 (b e ) = a b e 1 e ; = a b ; = a b ;
which is
a 1 b = a1 b1 + a2 b2 + a3 b3: a : b = a b ;
that is, we simply multiply corresponding entries in the a and b arrays, arranged as matrices, and then sum the products. Finally, on the computer we need to store only one form of component: a , a or a , a . We can always go from one to the other using the metric tensor, g , and its inverse, g , which are dened as
g =
e 1 e ; e 1 e :
and For
a =
g =
a 1 e (from above), (expressing a in component form); = a e 1 e (by the denition of g ): = a g Thus, a = g a ; similarly a = g a , and, by extension, for matrices,
a = g
g a
1.2.15
NOTATION
and
a
g =
= g g a
:
e e = e e = g :
1 1
The two sets of base vectors and components of vectors or matrices associated with them are named as follows:
e e
a (or a ) a (or a )
are covariant base vectors, are contravariant base vectors, are covariant components of a vector (or matrix), are contravariant components of a vector (or matrix).
Thus, the contravariant components are those associated with the covariant base vectors, a = a e , and vice versa. The simplest case is when the basis is a set of orthogonal unit vectors (a rectangular Cartesian system) because thenfrom the denition e 1 e = we see that e = e, and so and we need not distinguish the type of component. Whenever possible a rectangular a = a Cartesian system is chosen, so the type of component need not be distinguished. This system is discussed in more detail in the sections on beam elements and shell elements.
Components of a derivative
Consider a vector-valued function, b, which is expressed in component form on a basis system, e . Let the vector-valued function a depend on b: a(b). Then
@a edb da = @b
1
@a e ; @b @a ;
1
@b @
meaning
d =
Now suppose
@b
db :
1.2.16
NOTATION
Then
da =
@b
a db :
where
H =
@b
Readers who are familiar with general curvilinear tensor analysis will recognize H as the covariant derivative of a with respect to b , often written as aj . The advantage of the direct matrix notation is clear: because we can imagine a and b as vectors in space, we have a physical understanding of what we mean by @ a=@ b; it is the change in the vector-valued function a as a function of another vector-valued function b. For computations we must express a and b in component form. Then provides the necessary components once we have chosen convenient basis systems: e for for a. Typically e and E will both be the simple rectangular Cartesian bases
H = aj = @a E @ b e
1 1
@a E @ b e :
1 1
b and E
@_ @_ + @ @
u x
u x
T !
which requires the matrix @ u=@ x to be evaluated, where u is the velocity of the material currently _ _ _ owing through the point x in space. Let us now derive the components of " when the basis system _ for both u and x is the cylindrical system that we usually choose for axisymmetric problems, with the basis vectors e1 (radial) = (cos ; sin ; 0)
(in ABAQUS for axisymmetric cases we always take the components in this orderradial, axial, circumferential). These basis vectors are orthogonal and of unit length, so that e = e:
1.2.17
NOTATION
dx = dr e1 + dz e2 + r d e3 ;
so that Thus,
dx1
dr;
_ @u @x
dx2 = dz;
=
_ _ _ @u @u 1 @u ; ; @r @z r @
where so that
u = u r e 1 + uz e 2 + u e 3 ; _ _ _ _
@u _ @r @u _ @z _ 1 @u r @ @ e1 @
=( = = =
@ ur _ @ uz _ @ u _ e1 + e2 + e3 @r @r @r @ ur _ @u _ @u _ e1 + z e2 + e3 @z @z @z 1 @ ur _ _ _ 1 @ uz 1 @ u 1 @ e1 e1 + e2 + e3 + u r _ r @ r @ r @ r @
u _
@ e3 : @
We know that
so that
and thus,
@ ur =@z _ @ uz =@z _ @x @ u =@r @ u =@z _ _ The components of the strain rate are thus "rr _
rz _
and
=
2 @ u =@r _ @ u 4 r _ _ = @ uz =@r
@ ur _ ; @r @ ur _ @z
z _
+
(1=r )@ uz =@ _
0 u ) _
3 5:
"zz _
@ uz _ ; @z r _
" _
@u _
@
+ ur _ 1
= 2"rz = 2"zr = _ _
@ uz _ ; @r
= 2"r = 2"r = _ _ 1
@u _
r @
0 u _
+
@ u _ ; @r
@ uz _ @ u _ + : r @ @z For the case of purely axisymmetric deformation, u = 0 and @ ur =@ _ _ simplify to the familiar expressions
= 2"z = 2"z = _ _
@ uz =@ _
= 0,
so these results
"rr _
@ ur _ ; @r
"zz _
@ uz _ ; @z
" _
ur _ r
1.2.18
NOTATION
rz _
u _ _ u _ = @@zr + @@rz ; r = z = 0: _
In summary, direct matrix notation allows us to obtain all our fundamental results without reference to any particular choice of coordinate system. Careful application of the concept of the covariant derivative then allows these general results to be projected into component form for computation.
Virtual quantities
The concepts of virtual displacements and virtual work are fundamental to the development. Virtual quantities are innitesimally small variations of physical measures, such as displacement, strain, velocity, and so on. The virtual variation of a scalar quantity a is indicated by a; of a vector or matrix by . We extend this notation to such expressions as
@ v
@
which is the symmetric part of the spatial gradient of a virtual vector eld . This notation corresponds to the virtual rate of deformation (a measure of strain rate) if is a virtual velocity eld.
Most structural problems concern the description of the way a structure behaves as it is loaded and moves from its reference conguration. Thus, we often compare positions of a point in the current (deformed) conguration and a reference conguration that is usually chosen as the conguration when the structure is unloaded or, in the case of geotechnical problems, when the model is subject only to geostatic stresses. To distinguish these congurations, we use lowercase type ( ) to indicate the current position and uppercase type ( ) to indicate the initial position of the same material point in the same spatial coordinate frame. In ABAQUS we almost always store the rectangular Cartesian and . The exception is in axisymmetric structures, where radial (r) and axial (z ) components of components are stored.
Nodal variables
So far we have discussed quantities that are considered to be associated with all points in a model. The nite element approximation is based on assuming interpolations, by which displacement, position, andoftenother variables at any material point are dened by a nite number of nodal variables. In this manual we use uppercase superscripts to refer to individual nodal variables or nodal vectors and adopt the summation convention for these indices. Hence, the interpolation can be written quite generally as
a = N N aN ;
1.2.19
NOTATION
where is some vector-valued function at any point in the structure; N (g ), N = 1; 2 . . . up to the total number of variables in the problem, is a set of N vector interpolation functions (these are functions of the material coordinates, g ); and aN , N = 1; 2 . . . is a set of nodal variables. In some sections in this manual we need to describe operations on nodal variables for the complete system of nite element equations. In these sections we use the classical matrix-vector notation. In this notation fag represents a column vector containing nodal variables, bac represents a row vector, and a matrix is written as [A]. Common operations are the scalar product between two vectors,
c
= ba c f bg
(which is equivalent to
(which is equivalent to cM
MN bN in index notation).
1.2.110
ROTATION VARIABLES
1.3.1
ROTATION VARIABLES
Since ABAQUS contains such capabilities as structural elements (beams and shells) for which it is necessary to dene arbitrarily large magnitudes of rotation, a convenient method for storing the rotation at a node is required. The components of a rotation vector are stored as the degrees of freedom 4, 5, and 6 at any node where a rotation is required. The nite rotation vector consists of a rotation magnitude = kk and a rotation axis or direction in space, p = =kk. Physically, the rotation is interpreted as a rotation by radians around the axis p. To characterize this nite rotation mathematically, the rotation vector is used to dene an orthogonal ^ transformation or rotation matrix. To do so, rst dene the skew-symmetric matrix associated with by the relationships ^ ^ 1 = 0 and 1 v = 2 v for all vectors v :
^ is called the axial vector of the skew-symmetric matrix . In matrix components relative to the standard 1 2 3 T , then Euclidean basis, if =
2 ^ [] = 4
3 2
0 3
1
0
2 1 5 :
0
^ In what follows, a will be used to denote the skew-symmetric matrix with axial vector a. ^ A well-known result from linear algebra is that the exponential of a skew-symmetric matrix is an orthogonal (rotation) matrix that produces the nite rotation . Let the rotation matrix be C, such that C01 = CT . Then by denition, ^ ^ C = exp[] = I + + 1 2!
^
However, the above innite series has the following closed-form expression
^ C = exp[] = cos kkI + sin
kk kk
^
(1
:
(1.3.11)
In components,
Cij
= cos
ij
+ (1
where p = f p1
p2 123
cos )p
i pj
+ sin
ikj pk ;
231
312
= 1;
132
213
321
1;
= 0:
It is this closed-form expression that allows the exact and numerically efcient geometric representation of nite rotations.
1.3.11
ROTATION VARIABLES
Quaternion parametrization
Even though ABAQUS stores and outputs the rotation vector, quaternion parameters prove to be an efcient and convenient way to treat nite rotations computationally. Let q0 2 R be a scalar, and let 2 R3 be a vector eld. The quaternion q is simply the pairing
q = ( q0 ; q ) :
To associate
q with the nite rotation vector , dene the following: sin k=2k : q = cos k=2k and q =
0
kk
(1.3.12)
^ ^ exp[] = (2q2 0 1)I + 2q0q + 2qq : 0 ^ By the convention introduced above, q is the skew-symmetric matrix with axial vector q.
Compound rotations
For a more detailed discussion of quaternion algebra and its relation to other representations of nite rotations, see the discussion by Spring (1986).
A compound rotation is the successive application of two or more rotation elds. In geometrically linear problems compound rotations are obtained simply as the linear superposition of the individual . Let 1 (linearized) rotation vectors. This fact follows directly from the series expansion for 1, 2 and 2 be innitesimal rotations. Thus, 1 2 , and
exp[^]
In geometrically nonlinear analysis compound rotations are no longer additive. Furthermore, they are not commutative; that is, the order of application is important. A signicant exception occurs when the multiple rotations share the same rotation axis. This special case is investigated further below. A detailed example of a nite compound rotation is given in Conventions, Section 1.2.2 of the ABAQUS Analysis Users Manual. Let i be the orthogonal transformation representing the compound rotation dened as the product ; ; ; i. (For the case of specied of a set of individual or incremental rotations p , for p boundary conditions i is the nal product after i steps of all the specied rotations p ; for the iterative numerical solution procedure i is the total rotation after i increments, where p , for p ; ; i, is the converged rotation eld solution at each increment.) By denition, the compound rotation is the product
1C
= 1 2 ...
= 1 ...
1C 1C
C = 1 C 1 1 C 01 1 1 1 1 C 1 ;
i i i
1.3.12
ROTATION VARIABLES
C = 1 C C 01 :
i i 1 i
It is important to note that 1 , which is interpreted as the nite rotation superposed on the , which is interpreted as the nite rotation superposed nite rotation , is different from 1 . on the nite rotation Although compound rotations are dened in terms of orthogonal matrices, in a numerical context the rotation vectors (or equivalently the quaternion parameters) associated with the rotation matrices are the degrees of freedom. Compound rotations are performed as follows: Given a quaternion q0 ; and an incremental (nite) rotation q q0 ; , where q is parametrization q dened in terms of an incremental rotation vector by Equation 1.3.12, the total or compound r0 ; , which is calculated as rotation is given by the quaternion r
1C C
1C
C 1C
1C
= ( q)
= ( r) r = 1q q :
1 = (1 1q)
Here
(1.3.14)
Equation 1.3.14 allows for the update of rotation elds without ever calculating the orthogonal matrix from the quaternion and without performing a matrix multiplication. Furthermore, all operations are singularity free regardless of the magnitude of the incremental rotation eld . The nal (total) rotation vector can be calculated from the quaternion by inverting Equation 1.3.12. For the special case when compound rotations share the same rotation axis, the compound rotation reduces to an additive form. Let q and q have the same rotation axis . Then q k= k; k= k , q k = k; k = k , and
= (cos
1 2 sin 2 p) 1 = (cos 1 2 sin 1 2 p) 1q q = (cos k1=2k cos k=2k 0 sin k1=2k sin k=2k ; cos k1=2k sin k=2kp + cos k=2k sin k1=2kp) ; 1q q = (cos k( + 1)=2k; sin k( + 1)=2kp) :
which reduces to
For output purposes it is necessary to extract the rotation vector corresponding to a given quaternion. r0 ; be the quaternion, and let be the rotation The extraction procedure is as follows: Let r vector. Thus,
= ( r)
and
kk = 2 tan01
krk
r0
= kk
r : krk
(1.3.15)
1.3.13
ROTATION VARIABLES
It is important to note that the extraction of the rotation vector from the quaternion is not unique. ;0 ; ; ; ABAQUS The magnitude kk is determined only up to the addition of n, n will always choose that rotation vector such that kk < .
= . . . 1 0 1 . ..
As an example of the utility of the quaternion parameters, consider the incremental update of a director eld for either a beam or shell analysis. At some stage of the solution the director eld i i , the quaternion parametrization of the rotation eld qi , and the incremental rotation eld are , known at increment i. To update the director eld by the incremental rotation to increment i proceed as follows: First calculate the quaternion parametrization of the incremental rotation:
+1
and
c 1 i , where c is calculated with The director eld at i is then dened as i+1 Equation 1.3.13. Thus, the director is calculated directly from the quaternion as
i
+1
t = exp[1 ] t
i i i
exp[1 ]
i
Furthermore, the update of the rotation eld is obtained by quaternion multiplication q i+1 and is dened by
q
i
= 1q q
i
In the development of the balance equations, it is necessary to calculate the variation of the rotation eld. Consider the vector eld , which is obtained by rotation of the reference vector eld :
a = C1A: a = C 1 A ;
where is the linearized rotation matrix; that is, the variation of the orthogonal tensor other hand, the variation can be dened in terms of the linearized rotation eld :
C.
On the
b b a = 2 a = 1 a = 1 C 1 A :
b C = 1 C :
1.3.14
ROTATION VARIABLES
It is important to note that the linearized rotation , which is analogous to the angular velocity in dynamics, is not the variation of the rotation vector . By a straightforward (but involved) calculation, it can be shown that the variation of the rotation vector () is related to the linearized rotation by =
H ( )
;
+ 1
(1.3.16)
where
H()01 =
Let d represent an innitesimal change in the rotation eld. A direct calculation of the variation , which is equivalent to calculation of the second variation of either or , leads to an expression of d that is not symmetric in the variations and the changes d. However, it is shown in Simo (1992)
C a
that the correct denition of the Hessian operatorthat is, the covariant derivative of the weak form of the balance equationsrequires only the symmetric part (with respect to the variations) of the second variation. Thus, without loss of generality, we can write
d( 1 b c c b C) = 2 ( 1 d 1 C + d 1 1 C) :
C can be written as
3 3
12 2
2 (d 2 a) + d 2 ( 2 a)
2
0( 1 d )a + 1 2
(
Taking the time derivative of the rotation matrix, we nd with the same arguments as used in the calculation of the variations that
_ b C = ! 1C; b _ b b C = ! 1C +! 1! 1C;
b where ! is the angular velocity matrix. Equivalently, the rst and second time derivative of written as
a are
a = ! 2a; _ _ a = ! 2 a + ! 2 (! 2 a) :
1.3.15
ROTATION VARIABLES
The instantaneous angular velocity vector ! is related to the time rate of change of the rotation vector by the relation
_ ! = H() 1 ;
where H( ) is given by Equation 1.3.16. In the linearization of the dynamic balance equations, it is necessary to calculate the variation of the angular velocity, d! . This quantity, however, can be calculated only by linearizing the specic algorithm used for the time integration of the dynamic equations.
Coupling of rotations: constant velocity joint
Next, a more rigorous treatment of the two-dimensional constant velocity joint described in MPC, Section 25.2.13 of the ABAQUS Analysis Users Manual, is presented. This derivation exemplies some of the issues associated with the treatment of nite rotations. Uniform collapse of straight and curved pipe segments, Section 1.1.5 of the ABAQUS Benchmarks Manual, deals with a different nite rotation constraint and tackles additional complications. Let a, b, c (see Figure 1.3.11) be the nodes making up the joint, with a the dependent node.
a c c y b b
x
Figure 1.3.11 Nonlinear MPC exampleconstant velocity joint. The joint is operated by prescribing an axial rotation c = c ex at c and an out-of-plane rotation b = bez at b. The compounding of these two prescribed rotation elds will determine the total rotation at a. We can formally write this constraint as follows:
f (a ; b ; c ) = a 0 b c = 0:
The constraint can be written in terms of the rotation matrices as
(1.3.17)
ROTATION VARIABLES
With the previously dened variational expressions, the constraint can be linearized as
c a d 0 b 0 C( b) 1 c 1 CT (b) = 0; c
which can be written in vector form as
0 b 0 C(b ) 1 c = 0:
Since
C( b) = @ sin b
0
cos b
0 sin b
0
cos b
0 1
0A;
the linearized constraint is indeed identical to the one derived based on simple linear considerations in the ABAQUS Analysis Users Manual. The linearized constraint is used for the calculation of equilibrium. It can also be used for the recovery of the dependent rotation, a , as is done in the ABAQUS Analysis Users Manual. The resulting rotation will satisfy the constraint approximately (unless one of the angles b or c is constant, in which case the constraint is linear and the recovery is exact). For an exact enforcement of the constraint, user subroutine MPC must dene the components of the total rotation vector a exactly. To do so, a must be updated based on the current values of b c and . This is most easily accomplished with the aid of the quaternion parameters. Let q b = b=2); sin(b =2)e ) and q c = (cos(c =2); sin(c =2)e ) be the quaternion parameterizations (cos( z x associated with the nite rotation vectors b and c , respectively. The total compound rotation a is a given by the quaternion q a = (q0 ; qa ), where
qa = cos(b =2) sin(c =2)ex + sin(b =2) sin(c =2)ey + cos(b =2) sin(c =2)ez ;
a according to the quaternion compound formula Equation 1.3.14. The rotation vector is extracted a as follows: from the quaternion q
= a
qa kqa k with
a = 2 tan01
kq a k ;
a q0
where kqa k is the norm of the vector qa . MPC, Section 25.2.13 of the ABAQUS Analysis Users Manual, shows the implementation of the linearized form of the constraint in user subroutine MPC. The implementation of the exact nonlinear constraint is shown below:
1.3.17
ROTATION VARIABLES
SUBROUTINE MPC(UE,A,JDOF,MDOF,N,JTYPE,X,U,UINIT,MAXDOF,LMPC, * KSTEP,KINC,TIME,NT,NF,TEMP,FIELD) C INCLUDE ABA_PARAM.INC C DIMENSION UE(MDOF), A(MDOF,MDOF,N), JDOF(MDOF,N), X(6,N), * U(MAXDOF,N), UINIT(MAXDOF,N), TIME(2), TEMP(NT,N), * FIELD(NF,NT,N) PARAMETER( SMALL = 1.E-14 ) C IF ( JTYPE .EQ. 1 ) THEN A(1,1,1) = 1. A(2,2,1) = 1. A(3,3,1) = 1. A(3,1,2) = -1. A(1,1,3) = -COS(U(6,2)) A(2,1,3) = -SIN(U(6,2)) C JDOF(1,1) = 4 JDOF(2,1) = 5 JDOF(3,1) = 6 JDOF(1,2) = 6 JDOF(1,3) = 4 C CPHIB = COS(0.5*U(6,2)) SPHIB = SIN(0.5*U(6,2)) CPHIC = COS(0.5*U(4,3)) SPHIC = SIN(0.5*U(4,3)) C QA0 = CPHIB*CPHIC QAX = CPHIB*SPHIC QAY = SPHIB*SPHIC QAZ = CPHIB*SPHIC C QAMAG = SQRT( QAX*QAX + QAY*QAY + QAZ*QAZ ) IF ( QAMAG .GT. SMALL ) THEN PHIA = 2.*ATAN2( QAMAG , QA0 ) UE(1) = PHIA*QAX/QAMAG UE(2) = PHIA*QAY/QAMAG UE(3) = PHIA*QAZ/QAMAG ELSE UE(1) = 0. UE(2) = 0.
1.3.18
ROTATION VARIABLES
1.3.19
DEFORMATION
1.4.1
DEFORMATION
In any structural problem the analyst describes the initial conguration of the structure and is interested in its deformation throughout the history of loading. The material particle initially located at some position X in space will move to a new position x: since we assume material cannot appear or disappear, there will be a one-to-one correspondence between x and X, so we can always write the history of the location of a particle as
x = x(X; t) (1.4.11) and this relationship can be invertedwe know X when we know x and t. Now consider two neighboring particles, located at X and at X + dX in the initial conguration. In the current conguration we must
have
dx =
using the mapping Equation 1.4.11. The matrix
@x 1 dX @X
(1.4.12)
@x F = @X
is called the deformation gradient matrix, and Equation 1.4.12 is written
(1.4.13)
dx = F 1 dX :
(1.4.14)
As the material behavior depends on the straining of the material and not on its rigid body motion, those parts of the motion in the vicinity of a material point must be distinguished. Looking at an innitesimal gauge length dX emanating from the particle initially at X, we can measure its initial and current lengths as dL2 = dXT 1 dX and dl2 = dxT 1 dx; so the stretch ratio of this gauge length is
=
dl dL
r dx T
: dX T 1 dX
dx
(1.4.15)
If = 1, there is no strain of this innitesimal gauge lengthit has undergone rigid body motion only. Now using Equation 1.4.14,
dx T 1 dx = dX T
so that, from Equation 1.4.15,
F T F dX ;
1 1
1.4.11
DEFORMATION
2 =
T X p dX 1 FT 1 F 1 p dT T 1 dX dX dX 1 dX T T = N 1 F 1 F 1 N;
(1.4.16)
where N is a unit vector in the direction of the gauge length dX. Equation 1.4.16 shows how to measure the stretch ratio associated with any direction, N, at any material point dened by X (or by x). Useful results are obtained when we vary the direction dened by N at a particular material point and look for stationary values of the stretch ratio, . Since N must always be a unit vector, stationary values of 2 are obtained by solving the constrained variational equation
8 T N
1 FT 1 F 1 N 0 p (NT 1 N 0 1)
NT 1 N = 1 :
= 0;
Taking the variation gives back the constraint (conjugate to p) and, conjugate to N, gives
(
F T 1 F 0 p I) 1 N = 0 : N
Taking the dot product of the left-hand side of this equation with Equation 1.4.16 identies p = 2 , so Equation 1.4.17 is
(
F T 1 F 0 2 I) 1 N = 0 :
(1.4.18)
This problem is an eigenvalue one that can be solved for the three extreme values of 2. Since is always real and positive (and nonzero), 2 > 0, and hence FT 1 F must be positive denite. Equation 1.4.18 thus gives three real, positive eigenvalues, I , II , III , the principal stretches, with three corresponding eigenvectors, NI , NII , NIII , which will be orthogonal, by Equation 1.4.18, if the corresponding eigenvalues are different, and can be orthogonalized otherwise. The NI are the principal directions of strain. Now let nI , nII , nIII be unit vectors corresponding to NI , NII , NIII , but in the current conguration, so that, using Equation 1.4.14,
1 nI = F 1 NI ; etc:
I
Then
nI T 1 nII = 1 NI T 1 FT 1 F 1 NII
I II
I II
2 II
NI T 1 NII
=0
1.4.12
DEFORMATION
where is the same pure rigid body rotation matrix in each of these equations. A pure rigid body motion T = 01. Comparing the principal stretch matrix has the property that its inverse is its transpose: directions in the current and original congurations, therefore, isolates the rigid body rotation and the stretch. Finding the principal stretch ratios and their directions thus provides one solution to the problem of isolating straining motion and rigid body motion in the vicinity of a material point. Now consider a gauge length in the reference conguration, d I , directed along I . The same innitesimal material line in the current conguration will be along I and will be stretched by I , so that
nI , nII , and nIII will also be an orthogonal set. Since nI = R N I ; nII = R NII ; nIII = R NIII ;
1 1 1
dx I
Similarly, along the other principal directions,
= I R dX I :
1
dxII
and
= II R dXII
1
Since ( I , II , III ) is an orthonormal set of base vectors in the reference conguration, any innitesimal can be written in terms of its components in this basis: material line (gauge length) d at
N N N
dxIII
= III R dXIII :
1
X X
XI moves and stretches to the corresponding dxI , as dened above. Thus, the current x dx = dxI + dxII + dxIII = I R dXI + II R dXII + III R dXIII = I R NI NI T + II R NII NII T + III R NIII NIII T dX = I nI NI T + II nII NII T + III nIII NIII T dX 0 1 = I nI nI T + II nII nII T + III nIII nIII T R dX
1 1 1 1 1 1 1 1 1 1
dX I
= N I N I T dX ;
1
etc.
which we write as
dx = V 1 R 1 dX ;
1.4.13
(1.4.19)
DEFORMATION
where
(1.4.110)
F = V R;
1
(1.4.111)
which is the polar decomposition theoremthat any motion can be represented as a pure rigid body rotation, followed by a pure stretch of three orthogonal directions. The polar decomposition theorem is important because it allows us to distinguish the straining part of the motion from the rigid body rotation. Specically, F completely denes the relative motions of material particles in the innitesimal neighborhood of the material particle that was at X in the reference conguration; and the left stretch matrix, V, completely denes the deformation of the material particles at X. The rotation matrix R denes the rigid body rotation of the principal directions of strain (NI in the reference conguration; nI in the current conguration). R represents only the rigid body rotation of the material at the point under consideration in some average sense: in a general motion, each innitesimal gauge length emanating from a material particle has a different amount of rotation. This distinction between the rotation of the principal directions of strain, R, and the rotations of individual directions in the material becomes signicant when we must discuss large deformations of nonisotropic materials. Nevertheless, we have established an important result: if F = R only, we know there is no deformation of the material in the immediate neighborhood of the point originally at X and currently at x, since in this case V = I and so I = II = III = 1. V 0 I must be nonzero for there to be any deformation of the material at the point in question: in this sense V 0 I (and, hence, V itself) is sufcient to dene the deforming part of the motion (it contains complete information about all except pure rigid body rotation of the point). For this reasonso that, later in the development, we will be able to link the kinematics to the stressing of the materialwe will need to be able to isolate V from F. It is easy to obtain V 1 V, for
F 1 FT = V 1 R 1 R T 1 V T = V 1 V;
since RT = R01 and V is symmetric. Since we originally dened V from the principal stretches and their principal directions in the current conguration as
then
V 1 V = I 2 nI nI T + II 2nII nII T + III 2 nIII nIII T : (1.4.112) We see that the eigenvalues of F 1 FT , are I 2, II 2 , and III 2 , and the corresponding eigenvectors are nI , nII , and nIII . We can then construct V. The deformation at the point is, thus, readily obtained by
=
F 1 FT
1.4.14
DEFORMATION
multiplying a 3 2 3 matrix with its transpose (F 1 FT ) and solving the real eigenproblem for the resulting (symmetric) matrix. We can then obtain the rotation R as
R = V 01 F :
1
Since
V has been constructed from its eigenvalues and eigenvectors, its inverse is immediately available: V 01 = 1 n n T + 1 n n T + 1 n n T :
I I I II II II III III III
So far we have written the results quite generally, without reference to any particular coordinate system. To perform computations we must choose a basis system to express these results as arrays of individual numbers. We now do so with some generality with respect to the choice of basis system. The justication for retaining generality at this stage is twofold: as an exercise, to provide a little more familiarity in the notation system we have chosen to use in this manual, and because we do need somebut, as it turns out, not allof the generality when we have to deal with shell elements, where it is undesirable to use the rectangular Cartesian base vectors of the global, spatial system because the natural orientation of the shell reference surface causes us to prefer to choose two of the base vectors to be tangent to the shells reference surface and the other to be normal to this surface. This preference causes us to need two basis systems: one associated with the body in its current conguration, when the point in question is at x, and one associated with the body in its reference conguration, when the same point was at X, because the orientation of the shells reference surfacewhich determines our choice of basis vectorswill be quite different in these two congurations. We will write e , = 1; 2; 3, as the basis vectors chosen to write components associated with the current conguration (so that any vector a associated with the current conguration is written as a = a e) and E , = 1; 2; 3, as the basis at the same material point but in the reference conguration. (Since we assume that both of these basis systems are adequate to express any vector-valued function by its components in the basis systemthat is, the basis vectors are not linearly dependenteither would, in principal, serve for both congurations. We introduce two distinct systems by preference, because each is chosen as particularly suitable for a particular conguration.) Since we do not yet impose any particular restrictions on the e or the E (except for the requirement that the vectors must not be linearly dependent), we cannot assume that they will be orthogonal or of unit length: we will, therefore, need to use the corresponding contravariant vectors dened by
e e = and E E = ;
1 1
g = e 1 e and G = E 1 E :
We can express the deformation gradient,
F:
F = eF E : @x dX : dx = F dX = @X
1 1
1.4.15
DEFORMATION
dx = e 1 F 1 E dX @x = e 1 dX : @X
Thus, writing dx
=
F dX denes F = e 1 F 1 E
=
@x : @X
We must continue to bear in mind that the rst index of F is associated with a component of F along a base vector in the current conguration (e in this case), while its second index is associated with a component of F along a base vector in the reference conguration (E ). From Equation 1.4.113 we can write
F FT = e F E E
F
e
= e F G F
e ;
1 1
where G is the contravariant metric of the basis system that we have chosen in the reference conguration. The eigenproblem for the squared principal stretch ratios and their directions is solved by nding the eigenvalues of the matrix of numbers F G F . The eigenvectors will appear as the components (nI ) along the e base vectors in the current conguration. Since we have dened the left stretch on the current conguration as
V
and, since
e V e ;
1 1
01 ) =
R = V 01 F 01 ) e
e F E = e (V
01 ) g
F E ; = e (V
1 1
so
R = (V 01)
g
F ;
1.4.16
DEFORMATION
where g is the contravariant metric tensor of the basis system we have chosen to use in the current conguration andas with F we see that the rst index of R is associated with the contravariant base vector e in the current conguration, while the second index is associated with the contravariant base vector E in the reference conguration. We should take care to understand the distinction between the direct matrix expression of the rigid body rotation of the principal directions of strain of the material, R, and the components of R expressed on a particular basis. Suppose, for example, that the rigid body rotation at a point is zero (that is, R = I) but we, nevertheless, have chosen different basis systems e and E . In this case R = e 1 R 1 E = e 1 I 1 E = e 1 E . This implies that, even though R is a unit matrix (in the sense that operating on any vector with this matrix makes no change in that vector), the numerical values we have chosen to store the matrixthe R do not form a unit matrix of numbers unless the e and the E are coincident and orthonormal. Thus, our choice of quite general basis systems that are not the same in the current and reference congurations (introduced as being natural for writing results for shells) somewhat complicates the interpretation of the numbers we store. In the previous few paragraphs we have chosen to explore the expression of the basic results we have derived so far for the kinematics of the total motion in terms of quite general basis systems, e and E . In ABAQUS we wish to express results as simply and directly as possible, and we can do so by choosing particular sets of basis vectors that offer the most convenience for our purposes. First, we take the e (and, by extension, the E , since these are just the e at the beginning of the motion) to be a local, orthonormal system at each point. Although it is not possible to construct a Cartesian system with orthonormal base vectors over a general shell surface, we can always project the general results onto such a system when that system is chosen specically at each point where we need to make the projectiontypically at the integration points of the elements. The choice of which system is used as this local orthonormal basis is made in ABAQUS at two levels: we distinguish continuum (solid) elements from structural (shell and beam) elements, and we distinguish the default choice of directions from the particular choice of directions (orientation) specied by the user. For continuum elements the default E are unit vectors along the axes of the global Cartesian system chosen for the problem. At points where the orientation is dened by the user, the specied E are used. For shells (and membranes) we take E1 and E2 tangent to the shells reference surface and E3 normal to that surface at the point under consideration. By default, E1 is the projection of the global x-axis onto the reference surface or, if the global x-axis is almost normal to that surface at the point, E1 is the projection of the global z -axis onto the surface. If the orientation is dened by the user, E1 and E2 are the projections of the two specied axes onto the reference surface at the point. In all cases E3 is normal to the shells reference surface. For beams E1 is along the beam axis, with E2 and E3 dened from the beam section denition option and beam normals given as part of the nodal coordinate denition. For continuum elements the same schemes are applied by default to dene the basis system in the current conguration. For continuum elements with the orientation specied by the user and in all cases for shells, beams, and membranes, the e are dened by
e = R E :
1
These schemes all have the same property: at any point in time the e are orthonormal vectors: e e = , so e = e and, thus, g = g = g = , andin particularE E = and, = . This simplies the understanding of all the quantities we write, since the components of thus, G
1 1
1.4.17
DEFORMATION
any tensor T ... are always the physical projections of that tensor-valued quantity on the local orthogonal basis system e and we need not distinguish covariant and contravariant components as we did in the general development above. In practical terms the only price we must pay for this simplicity is in shells when we have to use a separate basis system at each point under study, since we cannot construct a single system with the orthonormal property on a general curved surface. (In an axisymmetric system we also have to use dx3 = r d to ensure that the e3 base vector is a unit vector, but this is a minor point.) The simplications are valuable and, from our perspective of studying nite element formulations, they are bought at modest cost, since we generally only consider a single integration point at a time. Throughout the rest of this manual, whenever we need to write down particular components of a tensor, we shall assume that the basis on which they are written has the orthonormal property e 1 e = . The material also undergoes rigid body translation, but this is not important in the development since we need consider only relative motion of neighboring points because we are interested in the deformation of the material to link the kinematics of the motion to the materials constitutive behavior. Numerically, rigid body translation is signicant only for two reasons. One is that the spatial discretization must allow rigid body translation without giving strain, which is important in choosing interpolation functions for the nite elements. The other is that care must be exercised to ensure that the strain and rotation are calculated accurately when the rigid body motion is large, since then the strain and rotation depend on the difference between two very large motions.
Reference
1.4.18
STRAIN MEASURES
1.4.2
STRAIN MEASURES
Strain measures used in general motions are most simply understood by rst considering the concept of strain in one dimension and then generalizing this to arbitrary motions by using the polar decomposition theorem just derived.
Strain in one dimension
We already have a measure of deformationthe stretch ratio . In fact, is itself an adequate measure of strain for a number of problems. To see where it is useful and where not, rst notice that the unstrained value of is 1.0. A typical soft rubber component (such as a rubber band) can change length by a large factor when it is loaded, so the stretch ratio would often have values of 2 or more. In contrast, a typical structural steel component will be designed to respond elastically to its working loads. Such a material has an elastic modulus of about 200 2 103 MPa (30 2 106 lb/in2 ) at room temperature and a yield stress of about 200 MPa (30 2 103 lb/in2 ), so the stretch at yield will be about 1.001 in tension, 0.999 in compression. The stretch ratio is an unsatisfactory way of measuring deformation for this case because the numbers of interest begin in the fourth signicant digit. To avoid this inconvenience, the concept of strain is introduced, the basic idea being that the strain is zero at = 1, when the material is unstrained. In one dimension, along some gauge length d , we dene strain as a function of the stretch ratio, , of that gauge length:
" = f ( ):
The objective of introducing the concept of strain is that the function f is chosen for convenience. To see what this implies, suppose " is expanded in a Taylor series about the unstrained state:
2 1 df 2d f + ( 0 1) + 111 (1.4.21) d 2! d2 We must have f (1) = 0, so " = 0 at = 1 (this was the main reason for introducing this idea of strain instead of just using the stretch ratio). In addition, we choose df=d = 1 at = 1 so that
" = f (1) + ( 0 1)
for small strains we have the usual denition of strain as the change in length per unit length. This ensures that, in one dimension, all strain measures dened in this way will give the same numerical value to the order of the approximation when strains are small (because then the higher-order terms in the Taylor series are all negligible)regardless of the magnitude of any rigid body rotation. Finally, we require that df=d > 0 for all physically reasonable values of (that, is for all > 0) so that strain increases monotonically with stretch; hence, to each value of stretch there corresponds a unique value of strain. (The choice of df=d > 0 is arbitrary: we could equally well choose df=d < 0, implying that the strain is positive in compression when < 1. This alternative choice is often made in geomechanics textbooks because geotechnical problems usually involve compressive stress and strain. The choice is a matter of convenience. In ABAQUS we always use the convention that positive direct strains represent tension when > 1. This choice is retained consistently in ABAQUS, including in the geotechnical options.)
1.4.21
STRAIN MEASURES
With these reasonable restrictions (f = 0 and df =d = 1 at = 1, and df=d > 0 for all > 0), many strain measures are possible, and several are commonly used. Some examples are Nominal strain (Biots strain): f () = 0 1: In a uniformly strained uniaxial specimen, where l is the current and L the original gauge length, this strain is measured as (l=L) 0 1. This denition is the most familiar one to engineers who perform uniaxial testing of stiff specimens. Logarithmic strain: f ()
= ln :
This strain measure is commonly used in metal plasticity. One motivation for this choice in this case is that, when true stress (force per current area) is plotted against log strain, tension, compression and torsion test results coincide closely. Later we will see that this strain measure is mathematically appropriate for such materials because, for these materials, the elastic part of the strain can be assumed to be small. 1 10 2 01 : Greens strain: f () =
2
This strain measure is convenient computationally for problems involving large motions but only small strains, because, as we will show later, its generalization to a strain tensor in any three-dimensional motion can be computed directly from the deformation gradient without requiring solution for the principal stretch ratios and their directions. All of these strains satisfy the basic restrictions. Obviously many strain functions are possible: the choice is strictly a matter of convenience. Since strain is usually the link between the kinematic and the constitutive theories, the convenience of this choice in the context of nite elements is based on two considerations: the ease with which the strain can be computed from the displacements, since the latter are usually the basic variables in the nite element model, and the appropriateness of the strain measure with respect to the particular constitutive model. For example, as mentioned above, it appears that log strain is particularly appropriate to plasticity, while large-strain elasticity analysis (for rubbers and similar materials) can be done quite satisfactorily without ever using any strain measure except the stretch ratio .
Strain in general three-dimensional motions
Having dened the basic concept of strain in one dimension, we now generalize the idea to three dimensions. In Deformation, Section 1.4.1, we established that the deforming part of the motion in the immediate neighborhood of a material point is completely characterized by six variables: the three principal stretch ratios (I , II , and III ) and the orientation of the three principal stretch directions in the current (or in the reference) conguration. This immediately gives the generalization of the one-dimensional strain function introduced above. We rst choose the function f that will be used as the strain measure. "I = f (I ) will be the strain along the rst principal direction, nI ; "II = f (II ) will be the strain along nII ; and "III = f (III ) will be the strain along nIII . The matrix
1.4.22
STRAIN MEASURES
" = " I nI nI T
+ "II
(1.4.22)
completely characterizes the state of strain at the material point. Notice the resemblance to the denition of the stretch matrix, Equation 1.4.110: we might consider " to be dened by the matrix function
" = f ( V) ;
where we understand a matrix function to mean that the two matrices have the same principal directions with their principal values related by the denition of f , which is a convenient shorthand way of indicating a relationship between two matrices. In Equation 1.4.22 we have written the matrix " by using the principal strain directions in the current conguration. We could equally have begun with the polar decomposition into a stretch followed by rotation of the principal directions of stretch: " would be dened in a similar way and would then be associated with its principal directions in the reference conguration. ABAQUS generally reports strains referred to directions in the current conguration. There is no obvious reason for this choice: either approach would sufce so long as the user knows which is being used. The strain measures reported by ABAQUS are enumerated in Conventions, Section 1.2.2 of the ABAQUS Analysis Users Manual. In a nite element code the deformation gradient F is usually computed at each material calculation point from the displacement solution at the nodes of each element and the interpolation function chosen for the element. We now need an algorithm to obtain ", given a choice of strain measure. This algorithm is available immediately from Equation 1.4.112: the eigenvalues and eigenvectors of the 3 2 3 matrix F 1 FT are I 2 ; II 2 and III 2; and nI , nII ; and nIII . We can then calculate "I = f (I ), etc. for the function f chosen as the strain measure and, thus, construct
" = "I nI nI T
+ "II
This algorithm also gives principal strain and stretch valuesoften a useful output because they give a concise description of the state of deformation at a point. However, the algorithm requires computation of the eigenvalues and eigenvectors of a 3 2 3 matrix at each of many points in the model at each of many iterations, which involves some computational cost. Thus, it would be useful if " could be computed less expensively from F, which is possible only for certain choices of the strain measure, f (). We now consider one such possibility. The unit matrix I can be written as
(1.4.23)
STRAIN MEASURES
"G =
that
1 2
(
0 1):
Comparing this one-dimensional denition with Equation 1.4.22 and Equation 1.4.23, we see
"g
1 2
F F T 0 I)
1
is then a generalization of Greens strain in one dimension. (The more standard denition of Greens strain matrix is obtained by using FT 1 F instead of F 1 FT , so the strain matrix is taken on the reference conguration instead of the current conguration as a basis:
"G =
1 2
F T 1 F 0 I ):
The denition we have adopted is consistent with taking the strain matrix on the current conguration. The only difference between the two denitions is the conguration in which the matrix is dened whether we think of the motion as rigid body rotation of the principal axes of stretch, R, followed by stretch along those axes, V, or stretch along the principal axes, U, followed by rigid body rotation of those axes, R. The choice is arbitrary.) Greens strain matrix is, thus, available directly from the deformation gradient without rst having to solve for the principal directions. This advantage makes Greens strain computationally attractive. Recall that strain is the link between the kinematics and the constitutive theory, so the strain choice should be optimal based on the two considerations of convenience and appropriateness. We have already suggested that logarithmic strain is the most appropriate for elastic-plastic or elasticviscoplastic materials in which the elastic strains are always small (because the yield stress is small compared to the elastic modulus), so it appears that the computational convenience of Greens strain cannot be used to advantage. However, the choice of a strain function, f (), was restricted so that, for small strains but arbitrary rotations, all strain measures are the same to the order of the approximation. Thus, for such cases Greens strain is a very convenient choice for computing the strain. The small-strain, large-rotation approximation is often usefulespecially in structural problems (shells and beams) because there the thinness or slenderness of the members often allows large rotations to occur with quite small-strainsand Greens strain is commonly used in large-rotation, small-strain formulations for such problems as shell buckling. Finally, it is worth remarking that the familiar small-strain measure used in most elementary elasticity textbooks, !
"=
@u @u T ; + @X @X
is useful only for small displacement gradientsthat is, both the strains and the rotations must be small for this strain measure to be appropriate. This can be demonstrated by considering pure rotation of a specimen: even though the material is not stretched, the components of this measure of strain become nonzero as the rotation increases.
Reference
1.4.24
RATE OF DEFORMATION
1.4.3
Many of the materials we need to model are path dependent, so usually the constitutive relationships are dened in rate form, which requires a denition of strain rate. The velocity of a material particle is
v= x
@ @t
where the partial differentiation with respect to time (t) means the rate of change of the spatial position, x, of a xed material particle. Here, again, we take the Lagrangian viewpoint: we observe a material particle and follow it through the motion, rather than looking at a xed point in space and watching the material owing through this point. The Lagrangian point of view is used for the mechanical modeling capabilities in ABAQUS because we are usually dealing with history-dependent materials and the Lagrangian perspective makes it easy to record and update the state of a material point since the mesh is glued to the material. The velocity difference between two neighboring particles in the current conguration is
d
v= v x=L x x
@ @ 1d 1d
where
v L= x
@ @
(1.4.31)
is the velocity gradient in the current conguration. In Deformation, Section 1.4.1, we introduced the denition of the deformation gradient matrix, F:
d
x=F X
1d 1d 1
so
d
v=L x=L F X
1d @ @t
v=
_ F X) = F X
1d 1d
where
_ F= F
@ @t
because dv is dened as the velocity difference between two neighboring material particles and, having chosen these particles, the gauge length between them in the reference conguration, dX, is the same throughout the motion and, so, has no time derivative. Comparing the two expressions for dv in terms of the reference conguration gauge length dX, we see that
_ L F=F
1
1.4.31
RATE OF DEFORMATION
or
Now L will be composed of a rate of deformation and a rate of rotation or spin. Since these are rate quantities, the spin can be treated as a vector; thus, we can decompose L into a symmetric strain rate matrix and an antisymmetric rotation rate matrix, just as in small motion theory we decompose the innitesimal displacement gradient into an innitesimal strain and an innitesimal rotation. The symmetric part of the decomposition is the strain rate (it is called the rate of deformation matrix in many textbooks) and is
_ L = F F 01 :
1
_ "=
1 0L + LT 1 2 ! @v T 1 @v = 2 @ x + @x :
= 1 0L 0 LT 1 2 ! @v T @v = 1 @x 0 @x : 2 _ These are particularly simple and familiar forms; for example, " is identical to the elementary denition _ of small strain if we replace the particle velocity, v, with the displacement, u. In one dimension " is
"= _
which identies " as the rate of logarithmic strain,
dv ; dx
" = ln :
This interpretation would also be correct if the principal directions of strain rotate along with the rigid body motion (because the identication can be applied to each principal value of the logarithmic strain matrix). In the general case, when the principal strain directions rotate independent of the material, " is not integrable into a total strain measure. Nevertheless, the identication of " with the rate of logarithmic strain in the particular case of nonrotating principal directions provides a useful interpretation of the logarithmic measure of strain as a natural strain if we think of " , as it is dened above as the symmetric part of the velocity gradient with respect to current spatial position, as a natural measure of strain rate. The typical inelastic constitutive model requires as input a small but nite strain increment ", as well as vector and tensor valued state variables (such as the stress) that are written on the current conguration. In ABAQUS/Explicit and for shell and membrane elements in ABAQUS/Standard, a slightly different algorithm is used to calculate R. For most element types in ABAQUS/Standard we approach this problem by rst using the polar decomposition in the increment to dene the change in the average material rotation over the increment, R, from the total deformation in the increment, F:
1F = 1 V 1 1R:
1.4.32
RATE OF DEFORMATION
All vectors and tensors associated with the material (whose values are available at the beginning of the increment from previous calculations) can now be rotated to the conguration at the end of the increment, solely to account for the rigid body rotation in the increment:
a 1R a
) 1
A 1R A 1R T
) 1 1
for a tensor. These rotated variables are now passed to the constitutive routines, which may provide further updates to them because of constitutive effects. These constitutive effects will be associated with deformation, which must be supplied in the form of the strain increment ". For this we proceed as follows. rotates the deformation basisin the sense that it rotates the principal axes of Since we assume deformation and, thus, provides a measure of average material rotationwe can dene the velocity gradient at any time during the increment, referred to the xed basis at t t, as
1R
1
L = 1R(t + 1t) 1R(t + )T L 1R(t + ) 1R(t + 1t)T ; 0 1t: (1.4.32) _ Then our integration of " is the matrix 1", on the basis at the end of the increment, and dened by Z 1t 1" = sym0L(t + )1 d:
1 1 1
+1
Z
1
1t 0
T t
(1.4.33)
d L = dt (1F) 1F01;
1
we can make use of the polar decomposition of the increment of deformation into a stretch on the axes at 1 ) to write the start of the increment followed by rotation (
1F = 1R 1U
1
d = dt (1R) 1RT + 1R
1
1U dt
d d dt
1
1 U0 1 1 R T
1 1
1R
1
L+L
1
d dt
(1U):
1.4.33
RATE OF DEFORMATION
We now assume that the incremental stretch at any time in the increment written on the basis at the , always has the same principal directions I , II , III , so that beginning of the increment,
and, hence,
d (1 dt
and
1 (1I 0 1)NI NI + (1II 0 1)NII NII + (1III 0 1)NIII NIII U) = 1t 1 1 01 = 1U + 1 + (1I 0 1)( =1t) NI NI 1 + (1II 0 1)( =1t) NII NII + 1 + (1 10 1)( =1t) NIII NIII :
III
1 d (1U) 1 1U01+1U01 1 d (1U) = (1I 0 1)(1=1t) NI NI 2 dt dt 1 + (1I 0 1)( =1t) (1II 0 1)(1=1t) NII NII + (1III 0 1)(1=1t) NIII NIII + 1 + (1II 0 1)( =1t) 1 + (1III 0 1)( =1t)
Z 1t 1 d
and, hence,
dt
1" = ln(1I )nI nI + ln(1II )nII nII + ln(1III )nIII nIII = ln 1V: 1U( + ) = 1U( + 1 ) 0 1
Thus, as long as we assume that the stretch at any time during the increment has the same principal directions as the total increment of stretch (on the xed basis at the start of the increment), the logarithmic denition of incremental strain provides the required integral of the strain rate expressed as the rate of deformation. This assumption amounts to requiring that the components of stretch grow proportionally t p t t , where p is any scalar that we take to grow during the increment: that monotonically from 0 to 1 during t. This assumption might be questionable if the increments are very large, but it is consistent with the levels of approximation used in the integration of the inelastic constitutive models. We, therefore, have a simple method for calculating the strain increment for use in this type of constitutive model without any additional loss of accuracy compared to what we already accept in the constitutive integration itself.
Reference
1.4.34
1.4.4
Many useful materials, such as conventional structural metals, can carry only very small amounts of elastic strain (the elastic modulus is typically two or three orders of magnitude larger than the yield stress). We can take advantage of this behavior to simplify the description of the deformation of such a material. Since the behavior is so common, the assumption that the elastic strains are always small forms the basis of almost all of the inelastic material models provided in ABAQUS. This section discusses the description of the deformation for this case. We begin by assuming that the material has a natural elastic reference state in the sense that, at any time in the deformation, we can imagine isolating the immediate neighborhood of a single point in the material, preventing any further inelastic deformation, removing all external forces from the isolated piece, and allowing the material to unload: the deformation associated with this unloading will then be (Fel )01 , the reverse of the elastic deformation. The deformation between the original reference state and this elastically unloaded state is then the inelastic deformation, Fpl :
F = Fel Fpl ;
1
(1.4.41)
from which we can obtain the velocity gradient with respect to position in the current conguration, _ L = F 1 F01 , as
_ _ L = Fel (Fel )01 + Fel Fpl (Fpl )01 (Fel )01 ;
1 1 1 1
which we write as
analogy with the denition of the total velocity gradient. The motion dened by F consists of rigid body motion and deformation. For the general case there is no advantage in associating rigid body motion with both the inelastic and the elastic deformation: we lose nothing by writing where R is the rigid body rotation of the principal axes of deformation and Vel and Vpl are each symmetric. Thus, we are writing the deforming part of the motion as an inelastic stretch along the principal axes of the total deformation and an elastic stretch along these same axes. For the materials of concern here we now assume that, if we write Vel = I + "el , the principal values of the nominal elastic strain "el are all very small compared to unity: j"el I j << 1. Then
L = Lel + Fel Lpl (Fel )01; _ by dening the elastic and plastic velocity gradients, Lel = Fel (Fel )01
1 1 1
(1.4.42) and
Lpl
pl = _
Fpl )01, by
F = Vel Vpl R;
1 1
Fel = Vel
= (1 + " I ) I
el
= Lel + (I + " el )
Lel + Lpl :
(1.4.43)
_ _ _ where " = sym(L), " = sym(Lel ), and " = sym(Lpl ) are the total, elastic, and inelastic strain rates. Equation 1.4.43 is the classical strain rate decomposition of plasticity theory. We see that it derives from the general decomposition Equation 1.4.41 when we use the symmetric part of the velocity gradient _ with respect to current position (the rate of deformation, ") as the measure of strain rate and when the total elastic strain is always small compared to one. The strain rate decomposition is used in this form in almost all of the inelastic constitutive models in ABAQUS: whenever we refer to an elastic or inelastic _ _ strain rate, we imply " el or "pl .
Reference
1.4.42
1.5.1
Many of the problems to which ABAQUS is applied involve nding an approximate (nite element) solution for the displacements, deformations, stresses, forces, andpossiblyother state variables such as temperature in a solid body that is subjected to some history of loading, where loading implies some series of events to which the bodys response is sought. The exact solution of such a problem requires that both force and moment equilibrium be maintained at all times over any arbitrary volume of the body. The displacement nite element method is based on approximating this equilibrium requirement by replacing it with a weaker requirement, that equilibrium must be maintained in an average sense over a nite number of divisions of the volume of the body. In this section we develop the exact equilibrium statement and write it in the form of the virtual work statement for later reduction to the approximate form of equilibrium used in a nite element model. Let V denote a volume occupied by a part of the body in the current conguration, and let S be the surface bounding this volume. (Again, we should emphasize that we are adopting a Lagrangian viewpoint: the volume being considered is a volume of material in the bodyspecically, V is the volume of space occupied by this material at the current point in time, which is distinct from the Eulerian approach, where we are examining a volume in space and watch material owing through that volume.) Let the surface traction at any point on S be the force t per unit of current area, and let the body force at any point within the volume of material under consideration be f per unit of current volume. Force equilibrium for the volume is then
dS +
dV
= 0:
(1.5.11)
;
(1.5.12)
where
is the unit outward normal to S at the point. Using this denition, Equation 1.5.12 is
n1
dS +
dV
= 0:
n1( )
dS =
Z @
V
@x
( ); dV ;
where ( ) is any continuous functionscalar, vector or tensor. Applying the Gauss theorem to the surface integral in the equilibrium equation gives
n1
dS =
Z @
V
@x
dV :
1.5.11
Since the volume is arbitrary, this equation must apply pointwise in the body, thus providing the differential equation of translational equilibrium:
@
@x
+ f = 0:
(1.5.13)
These are the three familiar differential equations of force equilibrium. In deriving them we have made no approximation with respect to the magnitude of the deformation or rotationthe equations are an exact statement of equilibrium so long as we are precise about our denitions of surface tractions, body forces, stress (Cauchy stress, dened by Equation 1.5.12), volume, and area. Moment equilibrium is most simply written in the general case by taking moments about the origin:
(x
t) dS +
(x 2 f ) dV = 0:
Use of the Gauss theorem with this equation then leads to the result that the true (Cauchy) stress matrix must be symmetric:
= T ;
(1.5.14)
so that at each point there are only six independent components of stress. Conversely, by taking the stress matrix to be symmetric, we automatically satisfy moment equilibrium and, therefore, need only consider translational equilibrium when explicitly writing the equilibrium equations. (The moment equilibrium equation written above assumes that there are no point couples acting on the volume. If there are, the stress matrix does not have the symmetry property of Equation 1.5.14. Continuum mechanics models that allow for such point couples have been developed, but they are not relevant to any of the models provided in ABAQUS.) The basis for the development of a displacement-interpolation nite element model is the introduction of some locally based spatial approximation to parts of the solution. To develop such an approximation, we begin by replacing the three equilibrium equations represented by Equation 1.5.13 by an equivalent weak forma single scalar equation over the entire body, which is obtained by multiplying the pointwise differential equations by an arbitrary, vector-valued test function, dened, with suitable continuity, over the entire volume, and integrating. As the test function is quite arbitrary, the differential equilibrium statement in any particular direction at any particular point can always be recovered by choosing the test function to be nonzero only in that direction at that point. For this case of equilibrium with a general stress matrix, this equivalent weak form is the virtual work principle. The test function can be imagined to be a virtual velocity eld, v, which is completely arbitrary except that it must obey any prescribed kinematic constraints and have sufcient continuity: the dot product of this test function with the equilibrium force eld then represents the virtual work rate. Taking the dot product of Equation 1.5.13 with v results in a single scalar equation at each material point that is then integrated over the entire body to give
Z @
V
@x
1
+f
v dV = 0:
(1.5.15)
1.5.12
@ @
x ( v) =
1 1
@ @
v ; v + : @x
@
so that
Z @
V
@
dV
amp
; (using the Gauss theorem with the rst term) v Z Z amp; = t 1 v dS 0 : @x dV @ S V amp; (using the denition of Cauchy stress with the rst term).
v dV @ @ v x Z ZV @ x = n 1 1 v dS 0 : @ x dV ; S V
=
1
Z @
(
v) 0 :
@
t 1 v dS + f 1 v dV =
V
@
@
v dV : @x
v = L @x
as the virtual velocity gradient in the current conguration. We can decompose the gradient into a symmetric and an antisymmetric part:
L = D +
;
where
= asym( L) = 1 ( L 0 LT ) 2
L = : D + :
:
Since is symmetric,
:
= 1 : L 0 1 : LT = 1 : L 0 1 : L = 0 : 2 2 2 2
1.5.13
tf
: D dV =
;
Z
S
v t dS +
1
Z
V
v f dV :
1
(1.5.16)
t=n
@
@
+ f = 0;
= T ;
1 D = 2
@
v + @v T ; @x @x v
and is compatible with all kinematic constraints. We can show that any two of these three statements (virtual work, equilibrium, and compatibility of the test function ) imply the other: we can thus use the virtual work principle, with a suitable test function, as a statement of equilibrium. The virtual work statement has a simple physical interpretation: the rate of work done by the external forces subjected to any virtual velocity eld is equal to the rate of work done by the equilibrating stresses on the rate of deformation of the same virtual velocity eld. The principle of virtual work is the weak form of the equilibrium equations and is used as the basic equilibrium statement for the nite element formulation that will be introduced in Procedures: overview and basic equations, Section 2.1.1. Its advantage in this regard is that it is a statement of equilibrium cast in the form of an integral over the volume of the body: we can introduce approximations by choosing test functions for the virtual velocity eld that are not entirely arbitrary, but whose variation is restricted to a nite number of nodal values. This approach provides a stronger mathematical basis for studying the approximation than the alternative of direct discretization of the derivative in the differential equation of equilibrium at a point, which is the typical starting point for a nite difference approach to the same problem.
Reference
1.5.14
STRESS MEASURES
1.5.2
STRESS MEASURES
The virtual work statement (Equation 1.5.16) expresses equilibrium in terms of Cauchy (true) stress and the conjugate virtual strain rate, the rate of deformation. (Here conjugate means work conjugate, in the sense that the product of the stress and the strain rate denes work per current volume.) It is natural to think of stress and strain as conjugate quantities, but so far we only have true stress and a wide range of possible strain measures. By dening the concept of conjugacy more precisely, we can dene a stress matrix conjugate to any strain matrix that we might choose to use. This exercise has some value, althoughas we develop the argumentit is worth remembering that the Cauchy (true) stress isfrom the engineers viewpointprobably the only measure of stress of practical interest as an output value from a computer code like ABAQUS, because it is a direct measure of the traction being carried per unit area by any internal surface in the body under study. For this reason ABAQUS always reports the stress as the Cauchy stress. One of the alternative stress denitions developed in this section (Kirchhoff stress) is relevant to the constitutive development in Chapter 4, Mechanical Constitutive Theories. The other (second Piola-Kirchhoff stress) is discussed because it is frequently mentioned in standard texts. It is convenient to think of a solid material as having a natural, elastic, reference state to which it will return upon unloading. For a fully elastic material like rubber, this state will always be the original, unstressed, state. For a material that yields, such as a metal, this reference state will be modied by the inelastic deformation to which the material is subjected. Further, we expect the elasticity of the material to be derivable from a thermodynamic potential written about this reference state so that, for isothermal deformations, there will be a potential function for the elastic strain energy per unit of the natural reference volume. On this basis we formalize the concept of conjugacy by writing the work rate per unit of volume in this elastic reference state as
where " is a particular choice of strain matrix, derived on the basis of the discussion in Strain measures, " Section 1.4.2, and is now the stress matrix that is work conjugate to d". Equation 1.5.21 denes a conjugate stress measure for any chosen strain measure. The internal virtual work rate was expressed in Equation 1.5.16 directly in terms of Cauchy stress, and the virtual velocity gradient. This internal virtual work rate may be rewritten as an integral over the natural reference volume: Z Z
V V0 0 is the Jacobian of the elastic deformation between the natural reference and the current where J = dV=dV volumethe ratio of the materials volume in the current and natural congurations. According to the work conjugacy concept just dened, the stress measure dened by
" dW 0 = : d" ;
(1.5.21)
: D dV
J : D dV 0 ;
J
(1.5.22)
is work conjugate to the strain measure whose rate is the rate of deformation
= sym
@v
@x
1.5.21
STRESS MEASURES
This measure of stress is called Kirchhoff stress. It is useful in the development of constitutive models at large strain because it is the most directly available stress measure when we wish to think of the strain rate measured by the rate of deformation and are considering a material with an elastic reference state. The discussion of strain suggested that Greens strain is convenient for the description of problems involving small strains but rotations that are not small, because Greens strain matrix, " G , can be computed directly from the deformation gradient . We now develop the stress measure work conjugate to Greens strain. From Strain measures, Section 1.4.2, the standard Greens strain matrix was dened with respect to the reference conguration as
"G =
1 2
F T F 0 I) ;
1
_ "G =
(_
_ F T 1 F + FT 1 F ) :
From the discussion of the rate of deformation (Rate of deformation and strain increment, Section 1.4.3) we have _ = 1 so that
F L F
_ "G =
F T 1 0L + L T 1 1 F 2 T = F 1D1F
1
and, thus,
D = F0T 1 "_ G 1 F01; where F0T means the inverse of the transpose of F.
Since the work rate per unit reference volume is
dW 0 = J : D;
it follows that
_ _ dW 0 = J : (F0T 1 "G 1 F01 ) = J (F01 1 1 F0T ) : "G :
(This last manipulation is most readily seen by looking at the equation in component form.) Thus,
_ is the stress that is work conjugate to "G . This stress measure, , is known as the second Piola-Kirchhoff stress tensor. For general motions including large strains is not readily interpreted physically. But for the important case of large rotations and small strains, the second Piola-Kirchhoff stress is readily interpreted. We can perform the polar decomposition as = 1 , where is the rotation of the principal axes of deformation and is the right stretch matrix (the stretch written on the reference conguration). If we write the principal stretches in terms of nominal principal strains,
S = J F01 1 1 F0T S
F RU
I
= 1+
"I ; II
= 1+"
II III = 1 + "III ;
STRESS MEASURES
F = R (I + " );
1
F 0 1 = ( I + " )0 1 R T (I 0 " ) 1 R T ;
1
sincefor the small-strain caseall entries in " are very much smaller than one. In addition,
J
Therefore,
= det
Neglecting terms of order strain compared to unity (since this is the small-strain approximation), we obtain
This result gives a very simple physical interpretation of the second Piola-Kirchhoff stress for small strains but arbitrarily large rotations: the components of S are the rotated axis components of . That is, the components of S are the stress components, associated with directions in the reference conguration. Thus, if we use a rectangular Cartesian basis system, the (1; 1) component of S, S 11 , is the normal component of force per unit area acting on a surface that was normal to the X -axis in the reference conguration, regardless of the current orientation of that surface. For example, consider a beam whose axis was initially parallel to the X -axis. Then, throughout the deformation, S 11 will always be the axial stress in the beam, no matter how much the beam is rotated or bent (provided the strains remain small compared to unity). Thus, for this case we can think of S as a material or corotational stress: the material stress and strain are unique, to the order of the approximation, provided strains remain small. When the small-strain approximation is no longer valid, it is essential to use appropriate measures of stress and strain. From a constitutive viewpoint we have already introduced the basic idea of the approach we will follow: we identify the natural reference for the materials elastic response and use stress and strain measures that provide a conjugate pairing so that the elastic potential can be readily expressed. Since we are often interested in the rate behavior of a material, and also because we prefer to use Cauchy stress as the most natural expression of the stress at a point, it is attractive to consider the usage of the strain measure whose rate is the rate of deformation. (We have identied this in one dimension as the logarithmic strain.) We then use the Kirchhoff stress, = J , with respect to the reference state for the materials elasticity, as the stress measure for our constitutive denitions; it is this stress measure that is used in forming constitutive models in ABAQUS at large strains, as will be seen in Chapter 4, Mechanical Constitutive Theories. The work conjugacy principle implies that, for small strains, all stress measures are indistinguishable, because in this case the strain measures are the same. One interpretation of this is that, if the stress-strain curve for a material is plotted using different stress and strain measures (for example, true stress versus
1.5.23
STRESS MEASURES
log strain, and as nominal stress versus engineering strain) the small-strain approximation is no longer appropriate at strain levels where these two plots differ to any degree considered important to the analysis.
Reference
1.5.24
STRESS RATES
1.5.3
STRESS RATES
Many of the materials we wish to model with ABAQUS are history dependent, and it is common for the constitutive equations to appear in rate form. In Stress measures, Section 1.5.2, it was suggested that an appropriate stress measure for stress-sensitive materials (such as yielding materials) is the Kirchhoff stress. We, therefore, need to dene the rate of Kirchhoff stress for use in the constitutive equations. This denition is not simply the material time rate of Kirchhoff stress, because the Kirchhoff stress components are associated with spatial directions in the current conguration (recall that the Kirchhoff stress is J , where J is the volume change from the reference conguration and is the Cauchy stress, dened by = 1 , where and are vectors in the current conguration). To illustrate the issue, consider a uniaxial tension specimen under constant axial force P , lying along the x-axis at time t1 and rotatedwith the axial force held constantto lie along the y-axis at time t2 (see Figure 1.5.31). Write the stress components on the global (1; 2; 3) rectangular Cartesian basis. At time t1, 11 = P=A, and all other ij = 0, while at time t2 , 22 = P=A, and all other ij = 0. Obviously during t1 ! t2 , d11 6= 0 and d22 6= 0, but equally clearly this rate of change of stress has nothing to do with the constitutive response of the material making up the bar. (A materially based stress, such as the second Piola-Kirchhoff stress, would stay constant during the above rotation, because its components are associated with a material basis.) The problem, then, is that the components of or are associated with current directions in space and, therefore, d and d will be nonzero if there is pure rigid body rotation, even though from a constitutive point of view the material is unchanged. Thus, we must divide the increment of or into two partsone attributable to rigid body motion only and a remainder that is then, presumably, associated with the rate form of the stress-strain law. We can derive a simple result for this purpose for any matrix whose components are associated with spatial directions. At some time t imagine attaching to a material point a set of base vectors, , = 1; 2; 3: These vectors cannot stretch but are dened to spin with the same spin as the material. Recall that the spatial gradient of the material particle velocity at a point, @ =@ , was decomposed into a rate of deformation and a spin,
v x
@v @v D = 1 @x + @x 2
T !
and
e =
1 e : _
1 @v
= 2 @x 0 @v @x
T !
For shell and membrane elements a slightly different approach is used to rotate the base vectors. The differences are signicant only if nite rotation of a material point is accompanied by nite shear. based on the current conguration: we can write it in terms of its Now consider any matrix components in the directions:
T = T e eT :
_ _ _ T = T_ e eT + T e eT + T e eT :
1.5.31
STRESS RATES
y,2
P time t2
x,1
Figure 1.5.31 Rotated specimen. The second and third terms are the rate of T caused by the rigid body spin, so the rst term is that part of T caused by other effects (in the case of stress, the rate associated with the constitutive response), called the corotational, or Jaumann, rate of T. The corotational rate of T for shells and membranes is called the Green-Naghdi rate and differs slightly from the Jaumann rate. We will write this as Tr , so
_ T = Tr +
e _
T e
_T e :
From the denition of e as rigid base vectors that spin with the local rigid body motion, we can rewrite this as
_ T = Tr +
1 e
T e
T e
T = Tr +
T + T
T
1 1
1.5.32
STRESS RATES
We, thus, have the total rate of any matrix associated with spatial directions in the current conguration as the sum of the corotational rate of the matrix and a rate caused purely by the local spin. For example, the rate of change of Kirchhoff stress can be written as
reference volume, in terms of the rate of deformation and past history, so this equation provides a convenient link between that material model and the overall change in true (Cauchy) stress (which is the stress measure dened directly from the equilibrium equations). In Chapter 4, Mechanical Constitutive Theories, where the constitutive models in ABAQUS are discussed, stress rate per reference volume r will mean d (J ), the corotational rate of Kirchhoff stress, which is the stress measure work conjugate dt to the rate of deformation.
Reference
r d (J ) = d (J ) + J
1 + 1
T : dt dt r We are assuming that the constitutive theory will dene d (J ), the corotational stress rate per dt
1.5.33
STATE STORAGE
1.5.4
STATE STORAGE
Many of the constitutive models in ABAQUS require tensors to be stored to dene the state at a material calculation point. Such material state tensors are stored as their components in a local, orthonormal, system at the material calculation point. The orientation of that system with respect to the global X; Y; Z spatial system is stored as a rotation from the global axis system. The purpose of this section is to dene the manner in which such tensors are stored and updated. Three types of local basis are used in ABAQUS for material calculations. For isotropic materials in continuum elements the global, spatial, X; Y; Z system is usedthe material basis is xed in time. For isotropic materials in structural surface elements (shells and membranes) the local system is dened by the standard ABAQUS convention described in Conventions, Section 1.2.2 of the ABAQUS Analysis Users Manual; and for beams and trusses it is dened with the 1-direction along the axis of the member and the 2- and 3-directions in material directions in the cross-section. Thus, with isotropic materials the material basis is always the same as the element basis, although for structural elements the material basis changes with time. For anisotropic materials the material basis must be dened by the user and rotates with the average rigid body spin of the material. In this case the material basis and the element basis are not the same. We refer to this local material basis at time t as eM jt , where the superscript M indicates that the basis i is associated with material calculations and jt means that the basis is taken at time t. In this section Latin subscripts (like the i above) take the range 13, while Greek subscripts will take the range 12. Any tensor associated with the materials state, a, say (such as the stress tensor ), is stored in terms of its components along the material basis:
a = aij eM eM : i j
The increment from time t to time t by the incremental deformation gradient,
This matrix is calculated from the gradient interpolator of the nite element and the coordinates of the t. elements nodes at times t and t The polar decomposition of F is
+1 1
1F = 1V 1R;
1
where
1R is the average rigid body rotation at the material point and 1V is a pure stretch matrix: 1V =
3 X1
I
(here
=1
I n I n I
STATE STORAGE
body rotation of the material. Since material tensors are written in terms of their components in the material basis system, this update is computed as
j j
a t+1t = 1ar + 1R a t 1RT ; where 1ar is the change in a caused by constitutive behavior and 1R is the average incremental rigid
j 1 j 1
(aij ) t+1t = 1ar + 1Rik(akl) t1Rjl: ij It is, therefore, necessary to project 1R onto the material basis systems at the start and end of the increment
For isotropic materials the M have been chosen for geometric convenience only, so the Rij are i quite general. For anisotropic materials the material basis system, M , rotates with the average rigid body rotation i , and so is updated by of the material,
1Rij = eM t+1t 1R eM t: i j
j 1 1 j
1R
e
1
eM t+1t = 1R eM t: i i
j j
However, since in this case the material basis system is not the same as the element basis system, E i (which is chosen for geometric convenience for element calculations), transformations must be done to change basis system. Specically, at the start of the material calculation routines, the strain increments are rotated from the element basis into the material basis:
Likewise, at the end of the material calculation routines, the stress increments are rotated back to the element basis for integration into the discretized approximation to the equilibrium equations:
def =
@ij @"kl
must also be rotated from the material basis to the element basis:
E Dijkl
For a shell or membrane only two-dimensional rotations are requiredfor example, since M 3
E because both are unit vectors along the normal to the surface. 3
Reference
1.5.42
ENERGY BALANCE
1.5.5
ENERGY BALANCE
The conservation of energy implied by the rst law of thermodynamics states that the time rate of change of kinetic energy and internal energy for a xed body of material is equal to the sum of the rate of work done by the surface and body forces. This can be expressed as
1 d ( v 1 v + U )dV dt V 2
Z
=
Z
S
v 1 tdS +
f 1 v dV;
(1.5.51)
where
v
is the current density, is the velocity eld vector, is the internal energy per unit mass, is the surface traction vector, is the body force vector, and is the normal direction vector on boundary S .
U
t f n
Using Gauss theorem and the identity that t = 1 n on the boundary S , the rst term of the right-hand side of Equation 1.5.51 can be rewritten as
Z
S
v 1 tdS =
(1.5.52)
since we also know (see Equilibrium and virtual work, Section 1.5.1) that
@v _ : = " : ; @x
_ where " is the symmetric part of the velocity gradient tensor (see Rate of deformation and strain increment, Section 1.4.3). Substituting Equation 1.5.52 into Equation 1.5.51 yields 1 d ( v 1 v + U )dV dt V 2
Z
=
@ _ 1 + f ) 1 v + : " ]dV: V @x
[(
(1.5.53)
@ 1 +f @x
dv : dt
1.5.51
ENERGY BALANCE
The internal energy, EU , is then dened as
dU dt
EU
U dV
Z t Z
=
_ : "dV
dt:
To make the energy balance (Equation 1.5.51) more convenient to use, we integrate it in time: Z Z t Z 1 _ v 1 vdV + U dV = EW F dt + constant; V 2 V 0
_ where EWF is the rate of work done to the body by external forces and contact friction forces between the contact surfaces, dened as Z Z _ v 1 tdS + f 1 v dV: EWF = S V We can further split the traction, t, into the surface distributed load, tl , and the frictional traction, tf . _ EWF can be written as
Z
_ EWF
v1t
l dS +
f 1 v dV
0 0
1 tf dS
_ = EW
_ 0 EF ;
_ _ where EW is the rate of work done to the body by external forces and EF is the rate of energy dissipated by contact friction forces between the contact surfaces. An energy balance for the entire model can be written as
EU
EK
+ EF
0 EW = constant;
1
(1.5.54)
v 1 vdV: V 2 For convenience, the internal energy is split into two contributions: Z t Z Z t Z 0 c 1 _ _ EU = : " dV dt = + d : "dV dt V V 0 0 Z t Z Z t Z _ _ c : " dV dt + d : "dV dt = V V 0 0 = E I + EV ; EK
=
1.5.52
ENERGY BALANCE
where c is the stress derived from the user-specied constitutive equation and d is the viscous stress (dened for bulk viscosity, material damping, and dashpots), EV is the energy dissipated by viscous effects, and EI is the remaining energy, which we continue to call the internal energy. If we introduce the strain _ _ _ _ _ _ _ decomposition, " = "el + "pl + " cr (where "el , "pl , and " cr are elastic, plastic, and creep strain rates, respectively), the internal energy, EI , can be expressed as
EI
Z t Z
_ c : "el dV = V 0 = ES + EP + EC ;
Z0t ZV
_ c : "dV dt
dt +
Z t Z
0
_ c : " pl dV dt +
Z t Z
0
_ c : "cr dV dt
(1.5.55)
where ES is the recoverable elastic strain energy, EP is the energy dissipated by plasticity, and EC is the energy dissipated by time-dependent deformation (creep, swelling, and viscoelasticity).
References
ABAQUS/Standard output variable identiers, Section 4.2.1 of the ABAQUS Analysis Users Manual ABAQUS/Explicit output variable identiers, Section 4.2.2 of the ABAQUS Analysis Users Manual
1.5.53
Chapter 2 Procedures
Overview Nonlinear solution methods Buckling and postbuckling Nonlinear dynamics Modal dynamics Complex harmonic oscillations Steady-state transport analysis Analysis of porous media Coupled uid-solid analysis Piezoelectric analysis Heat transfer Coupled thermal-electrical analysis Mass diffusion Substructuring Submodeling Fracture mechanics Stress linearization Design sensitivity analysis 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 2.10 2.11 2.12 2.13 2.14 2.15 2.16 2.17 2.18
PROCEDURES OVERVIEW
2.1.1
ABAQUS is designed as a exible tool for nite element modeling. An important aspect of this exibility is the manner in which ABAQUS allows the user to step through the history to be analyzed. This is accomplished by dening analysis procedures. A basic concept in ABAQUS is the division of the problem history into steps. A step is any convenient phase of the historya thermal transient, a creep hold, a dynamic transient, etc. In its simplest form in ABAQUS/Standard, a step is just a static analysis of a load change from one magnitude to another. In each step the user chooses a procedure, thus dening the type of analysis to be performed during the step: dynamic stress analysis, eigenvalue buckling, transient heat transfer analysis, etc. The procedure choice can be changed from step to step in any meaningful way. Since the state of the model (stresses, strains, temperatures, etc.) is updated throughout all analysis steps, the effects of previous history are always included in the response in each new step. Thus, for example, if natural frequency extraction is performed after a geometrically nonlinear static analysis step, the preload stiffness will be included. ABAQUS/Standard provides both linear and nonlinear response options. The program is truly integrated, so linear analysis is always considered as linear perturbation analysis about the state at the time when the linear analysis procedure is introduced. This linear perturbation approach allows general application of linear analysis techniques in cases where the linear response depends on preloading or the nonlinear response history of the model. In nonlinear problems the objective is to obtain a convergent solution at a minimum cost. The nonlinear procedures in ABAQUS/Standard offer two approaches to this. Direct user control of increment size is one choice, whereby the user species the incrementation scheme. Automatic control is the alternate approach: the user denes the step and species certain tolerances or error measures. ABAQUS/Standard then automatically selects the increments as it develops the response in the step. This approach is usually more efcient, because the user cannot predict the response ahead of time. Automatic control is particularly valuable in cases where the time or load increment varies widely through the step, as is often the case in diffusion type problems (such as creep, heat transfer, and consolidation). In ABAQUS/Explicit the time incrementation is controlled by the stability limit of the central difference operator. The time incrementation scheme is, hence, fully automatic and requires no user intervention. User-specied time incrementation is not available because it would always be nonoptimal. ABAQUS/Standard and ABAQUS/Explicit are separate program modules with different data structures; hence, the explicit dynamics procedure cannot be used in the same analysis as any of the procedures in ABAQUS/Standard. However, ABAQUS provides a capability to import a deformed mesh and associated material state from ABAQUS/Explicit into ABAQUS/Standard and vice versa. This procedure is described in Transferring results between ABAQUS/Explicit and ABAQUS/Standard, Section 7.7.2 of the ABAQUS Analysis Users Manual. In this chapter the basic equations for the most important analysis procedures in ABAQUS/Standard and ABAQUS/Explicit are described. In some sections specic aspects of an analysis procedure (i.e., damping, cavity radiation, etc.) are discussed.
Procedures
2.1.11
PROCEDURES OVERVIEW
This section describes the basic equations for standard displacement-based nite element analysis. We begin with the equilibrium statement, written as the virtual work principle, Equation 1.5.16:
: dV
tT v dS + f T v dV :
1
Following the discussion in Equilibrium and virtual work, Section 1.5.1, the left-hand side of this equation (the internal virtual work rate term) is replaced with the integral over the reference volume of the virtual work rate per reference volume dened by any conjugate pairing of stress and strain:
where c and " are any conjugate pairing of material stress and strain measures. The particular choice of " depends on the individual elementsee Chapter 3, Elements. The nite element interpolator can be written in general as
V0
" c : " dV 0 =
tT v dS + f T v dV ;
1
(2.1.11)
where N are interpolation functions that depend on some material coordinate system, uN are nodal variables, and the summation convention is adopted for the uppercase subscripts and superscripts that indicate nodal variables. The virtual eld, , must be compatible with all kinematic constraints. Introducing the above interpolation constrains the displacement to have a certain spatial variation, so must also have the same spatial form:
u = N N uN ;
v = NN vN :
The continuum variational statement Equation 2.1.11 is, thus, approximated by a variation over the nite set vN . " Now " is the virtual rate of material strain associated with , and because it is a rate form, it must be linear in . Hence, the interpolation assumption gives
where N is a matrix that depends, in general, on the current position, , of the material point being considered. The matrix N that denes the strain variation from the variations of the kinematic variables is derivable immediately from the interpolation functions once the particular strain measure to be used is dened. Without loss of generality we can write N = N ( ; N ), andwith this notationthe equilibrium equation is approximated as
x N
Z
vN
N : c dV 0 = vN 0
Z
NT N
t dS +
NT 1 f dV N
2.1.12
PROCEDURES OVERVIEW
since the v N are independent variables, we can choose each one to be nonzero and all others zero in turn, to arrive at a system of nonlinear equilibrium equations:
This system of equations forms the basis for the (standard) assumed displacement nite element analysis procedure and is of the form
V0
N : c dV 0 =
NT t dS + NT f dV : S N V N
1 1
(2.1.12)
F N (uM ) = 0
discussed above. The above equations are valid for static and dynamic analysis if the body force is assumed to contain the inertia contribution. In dynamic analysis, however, the inertia contribution is more commonly considered separately, leading to the equations
Procedures
M NM uM + F N (uM ) = 0:
For the Newton algorithm (or for the linear perturbation procedure) used in ABAQUS/Standard, we need the Jacobian of the nite element equilibrium equations. To develop the Jacobian, we begin by taking the variation of Equation 2.1.11, giving
V0
(d
(2.1.13)
where d( ) represents the linear variation of the quantity ( ) with respect to the basic variables (the degrees of freedom of the nite element model). In the above expression J = jdV=dV 0 j is the volume change between the reference and the current volume occupied by a piece of the structure and, likewise, Ar = jdS=dS 0 j is the surface area ratio between the reference and the current conguration. The Jacobian matrix is obtained by restricting the above variation, allowing variations in the nodal variables, uN , only. Let such a restricted variation be indicated by @N = @=@uN . Examining Equation 2.1.13 term by term with this in mind, we proceed as follows. The rst term contains d c . We now assume that the constitutive theory allows us to write
d c
where and are dened in terms of the current state, direction of straining, etc., and on the kinematic assumptions used to form the generalized strains. See Chapter 4, Mechanical Constitutive and for the material models currently available Theories, for a detailed discussion of forming in ABAQUS. From the choice of generalized strain measure and interpolation function,
" H : d" + g;
@N " =
From the above constitutive assumption,
N:
@N c =
H : N:
2.1.13
PROCEDURES OVERVIEW
" Now, since " is the rst variation of " with respect to nodal variables, " " = @M "uM Z
= M uM :
H : N dV 0;
the usual small-displacement stiffness matrix, except that, since the strain measure " will always be nonlinear in displacement, the N in this term will be a function of displacement. The second term in Equation 2.1.13 is
V
This is rewritten as
" c : d" dV 0 :
Z Z
V0 V0
" c : @N " dV 0 ;
which is
c : @N M dV 0:
This term contributes to the Jacobian and is the initial stress matrix. The external load rate terms in Equation 2.1.13 are considered next. In general, these load vectors can be written
where represents the externally prescribed loading parameters. Whether the load depends on position or not depends on the particular load type, but common types of loading (pressure, centrifugal load) do depend on positionfor example, if is caused by pressure on the surface, depends on the pressure magnitude, on the direction of the normal to the surface, and on the current surface area: the latter two are functions of the current position of points on the surface. The variation of the load vector with nodal variables can then be written symbolically as
@N
1 t + t A r @N A r = Q S ; N 1 f + f J @N J = Q V ; N
@N
and then writing
where M is obtained directly from the interpolation functions, we can write the Jacobian terms pertaining to the last four terms of Equation 2.1.13 as
v = NM vM ;
Z
T N 0 NM 1 QS dS 0 NT 1 QV dV : S V M N
2.1.14
PROCEDURES OVERVIEW
These are commonly called the load stiffness matrix. The actual form of the load stiffness is very much dependent on the type of load being consideredsee Chapter 3, Elements, and Hibbitt (1979) for examples. The complete Jacobian matrix is then
KMN
: H : N dV V0 M
+ 0 c : @N M dV 0 0 NT 1 QS dS 0 NT 1 QV dV : V S M N V M N
(2.1.14)
Thus, Equation 2.1.12 and Equation 2.1.14 provide the basis for the Newton incremental solution, given specication of the interpolation function and constitutive theories to be used.
Reference
Procedures
2.1.15
2.2.1
The nite element models generated in ABAQUS are usually nonlinear and can involve from a few to thousands of variables. In terms of these variables the equilibrium equations obtained by discretizing the virtual work equation can be written symbolically as
u is the value of where F the M th variable. The basic problem is to solve Equation 2.2.11 for the uM throughout the history of interest. Many of the problems to which ABAQUS will be applied are history-dependent, so the solution must be developed by a series of small increments. Two issues arise: how the discrete equilibrium statement Equation 2.2.11 is to be solved at each increment, and how the increment size is chosen. ABAQUS/Standard generally uses Newtons method as a numerical technique for solving the nonlinear equilibrium equations. The motivation for this choice is primarily the convergence rate obtained by using Newtons method compared to the convergence rates exhibited by alternate methods (usually modied Newton or quasi-Newton methods) for the types of nonlinear problems most often studied with ABAQUS. The basic formalism of Newtons method is as follows. Assume that, after an iteration i, an approximation M M ui , to the solution has been obtained. Let ci+1 be the difference between this solution and the exact solution to the discrete equilibrium equation Equation 2.2.11. This means that F
N ( M) = 0
u
(2.2.11)
Procedures
N( M + M ) = 0 i i+1
u c u c
: u
Expanding the left-hand side of this equation in a Taylor series about the approximate solution gives
F
If u c the rst two terms above can be neglected giving a linear system of equations:
K
N M P P ( i ) i+1 2 N M P Q + P Q ( i ) i+1 i+1 + . . . = 0 M is a close approximation to the solution, the magnitude of each M will be small, and so all but i i+1 N( M) + i
u @F @u @ F @u @u u c c :
M i
then
where
K
iNP P+1 = 0 i NP = N ( i P
c @F @u F u
iN
(2.2.12)
M) i
:
iN = N ( M ) i
u c
M = M+ M i+1 i i+1
2.2.11
and the iteration continues. Convergence of Newtons method is best measured by ensuring that all entries in FiN and all entries N are sufciently small. Both these criteria are checked by default in an ABAQUS/Standard solution. in ci+1 ABAQUS/Standard also prints peak values in the force residuals, incremental displacements, and corrections to the incremental displacements at each iteration so that the user can check for these contingencies himself. Newtons method is usually avoided in large nite element codes, apparently for two reasons. First, the complete Jacobian matrix is sometimes difcult to formulate; and for some problems it can be impossible to obtain this matrix in closed form, so it must be calculated numericallyan expensive (and not always reliable) process. Secondly, the method is expensive per iteration, because the Jacobian must be formed and solved at each iteration. The most commonly used alternative to Newton is the modied Newton method, in which the Jacobian in Equation 2.2.12 is recalculated only occasionally (or not at all, as in the initial strain method of simple contained plasticity problems). This method is attractive for mildly nonlinear problems involving softening behavior (such as contained plasticity with monotonic straining) but is not suitable for severely nonlinear cases. (In some cases ABAQUS/Standard uses an approximate Newton method if it is either not able to compute the exact Jacobian matrix or if an approximation would result in a quicker total solution time. For example, several of the models in ABAQUS/Standard result in a nonsymmetric Jacobian matrix, but the user is allowed to choose a symmetric approximation to the Jacobian on the grounds that the resulting modied Newton method converges quite well and that the extra cost of solving the full nonsymmetric system does not justify the savings in iteration achieved by the quadratic convergence of the full Newton method. In other cases the user is allowed to drop intereld coupling terms in coupled procedures for similar reasons.) Another alternative is the quasi-Newton method, in which Equation 2.2.12 is symbolically rewritten
ci+1
= 0[KiNP ]01FiN
and the inverse Jacobian is obtained by an iteration process. There are a wide range of quasi-Newton methods. The more appropriate methods for structural applications appear to be reasonably well behaved in all but the most extremely nonlinear casesthe tradeoff is that more iterations are required to converge, compared to Newton. While the savings in forming and solving the Jacobian might seem large, the savings might be offset by the additional arithmetic involved in the residual evaluations (that is, in calculating the Fi ), and in the cascading vector transformations associated with the quasi-Newton iterations. Thus, for some practical cases quasi-Newton methods are more economic than full Newton, but in other cases they are more expensive. ABAQUS/Standard offers the BFGS quasi-Newton method: it is described in Quasi-Newton solution technique, Section 2.2.2. When any iterative algorithm is applied to a history-dependent problem, the intermediate, nonconverged solutions obtained during the iteration process are usually not on the actual solution path; thus, the integration of history-dependent variables must be performed completely over the increment at each iteration and not obtained as the sum of integrations associated with each Newton iteration, ci . In ABAQUS/Standard this is done by assuming that the basic nodal variables, u, vary linearly over the increment, so that
u ( ) = 01 0 1t 1u(t) + 1t u(t + 1t);
2.2.12
where we compute
1t represents time during the increment. Then, for any history-dependent variable, g(t),
g t
( + 1 t) = g (t) +
Z t+1t
t
dg d
( ) d
at each iteration. The issue of choosing suitable time steps is a difcult problem to resolve. First of all, the considerations are quite different in static, dynamic, or diffusion cases. It is always necessary to model the response as a function of time to some acceptable level of accuracy. In the case of dynamic or diffusion problems time is a physical dimension for the problem and the time stepping scheme must provide suitable steps to allow accurate modeling in this dimension. Even if the problem is linear, this accuracy requirement imposes restrictions on the choice of time step. In contrast, most static problems have no imposed time scale, and the only criterion involved in time step choice is accuracy in modeling nonlinear effects. In dynamic and diffusion problems it is exceptional to encounter discontinuities in the time history, because inertia or viscous effects provide smoothing in the solution. (One of the exceptions is impact. The technique used in ABAQUS/Standard for this is discussed in Intermittent contact/impact, Section 2.4.2.) However, in static cases sharp discontinuities (such as bifurcations caused by buckling) are common. Softening systems, or unconstrained systems, require special consideration in static cases but are handled naturally in dynamic or diffusion cases. Thus, the considerations upon which time step choice is made are quite different for the three different problem classes. ABAQUS provides both automatic time step choice and direct user control for all classes of problems. Direct user control can be useful in cases where the problem behavior is well understood (as might occur when the user is carrying out a series of parameter studies) or in cases where the automatic algorithms do not handle the problem well. However, the automatic schemes in ABAQUS are based on extensive experience with a wide range of problems and, therefore, generally provide a reliable approach. For static problems a number of schemes have been suggested for automatic step control (see, for example, Bergan et al., 1978). ABAQUS/Standard uses a scheme based predominantly on the maximum force residuals following each iteration. By comparing consecutive values of these quantities, ABAQUS/Standard determines whether convergence is likely in a reasonable number of iterations. If convergence is deemed unlikely, ABAQUS/Standard adjusts the load increment; if convergence is deemed likely, ABAQUS/Standard continues with the iteration process. In this way excessive iteration is eliminated in cases where convergence is unlikely, and an increment that appears to be converging is not aborted because it needed a few more iterations. One other ingredient in this algorithm is that a minimum increment size is specied, which prevents excessive computation in cases where buckling, limit load, or some modeling error causes the solution to stall. This control is handled internally, with user override if needed. Several other controls are built into the algorithm; for example, it will cut back the increment size if an element inverts due to excessively large geometry changes. These detailed controls are based on empirical testing. In dynamic analysis when implicit integration is used, the automatic time stepping is based on the concept of half-step residuals (Hibbitt and Karlsson, 1979). The basic idea is that the time stepping operator t in terms of displacement at the end denes the velocities and accelerations at the end of the step t t ; which of the step and conditions at the beginning of the step. Equilibrium is then established at t ensures an equilibrium solution at the end of each time step and, thus, at the beginning and end of any
Procedures
( +1 )
( +1 )
2.2.13
individual time step. However, these equilibrium solutions do not guarantee equilibrium throughout the step. The time step control is based on measuring the equilibrium error (the force residuals) at some point t , to during the time step, by using the integration operator, together with the solution obtained at t t= . If the maximum interpolate within the time step. The evaluation is performed at the half step t entry in this residual vectorthe maximum half-step residualis greater than a user-specied tolerance, the time step is considered to be too big and is reduced by an appropriate factor. If the maximum half-step residual is sufciently below the user-specied tolerance, the time step can be increased by an appropriate factor for the next increment. Otherwise, the time step is deemed adequate. The algorithm is somewhat more complicated at traumatic events such as impact. Here, the time step can also be adjusted based on the magnitude of residuals in the rst or second iteration following such events. Clearly, if these residuals are several orders of magnitude greater than those permitted, convergence is unlikely and the time step is altered immediately to avoid unproductive iteration. These algorithms are discussed in more detail in Intermittent contact/impact, Section 2.4.2, as well as in the ABAQUS Analysis Users Manual. They are products of experience and many numerical experiments and have been shown to be effective in several problem areas of interest.
( + 1 2)
( +1 )
References
Convergence criteria for nonlinear problems, Section 8.3.3 of the ABAQUS Analysis Users Manual Time integration accuracy in transient problems, Section 8.3.4 of the ABAQUS Analysis Users Manual
2.2.14
2.2.2
A major contribution to the computational effort involved in nonlinear analysis is the solution of the nonlinear equations (Equation 2.2.11). In most cases ABAQUS/Standard uses Newtons method to solve these equations, as described in Nonlinear solution methods in ABAQUS/Standard, Section 2.2.1. The principal advantage of Newtons method is its quadratic convergence rate when the approximation at NM provide an iteration i is within the radius of convergencethat is, when the gradients dened by Ki improvement to the solution. The method has two major disadvantages: the Jacobian matrix has to be calculated, and this same matrix has to be solved. The calculation of the Jacobian matrix is a problem because, in many important cases, it is difcult to derive the form of the matrix algebraically. The solution of the Jacobian is a problem because of the computational effort involved: as the problem size increases, the direct solution of the linear equations can dominate the entire computational effort. There are a number of important nonlinear applications where the Jacobian is symmetric, is fairly well conditioned, and does not change greatly from one iteration to the next. Examples are implicit dynamic time integration with small time increments relative to the periods of the natural vibrations that participate in the response or small-displacement elastic-plastic analysis where the yielding is conned (such as occurs in many practical fracture mechanics applications). In such cases, especially when the problem is large, it can be less expensive to use an alternative to the Newton approach to the solution of the nonlinear equations. The quasi-Newton methods are such an approach; and Matthies and Strang (1979) have shown that, for systems of equations with a symmetric Jacobian matrix, the BFGS (Broyden, Fletcher, Goldfarb, Shanno) method can be written in a simple form that is especially effective on the computer and is successful in such applications. This method is implemented in ABAQUS/Standard and is described in this section. The user must select this method explicitly: by default, ABAQUS/Standard uses the standard Newton method. The basis of quasi-Newton methods is to obtain a series of improved approximations to the Jacobian e NM matrix, Ki , that satisfy the secant condition:
e NM approaches K NM as the iterations proceed. Equation 2.2.21 is the basic quasi-Newton so that Ki i equation. For convenience we dene the change in the residual from one iteration to the next as e NM uM F N (u M ) 0 F N (u M 1 ) = K i i i0 i 0 1 0 uM 1 ; i0
Procedures
(2.2.21)
N
i
so that Equation 2.2.21 can be written
FiN
0 FiN 1; 0
(2.2.22)
where c0M is the correction to the solution from the previous iteration, dened in Nonlinear solution i methods in ABAQUS/Standard, Section 2.2.1. Matthies and Strangs implementation of the BFGS method is a computationally inexpensive method h i
e NM of creating a series of approximations to Ki
N i
e =K
iNM cM ; i
01
2.2.21
eN e NM and positive deniteness of Ki M . They accomplish this by updating Ki01 product plus increment form: h e MN Ki i 01
e NM to Ki
using a
L I N L 0 i c N i i
3h
e PL K i0 1
i 01 2
IP M
101
0 i
iP cM + i cN cM ; i i i
:
h i 01
(2.2.23)
where
i
h i01
M cM i i
e MN e MN , is used (as the decomposition of KI ), and the update is accomplished a kernel matrix, KI by premultiplication of the kernel matrix by the terms 2
L I NL 0 j cN
j j
IP M
0 j jP cM j
for j = I + 1; I + 2; . . . i. Because of the form of these terms, the premultiplication and postmultiplication operations result in inner products of vectors and the scaling of vectors by constants: it is this organization that makes the method computationally attractive. However, too many such products (i0I being bigger than, say, 510) are not attractive, so usually a new kernel matrix is formed and stored after some iterations. In the ABAQUS/Standard implementation the kernel is the actual Jacobian matrix @F N =@uM . It is formed whenever a specied number of iterations have been done without obtaining a convergent solution; the default number of iterations is 8. ABAQUS/Standard does not reform the kernel unless this value is exceeded, so the same kernel can be used for several increments if the BFGS updates are successful. In general, the rate of convergence of the quasi-Newton method is slower than the quadratic rate of convergence of Newtons method, though faster than the linear rate of convergence of the modied Newton method.
Reference
Convergence criteria for nonlinear problems, Section 8.3.3 of the ABAQUS Analysis Users Manual
2.2.22
2.2.3
The classical approach in ABAQUS/Standard to obtain the stabilized response of an elastic-plastic structure subjected to cyclic loading is to apply the periodic loading cycles repetitively to the unstressed structure until a stabilized state is obtained. At each instant in time it typically involves using Newtons method to solve the nonlinear equilibrium equations
Procedures
where K is the elastic stiffness matrix and i stands for the iteration number. Because the elastic stiffness serves as the Jacobian matrix throughout the analysis, the equation system is solved only once. Therefore, the direct cyclic algorithm is likely to be less expensive to use than the full Newton approach to the solution of the nonlinear equations, especially when the problem is large. We also expand the residual vector in a truncated Fourier series in the same form as the displacement solution: n [Rs sin k!t + Rc cos k!t] ; R (t) = R 0 + k k k=1 where each residual vector coefcient R0; Rs , and Rc in the Fourier series corresponds to a displacement k k coefcient. The conversion of R(t) into Fourier terms is done incrementally on an element-by-element basis: 2 T R0 = R(t)dt;
2.2.31
Rs k
=T
T
2
ZT Z T
0 0
Rc = k c0 ; cs ; and cc . k k
At the end of each loading cycle, we solve for the corrections to the displacement Fourier coefcients The next displacement coefcients are then (i+1) (i ) (i+1)
uc(i+1) = uc(i) + cc(i+1) ; k k k s(i+1) s (i ) s(i+1) uk = u k + ck : The updated displacement coefcients are used in the next iteration to obtain displacements at each instant in time. This process is repeated until convergence is obtained. Each pass through the complete load cycle can therefore be thought of as a single iteration of the solution to the nonlinear problem. Convergence of the direct cyclic method is best measured by ensuring that all the entries in Rk and ck are sufciently small. By default, both these criteria are checked in an ABAQUS/Standard solution. There are two accuracy aspects to this algorithm: the number of Fourier terms and the number of iterations to obtain convergence. The number of Fourier terms needed to obtain a solution depends on the time variation of the cyclic load as well as the variation of the structure response. In determining the number of terms, keep in mind that the objective of this kind of analysis is to make low-cycle fatigue life predictions. Hence, the goal is to obtain a good approximation of the plastic strain cycle at each point; local inaccuracies in the stress are less important. More Fourier terms usually provide a more accurate solution but at the expense of additional data storage and computational time. ABAQUS/Standard uses an adaptive algorithm to determine the number of Fourier terms during the analysis. Both automatic time incrementation and direct user control of the time incrementation can be used in the direct cyclic method. Since the direct cyclic algorithm uses the modied Newton method, in which a constant elastic stiffness matrix serves as the Jacobian throughout the analysis, interface nonlinearities such as contact and friction are not taken into account. These nonlinearities are severe and would probably lead to convergence difculties if they were included in the direct cyclic algorithm. By default, the periodicity condition, in which the solution of an iteration starts with the solution at the end of the previous iteration, is always imposed from the beginning of an analysis. However, in cases where the periodic solution is not easily found (for example, when the loading is close to causing ratchetting), the state around which the periodic solution is obtained may show considerably more drift than would be obtained in a transient analysis. In such cases the user may wish to delay the application of the periodicity condition as an articial method to reduce this drift. ABAQUS/Standard allows the user to choose when to impose the periodicity condition. By delaying the application of the periodicity condition, the user can inuence the mean stress and strain level, without affecting the shape of the stress-strain curves or the amount of energy dissipated during the cycle. Therefore, this is rarely necessary since the average stress and strain levels are usually not needed for low-cycle fatigue life predictions.
Reference
u0
u0
+ 0
Direct cyclic analysis, Section 6.2.6 of the ABAQUS Analysis Users Manual
2.2.32
2.3.1
ABAQUS/Standard contains a capability for estimating elastic buckling by eigenvalue extraction. This estimation is typically useful for stiff structures, where the prebuckling response is almost linear. The buckling load estimate is obtained as a multiplier of the pattern of perturbation loads, which are added to a set of base state loads. The base state of the structure may have resulted from any type of response history, including nonlinear effects. It represents the initial state to which the perturbation loads are added. The response to the perturbation loads must be elastic up to the estimated buckling load values for the eigenvalue estimates to be reasonable. The following physical problem is addressed in eigenvalue buckling analysis: from an arbitrarily achieved base conguration with stresses B in equilibrium with surface traction tB and body forces qB , we consider an elastic deformation with small displacement gradients under additional surface tractions t, body forces b, and boundary displacements u, where the additional tractions and displacements are applied on mutually complementary parts of the boundary. Such a deformation is a linear perturbation on a predeformed state. A consistent application of the small-displacement gradient assumption to the kinematics and the constitutive equation from an initially stressed state leads to the solution of a linear problem as the response to the additional loading. Since the problem is linear, if is the stress response to the loads t, q, and u, then for loads t, q, and u the stress response will be . Each distinct value of corresponds to a linear perturbation of the base state. Among these perturbed states we seek special values of that allow for the existence of nontrivial incremental displacement elds with arbitrary magnitudes as valid solutions to the problem. Such nontrivial incremental displacement elds are referred to as buckling modes. In the buckling analysis procedure in ABAQUS we do not distinguish between the geometry of the base state and the linearly perturbed congurations. As a result of this assumption we can seek the buckling modes as incremental displacements out of the base state geometry with stresses B , applied tractions tB t, and applied body forces qB q. The equations of equilibrium for an arbitrarily chosen conguration during buckling, referred to as the current conguration, are written in terms of the nominal stress P in the base state. If X represents the position of a material point in the base state, the equilibrium equations can be expressed as
Procedures
1 1
1 1
+ 1
+ 1
Z
+ 1
VB
P:
dV B @v @X
SB
p 1 v dS B
VB
b v dV B ;
1
where v is an arbitrary virtual velocity eld, p is the nominal traction on the boundary S B of the body in the base state, b represents the body force per unit volume in the base state, and V B is the volume that the body occupies in the base state. The corresponding rate form is given by
VB
_ @v P : @ X dV B =
SB
_ p v dS B +
1
VB
_ b v dV B :
1
(2.3.11)
Since we have assumed that the base state and the current state are indistinguishable, we now proceed to express the left-hand side in terms of the rate of Kirchhoff stress , the velocity gradient L, the virtual 1 F0T , where is the velocity gradient L, and the deformation gradient F. Using the relations P
2.3.11
Kirchhoff stress based on the base state as the reference conguration, and F takes the form Z Z
VB
[_ : L 0 ( 1 LT ) : L] dV B : _ We now use the relation between the rate of Kirchhoff stress , the material spin ! = 1 (L 0 LT ), and the 2 _ P: =
VB
Z
@v dV B @X
Z
_ =L
F, Equation 2.3.11
VB
_ P:
@v dV B @X
= =
VB
[ r : D + : (LT 1 L 0 2D 1 D)] dV B :
In addition, we can replace the Kirchhoff stress with the Cauchy stress since it is assumed that the current and reference congurations are indistinguishable. For the right-hand side of Equation 2.3.11 we note that the nominal tractions p and body forces b are given by p t dS=dS B and b q dV =dV B , where dS and dV are the elements of surface area and volume in the current conguration. For any material point the changes in t and q during buckling are completely characterized by the change of the deformation gradient at that point; loosely speaking, the magnitude of the applied forces at any material point is kept xed, and the change in the applied tractions and body force intensities arises due to the change in geometry. For example, for a pressure load the magnitude of the pressure remains constant but the surface normal changesa change that is completely characterized by the change in the deformation gradient. Since the ratios of the surface area and volume measures between the reference and current congurations can be viewed as functions of the deformation gradient F only, it follows that p and b at any given material point also change only through their dependence on the deformation gradient; hence, their rates of change can be written as
_ p=
p= _
@p _ :F @F @p :L @F
and and
_ b= _ b=
@b _ : F; @F @b : L: @F
r = C( ) : D;
where C is a fourth-order tensor that can depend on the current stress, the governing equation for the buckling analysis becomes
()
(2.3.12)
where B and are the nominal tractions generated during buckling corresponding to the base state , respectively; similar denitions apply for the nominal tractions B and the linear perturbation tractions body force terms. The constitutive relation can represent elasticity, hypoelasticity, and hyperelasticity; rate effects and plasticity are ignored. The effective moduli are evaluated for the value of the stress and deformation in the base state. To derive the nite element discretization for the expression above, we introduce the interpolated velocity eld
1p
1t
where represents the position in the base state. Using the standard nite element approach, the governing equations for buckling then take the form of the standard eigenvalue problem:
v = v N N N (X );
Procedures
NM is the base state stiffness and K NM is the differential stiffness. The base state stiffness is where K0 1 the sum of the hypoelastic tangent stiffness, the initial stress stiffness, and the load stiffness:
K NM
Z
respect to the nodal displacements. Since we do not distinguish between the current conguration and the reference conguration, the partial derivatives appearing in the load stiffness terms are all evaluated at uM corresponding to . For example, the load stiffness term for the surface tractions appearing in Equation 2.3.12,
M @ NN B ) : @N : C ( dV B + 0 = B @x @ x sym V sym " N T M N # Z @N @ NM N B : @N dV B 0 0 2 @x 1 1 @@ x @x @x VB sym sym Z Z B @ bB N 1 @ p dS B 0 N @uM NN 1 @uM dV B ; SB VB where @ pB =@uM and @ bB =@uM are the derivatives of the nominal surface tractions and body forces with
=0
F=I
v1
@ pB @F
:L
NN
with
@ pB
: @F
@ NM vM @X
@ pB = NN 1 @uM
vM ;
The differential stiffness consists of the sum of the initial stress stiffness due to the perturbation stresses and the load stiffness due to the perturbation loads:
N F = I + @@X
M 1 @N @x
uK :
"
T
The contribution in this expression that is derived from the stress is symmetric; however, the contribution derived from the applied loads (the load stiffness) is symmetric only if the applied loading is conservative that is, if the loads can be derived from an energy potential. If the load stiffness is nonsymmetric, the contribution will be symmetrized since ABAQUS can solve eigenvalue problems only with symmetric matrices. If the generalized nodal loads resulting from both applied forces tB and qB as well as prescribed displacements uB are denoted by P N and those due to t, q, and u are denoted by QN , the eigenvalues i represent the multipliers that provide the estimated generalized buckling load as P N i QN , while the N corresponding eigenvectors vi give the associated buckling modes. Although in most analyses the lowest mode is the only one of interest, ABAQUS is able to extract several modes simultaneously. It is also worth noting that the common case of an antisymmetric buckling mode on a symmetric base state and buckling load is easily done with ABAQUS. NM K NM (that is, the structure is not stiff in If the tangent stiffness is predicted poorly by K0 1 the sense that the response is nonlinear prior to buckling), a nonlinear analysis using the Riks method is required to obtain a reliable estimate for the load carrying capacity of the structure.
1 1
Reference
Eigenvalue buckling prediction, Section 6.2.3 of the ABAQUS Analysis Users Manual
2.3.14
2.3.2
It is often necessary to obtain nonlinear static equilibrium solutions for unstable problems, where the loaddisplacement response can exhibit the type of behavior sketched in Figure 2.3.21that is, during periods of the response, the load and/or the displacement may decrease as the solution evolves. The modied Riks method is an algorithm that allows effective solution of such cases.
Procedures
displacement minimum
Figure 2.3.21 Typical unstable static response. It is assumed that the loading is proportionalthat is, that all load magnitudes vary with a single scalar parameter. Also, we assume that the response is reasonably smooththat sudden bifurcations do not occur. Several methods have been proposed and applied to such problems. Of these, the most successful seems to be the modied Riks methodsee, for example, Criseld (1981), Ramm (1981), and Powell and Simons (1981)and a version of this method has been implemented in ABAQUS. The essence of the method is that the solution is viewed as the discovery of a single equilibrium path in a space dened by the nodal variables and the loading parameter. Development of the solution requires that we traverse this path as far as required. The basic algorithm remains the Newton method; therefore, at any time there
2.3.21
will be a nite radius of convergence. Further, many of the materials (and possibly loadings) of interest will have path-dependent response. For these reasons, it is essential to limit the increment size. In the modied Riks algorithm, as it is implemented in ABAQUS, the increment size is limited by moving a given distance (determined by the standard, convergence rate-dependent, automatic incrementation algorithm for static case in ABAQUS/Standard) along the tangent line to the current solution point and then searching for equilibrium in the plane that passes through the point thus obtained and that is orthogonal to the same tangent line. Here the geometry referred to is the space of displacements, rotations, and the load parameter mentioned above.
Basic variable denitions
Let P N N ; ; the degrees of freedom of the model) be the loading pattern, as dened with one or more of the loading options in ABAQUS. Let be the load magnitude parameter, so at any time the actual load state is P N , and let uN be the displacements at that time. The solution space is scaled to make the dimensions approximately the same magnitude on each axis. In ABAQUS this is done by measuring the maximum absolute value of all displacement variables, 1 u, in the initial (linear) iteration. We also dene P P N P N 2 . The scaled space is then spanned by
( = 1 2 ... =
=(
and the solution path is then the continuous set of equilibrium points described by the vector uN in this scaled space. All components of this vector will be of order unity. The algorithm is shown in Figure 2.3.22 and is described below. Suppose the solution has been developed to the point Ao uN o . The tangent stiffness, o NM , is formed, and we solve Ko
load P N ; P N P N =P , uN =u displacements uN
= ~
~ = =~ =(
);
(~ ; )
The increment size Ao to A1 in Figure 2.3.22) is chosen from a specied path length, solution space, so that
NM M K o vo
= PN:
= (~ ; )
1l, in the
1 o = 0 N 6 1 l 1 1 vo vo + 1 2 ~ ~N (here vo is vo scaled by u). The value 1l is initially suggested by the user and is adjusted by ~N N the ABAQUS/Standard automatic load incrementation algorithm for static problems, based on the convergence rate. The sign of 1o the direction of response along the tangent lineis chosen so vN uN that the dot product of 1o (~o ; 1) on the solution to the previous increment, (1~01 ; 101), is positive: 1o(~o ; 1) : (1~N1; 101) > 0; vN u0
and, hence, that is
~ (vN ;1) 1
1
~N (v ;1) 0
Procedures
Equilibrium surface
~N u
Figure 2.3.22 Modied Riks algorithm. It is possible that in some cases, where the response shows very high curvature in the uN space, this criterion will cause the wrong sign to be chosensee, for example, Figure 2.3.23.
~ (uN ; -1) -1
(~ ; )
A0
~ u
This direction will be chosen and will take the solution the wrong way along the equilibrium path.
1 .
2.3.23
The wrong sign is rarely chosen in practical cases, unless the increment size is too large or the solution bifurcates sharply. To check for such cases is computationally expensive: one approach would be for the solution to be found at o 0 " 01; < " << , so that we obtain a vector that gives a close approximation of the directed tangent at Ao . Because the case is so rare, such a check is not included, and the simple dot product given above is used alone to determine the sign of o . Thus, we have N o vo o o in Figure 2.3.22. The solution is now corrected now found the point A1 uN o N , by the following onto the equilibrium path in the plane passing through A1 and orthogonal to vo iterative algorithm.
(~ + 1 ~ ; + 1 )
1 (~ ; 1)
N 1 = 1 1 = 1 0 vo = ( = 1 2 3 etc:):
b. Check equilibrium:
N : dV ; and K NM
@I = @uM ;
If all the entries in RN are sufciently small, the increment has converged. If not, we i proceed. c. Solve: NM 8 M M 9 8 N N 9 That is, we solve simultaneously with two load vectors, P N and RN , and obtain two N displacement vectors, vi and cN . i
= ( o + 1 i )P N 0 I N :
vi ; ci
; Ri
N , and add it to cN i where i RN P N =P is the projection d. Now scale the vector vi i i of the scaled residuals onto P N so that we move from Ai to Ai+1 in the plane orthogonal N see Figure 2.3.22. This gives the equation to vo
(~ ; 1)
(~ ; 1)
(~ ; )
(uN + 1uN + cN + viN ; o + 1i + ): o i i 1uN+1 = 1uN + cN + viN i i i 1i+1 = 1i + i = i + 1;
2.3.24
and return to (a) above for the next iteration. The implementation in ABAQUS/Standard includes the additional update after each iteration:
v
N o
= iN
v
This causes the equilibrium search to be orthogonal to the last tangent, rather than to the tangent at the beginning of the increment. The main motivation for this additional modication comes from the use of the method in plasticity problems, where the rst iteration of each increment uses the elastic material stiffness to establish the direction of straining and so provides a stiffness that is not representative of the tangent to the equilibrium path if active plasticity is occurring. The total path length traversed is determined by the load magnitudes supplied by the user on the loading options; while the number of increments is determined by the user-specied time increment data, assisted by ABAQUS/Standards automatic incrementation scheme if that is chosen.
Reference
Procedures
Unstable collapse and postbuckling analysis, Section 6.2.4 of the ABAQUS Analysis Users Manual
2.3.25
2.4.1
ABAQUS offers dynamic analysis options for both linear and nonlinear problems. In the case of purely linear systems, methods based on the eigenmodes of the system are almost always chosen because they can provide insight into the structures behavior that is not otherwise available and because they are usually signicantly more cost-effective than the direct integration methods that are usually used for nonlinear problems. The linear dynamic analysis methods provided in ABAQUS/Standard are discussed in Modal dynamics, Section 2.5. For mildly nonlinear dynamic analysis problems the modal projection method is provided. The basis of that method is to use the eigenmodes of the linear system (extracted from an eigenfrequency analysis) as a set of global Ritz functionsa set of global interpolation functions, in the terminology of the nite element methodwhose amplitudes dene the response. ABAQUS/Standard provides direct time integration using the explicit, central difference operator for this option. For any more severely nonlinear case the dynamic response is obtained by direct time integration of all of the degrees of freedom of the nite element model. The methods provided for this type of analysis are discussed in this section. The choice of operator used to integrate the equations of motion in a dynamic analysis is inuenced by many factors. ABAQUS/Standard is designed to analyze structural components, by which we mean that the overall dynamic response of a structure is sought, in contrast to wave propagation solutions associated with relatively local response in continua. Belytschko (1976) labels these inertial problems and classies them by stating that wave effects such as focusing, reection, and diffraction are not important. Structural problems are considered inertial because the response time sought is long compared to the time required for waves to traverse the structure. Dynamic integration operators are broadly characterized as implicit or explicit. Explicit schemes, as t based entirely on available values used in ABAQUS/Explicit, obtain values for dynamic quantities at t at time t. The central difference operator, which is the most commonly used explicit operator for stress analysis applications, is only conditionally stable, the stability limit being approximately equal to the time for an elastic wave to cross the smallest element dimension in the model. Implicit schemes remove this t based not only on values upper bound on time step size by solving for dynamic quantities at time t t. But because they are implicit, nonlinear equations must at t, but also on these same quantities at t be solved. In structural problems implicit integration schemes usually give acceptable solutions with time steps typically one or two orders of magnitude larger than the stability limit of simple explicit schemes, but the response prediction will deteriorate as the time step size, t, increases relative to the period, T , of typical modes of response. See, for example, Hilber, Hughes and Taylor (1978) for a discussion of such errors. Three factors should be considered when selecting the maximum allowable time step size: the rate of variation of the applied loading, the complexity of the nonlinear damping and stiffness properties, and the typical period of vibration of the structure. In general, a maximum increment versus period ratio t=T < 1/10 is a good rule of thumb for obtaining reliable results. Thus, the relative economy of the two techniques of integration depends on the stability limit of the explicit scheme, the ease with which the nonlinear equations can be solved for the implicit operator, the relative size of time increments that can provide acceptable accuracy with the implicit scheme compared to the stability limit of the explicit scheme, and the size of the model.
Procedures
+1
+1
+1
2.4.11
In ABAQUS/Standard the time step for implicit integration can be chosen automatically on the basis of the half-step residual, a concept introduced in Hibbitt and Karlsson (1979). By monitoring the values t= once the solution at t t has been obtained, the accuracy of the of equilibrium residuals at t solution can be assessed and the time step adjusted appropriately. To discuss the dynamic procedures further, we write out the dAlembert force in the overall equilibrium equation, Equation 2.1.11. The body force at a point, f , can be written as an externally prescribed body force, F, and a dAlembert force:
+1 2
+1
f = F 0 u ;
where is the current density of the material at this point and body force term in the virtual work equation is
The
Z
V
f 1 v dV =
Z
V
F 1 v dV 0
Z
V
u 1 v dV:
The dAlembert term can be written more conveniently in terms of the reference volume and reference density, 0 , as Z where u is the acceleration eld. When implicit integration is used, the equilibrium equations are written t), and u is calculated from the time integration operator. The at the end of a time step (at time t interpolator approximates the displacement at a point as
V0
0 u 1 v dV0;
+1
u = N N uN ;
so that provided that NN are not displacement dependent. This is the case for most of the elements in ABAQUS; the form taken for the dAlembert force terms in those instances where it is not true (the Hermite cubic beams, B23 and B33) is discussed in detail in Chapter 3, Elements. With this interpolation assumption, the dAlembert force term is Z
u = N N uN ;
V0
0 N N
1 NM dV0 uM ;
that is, the consistent mass matrix times the accelerations of the nodal variables. The nite element approximation to equilibrium, Equation 2.1.12, is
M NM uM
where
M NM
+ I N 0 P N = 0;
0 N N
(2.4.11)
IN
V0 Z
1 NM dV0
V0
N : dV0
2.4.12
PN
NN t dS +
1
NN F dV
1
is the external force vector. In this context the terms matrix and vector refer to matrices and vectors in the space of the nodal variables uN . The denition of the mass matrix introduced above is the consistent mass: the mass matrix obtained by consistent use of the interpolation. The rst-order elements in ABAQUS all use lumped mass, where the mass matrix is a diagonal matrix. The lumped matrix is obtained by adding each row of the consistent matrix onto the diagonal. For these rst-order elements the lumped mass matrix gives more accurate results in numerical experiments that calculate the natural frequencies of simple models. The implicit operator used for time integration of the dynamic problem is the operator dened by Hilber, Hughes, and Taylor (1978). This operator is a single parameter operator with controllable numerical dampingthe damping being most valuable in the automatic time stepping scheme, because the slight high frequency numerical noise inevitably introduced when the time step is changed is removed rapidly by a small amount of numerical damping. The operator replaces the actual equilibrium equation (Equation 2.4.11) with a balance of dAlembert forces at the end of the time step and a weighted average of the static forces at the beginning and end of the time step:
Procedures
operator denition is completed by the Newmark formulae for displacement and velocity integration:
0 1 u ujt+1t = ujt + 1t ujt + 1t2 ( 0 )jt + ujt+1t _ 1
M NM uM jt+1t + (1 + )(I N jt+1t 0 P N jt+1t ) 0 (I N jt 0 P N jt) + LN jt+1t = 0; (2.4.12) where LN jt+1t is the sum of all Lagrange multiplier forces associated with degree of freedom N . The
(2.4.13)
and
ujt+1t = ujt + 1t _ _
with
(1 0
)jt +
ujt+1t1; u
and
(2.4.14)
1 2 4 (1 0 ) ;
= 0
1 2
0 1 0: 3
Hilber, Hughes, and Taylor (1978) present cogent arguments for the use of Equation 2.4.12 Equation 2.4.14 for integrating structural dynamics problems. The main appeal of the operator is its controllable numerical damping and the form this damping takes, slowly growing at low frequencies, with more rapid growth in damping at high frequencies. Control over the amount of numerical damping is , there is no damping and the operator is the trapezoidal rule provided by the parameter : with = ); while with 0 = , signicant damping is available. This operator is used (Newmark, primarily because the slight numerical damping it offers is needed in the automatic time stepping scheme. Each time step change introduces some slight noise or ringing into the solution; a little numerical damping ( 0 : seems a good choice) quickly removes this high frequency noise without having any signicant effect on the meaningful, lower frequency response. An energy content output is available and should be printed to monitor the overall energy balance. This has been done in all of the dynamic examples in the
=14
=0 = 13
= 0 05
2.4.13
ABAQUS Example Problems Manual and shows that the numerical dissipation is always quite small (less than 1% of the total energy). The integration of rotations during implicit dynamic calculations is done to preserve accuracy in cases where the rotary inertia is different in different directions in a body. For this purpose the accelerations are integrated in the body axis system, so that Newmarks formula gives the change in velocity as
_ _ jt+1t = jt + 1t
jt+1t + (1 0
)jt ;
where is the angular velocity of the node and is its angular acceleration, both taken in the current t; t is the time increment. direction of the body axis at time t or at time t In the global system this is
+1 1
= ()
Then
_ _ (2.4.15) jt+1t = 1t
jt+1t + 1C 1 jt + 1t(1 0
) jt ; where 1C is the incremental rotation matrix. Let 1 be the increment in rotation from time t to t + 1t.
See Rotation variables, Section 1.3.1, for a discussion of rotation variables, rotation matrices, and the exponential mapping of a skew-symmetric matrix. Newmarks formula for the time integration of the rotation increment in the body axis system gives
1C is
1 _ 1 = 1tjt + 1t2 2 0 jt + jt+1t : Since 1C 1 1 = exp[1^] 1 1 = 1 , the components 1 are the same relative to the body axes at time t or t + 1t. In the global system this is 2 jt+1t + 1C 1 1tjt + 1t2 1 0 jt : _ 1 = 1t 2 Solving for the unknown velocity and acceleration at time t + 1t, the velocity update equation is _ _ jt+1t =
1 + 1C 1 1 0
jt + 1t 1 0
jt ; 1t 2
and the acceleration update equation becomes
The automatic time stepping for dynamic problems in ABAQUS is based on the half-step residual rst proposed in Hibbitt and Karlsson (1979). The concept is quite appealing intuitively. Satisfaction of Equation 2.4.11 at the end of each time step (or actually of the Hilber-Hughes-Taylor form, Equation 2.4.12) ensures equilibrium in the discrete sense of the nite element model at these points in time but does not say anything about the quality of equilibrium at intermediate time points. The idea of the half-step residual is to calculate the equilibrium residual error (the left-hand side of Equation 2.4.11) t= ) and to assess the error in the dynamic response at some intermediate time point (chosen as t prediction by the magnitude of that error. The half-step residual is based on the assumption that the accelerations vary linearly over the time interval (this is the basis of Newmarks formulae) so that, for any nodal displacement or rotation component, u:
+1 2
Procedures
= (1 0 )jt + ujt+1t; 0 1: u Having already solved for the state at t + 1t, this equation, together with Newmarks formulae now written for the time interval from t to t + 1t, requires that 2 1uj = 31ujt+1t + (1 0 2)1tujt + 2(1 0 ) 12t ujt; _
u j = _
1t 1uj + 1 0 ujt + 1 0 2
1tujt; and _ 1 u jt + 1 0 1 u jt ; _ u j = 2 2 1 u j 0 11t 1t 2
where
u j
With these equations it is possible to evaluate the equilibrium residual at any time within the step. Presumably, if the solution is accurate, this residual will be small compared to signicant forces in the problem. The residual at the end of the time step is
1ujt+1t = ujt+1t 0 ujt is the increment in displacement obtained for the time step, 1t.
t0 = t +1t=2
where the
I N jt+1t=2 ; etc.
(t + 1t=2) and
LN jt+1t=2 def =
1 N N 2 (L jt+1t + L jt):
The half-step residual, Rjt+1t=2, is dened as the magnitude of the largest entry in RN jt+1t=2 and provides a measure of accuracy of the time-stepping solution. The motivation behind the calculation of the half-step residual is to provide a measure of the accuracy of the solution for a given time step. Numerical tests show that it provides a sensitive accuracy check on dynamic solutions and suggest that, if P is a typical magnitude of real forces in an undamped elastic system (for which the high frequency response must be modeled reasonably accurately), then if
Rjt+1t=2 0:1P consistently, the time stepping solution has high accuracy; if Rjt+1t=2 P consistently, the time stepping solution has moderately good accuracy; if Rjt+1t=2 10P consistently, the time stepping solution is rather coarse.
Problems with large amounts of natural dissipation of energy, such as elastic-plastic systems, are usually less sensitive to time step choice than purely elastic problems, because the energy that appears in higher frequency modes is quickly dissipated. In such cases maximum half-step residuals in the range of 110 times typical forces indicate quite acceptable accuracy for most studies, and even values of 10100 times typical forces can give useful results for primary effects, such as overall deformation. Thus, the method can offer relatively cost-effective solutions for highly dissipative systems for which we require only moderately accurate prediction of the overall response. The half-step residual is the basis of the adaptive time incrementation scheme. If the half-step residual is small, it indicates that the accuracy of the solution is high and the time step can be increased safely; conversely, if the half-step calculation shows the solution is coarse, the time step used in the solution should be reduced. The algorithm is described in detail in Time integration accuracy in transient problems, Section 8.3.4 of the ABAQUS Analysis Users Manual. The algorithm is purely empirical, but experience shows that it works quite well (Hibbitt and Karlsson, 1979), most especially in initially excited problems with high dissipation, such as impulsively loaded problems (or problems with short duration forcing) with extensive plasticity. In these cases the scheme is economic because the time step naturally increases as the solution progresses and the high frequency response is dissipated. The above observations are based on using the Hilber-Hughes-Taylor operator with 0 : : The slight numerical damping that the operator introduces removes the noise that inevitably enters the solution when the time step is changed. Even in quite lengthy problems the overall energy totals, computed on the basis of physical mechanisms in the model, balance well. The energy balance calculation is useful in assessing a solutionfor example, the extent to which energy is dissipated by plasticity can be measured and it is recommended that the user request occasional printout of the energy balance calculation when doing any analysis with ABAQUS.
= 0 05
Reference
Implicit dynamic analysis using direct integration, Section 6.3.2 of the ABAQUS Analysis Users Manual
2.4.16
INTERMITTENT CONTACT/IMPACT
2.4.2
INTERMITTENT CONTACT/IMPACT
Many dynamic problems involve intermittent contact, often with severe impact occurring when the structures hit. The contact algorithms in ABAQUS/Standard are designed to handle such cases. An example is the pipe whip problem, in which the uid escaping from a ruptured pipe causes pipe motion and possible impact with restraints along the pipe. Depending on the geometry of the pipe and the position of the postulated break, severe impact can occur since the pipe may acquire high velocity before hitting a motionless restraint. Impact conditions with hard contact are modeled in ABAQUS/Standard with the assumption that, at the time of impact, the two impacting surfaces instantaneously acquire the same velocity in the direction of the impact. This fully plastic impact concept is essential to any discrete model, the accuracy of the physical representation of local effects being dependent on the spatial model adopted. It should be emphasized that the plastic impact assumption is local: it is assumed that, at impact, energy is dissipated by some mechanism that is not modeled and whose spatial and temporal scale is innitesimal compared to the discrete model. In the case of a beam model of a pipe hitting some other structure, this would presumably be the local plastic deformation of the pipe wall where it impacts the other structure. The points that acquire the same velocity sometimes separate just after the impact: part of the dynamic contact algorithm is designed to handle this case. The local plastic impact concept is quite different from the simple assumed coefcient of restitution methods sometimes used in scoping analyses in place of detailed analysis of the history of contact forces. The method discussed here is specically designed to allow as muchor as littleof this local detail to be obtained, as required. The usual time integration procedures in ABAQUS maintain the energy balance in terms of the energy mechanisms of the discrete model, so the instantaneous jumps that occur in the velocity and acceleration at impact imply that some other system of equations must govern the solution during impact. Thus, we view the impact events as separate from the usual time stepping and develop a set of impulse equations that allow the propagation of the solution over these instants of time and, hence, provide initial conditions from which the normal time stepping solution can continue to the next such event. To develop the governing equations for impact, assume that, at some time t0, a part I of the surface of two bodies, A and B , comes into contact. Denote the velocities and accelerations of corresponding parts just before impact as
u
Procedures
0 _ 0 _ BI 0 0 AI AI BI
; u ; u ; u u
and
_ + _ + + + AI BI AI BI
;u ;u ;u
just after impact. The fully plastic impact assumption requires that corresponding points acquire the same velocity and acceleration in the direction of impact immediately after impact so that, at time t+ : 0
Here
n is the current normal to the interface surface I . Writing u u+ 0 u0 to describe the velocity jump at
(2.4.21)
1_ = _
0 at a point, we have
2.4.21
INTERMITTENT CONTACT/IMPACT
0 _ n 1 u0
AI
_ + 1u
1
AI
_ _ = n 0u 0 + 1 u
1
BI
1
BI
(2.4.22)
The component of the force per unit area between the bodies across I in the normal direction n, N I , must satisfy
I NA I NA
(2.4.23)
Since nite velocity jumps are occurring at the time of impact in innitesimal time (compared to the time scale of the simulation), during the innitesimal interval t0 to t+ , N I dominates all other forces in 0 0 the system except for the dAlembert forces. This simplies the virtual work equation during t0 to t+ to 0 0
X hZ
A;B
u u dV
1
NA n uA dS
NB
I
n 1 u
dS
= 0:
Integrating from t0 to t+ , 0 0
+ XhZ Z t0
V t
n 1 uB at t0, and NB 0NA , so the second term in Equation 2.4.24 is zero. But n 1 uA However, the constraint (Equation 2.4.22) must be satised by augmenting Equation 2.4.24 with a Lagrange multiplier term with H as the multiplier:
X
A;B
A;B
0 0
u u dV
1
Z Z t+
I
0 0 t 0
NA n uA
+N
n 1 u
B
dS
= 0:
(2.4.24)
Z
V
_ 1u 1 u dV +
Z
I
Hn
0 _ _ _ _ 1 u 0 + 1u 0 u0 0 1u
AI AI BI
1
BI
dS:
The rst term has been integrated over t0 to t+ to give the velocity jump term. Taking the variation 0 0 in the second term and noting that n cannot rotate in the innitesimal time interval because there is no discontinuity in displacement, we obtain
X
A;B
Z
V
1u 1 u dV + _
Z Z
I
Hn
1 u 0 u
AI AI BI
1
BI
dS
_ _ + (1u 0 1u ) 1 n H dS =0 ( )=0
I
(2.4.25)
0
I
_ u0 0 u0 1 n H dS: _
AI BI
This equation is the impulse condition, which can be solved for the velocity jump u at all nodes. The solution also provides the impulse per unit area, H : the time integral of the pressure between the surfaces over the innitesimal time of impact. The equilibrium equation written at time t+ and including the 0 can then be used to obtain the initial accelerations immediately after constraint that n 1 uAI 0 uBI impact.
1_
2.4.22
INTERMITTENT CONTACT/IMPACT
The two sets of equations (for the velocity jumps and the initial accelerations at t+ ) require solution 0 of the mass matrix, augmented by the constraint, with different right-hand sides for the initial acceleration and the velocity jump equations. When the elements attached to the nodes that impact have consistent mass matrices, the equations will give velocity and acceleration jumps throughout that part of the model and not just at the impacting nodes themselves. From these initial conditions at t+ the usual time stepping 0 R _ _ equations can continue, augmented by the constraint that I n 1 (uA 0 uA) dS = 0, which is imposed by a Lagrange multiplier. Since this multiplier represents the interface pressure, its value is monitored for possible separation: if a negative value is seen (that is, if a state of tension exists between A and B across I ), the constraint is removed: this requires another solution for the initial accelerations just after removal of the constraint, although there will be no velocity jump at that time. Separation can occur immediately after impact. If so, the equilibrium equations must be solved again without the constraint to nd the corresponding accelerations. In any real problem, impact and separation will occur at some intermediate point in a time step. To accommodate this, ABAQUS/Standard rst solves the time step by ignoring impact, then estimates (by linear interpolation) the average time of impact or separation of all points that change in the increment, tI , and again solves the increment to that time. All surface contact changes are assumed to take place at that interpolated time point, with the surface clearances adjusted as necessary. For reasonable time steps this geometric adjustment is slight. Typically, the time step used in the solution following a severe impact event is one or two orders of magnitude smaller than that preceding the event. As the high frequency noise generated by impact dissipates (through plasticity and the articial damping introduced by the parameter in the time integration operator), the time step is expanded by the automatic time stepping algorithm and the solution proceeds accordingly. Soft contact conditions treat impact as an elastic event that does not destroy any kinetic energy. The change in velocity is determined by the amount of interpenetration. Thus, with soft contact the standard implicit integration procedure is used and no impulse equations need be solved. Under truly high velocity impact conditions, this may result in nodes bouncing back immediately after impact. This may lead to excessive contact chattering, resulting in convergence problems and small time increments. For this reason, softened contact is not recommended in true impact calculations. However, in certain dynamic calculations where impact effects are not criticalsuch as sheet forming, drop forging, or rolling operationssoft contact can work well because cutbacks due to impact calculations are avoided. The soft contact constraint is enforced via Lagrange multipliers. If the soft contact constraint compatibility is not satised within the given tolerance, a severe discontinuity iteration is performed.
Reference
Procedures
Implicit dynamic analysis using direct integration, Section 6.3.2 of the ABAQUS Analysis Users Manual
2.4.23
SUBSPACE DYNAMICS
2.4.3
SUBSPACE DYNAMICS
An alternative approach provided in ABAQUS/Standard for analysis of nonlinear dynamic problems is the subspace projection method. This method uses direct, explicit integration of the dynamic equations of equilibrium written in terms of a vector space spanned by a number of eigenvectors. The eigenmodes of the system, extracted in an eigenfrequency step prior to the dynamic analysis, are used as the global basis vectors. This method can be very effective for systems with mild nonlinearities that do not substantially change the mode shapes. The method cannot be used for contact problems. As with the other direct integration methods, it is more expensive in terms of computer time than the modal methods of purely linear dynamic analysis, but it is often signicantly less expensive than the direct integration of all of the equations of motion of the model. This method is implemented in ABAQUS/Standard using the explicit (central difference) operator to integrate the equations of motion projected onto the eigenmodes. In this procedure the explicit integration method is particularly effective because the eigenmodes are orthogonal with respect to the mass matrix so that the projected system always has a diagonal mass matrix and because the stability limit is determined by the highest eigenfrequency associated with the modes used in the analysis and not by the highest eigenfrequency of the structure. Let us write the nite element approximation to the virtual work equations as
Procedures
uN M N M uM + I N 0 P N = 0;
where
(2.4.31)
M NM = IN PN =
Z Z
V0
0 NN 1 NM dV0 N : dV0
V0
Z
S
NN 1 t dS +
Z
V
NN 1 F dV
is the external force vector. The nodal displacements, velocities, accelerations, and the variations in displacement are expressed in terms of eigenmodes:
uN = uN
X
X
8 N q ; 8 N q ;
uN = _
8N q_; X uN = 8N q;
(2.4.32)
_ where q , q, q , and q represent generalized displacement, velocity, acceleration, and displacement variation, respectively. Substitution into the virtual work expression yields the formula for the acceleration associated with mode as
2.4.31
SUBSPACE DYNAMICS
where
1 8N P N 0 I N q = m
(no sum on ),
(2.4.33)
From Equation 2.4.33 it is seen that the element residuals are projected onto the vector space spanned by the chosen number of eigenmodes. Having calculated the generalized acceleration for each mode, the generalized displacement and velocity are calculated with the central difference operator
(2.4.34)
The nodal values for all kinematic variables are obtained using the formul in Equation 2.4.32. When initial velocities are applied, specied either explicitly by the user or implicitly by continuation of the previous dynamic step, the initial velocity vector u0 has to be projected into the eigenspace. This leads to an initial generalized velocity for the mode in the form
q = _0
8 M m
N
NM uM : _
As in standard explicit dynamic integration, the method is conditionally stable. The stability limit is determined by the highest eigenfrequency of the modes used in the analysis:
where !max is the highest circular frequency of the eigenmodes that are used as the basis of the solution. Since we will generally use a relatively small number of modes, this stability limit is usually much less restrictive than the stability limit for standard explicit integration. Throughout the procedure a xed time increment is used: the value is chosen as the smaller of the time increment specied by the user, or 80% of the stable time increment. The 80% factor is intended as a safety factor, so a small increase in the highest eigenfrequency caused by nonlinear effects will not cause the integration to become unstable. The number of eigenvectors spanning the solution space for a subspace dynamic analysis can be specied by the user. The default number of vectors will be equal to the number of eigenvectors extracted in the eigenfrequency calculation. It is possible to perform a subspace dynamic simulation for some time and then reextract the modes based on the current, stressed geometry (by using another eigenfrequency extraction step), followed by continuation of the analysis with the new modes as the subspace basis system. This can improve the accuracy of the method in certain cases. Note that the method is noniterative; hence, there are no tolerances required for the procedure.
Reference
1tstable = ! 2
max
Implicit dynamic analysis using direct integration, Section 6.3.2 of the ABAQUS Analysis Users Manual
2.4.32
2.4.4
It is often useful to obtain the equivalent rigid body motion of part of a model (or of the whole model): the position and translational velocity of the parts center of mass and its angular rotation and velocity about the same center of mass. ABAQUS/Standard provides such output, based on equivalent momentum. This section denes how these values are calculated. Let V be the volume of a part for which the equivalent rigid body motion values are requested. The density of the part in its initial conguration is 0(Si ), where Si , i = 1; 2; 3, are material coordinates in the part. The spatial position of a material particle in its initial conguration is (Si ) and in the current conguration is x(Si ), resulting in a displacement of u(Si ). We wish to compute the current spatial _ position of the center of mass of the part, x ; the translational velocity of an equivalent rigid body, x ; the angular velocity of this equivalent rigid body, ! ; the translational displacement of an equivalent rigid body motion, u; and the rotation of an equivalent rigid body motion around the center of mass, . In these denitions an equivalent rigid body means a rigid body with the same mass distribution and the same translational and angular momentum as the actual deforming part in the current conguration. For simplicity of notation we dene some quantities. The mass of the part is
Procedures
M def =
Its rst mass moment about the origin is
V0
Z
0 dV 0:
J def =
Its second mass moment about the origin is
V0
0 x dV 0 :
I def =
V0
0 (x 1 x I 0 x x) dV 0 ;
where I is a unit matrix. It is convenient to invoke the relation x = N N xN (the summation convention is assumed), where N N are the nite element interpolation functions associated with each degree of freedom and xN is the vector of current nodal positions. We can now write
I
def
=
V0
0 (N N N M xN 1 xM I 0 N N N M xN xM ) dV 0 :
V0
0 N N N M dV 0;
we have
I def M NM (xN 1 xM I 0 xN xM ): =
2.4.41
= J =M
0 x dV 0 =M: _
and, by equating the translational momentum of the equivalent and the actual body,
x = _
V0
The angular velocity of the part is dened by equating the angular momentum of the part and of the equivalent rigid body about the center of mass:
V0
_ 0 (x 0 x ) 2 x dV 0
I 1 !;
where
I = I + M (x x 0 x 1 x I )
is the second mass moment of the part about its center of mass. ABAQUS uses the lumped mass formulation for low-order elements. As a consequence, the second mass moments of inertia can deviate from the theoretical values, especially for coarse meshes. Certain ABAQUS elements produce lumped or structural contributions to this second mass moment (rotary inertias) not shown in these equations. This provides ! = I 01 1 [H 0 J 2 x ]; _ where
_ _ 0 x 2 x dV 0 = M NM xN 2 xM V0 is the angular momentum of the part about the origin. The perceived translational motion of the center of mass in an equivalent rigid body motion is calculated as Z u = 0 u dV 0 =M:
H
def
=
V0
The equivalent rigid body rotation of the part with respect to its center of mass requires some conceptual approximations as follows. Denote the relative positions of a material particle with respect to the center of mass in the undeformed conguration and in the deformed conguration X and x, respectively. Consider that the congurations are known and that the axis of rotation of the body is denoted by the unit vector p. A material particle sees such rotation relative to the center of mass as
where the subscript p denotes the projection of a vector into a plane normal to p. We now generalize this concept by integration of the constituent parts. Dene
a=
Xp 2 xp dV;
b=
Xp 1 xp dV:
2.4.42
av
These integrals are not easily calculated, but with the modications below they can be expanded in such a way that the (initially unknown) current center of mass, x , only appears in products with the position of particles in the known current conguration. A necessary condition for the validity of the intuitive generalization above is that if the part undergoes an arbitrary rigid body rotation, the formula returns the rotation. That can easily be proved as follows. Every material particle rotates exactly an angle = av in such a way that
Z
a = p sin
Procedures
b = cos
and, therefore,
jaj = p ;
jaj = tan :
b
In all of these equations the direction vector p is unknown. To determine p we consider the characteristics of the displacement eld of a rigid body rotation. For such a eld, 1) the displacement of a particle is orthogonal to the rotation vector, and 2) the displacement is orthogonal to the position vector at half the motion. In a deformable body context we try to determine p by forcing these two statements to be true in an average sense. Considering that we are looking strictly at a rotation with respect to the center of mass, its denition automatically ensures that the rst statement is satised. The second condition can be written in the form
Z
p 2 (x + X )
2 [x 0 X] dV
= 0;
which is a homogeneous set of equations in the components of p with coefcients made out of integrals of known quantities, from which p can be solved. We can then calculate the projection of the old and new position onto the plane normal to p with
Xp = X
0 p p 1 X;
xp = x
0 p p 1 x;
= pp
X0 X1p p
1 3
2 [x 0 (x 1 p) p] dV
X 2 x dV;
where the vector V X 2 x dV is easily calculated from available quantities. With p known, a becomes determined. The quantity b can be calculated using with the same expressions, which yields
Z
b=
X 0 ( X 1 p )p
1 [x 0 (x 1 p)p] dV
2.4.43
= (I
0 p p) :
X x dV:
Once a and b are known, is readily determined. The determination of the equivalent rigid body rotation is based on average particle translations. Rotational degrees of freedom are ignored in the calculation of this variable; it is assumed that such rotations will produce motions of points that will measurably contribute to the calculation. However, it is possible to nd pathological cases in which that would not be the case; for instance, if the part of the model considered consists of rotary inertia elements only, the calculated average rigid body rotation will be calculated as zero, even if the elements have indeed rotated.
Reference
Implicit dynamic analysis using direct integration, Section 6.3.2 of the ABAQUS Analysis Users Manual
2.4.44
2.4.5
The explicit dynamics analysis procedure in ABAQUS/Explicit is based upon the implementation of an explicit integration rule together with the use of diagonal or lumped element mass matrices. The equations of motion for the body are integrated using the explicit central difference integration rule
The explicit integration rule is quite simple but by itself does not provide the computational efciency associated with the explicit dynamics procedure. The key to the computational efciency of the explicit procedure is the use of diagonal element mass matrices because the inversion of the mass matrix that is used in the computation for the accelerations at the beginning of the increment is triaxial:
+1 u 2 u =u +1 u 1 _ 2 _ where u is velocity and u is acceleration. The superscript ( ) refers to the increment number and 0 and + refer to midincrement values. The central difference integration operator is explicit in that the _ 1 kinematic state can be advanced using known values of u 0 2 and u from the previous increment.
t
u(i+ 1 ) = u(i0 1 ) + 1 _ 2 _ 2
(i+1) (i )
(i+1)
(i )
(i )
Procedures
(i+1)
(i + )
1 2
1 2
(i
(i )
u (i ) = M 0 1 1 ( F (i ) 0 I (i ) ) _ _
where M is the diagonal lumped mass matrix, F is the applied load vector, and I is the internal force vector. The explicit procedure requires no iterations and no tangent stiffness matrix. 1 1 Special treatment of the mean velocities u(i+ 2 ) , u(i0 2 ) etc. is required for initial conditions, certain constraints, and presentation of results. For presentation of results, the state velocities are stored as a linear interpolation of the mean velocities:
The central difference operator is not self-starting because the value of the mean velocity u(0 2 ) needs ) of velocity and acceleration are set to zero unless they are to be dened. The initial values (at time t specied by the user. We assert the following condition:
=0
(1)
Substituting this expression into the update expression for u(i+ 2 ) yields the following denition of u(0 2 ) :
1
u (0 1 ) _ 2
_ 1 = u 0 12 u _
(0)
t
(0)
(0)
2.4.51
Stability
The explicit procedure integrates through time by using many small time increments. The central difference operator is conditionally stable, and the stability limit for the operator (with no damping) is given in terms of the highest eigenvalue in the system as
1t ! 2
max
In ABAQUS/Explicit a small amount of damping is introduced to control high frequency oscillations. With damping the stable time increment is given by
1 t ! 2 ( 1 + 2 0 );
max
where is the fraction of critical damping in the highest mode. Contrary to our usual engineering intuition, introducing damping to the solution reduces the stable time increment. The time incrementation scheme in ABAQUS/Explicit is fully automatic and requires no user intervention. ABAQUS/Explicit uses an adaptive algorithm to determine conservative bounds for the highest element frequency. An estimate of the highest eigenvalue in the system can be obtained by determining the maximum element dilatational mode of the mesh. The stability limit based upon this highest element frequency is conservative in that it will give a smaller stable time increment than the true stability limit that is based upon the maximum frequency of the entire model. In general, constraints such as boundary conditions and contact have the effect of compressing the eigenvalue spectrum, which the element by element estimates do not take into account. ABAQUS/Explicit contains a global estimation algorithm, which determines the maximum frequency of the entire model. This algorithm continuously updates the estimate for the maximum frequency. ABAQUS/Explicit initially uses the element by element estimates. As the step proceeds, the stability limit will be determined from the global estimator once the algorithm determines that the accuracy of the global estimation is acceptable. The global estimation algorithm is not used when any of the following capabilities are included in the model: uid elements, JWL equation of state, innite elements, material damping, dashpots, thick shells (thickness to characteristic length ratio larger than 0.92) or thick beams (thickness to length ratio larger than 1.0), and nonisotropic elastic materials with temperature and eld variable dependency. A trial stable time increment is computed for each element in the mesh using the following expression:
max element is the element maximum eigenvalue. A conservative estimate of the stable time where !max
2 1t = !element ;
increment is given by the minimum taken over all the elements. The above stability limit can be rewritten as
1t = min( Le ); c
d
2.4.52
where Le is the characteristic element dimension and cd is the current effective, dilatational wave speed of the material. The characteristic element dimension is derived from an analytic upper bound expression for the maximum element eigenvalue. Considering the 4-node uniform strain quadrilateral (CPE4R), the characteristic element dimension is
Le
= pBA B
iI
iI
where BiI is the element gradient operator. Similar characteristic element dimensions are derived for all element types found in ABAQUS/Explicit. The current dilatational wave speed is determined in ABAQUS/Explicit by calculating the effective hypoelastic material moduli from the materials constitutive response. Effective Lams constants, and , are determined in the following manner. We dene p as the increment in the equivalent , S as the increment in the deviatoric stress, vol as the increment pressure stress, p 0 1 3 of volumetric strain, and e as the deviatoric strain increment. We assume a hypoelastic stress-strain rule of the form
trace( ) 1 1
Procedures
where K is the effective bulk modulus. The effective moduli can then be computed as
If the strain increments are insignicant, these relationships will not yield numerically meaningful results. In this circumstance ABAQUS/Explicit sets the effective Lams constants to the initial values for the material, o and o . In the case where the volumetric strain increment is signicant but the deviatoric stress increment is not, the effective shear modulus is estimated to be
^ 2^ = 1 (3(o + 2o ) 0 3K ) : 2
cd = ( + 2^) :
These effective moduli represent the element stiffness and determine the current dilatational wave speed in the element as
s^
Bulk viscosity
Bulk viscosity introduces damping associated with the volumetric straining. Its purpose is to improve the modeling of high-speed dynamic events.
2.4.53
There are two forms of bulk viscosity in ABAQUS/Explicit. The rst is found in all elements and is introduced to damp the ringing in the highest element frequency. This damping is sometimes referred to as truncation frequency damping. It generates a bulk viscosity pressure, which is linear in the volumetric strain:
pbv1
= b1 cd Le _vol ;
where b1 is a damping coefcient (default=.06), is the current material density, cd is the current dilatational wave speed, Le is an element characteristic length, and vol is the volumetric strain rate. The second form of bulk viscosity pressure is found only in solid continuum elements (except CPS4R). This form is quadratic in the volumetric strain rate:
pbv2
= (b2Le_vol)2;
where b2 is a damping coefcient (default=1.2) and all other quantities are as dened for the linear bulk viscosity. The quadratic bulk viscosity is applied only if the volumetric strain rate is compressive. The quadratic bulk viscosity pressure will smear a shock front across several elements and is introduced to prevent elements from collapsing under extremely high velocity gradients. Consider a simple one element problem in which the nodes on one side of the element are xed and the nodes on the other side have an initial velocity in the direction of the xed nodes. If the initial velocity is equal to the dilatational wave speed of the material, the elementwithout the quadratic bulk viscositywould collapse to zero volume in one time increment (because the stable time increment size is precisely the transit time of a dilatational wave across the element). The quadratic bulk viscosity pressure will introduce a resisting pressure that will prevent the element from collapsing. The bulk viscosity pressure is not included in the material point stresses because it is intended as a numerical effect onlyit is not considered to be part of the materials constitutive response. The bulk viscosity pressures are based upon the dilatational mode of each element. The fraction of critical damping in the dilatational mode of each element is given by
Linear bulk viscosity is always included in ABAQUS/Explicit. The parameters b1 and b2 can be : and b2 : . The bulk viscosity parameters redened by the user. The default values are b1 can be changed from step to step. If the default values are changed in a step, the new values will be used in any subsequent steps unless they are redened.
= 0 06
=12
For the displacement degrees of freedom, bulk viscosity introduces damping associated with volumetric straining. Linear bulk viscosity or truncation frequency damping is used to damp the high frequency ringing that leads to unwanted noise in the solution or spurious overshoot in the response amplitude. For the same reason, in shells the high frequency ringing in the rotational degrees of freedom is damped with linear bulk viscosity acting on the mean curvature strain rate. This damping generates a bulk viscosity pressure moment, m, which is linear in the mean curvature strain rate:
m
2 = b1 h0 cd L 1 ; 12 1t
2.4.54
where b1 is a damping coefcient (default = 0.06), h0 is the original thickness, is the mass density, cd is the current dilatational wave speed, L is the characteristic length used for rotary inertia and 11 22 is twice the increment in mean curvature. transverse shear stiffness scaling, and The dilatational wave speed is given in terms of the effective Lam constants as
1 = 1 +1
cd =
s^
( + 2^) :
Procedures
The resultant pressure moment mh, where h is the current thickness, is added to the direct components of the moment resultant.
Reference
Explicit dynamic analysis, Section 6.3.3 of the ABAQUS Analysis Users Manual
2.4.55
EIGENVALUE EXTRACTION
2.5.1
EIGENVALUE EXTRACTION
There are many important areas of structural analysis in which it is essential to be able to extract the eigenvalues of the system and, hence, obtain its natural frequencies of vibration or investigate possible bifurcations that may be associated with kinematic instabilities. For example, structural evaluation for seismic events is often based on linear analysis, using the structures modes up to a limiting cutoff frequency, which is usually taken as 33 Hz (cycles/second). Once the modes are available, their orthogonality property allows the linear response of the structure to be constructed as the response of a number of single degree of freedom systems. This opens the way to several response evaluation methods that are computationally inexpensive and provide useful insight into the structures dynamic behavior. Several such methods are provided in ABAQUS/Standard and are described in the following sections. The mathematical eigenvalue problem is a classical eld of study, and much work has been devoted to providing eigenvalue extraction methods. Wilkinsons (1965) book provides an excellent compendium on the problem. The eigenvalue problems arising out of nite element models are a particular case: they involve large but usually narrowly banded matrices, and only a small number of eigenpairs are usually required. For many important cases the matrices are symmetric. The eigenvalue problem for natural modes of small vibration of a nite element model is
0 2 M
Procedures
MN + C MN + K MN 1 N = 0
[M ] + [C ] + [K ] fg = 0;
1
(2.5.11)
where [M ] is the mass matrix, which is symmetric and positive denite in the problems of interest here; [C ] is the damping matrix; [K ] is the stiffness matrix, which may include large-displacement effects, such as stress stiffening (initial stress terms), and, therefore, may not be positive denite or symmetric; is the eigenvalue; and fg is the eigenvectorthe mode of vibration. This equation is available immediately from a linear perturbation of the equilibrium equation of the system. The eigensystem (Equation 2.5.11) in general will have complex eigenvalues and eigenvectors. This system can be symmetrized by assuming that [K ] is symmetric and by neglecting [C ] during eigenvalue extraction. The symmetrized system has real squared eigenvalues, 2, and real eigenvectors only. Typically, for symmetric eigenproblems we will also assume that [K ] is positive semidenite. In this case becomes an imaginary eigenvalue, = i!, where ! is the circular frequency, and the eigenvalue problem can be written as
0
0!2 [M ] + [K ] fg = 0:
(2.5.12)
If the model contains hybrid elements, contact pairs, or contact elements, the system of equations contains Lagrange multipliers and the stiffness matrix [K ] becomes indenite. However, all the terms of the mass matrix corresponding to the Lagrange multipliers are equal to zero. Therefore, all the eigenvalues are imaginary, and the eigenvalue problem can still be written as Equation 2.5.12.
2.5.11
EIGENVALUE EXTRACTION
ABAQUS provides eigenvalue extraction procedures for both symmetric and complex eigenproblems. For symmetrized eigenproblems ABAQUS/Standard offers two approaches: Lanczos and subspace iteration methods. For complex eigenproblems the subspace projection method is used.
Eigenvalue extraction for symmetric systems
The structural eigenvalue problem has received considerable attention since the advent of nite element models. Ramaswami (1979) summarizes available methods for the problem: the most attractive appear to be the Lanczos method (see, for example, Newman and Pipano, 1973; Parlett, 1980) and the subspace iteration method, a classical method that was introduced into nite element applications by Bathe and Wilson (1972). Both the subspace iteration and the Lanczos methods, using the Householder and Q-R algorithm for the reduced eigenproblem, have been implemented in ABAQUS/Standard. The various parts of these algorithms are discussed in the remainder of this section.
Subspace iterationthe basic algorithm
The application of the subspace iteration method to eigenproblems arising from nite element models of the dynamic behavior of structures has been discussed by a number of authorssee Ramaswami (1979) for references. The basic idea is a simultaneous inverse power iteration. A small set of base vectors is created, thus dening a subspace: this subspace is then transformed, by iteration, into the space containing the lowest few eigenvectors of the overall system. Assume that, at the i-th iteration, a set of m vectors, fV gni) , n = 1; . . . ; m, exists, where m ( is less than the number of variables in the nite element model. These are considered as the base vectors that dene the m-dimensional subspace out of the n dimensions dened by the variables in the nite element model. We arrange these vectors as the columns in the matrix [V ](i) . The number of rows of [V ](i) is, thus, the size of the complete set of equations (the number of variables in the nite element model, N ), and the number of columns is the dimension of the subspace, m. The rst step in the algorithm is to dene a new set of base vectors by solving
^ [K ][V ](i+1) = [M ][V ](i) :
(2.5.13)
This operation, a generalized inverse power sweep with m vectors, involves the solution of the complete set of linear stiffness equations for several right-hand-side vectors (with, after the rst iteration of the method, a previously decomposed matrix). The stiffness and mass matrices of the structure are then projected onto the subspace by
3 ^ T ^ [K ] = [V ](i+1) [K ][V ](i+1) ; 3
thus dening a mass matrix, [M 3 ], and a stiffness matrix, of dimension m by m. The eigenproblem
(
[ K 3 ],
0!2 [M 3] + [K 3])fg = 0
2.5.12
(2.5.14)
EIGENVALUE EXTRACTION
is now solved completely in the subspace using the Householder and Q-R methods, which are discussed below. The eigenvectors have now been dened in the reduced space and can be transformed back to the full space of the structural problem to dene [V ] for the next iteration:
^ [V ](i+1) = [V ](i+1) [];
where [] = [fg1 jfg2j 1 1 1jfgm ]. This completes an iteration. With m = 1 (a one-dimensional subspace), the method reduces to the simple inverse power sweep method (see Wilkinson, 1965, or Strang, 1976). The advantage of the subspace method is the extraction of the eigenvalues in reduced space, which will cause a rapid convergence to the eigenvectors in full space. The number of base vectors carried in the iterations and the choice of initial base vectors are, therefore, important for an economical 2 2 2 solution. The convergence rate of a particular eigenvalue is proportional to !i =!(m+1) , where !(m+1) is the eigenvalue corresponding to the next vector, m + 1 beyond the m vectors used to dene the subspace. The default value of m used in ABAQUS is m = min(2p; p + 8), where p is the number of eigenvectors requested. If more vectors are used, the number of required iterations is reduced, but each iteration takes longer because of the greater number of right-hand sides. Increasing m can sometimes improve the performance of the algorithm signicantly. The choice of starting vectors is also important. There is no requirement that these should be close to the eigenvectors for rapid convergence. The Householder and Q-R steps rotate the vectors into the eigenvectors as long as the base vectors span the space of the eigenvectors. The approach used is to choose initial vectors that span this space as completely as possible. The algorithm used in ABAQUS/Standard for this purpose is that of Bathe and Wilson (1972). They recommend using the diagonal mass terms as one vector: the other vectors are unit vectors, providing a set of single unit entries at the nodes and the degrees of freedom with the largest mass terms. A check is made to ensure that all degrees of freedom are represented in these additional vectors. This completes the description of the basic subspace iteration algorithm as it is implemented in ABAQUS/Standard. The small eigenproblem, for which all eigenpairs must be found, is solved by the Householder and Q-R algorithms. These algorithms are described at the end of this section.
Lanczos eigensolver
Procedures
The implementation of the Lanczos eigensolver as a powerful tool for extraction of the extreme eigenvalues and the corresponding eigenvectors of a sparse symmetric generalized eigenproblem has been discussed by a number of authors; see, for example, Parlett (1980), Parlett and Nour-Omid (1989), Simon (1984), and Ericsson and Ruhe (1980). A shifted block Lanczos algorithm was developed and described in detail by Grimes, Lewis, and Simon (1994). The Lanczos procedure in ABAQUS/Standard consists of a set of Lanczos runs, in each of which a set of iterations called steps is performed. For each Lanczos run the following spectral transformation is applied:
[M ] ([K ]
(2.5.15)
EIGENVALUE EXTRACTION
where is the shift, is the eigenvalue, and fg is the eigenvector. This transformation allows rapid convergence to the desired eigenvalues. The eigenvectors of the symmtrized problem (Equation 2.5.12) and the transformed problem (Equation 2.5.15) are identical, while the eigenvalues of the original problem and the transformed problem are related in the following manner:
!2
= 1 + :
A Lanczos run will be terminated when its continuation is estimated to be inefcient. In general, only tens of eigenvalues (closest to the shift value) are computed in a single Lanczos run. The possibility of computing many eigenmodes by carrying out several runs with different shift values is an important feature of the Lanczos eigensolver. Within each run a sequence of Krylov subspaces is created, and the best possible approximation of the eigenvectors on each subspace is computed in a series of steps. In each Lanczos step the dimension of the subspace grows, allowing better approximation of the desired eigenvectors. This is in contrast to the subspace iteration method, in which the dimension of the subspace used to approximate the eigenvectors is xed. In theory the basic Lanczos process (in the assumption of exact computations without taking into account round-off errors) is able to determine only simple eigenvalues. The shifting strategy (and the Sturm sequence check as a part of it) detects missing modes and enforces computation of all the modes during the subsequent Lanczos runs. However, this strategy is expensive if the multiplicity of certain eigenvalues is high. Therefore, a blocked version of the Lanczos algorithm is implemented in ABAQUS/Standard. The idea is to start with a block of orthogonal vectors and to increase the dimension of the Krylov subspaces by the block size at each Lanczos step. This approach allows automatic computation of all multiple eigenvalues if the largest multiplicity does not exceed the block size. Another important advantage of the blocked Lanczos method is that it allows efcient implementation of expensive computational kernels such as matrix-blocked vector multiplications, blocked back substitutions, and blocked vector products. As discussed above, the Lanczos process consists of several Lanczos runs. Each Lanczos run is associated with some shift value that remains constant during the run. The initial shift value, 1 , is determined (by a heuristic approach) using the geometric mean of the centers of the Gershgorin circles, termed the problem scale. At the beginning of each Lanczos run a factorization of the shifted matrix [K ] 0 p [M ] (where p is the run number) is carried out, after which a sequence of Lanczos steps is performed. Each block Lanczos step i is implemented in ABAQUS/Standard in the following manner: 1. Solve the system of linear equations with b right-hand sides (b is the Lanczos block size) using the factorized shifted matrix:
([K ] ^ 0 p[M ]) [V ](i+1) = [M ][V ](i);
where [V ](i) is a block of Lanczos vectors. The number of rows of [V ](i) is the number of variables in the nite element model, and the number of columns is the Lanczos block size b. The initial block of vectors [V ](1) is a set of random vectors orthonormalized using the procedure: [I ] = [V ]T [M ][V ](1) ; where [I ] is the identity matrix. (1)
2.5.14
EIGENVALUE EXTRACTION
Procedures
where i+1 is a b 2 b upper triangular matrix. 6. Estimate the loss of orthogonality between [V ](i+1) and [V ](j ) ; for j = 1; 2; :::;i 0 1, using the partial reorthogonalization technique; and perform reorthogonalization if necessary (see Simon, 1984; Grimes, Lewis, and Simon, 1994). 7. Perform local reorthogonalization: recompute [V ](i+1) to provide
[T ][S ] = [S ] [] ;
where
and [S ] and [] are, respectively, the matrices containing the eigenvectors and eigenvalues of the reduced eigenproblem. This problem is solved by the Householder and Q-R algorithms, which are discussed below.
2.5.15
EIGENVALUE EXTRACTION
9. Determine the error bounds in eigenvalue approximation for the symmetrized eigenvalue problem (Equation 2.5.12) (see Parlett, 1980; Grimes, Lewis, and Simon, 1994). Check the termination conditions of the Lanczos run. The Lanczos run terminates if one of the following conditions is satised: All the eigenvalues required for the current run are extracted. The triangular matrix i+1 is singular or ill-conditioned. The number of Lanczos steps exceeds the maximum number allowed. Continuation of the current run is evaluated to be inefcient. This decision is based on the estimation of the cost per eigenvalue over the next few steps (the Lanczos run is continued as long as the cost per eigenvalue is decreasing). After termination of each Lanczos run, the converged eigenvectors of the symmetrized problem (Equation 2.5.12) are recovered using the blocks of vectors [V ](j ) ; j = 1; 2; :::; i and the eigenvectors [ S ]. Once the Lanczos run for the shift p is completed, the shift value p+1 is computed using the results of all the previous Lanczos runs. To describe the shifting strategy, the following concepts are introduced: The computational interval is the interval between the minimum and maximum eigenvalues of interest. This interval can be nite (both ends are nite), semi-innite (only one end is nite), or innite (both ends are innite). The center point is the point in the computational interval nearest to the desired eigenvalues. The Sturm sequence number, ([A]), of a real nonsingular symmetric matrix [A] is the number of negative eigenvalues. This number is equal to the number of negative terms in the diagonal matrix [D ] of the Cholesky decomposition [A] = [L][D ][L]T and is, therefore, available after the Cholesky decomposition is completed. Let 1 and 2 be two shift values such that (1 < 2) : Then the number of eigenvalues of the symmetrized problem (Equation 2.5.12, assuming that [M ] is positive denite or positive semidenite) in the interval [1 ; 2] is equal to ([K ] 0 2[M ]) 0 ([K ] 0 1[M ]) : If this number is equal to the number of eigenvalues actually determined by the Lanczos algorithm, the interval [1; 2] is called a trust interval. The trust interval containing the center point is referred to as a primary trust interval (denoted by the key + in the trust intervals list printed in the message le). The shifting strategy is aimed at constructing the primary trust interval inside the computational interval containing the required number of eigenvalues closest to the center point. One of the most important features in the formulation of the shift update strategy is the concept of a sentinel. The sentinels are the endpoints of the intervals containing exclusively converged eigenvalues closest to the current shift value (in each direction). The sentinels are computed during each Lanczos run and are updated at the end of each step after the eigenvalue analysis of the reduced matrix [T ] is completed. The basic assumption is that there are no missing eigenvalues inside the sentinels; therefore, they are excluded from the computation intervals for the upcoming Lanczos runs. This assumption is later veried on the basis of the Sturm sequence check, and a special procedure is activated if some eigenvalues are missing. If no missing eigenvalues are detected, the sentinels transform directly into the corresponding trust intervals.
2.5.16
EIGENVALUE EXTRACTION
The new shift values are selected on the basis of the nonconverged eigenvalue approximations after the Lanczos run is terminated. Assuming the same convergence properties for the upcoming runs, the new shift values are selected in such a way that the number of eigenvalues expected to be found in the upcoming runs will be close to the number of eigenvalues found inside the corresponding sentinels on the previous run.
The Householder method with quarter rotation
The Householder method is used to reduce a general matrix to a symmetric tridiagonal form. A tridiagonal matrix is one whose only nonzero entries are on or immediately adjacent to the diagonal. The rst step is to transform Equation 2.5.14 to the form
0
Procedures
[A] 0 !2 [I ]
f g = 0 ;
where [I ] is the identity matrix. This is done by using a Cholesky decomposition of [M 3 ] and then premultiplying and postmultiplying [K 3 ] by the inverse of the lower and upper triangular matrices that are, thus, obtained:
To preserve symmetry in [A], it is necessary that the decomposition of [M 3 ] result in two matrices that are the transpose of each other. A Cholesky decomposition produces the desired result but adds the requirement that the matrix [M 3 ] be positive denite. [M 3] should always be positive denite (because [M ] is positive denite in all of the problems considered here) if the base vectors dening the subspace are not linearly dependent across [M ]. In practice the choices made for starting values of the base vectors usually satisfy this requirement, although cases can arise when this is not true. Then ABAQUS/Standard will reduce the dimensionality of the subspace to obtain a positive denite [ M 3 ]. The Householder transformation starts with the matrix [A] and, proceeding one column at a time, reduces all the entries outside the tridiagonal part of the matrix to zero by the transformation
[L][L]T = [M 3]:
[ A ] 2 = [ U ] 0 1 [ A ] 1 [ U ]1 : 1
The matrix [U ] is of the form
(2.5.16)
1 C C: A
1 B0 U = B0 B @ 0 0
0 0 1 0 0 0 0
0 0 [H ]
0 0C
For the rst transformation [H ] is of the same order as [A] and [U ]; that is, m 2 m, the dimension of the subspace. However, each transformation reduces the order of [H ] by one, the leading parts of [U ] being 1 on the diagonal and 0 outside the diagonal, as illustrated above.
2.5.17
EIGENVALUE EXTRACTION
The matrix 1.
[H ] is obtained by the following algorithm: If this is the kth iteration, write the kth column of [A] jX j = bX cfX g :
fX g = ba +1 ; a +2 . . . a c :
2. Calculate the norm 3. Dene a vector,
;k
;k
n;k
bV c = ba +1 c + jX jba +1 . . . a c :
k ;k k ;k n;k T
4. Then,
1 [H ] = [I ] 0 2 bV cfV g fV gbV c:
This algorithm is developed in detail in Strangs (1976) book. The Householder algorithm produces a symmetric tridiagonal matrix, which has the same eigenvalues as the original matrix, because the transformation (Equation 2.5.16) does not alter the eigenvalues. The next step is to calculate the eigenvalues of the tridiagonal matrix. This is done by the Q-R algorithm. In this method the matrix [A], which is now tridiagonal (although this is not a requirement for the method to work), is factored into [Q][R], where [Q] is an orthonormal matrix (that is, [Q]T Q = [I ]) and [R] is an upper-triangular matrix (that is, all the terms in [R] below the diagonal are zero). The next matrix in the iteration is obtained by premultiplying and postmultiplying [A] by [Q]:
[A] +1 = [Q] [A] [Q] = [Q] [Q] [R] [Q] = [R] [Q] :
i T i i i T i i i i i i
Because [Q] is orthonormal, this will not affect the eigenvalues. This process gradually reduces the offdiagonal terms of [A] so that the diagonal terms approach the eigenvalues. To speed up convergence, a method of shifting is introduced, leading to the following iterative loop:
[B ] = [ A ] 0 [I ]
i i i
The shift cannot be used until the iteration is converging toward an eigenvalue; but once that is obvious, the shift value i can be set to the expected eigenvalue. This will lead to a signicant improvement in the convergence rate. If the shift is done too early, with a value that is not close to an eigenvalue, it is quite possible that the process will converge to an incorrect number. The Q-R process will converge to the eigenvalues in ascending order; and as soon as an eigenvalue is obtained, the order of the matrix can be reduced by one.
2.5.18
EIGENVALUE EXTRACTION
The nal step after the eigenvalues have been obtained is to calculate the eigenvectors by using the inverse power method to solve for an eigenvector, given any right-hand side:
2
[K 3 ] 0 [M 3 ] = V ;
38 9
8 9
where is the eigenvalue just obtained and fV g is any vector. Because the left-hand-side matrix is singular in the direction of the eigenvector fg, this vector will be obtained regardless of the right-hand-side vector fV g, as long as fV g is not orthogonal to the eigenvector. To ensure that consecutive vectors are orthogonal, especially in the case of multiple 2 eigenvalues, fV3g is always chosen to be orthogonal to the previously extracted eigenvectors. Since [K 3] 0 [M 3 ] is singular, a slight numerical shift must be included to decompose it and, thus, solve for fg. In the standard notation used in this manual, the matrix of eigenvectors [] is written as N , where N refers to the nodal variable in the problem and = 1; 2; . . . ; p is the mode number.
Complex eigenvalue extraction
Procedures
ABAQUS/Standard offers the subspace projection method to solve for complex eigenvalues and right eigenvectors of the original eigenproblem (Equation 2.5.11).
Subspace projection method
In the subspace projection method the original eigensystem (Equation 2.5.11) is projected onto a subspace spanned by the eigenvectors of the undamped, symmetric system (Equation 2.5.12). Thus, the symmetrized eigenproblem must be solved prior to the complex eigenvalue extraction procedure to create the subspace onto which the original system will be projected. Next, the original mass, damping, and stiffness matrices are projected onto the subspace of N eigenvectors:
[ M 3 ] + [ C 3 ] + [K 3 ] f 3 g = 0 :
Typically, the number of eigenvectors is relatively small; a few hundred is common. This small complex eigenvalue system is solved using the standard QZ method for generalized nonsymmetric eigenproblems. Complex eigenvalues of the projected system are the approximation of the eigenvalues of the original system (Equation 2.5.11), but the right eigenvectors of the original system need to be recovered: where
EIGENVALUE EXTRACTION
The only energy density calculated in the eigenvalue extraction procedures is the elastic strain energy density, which for the complex eigenvalue extraction is dened as follows:
ES
p
(
r : "r
where r ; i are real and imaginary components of the eigenstress tensor and " r ; "i are real and imaginary components of the eigenstrain tensor.
References
Eigenvalue buckling prediction, Section 6.2.3 of the ABAQUS Analysis Users Manual Natural frequency extraction, Section 6.3.5 of the ABAQUS Analysis Users Manual Complex eigenvalue extraction, Section 6.3.6 of the ABAQUS Analysis Users Manual
2.5.110
MODAL VARIABLES
2.5.2
After an eigenfrequency step has been used to nd the eigenvalues of a model, ABAQUS/Standard automatically calculates the participation factor, the effective mass, and the composite modal damping for each mode so that these variables are available for use in subsequent linear dynamic analysis.
Generalized mass
is
(no sum over );
Procedures
m = N M NM M
where M NM is the structures mass matrix and N is the eigenvector for mode . The superscripts N and M refer to degrees of freedom of the nite element model. ABAQUS/Standard allows the user to choose between two types of eigenvector normalization: the eigenvectors can be scaled so that the largest entry in each vector is unity, or they can be normalized so that the generalized mass for each vector is unity (m = 1). The choice of eigenvector normalization type does not inuence the results of subsequent modal dynamic steps. The normalization type determines only the manner in which the eigenvectors are represented.
Modal participation factors
The participation factor for mode in direction i, 0i , is a variable that indicates how strongly motion in the global x-, y- or z -direction or rotation about one of these axes (indicated by i, i = 1, 2, . . ., 6) is represented in the eigenvector of that mode. It is dened as
0i =
1 N NM M M Ti m
where TiN denes the magnitude of the rigid body response of a degree of freedom in the model (N ) to imposed rigid body motion (displacement or innitesimal rotation) in the i-direction. For example, at a node with the usual three displacement and three rotation components, TiN is
01 B0 B0 B B0 B @0
0 1 0 0 0 0 0
0 0 1 0 0 0
( z 0 z0 ) 0 (z 0 z 0 ) 0 ( y 0 y 0 ) 0 (x 0 x 0 ) 1 0 0 1 0 0 0
where ei is unity; and all other ep zero, x, y, and z are the coordinates of the node; and x0, y0 , and ^ ^ z0 represent the coordinates of the center of rotation. The participation factors are, thus, dened for the translational degrees of freedom and for rotation around the center of rotation.
2.5.21
MODAL VARIABLES
For coupled acoustic-structural eigenfrequency analyses, an additional acoustic participation factor is computed for each mode as outlined in Coupled acoustic-structural medium analysis, Section 2.9.1.
Modal effective mass
If the effective masses of all modes are added in any particular direction, the sum should give the total mass of the model, except for mass at kinematically restrained degrees of freedom. Thus, if the effective masses of the modes used in the analysis add up to a value that is signicantly less than the models total mass, this suggests that modes that have signicant participation in excitation in that direction have not been extracted. For coupled acoustic-structural eigenfrequency analyses, an additional acoustic effective mass is computed for each mode as outlined in Coupled acoustic-structural medium analysis, Section 2.9.1.
Composite modal damping
ABAQUS/Standard provides an option to dene a composite damping factor for each material. These 0 1 are assembled into fractions of critical damping values for each mode, c , according to
0
1 N c = m
1
NM a M a
M
where a is the critical damping fraction given for material mass matrix made up of material a.
References
NM a and Ma
Natural frequency extraction, Section 6.3.5 of the ABAQUS Analysis Users Manual Material damping, Section 12.1.1 of the ABAQUS Analysis Users Manual
2.5.22
LINEAR DYNAMICS
2.5.3
Linear dynamic analysis using modal superposition is computationally inexpensive and can provide useful insight into the dynamic behavior of a system. With modern eigenvalue/eigenvector extraction techniques such as the subspace method available in ABAQUS/Standardthe cost of obtaining a sufcient basis of eigensolutions is not excessive; and the subsequent computational effort involved in obtaining the dynamic response by modal superposition methods is relatively small, especially when compared to the cost of the direct integration methods used for general nonlinear analysis (Implicit dynamic analysis, Section 2.4.1). The basic concept of modal superposition is that the response of the structure is expressed in terms of a relatively small number of eigenmodes of the system. The orthogonality of the eigenmodes uncouples this system. Furthermore, only eigenmodes that are close to the frequencies of interest are usually needed; for example, only the lowest few frequencies are usually required to obtain an accurate estimate of a structures linear dynamic response to relatively long-term loading (for example, its steady-state response to low frequency excitation). The technique can be extended in a limited way into the nonlinear rgime, but the superposition and orthogonality principles apply only to purely linear systems: for this reason the methods described in this section are implemented only for linear analysis. ABAQUS/Standard has two subspace proceduresone for nonlinear dynamic and the other one for steady-state dynamic analysisthat use some of the eigenmodes of the system on which the equilibrium equations are projected. In both cases the systems eigenmodes are used as a set of global basis vectors for computing the dynamic response, even though the system exhibits nonlinear or frequency-dependent effects during the dynamic response. These methods are cost-effective compared to fully nonlinear dynamic response analysis developed in terms of all the systems degrees of freedom. The subspace projection method for steady-state response is described in Subspace-based steady-state dynamic analysis, Section 2.6.2. The time-domain subspace projection method is described in Subspace dynamics, Section 2.4.3. The procedures provided for modal dynamic analysis of linear systems are summarized below: a. Modal dynamic time history analysis (see Modal dynamic analysis, Section 2.5.5). This procedure can be used to obtain the time history response of a system to loading conditions that are given as functions of time. The response is integrated through time: the integration method used is exact for loadings that vary piecewise linearly with time. Thus, the only approximations in this analysis procedure are the linearization of the problem, the spatial modeling (that is, the choice of the nite element model), the loading denitions, and the choice of the number of eigenmodes used to represent the system. b. Response spectrum analysis (see Response spectrum analysis, Section 2.5.6). Response spectrum analysis is often used to obtain an approximate upper bound to the peak signicant response of a system to an input spectrum as a function of frequency: it gives the maximum response of a one degree of freedom system as a function of its fundamental frequency of vibration and of its damping ratio. The method has very low computational cost and gives useful information about the spectral behavior of a system with respect to frequency. c. Steady-state harmonic response analysis (see Steady-state linear dynamic analysis, Section 2.5.7, and Subspace-based steady-state dynamic analysis, Section 2.6.2).
Procedures
2.5.31
LINEAR DYNAMICS
This procedure is used when the steady-state response of a system to harmonic excitation is required. The solution is given as the peak amplitudes and phase relationships of the solution variables (stress, displacement, etc.) as functions of frequency: postprocessing options are provided to display such results conveniently. A similar option is provided for direct harmonic response analysis without using the eigenmodes as a basis. The direct method is signicantly more expensive computationally than the modal method: it is needed if the system is nonsymmetric (because ABAQUS presently does not have a nonsymmetric eigenvalue extraction capability) or if the systems behavior includes frequency-dependent parameters. The subspace method is typically less expensive than the direct method. It is generally used for nonsymmetric systems or when the systems behavior includes frequency-dependent parameters or discrete damping. d. Random response analysis (see Random response analysis, Section 2.5.8). This procedure is used when the structure is excited continuously and the loading can be expressed statistically in terms of a Power Spectral Density Function. The response is calculated in terms of statistical quantities, such as the mean value and the standard deviation of nodal and element variables.
Reference
Dynamic analysis procedures: overview, Section 6.3.1 of the ABAQUS Analysis Users Manual
2.5.32
2.5.4
For linear dynamic analysis based on modal superposition, several options are provided in ABAQUS/Standard to introduce damping, as follows:
Critical damping factors
The damping in each eigenmode can be given as a fraction of the critical damping for that mode. The equation of motion for a one degree of freedom system (one of the eigenmodes of the system) is
Procedures
mq + cq + kq = 0; _ q = A exp t;
where m is the mass, c the damping, The solution is of the form where
A is a constant, and
0 c 6 c2 0 k : = 2m 4m 2 m
ccr = 2 mk:
The solution will be oscillatory if the expression under the root sign is negative. Critical damping is dened as the damping that makes this expression zero:
If the system is critically damped, after any disturbance the system will return to a static equilibrium state as rapidly as possibly without any oscillation. Typically, when damping is given as a fraction of critical damping associated with each mode, the values used are in the range of 1% to 10% of critical damping. This method of introducing damping has no physical basis in the nite element model: it is a purely mathematical concept introduced in association with the eigenmodes of the system. Thus, the concept cannot be extended to nonlinear applications where the equations of motion of the system are integrated directly and where the natural frequencies of the system are constantly changing because of nonlinearities.
Rayleigh damping
Rayleigh damping is dened by a damping matrix formed as a linear combination of the mass and the stiffness matrices:
C MN = M MN + K MN :
With Rayleigh damping the eigenvectors of the damped system are the same as the eigenvectors of the undamped system. Rayleigh damping can, therefore, be converted into critical damping fractions for each mode: this is the way Rayleigh damping is handled in ABAQUS/Standard.
2.5.41
A form of Rayleigh damping is also provided in ABAQUS for nonlinear analysis. When the problem is nonlinear the mass damping factor can be used directly: the stiffness damping factor is interpreted as creating viscoelastic behavior in which the viscosity is proportional to the elasticity, which gives exactly the stiffness proportional damping effect dened above for the linear case.
Composite modal damping
When composite modal damping is used, a damping value is dened for each material as a fraction of critical damping to be associated with that material. These values are converted into a weighted average for each eigenmode, weighted by the mass matrix according to the equation
=
M M MN N m m m
where is the critical damping ratio used in mode ; m is the critical damping fraction dened for MN material m; Mm is the mass matrix associated with material m; M is the eigenvector of the th mode; and m is the generalized mass associated with the th mode (m = M M MN N , no sum over ).
Structural damping
N where I N are the forces caused by stressing of the structure, FD are the damping forces, s is the p structural damping factor, and i = 01. Any combination of damping options can be used in an analysis: the effects will be added if several damping denitions are chosen.
Reference
Structural damping assumes that the damping forces are proportional to the forces caused by stressing of the structure and are opposed to the velocity. This form of damping can be used only if the displacement and velocity are exactly 90 out of phase, which is the case when the excitation is sinusoidal, so structural damping can be used only in steady-state and random response analysis. The damping forces are then N FD = isI N ;
2.5.42
2.5.5
The modal dynamic procedure provides time history analysis of linear systems. The excitation is given as a function of time: it is assumed that the amplitude curve is specied so that the magnitude of the excitation varies linearly within each increment. When the model is projected onto the eigenmodes used for its dynamic representation, we obtain a set of uncoupled one degree of freedom systems, for any of which the equilibrium equation at time t is
q
where is the critical damping ratio (the ratio of the damping term in this equation to that damping that would cause critical damping of the equation); q is the generalized coordinate of the mode (the amplitude k=m is the natural frequency of the undamped mode (obtained as of the response in this mode); ! the square root of the eigenvalue in the eigenfrequency step that precedes the modal dynamic time history analysis); f is the magnitude of the loading projected onto this mode (the generalized load for the mode); and f is the change in f over the time increment, which is t. The solution to this equation is readily obtained as a particular integral for the loading and a solution to the homogeneous equation (with no right-hand side). These solutions can be combined and written in the general form
(2.5.51)
Procedures
(2.5.52) = ; _ _ + f +1 where a and b , i; j = 1; 2, are constants, since we have assumed that the loading only varies linearly over the time increment (that is, 1f =1t is constant). There are three cases of this solution for nonrigid-body motion (! 6= 0), depending on whether the damping in the modal equilibrium equation is greater than, equal to, or less than critical damping (that is, depending on whether ( 2 0 1) is positive, zero, or negative). For convenience, we dene p ! = ! 1 0 2:
qt qt ft
t t
ij ij
qt+1t qt+1t
a11 a21
a12 a22
b11 b21
b12 b22
= exp (0!1t) ! sin !1t + cos !1t ! 1 a12 = exp (0!1t) sin ! 1t ! ! sin !1t a21 = 0 exp (0!1t) p 1 0 2 ! a22 = exp (0!1t) cos ! 1t 0 ! sin !1t
2.5.51
2 2 + 22!0 1 sin !1t + !12 + !21t cos !1t + !31t 3 !! ! 1t 2 2 0 1 sin !1t + 2 cos !1t + 1 0 2 b12 = exp (0!1t) ! 2 ! 1t ! 3 1t !2 ! 3 1t 2 2 0 1 + b21 = 0 exp (0!1t) ( cos ! 1t 0 ! sin !1t) ! ! 2 ! 1t !! 1 + 2 0 1 0 ( sin !1t + ! cos !1t) !2 !3 1t ! ! 2 1t 201 2 2 ! ! !!21t + ( sin !1t + ! cos !1t) !31t b22 = 0 exp (0!1t) 0 ( cos ! 1t 0 ! sin ! 1t) + !211t
b11
In this case
= exp (0!1t) (1 + !1t) a12 = exp (0!1t) 1t a21 = 0 exp (0!1t) (!)2 1t a22 = exp (0!1t) (1 0 !1t) 1 + 2 + 1 b11 = 0 exp (0!1t) 0(1 + !1t) !2 ! 3 1t !2 2 0 1 + 1 0 2 b12 = exp (0!1t) (1 + !1t) ! 3 1t !2 !2 ! 3 1t 1 + 2 + 1 0 !1t 0 1 2 b21 = 0 exp (0!1t) (!) 1t !2 ! 3 1t ! 2 1t 1t!2 2 0 1 0 !1t + 1 2 b22 = 0 exp (0!1t) 0(!) 1t 3 ! 1t 1t!2 ! 2 1t
a11
In this case
= exp (0!1t) ! sinh !1t + cosh !1t ! 1 a12 = exp (0!1t) sinh ! 1t ! ! sinh !1t a21 = exp (0!1t) p 2 0 1 ! a22 = exp (0!1t) cosh ! 1t 0 ! sinh !1t
a11
2.5.52
2 2 2 0 1 2 + !2!1t sinh !1t + !12 + !231t cosh !1t + !31t 2 !! 2 0 1 sinh !1t + 2 cosh !1t + 1 0 2 b12 = exp (0!1t) 2 ! ! 1t ! 3 1t ! 2 ! 3 1t 2 2 0 1 + ! b21 = 0 exp (0!1t) ( cosh !1t 0 ! sinh ! 1t) 2 ! ! 1t ! ! 1 1 0 (0! sinh ! 1t + ! cosh !1t) !2 + !21t 0 !2 1t 3 2 ! 2! 0 1 b22 = 0 exp (0!1t) 0 ( cosh !1t 0 ! sinh !1t) 2 ! 1t 2 + 1 + (0! sinh !1t + ! cosh !1t) !31t !21t
b11 = 0 exp (0! 1t)
Rigid body mode with damping
Procedures
If there are rigid body modes in the nite element model, there will be one or several eigenvalues that are zero. The equation of motion (Equation 2.5.51) is reduced to
q + q = ft01t + _
Only Rayleigh damping can be specied for a rigid body mode, since the critical damping is zero. Furthermore, since it is a rigid body mode, only the mass damping factor, , appears (stiffness damping requires that there be straining of the body). For this case
1f 1t: 1t
(2.5.53)
a11 =1
b11 = 0 2 (1 0 exp (01t)) 1 + + 1t 1 t 1 1 1t 1 b12 = (1 0 exp (01t)) 2 0 2 + 1 t 2 1 1 + 1 (1 0 exp (01t)) 0 1 b21 = 1 t b22 = 0 2 (1 0 exp (01t)) + 1t
1
1 1 + 1t 0 1t 2
For the particular case of a rigid body mode without damping, the equation of motion (Equation 2.5.51) is reduced to
2.5.53
q = f t0 1 t +
For this case
1f 1t: 1t
(2.5.54)
The time integration is done in terms of the generalized coordinates, and the response of the physical variables is then immediately available by summation:
X q X "= " q X = q X R= R q ;
u=
where are the modes, " are the modal strain amplitudes, are the modal stress amplitudes, and R are the modal reaction force amplitudes corresponding to each eigenvector .
Initial conditions
At the beginning of the step initial displacements and initial velocities must be converted to equivalent values of the generalized coordinates, which can only be done exactly if the number of eigenvectors equals the number of degrees of freedom. Since this is usually not the case, the initial values of the generalized coordinate displacement and velocity are calculated as
q =
N M M MN x0 ; m
2.5.54
where m is the generalized mass for eigenvector , is the eigenvector, and xN are the initial displacements. 0 Similarly, for initial velocities
M MN
q = _
Base motion denition
M M MN xN : _0 m
Many linear dynamic problems involve nding the response of a structure to a base motion: a time history of displacement, velocity, or acceleration given for the points where the displacements of the structure are prescribed. In all cases these base motions are converted into an acceleration history. If velocities are given for tabular or equally spaced amplitude curve denitions, the acceleration is dened by the central difference rule
Procedures
If displacements are given, the acceleration is dened by the central difference rule
In the above expressions a superscript * indicates user-dened amplitude data. If the displacement or velocity history has nonzero values at time t 0, corrections to the above t. If velocities are given, the acceleration at acceleration histories are made at times t 0 and t t 0 is 3 3
=1 = 0 z 0 = 1t 0 : z_ 1t z_ If displacements are given, the accelerations at t = 0 and t = 1t are 3 05 3 z3 z0 = 12t and 1t = 21t 22 1t : 31t z z 1t z
uabs = z + x; _ _ uabs = z + x; _ uabs = z + x;
The response is calculated relative to the base. If total values of nodal variables are required, the motion at the base is added to the relative values:
where
Reference
Transient modal dynamic analysis, Section 6.3.7 of the ABAQUS Analysis Users Manual
2.5.55
2.5.6
Response spectrum analysis is intended to provide an inexpensive approach to estimating the peak response of a model (usually a model of a structure) subjected to base motion: the simultaneous motion of all nodes xed with boundary conditions. The approach assumes that the systems response is linear, so it can be analyzed in the frequency domain using its lowest eigenmodes8and eigenfrequencies ! extracted in a previous eigenfrequency step. The method is typically used to estimate the response of a building or of a piping system in a building to an earthquake. The method is not appropriate if the excitation is so severe that nonlinear effects in the system are important. In such a case the time history of the base excitation must be known and used with a dynamic analysis step to obtain the systems response. Even for a linear system the response spectrum method provides only estimates of the peak response. If more precise values are required, a modal dynamic analysis step can be used to integrate the system through time and, thus, develop its response to the given base excitation. In response spectrum analysis the estimates of peak values are obtained by combining the peak responses of the participating modes corresponding to user-specied spectra denitions. Several approximations are introduced by response spectrum analysis. The conversion from a time history of excitation into an equivalent frequency domain spectrum is based on the behavior of a single degree of freedom system. Different spectra are often applied in different excitation directions. Once the spectra are known, the peak modal responses can be calculated. The manner in which these peak modal responses are combined to estimate the peak physical response, together with the manner in which multidirectional excitations are combined, introduces approximations. Since no one method gives good approximations for all cases, several methods are offered. These methods are discussed in the Regulatory Guide 1.92 (1976) of the U.S. Nuclear Regulatory Commission, in the papers by Anagnastopoulos (1981), Der Kiureghian (1981), and Smeby (1984) and in the book by A. K. Gupta (1990). The choice of the summation rules depends on the particular case and is a matter of the users judgment. Since response spectrum analysis is commonly used as a basic design tool, spectra are dened in many design codes for such applications as seismic analysis of buildings. In such cases the user works from the given spectra. In other cases the time history of a known base excitation must rst be converted into a response spectrum by considering the response of a single degree of freedom system excited by the known base motion. For this purpose the single degree of freedom system is characterized by its undamped natural frequency, !, and the fraction of critical damping present in the system, . The equation of motion of the system is integrated through time to nd peak values of relative displacement, relative velocity, and absolute acceleration. The integration described in Modal dynamic analysis, Section 2.5.5, can be used for this purpose, since it is exact when the base motion record varies linearly with time. Thus, the maximum values of displacement, velocity, and acceleration are found for the linear, one degree of freedom system. This process is repeated for all frequency and damping values in the range of interest to construct displacement, velocity, and acceleration spectra, S D (!; ), S V (!; ); and S A (!; ). A FORTRAN program to build spectra in this way from an acceleration record is given in Analysis of a cantilever subject to earthquake motion, Section 1.4.13 of the ABAQUS Benchmarks Manual (le cantilever_spectradata.f.)
Procedures
2.5.61
S D , S V , and S A is given by
1 1 SD = SV = 2 SA : ! !
In ABAQUS/Standard it is assumed that the damping is always small, so these relationships are used whenever a conversion is needed. A response spectrum is dened by giving a table of values of S at increasing values of frequency, !, for increasing values of damping, . Linear interpolation on a logarithmic scale is used to compute the response for any required frequency and damping factor. Any number of spectra can be dened. The response spectrum procedure allows up to three spectra, which we denote by k, k = 1; 2; 3, to be applied to the model in orthogonal physical directions dened by their direction cosines, tk . These spectra can come from different excitations (with a certain level of correlation between them), or they can be components of a single base excitation acting in an arbitrary direction. When modal methods are used to dene a models response, the value of any physical variable is dened from the amplitudes of the modal responses (the generalized coordinates), q. The rst stage in the response spectrum procedure is to estimate the peak values of these modal responses. For mode and spectrum k this is
j D (!; ) is the kth displacement spectrum, tk is the j th where ck is a user-dened scaling parameter, Sk j direction cosine for the kth spectrum, and 0j is the participation factor for mode in direction j (see Variables associated with the natural modes of a model, Section 2.5.2, for the denition of 0j ). Similar _max expressions for (q )k and ( )k are obtained by using velocity or acceleration spectra in the above qmax
max D (q ) k = c k S k
Xt 0
k j j ;
formula. We now have estimates of the peak responses of the generalized coordinatesthe amplitudes of the responses of the natural modes of the system for excitation in each direction. If the input spectra in the different directions are components of a single base excitation acting in an arbitrary direction, for each mode we can combine these peak responses into a single value by specifying algebraic summation of the values for the different spatial directions:
max q =
X (q
k
max )k :
In this case the modal combinations discussed below still apply, but the subscript k is no longer relevant and should be ignored. 0 1max the peak response of some physical variable Ri (a component of Let us denote by Ri (!; ) k displacement, stress, section force, reaction force, etc.) caused by motion in the natural mode excited by the response spectrum in excitation direction k at frequency ! and with damping . Denote the component of the eigenvector associated with Ri by 8i . Then
2.5.62
and
We need to combine these estimates of the peak physical responses in the individual modes into 1max 0 estimates of the total peak response of the particular physical variable to the given spectrum, Ri (!; ) k . Since the peak responses in the different modes will not in general occur at the same time, this combination is only an estimate, so several formul are offered, as follows: Summation of the absolute values of the modal peak responses estimates
(Ri )max = k
X (Ri )max :
Procedures
This provides the most conservative estimate of the peak response, since it assumes that all modes provide peak response in phase at the same time. Square root of the summation of the squares (SRSS) estimates
(Ri )max k
X0
(Ri )max 2 : k
This summation usually provides a reasonable estimate if the natural frequencies of the modes are well separated. The Naval Research Laboratory method distinguishes the mode, , in which the physical variable has its maximum response, and adds the square root of the sum of squares of the peak responses in all other modes to the absolute value of the peak response of that mode. This gives the estimate
sX
6=
(Ri )max 2 : k
Again, the modes must be reasonably well-spaced in the frequency domain to obtain an accurate estimate with this method. A variety of methods are available that aim to improve the estimation for structures with closely spaced frequencies. ABAQUS/Standard provides two of them: the Ten Percent method recommended by Regulatory Guide 1.92 (1976) of the U.S. Nuclear Regulatory Commission and the complete quadratic combination method, which was rst introduced by Der Kiureghian (1981) and developed by Smeby and Der Kiureghian (1984). Both methods reduce to the SRSS method if the modes are well separated with no coupling among them. The Ten Percent method described in Regulatory Guide 1.92 modies the SRSS method by adding a contribution from all pairs of modes and whose frequencies are within 10% of each other, giving the estimate s
(Ri )max = k
((Ri )max )2 + 2 k
<
The complete quadratic combination method (CQC) combines the modal response with the formula
(
Ri )max = k
sX X
where are cross-correlation coefcients between modes frequencies and modal damping between the two modes:
=
where
2 0 r
2
2 ( + r ) r
+4
r
2 1 +
+4
2 2 2 + r
If double eigenvalues occur with the same damping coefcient, their correlation coefcient will be = 1. If modes are well-spaced, their cross-correlation coefcient will be small ( << 1) and the method will give the same results as the SRSS method. This method is usually recommended for asymmetrical building systems, since, in such cases, other methods can underestimate the response in the direction of motion and overestimate the response in the transverse direction (see Response spectra of a three-dimensional frame building, Section 2.2.3 of the ABAQUS Example Problems Manual). For the case of different base excitations acting along orthogonal directions, we still need to sum over the directions indicated by the subscript k. Regulatory Guide 1.92 species that this directional summation be based on the square root of the sum of the squares summation rule:
(
! r = : !
Ri )max =
s X0
Ri )max 2: k
This rule is appropriate when the base motions in different directions are statistically independent (uncorrelated)when they are acting along principal directions. The existence of a set of directions along which the ground motions can be considered uncorrelated is discussed by Penzien and Watabe (1975). Such considerations are especially important when the CQC modal combination method is used. In the ABAQUS implementation of the CQC method it is assumed that the horizontal components act along the principal directions and are of equal intensity. For details on the CQC method applied to more general cases, see Smeby and Der Kiureghian (1984).
Reference
Response spectrum analysis, Section 6.3.10 of the ABAQUS Analysis Users Manual
2.5.64
STEADY-STATE DYNAMICS
2.5.7
Steady-state linear dynamic analysis predicts the linear response of a structure subjected to continuous harmonic excitation. In many cases steady-state linear dynamic analysis in ABAQUS/Standard uses the set of eigenmodes extracted in a previous eigenfrequency step to calculate the steady-state solution as a function of the frequency of the applied excitation. ABAQUS/Standard also has a direct steady-state linear dynamic analysis procedure, in which the equations of steady harmonic motion of the system are solved directly without using the eigenmodes, and a subspace steady-state linear dynamic analysis procedure, in which the equations are projected onto a subspace of selected eigenmodes of the undamped system. These options are intended for systems in which the behavior is dependent on frequency, for when the model includes damping, or for systems in which the governing equations are not symmetric. This section describes the linear steady-state response procedure based on the eigenmodes. The projection of the equations of motion of the system onto the th mode gives
2 q + cq + ! q = _
Procedures
where q is the amplitude of mode (the th generalized coordinate), c is the damping associated with this mode (see below), ! is the undamped frequency of the mode, m is the generalized mass associated with the mode, and (f1 + if2 ) exp(i
t) is the forcing associated with this mode. The forcing is dened N N by the frequency,
, and the real and imaginary parts of the nodal equivalent forces, F1 and F2 , projected N: onto the eigenmode
1 (f + if ) exp(i
t); 2 m 1
(2.5.71)
In this equation summation is implied by the repeat of the superscript N indicating a degree of freedom in the model; but throughout this section we are working with only a single modal equation, so no summation is implied by the repeat of the mode subscript . The load vector is written in terms of its real and imaginary N N parts, F1 and F2 , since this is the manner in which the loading is dened in ABAQUS/Standard (as load N case 1 and load case 2). It is equivalently possible to write the loading in terms of its magnitude, F0 and N = F N exp i(
t + 9), where F N = F cos 9 and F N = F sin 9. phase, 9, as F 0 0 0 1 2 Several representations of modal damping are provided. Modal damping denes c = 2 !, where is the fraction of critical damping in the mode. Structural damping gives a damping force proportional to the modal amplitude:
2 c q = is !q; _
coefcient for the mode. Rayleigh damping is dened by Rayleigh coefcients damping low and high frequency modes, respectively. Rayleigh damping can be reproduced exactly by modal damping as
! = + : 2! 2
Introducing all of these damping denitions into Equation 2.5.71 gives
2.5.71
STEADY-STATE DYNAMICS
2 2 2 q + 2 !q + + ! q + is ! q + !q = _ _
1 (f + if ) exp (i
t): 1 2
(2.5.72)
where f0 = (f1)2 + (f2 )2 is the amplitude of the projected load vector and H0(
) is the amplitude of the complex transfer function for mode that denes the response in mode from the force projection onto that mode and is dened by its real and imaginary parts as
2 f 1 ! 0
2 + 2 f22
2 0
2)2 + (
)2 (! (! 0
)2 + (
)2 # " 0 2 1 f 2 ! 0
2 f 1
1 0 =(H) = m ; + 2 2 (! 0
2 )2 + (
)2 (! 0
2 )2 + (
)2
q = H0f0 exp i(
t + 9);
(2.5.73)
1 <(H) = m
"
s !2 2 = 2 ! + + ! + :
H0f0 = <(H)2 + =(H)2 f0 =
and the phase angle of the response is
1 f 0 ; 2 (! 0
)2 + (
)2
f 1 = 0 f 2 = 0
1 N M NM eM a exp (i
t); ^j 1j m 1 N M NM eM a exp (i
t); ^j 2j m
parts of the base acceleration. If the base motion is given as a velocity or displacement, the corresponding accelerations are a1 = 0
v1 , a2 = 0
v2 , where v1 and v2 are the real and imaginary parts of the velocity, or a1 = 0
2 u1 , a2 = 0
2 u2, where u1 and u2 are the real are imaginary parts of the displacement. The peak amplitude of any physical variable, uN , is available from the modal amplitudes as
NM is the structures mass matrix and eM is a vector that has unit magnitude in the direction of where M ^j the base acceleration at any grounded node and is otherwise zero; a1j and a2j are the real and imaginary
uN =
N q :
2.5.72
STEADY-STATE DYNAMICS
Steady-state response is given as a frequency sweep through a user-specied range of frequencies. Since the structural response peaks around the natural frequencies, a bias function is used to cluster the response points around the frequencies. The biasing is described in Mode-based steady-state dynamic analysis, Section 6.3.8 of the ABAQUS Analysis Users Manual.
Reference
Mode-based steady-state dynamic analysis, Section 6.3.8 of the ABAQUS Analysis Users Manual
Procedures
2.5.73
2.5.8
Random response linear dynamic analysis is used to predict the response of a structure subjected to a nondeterministic continuous excitation that is expressed in a statistical sense by a cross-spectral density (CSD) matrix. The random response procedure uses the set of eigenmodes extracted in a previous eigenfrequency step to calculate the corresponding power spectral densities (PSD) of response variables (stresses, strains, displacements, etc.) and, henceif requiredthe variance and root mean square values of these same variables. This section provides brief denitions and explanations of the terms used in this type of analysis. Detailed discussion of the theory of random response analysis is provided in the books by Clough and Penzien (1975), Hurty and Rubinstein (1964), and Thompson (1988). Examples of random response analysis are the study of the response of an airplane to turbulence; the response of a car to road surface imperfections; the response of a structure to noise, such as the jet noise emitted by a jet engine; and the response of a building to an earthquake. Since the loading is nondeterministic, it can be characterized only in a statistical sense. We need some assumptions to make this characterization possible. Although the excitation varies in time, in some sense it must be stationaryits statistical properties must not vary with time. Thus, if x(t) is the variable being considered (such as the height of the road surface in the case of a car driving down a rough road), then any statistical function of x, f (x), must have the same value regardless of what time origin we use to compute f : 0 1 0 1 f x(t) = f x(t + ) for any : We also need the excitation to be ergodic. This term means that, if we take several samples of the excitation, the time average of each sample is the same. These restrictions ensure that the excitation is, statistically, constant. In the following discussion we also assume that the random variables are real, which is the case for the variables that we need to consider.
Statistical measures
Procedures
We dene some measures of a variable that characterize it in a statistical sense. The mean value of a random variable x(t) is
E (x ) =
T !1 T
lim
Z T
0T 2
x(t) dt:
Since the dynamic response is computed about a static equilibrium conguration, the mean value of any dynamic input or response variable will always be zero:
E (x ) = 0 :
The variance of a random variable measures the average square difference between the point value of the variable and its mean:
2 r (x ) =
T !1 T
lim
Z T
2
0T 2
12
2.5.81
Since E (x) = 0 for our applications, the variance is the same as the mean square value:
E (x
) = lim
T !1 T
0T 2
x2(t) dt:
The units of variance are (amplitude)2 , so thatfor examplethe variance of a force has units of (force)2 . Generally we prefer to use the same units as the variable itself. Therefore, output variables p 2 in ABAQUS/Standard are given as root mean square (RMS) values, r (x) = r .
Correlation
Correlation measures the similarity between two variables. Thus, the cross-correlation between two random functions of time, x1 (t) and x2 (t), is the integration of the product of the two variables, with one of them shifted in time by some xed value to allow for the possibility that they are similar but shifted in time. (Such a case would arise, for example, in studying a car driving along a rough road. If the separation of the axles is d and the car is moving at a steady speed v, the back axle sees the same road prole as the front axle, but delayed by a time = d=v. Assume that the road prole moves each wheel which, in turn applies a force to the car frame through the suspension. If F2 is the force applied to the rear axle (as a concentrated load in ABAQUS) and F1 is the force applied to the front axle, F2 (t + d=v) = F1(t).) The cross-correlation function is, thus, dened as
Rx1 x2 ( ) =
T !1 T
lim
T
2
0T 2
x1(t)x2 (t + ) dt:
Since the mean value of any variable is zero, on average each variable has equal positive and negative content. If the variables are quite similar, their cross-correlation (for some values of ) will be large; if they are not similar, the product x1 x2 will sometimes be negative and sometimes positive so that the integral over all time will provide a much smaller value, regardless of the choice of .
x1(t) t
x2(t) t
2.5.82
A simple result is
R x1 x2 ( ) =
T !1 T T !1 T T !1 T
lim
Z Z Z
T
2
0T 2
T
2
= lim
0T 2
T
2
= lim =
R x2 x1 (0 ):
0T 2
For convenience the cross-correlation can be normalized to dene the nondimensional normalized cross-correlation: R ( ) : r x1 x2 ( ) = p 2 x1 x2 2 r (x 1 ) r (x 2 ) Thus, if x1 = x2, r(0) = 1; if x1 = 0x2, r(0) = 01. If x1 and x2 are entirely dissimilar, r ! 0. (When r = 0 the variables are said to be orthogonal.) Clearly 01 r 1, with small values of r indicating that x1 and x2 have quite different time histories. Now consider the cross-correlation of a variable with itself: the autocorrelation. Intuitively we can see that if the variable is very random, its autocorrelation will be very small whenever 6= 0: there will be no time shift that allows the variable to correlate with itself. However, if the variable is not so randomif it is just a vibration at a xed frequencythe autocorrelation will be close to 61 whenever is chosen to be some integer multiple of half the period of the vibration. Thus, the autocorrelation provides a measure of how random a variable really is. The autocorrelation of a variable x(t) is, therefore,
R ( ) =
T !1 T
lim
T
2
2 Clearly, as ! 0, R( ) ! r : the autocorrelation equals the variance (the mean square value). We can, therefore, also use the normalized autocorrelation:
0T 2
x(t)x(t + ) dt:
r ( ) =
Obviously R( ) is symmetric about
= 0:
R ( ) : 2 r
R(0 ) = R( );
and the value of R( ) never exceeds its value at
= 0:
2 R( ) R(0) = r :
The autocorrelation function of records with very similar amplitude over the wide range of frequencies drops off rapidly as increases. This kind of function is known as a wide band random function.
2.5.83
R()
Figure 2.5.82 Wide band noise record and its autocorrelation. The most extreme wide band random function would have an autocorrelation that is just a delta function: 2 r if = 0 R ( ) = ( ) = 0 if 6= 0. Such a function is called white noise. White noise has the same amplitude at all frequencies, and its autocorrelation is zero except at = 0.
R() z2
Figure 2.5.83 Autocorrelation of white noise. Let us now consider the opposite case, known as a narrow band function. The extreme case of such a function is a simple sinusoidal vibration at a single frequency: x = A sin !0t. Then R( ) 2 must also be periodic, since R( ) must attain the same value, R(0) = r , each time the shift, , corresponds to the period of vibration. Performing the integration through time,
R( ) = lim
T !1 T
=A
T !1 T
lim
cos !0
0T 2
2
T
2
T
2
sin !0 t cos !0 t dt
0T 2
sin !0 t dt + sin !0
0T 2
2.5.84
1 = A2 Tlim T !1 = 1 2
cos !0 =
1 2
0T 2
(1
T
2
1 2
0T 2
sin 2!0 t dt
A2 cos !0
2 r cos !0 :
Procedures
Figure 2.5.84
The autocorrelation, R( ), thus tells us about the nature of the random variable. If R( ) drops off rapidly as the time shift moves away from = 0, the variable has a broad frequency content; if it drops off more slowly and exhibits a cosine prole, the variable has a narrow frequency content centered around the frequency corresponding to the periodicity of R( ).
Narrow band response R() = ce -k || cos 0
Figure 2.5.85 Narrow band record and its autocorrelation. We can extend this concept to detect the frequency content of a random variable by crosscorrelating the variable with a sine wave: sweeping the wave over a range of frequencies and examining the cross-correlation tells us whether the random variable is dominated by oscillation at particular frequencies. We begin to see that the nature of stationary, ergodic random processes is best understood by examining them in the frequency domain.
2.5.85
As an illustration, consider a variable, x(t), which contains many discrete frequencies. We can write x(t) in terms of a Fourier series expanded in N steps of a fundamental frequency !0:
x (t) =
n=1
N X
The series begins with the n = 1 term because the mean of the variable must be zero. We can write this series more compactly as a complex Fourier series (keeping in mind that we will be interested only in the real part):
x (t) =
where An of An : and
=
<(An ) + i=(An ) is the complex amplitude of the nth term, A3 is the complex conjugate n
A 3 = < (A n ) 0 i = (A n ); n
(
(a
n=0N
N X
An exp(in!0 t);
An =
The variance of x(t) is
(a 0
r (x ) =
Z T
N X
r (x ) =
Z T
N X
where
xn = < An exp(in!0 t)
is the nth component of the Fourier series. Thus, thanks to the orthogonality of Fourier terms, the variance (the mean square value) of the series is the sum of the variances (the mean square values) of its components. In particular, we see that An A3 is the variance, or mean square value, of the variable at the frequency n!0. n 2 The contribution to the variance of x, r (x), at the frequency fn = n!0=(2), per unit frequency, is thus
S x (f n ) =
2
A A3 ; !0 n n
2.5.86
n=0N
N X
Sx (fn )1f:
As we examine x as a function of frequency, Sx fn tells us the amount of power (in the sense of mean square value) contained in x, per unit frequency, at the frequency fn . As we consider smaller and smaller intervals, f ! , Sx is the power spectral density (PSD) of the variable x:
( )
r (x ) =
2
01
Sx (f ) df;
Procedures
where f is the frequency in cycles per time (usually Hz). Notice that Sx has units of (variable)2 /frequency, where (variable) is the unit of the variable (displacement, force, stress, etc.). In this case frequency is almost always given in Hz, although since ABAQUS does not have any built-in unitsthe frequency could be expressed in any other units of cycles per time. However, Sx should not be given per circular frequency (radians per time): ABAQUS assumes that Sx f , not Sx ! .
()
()
Fourier transforms
Since the variables of interest in random response analysis are characterized as functions of frequency, the Fourier transform plays a major role in converting from the time domain to the frequency domain and vice versa. The Fourier transform of x t , which we write as x f , is dened by
x (t) =
or, in terms of the circular frequency !
()
()
01
x(f ) exp(i2ft) df
1 Z 120x(!) + x(0!)1 cos !t + i0x(!) 0 x(0!)1 sin !t3 d!; 2 0 1 0 so x(0 ) + x(0!) =(2) is the amplitude of the cosine term in x(t) at the circular frequency !, ! 1 while x(!) 0 x(0!) =(2) is the amplitude of the sine term. We, thus, see the physical meaning of the Fourier transformit provides the complex magnitude (the amplitude and phase) of the content of x(t) at a particular frequency. If x(t) is real only (which is the case for the variables we need to consider), this expression shows that we must have 0 1 0 1 = x(!) = 0= x(0!)
x (t) =
2.5.87
< x (! )
< x(0!) ;
x(0!) = x3 (!):
Z
x (f ) =
1 01
which shows that the transformations between the time and frequency domains are rather symmetrical. We now need Parsevals theorem:
Z
1 01
x1 (t)x3 (t) dt = 2
= =
Z01 1
x 3 (t) 2
01 Z 1 01
01 Z 1
x1 (f ) exp(i2ft) df dt
Z T
2
T !1 T
0T 2
1
01 T !1 T
=
lim
x3 (f ), and so
r (x ) =
By comparison,
1 01
1
S x (f ) =
We also see that
T !1 T
lim
S x (0 f ) = S x (f ):
2.5.88
Z1 Z0 1
0
[ S x (f ) + ~ Sx (f ) df;
Sx (0f )] df
for f
0.
Procedures
Since Sx (0f )
Sx (f ), we see that
~ S x (f ) = 2 S x (f ) :
R ( ) =
= lim
T !1 T
1
lim
T
2
0T 2
x (t)
Z
x(t)x(t + ) dt
Z
1 01
Z
=
T !1 T
0T 2
1
lim
01
Sx (f ) exp(i2f ) df;
(using the result above). Thus, the power spectral density is the Fourier transform of the autocorrelation function. The inverse transform is
S x (f ) =
Since R( ) is symmetric about
=0
S x (f ) =
01
0 Z 0
=2
R ( )
0i2f ) + exp(i2f )
d
so that
~ S x (f ) = 4
2.5.89
Cross-spectral density
Following a similar argument to that used above to develop the idea of the power spectral density, we can dene the cross-spectral density (CSD) function, Sx1 x2 (f ), which gives the cross-correlation between two variables, Rx1x2 ( ), as
R x1 x2 ( ) =
with the inverse transformation
Z 1
01
S x1 x2 (f ) =
Z
Z 1
01
T
2
R x1 x2 ( ) =
= lim 1
T !1 T 0 T x1(t)x2 (t + ) dt 2
lim
T
2
Z
=
T !1 T 0 T 2
1
lim 1
x 1 (t)
Z
Z 1
Z
=
01
x2 (f ) exp i2f (t + ) df dt
By comparison,
T !1 T x1 (f )x2(f ):
1
S x1 x2 (f ) =
Tlim T 2 !1 3 = S x2 x 1 ( f ) = S x 2 x 1 ( 0 f ) :
x3 (f )x1 (f )
3
The general concept of random response analysis is now clear. A system is excited by some random loads or prescribed base motions, which are characterized in the frequency domain by a matrix of cross-spectral density functions, SNM (f ). Here we think of N and M as two of the degrees of freedom of the nite element model that are exposed to the random loads or prescribed base motions. In typical applications the range of frequencies will be limited to those to which we know the structure will respondwe do not need to consider frequencies that are higher than the modes in which we expect the structure to respond.
2.5.810
The values of SN M (f ) might be provided by Fourier transformation of the cross-correlation of time records or by the Fourier transformation of the autocorrelation of a single time record, together with known geometric data, as in the case of the car driving along a roughly grooved road, where the autocorrelation of the road surface prole, together with the speed of the car and the axle separation, allow S12 (f ) to be dened for the front (1) and rear (2) axles, as shown above. (If, in this case, the road prole seen by the wheels on the left side of the car is not similar to that seen by the wheels on the right side of the car, the cross-correlation of the left and right road surface proles will also be required to dene the excitation.) The system will respond to this excitation. We are usually interested in looking at the power spectral densities of the usual response variablesstress, displacement, etc. The PSD history of any particular variable will tell us the frequencies at which the system is most excited by the random loading. We might also compute the cross-spectral densities between variables. These are usually not of interest, and ABAQUS/Standard does not provide them. (They might be needed if the analysis involves obtaining results that, in turn, will dene the loading for some other system. For example, the response of a building to seismic loading might be used to obtain the motions of the attachment points for a piping system in the building so that the piping system can then be analyzed. The only option would be to model the entire system together.) An overall picture is provided by looking at the variance (the mean square value) of any variable; the RMS value is provided for this purpose. The RMS value is used instead of variance because it has the same units as the variable itself. ABAQUS/Standard computes it by integrating the single-sided power spectral density of the variable over the frequency range, since
Procedures
r
pR
sZ
(0) =
1
0
~ Sx (f ) df:
This integration is performed numerically by using the trapezoidal rule over the range of frequencies specied for the random response step:
r
v u u10 t S (f ~ 1 2
2
0 f1 ) +
N 01 X i=2
0 fN 01 ) ;
~ ~ ~ ~ where S1 denotes Sx at the user-dened lower frequency range f1 , Si is the Sx at the frequency fi , and N is the number of points at which the response was calculated. N will depend on the number of eigenmodes used in the superposition and on the user-specied number of points between the eigenfrequencies. The user must ensure that enough frequency points are specied so that this approximate integration will be sufciently accurate. The transformation of the problem into the frequency domain inherently assumes that the system under study is responding linearly: the random response procedure is considered as a linear perturbation analysis step. What remains, then, is for us to consider how ABAQUS/Standard nds the linear response to the random excitation.
2.5.811
Random response is studied in the frequency domain. Therefore, we need the transformation from load to response as a function of frequency. Since the random response is treated as the integration of a series of sinusoidal vibrations, this transformation is based on the same steady-state response function used for a steady-state dynamic analysis and described in Steady-state linear dynamic analysis, Section 2.5.7. The discrete (nite element) linear dynamic system has the equilibrium equation
uN M N M uM + C NM uM + K NM uM _
= uN F N ;
where M NM is the mass matrix, C NM is the damping matrix, K NM is the stiffness matrix, F N are the external loads, uN is the value of degree of freedom N of the nite element model (usually a displacement or rotation component, or an acoustic pressure), and uN is an arbitrary virtual variation. We project the problem onto the eigenmodes of the system. To do this the modes are rst extracted from the undamped system:
0
2 M NM ! 0 K NM N
= 0:
(Here, and throughout the remainder of this section, repeated subscripts and superscripts are assumed to be summed over the appropriate range except when they are barred, like above. Roman superscripts and subscripts indicate physical degrees of freedom; Greek superscripts and subscripts indicate modal variables.) Typically the structural dynamic response is well represented by a small number of the lower modes of the model, so the number of modes is usually O(101 )O(102), while the number of physical degrees of freedom might be O(103)O(105). The eigenmodes are orthogonal across the mass and stiffness matrices:
N M NM M
=
m k
0
N K NM M =
if = if 6= ; if = if 6= .
(The eigenmodes are normalized so that the largest entry in N is 1.0. Thus, m 6= 1 in general. In contrast, some codes normalize the eigenmodes so that m = 1.) We assume that any damping is in the general form of Rayleigh damping:
C NM = M NM + K NM
so that C NM will also project into a diagonal damping matrix c. The problem, thus, projects into a set of uncoupled modal response equations,
m q + c q + k q = f ; _
2.5.812
where
is the generalized load for mode and q is the generalized coordinate (the modal amplitude) for mode . Steady-state excitation is of the form
f = N F N
f = A exp(i2ft)
and creates response of similar form that we write as
Procedures
q = HA exp(i2ft);
where H (f ) is the complex frequency response function dened in Steady-state linear dynamic analysis, Section 2.5.7.
Response development
F Random loading is dened by the cross-spectral density matrix SNM (f ), which links all loaded degrees of freedom (N and M ). Projecting this matrix onto the modes provides the cross-spectral density function for the generalized (modal) loads: f F S (f ) = N SNM (f )M :
The complex frequency response function then denes the response of the generalized coordinates as
Finally, the response of the physical variables is recovered from the modal responses as
q u SNM (f ) = N S (f )M
so that the power spectral density of degree of freedom uN is
q u SN N (f ) = N S (f )N :
The PSDs for the velocity and acceleration of the same variable are
u _ u SN N (f ) = (2f )2SN N (f )
and
u u SN N (f ) = (2f )4 SN N (f ): 1)O(102) eigenmodes but many more (O(103 )O(105 )) Recall that we may typically have O(10
physical degrees of freedom. Therefore, if many of the physical degrees of freedom are loaded (as in
2.5.813
the case of a shell structure exposed to random acoustic noise), it may be computationally expensive to perform operations such as
f F S (f ) = N SNM (f )M ;
which involve products over all loaded physical degrees of freedom and must be done at each frequency in the range considered. In contrast, operations such as
q f 3 S (f ) = H(f )S (f )H (f )
are relatively inexpensive since they are done independently for each combination of modes. Forming
q u SN N (f ) = N S (f )N
may be expensive if we choose to compute the results for a large selection of physical variables (displacements, velocities, accelerations, stresses, etc.). ABAQUS/Standard will calculate the response only for the element and nodal variables requested. However, if a restart analysis is requested with the random response procedure, all variables are computed at the requested restart frequency, which can add substantially to the computational cost. The user is advised to write the restart le for the last increment only. To reduce the computational cost of random response analysis, ABAQUS/Standard assumes that the cross-spectral density matrix for the loading can be separated into a frequencydependent scalar function (containing the units of the CSD) and a set of coupling terms that are independent of frequency as follows:
F SNM (f ) =
X P (f ) X 9
J
2
33
IJ NM
for
for
!
N < M;
N > M; :
P J (f )
9IJ NN
Here J is the number of cross-correlation denitions included in the random response step. Each cross-correlation denition refers to an input complex frequency function, P J (f ). The spatial crosscorrelations are then dened by the complex set of values 9IJ . NM f With this approach the cross-spectral density function for the generalized loads, S (f ), can be constructed as
X X f S (f ) = P J (f ) N M 9IJ NM J I h i3 f f for N > M; S (f ) = S (f ) !
for
N < M;
f S (f ) = <
"
P J (f ) N M
9IJ NM
!#
2.5.814
Since the N M I IJ are not functions of frequency, they can be computed once only, leaving NM the frequency-dependent operations to be done only in the space of the eigenmodes. Although this procedure is not natural for typical correlated loadings (like road excitation or jet noise), loadings can always be dened this way by using enough cross-correlation denitions. The approach then reduces the computational cost for models with many loaded physical degrees of freedom. The approach works well for uncorrelated and fully correlated loadings, which are quite common cases.
Decibel conversion
P9
ABAQUS/Standard allows the user to provide an input PSD (say P f ) in decibel units rather than units of power/frequency. There are various ways to convert from decibel units to units of power/ frequency, depending on how the frequencies in one octave band are related to the frequencies in the next. A general formula relates the center (midband) frequencies fc between octaves as
( f c i) f (i01) c
()
Procedures
= 2x; =1
where the superscript i denotes i octave band and x is a chosen value. For example, x for = for one-third octave band conversion. ABAQUS/Standard full octave band conversion, and x uses full octave band conversion to convert from decibel units to units of power/frequency. For full octave band conversion, as shown by the above equation, the center frequency doubles from octave band to octave band. Since decibel units are based upon log scales, the center frequency for an octave band bounded by lower frequency fL and upper frequency fU is given by
()
th =1 3
Since
fU = fc 20:5x
and
fL = fc 200:5x:
1f = f 0 f = f [2 : x 0 20 : x]:
U L c 05 05
To convert from one type of conversion formula to the next, we need the following general decibel to power/frequency conversion equation:
db(f ) = 10 log
P (f ) ; P x =1 f
( ) ref
2.5.815
where Pref is a reference power value. The subscript x means that the power reference is given or one-third for the type of conversion represented by x (e.g., full octave band conversion for x = ). When x , we will simply use the notation Pref . Thus, since octave band conversion for x p and ABAQUS/Standard uses a full octave band conversion, f fc =
(x )
()
=1 3
=1
=1
P f
( )= 2
fc
1 =
:
Pref
10
db(f )=10
The PSD data can be given with respect to some other type of octave band frequency scale. In that case we can convert the PSD data at those frequencies coinciding with the full octave band scale by computing an equivalent full octave band reference power based on the following ratio:
Pref
(x ) P ref (1=3)
= p2[2 10 20 ]
0:5x 0:5x
For example, if we are given Pref (i.e., one-third octave band frequency scale), the equivalent full octave band reference power value would be
Pref
= 3 0536
:
(1=3) : ref
This conversion would be valid only at the one-third octave band center frequencies that coincide with the full octave band center frequencies. Thus, only every third data point should be considered.
Reference
Random response analysis, Section 6.3.11 of the ABAQUS Analysis Users Manual
2.5.816
BASE MOTIONS
2.5.9
Structures subjected to ground motion by earthquakes or other excitations such as explosions or dynamic action of machinery are examples in which support motions may have to be considered in the analysis of dynamic response. For modal-based dynamic analyses with modal dynamic, steady-state dynamic, or random response steps, the support motions are simulated by prescribed excitations called base motions that are applied to the suppressed degrees of freedom. The suppressed degrees of freedom are grouped into one or more bases. Multiple bases are required if base motions cannot be described by a single set of rigid body motions. A common case is that of a bridge whose supports are subjected to the same earthquake record but with a time shift. The degrees of freedom that are suppressed without being assigned to a named base make up the primary base, which typically is the only base if the motion can be described by a single set of rigid body motions. The suppressed degrees of freedom that are associated with named boundary conditions make up the secondary base or bases. ABAQUS/Standard uses different approaches to handle primary and secondary base motions. The modal participation method is used for primary base motions, and the big mass method is used for secondary base motions. Multiple bases can be used only in modal dynamic and steady-state dynamic analyses.
Primary base motions
Procedures
Let us consider structural motions relative to the base motion, ub . The total response, fut g, of the dynamic system will now consist of the relative response, fug, and the applied base motion excitation, fu b g:
fu t g = fug + fu b g;
with similar expressions for velocities and accelerations. Substituting of motion gives
The base acceleration is converted into applied inertia loads 0[M ]fubg. Here it has been assumed that there is no damping on rigid body modes (i.e., Rayleigh damping with 6= 0 is not allowed). If the prescribed excitation is given in the form of a displacement or a velocity, ABAQUS/Standard differentiates it to obtain the acceleration. The base motion vector can be expressed in terms of the rigid body mode vectors, fT gj , and time dependent base motion values, zj ; j = 1; 2; :::; 6:
fu b g =
X fT g z :
6
j =1 j j
qm +
6 cm k qm + m qm = 0 (0m )j zj ; _ mm mm j =1
2.5.91
BASE MOTIONS
where qm and fgm denote the relative generalized coordinate and mode shape for the mode cm and mm are modal stiffness, modal damping, and modal mass, respectively; and
m ; km ,
(0m )j =
1 fgT [M ]fT gj m mm
is the modal participation factor for mode m and degree of freedom j . Kinematic boundary conditions dened without being assigned a base name in an eigenfrequency step cannot be changed in any of the subsequent modal-based procedures. The kinematic constraints are built into the eigenvectors and into the participation factors for each mode, which implies that all degrees of freedom in the primary base must be subjected to the same rigid body motion. The participation factors are used to calculate the equivalent forcing function, and the equation of motion is solved for the relative quantities (such as relative displacements, relative velocities, and relative accelerationsoutput variables U, V, and A, respectively). To obtain total kinematic quantities (such as total displacements, total velocities, and total accelerationsoutput variables TU, TV, and TA, respectively), the primary base motions are added to the relative responses.
Secondary base motions
The base motion treatment described above cannot be applied to secondary bases. Instead, ABAQUS/Standard uses a big mass approach to simulate the motion of secondary bases. In this approach a big mass (much bigger than the total mass of the structure) is added to each degree of freedom in a secondary base during the eigenfrequency step. This generates additional low frequency modes associated with the masses Mbig . As more big masses are applied, more low frequency modes will be extracted in the frequency analysis step. To keep the number of frequencies of interest the same, the number of eigenvalues extracted is automatically increased. Hence, the size of the subspace will grow proportionally to the number of degrees of freedom associated with secondary bases. The desired base motion is obtained by applying a point force to each degree of freedom in the modal superposition step: where Mbig is the big mass and us N is the applied acceleration prescribed for degree of freedom N associated with secondary supports. Using the notation used in the equation of motion for primary base motions, the equation of motion for combined primary and secondary base motions is readily written as
Ps N = Mbig us N ;
i i ] is the diagonal matrix containing the big masses for secondary base i and fui g is the s where [Mbig base motion applied to this base. The mass matrix [M ] now contains the mass of the structure as
2.5.92
fP s g =
X[Mbig]fuisg; i
BASE MOTIONS
well as the big masses associated with the secondary bases. Projecting the equation of motion into the eigenspace (expanded by the low frequency modes) we obtain
6 c k 1 qm + m qm + m qm = 0 (0m )j zj + _ fgT fP g: mm mm mm m s j =1
Again, the quantities solved for are relative to the primary support, including those obtained at the secondary supports. The big masses should be chosen as large as possible to obtain accurate base motions but should not be so large as to cause excessive round-off errors or overows. To provide six digits of numerical accuracy, ABAQUS/Standard chooses each big mass equal to 106 times the total mass of the structure and each big rotary inertia equal to 106 times the total moment of inertia of the structure. The big masses, which are introduced in the eigenfrequency step, are not included in the model for other steps in a multiple step analysis. Hence, the total mass of the structure and the printed messages about masses and inertia of the entire model are not affected. However, the presence of the masses will be noticeable in the output tables printed in the eigenvalue extraction step, as well as in the information for the generalized masses and effective masses. See Double cantilever subjected to multiple base motions, Section 1.4.12 of the ABAQUS Benchmarks Manual, for an example of the use of the base motion feature.
References
Procedures
Transient modal dynamic analysis, Section 6.3.7 of the ABAQUS Analysis Users Manual Mode-based steady-state dynamic analysis, Section 6.3.8 of the ABAQUS Analysis Users Manual Random response analysis, Section 6.3.11 of the ABAQUS Analysis Users Manual Amplitude curves, Section 19.1.2 of the ABAQUS Analysis Users Manual
2.5.93
2.6.1
For structures subjected to continuous harmonic excitation, ABAQUS/Standard offers a direct steadystate dynamic analysis procedure in addition to the modal procedure described in Steady-state linear dynamic analysis, Section 2.5.7, and the subspace procedure described in Subspace-based steadystate dynamic analysis, Section 2.6.2. This procedure belongs to the perturbation procedures, where the perturbed solution is obtained by linearization about the current base state. For the calculation of the base state, the structure may exhibit material and geometrical nonlinear behavior as well as contact nonlinearities. Viscous damping can be included in the procedure, using the Rayleigh damping coefcients specied under the material denition. Discrete damping (such as dashpot elements) can be included. The procedure can also be used for coupled acoustic-structural medium analysis, including radiation boundary conditions and innite elements (as described in Coupled acoustic-structural medium analysis, Section 2.9.1); with piezoelectric medium (as described in Piezoelectric analysis, Section 2.10.1); and with viscoelastic material modeling (as described in Frequency domain viscoelasticity, Section 4.8.3). All properties can be frequency dependent. The formulation is based on the dynamic virtual work equation,
Procedures
where _ and are the velocity and the acceleration, is the density of the material, c is the mass proportional damping factor (part of the Rayleigh damping assumption), is the stress, is the surface " traction, and " is the strain variation that is compatible with the displacement variation . The discretized form of this equation is
u 1 u dV +
Z
V
_ c u 1 u dV +
" " : dV 0
u 1 t dS = 0;
(2.6.11)
N M NM uM + C(NM uM + I N 0 P N = 0; u m) _ Z
NN 1 NM dV N : dV
t
(2.6.12)
M NM = C(NM = m) IN = PN =
ZV ZV ZV
is the mass matrix; is the mass damping matrix; is the internal load vector; is the external load vector: (2.6.13)
c NN 1 NM dV
NN 1 t dS
For the steady-state harmonic response we assume that the structure undergoes small harmonic vibrations about a deformed, stressed state, dened by the subscript 0. Since steady-state dynamics belongs to the perturbation procedures, the load and response in the step dene the change from the base state.
2.6.11
1I N
Z h
1 N : + N : 1
dV:
The change in stress can be written in the form where el is the elasticity matrix for the material and c is the stiffness proportional damping factor (the other part of the Rayleigh damping assumption). The strain and strain rate changes follow from the displacement and velocity changes:
1" = M 1uM ;
This allows us to write Equation 2.6.12 as
1_ = M 1uM : " _
uN
(2.6.14)
@ N : 0 + N @uM
Z
: Del : M
dV
NM C (k )
N : Del : M dV: V c
where < uM and = uM are the real and imaginary parts of the amplitudes of the displacement, < P N 0 1 and = P N are the real and imaginary parts of the amplitude of the force applied to the structure is the circular frequency. Substituting the expressions for harmonic excitation and response in and Equation 2.6.14 and writing the result in matrix form yields
0 1
1P N = 0< 0P N 1 + i = 0P N 11 exp i
t;
where
2 3 2 3 0 1 1 0 < 2ANM 3 = 2 NM 3 < 0uM 1 = < 0 N 1 ; A P = ANM 0< ANM = uM 0= P N 2 3 < 2ANM 3 = K NM 0
2 M NM = ANM = 0
(C(NM + C(NM ): m) k)
(2.6.15)
2.6.12
Note that both the real and imaginary parts of AN M are symmetric. The procedure is activated by dening a direct-solution steady-state dynamic analysis step. Both real and imaginary loads can be dened. As output ABAQUS/Standard provides amplitudes and phases for all element and nodal variables at the requested frequencies. For this procedure all amplitude references must be given in the frequency domain.
Reference
Direct-solution steady-state dynamic analysis, Section 6.3.4 of the ABAQUS Analysis Users Manual
Procedures
2.6.13
2.6.2
For structures subjected to continuous harmonic excitation, ABAQUS/Standard offers a subspace steadystate dynamic analysis procedure in addition to the modal procedure described in Steady-state linear dynamic analysis, Section 2.5.7, and the direct procedure described in Direct steady-state dynamic analysis, Section 2.6.1. The procedure is activated by dening a subspace-based steady-state dynamic analysis step. This procedure is a perturbation procedure, where the perturbed solution is obtained by linearization about the current base state. For the calculation of the base state the structure may exhibit material and geometrical nonlinear behavior as well as contact nonlinearities. Viscous damping can be included in the procedure using the Rayleigh damping coefcients specied under the material denition. The procedure can also be used for viscoelastic material modeling (Frequency domain viscoelasticity, Section 4.8.3). Discrete damping (such as dashpots) can be included. The main advantage of this method is that it allows frequency-dependent behavior to be considered at a relatively small cost increase over the purely linear analysis via the modal procedure described in Steady-state linear dynamic analysis, Section 2.5.7. The discretized form of the linearized dynamic virtual work equation for a solid or structural system can be written as
Procedures
uN M N M uM
where the following denitions apply:
+C
NM uM + K NM uM _
0 PN
= 0;
(2.6.21)
M NM = K NM PN
= =
Z " @ N
V V St
N
N 1 NM dV
@uM
:
0 + N
Del : M
#
dV
is the mass matrix; is the stiffness matrix; is the external load vector;
NN 1 t dS u t u D
C NM is the damping matrix, _ and are the velocity and the acceleration, is the density of the material, 0 is the stress in the base state, is the surface traction, and el is the elasticity matrix for the material. We assume that both the stiffness K NM and the damping C NM are frequency dependent. The eigenfrequency step prior to the subspace-based steady-state dynamic analysis has extracted neig
eigenmodes of the undamped system using
(
0!2 M MN + K MN )N = 0;
where ! is the eigenfrequency in radians/time. The procedure assumes that the complex displacement changes for the damped system can be written in the form
uM
M a ;
= 1; :::; neig ;
(2.6.22)
2.6.21
where a are the complex modal amplitudes. Using Equation 2.6.22 in Equation 2.6.21 and premultiplying with , the equation of motion projected onto the subspace is provided:
M a + C (
)_ + K (
)a = P (
) a
where
; = 1; :::; neig ;
(2.6.23)
and
is the excitation frequency. Since the eigenmodes are not orthogonal to the damping and stiffness matrices in the equilibrium equations (because of the frequency-dependent properties), the projected damping and stiffness matrices are not diagonal. For harmonic excitation and response we can write
M = N M NM M ; K (
) = N K NM (
)M ; C (
) = N C NM (
)M ; P (
) = N P N (
);
a = (< (a ) + i = (a )) exp i
t
and
P = (< (P) + i = (P)) exp i
t;
where < (a ) and = (a ) are the real and imaginary parts of the modal amplitudes and < (P) and = (P) are the real and imaginary parts of the amplitude of the force applied to the structure after projection onto the subspace. Substituting the expressions for harmonic excitation and response in Equation 2.6.23 and writing the result in matrix form yields
< [A ] = [A ] = [A ] 0< [A ]
< (a ) = < (P ) ; = (a ) 0= (P)
where
< [A ] = K 0
2 M ; = [A ] = 0
C :
The equation is solved for the real and imaginary part of the complex modal amplitudes a , and Equation 2.6.22 can be used to compute the real and imaginary part of the nodal displacements. For a coupled acoustic-structural analysis the development is similar. However, in this case the damping matrix also contains volumetric drag, impedance boundary, radiating boundary, and uid-solid coupling effects. Although the original formulation of the coupled acoustic-structural eigenvalue extraction problem is unsymmetric, the equations are rearranged such that the uid-solid coupling effects are retained without making the problem unsymmetric. The resulting modes and frequencies are associated with the coupled, undamped uid-solid system. The subspace-based steady-state dynamic procedure will project the full, coupled, and damped acoustic-structural system of equations onto the space spanned by the set of coupled uid-solid modes. For an acoustic-only analysis the development is very similar to that of the solid-only case. The acoustic material damping equivalent, volumetric drag, is treated in a similar manner to solid material damping effects.
2.6.22
As output ABAQUS/Standard provides amplitudes and phases for all element and nodal variables at the requested frequencies. All amplitude references must be given in the frequency domain.
Reference
Subspace-based steady-state dynamic analysis, Section 6.3.9 of the ABAQUS Analysis Users Manual
Procedures
2.6.23
2.7.1
ABAQUS/Standard provides a specialized analysis capability to model the steady-state behavior of a cylindrical deformable body rolling along a at rigid surface. The capability uses a reference frame that removes the explicit time dependence from the problem so that a purely spatially dependent analysis can be performed. For an axisymmetric body traveling at a constant ground velocity and constant angular rolling velocity, a steady state is possible in a frame that moves at the speed of the ground velocity but does not spin with the body in the rolling motion. This choice of reference frame allows the nite element mesh to remain stationary so that only the part of the body in the contact zone requires ne meshing.
Kinematics of steady-state rolling
Procedures
The kinematics of the rolling problem are described in terms of a coordinate frame that moves along with the ground motion of the body. In this moving frame the rigid body rotation is described in a spatial or Eulerian manner and the deformation in a material or Lagrangian manner. It is this kinematic description that converts the steady moving contact eld problem into a purely spatially dependent simulation. We consider the case shown in Figure 2.7.11, where the ground velocity of the body is described in terms of a constant cornering motion.
T n X0
Xc
Figure 2.7.11 Constant cornering motion. at 0 , which The body is rotating with a constant angular rolling velocity ! around a rigid axle in turn rotates with constant angular velocity
around the xed cornering axis through point c . Hence, the motion of a particle at time t consists of a rigid rolling rotation to position , described by
T X Y
Y = R s (X 0 X 0 ) + X 0 ;
1
2.7.11
followed by a deformation to point x, and a subsequent cornering rotation (or precession) around n to position y so that y = R c 1 (x 0 X c ) + X c ;
_ _ _ = y = R c 1 (x 0 X c ) + R c 1 x :
To describe the deformation of the body, we dene a map (Y; t), which gives the position of point X at time t as a function of its location Y at time t so that x = (Y; t): It follows that
x
@ Y _ = @ Y 1 @@t + @ ; @t
@Y _ = R s 1 (X 0 X 0 ) = ! T 2 (Y 0 X 0 ): @t ^ _ ^ Noting that Rs = ! 1Rs = !T1Rs, and introducing the circumferential direction S = T2(Y0X0 )=R, where R = jY 0 X0j is the radius of a point on the reference body, the velocity of the reference body can be written as @ Y=@t = !R S, so that
x
where
where S _ with Rc
= S 1 Y is the distance-measuring coordinate along the streamline. Using this result, together ^ =
1 Rc =
^ 1 Rc , the velocity of the particle can be written as n
v
@ @ _ = !R @ Y 1 S + @ = !R @S + @ ; @t @t
@ =
n 2 (y 0 Xc ) + !R Rc 1 @S + Rc 1 @ : @t
@ =
2 (nn 0 I) 1 (y 0 Xc ) + 2!
R n 2 Rc 1 @S + 2
n 2 Rc 1 @ @t @ @ + !2 R2 Rc 1 @S + 2!R Rc 1 @S@t + Rc 1 @ : 2 @t2
2 2 2
To obtain expressions for the velocity and acceleration in the reference frame tied to the body, we use the transformations
vr
= RT 1 v and c
ar
= RT 1 a c
so that we obtain
vr
@ =
n 2 (x 0 Xc ) + !R @S + @ @t
2.7.12
and
ar
@ =
2 (nn 0 I) 1 (x 0 Xc ) + 2!
R n 2 @S + 2
n 2 @ @t @ @ + !2R2 @S + 2!R @S@t + @ : 2 @t2
vr
2 2 2
and
ar
@ =
n 2 (x 0 Xc ) + !R @S
2
The rst term in the last expression can be identied as the acceleration that gives rise to centrifugal forces resulting from rotation about n. Noting that !R@ =@S is a measure of velocity, the second term can be identied as the acceleration that gives rise to Coriolis forces. The last term combines the acceleration that gives rise to Coriolis and centrifugal forces resulting from rotation about T. When the deformation is uniform along the circumferential direction, this Coriolis effect vanishes so that the acceleration gives rise to centrifugal forces only. The velocity of the center of the body X0 (which must lie on the axis T) is
v0
@ @ =
2 (nn 0 I) 1 (x 0 Xc ) + 2!
R n 2 @S + !2 R2 @S : 2
Procedures
=
n 2 (X 0 0 X c )
since the motions due to rolling and deformation vanish on the axis. To obtain the expression for straight line motion, as shown in Figure 2.7.12, we move Xc far away from the center of the body x0 but keep v0 the same. In that case
! 0 and, hence, in the limit
v
@ = v0 + vr = v0 + !R @S ;
a
@ = ar = !2 R2 @S ; 2
2
v0
T X0
2.7.13
Inertia
5=0
a v dV :
5 = 0
2 + !
2
Z
2
V Z
(x 0 Xc ) 1 (nn 0 I) 1 v dV 0 2!
R
@ @S
V
Z
@v @S
@ 2 @S 1 v dV
dV
5 = 0
2 + !
2
V Z
dx
1 (nn 0 I) 1 v dV 0 2!
R 1
@v @S dV :
@d @S
2 @d 1 v dV @S
For straight line rolling only the last term in each expression needs to be taken into account.
Contact conditions
To obtain the contact conditions, we start with the expressions for velocity derived in the previous section. For points on the surface of the deformable body
vD @ =
n 2 (x 0 Xc ) + !R @S + @ ; @t
where n is the cornering axis (which must be normal to the rigid surface) and
is the cornering angular velocity around n. Assuming that the velocity of a point on the foundation (or rigid surface) is vR , the relative motion becomes
v @ = vD 0 vR =
n 2 r + !R @S + @ @t
0 vR ;
where r = x 0 Xc . This equation can be split into normal and tangential components. The rate of penetration is
_ = 0n 1 v = n 1 vR 0 !R n 1 @ 0 n 1 @ : @S @t For any point in contact n 1 @=@S = 0; hence, h h _ = n 1 vR 0 n 1 @ ; @t
2.7.14
= t 1 v =
t 1 (n 2 r) + !R t 1 _
where t ; are two orthogonal unit vectors tangent to the contact surface so that n , so For steady-state conditions @=@t
( = 1 2) _
@ @S
+ t 1 @ 0 t 1 vR; @t
=0
= t 1 2 t2 .
Procedures
= t 1 v =
t 1 (n 2 r) + !R t 1 _
Variations in
yield
@ 0 t 1 vR : @S
=
t 1 (n 2 r) + !R t 1 _
For straight line rolling we can replace
n 2 r by v0 so that we obtain
@ + !R t 1 @S 0 t 1 vR @ 0 t 1 vR : @S
@ 0 t 1 vR : @S
= t 1 v0 _
and
= !R t 1 _
2 2 eq = 1 + 2 ; is equal to the critical stress, crit = p, where 1 and 2 are shear stresses along t, is the friction coefcient, and p is the contact pressure. On the other hand, when eq < crit , no relative motion
To complete the formulation, a relationship between frictional stress and slip velocity must be developed. A Coulomb friction law is provided for steady-state rolling. The law assumes that slip occurs if the frictional stress,
occurs. The condition of no relative motion is approximated in ABAQUS by stiff viscous behavior
= s
; _v
where
is the tangential slip velocity and s is the stick viscosity, which follows from the relation
s = crit : crit _
The allowable viscous slip velocity is dened as a fraction of the circumferential velocity
where Ff is a user-dened slip tolerance. These expressions contribute to the standard virtual work contribution for slip,
5 =
and rate of virtual work for slip,
Z
A
dA;
d 5 =
Z
A
(d + d )dA:
Reference
Steady-state transport analysis, Section 6.4.1 of the ABAQUS Analysis Users Manual
2.7.16
2.8.1
A porous medium is modeled in ABAQUS/Standard by the conventional approach that considers the medium as a multiphase material and adopts an effective stress principle to describe its behavior. The porous medium modeling provided considers the presence of two uids in the medium. One is the wetting liquid, which is assumed to be relatively (but not entirely) incompressible. Often the other is a gas, which is relatively compressible. An example of such a system is soil containing ground water. When the medium is partially saturated, both uids exist at a point: when it is fully saturated, the voids are completely lled with the wetting liquid. The elementary volume, dV , is made up of a volume of grains of solid material, dVg ; a volume of voids, dVv ; and a volume of wetting liquid, dVw dVv , that is free to move through the medium if driven. In some systems (for example, systems containing particles that absorb the wetting liquid and swell in the process) there may also be a signicant volume of trapped wetting liquid, dVt. The total stress acting at a point, , is assumed to be made up of an average pressure stress in the wetting liquid, uw , called the wetting liquid pressure, an average pressure stress in the other uid, ua , and an effective stress, 3 , dened by
Procedures
3 = + uw + (1 0 )ua I:
(2.8.11)
Stress components are stored so that tensile stress is positive, but uw and ua are pressure stress values. This explains the sign in this equation. is a factor that depends on saturation and on the surface tension of the liquid/solid system (Wu, 1976). is 1.0 when the medium is fully saturated and between 0.0 and 1.0 in unsaturated systems when its value depends on the degree of saturation of the medium. There is very sparse experimental evidence of its dependence on saturation; and because of this lack of data, we simply assume that is equal to the saturation of the medium (we dene saturation later in this section). We simplify the model by assuming that the pressure applied to the nonwetting uid is constant throughout the domain being modeled, does not vary with time, and is small enough that its value can be neglected. This requires that the nonwetting uid can diffuse through the medium sufciently freely so that its pressure, ua, never exceeds the pressure applied to this uid at the boundaries of the medium, which remains constant throughout the process being modeled. The most common example where this simplication applies is a porous medium that is quite permeable to gas ow, in which the nonwetting uid is air, exposed to atmospheric pressure. The dimensions of the region modeled must not be so large that the gravitational gradient of atmospheric pressure causes a signicant change in the air pressure, and there can be no external event that provides a transient change in the air pressure. This assumption allows ua to be removed from the equation, provided that the corresponding loading term (for example, atmospheric pressure on the boundary of the medium) is also omitted from the equilibrium equations and that ua is small enough that its effect on the deformation of the medium is not important (or that deformation is measured from the state = 0ua I). This simplication reduces the effective stress principle to
3 = + uw I:
(2.8.12)
In the case where trapped uid is present in the system, we assume that the effective stress is made up of two components weighted according to the relative volume of trapped uid and porous material:
2.8.11
3 = (1 0 nt ) 0 nt ptI;
(2.8.13)
where is the effective stress in the porous material skeleton, pt is the average pressure stress in the trapped liquid, and nt is the ratio of trapped uid volume to total volume, as dened later in this section. We assume that the constitutive response of the porous medium consists of simple bulk elasticity relationships for the liquid and for the soil grains, together with a constitutive theory for the soil skeleton whereby is dened as a function of the strain history and temperature of the soil:
n def =
dVv dV
=1
0 dVg 0 dVt : dV dV
Using the superscript 0 to indicate values in some convenient reference conguration allows the porosity in the current conguration to be expressed as
n = 10
so that
0 n 0 nt J g 0 n 0 0 n0 = J ; t
J
def
=
(2.8.14)
where
dV dV 0
is the ratio of the mediums volume in the current conguration to its volume in the reference conguration,
def dV Jg = g dV 0
g
2.8.12
is the ratio of the current to reference volume for the grains, and
nt
def
dVt dV
is the volume of trapped wetting liquid per unit of current volume. def ABAQUS generally uses void ratio, e = dVv =(dVg + dVt), instead of porosity. relationships are readily derived as
e
Conversion
1+
1 1+e
(2.8.15)
Procedures
Saturation, s, is the ratio of free (untrapped) wetting liquid volume to void volume:
s
def
=
dVw dVv
def
=
dVw dV
sn:
The total volume of wetting liquid (free liquid plus trapped liquid) per unit of current volume is
nf
sn
nt :
References
Coupled pore uid diffusion and stress analysis, Section 6.7.1 of the ABAQUS Analysis Users Manual Geostatic stress state, Section 6.7.2 of the ABAQUS Analysis Users Manual
2.8.13
2.8.2
Equilibrium is expressed by writing the principle of virtual work for the volume under consideration in its current conguration at time t:
Z
V
" : " dV
def
Z
=
t1
v dS + ^ v dV; f
V
1
" where is a virtual velocity eld, " = sym(@ =@ ) is the virtual rate of deformation, is the true (Cauchy) stress, are surface tractions per unit area, and ^ are body forces per unit volume. For our system ^ will often include the weight of the wetting liquid,
v x
Procedures
only with the weight of the dry porous medium. Thus, we write the virtual work equation as
fw = (sn + nt)w g; where w is the density of the wetting liquid and g is the gravitational acceleration, which we assume to be constant and in a constant direction (so that, for example, the formulation cannot be applied directly to a centrifuge experiment unless the model in the machine is small enough that g can be treated as constant). f For simplicity we consider this loading explicitly so that any other gravitational term in ^ is associated
Z
where are all body forces except the weight of the wetting liquid. In a nite element model equilibrium is approximated as a nite set of equations by introducing interpolation functions. The notation used to indicate such discretization are those quantities with uppercase superscripts (for example, v N ), which represent nodal variables, with the summation convention adopted for the superscripts. The interpolation is assumed to be based on material coordinates in the material skeleton (a Lagrangian formulation). For simplicity, in this section we consider only the case where the problem has no internal constraints such as incompressibilityand the discretization is made entirely by approximating equilibrium: this results in the displacement (or stiffness) method. Mixed formulation (hybrid) elements are available for porous medium analysis with ABAQUS/Standard, but consideration of such formulations does not require any important extension of the development at this stage. The virtual velocity eld is interpolated by
" : " dV
Z
=
t v dS + f v dV +
1
(sn +
n t ) w
g v dV;
1
(2.8.21)
v = NN vN ;
where N (Si ) are interpolation functions dened with respect to material coordinates, Si . The virtual rate of deformation is interpolated as
2.8.21
= sym
@ N ; @
N x
although more general forms are used in some of the elements in ABAQUS. The virtual work equation is thus discretized as
V S V V where the vN are assumed to be independent. The term conjugate to vN on the left-hand side of this equation is referred internal force array, I N : Z N def N : dV : I = V N , is taken from the right-hand side: Likewise, the external force array, P
PN
(P N
def
=
vN
N : dV
N = v
Z
NN 1 t dS +
NN
f dV +
N (sn + nt )w N 1 g dV ;
to subsequently as the
NN t dS + NN f dV +
1
(sn + n )
t w NN 1 g dV
includes any dAlembert forces). Choosing each vN to be nonzero in turn expresses equilibrium as a balance of internal and external forces:
IN
0 P N = 0:
(2.8.22)
These discretized equilibrium equations, together with the continuity equation discussed in Continuity statement for the wetting liquid phase in a porous medium, Section 2.8.4, dene the state of the porous medium. The equilibrium equations are written at the end of a time increment when implicit integration is used and, for all but the simplest cases, they are nonlinear. Newtons method is often used for their solution. Also, small, linear perturbations of the system are sometimes of interest (an example is the small vibration problem). These considerations imply a need for the Jacobian matrix of the system, which denes the variation of each term in the equations with respect to the basic variables of the discretized problem, whichfor this caseare the nodal positions, xN (or, equivalently, the displacements xN 0 X N ), and the nodal wetting liquid pressure values, uN . Symbolically we write such a variation of a term, f say, as df , w meaning
df
def
=
From the variation of discretized equilibrium, Equation 2.8.22, the term dP N gives rise to the mass matrix (for the dAlembert forces) and the load stiffness matrix in the Jacobian. The load stiffness matrix is discussed in Chapter 3, Elements, and Chapter 6, Loading and Constraints, for particular load types. The load stiffness term associated with the weight of the wetting liquid is
@f dxN @xN
@f P P du : @uw
1 2
VJ
d J (sn + nt )w
NN 1 g dV;
2.8.22
where
def
dV = dV 0
is the ratio of volume in the current conguration to volume in the reference conguration. The term dI N is Z
dI N
=d
V
N : dV
The rst term includes d(J ), which is the variation of stress caused by variations in nodal positions and pore liquid pressure values. In a continuum sense (that is, before the spatial discretization of the solution variables) this term is dened by the effective stress principle and by the constitutive assumptions used for the material and is discussed in more detail below. Introducing the spatial discretization into the second term provides a contribution to the initial stress matrix. Since the effective stress, , is generally stored as components associated with spatial directions, the rotation of the material during an increment must be included in the formulation. This issue is discussed in detail in Rate of deformation and strain increment, Section 1.4.3; Stress rates, Section 1.5.3; State storage, Section 1.5.4; and Solid element formulation, Section 3.2.2. For the purpose of the present development we assume that the variation of stress is d(J ) = dr (J (1 0 nt )) 0 d(Jntpt I) + J (d
1 + 1 d
T ) 0 d(Juw ) I; where dr (J ) is the variation in effective stress associated with constitutive response in the material (that def is, caused by variations in the strain or other state variables) and d
= asym(@dx=@ x) is the spin of the material. Using this assumption, the Jacobian contribution from stress in the porous medium is dI N
V J
1 N : d(J ) + : d N
dV:
Procedures
V J
def
(2.8.23)
References
Coupled pore uid diffusion and stress analysis, Section 6.7.1 of the ABAQUS Analysis Users Manual Geostatic stress state, Section 6.7.2 of the ABAQUS Analysis Users Manual
2.8.23
2.8.3
A porous medium in ABAQUS/Standard is considered to consist of a mixture of solid matter, voids that contain liquid and gas, and entrapped liquid attached to the solid matter. The mechanical behavior of the porous medium consists of the responses of the liquid and solid matter to local pressure and of the response of the overall material to effective stress. The assumptions made about these responses are discussed in this section.
Liquid response
Procedures
For the liquid in the system (the free liquid in the voids and the entrapped liquid) we assume that
Kw () is the
0 0 "th = 3w ( 0 w ) 0 3w jI ( I 0 w ) w
is the volumetric expansion of the liquid caused by temperature change. Here w () is the liquids thermal expansion coefcient, is the current temperature, I is the initial temperature at this point 0 in the medium, and w is the reference temperature for thermal expansion. Both u=Kw and "th are w assumed to be small.
Grains response
The solid matter in the porous medium is assumed to have the local mechanical response under pressure
1 g p 1 + K suw + 1 0 n 0 n 0 "th ; (2.8.32) g 0 g t g where Kg () is the bulk modulus of this solid matter, s is the saturation in the wetting uid, and
is its volumetric thermal strain. Here g () is the thermal expansion coefcient for the solid matter 0 and g is the reference temperature for this expansion. 1 0 g =0 is assumed to be small. g It is important to distinguish Kg and g as properties of the solid grains material. The porous medium as a whole will exhibit a much softer (and generally nonrecoverable) bulk behavior than is indicated by Kg and will also show a different thermal expansion. These effects are partially structural, caused by the medium being made up of irregular grains in partial contact. They may also be caused by the system being only partially saturated, with the voids containing a mixture of relatively compressible gas and relatively incompressible liquid.
2.8.31
Liquid entrapment
Entrapment of liquid is associated with specic materials that absorb liquid and swell into a gel. A simple model of this behavior is based on the idealization of this gel as a volume of individual spherical particles of equal radius ra . Tanaka and Fillmore (1979) show that, when a single sphere of such material is fully exposed to liquid, its radius change can be modeled as
ra = r f 0 a
XN
N
a
exp
0 t N
f where ra is the fully swollen radius approached as t ! 1 and N , aN and N are material parameters. Tanaka and Fillmore also show the rst term in the series dominates, so the model can be simplied to f ra = ra 0 a1 exp
0 t
r f 0 ra ra = a _ : 1
When the gel particles are only partially exposed to liquid (in an unsaturated system), it seems reasonable to assume that the swelling rate will be lessened according to the level of saturation. Further, we assume that the gel will swell only when the saturation of the surrounding medium exceeds the f f dry dry effective saturation of the gel, 1 0 [(ra )3 0 (ra )3 ]=[(ra )3 0 (ra )3 ], where ra is the radius of a gel particle that is completely dry. We combine these into a simple, linear effect:
ra = _
f r a 0 ra 1
s01+
f a 0 (r a ) 3 f dry (r a ) 3 0 (r a ) 3
(r ) 3
!+
where hf i = f if f > 0, hf i = 0 otherwise. The packing density and swelling may cause the gel particles to touch. In that case the surface available to absorb and entrap liquid is reduced until, if the gel particles occupy the entire volume except for solid material, liquid entrapment must cease altogether. With ka gel particles per unit reference volume, the maximum radius that the gel particles can achieve before they must touch (in a face center cubic arrangement) is
t = ra def
n0 pJ 4 2k a
3
1 3 1 3
and the volume is entirely occupied with gel and solid matter when the effective gel radius is
s ra =
n0 J 4 ka
2.8.32
ra = _
rf 0 ra a s01+ 1
0 (r a ) 3 f dry (r a ) 3 0 (r a ) 3
f (r a ) 3
!+
1
t ra 0 ra s t ra 0 r a
2!
(2.8.33)
dVt = ht dV 0;
where
ht
def 4
= 3
3 dry r a 0 (r a ) 3 k a ;
Procedures
where ra (J; s) is dened by the integration of Equation 2.8.33. This entrapped liquid can be compressed by pressure so that, when the porous medium is under stress, we assume
dVt =
and thus
1
u 0 Kw ht J
+ "w
th
htdV 0 ;
nt =
dVt dV
uw 10 Kw
th + "w
(2.8.34)
Combining this with Equation 2.8.31 and neglecting small terms compared to unity then provides
w n ht : 0 t w
(2.8.35)
We assume that, in the initial state, the effective saturation of the gel is the same as the saturation of the surrounding medium:
0 f f dry r a = (r a ) 3 0 (r a ) 3 0 (r a ) 3
(1
0 s0 )
31 3
The constitutive behavior of the gel containing entrapped uid is given by the elastic bulk relationship
where pt is the average pressure stress in the gel uid and "vol is its volumetric effective strain.
Effective strain
From Equation 2.8.32 we see that the volumetric strain 0uw =Kg + "th represents that part of the g total volumetric strain caused by pore pressure acting on the solid matter in the porous medium and by thermal expansion of that solid matter. In addition, entrapment of liquid in the medium may cause an additional volume change ratio:
1+
dVt 0 dVt0 dV 0
= 1 + Jnt
0 n0 : t
2.8.33
Finally, " ms (s) is a saturation driven moisture swelling strain that represents the volumetric swelling of the solid skeleton in partially saturated ow conditions. This moisture swelling can be isotropic or anisotropic. The remaining part of the strain in the medium,
"
def
I
0 "ms (s)
(2.8.36)
is the strain that is assumed to modify the effective stress in the medium. That is, we assume
is dened for each particular model in Chapter 4, Mechanical Constitutive Theories. where Also, for the effective pressure stress of the uid entrapped in the gel,
s 3K g
uw ds 3Kg duw
ht 3Kw (1 + Jnt 0 n0 ) t
I duw;
(2.8.37)
d(Jntpt
I) = JntKw II :
" 0 d" ds
ms ds du : duw w
" d" +
s 3K g
uw ds 3Kg duw
ht 3Kw (1 + Jnt 0 n0 ) t
I duw
(2.8.38)
dI N
d 0 N : ( + uw du )I w uw ds s t 0 3K + 3K du + 3K (1 +hJn 0 n0 ) f(1 0 nt)D + ntKw IIg : I g g w w t t "ms ds duw + du f(1 0 nt )D + nt Kw IIg : d" ds w 3 " + : (d N + 2 N 1 d
) 0 N : Iuw I : d" dV:
(2.8.39)
Reference
Pore uid ow properties, Section 12.7 of the ABAQUS Analysis Users Manual
2.8.34
2.8.4
ABAQUS/Standard provides capabilities for particular cases of uid ow through a porous medium. These cases are associated with having a relatively incompressible wetting liquid present in the medium. The medium may be wholly or partially saturated, with this liquid. When the medium is only partially saturated the remainder of the voids is lled with another uid. An example is a geotechnical problem, with soil containing water and air: continuity is written for the water phase. The wetting liquid can attach to and, thus, be trapped by certain solid particles in the medium: this volume of trapped liquid attached to solid particles forms a gel. A porous medium is modeled approximately in ABAQUS by attaching the nite element mesh to the solid phase. Liquid can ow through this mesh. A continuity equation is, therefore, required for the liquid, equating the rate of increase in liquid mass stored at a point to the rate of mass of liquid owing into the point within the time increment. This continuity statement is dened in this section. It is written in a variational form as a basis for nite element approximation. The liquid ow is described by introducing Darcys law or, alternatively, Forchheimers law. The continuity equation is satised approximately in the nite element model by using excess wetting liquid pressure as the nodal variable (degree of freedom 8), interpolated over the elements. The equation is integrated in time by using the backward Euler approximation. The total derivative of this integrated variational statement of continuity with respect to the nodal variables is required for the Newton iterations used to solve the nonlinear, coupled, equilibrium and continuity equations. This expression is also derived in this section. Consider a volume containing a xed amount of solid matter. In the current conguration this volume occupies space V with surface S . In the reference conguration it occupied space V 0 . Wetting liquid can ow through this volume: at any time the volume of such free liquid (liquid that can ow if driven by pressure) is written Vw . Wetting liquid can also become trapped in the volume, by absorption into the gel. The volume of such trapped liquid is written Vt . The total mass of wetting liquid in the control volume is
Procedures
w dVw + dVt
Z
=
w (nw + nt ) dV;
where w is the mass density of the liquid. The time rate of change of this mass of wetting liquid is
d dt
Z
Z
=
w (nw + nt) dV
Z
The mass of wetting liquid crossing the surface and entering the volume per unit time is
w nw n 1 vw dS;
where vw is the average velocity of the wetting liquid relative to the solid phase (the seepage velocity) and n is the outward normal to S .
2.8.41
Equating the addition of liquid mass across the surface S to the rate of change of liquid mass within the volume V gives the wetting liquid mass continuity equation
V J dt
d0
J w nw
( + nt )
dV
=0
w nw n
1 vw dS:
Using the divergence theorem and because the volume is arbitrary, this provides the pointwise equation
J dt
d0
J w nw
uw
J dt
d0
J w nw
( + nt)1 dV +
uw
@x
1 (w nw vw ) dV = 0; 1 (w nw vw ) dV 0 = 0:
where uw is an arbitrary, continuous, variational eld. This statement can also be written on the reference volume: Z Z
V0
uw d0 dt J w nw
( + nt)1 dV 0 +
1 0
V0
uw J
@x
In ABAQUS/Standard this continuity statement is integrated approximately in time by the backward Euler formula, giving
Z
V0
uw
J w nw
( + nt) t+1t 0
J w nw
( + nt ) t
Z
dV 0
+ 1t
uw
V0
uw J
1 ( w n w v w ) @x
@
t+1t
dV 0
= 0;
0 Z
+ 1t
t+1t
J w nw
( + nt ) t
@
dV
uw
1 ( w n w v w ) @x
t+1t
dV
= 0:
We now drop the subscript t+1t by adopting the convention that any quantity not explicitly associated with a point in time is taken at t+1t. The divergence theorem allows the equation to be rewritten as
Z
uw
+ 1t
w Z
uw
w 0 w
nw n
1 vw dS = 0;
wherefor conveniencewe have normalized the equation by the density of the liquid in the reference conguration, 0 . w
2.8.42
Since nw
= sn, this is the same as 1 w J (sn + nt) 0 1t w sn @ uw 1 vw dV w uw 0 (sn + nt ) 0 w J 0 0 @x w w t Z w + 1t uw 0 snn 1 vw dS = 0:
S w
(2.8.41)
Constitutive behavior
The constitutive behavior for pore uid ow is governed either by Darcys law or by Forchheimers law. Darcys law is generally applicable to low uid ow velocities, whereas Forchheimers law is commonly used for situations involving higher ow velocities. Darcys law can be thought of as a linearized version of Forchheimers law. Darcys law states that, under uniform conditions, the volumetric ow rate of the wetting liquid through a unit area of the medium, sn w , is proportional to the negative of the gradient of the piezometric head (Bear, 1972):
Procedures
sn w
where
b v = 0k 1 @ ; @x
u = z + gw ;
w
where z is the elevation above some datum and g is the magnitude of the gravitational acceleration, which acts in the direction opposite to z . On the other hand, Forchheimers law states that the negative of the gradient of the piezometric head is related to a quadratic function of the volumetric ow rate of the wetting liquid through a unit area of the medium (Desai, 1975):
sn w (1 +
where ; e is a velocity coefcient (Tariq, 1987). This nonlinear permeability can be dened to be dependent on the void ratio of the material. We see that, as the uid velocity tends to zero, , the two ow laws are identical. Forchheimers law approaches Darcys law. Also, if b can be anisotropic and is a function of the saturation and void ratio of the material. b has units of velocity (length/time). [Some authors refer to b as the hydraulic conductivity (Bear, 1972) and dene the permeability as
(x )
pv 1 v ) = 0k 1 @ ; b w w @x
=0
b b K = (1 + p1vw 1 vw ) k; g
where is the kinematic viscosity of the uid (the ratio of the uids dynamic viscosity to its density).] We assume that g is constant in magnitude and direction, so
@ @
1 @uw 0 g ; x = gw @x w
2.8.43
where g = 0g@z=@ x is the gravitational acceleration (we assume that w varies slowly with position). The permeability of a particular uid in a multiphase ow system depends on the saturation of the phase being considered and on the porosity of the medium. We assume these dependencies are separable, so b k = ks k;
def
where ks s provides the dependency on saturation, with ks : and k x; e is the permeability of the fully saturated medium. The function ks can be dened by the user. Nguyen and Durso (1983) observe that, in steady ow through a partially saturated medium, the permeability varies with s3 . We, therefore, take ks s3 by default. Introducing the ow constitutive law allows the mass continuity equation (Equation 2.8.41) to be written
()
(1) = 1 0
( )
Z
1 w J (sn + nt) uw w (sn + nt )0 0 w J 0 V w t @ uw ks + 1t 0 g(1 + pvw 1 vw ) @x 1 k 1 @uxw @ Zw w + 1t uw 0 snn 1 vw dS = 0: S w
0 w g
dV
(2.8.42)
The bulk behavior of the grains was discussed in Constitutive behavior in a porous medium, Section 2.8.3. From Equation 2.8.32,
g 0 g
1 = J1 1 + K uw + 1 0 np0 n 0 "th: g g g t
Combining this with Equation 2.8.14 and neglecting all but rst-order terms in small quantities, we obtain u p w 0 0 th 0 0 n 0 n t K 0 "g n 0 nt : Kg J g
+ 1 (1
Using Equation 2.8.31 and again neglecting second-order terms in small quantities, we obtain
1 w p n 1 0 nt 0 (1 0 n0 0 n0 ) + + uw t 0 J Kg w
1 0 n
Combining this result with Equation 2.8.34, and again approximating to rst-order in small quantities,
2.8.44
1 w p n 1 0 (1 0 n0 0 n0 + ht ) + + uw t 0 J Kg w
Kw
(1
0 n 0 0 n0 ) t
J
1 "th + (1 w
0n 0
0
Kg
0K
(2.8.43)
nt )("th w
0
"th ): g
Saturation
Because uw measures pressure in the wetting liquid and we neglect the pressure in the other uid phase in the medium (see Effective stress principle for porous media, Section 2.8.1), the medium is fully saturated for uw > 0. Negative values of uw represent capillary effects in the medium. For uw < 0 it is known (see, for example, Nguyen and Durso, 1983) that, at a given value of capillary pressure, 0uw , the saturation lies within certain limits. Typical forms of these limits are shown in Figure 2.8.41.
Procedures
exsorption
scanning absorption
0.0
1.0
saturation
Figure 2.8.41 Typical liquid absorption and exsorption behavior. We write these limits as sa s se , where sa (uw ) is the limit at which absorption will occur (so _ _ that s > 0), and se (uw ) is the limit at which exsorption will occur, and thus s < 0. We assume that
2.8.45
these relationships are uniquely invertible and can, thus, also be written as ua s during absorption w and ue s during exsorption. We also assume that some wetting liquid will always be present in the w medium: s > . Bear (1972) suggests that the transition between absorption and exsorption and vice versa takes place along scanning curves. We approximate these with a straight line, as shown in Figure 2.8.41. Saturation is treated as a state variable that may have to change if the wetting liquid pressure is outside the range for which its value is admissible according to that actual data corresponding to Figure 2.8.41. The evolution of saturation as a state variable is dened as follows. Assume that the saturation at time t, st, is known. It must satisfy the constraints
()
()
otherwise: We solve the continuity equation for uw t+1t , initially assuming s t+1t = s t . We then obtain s t+1t
st = 1:0
if
uw > 0:0;
s a (u w t ) s t s e (u w t )
if uw t+1t > 0:0 then s t+1t = 1:0 and s_ = 0:0; else if uw t+1t > ua (st) then s t+1t = sa (uw t+1t) and w
ds duw ds duw
= =
else if (
uw t+1t < ue st w
( ) then
s t+1t
= (
se uw t+1t
) and
dse duw uw
dsa duw uw
+1t
; ;
otherwise
s t+1t = s t +
where ds=duw s is the slope of the scanning line. These choices are shown in Figure 2.8.42.
Jacobian contribution
ds 1u duw s w
and
ds duw
ds ; duw s
+1t
The Jacobian contribution from the continuity equation is obtained from the variation of t. Equation 2.8.42 with respect to x and uw at time t Consider rst the surface integral. The surface divides into that part across which the liquid mass ow rate, w snn 1 vw , is prescribed and that part where the wetting liquid pressure, uw , is prescribed. Thus, the only contribution of this term to the Jacobian is the variation in the integral caused by change in surface area in that part where the mass ow is prescribed. We neglect this contribution. The remaining part of the variation of Equation 2.8.42 is
+1
V J
0 w g
dV:
0 p J w sn s J 1 + 0 w Kg
+ (1 0 n0 0 n0)("th 0 "th) t w g
u 1 + Kw 0 "th1 0 1 + n0 + n0 0 ht + uw (1 0 n0 0 n0) K 0 K1 w t t
2.8.46
-uw
uw
t+t
a >uw(st)
Procedures
-uw
t+t
0.0
st
st+t
1.0
-uw
-uw
t+t
e uwt+t<uw(st)
0.0
st+t
st
1.0
2.8.47
J 0 3K
I:
D + JI
:
" d"
I D I
Equation 2.8.35 shows that J (w =0 )nt ht, which is dened by the evolution equation given w in Constitutive behavior in a porous medium, Section 2.8.3, and so makes no contribution to the Jacobian. Finally, the Jacobian contribution from the permeability term is rather complex in the general case of the nonlinear Forchheimer ow law. Although we include it in the software, here we only write the linearized ow version reecting Darcys law ( = 0):
d Jks
@ uw @
x 1k1
@uw 0 w = @ dks ds @ uw @ uw " 1 1 @uw 0 w 1 1 @uw 0 w duw : d" + J Jks @ @ ds duw @ @ @uw @uw @ d T @ uw @ d @ uw 0 Jks @ 1 @ 1 1 @ 0 w 0 Jks @ 1 1 @ 1 @ du @ uw 1 1 @@ w + Jks @ 1 @ uw d 1 " + Jks 1 de 1 @uw 0 w (1 0 n)2 K : : 0 1 d" @ @ 3 g 1 ht 1 0 0 " duw + + ( (1 0 n 0 nt ) + nt ) : d" 0 J 9K g 9Kw (1 + ht 0 n0 ) t (1 0 n0 0 n0 ) t du + nt du : + w JKg Kw w
x k
g I x x g x x k x x k x k g x x I
g x k x x x k x x I D
2.8.48
Z
V
uw
n n + Kt + 1 0J K0 nt
0
x 1 k 1 @ x 0 w g I @uw 1 0 1 I : D + 1 (1 0 n0 0 n0) + ntI : d" dk " + de 1 @x 0 w g (1 0 n)2 3K t J g du u + @ @ xw 1 k 1 @ @ x w ds u + k1 dks du @ @xw 1 k 1 @uxw 0 w g @ s ds w 1 1 u t + (1 0 n)2 @ @xw 1 dk 1 @uxw 0 w g I :9D : I 0 K 0 K (1 +hh 0 n0) de @ Kg g w t t
+ 1t 0 g
ks @ uw @ @uw
0 n0 (J 0 1 + n + nt ) + s 1 0J K0 nt duw
0 0
1 t I : D : I Kg + Kw (1 +hht 0 n0) 9K g
s
Procedures
w
0
References
@ uw
1 @@dx 1 k 1 @x x
duw @uw
u 0 w g + @ @ xw 1 k 1 @x
0 @uw
1 @@dx @x x
1T
dV:
Coupled pore uid diffusion and stress analysis, Section 6.7.1 of the ABAQUS Analysis Users Manual Pore uid ow properties, Section 12.7.1 of the ABAQUS Analysis Users Manual
2.8.49
2.8.5
equilibrium: K MN cN
0 LMP cP = P M 0 I M ; u
and
Procedures
t+ t = t +
N
where 0
4 t ( t+ 4 t 0 t ) : With this operator the pore uid ow equation at time (t + 4t) can be rewritten as
4
M M b b t b (B MQ )T t+4t + 4tH QP uP+4t = 4tQQ 4t + (B MQ )T t : t+
Using the Newton linearization, the ow equation becomes
0 LMP cP = P M 0 I M u
and where
0(B MQ )T cM 0 4tH QP cP = RQ ; u
RQ =
These equations form the basis of the iterative solution of a time step in a coupled ow deformation solution in ABAQUS/Standard. They are, in general, nonsymmetric. The lack of symmetry may be due to a number of effects: changes in geometry, dependence of permeability on void ratio, changes in saturation in partially saturated cases, and inclusion of uid gravity load terms in total pore pressure analyses. The steady-state version of the coupled problem is also nonsymmetric. ABAQUS/Standard uses the nonsymmetric equation solver by default in all steady-state or partially saturated coupled analyses; in other cases it uses the symmetric solver by default. In the latter cases, if the effects of changes in geometry or nonlinear permeability are signicant, or if a total pore pressure (versus excess pore pressure) analysis is performed, the user is advised to activate the unsymmetric solver.
Reference
Coupled pore uid diffusion and stress analysis, Section 6.7.1 of the ABAQUS Analysis Users Manual
2.8.52
ACOUSTIC MEDIUM
2.9.1
ABAQUS provides a set of elements for modeling a uid medium undergoing small pressure variations and interface conditions to couple these acoustic elements to a structural model. These elements are provided to model a variety of phenomena involving dynamic interactions between uid and solid media. Steady-state harmonic (linear) response analysis can be performed for a coupled acoustic-structural system, such as the study of the noise level in a vehicle. The steady-state procedure is based on direct solution of the coupled complex harmonic equations, as described in Direct steady-state dynamic analysis, Section 2.6.1; on a modal-based procedure, as described in Steady-state linear dynamic analysis, Section 2.5.7; or on a subspace-based procedure, as described in Subspace-based steady-state dynamic analysis, Section 2.6.2. Mode-based linear transient dynamic analysis is also available, as described in Modal dynamic analysis, Section 2.5.5. The acoustic uid elements can also be used with nonlinear response analysis (implicit or explicit direct integration) procedures: whether such results are useful depends on the applicability of the small pressure change assumption in the uid. Often, in coupled uid-solid problems the uid forces in this linear regime are high enough that nonlinear response of the structure needs to be considered. For example, a ship subjected to underwater incident wave loads due to an explosion may experience plastic deformation, or large motions of internal machinery may occur. The acoustic medium in ABAQUS may have velocity-dependent dissipation, caused by uid viscosity or by ow within a resistive porous matrix material. In addition, rather general boundary conditions are provided for the acoustic medium, including impedance, or reactive, boundaries. The possible conditions at the surface of the acoustic medium are: 1. Prescribed pressure (degree of freedom 8) at the boundary nodes. (Boundary conditions can be used to specify pressure at any node in the model.) 2. Prescribed inward normal derivative of pressure per unit density of the acoustic medium through the use of a concentrated load on degree of freedom 8 of a boundary node. If the applied load has zero magnitude (that is, if no concentrated load or other boundary condition is present), the inward normal derivative of pressure (and normal uid particle acceleration) is zero, which means that the default boundary condition of the acoustic medium is a rigid, xed wall (Neumann condition). 3. Acoustic-structural coupling dened either by using surface-based coupling procedures (see Surfacebased acoustic-structural medium interaction, Section 5.2.7) or by placing ASI coupling elements on the interface between the acoustic medium and a structure. 4. An impedance condition, representing an absorbing boundary between the acoustic medium and a rigid wall or a vibrating structure or representing radiation to an innite exterior. 5. An incident wave loading, representing the inward normal derivative of pressure per unit density of the acoustic medium resulting from the arrival of a specied wave. The formulation of this loading case is discussed in Loading due to an incident dilatational wave eld, Section 6.3.1. It is applicable to problems involving blast loads and to acoustic scattering problems.
Procedures
2.9.11
ACOUSTIC MEDIUM
The ow resistance and the properties of the absorbing boundaries may be functions of frequency in steadystate response analysis but are assumed to be constant in the direct integration procedure. This section denes the formulation used in these elements.
Acoustic equations
The equilibrium equation for small motions of a compressible, adiabatic uid with velocity-dependent momentum losses is taken to be
@p f f +
(x ; i ) u + f (x; i ) u = 0; _ (2.9.11) @x where p is the excess pressure in the uid (the pressure in excess of any static pressure), x is the spatial _ position of the uid particle, uf is the uid particle velocity, uf is the uid particle acceleration, f is the density of the uid,
is the volumetric drag (force per unit volume per velocity), and i are i independent eld variables such as temperature, humidity of air, or salinity of water on which f and
may depend (see Acoustic medium, Section 12.3.1 of the ABAQUS Analysis Users Manual). The dAlembert term has been written without convection on the assumption that there is no steady ow of the uid. This is usually considered sufciently accurate for steady uid velocities up to Mach 0.1. The constitutive behavior of the uid is assumed to be inviscid, linear, and compressible, so p = 0Kf
(
x; i )
@ @x
1 uf ;
(2.9.12)
where Kf is the bulk modulus of the uid. For an acoustic medium capable of undergoing cavitation, the absolute pressure (sum of the static pressure and the excess dynamic pressure) cannot drop below the specied cavitation limit. When the absolute pressure drops to this limit value, the uid is assumed to undergo free expansion without a corresponding drop in the dynamic pressure. The pressure would rebuild in the acoustic medium once the free expansion that took place during the cavitation is reversed sufciently to reduce the volumetric strain to the level at the cavitation limit. The constitutive behavior for an acoustic medium capable of undergoing cavitation can be stated as
pv
where pc is the uid cavitation limit and p0 is the initial acoustic static pressure. A total wave formulation is used for a nonlinear acoustic medium undergoing cavitation. This formulation is very similar to the scattered wave formulation presented below except that the pseudo-pressure, dened as the product of the bulk modulus and the compressive volumetric strain, plays the role of the material state variable instead of the acoustic excess pressure. The acoustic excess pressure is readily available from this pseudo-pressure subject to the cavitation condition.
2.9.12
ACOUSTIC MEDIUM
Acoustic elds are strongly dependent on the conditions at the boundary of the acoustic medium. The boundary of a region of acoustic medium that obeys Equation 2.9.11 and Equation 2.9.12 can be divided into subregions S on which the following conditions are imposed:
S ,
Sfs , Sfrs ,
where the value of the acoustic pressure p is prescribed. where we prescribe the normal derivative of the acoustic medium. This condition also prescribes the motion of the uid particles and can be used to model acoustic sources, rigid walls (bafes), incident wave elds, and symmetry planes. the reactive acoustic boundary, where there is a prescribed linear relationship between the uid acoustic pressure and its normal derivative. Quite a few physical effects can be modeled in this manner: in particular, the effect of thin layers of material, whose own motions are unimportant, placed between acoustic media and rigid bafes. An example is the carpet glued to the oor of a room or car interior that absorbs and reects acoustic waves. This thin layer of material provides a reactive surface, or impedance boundary condition, to the acoustic medium. This type of boundary condition is also referred to as an imposed impedance, admittance, or a Dirichlet to Neumann map. the radiating acoustic boundary. Often, acoustic media extend sufciently far from the region of interest that they can be modeled as innite in extent. In such cases it is convenient to truncate the computational region and apply a boundary condition to simulate waves passing exclusively outward from the computational region. where the motion of an acoustic medium is directly coupled to the motion of a solid. On such an acoustic-structural boundary the acoustic and structural media have the same displacement normal to the boundary, but the tangential motions are uncoupled. an acoustic-structural boundary, where the displacements are linearly coupled but not necessarily identically equal, due to the presence of a compliant or reactive intervening layer. This layer induces an impedance condition between the relative normal velocity between acoustic uid and solid structure and the acoustic pressure. It is analogous to a spring and dashpot interposed between the uid and solid particles. As implemented in ABAQUS, an impedance boundary condition surface does not model any mass associated with the reactive lining; if such a mass exists, it should be incorporated into the boundary of the structure. a boundary between acoustic uids of possibly differing material properties. On such an interface, displacement continuity requires that the normal forces per unit mass on the uid particles be equal. This quantity is the natural boundary traction in ABAQUS, so this condition is enforced automatically during element assembly. This is also true in onedimensional analysis (i.e., piping or ducts), where the relevant acoustic properties include the cross-sectional areas of the elements. Consequently, uid-uid boundaries do not require special treatment in ABAQUS.
Procedures
S ,
2.9.13
ACOUSTIC MEDIUM
In ABAQUS the nite element formulations are slightly different in direct integration transient and steady-state or modal analyses, primarily with regard to the treatment of the volumetric drag loss parameter and spatial variations of the constitutive parameters. To derive a symmetric system of ordinary differential equations for implicit integration, some approximations are made in the transient case that are not needed in steady state. For linear transient dynamic analysis, the modal procedure can be used and is much more efcient. To derive the partial differential equation used in direct integration transient analysis, we divide Equation 2.9.11 by f , take its gradient with respect to x, neglect the gradient of
=f , and combine the result with the time derivatives of Equation 2.9.12 to obtain the equation of motion for the uid in terms of the uid pressure:
1 Kf
p+
f K f
p0 _
@ @x
1
@p f @ x
= 0:
(2.9.13)
The assumption that the gradient of
=f is small is violated where there are discontinuities in the quantity
=f (for example, on the boundary between two elements that have a different
=f value).
Variational statement
An equivalent weak form for the equation of motion, Equation 2.9.13, is obtained by introducing an arbitrary variational eld, p, and integrating over the uid:
Vf
p
1 Kf
p+
f K f
p0 _
@ @x
@p f @ x
1
dV
= 0:
Z
Vf
p
1 Kf
p+
f K f
p _
+
@p @p 1 dV f @ x @ x
1
Z
+
@p dS = 0: p n0 1 f @x S
1
(2.9.14)
Assuming that p is prescribed on Sfp , the equilibrium equation, Equation 2.9.11, is used on the remainder of the boundary to relate the pressure gradient to the motion of the boundary:
n0 1 (
Using this equation, the term
1 @p f @ x
@p n0 1 @ x
f _ u f
+ )=0
uf
on
S 0 Sfp :
(2.9.15)
Z
Vf
p
1 Kf
p+
f K f
p _
+
@p @p 1 dV f @ x @ x
1
Z
S 0Sfp
p (T (x)) dS
= 0;
(2.9.16)
2.9.14
ACOUSTIC MEDIUM
T (x) = n0 1 ( f u
f 1 @p _ ) u ) = 0n0 1 ( f f @ x
on
S 0 Sfp
(2.9.17)
has been introduced. Except for the imposed pressure on Sfp , all of the other boundary conditions described above can be formulated in terms of T (x). This term has dimensions of acceleration; in the absence of volumetric drag this boundary traction is equal to the inward acceleration of the particles of the acoustic medium:
T (x ) = n 0 1 u f
on
S 0 Sfp :
(2.9.18)
When volumetric drag is present, the boundary traction is the normal derivative of the pressure eld, divided by the true mass density: a force per unit mass of uid. Consequently, when volumetric drag exists in a transient acoustic model, a unit of T (x) yields a lower local volumetric acceleration, due to drag losses. In direct integration transient dynamics we enforce the acoustic boundary conditions as follows: On Sfp , On Sft ,
Procedures
p is prescribed and p = 0. where we prescribe the normal derivative of the acoustic pressure per unit density: Tft (x) T0 :
In the absence of volumetric drag in the medium, this enforces a value of uid particle acceleration, n0 1 uf = T0 = ain. An imposed T0 = ain can be used to model the oscillations of a rigid plate or body exciting a uid, for example. A special case of this boundary condition is ain = 0, which represents a rigid immobile boundary. As mentioned above, if the medium has nonzero volumetric drag, a unit of T0 imposed at the boundary will result in a relatively lower imposed particle acceleration. Incident wave elds on a boundary of a uid are modeled with a T0 that varies in space and time, corresponding to the effect of the arrival of the wave on the boundary. See Loading due to an incident dilatational wave eld, Section 6.3.1. the reactive boundary between the acoustic medium and a rigid bafe, we apply a condition that relates the velocity of the acoustic medium to the pressure and rate of change of pressure:
_ 0n0 1 uf = ( k1 p + c1 p) _
On Sfr ,
on
Sfr ;
(2.9.19)
where 1=k1 and 1=c1 are user-prescribed parameters at the boundary. This equation is in the form of an admittance relation; the impedance expression is simply the inverse. The layer of material, in admittance form, acts as a spring and dashpot in series distributed between the acoustic medium and the rigid wall. The spring and dashpot parameters are k1 and c1, respectively; they are per unit area of the acoustic boundary. Using this denition for the uid velocity, the boundary tractions in the variational statement become
2.9.15
ACOUSTIC MEDIUM
Tfr (x) 0
On S ,
1 p+ f c1
1 f k1
c1
p+ _
k1
p :
(2.9.110)
the radiating boundary, we apply the radiation boundary condition by specifying the corresponding impedance:
T (x ) 0
1
c1
p+ _
a1
p ;
On Sfs ,
using the admittance parameters of Equation 2.9.143 and Equation 2.9.144, dened below. the acoustic-structural interface, we apply the acoustic-structural interface condition by equating displacement of the uid and solid, which enforces the condition
n0 1 uf = n0 1 um ;
where um is the displacement of the structure. In the presence of volumetric drag it follows that the acoustic boundary traction coupling uid to solid is
T (x) = n0 1 ( m + u
m _ u ): f
In ABAQUS/Standard the formulation of the transient coupled problem would be _ made nonsymmetric by the presence of the term n0 1 ( f um ). In the great majority of practical applications the acoustic tractions associated with volumetric drag are small compared to those associated with uid inertia,
um
m u _ f
8 u m (t);
Tfs (x) n0 1 um :
On Sfrs , the mixed impedance boundary and acoustic-structural boundary, we apply a condition that relates the relative outward velocity between the acoustic medium and the structure to the pressure and rate of change of pressure:
p on Sfrs : (2.9.111) k1 c1 This relative normal velocity represents a rate of compression (or extension) of the intervening layer. Applying this equation to the denition of T (x), we obtain for the transient case:
n 0 1 (u m 0 u f ) = _ _
p+ _
2.9.16
ACOUSTIC MEDIUM
1 1 1 1 Tfrs (x) = n0 1 ( m ) 0 u p0 + p 0 p: _ f c1 f k1 c1 k1
(2.9.112)
This expression for T (x) is the sum of its denitions for Sfs and Sfr . In the steady-state case the effect of volumetric drag on the structural displacement term in the acoustic traction is included:
Tfrs (x) = n0 1 ( m + u
m 1 1 1 1 _ p0 + u )0 p 0 p: _ f f c1 f k1 c1 k1
Procedures
(2.9.113)
These denitions for the boundary term, T (x), are introduced into Equation 2.9.16 to give the nal variational statement for the acoustic medium (this is the equivalent of the virtual work statement for the structure):
Z
Vf
p
1 Kf
p+ p
Z
+
Sfrs
1 1 p+ + p + p dS _ f c1 f k1 c1 k1 S 1 Z Z 1 + p p + p dS 0 _ p n0 1 um dS c1 a1 S S
1 1 p p+ + p + p 0 n0 1 um dS = 0: _ c k c k Z
+
fr fi fs
f K f
p _
@p @p 1 dV 0 f @x @x
1
Sft
p T0 dS
f 1
f 1
(2.9.114)
" " : dV
Z
+
c um 1 um dV _
where is the stress at a point in the structure, p is the pressure acting on the uid-structural interface, n is the outward normal to the structure, is the density of the material, c is the mass proportional damping factor (part of the Rayleigh damping assumption for the structure), um is the acceleration of a point in the structure, t is the surface traction applied to the structure, um is a variational " displacement eld, and " is the strain variation that is compatible with um . For simplicity in this equation all other loading terms except the uid pressure and surface traction t have been neglected: they are imposed in the usual way.
Sfs
pum 1 n dS 0
Z
St
um 1 um dV
(2.9.115)
um 1 t dS = 0;
2.9.17
ACOUSTIC MEDIUM
Equation 2.9.114 and Equation 2.9.115 dene the variational problem for the coupled elds m and p. The problem is discretized by introducing interpolation functions: in the uid p = H P pP , P = 1; 2 . . . up to the number of pressure nodes and in the structure m = N uN , N = 1; 2 . . . up to the number of displacement degrees of freedom. In these and the following equations we assume summation over the superscripts that refer to the degrees of freedom of the discretized model. We also use the superscripts P , Q to refer to pressure degrees of freedom in the uid and N , M to refer to displacement degrees of freedom in the structure. We use a Galerkin method for the structural system; the variational eld has the same form as the displacement: m = N uN . For the uid we use p = H P pP but with the subsequent Petrov-Galerkin substitution
pP
d2 dt2
( p ): ^
(2.9.116)
^ The new function pP makes the single variational equation obtained from summing Equation 2.9.114 and Equation 2.9.115 dimensionally consistent:
PQ p PQ _ PQ PQ PM 0 pP (MfPQ + Mfr )Q + (CfPQ + Cfr )pQ + (KfPQ + Kfr + K )pQ 0 Sfs uM 0 PfP ^ QN N I N + M NM uM + C(NM uM + hSfs iT pQ 0 P N = 0; +u m) _
(2.9.117) where, for simplicity, we have introduced the following denitions:
MfPQ
fr f
Z
=
M PQ = C PQ =
ZV
1
f K
H P H Q dV;
1
PQ Cfr
K PQ
f
PQ Kfr
K PQ =
S ZS [
1 H P H Q dV; f Vf f K Z
fr frs
k1
H P H Q dS;
1
fi
c1
H P H Q dS;
fi
a1
H P H Q dS;
2.9.18
ACOUSTIC MEDIUM
PM Sfs =
Z Z S [S
fs ft frs
H P n0 1 NM dS;
QN pQ are In the case of coupled systems where the forces on the structure due to the uid Sfs very small compared to the rest of the structural forcesthe system can i solved in a sequentially be h
h i
where N is the strain interpolator. This equation denes the discretized model. We see that the volumetric drag-related terms are mass-like; i.e., proportional to the uid element mass matrix. P The term Pf is the nodal right-hand-side term for the acoustical degree of freedom pP , or the applied force on this degree of freedom. This term is obtained by integration of the normal derivative of pressure per unit density of the acoustic medium over the surface area tributary to a boundary node.
Procedures
QN pQ term omitted; i.e., in coupled manner. The structural equations can be solved with the Sfs 2 PM 3 uM an analysis without uid coupling. Subsequently, the uid equations can be solved, with Sfs imposed as a boundary condition. This two-step analysis is less expensive and advantageous for systems such as metal structures in air.
Time integration
The equations are integrated through time using the standard implicit (ABAQUS/Standard) and explicit (ABAQUS/Explicit) dynamic integration options. From the implicit integration operator we obtain relations between the variations of the solution variables (here represented by f ) and their time derivatives:
Da def =
f pP def f_ = P ; Dv = f : f p ^
The equations of evolution of the degrees of freedom can be written for the implicit case as
1 PQ PQ _ 0pP D (MfPQ + Mfr )pQ + (CfPQ + Cfr )pQ a PQ + K PQ + K PQ )pQ 0 Sfs uM 0 PfP PM + ( Kf fr QN N I N + M NM uM + C(NM uM + hSfs iT pQ 0 P N = 0: + u m) _
2.9.19
ACOUSTIC MEDIUM
0pP
PQ + M PQ ) + Dv (C PQ + C PQ ) + 1 (K PQ + K PQ + K PQ ) (M f fr fr f fr Da f Da h iT QN dpQ P PM duM + pP 1 PfP + uN Sfs + p Sfs Da N K NM + Da M NM + Dv (C(NM + C(NM ) duM = 0; + u m) k)
dp
where dp and du are the corrections to the solution obtained from the Newton iteration, K NM is the NM structural stiffness matrix, and C(k) is the structural damping matrix. These equations are symmetric if the structural stiffness is symmetric. For explicit integration the uid mass matrix is diagonalized in a manner similar to the treatment of structural mass. The explicit central difference procedure described in Explicit dynamic analysis, Section 2.4.5, is applied to the coupled system of equations.
Summary of additional approximations of the direct integration transient formulation
As mentioned above, derivation of symmetric ordinary differential equations in the presence of volumetric drag requires some approximations, in addition to those inherent in any nite element method. First, the spatial gradients of the ratio of volumetric drag to mass density in the uid are neglected. This may be important in lossy, inhomogeneous acoustic media. Second, to maintain symmetry, the effect of volumetric drag on the uid-solid boundary terms is neglected. Finally, the effect of volumetric drag on the radiation boundary conditions is approximate. If any of these effects is expected to be signicant in an analysis, the user should realize that the results obtained are approximate.
Formulation for eigenvalue extraction and mode-based procedures
From the discretized equation, Equation 2.9.117, neglecting any damping terms and any terms associated with a reactive surface, the eigenvalue problem can be stated as
"
K
NM
0
Sfs
QN T # uM 2 M NM PM PQ pQ 0 ! 0Sfs K
f
PQ M
0
u p
M Q = 0:
(2.9.118)
As stated, this problem cannot be solved in ABAQUS due to the unsymmetric stiffness and mass = p=! 2 , augmenting the system of equations with matrices. Introducing an auxiliary variable, Kf p = !2 Kf , and manipulating the equations yields
Kf
PQ 7 0 !2 4 5
NM
0 0
0 0 0
Kf
31 8 M 9 <u = C Q 0 5A PQ : p Q ; = 0:
0
(2.9.119)
This augmented system of equations is used for the eigenvalue extraction. The auxiliary variable is internal to ABAQUS/Standard and is not available for output. This solution method for the coupled
2.9.110
ACOUSTIC MEDIUM
acoustic-structural eigenvalue problem is implemented only for the Lanczos eigensolver. If Kf is singular, orthogonalization against the singular acoustic modes is done in the Lanczos eigensolver. Since the original system of equations, Equation 2.9.118, is unsymmetric, there are left- and right-hand-side eigenvectors. It can be shown that the left and right eigenvectors are as follows:
M M L = v Q ; R = u Q ; p
where
vM =
For any nonsingular acoustic circular eigenfrequency. The left and right eigenvector subspaces are then used to compute modal quantities (generalized mass, participation factors, effective mass) and to project the mass, stiffness, and damping matrices in modebased procedures (such as subspace-based steady-state dynamics or transient modal dynamic analysis) to obtain a reduced system of equations. Most of these computations are conducted in a very similar fashion to the way they are carried out in the pure structural problem and will not be discussed here. The only exception worth a brief discussion is the choice for the calculation of the acoustic participation factors and effective masses, as follows. First, a rigid body acoustic mode, Ta , analogous to the rigid body modes for the structural problem outlined in Variables associated with the natural modes of a model, Section 2.5.2, is chosen to be a constant pressure eld of unity. A total acoustic mass is then dened as mac = TaT Mf Ta. Left and right acoustic participation factors are dened as
Procedures
0L =
1 vM T M NM 0 0Sfs Mf m Q
1 Q 0 Ta = m M f Ta
and
0R =
1 0 T M NM 0 u M = 1 00 T S u M + T M p Q 1 : a fs a f 0Sfs Mf pQ m Ta m
0ac =
and an acoustic effective mass computed as
0L 0R =mac
e mac = (0ac)2 m :
The scaling by mac in the equation for 0ac is arbitrary. However, this scaling ensures that when all eigenmodes are extracted, the sum of all acoustic effective masses is 1.0 (minus the contributions from nodes constrained in the acoustic degree of freedom).
Formulation for steady-state response
The direct-solution steady-state dynamic analysis procedure is the preferred solution method for acoustics in ABAQUS if volumetric drag and/or acoustic radiation are signicant. If these effects are not signicant, the mode-based procedure is preferred because of its efciency.
2.9.111
ACOUSTIC MEDIUM
All model degrees of freedom and loads are assumed to be varying harmonically at an angular frequency
, so we can write ~ f = f exp i
t;
f = 0
2 f:
@p @x
_ +
u f + f u f = 0 + i
)~ f = 0:
u + i
(2.9.120)
@p ~ 0
2 ( f @x
f ~
and, thus, write
@p ~ 0
2 (~) uf = 0: ~ (2.9.121) @x The equilibrium equation is now in a form where the density is complex and the acoustic medium ~ velocity does not enter. We divide this equation by and combine it with the second time derivative of the constitutive law, Equation 2.9.12, to obtain
~ 1 p 0 @ 1 1 @ p = 0: 0
K ~ @ x @ x ~ f
2
(2.9.122)
We have not used the assumption that the spatial gradient of r=f is small, as was done in the transient dynamics formulation.
Variational statement
The development of the variational statement parallels that for the case of transient dynamics, as though the volumetric drag were absent and the density complex. The variational statement is
Vf
p
1 @ p ~ 1 ~ 0
2 K p 0 @@x 1 @ x dV = 0: ~ f
~ ~
2 p dV + Z 1 @p 1 @ p dV + Z p 1 @ p 1 n0 dS = 0: 0 p K ~ @x @x ~ @x ~ f Vf Vf S
2.9.112
ACOUSTIC MEDIUM
~ T (x ) 0
~ 1 @ p 1 n0 = 0
2n0 1 uf = n0 1 uf : ~ @x ~
This expression is not the Fourier transform of the boundary traction dened above for the transient case. The steady-state denition is based on the complex density and includes the volumetric drag effect in such a way that it is always equal to the acceleration of the uid particles. The application of boundary conditions may be slightly different for some cases in steady state, due to this denition of the traction. On Sfp , On Sft ,
Procedures
the reactive boundary between the acoustic medium and a rigid bafe, we apply
2 ~fr (x) 0 i
0
p: ~ T c k
1 1
(2.9.123)
On S ,
the radiating boundary, we apply the radiation boundary condition impedance in the same form as for the reactive boundary but with the parameters as dened in Equation 2.9.138 and Equation 2.9.139. the acoustic-structural interface, we equate the displacement of the uid and solid as in the transient case. However, the acoustic boundary traction coupling uid to solid, ~ ~ T (x ) = 0
2 n 0 1 u m ; can be applied without affecting the symmetry of the overall formulation. Consequently, the acoustic tractions in the steady-state case make no assumptions about volumetric drag.
On Sfs ,
On Sfrs ,
_ _ n 0 1 (u m 0 u f ) =
results in the denition:
k1
1 p + 1 p on _ c
1
Sfrs
~ u Tfrs (x) = 0
2 n0 1 (~ m ) 0
2 i
p+ ~
p: ~ c1 k1
2.9.113
ACOUSTIC MEDIUM
@p ~ 1 ~ 1 0
2p K p + @p @ x dV ~ @x f Vf i
2 Z Z p 0 k p dS ~ 0 p ain dS + c1 1 S S [S Z ~ + p
2 n0 1 um dS S i
2 Z p p0 ~ k p +
2 n0 1 um dS = 0: ~ ~ + c
1
ft fr fi fs
Sfrs
This equation is formally identical to Equation 2.9.14, except for the pressure stiffness term, the radiation boundary conditions, and the imposed boundary traction term. Because the volumetric drag effect is contained in the complex density, the acoustic-structural boundary term in this formulation does not have the limitation that the volumetric drag must be small compared to other effects in the acoustic medium. In addition, in this formulation the applied ux on an acoustic boundary represents the inward acceleration of the acoustic medium, whether or not the volumetric drag is large. Finally, the radiation boundary conditions do not make any approximations with regard to the volumetric drag parameter. ~ The above equation uses the complex density, 1=. We manipulate it into a form that has real coefcients and an additional time derivative through the relations
~
to obtain
1=
2 f
f
=
+
2 =
2 + i 2 +
2 =
2 ; f
@p @x
1 @p _ =
@x ;
Z
Vf
2 @p ~
=
2 @p ~ ~ 0p K p + 2 +
f2 =
2 @p 1 @ x + (i
) 2 +
2 =
2 @p 1 @ x dV @x @x f f Z Zf ~ p
2 n0 1 um dS 0 p ain dS + S S [S i
2 Z p p0 ~ k p dS = 0: ~ + c
ft fs frs
(2.9.124)
Applying Galerkins principle, the nite element equations are derived as before. We arrive again at Equation 2.9.117 with the same matrices except for the damping and stiffness matrices of the
2.9.114
ACOUSTIC MEDIUM
acoustic elements and the surfaces that have imposed impedance conditions, which now appear as
CfPQ KfPQ
Z Z
= c1 H P H Q dS; S 1 PQ = 0; Kfr PQ K = 0 :
PQ Cfr
fr
Vf 2 f
2 Z Vf f
=
2 @H P +
2 =
2 @ f @H P 2 =
2 @ +
@H Q dV; @ @H Q dV; @
x x
Procedures
The matrix modeling loss to volumetric drag is proportional to the uid stiffness matrix in this formulation. For steady-state harmonic response we assume that the structure undergoes small harmonic vibrations, identied by the prex , about a deformed, stressed base state, which is identied by the subscript . Hence, the total stress can be written in the form
= 0 + 1 = 0 +
where 0 is the stress in the base state; el is the elasticity matrix for the material; c is the stiffness proportional damping factor chosen for the material (to give the stiffness proportional contribution to the Rayleigh damping, thus introducing the viscous part of the material behavior); and, from the discretization assumption,
1" = M 1uM :
To solve the steady-state problem, we assume that the governing equations are satised in the base state, and we linearize these equations in terms of the harmonic oscillations. For the internal force vector this yields
h i PQ PQ PM u ~ p 0 pP 0
2 (MfPQ + Mfr ) + i
(CfPQ + Cfr ) + KfPQ 1~Q +
2Sfs 1~M 0 1PfP ^ h i h QN iT ~ u p +uN 0
2M NM + i
(C(NM + C(NM ) + K NM 1~M + Sfs 1~Q 0 1P N = 0; m) k)
with
K NM =
(2.9.125)
ACOUSTIC MEDIUM
(this stiffness includes the initial stress matrix, so stress stiffening and load stiffness effects associated with the base state stress and loads are included), and
NM C (k )
Z h c N
: Del : M
dV:
We assume that the loads and (because of linearity) the response are harmonic, and, hence, we can write 0 Q 11 Q 0 0 Q1
1~ = 0< 0p 1+ i = 0p 11 i
t p ~ ~ exp M = < uM + i = uM exp i
t ~ ~ 1~ u
~ ~ ~ 1P N = < P N + i = P N exp i
t ~ ~ ~ 1PfP = < PfP + i = PfP exp i
t; 0 1 where < pQ , < uM , = pQ , and = uM are the real and imaginary parts of the amplitudes of ~ 0~ 1 0 ~ 1 0 ~ 1 ~ ~ the response; < P N and = P N are the real and imaginary parts of the amplitude of the force ~P ~P applied to the structure; <(Pf ) and =(Pf ) are the real and imaginary parts of the amplitude of the acoustic traction (dimensions of volumetric acceleration) applied to the uid; and
is the circular frequency. We substitute these equations into Equation 2.9.125 and use the time-harmonic form of ^ Equation 2.9.116, pP = 0
0 pP , which yields the coupled complex linear equation system
and
2
2 h
(2.9.126)
where
i h PQ PQ < APQ =
02 (KfPQ + Kfr ) 0 (MfPQ + Mfr ) f i h PQ = APQ = 0
01(CfPQ + Cfr ) f i h < ANM = K NM 0
2 M NM s i h = ANM = 0
(C(NM + C(NM ): s m) k)
and
If K NM is symmetric, Equation 2.9.126 is symmetric. The system may be quite large, because the real and imaginary parts of the structural degrees of freedom and of the pressure in the uid all appear in the system. This set of equations is solved for each frequency requested in the direct-solution steady-state dynamics procedure. If damping is absent, the user can specify that only the real matrix
2.9.116
ACOUSTIC MEDIUM
equation should be factored in the analysis. Nonzero r values for the acoustic medium and nonzero 1=c1 values for the impedances represent damping. As mentioned above for the transient case, the coupled system can be split into an uncoupled structural analysis and an acoustic analysis driven by the structural response, provided the uid forces on the structure are small.
Volumetric drag and uid viscosity
The medium supporting acoustic waves may be owing through a porous matrix, such as berglass used for sound deadening. In this case the parameter
is the ow resistance, the pressure drop required to force a unit ow through the porous matrix. A propagating plane wave with nominal _ particle velocity uf loses energy at a rate
_ E =
Procedures
2 3 _ 0 u f 2 :
(2.9.127)
Fluids also exhibit momentum losses without a porous matrix resistive medium, through coefcients of shear viscosity and bulk viscosity . These are proportionality constants between components of the stress and spatial derivatives of the shear strain rate and volumetric strain rate, respectively. In uid mechanics the shear viscosity term is usually more important than the bulk term ; however, acoustics is the study of volumetrically straining ows, so both constants can be important. The linearized Navier-Stokes equations for adiabatic perturbations about a base state can be expressed in terms of the pressure eld alone (Morse and Ingard, 1968):
@ @p 1 @ x @x
+ 4 @ @p f 3 1 _: p0 Kf Kf @ x @ x
3
(2.9.128)
In steady state this linearized equation can be written in the form of Equation 2.9.121, with
2
< ( f ) = 6 ~ 4
1+
Kf Kf
2 0 0
+ 4 3
7 12 5 f ; 3 7 12 5 f ;
= ( f ) = 0 6 ~ 4
2
1+
Kf
2 0
1 + 4 3
+ 4 3
3
so that the viscosity effects can be modeled as a volumetric drag parameter with the value
=4 1+
2 0 + Kf 2 0
Kf
4 1 3 7 12 5 f K f : + 4 3
@ @p 1 @x @x
Kf p;
(2.9.129)
2.9.117
ACOUSTIC MEDIUM
so we can write
Kf
In steady-state form
@ @p 1 @x @x
(2.9.130)
@ @p 1 f @ x @ x
1 0 i
+ 4 3 K K2
f f
p = 0;
(2.9.131)
where
is the forcing frequency. This leads to the following analogy between viscous uid losses and volumetric drag or ow resistance:
2 f
=
Kf
4 + 3
(2.9.132)
with density constant with respect to frequency. The energy loss rate for a propagating plane wave in this linearized, adiabatic, small-viscosity case is
_ E
Acoustic output quantities
=0
+
4 3
2 f 2u f 3 2 : _ K
f
(2.9.133)
Several secondary quantities are useful in acoustic analysis, derived from the fundamental acoustic pressure eld variable. In steady-state dynamics the acoustic particle velocity at any eld point is
v
@p ~ ~ = i! @ x : ~
~ = 0 1 1 v: ^ I 2 ~
In an acoustic medium the stress tensor is simply the acoustic pressure times the identity tensor, = 0pI, so this expression simplies to
p ~ ~ = 01 p @ ^ ~@x I ^
"
2i! ~
The hats denote complex conjugation. The real part of the intensity is referred to as the active intensity, and the imaginary part is the reactive intensity.
Impedance and admittance at uid boundaries
Equation 2.9.111 (or alternatively Equation 2.9.19) can be written in a complex admittance form for steady-state analysis:
2.9.118
ACOUSTIC MEDIUM
n0 1 (um 0 uf ) = ( _ _
where we dene
1 + i
)p = 1 p = 0T (x)(i
)01 ; c k Z (
)
1 1
(2.9.134)
1 1 + i
: Z (
) c1 k1
(2.9.135)
Procedures
pK = rK ( Z (
) = ~
f f
+ i ):
(2.9.136)
For the impedance-based nonreective boundary condition in ABAQUS/Standard, the equations above are used to determine the required constants 1=c1 and 1=k1. They are a function of frequency if the volumetric drag is nonzero. The small-drag versions of these equations are used in the direct time integration procedures, as in Equation 2.9.142.
Radiation boundary conditions
Many acoustic studies involve a vibrating structure in an innite domain. In these cases we model a layer of the acoustic medium using nite elements, to a thickness of 1=3 to a full wavelength, out to a radiating boundary surface. We then impose a condition on this surface to allow the acoustic waves to pass through and not reect back into the computational domain. For radiation boundaries of simple shapessuch as planes, spheres, and the likesimple impedance boundary conditions can represent good approximations to the exact radiation conditions. In particular, we include local algebraic radiation conditions of the form
~ ~ ~ where k =
=Kf is the wave number and is the complex density (see Equation 2.9.120). f is a geometric factor related to the metric factors of the curvilinear coordinate system used on the boundary, and is a spreading loss term (see Table 2.9.11). Comparison of Equation 2.9.137 and Equation 2.9.19 reveals that, for steady-state analysis, there exists a direct analogy to the reactive boundary equation, Equation 2.9.123, with
k1
and
n0 1
@p @x
~ = M p = f (ik + )p;
(2.9.137)
1 = =(
Kf ~
pf
f ) 0
2 (1 + (
=
)2 ) ; f f
(2.9.138)
2.9.119
ACOUSTIC MEDIUM
f r=f 1 = <( p f ) + : 2 (1 + (
=
)2 ) c1
f Kf ~ f
(2.9.139)
For transient procedures the treatment of volumetric drag in the acoustic equations and the radiation conditions necessitates an approximation. In the acoustics equation we use the boundary term
@p _ 0n0 1 @ x 1 = n0 1 uf +
uf :
f f
(2.9.140)
Combining Equation 2.9.137 with Equation 2.9.140, expanding about rst-order terms leads to
= 0, and
p:
retaining only
@p 1 n0 1 = f p i
@ x f
"
f K f
p+f f
"
p 2 f f K f
!#
(2.9.141)
The Fourier inverse of the steady-state form results in the transient boundary condition
@p 1 n0 1 =f p 1 @ x f
"
f K f
p+f _ f
"
p p: 2 f f K f
(2.9.142)
This expression involves independent coefcients for pressure and its rst derivative in time, unlike the transient reactive boundary expression (Equation 2.9.110), which includes independent coefcients for the rst and second derivatives of pressure only. Consequently, to implement this expression, we dene the admittance parameters
" 1 = pf c
1
f K f
(2.9.143)
and
1 =f a
1
"
f
p ; 2 f f K f
(2.9.144)
so the boundary traction for the transient radiation boundary condition can be written
n0 1
@p 1 @ x f
= c1 p + a1 _
1
p:
The values of the parameters f and vary with the geometry of the boundary of the radiating surface of the acoustic medium. The geometries supported in ABAQUS are summarized in Table 2.9.11. In the table refers to the eccentricity of the ellipse or spheroid; r1 refers to the radius of the circle, sphere, or the semimajor axis of the ellipse or spheroid; xg is the vector locating the integration point on the ellipse or spheroid; x0 is the vector locating the center of the ellipse or spheroid; and em is the vector that orients the major axis. These algebraic boundary conditions are approximations to the exact impedance of a boundary radiating into an innite exterior. The plane wave condition is the exact impedance for plane waves
2.9.120
ACOUSTIC MEDIUM
Table 2.9.11 Boundary condition parameters. Geometry Plane Circle or circular cylinder Ellipse or elliptical cylinder Sphere Prolate spheroid
q
1
1 1 1
2r 1 1 2 r1
Procedures
q
1
r1
1
r1
normally incident to a planar boundary. The spherical condition exactly annihilates the rst Legendre mode of a radiating spherical surface; the circular condition is asymptotically correct for the rst mode (Bayliss et al., 1982). The elliptical and prolate spheroidal conditions are based on expansions of elliptical and prolate spheroidal wave functions in the low-eccentricity limit (Grote and Keller, 1995); the prolate spheroidal condition exactly annihilates the rst term of its expansion, while the elliptical condition is asymptotic.
An improvement on radiation boundary conditions for plane waves
As already pointed out, the radiation boundary conditions derived in the previous section for plane waves are actually based on the presumption that the sound wave impinges on the boundary from an orthogonal direction. But this is not always the case. Figure 2.9.11 shows a general example for plane waves in which the sound wave direction differs from the boundary normal by an angle of . To consider this situation accurately, we adopt the plane-wave-radiation equation used in Sandler (1998); i.e.,
1 _ = C p = cos p; _ (2.9.145) C n p ~ ~ _ where C = Kf = is the sound speed with = f +
p=p and Cn is the corresponding speed normal
n0 1
@p @x @p 1 @ x f
to the boundary. This exact description of the geometry is the starting point for many developments of approximate absorbing boundary conditions (see, for example, Engquist and Majda, 1977). Thus we have
n0 1
cos =p
f K f
p 1 +
p p: _ f _
1
(2.9.146)
Using the rst-order expanding approximation to the second term in the square root in the above equation (similar to what we did to reach Equation 2.9.141), we can obtain an improved radiation boundary condition
2.9.121
ACOUSTIC MEDIUM
s sound wave Cs C n
-
= C K f r~
~ r= rf+ rp p
Cn
boundary
@p 1 n0 1 @ x f
cos =p K
f
f
p+ _
2 f p :
(2.9.147)
It can be found from comparison that this equation differs from Equation 2.9.142 only by a factor of cos for plane waves. In two dimensions the cos can be calculated as
@p cos = j @ n j=
s 2 2 @p + @p : @n @s
(2.9.148)
The normal and tangential derivatives @p=@ n and @p=@ s at the integration points can be evaluated using the corresponding element along the radiation boundary surface (see Figure 2.9.12); i.e.,
where pP are the nodal pressure values of the element. The method described in this section can be used only for direct integration transient dynamics, and it cannot be used with steady-state or modal response. In addition, it is available for planar, axisymmetric, and three-dimensional geometries. Finally, the method makes the equilibrium equations nonlinear, as shown in Equation 2.9.148. Although in theory the iteration process in ABAQUS/Standard can solve the nonlinear equilibrium equations accurately, the use of a small half-step residual tolerance is strongly suggested since in many cases the pressure and its related residual along the radiation boundaries are very weak relative to the
@p @n
=n
@H P P p @x
and
@p @s
=s
@H P P p ; @x
(2.9.149)
2.9.122
ACOUSTIC MEDIUM
x element x
integration points
Procedures
boundary
Figure 2.9.12 An element along the boundary. other places in the modeled domain. The computation of cos at the integration point is based on the nodal pressures. The nodal pressures are updated using the explicit central difference procedure described in Explicit dynamic analysis, Section 2.4.5.
References
Acoustic, shock, and coupled acoustic-structural analysis, Section 6.9.1 of the ABAQUS Analysis Users Manual Acoustic medium, Section 12.3.1 of the ABAQUS Analysis Users Manual Acoustic loads, Section 19.4.4 of the ABAQUS Analysis Users Manual
2.9.123
PIEZOELECTRIC ANALYSIS
2.10.1
PIEZOELECTRIC ANALYSIS
The piezoelectrical effect is the coupling of stress and electrical eld in a material: an electrical eld causes the material to strain, and vice versa. ABAQUS/Standard has the capability to perform fully coupled piezoelectric analysis. The elements that are used in this case contain both displacement degrees of freedom and the electric potential as nodal variables.
Equilibrium and ux conservation
Procedures
The piezoelectric effect is governed by coupled mechanical equilibrium and electric ux conservation equations. The mechanical equilibrium equation is
f 1 u dV; (2.10.11) V S V where is the true (Cauchy) stress at a point currently at x; t is the traction across a point of the surface of the body; f is the body force per unit volume in the body (such as the dAlembert force def " f = 0 u, in which is the density of the body); and " = sym(@ u=@ x), where u is an arbitrary, continuous vector eld (the virtual velocity eld). The electrical ux conservation equation is
" : " dV =
t 1 u dS +
1 E dV
qS ' dS +
qV ' dV;
(2.10.12)
where q is the electric ux vector; qS is the electric ux per unit area entering the body at a point on def its surface; qV is the electric ux entering the body per unit volume; and = 0@'=@ x, where ' is an arbitrary, continuous, scalar eld (the virtual potential). These quantities are also known by other terms that are frequently used within electrical engineering references. The electric ux vector q is known as the electrical displacement, and the potential gradient E is known as the electrical eld.
Currently the assumption of linear materials is utilized. The basic equations for a piezoelectric linear medium are dened in the following. Following Ikeda (1990), three alternative forms of the constitutive equations are presented: e-form:
ij = Dijkl "kl
0 e ' Em ; mij
' (" ) '
Ej ;
'
'
' ( )
Ej ;
2.10.11
PIEZOELECTRIC ANALYSIS
g-form:
' where Sijkl are the material compliances; Dijkl are the material stiffnesses; e' ; d' ; and gijk are ijk ijk ' piezoelectric constants; and Dij are the dielectric constants. In these equations the superscript E; q; ", or above a particular property indicates that the property is dened at zero electrical gradient, at zero electrical displacement, at zero strain, and at zero stress, respectively. Since all these forms describe the same constitutive relationships, the different mechanical and piezoelectrical constants can be expressed in terms of one another. The following relationships exist among the properties (Ikeda, 1990):
"ij
e' mij
' ( ' d' = Dmk ) gkij mij q E Dijkl 0 Dijkl = e' (D'(") )01 e' mn nkl mij ' ' Dij() 0 Dij(") = d' e' ikl jkl
ij
(2.10.13)
These are expressed in terms of the piezoelectric stress coefcient matrix e' . However, ABAQUS mkl also allows the input of piezoelectric constants in terms of the piezoelectric strain coefcient matrix d' . mkl The constitutive equations in the g-form can also be expressed as
ij
and
These equations can be convenient in interpreting and verifying the results of piezoelectrical analyses.
Kinematics
For the piezoelectric elements both displacements and electric potentials exist at the nodal locations. The displacements and electrical potentials are approximated within the element as
u = N N uN
2.10.12
PIEZOELECTRIC ANALYSIS
and
where N is the array of interpolating functions and uN and 'N are nodal quantities. The body forces and charges as well as the surface forces and charges are interpolated in a similar manner. The strains and electrical potential gradients are given as
'=
N N 'N ;
"=
and
B N uN u
Procedures
E = 0B N 'N ; ' N and BN are the spatial derivatives of NN . In geometrically nonlinear analyses these spatial where Bu '
With these approximate elds and the constitutive properties given above, in conjunction with the equilibrium and conservation equations, the following system of equations is derived in terms of nodal quantities:
M MN uN
and
(2.10.14)
MN K'u uN
where
(2.10.15)
M MN
NN dV
is the mass matrix (no inertia terms exist for the electrical ux conservation equation), is the mass density, Z MN M: N Kuu = m : u dV u
D B
1
MN K''
is the dielectric stiffness matrix,
Z
V
MN K'u
is the piezoelectric coupling matrix,
Z
V
BM e : BN dV ' u
1
PM
NM Pv dV + NM Ps dS + PcM
1
2.10.13
PIEZOELECTRIC ANALYSIS
QM =
Z
V
NM Qv dV + NM Qs dS + QM c
1
is the electrical charge vector. In these expressions the constitutive properties are specied in a matrix form where m is the mechanical relationship, ' is the electrical relationship, and is the piezoelectrical relationship. The load vectors include the body, surface, and concentrated quantities, as shown. The unknowns are the nodal displacements and potentials. Once these are determined, the strains and potential gradients can be computed using the expressions given above. The stresses and electrical ux densities are computed by means of constitutive relationships.
Reference
2.10.14
2.11.1
The ABAQUS/Standard capability for uncoupled heat transfer analysis is intended to model solid body heat conduction with general, temperature-dependent conductivity; internal energy (including latent heat effects); and quite general convection and radiation boundary conditions. This section describes the basic energy balance, constitutive models, boundary conditions, nite element discretization, and time integration procedures used. Heat transfer in owing materials (convection) is discussed in Convection/diffusion, Section 2.11.3. Radiation heat transfer in cavities is discussed in Cavity radiation, Section 2.11.4. All such heat transfer mechanisms can be present in a model.
Energy balance
Procedures
Z
V
_ U dV =
Z
S
q dS +
Z
V
r dV ;
(2.11.11)
_ where V is a volume of solid material, with surface area S ; is the density of the material; U is the material time rate of the internal energy; q is the heat ux per unit area of the body, owing into the body; and r is the heat supplied externally into the body per unit volume. It is assumed that the thermal and mechanical problems are uncoupled in the sense that U = U ( ) only, where is the temperature of the material, and q and r do not depend on the strains or displacements of the body. For simplicity a Lagrangian description is assumed, so volume and surface mean the volume and surface in the reference conguration.
Constitutive denition
This relationship is usually written in terms of a specic heat, neglecting coupling between mechanical and thermal problems:
c ( ) = dU ; d
except for latent heat effects at phase changes, which are given separately in terms of solidus and liquidus temperatures (the lower and upper temperature bounds of the phase change range) and the total internal energy associated with the phase change, called the latent heat. When latent heat is given, it is assumed to be in addition to the specic heat effect (see Figure 2.11.11). For many cases it is reasonable to assume that the phase change occurs within a known temperature range, which can be specied by the user. However, in some cases it may be necessary to include a kinetic theory for the phase change to model the effect accurately (an example would be the prediction of crystallization in a polymer casting process). For such cases the user can model the process in considerable detail, using the solution-dependent state variable feature in ABAQUS, with user subroutine HETVAL.
2.11.11
U, internal energy
, temperature
c() =
dU d
Specific heat, c ()
Solidus
Figure 2.11.11
Liquidus
Specic heat, latent heat denition.
, temperature
2.11.12
where k is the conductivity matrix, k = k( ); f is the heat ux; and k can be fully anisotropic, orthotropic, or isotropic.
Boundary conditions
@ = 0k @ x ;
(2.11.12)
x
Boundary conditions can be specied as prescribed temperature, = (x; t); prescribed surface heat ux, q = q(x; t) per area; prescribed volumetric heat ux, q = r(x; t) per volume; surface convection: q = h( 0 0), where h = h(x; t) is the lm coefcient and 0 = 0 (x; t) is the sink temperature; and radiation: q Stefan-Boltzmann constant) and Z is the absolute zero on the temperature scale used. Surfaces can also participate in cavity radiation effects. The cavity radiation formulation in ABAQUS is described in Cavity radiation, Section 2.11.4.
= A (
Procedures
0 Z ) 4 0 ( 0 0 Z ) 4
Spatial discretization
A variational statement of the energy balance, Equation 2.11.11, together with the Fourier law, Equation 2.11.12, is obtained directly by the standard Galerkin approach as
_ U dV +
@
@x
@ 1 k 1 @ x dV
r dV +
sq
q dS;
(2.11.13)
where is an arbitrary variational eld satisfying the essential boundary conditions. The body is approximated geometrically with nite elements, so the temperature is interpolated as
= N N (x ) N ; N = 1 ; 2; . . . ;
where N are nodal temperatures. The Galerkin approach assumes that , the variational eld, is interpolated by the same functions:
= N N N :
First- and second-order polynomials in one, two, and three dimensions are used for the N N . With these interpolations the variational statement, Equation 2.11.13, becomes
N
Z
_ N N U dV
Z
+ =
@N N @x
ZV
V
N N r dV +
@ 1 k 1 @ x dV Z
Sq
N N q dS ;
and since the N are arbitrarily chosen, this gives the system of equations
2.11.13
Z
V
N N U dV
+ =
@ @N N 1k1 dV @x ZV @ x Z
N N r dV
Sq
N N q dS:
(2.11.14)
This set of equations is the continuous time description of the geometric approximation.
Time integration
Ut+1t
= (Ut+1t 0 Ut)(1=1t): )_ + _ 1
(2.11.15)
This operator is chosen for a number of reasons. First of all, we choose from one-step operators of the form 0 1 ft+1t ft 0 ft ft+1t t
= + (1 12
because of their simplicity in implementation (for example, no special starting procedures are needed) and well-understood behavior. For < = such operators are only conditionally stable for linear heat transfer problems. We prefer to work with unconditionally stable methods, because ABAQUS is most commonly applied to problems where the solution is sought over very long time periods ), and so choose = . (compared to the stability limit for the explicit form of the operator, = , has the highest accuracy. However, that Of these operators the central difference method, form of the operator tends to produce oscillations in the early time solution that are not present in : backward difference. Introducing the operator, the backward difference form. Thus, we use Equation 2.11.15, into the energy balance Equation 2.11.14 gives
=1 2
=0
12
=1
(2.11.16)
This nonlinear system is solved by a modied Newton method. The method is modied Newton because the tangent matrix (the Jacobian matrix)that is, the rate of change of the left-hand side of N Equation 2.11.16 with respect to t+1t is not formed exactly. The formation of the terms in this tangent matrix is now described. The internal energy term gives a Jacobian contribution:
1 Z N N dU N M dV : 1t V d t+1t is the specic heat, c(), outside the latent heat range, and is c + L=(L 0 S ) if
L > t+1t > S at the integration point, where L and S are the liquidus and solidus temperatures and L is the latent heat associated with this phase change.
(dU=d)jt+1t
2.11.14
In severe latent heat cases this term can result in numerical instabilities, as the stiffness term
dU=d is small outside the solidus-liquidus temperature range and is very stiff inside that rather
narrow range. To avoid such instabilities in those cases this term is modied to a secant term during the early iterations of the solution to a time step. Since the modication occurs only in cases involving latent heat, it affects only those problems. The conductivity term gives a Jacobian contribution:
Z
V
@N N @x
+1
1
t
@N M @x
dV
Z
V
@N N @x
@k
t
@
+1
1
t
@x
@
t
+1
N M dV :
The second of these terms is typically small, since the conductivity usually varies only slowly with temperature. Because of this, and because the term is not symmetric, it is usually more efcient to omit it. This term is omitted unless the unsymmetric solver is chosen. Prescribed surface uxes and body uxes can also be temperature dependent and will then give rise to Jacobian contributions. With lm and radiation conditions, the surface ux term gives a Jacobian contribution:
Z
Procedures
NN
S
@q @
t
For lm conditions, q
= h()( 0 );
o
+1
N M dS:
@q
= A ( 4 0 4 );
o
@
= @h ( 0 ) + h; @
o
@q @
= 4A3:
M
These terms are included in exactly this form in the Jacobian. The modied Newton method is then
1 1t
NN
V
dU N M dV d t+1t
+ + =
Z
V
@N N @x N
N
1 k +1 1 @Nx @
t t
dV
@h @
0 ) + h + 4A3
o
N
Z
V
dS cM
N N r dV
Z
V
+
;i
N N q dS
Sq
1 0 1t
N N Ut+1t
0 U ) dV
t
(2.11.17) N For purely linear systems Equation 2.11.17 is linear in cM and, hence, in t+1t , so a single equation N solution provides the t+1t . Since the method usually is only a minor modication of Newtons method, convergence is rapid.
N with t+1t;i+1
= +1 + c
N t t;i
@N N @x
@ 1 k 1 @ x dV ;
N
iteration number.
2.11.15
ABAQUS/Standard uses an automatic (self-adaptive) time stepping algorithm to choose t. This is based on a user-supplied tolerance on the maximum temperature change allowed in a time increment, and the increment is adjusted according to this parameter, as well as the convergence rate of Equation 2.11.17 in nonlinear cases. The rst-order heat transfer elements (such as 2-node link, 4-node quadrilateral, and 8-node brick) use a numerical integration rule with the integration stations located at the corners of the element for the heat capacitance terms. This means that the Jacobian term associated with the internal energy rate is diagonal. This approach is especially effective when strong latent heat effects are present. The second-order elements use conventional Gaussian integration. Thus, second-order elements are to be preferred for problems when the solution will be smooth (without latent heat effects), whereas the rst-order elements should be used in nonsmooth cases (with latent heat). The HEATCAP element is available for modeling lumped heat capacitance at a point. The associated concentrated lm and concentrated radiation loading options are specied by the user. These loading options are also allowed in coupled temperature-displacement and coupled thermalelectrical analysis.
Reference
Uncoupled heat transfer analysis, Section 6.5.2 of the ABAQUS Analysis Users Manual
2.11.16
2.11.2
This section describes the formulation used in the shell heat conduction elements in ABAQUS/Standard. The basis of the elements is a combination of piecewise quadratic interpolation of temperature through the thickness of the shell and either linear interpolation (in elements DS3 and DS4) or quadratic interpolation (in elements DS6 and DS8) on the reference surface of the shell. The isoparametric interpolation functions for the shell reference surface are identical in form to those used for the solid quadrilateral and triangular elements and can be found in Solid isoparametric quadrilaterals and hexahedra, Section 3.2.4, and Triangular, tetrahedral, and wedge elements, Section 3.2.6, respectively. Nodal temperature values are stored at a set of points through the thickness (points P below) at each node of the element (nodes N below). For the purpose of numerical integration of the nite element equations, a 2 2 2 Gauss integration scheme with a 2 2 2 nodal integration scheme for the internal energy and specic heat term is used for the quadrilateral element DS4 and a 3 2 3 Gauss integration scheme is used for the quadrilateral element DS8. Three- and six-point integration schemes are used for the triangular elements DS3 and DS6, respectively, the details of which can be found in Triangular, tetrahedral, and wedge elements, Section 3.2.6. Let 1 ; 2 be material coordinates of a point in the reference surface of the shell, and let s3 measure position through the thickness of the shell so that 0h= 0 z0 s3 h= 0 z , where h is the thickness of the shell, z0 is the offset of the reference surface from the midsurface as discussed in Transverse shear stiffness in composite shells and offsets from the midsurface, Section 3.6.8. The position of any point in the shell is given by
Procedures
where
xo n
is the position of a point in the reference surface, and is the unit normal to the reference surface of the shell.
(2.11.21)
where
M P s3
N N 1 ; 2
( ) ( ) NP
Z
is a piecewise parabolic interpolation, is an interpolator in the reference surface, and are nodal temperature values (at node N , point P through the thickness).
The basic heat energy balance is Equation 2.11.13, with the approximate Jacobian matrix for the Newton method based on
dt+1t
dU
d dV
@
@x
1k1
@d @x
dV
where d is the correction to the temperature solution at time t t. The derivation of this form is discussed in Uncoupled heat transfer analysis, Section 2.11.1. The form of these terms for shell heat
+1
@t+1t
@q
d dS;
2.11.21
conduction elements is obtained by introducing the interpolator, Equation 2.11.21, and neglecting the change in area, with respect to s3 , of surfaces parallel to the reference surface. The internal energy rate term (the rst term in Equation 2.11.13) contributes, to the residual,
Z
A
h=2
0h=2
P
_ U
t+1t ds3
dA;
Z
A
(s
of the shell, set up according to the standard convention in ABAQUS for such local systems in shells. The term is formed by rst introducing an intermediate set of temperature values,
1; s2 ; s3), where s1 and s2 measure distance along local base vectors e1 and e2 , in the reference surface
@P @P ; ; ; @s1 @s2
P
For the second term the temperature derivatives are taken with respect to a local orthogonal system
corresponding to each temperature value point through the thickness, at each section where integration through the thickness is performed. Since the number of temperature values on the section is the same as the number of integration points, M P is unity in the appropriate locations and zero everywhere else. Then we can interpolate to the section by
8 P > > P > @ > < @s1 > P > @ > > : @s2
where
9 MP M ;
8 M > N > > M > 8 M 9 < @N = @s1 > > @N M > > : @s2 8 9 > @ > > > @s > > 1> > > > > < @ > h = P = > @s2 > > > > > @ > > > > > > : ; @s3
8 P > > P > < i > @ @s1 > P > @ > > : @s2
2.11.22
where
=4
dM P =ds3
3T 2 32
0 0
MP
0 0
MP 5 :
"Z
h=
22
0 2
h=
3 Q
ds3
M dA;
Procedures
MQ where [ k ] is the local conductivity matrix and the same term multiplied by t+1t appears in the residual. Finally, the external ux terms contribute
N
A
h=
0 2
h=
M r ds3 dA +
P
Z
A
N N I A qA + I B qB dA
#
"
NN
A
dq dq IA N M dA + I B d d t+1t;A t+1t;B
IA IA
=1 =0
at point A through the thickness at all other points through the thickness,
and points A and B are on the top and bottom surfaces of the shell.
Reference
Three-dimensional conventional shell element library, Section 15.6.7 of the ABAQUS Analysis Users Manual
2.11.23
CONVECTION/DIFFUSION
2.11.3
CONVECTION/DIFFUSION
The formulation in this section describes a capability for modeling heat transfer with convection in ABAQUS/Standard. The resulting elements can be used in any general heat transfer mesh. These elements have a nonsymmetric Jacobian matrix: the nonsymmetric capability is invoked automatically if elements of this type are included in the model. Both steady-state and transient capabilities are provided. The transient capability introduces a limit on the time increment (the limit is dened below): the time increment is adjusted to satisfy this limit if necessary. The steady-state versions of the elements can be used in a transient analysis, which means that transient effects in the uid are not included in the model. The formulation is based on the work of Yu and Heinrich (1986, 1987).
Thermal equilibrium equation
Procedures
The thermal equilibrium equation for a continuum in which a uid is owing with velocity v, is
c
@
@t
+v
@ @x
0 @x 1
1 @x 0 q
@
dV +
Sq
n k
1 1 @x 0 qs
@
dS = 0;
where (x; t) is the temperature at a point, (x; t) is an arbitrary variational eld, () is the uid density, c() is the specic heat of the uid, k() is the conductivity of the uid, q is the heat added per unit volume from external sources, qs is the heat owing into the volume across the surface on which temperature is not prescribed (Sq ), n is the outward normal to the surface, x is spatial position, and t is time. Although most uids will have isotropic conductivity, so that k = k I (where k() is a scalar and I is a unit matrix), we provide for anisotropic conductivity to cover such cases as that of uid owing through a set of bafe plates whose conductivity is smeared into that of the uid. The boundary conditions are that (x) is prescribed over some part of the surface, S , and that the heat ux per unit area entering the domain across the rest of the surface, qs (x), is prescribed or is dened by convection and/or radiation conditions. For example, the boundary layer between uid convection elements and solid elements might be modeled by DINTERx-type elements. The boundary term in the thermal equilibrium equation denes
qs = @ 0n 1 k 1 @x :
This implies that qs is the ux associated with conduction across the surface onlyany convection of energy across the surface is not included in qs. This makes no difference if the surface is part of a solid body (where qs would be dened by heat transfer into the adjacent body), since then the normal velocity into that body, v 1 n, is zero. But it does make a difference when there is uid crossing the surface, asfor exampleon the upstream and downstream boundaries of the mesh. In this case the choice of qs for the natural boundary condition (instead of using the total ux crossing the surface) is desirable because it avoids spurious reections of energy back into the mesh as the uid ows through the surface.
2.11.31
CONVECTION/DIFFUSION
These equations are discretized with respect to position by using rst-order isoparametric elements. The uid velocity, v, is assumed to be known. (ABAQUS actually requires that the mass ow rate of the uid per unit area be dened, because this is generally more convenient for the user. The velocity is computed from the mass ow rate and the density of the uid.) t from the known solution at time t. The time discretization generates the solution at time t The interpolation for the temperature, , is dened over an element and over a time increment as
+1
for
NN
At = 1 0
where t is the time increment and t. The Petrov-Galerkin discretization proposed by Yu and Heinrich couples this linear interpolation with the weighting functions
1t
; At+1t =
1t ;
= W N N
NNA +
A +
1t dA v 1 @N N N ; 2 dt jvj @x
1 0 1t ; 1t2 v is the average uid velocity over the element; jvj is its magnitude; and h is a characteristic element length measure, dened below. and are control parameters. The term in the weighting is introduced to eliminate articial diffusion of the solution, while the term is introduced to avoid
where
A=6
numerical dispersion. Yu and Heinrich show that the optimal choices are
2 = coth 0 2
where
and
2 0
C; 3 1t :
h
= jv jh
c k
and
C = jv j
The above expression for yields negative values for very small uid velocities, which may destabilize the solution; hence, for low velocities dispersion control is switched off. The characteristic element length measure, h, is dened by Yu and Heinrich as follows. Let h be the isoparametric line across the element passing through its centroid. The projection of h in the direction of the uid velocity vector at the elements centroid is
h = h 1
v jv j :
2.11.32
CONVECTION/DIFFUSION
Then we dene
h as
h=
X jh j:
When is nonzero, these elements require that C for numerical stability. Since the weighting functions are biased (upwinding), they are discontinuous from one element to the next. Some care is, therefore, required in manipulating the weak form of the thermal equilibrium equation (see Hughes and Brooks, 1982). In particular, the usual integration by parts of the conduction term
can be performed only for the continuous part of the weighting functions used to discretize : otherwise, continuity of heat ux between elements is not assured. For convenience we write the discontinuous part of the weighting as
@ @ 1 k 1 @x @x
Procedures
PN
=h 2
A +
1t dA v 1 @N N : 2 dt jvj @x
N M
Z
W N c N M
M dAn + v 1 @Nx An dt @
+ A @Nx 1 k 1 @Nx @ @
@2N M n A dV @ x@ x Z Z 0 W N q dV 0 A N N qs dS = 0: An 0 P N k :
S
A
h v @N N M 2 jvj 1 @x N dV Z n c N N + h v 1 @N N v 1 @N M dV + AA 2 jv j @ x @x Z N M Z v @N N M h1t dA dAn n c v 1 @N v 1 @N dV + 4 dt dt c jvj 1 @x N dV + A jv @ x j @x Z @N N @N M h v @N N @2N M 1 k 1 @x 0 2 v 1 k: dV + AAn @x @ x@ x Z v @N N @ 2jN jM @ x 0 h1t dA An jvj 1 @ x k : @ x@ x dV Z Z4 dt Z N N + h v 1 @N N q dV 0 dA 1t v 1 @N N q dV 0 A N N qs dS = 0: 0A 2 jv j @ x dt V 2 jvj @ x V S We now integrate this equation from time t to t +1t to provide an average equilibrium statement c N N +
dAn dt
1t
A dt = 1;
Z dA dt = 0; dt
1t
2.11.33
CONVECTION/DIFFUSION
1t
and
+1 A dA
t
dt
dt = 0
Z
1t
dA A
dt
1 dt = ; 1t
Z
1t
AA +1
t
dt =
1t
AAt dt = ;
1 2
dA dAt+1t dA dAt dt = dt = 0; 1t dt dt 1t dt dt
N N M
1 dA t+1t dA t A A dt = 0 ; dt = 0 1t 1t dt 1t dt
Z
M ;t t M;t
to give
M;t
0 1 0 h c jv j 1 @Nx v 1 @Nx dV +1 0 4 v @ @ Z Z 2N + 1 @Nx 1 k 1 @Nx dV 0 h jvj 1 @Nx k : @@x@x dV 0 +1 + 2Z @ @ 4 v @ @2N h v @N + 4 jvj 1 @x k : @x@x dV 0 +1 0 1
M;t t N M M;t t M;t
M;t
Z
h v @N N + 2 jv j 1 @ x
N
q dV 0
N q dS = 0:
N s
For the steady-state case the third term in this equation is omitted. In both transient and steadystate forms the contribution of such a convective element to the system of equations for the heat transfer model is not symmetric, requiring the use of the nonsymmetric matrix storage and solution scheme.
Reference
Uncoupled heat transfer analysis, Section 6.5.2 of the ABAQUS Analysis Users Manual
2.11.34
CAVITY RADIATION
2.11.4
CAVITY RADIATION
The formulation described in this section provides a capability for modeling heat transfer with cavity thermal radiation (in addition to the radiation boundary conditions described in Uncoupled heat transfer analysis, Section 2.11.1). Cavities are dened in ABAQUS/Standard as collections of surfaces that are composed of facets. In axisymmetric and two-dimensional cases a facet is a side of an element; in threedimensional cases a facet can be a face of a solid element or a surface of a shell element. For the purposes of the cavity radiation calculations, each facet is assumed to be isothermal and to have a uniform emissivity. Based on the cavity denition, cavity radiation elements are created internally by ABAQUS. These elements can generate large matrices since they couple the temperature degree of freedom of every node on the cavity surface. Their Jacobian matrix is nonsymmetric: the nonsymmetric solution capability is automatically invoked if cavity radiation calculations are requested in the analysis. Both steady-state and transient capabilities are provided. The theory on which this cavity radiation formulation is based is well-known and can be found in Holman (1990) and Siegel and Howell (1980). This section describes the formulation of the cavity radiation ux contributions and respective Jacobian for the Newton method used for the solution of the nonlinear radiation problem. The geometrical issues associated with the calculation of radiation viewfactors necessary in the formulation are addressed in Viewfactor calculation, Section 2.11.5.
Thermal radiation
Procedures
Our formulation is based on gray body radiation theory that means that the monochromatic emissivity of the body is independent of the wavelength of propagation of the radiation. Only diffuse (nondirectional) reection is considered. Attenuation of the radiation in the cavity medium is not considered. Using these assumptions together with the assumption of isothermal and isoemissive cavity facets, we can write the radiation ux per unit area into a cavity facet as
qi
= i A
i j
X X
j
Fik Ckj
01
( j
0 Z ) 4 0 ( i 0 Z ) 4
(no summation)
(2.11.41)
where
Cij = ij
0 0 (1 A i) Fij ;
i
and Ai is the area of facet i (seeing all cavity facets j = 1; n); i; j are the emissivities of facets i; j ; is the Stefan-Boltzmann constant; Fij is the geometrical viewfactor matrix; i ; j are the temperatures of facets i; j ; Z is the absolute zero on the temperature scale used; and ij is the Kronecker delta. In the special case of blackbody radiation, where no reection takes place (emissivity equal to one), Equation 2.11.41 reduces to
c qi
i Ai
X
j
j Fij
( j
0 Z ) 4 0 (
0 Z )4
(2.11.42)
2.11.41
CAVITY RADIATION
Spatial interpolation
The variables used to solve the discrete approximation of the heat transfer problem with cavity radiation are the temperatures of the nodes on the cavity surface. Since we assume that for cavity radiation purposes each facet is isothermal, it is necessary to calculate an average facet temperature radiation power. To do so, we rst dene temperature radiation power as
i
i
0 Z
14
N
N
0 Z
14
where the subscript i refers to facet quantities and the superscript N refers to nodal quantities. Then, we interpolate the average facet temperature radiation power from the facet nodal temperatures as
i
where N is the number on nodes forming the facet and PiN are nodal contribution factors calculated from area integration as Z
PiN
1 =A
PiN N ;
(2.11.43)
where NiN are the interpolation functions for facet i. The radiation ux into facet i can now be written as
NN i Ai i
dAi ;
c Q i = qi A i =
where
Rij (j
0 i ) ;
and
Dij =
0 Fik Ckj1:
Rij j ;
(2.11.44)
!
where
Rij = Rij
Rik
ij :
2.11.42
CAVITY RADIATION
The nodal contributions from the radiation ux on each facet can now be written as
QN i
Ai
c qi NiN dAi
= PiN Qi ;
Procedures
QN
QN i
PiN Qi :
QN
where
NM M ;
PiN Rij PjM :
(2.11.45)
NM
XX
c The radiation ux qi is evaluated based on temperatures at the end of the increment, coordinates at the end of the increment, and emissivities at the beginning of the increment. Any time variation of the coordinates during the heat transfer analysis is predened as translational and/or rotational motion and, therefore, provides no contribution to the Jacobian. Any variation of the emissivities as a function of temperature and predened eld variable changes with time is treated explicitly (values at the beginning of the increment are used) and, therefore, also provides no contribution to the Jacobian. The user can specify the maximum allowable emissivity change during an increment of the heat transfer analysis. Thus, the only Jacobian contribution is provided by temperature variations. The Jacobian contribution arising from the cavity radiation ux is then written trivially as
J NM
= @QM = 4RNM @
M
0 Z
13
(no summation):
(2.11.46)
In all practical cases the Jacobian is unsymmetric. This exact unsymmetric Jacobian is always used when cavity radiation analysis is performed.
Reference
2.11.43
VIEWFACTOR CALCULATION
2.11.5
VIEWFACTOR CALCULATION
Cavity radiation occurs when surfaces of the model can see each other and, thus, exchange heat with each other by radiation (Figure 2.11.51). Such exchange depends on viewfactors that measure the relative interaction between the surfaces composing the cavity. Viewfactor calculation is rather complicated for anything but the most trivial geometries. ABAQUS offers an automatic viewfactor calculation capability for two- and three-dimensional cases as well as for axisymmetric situations. This capability can take into account general surface blocking (or shadowing) as well as the most common forms of radiation symmetry. The viewfactor calculation can also be automatically repeated a number of times throughout the analysis history (this is user-controlled) if cavity surfaces are moved in space causing the viewfactors to change.
Procedures
Figure 2.11.51
ABAQUS, Inc., is pleased to acknowledge that the viewfactor calculation technology implemented in ABAQUS was derived from technology originally developed by the Atomic Energy Authority of the United Kingdom; see, for example, Johnson (1987). The remainder of this section contains a general description of this technology.
Viewfactor between elementary areas
cosi cosj dAi dAj ; (2.11.51) R2 Ai Aj where R is the distance between the two areas and i; j are the angles between R and the normals to the surfaces of the areas (Figure 2.11.52). This formula applies when the areas Ai and Aj are small compared with the distance R. We compute viewfactors in an area-lumped fashion, so that the Fij =
integral of Equation 2.11.51 would become
and
Fij =
(2.11.52)
VIEWFACTOR CALCULATION
dA i i
ni
R j nj
dA j
Figure 2.11.52
However, this expression is singular in Rij , so it is not well behaved for general surfaces. Thus, we use
Fij =
4 Ai Aj
2
atan
p
"p
#
(2.11.53)
Cavities are composed of surfaces; and surfaces, in turn, are made up of nite element faces. For the purpose of viewfactor calculations, one can then think of a cavity as a collection of element faces (or facets) corresponding to the nite element discretization around the cavity.In the two- and three-dimensional cases the element faces composing cavities can be treated as elementary areas and, thus, Equation 2.11.51 applies. In axisymmetric cases the element faces represent rings so that the viewfactors involve two ring surfaces looking at each other. This requires an integration over 2 where it is important to account for horizon effects (Johnson, 1987). In so far as the viewfactor calculations are concerned, rst- and second-order element faces are treated similarly in the sense that the midside nodes of the faces in the second-order elements are ignored. This means that a pair of four-noded faces looking at each other will produce the same viewfactor as a pair of eight-noded faces with corner nodes coinciding with the nodes of the fournoded faces.
2.11.52
VIEWFACTOR CALCULATION
Radiation blocking
Radiation within a cavity implies that every surface exchanges heat with every other surface. The problem is made more complex when solid bodies are interposed between radiating surfaces blocking (or shadowing) off some but not all the possible paths along which heat can be radiated from the facets of one surface to the facets of another surface (Figure 2.11.53).
A
Procedures
Figure 2.11.53
It is inconceivable that the user could handle this complexity in all but the simplest situations. Therefore, by default, ABAQUS automatically checks if blocking takes place for every possible radiation path in a cavity. This requires that the program check if the ray joining the centers of each pair of facets intersects any other facet. For cavities with a large number of facets this can be very time consuming. For this reason ABAQUS allows the user to guide its blocking algorithm by accepting input of which surfaces cause blocking, thus signicantly reducing the computational effort required. If a ray between two facets intersects any other facet, then in the two- and three-dimensional cases that ray is eliminated and no radiative heat transfer takes place between the facets. In the axisymmetric case blocking is much more complicated since each element face in the nite element model represents a ring. This is handled automatically and requires that the program calculate which part of the 2 extent of the ring is blocked.
Radiation symmetries
Use of symmetry can greatly reduce the size of a problem, butin the case of cavity radiation it requires that special facilities for denition and handling of symmetries be available. ABAQUS provides capabilities for three different kinds of symmetries: simple reection symmetry, periodic symmetry, and cyclic symmetry. Reection symmetry allows one additional image of the model to be created by reection through a line in two dimensions or reection through a plane in three dimensions. Periodic symmetry can be used to create multiple images of the model by periodic repetition in twoor three-dimensional space according to a periodic distance vector. Cyclic symmetry creates multiple
2.11.53
VIEWFACTOR CALCULATION
images of the model by cyclic repetition about a point in two dimensions or by cyclic repetition about an axis in three dimensions. Combinations of the different types of symmetry are supported. To illustrate the handling of symmetries during viewfactor calculation, consider the case of a simple reection symmetry in two-dimensional space (Figure 2.11.54). Radiation between facet i (with its centroid at point A) and facet j (with its centroid at point B ) has two contributions: one arising from the ray between points A and B and the other coming from the ray between points A and 0 0 B , where B is the mirror image of B . The length of ray AB is dened directly in the model. The denition of the length of ray AB 0 requires that point C on the reection symmetry line be located such that AC and BC make equal angles to it; ray AB 0 then has length AC + BC . Similar logic can be extended to the three-dimensional case.
j B C B'
Figure 2.11.54 Reection symmetry example. In axisymmetric cases symmetry about the axis of symmetry of the model is always implied, and the only other symmetries allowed are simple reection through a plane normal to the axis of symmetry or periodic repetition in the direction of the axis of symmetry.
Viewfactor checking
The viewfactor is a purely geometrical quantity, and it has some special properties. One property that allows us to check the accuracy of the calculation is that for a completely enclosed cavity:
1
Ai
X
j
Fij
=1
(2.11.54)
indicating that any ray from surface surface in the enclosed cavity.
2.11.54
VIEWFACTOR CALCULATION
The quantity in Equation 2.11.54 is calculated for every facet of each cavity, and its value is used to provide a check to control the accuracy of viewfactor calculation.
Radiation to ambient
The quantity calculated in Equation 2.11.54 can deviate from unity so long as the cavity is not fully enclosed. The user can dene such an open cavity by giving the value of the ambient temperature in the cavity denition. In this case the difference between one and the quantity calculated in Equation 2.11.54 for each facet of the open cavity is considered to be the fraction radiating from that facet to the surrounding medium.
Procedures
References
Thermal contact properties, Section 22.2.1 of the ABAQUS Analysis Users Manual Cavity radiation, Section 24.1.1 of the ABAQUS Analysis Users Manual
2.11.55
2.12.1
Joule heating arises when the energy dissipated by an electrical current owing through a conductor is converted into thermal energy. ABAQUS/Standard provides a fully coupled thermal-electrical procedure for analyzing this type of problem. Coupling arises from two sources: the conductivity in the electrical problem is temperature dependent, and the internal heat generated in the thermal problem is a function of electrical current. The thermal part of the problem includes all the heat conduction and heat storage (specic and latent heat) features described in Uncoupled heat transfer analysis, Section 2.11.1. (Forced heat convection caused by uid owing through the mesh is not considered.) The thermal-electrical elements have both temperature and electrical potential as nodal variables. This section describes the governing equilibrium equations, the constitutive model, boundary conditions, the surface interaction model, nite element discretization, and the components of the Jacobian used.
Governing equations
Procedures
The electric eld in a conducting material is governed by Maxwells equation of conservation of charge. Assuming steady-state direct current, the equation reduces to
rc dV; (2.12.11) S V where V is any control volume whose surface is S , n is the outward normal to S , J is the electrical current density (current per unit area), and rc is the internal volumetric current source per unit volume. The divergence theorem is used to convert the surface integral into a volume integral:
J 1 n dS =
Z
V
@x
J 0 rc
dV
= 0;
and since the volume is arbitrary, this provides the pointwise differential equation
@x @
1 J 0 rc = 0 :
dV
The equivalent weak form is obtained by introducing an arbitrary, variational, electrical potential eld, ', and integrating over the volume:
Z
V
'
@x
1 J 0 rc
Z
S
= 0:
Z
V
' rc dV;
Using rst the chain rule and then the divergence theorem, this statement can be rewritten as
V @x
@'
1 J dV =
' J dS
2.12.11
where
J def 0J 1 n =
Constitutive behavior
J = E 1 E;
where E (; f ) is the electrical conductivity matrix; is the temperature; and f ; = 1; 2::: are any predened eld variables. The conductivity can be isotropic, orthotropic, or fully anisotropic. E(x) is the electrical eld intensity dened as
E = 0 @' : @x
(2.12.12)
Since a potential rise occurs when a charged particle moves against the electrical eld, the direction of the gradient is opposite to that of the electrical eld. Using this denition of the electrical eld, Ohms law is rewritten as
J = 0 E 1 @' : @x
(2.12.13)
The constitutive relation is linear; that is, it assumes that the electrical conductivity is independent of the electrical eld. Introducing Ohms law, the governing conservation of charge equation becomes
@' E @' 1 1 @ x dV V @x
Z
=
' rc dV
Z
+
' J dS:
(2.12.14)
The heat conduction behavior is described by the basic energy balance relation
Z
V
U dV
_
Z
+
@ @ 1 k 1 @ x dV @x V
Z
=
r dV
Z
+
q dS;
(2.12.15)
where V is a volume of solid material, with surface area S ; is the density of the material; U is the internal energy; k is the thermal conductivity matrix; q is the heat ux per unit area of the body, owing into the body; and r is the heat generated within the body. The thermal problem is discussed in detail in Uncoupled heat transfer analysis, Section 2.11.1. Equation 2.12.14 and Equation 2.12.15 describe the electrical and thermal problems, respectively. Coupling arises from two sources: the conductivity in the electrical problem is temperature dependent, E = E (), and the internal heat generation in the thermal problem is a function of electrical current, r = rec (J), as described below.
2.12.12
Joules law describes the rate of electrical energy, Pec, dissipated by current owing through a conductor as Pec = E 1 J: Using Equation 2.12.12 and Equation 2.12.13, Joules law is rewritten as
Pec
=E
E
E:
Procedures
In a steady-state analysis Pec is evaluated at time t Pec is obtained over the increment
Pec
1Z = 1t =E
1
E
1t
1
Pec dt
where
E and E are values at time t + 1t. The amount of this energy released as internal heat is
r
E 0 E 1 E 1 1E + 1 1E 1 E 1 1E; 3
= v Pec;
The surfaceS of the body consists of parts on which boundary conditions can be prescribedSp and parts that can interact with nearby surfaces of other bodiesSi . Prescribed boundary conditions ' x; t ; temperature, x; t ; electrical current density, include the electrical potential, ' J J x; t ; heat ux, q q x; t ; and surface convection and radiation conditions. The surface interaction model includes heat conduction and radiation effects between the interface surfaces and electrical current owing across the interface. Heat conduction and radiation are modeled by
= ( )
= ( ) = ( )
qc
= ( )
= k g ( B 0 );
and
qr
= F B ( B 0 z ) 4 0 F ( 0 z ) 4 ; = ( + )
respectively, where is the temperature on the surface of the body under consideration, B is the temperature on the surface of the other body, z is the value of absolute zero temperature on the temperature scale being used, kg ; f is the gap thermal conductance, 1 B is the average 2 1 f f is the average of any predened eld variables at A and B, interface temperature,f 2 A B and F and FB are constants.
( ) = ( + )
2.12.13
= g (' B 0 ' );
where ' is the electrical potential on the surface of the body under consideration, 'B is the electrical potential on the surface of the other body, and g (; f ) is the gap electrical conductance. The electrical energy dissipated by the current owing across the interface,
Pec = J ('B
B qec = (1
0 f )g Pec;
where g is an energy conversion factor and f species how the total heat is distributed between the interface surfaces. Pec is evaluated at the end of the time increment in a steady-state analysis, and an averaged value over the time increment is used in a transient analysis. This is described in detail in Heat generation caused by electrical current, Section 5.2.6. Introducing the surface interaction effects and electrical energy released as thermal energy, the governing electric and thermal equations become
V @x
and
@'
1 E 1 @' dV =
@x
Z
V
' rc dV +
Z
Sp
' J dS +
Z
Si
' g ('B
0 ') dS;
(2.12.16)
Z
V
_ U dV + +
Z V @x
Sp
@
@ 1 k 1 @ x dV
Z
=
Z
V
r dV +
v Pec dV
q dS +
Si
(2.12.17)
Spatial discretization
In a nite element model equilibrium is approximated as a nite set of equations by introducing interpolation functions. Discretized quantities are indicated by uppercase superscripts (for example, 'N ). The summation convention is adopted for the superscripts. The discretized quantities represent nodal variables, with nodes shared between adjacent elements and appropriate interpolation chosen to provide adequate continuity of the assumed variation. The virtual electrical potential eld is interpolated by
' = NN 'N ;
2.12.14
where
N '
(N Z
are the interpolation functions. The discretized electrical equation is then written as
@N N @
@' @
x dV = Z
V
rc dV
+
N
Sp
J dS
Si
g ' B
( 0 ') dS
Z
V
@N N @
x 1 1 @ x dV 0
E
@'
rc dV
Z
Sp
J dS
Z
Si
N N g ' B
( 0 ') dS = 0: (2.12.18)
Procedures
The temperature eld in the thermal problem is approximated by the same set of interpolation functions:
= NP P :
@N P
Using these interpolation functions and a backward difference operator to integrate the internal energy rate, U , the thermal energy balance relation is obtained:
P I
1Z = 1t
Z
V
N P Ut+1t
N P v Pec dV
0 Ut ) dV + Z
Sp
Z @ 1 k 1 @ x dV 0 N P r dV V Z @x V P P N q dS 0 N (qc + qr + qec) dS = 0: Z
Si
(2.12.19)
Jacobian contributions
The Jacobian contributions are obtained by taking variations of Equation 2.12.18 and t. Equation 2.12.19 with respect to the electrical potential, ', and the temperature, , at time t This yields
+1
KN M = @'' = '' M
@I N
@I N
@N N
V
1 E 1 @Nx @x @
@N N
dV
Si
N N g N M dS;
Z E 1 @ 1 E N Q dV 0 1 N N @g ('B 0 ') N Q dS; @ 2 S @ V @x Z Z @I P 1 M KP M = @' = N P v J 0 3 1J 1 @Nx dV + 2 N P f g g ('B 0 ') N M dS; ' M @ V S Z Z P P Q P @ 1Z KP Q = @I = 1t N P dU N Q dV + @N 1 k 1 @Nx dV + @N 1 @ k 1 @ x N Q dV Q @ @ @ V @x V @x Z V d @ E @ E 1 1E 1 @E 1 1E N Q dV 0 N P v E 1 @ 1 E 0 E 1 @ 1 1E + 3 @ Z ZV @qr @qec @qc P @q Q P 0 N +@ N dS 0 N @ + @ + @ N Q dS: KN Q = @' = 0 ' Q Z
i i
Sp
Si
2.12.15
The term @q=@ in the P Q component includes prescribed surface convection and radiation conditions. The surface interaction terms @qc =@, @qr =@, and @qec =@ are evaluated in Heat generation caused by electrical current, Section 5.2.6. The Jacobian contributions give rise to an unsymmetric system of equations, requiring the use of the nonsymmetric matrix storage and solution scheme.
Reference
Coupled thermal-electrical analysis, Section 6.6.2 of the ABAQUS Analysis Users Manual
2.12.16
MASS DIFFUSION
2.13.1
ABAQUS/Standard provides for the modeling of the transient or steady-state diffusion of one material through another, such as the diffusion of hydrogen through a metal (Crank (1956), deGroot and Mazur (1962)). The governing equations are an extension of Ficks equations, to allow for nonuniform solubility of the diffusing substance in the base material. The basic solution variable (used as the degree of freedom at the nodes of the mesh) is the normalized def concentration (often referred to as the activity of the diffusing material), = c=s, where c is the mass concentration of the diffusing material and s is its solubility in the base material. This means that when the mesh includes dissimilar materials that share nodes, the normalized concentration is continuous across the interface between the different materials. Since is the square root of the partial pressure of the diffusing phase, the partial pressure is the same on both sides of the interface; Sieverts law is assumed to hold at the interface.
Governing equations
Procedures
The diffusion problem is dened from the requirement of mass conservation for the diffusing phase:
dV + n 1 J dS = 0; (2.13.11) V dt S where V is any volume whose surface is S , n is the outward normal to S , J is the ux of concentration of the diffusing phase, and n 1 J is the ux of concentration leaving S . Using the divergence theorem,
dc
Z
V
dc
dt
@x
dV = 0:
@x
J = 0:
Z
V
dc dt
@x
dV = 0;
where is an arbitrary, suitably continuous, scalar eld. This statement can be rewritten as
Z dc
V
dt
@x
( J)
0J1
@ @x
dV = 0:
2.13.11
MASS DIFFUSION
Z
V
Constitutive behavior
dc dt
0 @x 1 J
@
dV +
n 1 J dS = 0:
(2.13.12)
The diffusion is assumed to be driven by the gradient of a chemical potential, which gives the general behavior
@p J = 0sD 1 @ + s @@x ln( 0 Z ) + p @x ; (2.13.13) @x where D(c; ; f ) is the diffusivity; s(; f ) is the solubility; s(c; ; f ) is the Soret effect factor, providing diffusion because of the temperature gradient; is the temperature; Z is the absolute zero on the temperature scale used; p (c; ; f ) is the pressure stress factor, providing diffusion driven by the def gradient of the equivalent pressure stress, p = 0trace( )=3; and f are any predened eld variables.
An example of a particular form of this constitutive model is the assumption made for hydrogen diffusion in a metal:
@ J = 0 R(DcZ ) 1 @x ; 0
0 Z ) ln + pV H ;
where 0 is a xed datum, R is the universal gas constant, and V H is the partial molar volume of hydrogen in the solid solution. This form is similar to that used by Sofronis and McMeeking (1989) and results in a constitutive expression of the form
J = 0sD 1 @x + ln @x
@ s = ln
@ 0
Z1 ln( 0 ) +
R (
VH
0 Z ) @ x
@p
:
To implement this particular form, data for s and p must be calculated from the equations
and
p =
R (
VH
0 Z ) :
Changing variables (c = s) and introducing the constitutive assumption of Equation 2.13.13 into Equation 2.13.12 yields
Z
s d dt ds d d dt
+
+
@
1 sD 1 @x
@ @
(
0 Z ) @ x
s
@
+
p @x
@p
dV =
q dS; (2.13.14)
where
q =
def
0n 1 J
2.13.12
MASS DIFFUSION
Equilibrium in a nite element model is approximated by a nite set of equations through the introduction of appropriate interpolation functions. Discretized quantities are indicated by uppercase superscripts (for example, N ). The summation convention is adopted for the superscripts. These represent nodal variables, with nodes shared between adjacent elements and appropriate interpolation chosen to provide adequate continuity of the assumed variation. The interpolation is based on material coordinates Si, i = , 2, 3. The virtual normalized concentration eld is interpolated by
Procedures
= NN N ;
where
NN (Si) are interpolation functions. Then, the discretized equations are written as
s d
(2.13.15) Time integration in transient problems utilizes the backward Euler method (the modied CrankNicholson operator). Adopting the convention that any quantity not explicitly associated with a point in time is taken at t+1t, we can drop the subscript t+1t and write the integrated equations as
NN
Z
NN
( 0 t) + ds d + @ NN 1 sD 1 @ + s @ + p @p dV = Z NN q dS: s 1t d dt @x @x ( 0 Z ) @ x @x S
(2.13.16)
Jacobian contribution
The Jacobian contribution from the conservation equation is obtained from the variation of t. This yields Equation 2.13.16 with respect to at time t
@ @ s @p NN 1t d + ds d d + @@N 1 s @D 1 @x + ( 0sZ ) @x + p @x d dt x @ V @d @ NN 1 @ @s d + @p @p d dV: + @x 1 sD 1 @x + ( 0 Z ) @x @ @ x @ N dN , we obtain Rearranging and using the interpolation d = N Z s ds d N NM + d dt NN NM + @@N 1 sD 1 @ @x x V 1t
Z
+1
d
2.13.13
MASS DIFFUSION
@ NN @
+s
@ D @
@
1
x + ( 0 Z ) @ x + p @ x
@
@p
+
D1
(
0 Z ) @ x @
@ @
s + @p @p
@
x @
M N
dV:
Inspecting the above equation, we observe that the Jacobian becomes unsymmetric whenever the diffusivity, ; the temperature-driven diffusion coefcient, s ; or the pressure-driven diffusion coefcient, p , is dened as a function of concentration.
Reference
Mass diffusion analysis, Section 6.8.1 of the ABAQUS Analysis Users Manual
2.13.14
SUBSTRUCTURE ANALYSIS
2.14.1
The basic substructuring idea is to consider a substructure (a part of the model) separately and eliminate all but the degrees of freedom needed to connect this part to the rest of the model so that the substructure appears in the model as a substructure: a collection of nite elements whose response is dened by the stiffness (and mass) of these retained degrees of freedom denoted by the vector, fuR g. In ABAQUS/Standard the response within a substructure, once it has been reduced to a substructure, is considered to be a linear perturbation about the state of the substructure at the time it is made into a substructure. Thus, the substructure is in equilibrium with stresses 0 , displacements u0, and other state variables h0 when it is made into a substructure. Then, whenever it responds as a substructure, the total value of a displacement or stress component at some point within the substructure is
Procedures
where bLR x c and bLR x c are linear transformations between the retained degrees of freedom of the u substructure and the component of displacement or stress under consideration. The substructure must be in a self-equilibriating state when it is made into a substructure (except for reaction forces at prescribed boundary conditions that are applied to internal degrees of freedom in the substructure). If the substructure has been loaded to a nonzero state with some of its retained degrees of freedom xed, these xities are released at the time the substructure is created and any reaction forces at them converted into concentrated loads that are part of the preload state. This means that the contribution of the substructure to the overall equilibrium of the model is dened entirely by its linear response. Since the purpose of the substructuring technique is to have the substructure contribute terms only to the retained degrees of freedom, we need R to dene its external load vector fP g, formed from the nonzero substructure load cases applied to the R substructure, and its internal force vector, fI g, as a sum of linear transformations of the retained variables f uRg and their velocities and accelerations:
()
()
We refer to M as the reduced mass matrix for the substructure, C as its reduced damping matrix, and K as its reduced stiffness. These reduced mass, damping, and stiffness matrices connect the retained degrees of freedom only. The reduced stiffness matrix is easily derived when only static response is considered. Since the response of a substructure is entirely linear, its contribution to the virtual work equation for the model of which it is a part is
[ ]
[ ]
[ ]
W
= b uR
uE
where f P Rg and f P E g are consistent nodal forces applied to the substructure during its loading as a substructure (they do not include the self-equilibriating preloading of the substructure) and
1P R 0 K RR 1P E K ER
K RR K ER K RE K EE
K RE K EE
1uR ; 1 uE
[K ] =
2.14.11
SUBSTRUCTURE ANALYSIS
is its tangent stiffness matrix. Since the internal degrees of freedom in the substructure, fuE g, appear only within the substructure, the equilibrium equations conjugate to fuE g in the contribution to the virtual work equation given above are complete within the substructure, so that
(2.14.11)
= 4uR5 0(f1P Rg 0 [K RE ][K EE ]01f1P E g) 0 ([K RR] 0 [K RE ][K EE ]01[K ER])f1uRg1: [K ] = [K RR] 0 [K RE ][K EE ]01[K ER];
and the contribution of the substructure load cases applied to the substructure is the load vector
fuE g = 0[K EE ]01[K ER ]fuRg + fE gq fuE g = 0[K EE ]01[K ER ]fuR g + fE gq_ _ _ E g = 0[K EE ]01[K ER ]fuRg + fE g q: fu
The fE g are the eigenmodes of the substructure, obtained with all retained degrees of freedom constrained, and the q are the generalized displacementsthe magnitudes of the response in these normal modes.
2.14.12
SUBSTRUCTURE ANALYSIS
The contribution of the substructure to the virtual work equation for the dynamic case is
f uR
uE g
1P R 0 M RR 1P E M ER
M RE M EE
uR uE
0 C RR ER C RR K
0
C RE C EE
K ER
_ _ RE 1uR K K EE 1 uE ;
uR uE
where
[M ] = [C ] =
M EE M RE C EE C RE
M ER M RR C ER C RR
Procedures
R 1P fP g = 1 P E
is the nodal force vector in the substructure. With the assumed dynamic response within the substructure, the internal degrees of freedom in this contribution ( uE and its time derivatives) can be transformed to the retained degrees of freedom and the normal mode amplitudes, reducing the system to
b uR
where
q
[T ]T fP g 0 [T ]T [M ][T ]
uR q
in which E is the matrix of eigenvectors, fqg is the vector of generalized degrees of freedom, I is a is a null matrix. unit matrix, and
[ ]
[0]
[]
Large-rotation substructures
Large-rotation substructures require rst the computation of an equivalent rigid body rotation matrix associated with the substructures motion. Since the substructure exhibits only small deformations, one can use the original and current positions of two nodes in two-dimensional analyses or three nodes in three-dimensional analyses to compute two rectangular local systems and then the rotation matrix. For example, in three dimensions ABAQUS/Standard computes a reference (average) point using the 1 three nodes in the original conguration. The rst unit direction vector, Ej , points from this average 3 , is taken to be the normal to the plane dened by the point to the rst node. The third direction, Ej 2 three nodes and the second direction, Ej , is simply the cross-product of the third and rst directions. The process is repeated in the current conguration to compute a local system, ek . The rotation matrix i k can then be easily computed as Rij ek Ej . i
2.14.13
SUBSTRUCTURE ANALYSIS
ABAQUS/Standard automatically picks the two or three nodes used for the computation of the from the substructures retained nodes. In most cases only retained nodes with at rotation matrix least all translational degrees of freedom retained can be canditates. For example, in three-dimensional analyses the rst node used for the equivalent rigid body calculation is chosen to be the node with the highest stiffness (largest diagonal value) in the substructure. The second node is chosen to be the retained node farthest apart from the rst node with the provision that its nodal stiffness is high enough (at least 0.01% of the stiffness of the rst node). The third node is picked to be the retained node for which the distance to the line dened by the rst and second node is maximum (with the same stiffness requirement as for the second node). In the rare case when less than three (in threedimensional analyses) valid candidate nodes are retained, the matrix is computed directly from the nodal rotations of the stiffest node with all rotational degrees of freedom retained. To compute internal forces associated with a substructure in large rotations, ABAQUS/Standard computes strain-inducing displacements/rotations by subtracting the rigid body motion from the substructures nodal displacements/rotations. For translational degrees of freedom the strain-inducing displacements u at a node can be computed using
u = x 0 x0 0 R(X 0 X0 );
where X and x are the original and current positions of the node and X0 and x0 are the original and current coordinates of an average point (calculated as outlined above). For rotational degrees of freedom the total rotation matrix at a node (Rtot ) is the compound rotation between the straininducing rotation matrix and the rigid body rotation
hi ^ Rtot = exp R:
F = Krot u; Krot = RKRT
u= u
is the strain-inducing displacement-rotation vector. Similarly, in dynamics the reduced mass (including the coupling between nodal displacements/ rotations and eigenmode contributions) is rigid body rotated before any mass contributions are included in the virtual work associated with the substructure.
Fixed direction gravity loading
When gravity loading that acts in a xed direction is dened, ABAQUS/Standard will create internally at the generation level a number of load cases (two in two-dimensional analyses and three in threedimensional analyses) corresponding to unit gravity loads in the substructures directions. At the
2.14.14
SUBSTRUCTURE ANALYSIS
usage level the total rotation matrix of the substructure (includes both the user-specied rotation/ mirroring and the rotation of the substructure in nonlinear geometry analyses) is rst used to rotate for the gravity load. Thus, the user-specied directions are back the user-specied unit direction now expressed in the local (rotating) system associated with the substructure ( = T 1 ). The internally generated unit load cases are then scaled by the components of and by the appropriate magnitude and amplitude and then added to the external force in the model.
n R N
Reference
Procedures
2.14.15
SUBMODELING ANALYSIS
2.15.1
SUBMODELING ANALYSIS
Submodeling is the technique of studying a local part of a model with a rened mesh, based on interpolation of the solution from an initial, global model onto the nodes on the appropriate parts of the boundary of the submodel. The method is most useful when it is necessary to obtain an accurate, detailed solution in the local region and the detailed modeling of that local region has negligible effect on the overall solution. The response at the boundary of the local region is dened by the solution for the global model and it, together with any loads applied to the local region, determines the solution in the submodel. The technique relies on the global model dening this submodel boundary response with sufcient accuracy. Submodeling can be applied quite generally in ABAQUS. With a few restrictions different element types can be used in the submodel compared to those used to model the corresponding region in the global model. Both the global model and the submodel can use solid elements, or they can both use shell elements. A special option is available to use a submodel consisting of solid elements with a global model consisting of shell elements. The material response dened for the submodel may also be different from that dened for the global model. Both the global model and the submodel can have nonlinear response and can be analyzed for any sequence of analysis procedures. The procedures do not have to be the same for both models. The submodel is run as a separate analysis. The only link between the submodel and the global model is the transfer of the time-dependent values of variables to the relevant boundary nodes (the driven nodes) of the submodel. The only information in the global model available to the submodel analysis is the le output data written during the global model analysis. It contains, by default, the undeformed coordinates of all global model nodes and element information for all elements in the global model (see Results le output format, Section 5.1.2 of the ABAQUS Analysis Users Manual). The user must have requested nodal responses in the area where the submodel boundary is located. These responses are used to prescribe boundary conditions at the driven nodes in the submodel.
Interpolation procedure and tolerance checking
Procedures
In the solid-to-solid case the positions of the submodel boundary nodes (the driven nodes) are determined with respect to the global model, and the appropriate element interpolation functions are used to obtain the values of the degrees of freedom at the driven nodes. An exterior tolerance, which the user can specify, is used to check whether it is valid to extrapolate values from the global model. The extrapolation is valid if the distance between the driven nodes and the free surface of the global model falls within the specied tolerance. A similar check is done along the global model boundaries for the shell-to-shell submodeling case. We also check whether the driven nodes of the submodel lie sufciently close to the midsurface of the shell elements in the global model. To simplify the calculations, the closest point in the global model is approximated by measuring the distance in the direction normal to a at approximation to each shell element in the global model, as shown in Figure 2.15.11. For the shell-to-solid case ABAQUS uses two kinds of tolerances to determine the relation between the submodel and the global model. First, the closest point on the shell midsurface of the global model is determined. This point will subsequently be referred to as the image node of the driven node. The exterior tolerance
2.15.11
SUBMODELING ANALYSIS
plane approximation
Figure 2.15.11
parameter is used to check if the image node lies within the domain of the global model. Then the distance between the driven node and its image is checked against half of the maximum shell thickness specied by the user (see Figure 2.15.12).
global model shell elements
t A center zone D
shell midsurface
shell thickness AI
A - driven node AI - driven node image on the shell midsurface solid element submodel mesh
Figure 2.15.12 Center zone in shell-to-solid submodeling. If the node is within the half thickness plus the exterior tolerance, it is accepted. This check is only approximate if the global model has varying shell thickness, and in that case it will not protect
2.15.12
SUBMODELING ANALYSIS
the user in parts of the global model that have a small thickness compared to the maximum thickness specied by the user. After the locations of the driven nodes (or image nodes for the shell-to-solid case) are determined, the prescribed values of the driven variables are interpolated from the values written to the le output for the global model. These must have been written with a sufciently high frequency to obtain accurate values at the driven nodes. All components of displacements, temperatures, charges, and for complex steady-state dynamic analysisthe phase angles as well as the amplitudes have to be written for the global model nodes from which the values for the driven nodes will be interpolated. For small global models responses will typically be written for all nodes. For large global models node sets can be created that contain the nodes in the regions around the submodel boundary. For solid-to-solid and shell-to-shell submodeling, the interpolated values of displacements, rotations, temperatures, etc. are applied directly to the driven nodes. For these nodes the user can specify the individual degrees of freedom that are driven.
Driven variables for shell-to-solid submodeling
Procedures
In the shell-to-solid case the driven degrees of freedom are chosen automatically, depending on the distance between the driven node and the midsurface of the shell. If the node lies within the center zone (specied by the user; see Figure 2.15.12), all displacement components are driven. If the node lies outside the center zone, only the displacement components parallel to the shell midsurface are driven. By default, the size of the center zone is taken as 10% of the maximum shell thickness. The procedure is described in detail below. The center zone should be large enough so that it contains at least one layer of nodes. If the transverse shear stresses at the submodel boundary are high and the submodel is highly rened in the thickness direction, this can result in high local stresses, since the shear force at the submodel boundary is only transferred at the driven nodes within the center zone. High transverse shear stresses occur only in regions where bending moments vary rapidly, and it is better not to locate the submodel boundary in such regions. It is best to locate the submodel boundary in areas of low transverse shear stress in the global model. All displacement degrees of freedom are driven when the driven node lies within the center zone. For geometrically linear analysis these prescribed displacements are obtained from the displacements and rotations of the image node as
where A is the prescribed displacement of driven node A, AI and AI are the interpolated is the vector connecting the image node to displacement and rotation of the image node, and the driven node:
(2.15.11)
For large-displacement analysis nite rotations must be taken into account. The nite rotation equivalent of Equation 2.15.11 is
uA = uAI + (C 0 I) 1 D = uAI + d 0 D;
2.15.13
(2.15.12)
SUBMODELING ANALYSIS
where is the rotation matrix as dened in Rotation variables, Section 1.3.1; is the identity tensor; and is the rotated vector connecting the image node to the driven node in the current conguration:
d = C D:
1
For driven nodes outside the center zone only the displacement components parallel to the shell midsurface are driven. For the geometrically linear case this leads to the constraints
where 1 and 2 are two (unit) vectors orthogonal to geometrically nonlinear case are
= = ( ( + +
T1 uA = T1 uAI T2 1 uA T2 1 uAI
1 1
( (
AI 2 D); AI 2 D); +
+
(2.15.13)
D.
(2.15.14)
Since the submodeling capability in ABAQUS is quite general and allows the use of different procedure types in both analyses, there are several possibilities for the evaluation of the values at driven nodes as follows. In all cases ABAQUS assumes that the global model and the submodel both use small- or large-displacement theory. In the schemes listed below the rst procedure type applies to the global analysis and the second to the submodel analysis. 1. General procedure to general procedure for small-displacement theory: Equation 2.15.11 is used inside the center zone, and Equation 2.15.13 is used outside the center zone. 2. General procedure to general procedure for large-displacement theory: Equation 2.15.12 is used inside the center zone, and Equation 2.15.14 outside the center zone. 3. General procedure to linear perturbation procedure for small-displacement theory:
1uA uAI
=
+
AI 2 D 0 uA;
0
+ + ) )
(2.15.15)
T1 1 1uA T1 1 uAI AI 2 D 0 uA ; 0 A T2 1 uAI AI 2 D 0 uA ; T2 1 1u 0 outside the center zone; uA denotes the base state in the submodel. 0
= = ( (
(2.15.16)
1uA uAI d 0 D 0 uA 0
= +
(2.15.17)
2.15.14
SUBMODELING ANALYSIS
A 1uA = t1 (uAI + d 0 D 0 u0 ); (2.15.18) A = t 1 (uAI + d 0 D 0 uA) t2 1 1u 2 0 outside the center zone, where t denotes the tangent vector. The exact formulation would require the use of the base state normal vector d0 and the base state tangent vector t0 . Since they are not available, ABAQUS approximates them with the current normal vector d and current tangent vector t.
t1 1
1
uA = 1uAI + 1AI 2 D + uA 0
inside the center zone, and
Procedures
(2.15.19)
T1 1 uA = T1 1 (1uAI + 1AI 2 D + uA); 0 T2 1 uA = T2 1 (1uAI + 1AI 2 D + uA) 0 A denotes the base state in the submodel. outside the center zone; u0
6. Linear perturbation procedure to general procedure for large-displacement theory:
(2.15.110)
uA = 1uAI + 1AI 2 D + uA 0
inside the center zone, and
(2.15.111)
governed by the same equations. The approximation will give good results for cases with a small base state rotation eld in the global analysis. 7. Linear perturbation procedure to linear perturbation procedure for small-displacement theory:
T1 1 uA = T1 1 (1uAI + 1AI 2 D + uA); 0 (2.15.112) T2 1 uA = T2 1 (1uAI + 1AI 2 D + uA) 0 outside the center zone. Since the base state is not available, an approximate form is used, where D is used in place of d0 and T is used for t0 . With the above assumptions cases 5 and 6 are 1uA = 1uAI + 1AI 2 D
inside the center zone, and
(2.15.113)
T1 1 1uA = T1 1 (1uAI + 1AI 2 D); T2 1 1uA = T2 1 (1uAI + 1AI 2 D) 1uA = 1uAI + 1AI 2 D
inside the center zone, and
2.15.15
(2.15.114)
outside the center zone. 8. Linear perturbation procedure to linear perturbation procedure for large-displacement theory: (2.15.115)
SUBMODELING ANALYSIS
(2.15.116)
outside the center zone. Since the base state is not available, D is used in place of 0 and in place of 0 . With the above assumptions cases 7 and 8 are governed by the same equations. The approximation will give good results for cases with a small base state rotation eld in the global analysis.
Reference
2.15.16
J-INTEGRAL EVALUATION
2.16.1
J -INTEGRAL EVALUATION
The J -integral is widely accepted as a fracture mechanics parameter for both linear and nonlinear material response. It is related to the energy release associated with crack growth and is a measure of the intensity of deformation at a notch or crack tip, especially for nonlinear materials. If the material response is linear, it can be related to the stress intensity factors. Because of the importance of the J -integral in the assessment of aws, its accurate numerical evaluation is vital to the practical application of fracture mechanics in design calculations. ABAQUS/Standard provides a procedure for such evaluations of the J -integral, based on the virtual crack extension/domain integral methods (Parks, 1977, and Shih, Moran, and Nakamura, 1986). The method is particularly attractive because it is simple to use, adds little to the cost of the analysis, and provides excellent accuracy, even with rather coarse meshes.
Procedures
J -integral in two-dimensions
In the context of quasi-static analysis the J -integral is dened in two dimensions as
1 1 (2.16.11) 0 where 0 is a contour beginning on the bottom crack surface and ending on the top surface, as shown in Figure 2.16.11; the limit 0 ! 0 indicates that 0 shrinks onto the crack tip; is a unit vector in is given by the virtual crack extension direction; and is the outward normal to 0.
For elastic material behavior W is the elastic strain energy; for elastic-plastic or elastic-viscoplastic material behavior W is dened as the elastic strain energy density plus the plastic dissipation, thus representing the strain energy in an equivalent elastic material. This implies that the J -integral calculation is suitable only for monotonic loading of elastic-plastic materials.
H = W I 0 1 @u : @x
x2
x1
n
Figure 2.16.11 Contour for evaluation of the J -integral.
2.16.11
J-INTEGRAL EVALUATION
J
where
=0
C + C+ +0+ C0 and has the value q = q on 0 and q = 0 on C ; and m is the outward normal to the domain enclosed by the closed contour, as shown in Figure 2.16.12. m = 0n on 0; and t = m 1 is the surface traction on the crack surfaces C+ and C0 .
m
q is a sufciently smooth weighting function within the region enclosed by the closed contour
C+C+ +0+C0
m 1 H 1 q d0 0
C+ +C0
t 1 @u 1 q d0; @x
(2.16.12)
q C+ m n C A C
Figure 2.16.12 Closed contour C + C+ + 0 + C0 encloses a domain A that includes the crack-tip region as 0 ! 0: Using the divergence theorem, we convert the closed contour integral into the domain integral
C+ +C0 where A is the domain enclosed by the closed contour C + C+ + 0 + C0. It is worth noting that the domain A includes the crack-tip region as 0 ! 0. If equilibrium is satised and W is a function of the mechanical straini.e., W = W ("m )we have " "th "m @W @" @ 1 + f = 0 and @W = @" m : @"x = : @ x 0 @"x ; " @x @x @ @ th is the thermal strain. Substituting the above two where f is the body force per unit volume and "
equations into Equation 2.16.13 gives
J =0
Z @
A
1 (H 1 q ) d 0 0 @x
t 1 @ u 1 q d0 ; @x
(2.16.13)
2.16.12
J-INTEGRAL EVALUATION
=0
H : @q + @x A
"th f 1 @ u 0 : @"x @x @
Z 1 q d0 0
C+ +C0
t 1 @u 1 q d0: @x
(2.16.14)
To evaluate these integrals, ABAQUS denes the domain in terms of rings of elements surrounding the crack tip. Different contours (domains) are created. The rst contour consists of those elements directly connected to crack-tip nodes. The next contour consists of the ring of elements that share nodes with the elements in the the rst contour as well as the elements in the rst contour. Each subsequent contour is dened by adding the next ring of elements that share nodes with the elements in the previous contour. is chosen to have a magnitude of zero at the nodes on the outside of the contour and to be one (in the crack direction) at all nodes inside the contour except for the midside nodes (if they exist) in the outer ring of elements. These midside nodes are assigned a value between zero and one according to the position of the node on the side of the element.
Procedures
J -integral in three-dimensions
The J -integral can be extended to three dimensions by considering a crack with a tangentially continuous front, as shown in Figure 2.16.13. The local direction of virtual crack extension is again given by , which is perpendicular to the local crack front and lies in the crack plane. Asymptotically, as r ! 0, the conditions for path independence apply on any contour in the x1x2 plane, which is perpendicular to the crack front at s. Hence, the J -integral dened in this plane can be extended to represent the pointwise energy release rate along the crack front as
J (s) = lim
0!0 0
n 1 H 1 q d0 :
(2.16.15)
x2 (perpendicular to plane of crack) r x1 (normal to crack front) x3 (tangent to crack front) s crack front
Figure 2.16.13
Denition of local orthogonal Cartesian coordinates at the point s on the crack front; the crack is in the x1 x3 plane.
2.16.13
J-INTEGRAL EVALUATION
For a virtual crack advance (s) in the plane of a three-dimensional crack, the energy release rate is given by
(s)n 1 1 dA ; (2.16.16) 0!0 At where L denotes the crack front under consideration; dA is a surface element on a vanishingly small tubular surface enclosing the crack tip (i.e., dA = dsd0); and is the outward normal to dA. J can be calculated by the domain integral method similar to that used in two dimensions. To do so, we rst convert the surface integral in Equation 2.16.16 to a volume integral by introducing a contour surface Ao , outside surface At , external surfaces Aends at the ends of the crack front (the surfaces Aends vanish for the crack whose front forms a closed loop), and the crack faces Acracks, as shown in Figure 2.16.14. It can be seen that A = At + Ao + Aends + Acracks encloses a volume V . A weighting function is dened such that it has a magnitude of zero on Ao and = (s) on At . is assumed to vary smoothly between these values within A. On the external surfaces Aends where is not tangential to the surfaces, it must be made so. This can be done in ABAQUS by dening the surface normals explicitly. Then, we can rewrite Equation 2.16.16 as
J=
J (s)(s) ds = lim
H q
q q
Z m 1 H 1 q dA 0 t 1 @u 1 q dA; (2.16.17) A Aends +Acracks @ x where m is the outward normal to A (and m = 0n on At ). t = m 1 is the surface traction on
J =0
Ao
At
Acracks Aends V
Figure 2.16.14 Surface A = At + Ao + Aends + Acracks encloses a domain volume V that includes the crack-front region as 0 ! 0:
2.16.14
J-INTEGRAL EVALUATION
J =0
Z
V
Z "th H : @ q + f 1 @u 0 : @"x 1 q dV 0 @x @x @
Aends +Acracks
t 1 @u 1 q dA: @x
(2.16.18)
To obtain J (s) at each node set P along the crack front line, (s) is discretized with the same interpolation functions as those used in the nite elements along the crack front:
(s ) = N Q (s ) Q ;
where Q = 1 at the node set P and all other Q are zero. This expression for (s) is substituted into Equation 2.16.18. Finally, the J -integral value at each node set P along the crack front can be calculated as
Procedures
JP = JP =
N P ds:
(2.16.19)
Reference
Contour integral evaluation, Section 7.9.2 of the ABAQUS Analysis Users Manual
2.16.15
2.16.2
The stress intensity factors KI , KII , and KIII play an important role in linear elastic fracture mechanics. They characterize the inuence of load or deformation on the magnitude of the crack-tip stress and strain elds and measure the propensity for crack propagation or the crack driving forces. Furthermore, the stress intensity can be related to the energy release rate (the J -integral) for a linear elastic material through
where = bKI ; KII ; KIII cT and is called the pre-logarithmic energy factor matrix (Shih and Asaro, 1988; Barnett and Asaro, 1972; Gao, Abbudi, and Barnett, 1991; Suo, 1990). For homogeneous, isotropic materials is diagonal and the above equation simplies to
= 81 KT B01 K;
1 1
Procedures
where E = E for plane stress and E = E=(10 2) for plane strain, axisymmetry, and three dimensions. For an interfacial crack between two dissimilar isotropic materials with Youngs moduli E1 and E2, Poissons ratios 1 and 2, and shear moduli G1 = E1 =2(1 + 1) and G2 = E2 =2(1 + 2),
J=
where
1 0 2
E3
2 2 (KI + KII ) +
1 K2 ; 2G3 III
E3
and = 3 0 4 for plane strain, axisymmetry, and three dimensions; and = (3 0 )=(1 + ) for plane stress. Unlike their analogues in a homogeneous material, KI and KII are no longer the pure Mode I and Mode II stress intensity factors for an interfacial crack. They are simply the real and imaginary parts of a complex stress intensity factor, whose physical meaning can be understood from the interface traction expressions: (KI + iKII )ri" KIII p (22 + i12)=0 = ; (23 )=0 = p ; 2r 2r where r and are polar coordinates centered at the crack tip. The bimaterial constant " is dened as
"=
10 1 ln : 2 1 +
In this section we describe an interaction integral method (Shih and Asaro, 1988) to extract the individual stress intensity factors for a crack under mixed-mode loading. The method is applicable to cracks in isotropic and anisotropic linear materials.
2.16.21
J=
1 0 0 [K B 01K + 2KI B121 KII + 2KI B131 KIII 8 I 11 I + (terms not involving KI )]:
where I; II; III correspond to 1; 2; 3 when indicating the components of B . We dene the J -integral for an auxiliary, pure Mode I, crack-tip eld with stress intensity factor kI , as
I Jaux =
1 k 1 B 01 1 k : 8 I 11 I
I Jtot =
1 0 0 [(K + k )B 01 (K + k ) + 2(KI + kI )B121 KII + 2(KI + kI )B131 KIII 8 I I 11 I I + (terms not involving KI or kI )]:
KI
I or kI in Jtot and
k 0 0 0 I I I Jint = Jtot 0 J 0 Jaux = I (B111 KI + B121 KII + B131 KIII ): 4 If the calculations are repeated for Mode II and Mode III , a linear system of equations results: k 01 Jint = B K ; (no sum on = I; II; III ); 4 If the k are assigned unit values, the solution of the above equations leads to
K = 4B 1 Jint ;
I II III where int = bJint; Jint; Jint cT . The calculation of this integral is discussed next. Based on the denition of the J -integral, the interaction integrals Jint can be expressed as Jint = lim 0! 0
with
M given as
n 1 M 1 q d0
@u 0 1 @ u : @ x aux aux @ x
M = : " I 0 1 aux
aux represents three auxiliary pure Mode I, Mode II, and Mode III crack-tip elds for = I; II; III , respectively. 0 is a contour that lies in the normal plane at position s along the crack
The subscript front, beginning on the bottom crack surface and ending on the top surface (see Figure 2.16.21). The limit 0 ! 0 indicates that 0 shrinks onto the crack tip.
2.16.22
x2 (perpendicular to plane of crack) r x1 (normal to crack front) x3 (tangent to crack front) s crack front
Procedures
Figure 2.16.21
Denition of local orthogonal Cartesian coordinates at the point s on the crack front; the crack is in the x1 x3 plane.
Following the domain integral procedure used in ABAQUS/Standard for calculating the J -integral, we dene an interaction integral for a virtual crack advance (s):
Jint =
Jint(s)(s) ds =
(s ) n 1
M qdA ;
1
where L denotes the crack front under consideration; dA is a surface element on a vanishingly small tubular surface enclosing the crack tip (i.e., dA = dsd0); is the outward normal to dA; and is the local direction of virtual crack propagation. The integral Jint can be calculated by the same domain integral method as that used for calculating the J -integral. To obtain Jint at each node set P along the crack front line, is discretized with the same interpolation functions as those used in the nite elements along the crack front:
(s ) = N Q (s ) Q ;
where Q = 1 at the node set P and all other Q are zero. The result is substituted into the expression for Jint. Finally, the interaction integral value at each node set P along the crack front Z can be calculated as
P P Jint = Jint =
N P ds:
Reference
Contour integral evaluation, Section 7.9.2 of the ABAQUS Analysis Users Manual
2.16.23
T-STRESS EXTRACTION
2.16.3
T -STRESS EXTRACTION
The asymptotic expansion of the stress eld near a sharp crack in a linear elastic body with respect to r, the distance from the crack tip, is
ij
= pKI
2r
I fij () +
K KIII III II p2II fij () + p2r fij () + T 1i1j + (T + E"33)3i3j + O(r1=2) r
(Williams, 1957), where r and are the in-plane polar coordinates centered at the crack tip. The local axes are dened so that the 1-axis lies in the plane of the crack at the point of interest on the crack front and is perpendicular to the crack front at this point; the 2-axis is normal to the plane of the crack (and thus is perpendicular to the crack front); and the 3-axis lies tangential to the crack front. "33 is the extensional strain along the crack front. In plane strain "33 = 0; in plane stress the term (T + "33)3i 3j vanishes. The T -stress represents a stress parallel to the crack faces. It is a useful quantity, not only in linear elastic crack analysis but also in elastic-plastic fracture studies. The T -stress usually arises in the discussions of crack stability and kinking for linear elastic materials. For small amounts of crack growth under Mode I loading, a straight crack path has been shown to be stable when T < 0, whereas the path will be unstable and, therefore, will deviate from being straight when T > 0 (Cotterell and Rice, 1980). A similar trend has been found in three-dimensional crack propagation studies by Xu, Bower, and Ortiz (1994). Hutchinson and Suo (1992) also showed how the advancing crack path is inuenced by the T -stress once cracking initiates under mixed-mode loading. (The direction of crack initiation can be otherwise predicted using the criteria discussed in Prediction of the direction of crack propagation, Section 2.16.4.) The T -stress also plays an important role in elastic-plastic fracture analysis, even though the T stress is calculated from the linear elastic material properties of the same solid containing the crack. The early study of Larsson and Carlsson (1973) demonstrated that the T -stress can have a signicant effect on the plastic zone size and shape and that the small plastic zones in actual specimens can be predicted adequately by including the T -stress as a second crack-tip parameter. Some recent investigations (Bilby et al., 1986; Al-Ani and Hancock, 1991; Betegn and Hancock, 1991; Du and Hancock, 1991; Parks, 1992; and Wang, 1991) further indicate that the T -stress can correlate well with the tensile stress triaxiality of elastic-plastic crack-tip elds. The important feature observed in these works is that a negative T -stress can reduce the magnitude of the tensile stress triaxiality (also called the hydrostatic tensile stress) ahead of a crack tip; the more negative the T -stress becomes, the greater the reduction of tensile stress triaxiality. In contrast, a positive T -stress results only in modest elevation of the stress triaxiality. It was found that when the tensile stress triaxiality is high, which is indicated by a positive T -stress, the crack-tip eld can be described adequately by the HRR solution (Hutchinson, 1968; Rice and Rosengren, 1968), scaled by a single parameter: the J -integral; that is, J -dominance will exist. When the tensile stress triaxiality is reduced (indicated by the T -stress becoming more negative), the crack-tip elds will quickly deviate from the HRR solution, and J -dominance will be lost (the asymptotic elds around the crack tip cannot be well characterized by the HRR elds). Thus, using the T -stress (calculated based on the load level and linear elastic material properties) to characterize the triaxiality of the crack-tip stress state and using the J -integral (calculated based on the actual elastic-plastic deformation eld) to measure the scale of the
Procedures
2.16.31
T-STRESS EXTRACTION
crack-tip deformation provides a two-parameter fracture mechanics theory to describe the Mode I elasticplastic crack-tip stresses and deformation in plane strain or three dimensions accurately over a wide range of crack congurations and loadings. To extract the T -stress, we use an auxiliary solution of a line load, with magnitude f , applied in the plane of crack propagation and along the crack line:
L 11
f = r cos3 ;
L 22
f r
cos sin2 ;
L 13
L 12
f r
sin cos2 ;
L 33
f r
cos ;
L = 23 = 0:
L The term 33 = 0 for plane stress. The interaction integral used is exactly the same as that for extracting the stress intensity factors:
Iint
= lim
0!0 0
n 1 M 1 q d0 ;
with M as
M
= : "L I 0 1 aux
L
@u @x
aux
0 L 1 @u : aux @ x
i
;
=E
0 Iint(s) 0 "33(s) f
where E = E for plane stress and E = E=(1 0 2) for plane strain, axisymmetry, and three dimensions. "33 is zero for plane strain and plane stress. Iint(s) can be calculated by means of the same domain integral method used for J -integral calculation and the stress intensity factor extraction, which has been described in J -integral evaluation, Section 2.16.1, and Stress intensity factor extraction, Section 2.16.2.
Reference
Contour integral evaluation, Section 7.9.2 of the ABAQUS Analysis Users Manual
2.16.32
2.16.4
Various criteria have been proposed to predict the angle at which a pre-existing crack will propagate. Among these criteria are the maximum tangential stress criterion (Erdogan and Sih, 1963), the maximum principal stress criterion (Maiti and Smith, 1983), the maximum energy release rate criterion (Palaniswamy and Knauss, 1978, and Hussain, Pu, and Underwood, 1974), the minimum elastic energy density criterion (Sih, 1973), and the T-criterion (Theocaris, 1982). These criteria predict slightly different angles for the initial crack propagation, but they all have the implication that KII = 0 at the crack tip as the crack extends (Cotterell and Rice, 1980). In ABAQUS/Standard we provide three criteria for homogeneous, isotropic linear elastic materials: the maximum tangential stress criterion, the maximum energy release rate criterion, and the KII = 0 criterion. KIII is not taken into account in what follows, since a generally accepted 6 theory for crack propagation with KIII = 0 remains to be developed.
Maximum tangential stress criterion
Procedures
The near-crack-tip stress eld for a homogeneous, isotropic linear elastic material is given by
= r =
p1 p
2r 1 2r
cos cos
1 2 1 2
(KI cos2
1 2
where r and are polar coordinates centered at the crack tip in a plane orthogonal to the crack front. The direction of crack propagation can be obtained using either the condition @ =@ = 0 or r = 0; i.e., ! p
^ = cos01
2 3KII +
2 2 KI + 9KII
4 2 2 KI + 8KI KII
^ ^ where the crack propagation angle is measured with respect to the crack plane. = 0 represents ^ < 0 if K ^ > 0 if K the crack propagation in the straight-ahead direction. II > 0, while II < 0.
2.16.41
tan-1(Kll/Kl)
Figure 2.16.41 Contour for evaluation of the J -integral.
For the crack segment we also have the relation 1 k k2 k 2 G = (KI + KII ): E The maximum energy release rate criterion postulates that the parent crack initially propagates in the direction that maximizes Gk .
KII = 0 criterion
k This criterion simply postulates that a crack will initially propagate in the direction that makes KII = 0. It can be seen from Figure 2.16.41 that the maximum energy release rate criterion and the KII = 0 criterion predict nearly coincident crack propagation angles. By comparison, the maximum tangential stress criterion predicts smaller crack propagation angles.
Reference
Contour integral evaluation, Section 7.9.2 of the ABAQUS Analysis Users Manual
2.16.42
STRESS LINEARIZATION
2.17.1
STRESS LINEARIZATION
Stress linearization is the separation of stresses through a section into constant membrane, linear bending, and nonlinearly varying peak stresses. The capability for calculating linearized stresses is available in the Visualization module of ABAQUS/CAE; it is most commonly used for two-dimensional axisymmetric models. See Chapter 36, Calculating linearized stresses, of the ABAQUS/CAE Users Manual for detailed information on how to obtain linearized stresses; the computational methods used by ABAQUS are discussed here.
Procedures
Linearized stress components are computed using user-dened sections traversing a nite element model structure. Stress values are extracted at regular intervals along the dened section, and integration is performed numerically using the extracted stress values. Membrane, bending, and peak stress values are computed. These stresses are dened as follows:
Membrane stress
The constant portion of the normal stress such that a pure moment acts on a plane after the membrane stress is subtracted from the total stress.
Bending stress
The variable portion of the normal stress equal to the equivalent linear stress or, where no peak stresses exist, equal to the total stress minus the membrane stress.
Peak stress
The portion of the normal stress that exists after the membrane and bending stresses are subtracted from the total stress. For the best results, the endpoints of the section should be chosen so that the section is normal to the interior and exterior surfaces of the model. This orientation minimizes problems with shear stresses since they will be approximately zero at the ends of the line (Kroenke, 1973).
Three-dimensional structures
The membrane values of the stress components are computed using the following equation:
where
m
t=2 Z
0t=2
dx;
2.17.11
STRESS LINEARIZATION
is the thickness of the section, is the stress along the path, and x is the coordinate along the path.
t
The linear bending values of the stress components are computed using the following equations:
b B
=0
b A
= t2
6
t=2 Z
0t=2
xdx;
where b and b are the bending values of the stress at point A and point B (the endpoints of the A B section; see Figure 2.17.11).
B axis of symmetry E A = 20
2
= C
midplane
Figure 2.17.11
The integration is performed numerically. Assuming the path between point A and point B is divided uniformly into n intervals, the integrals are evaluated as follows:
m =
1 2n
X
n j =1
( j + j +1 )
2.17.12
STRESS LINEARIZATION
and
b = 0 A
nt j =1
X
n
( j xj + j +1 xj +1 );
The derivation of the above equations is similar for the axisymmetric case, except for the fact that the neutral axis is shifted radially outward. Separate expressions are obtained for the stresses in the thickness, meridional, and hoop directions. In ABAQUS/CAE these are represented as local directions 1, 2, and 3, respectively (see Figure 2.17.12).
Procedures
meridional direction 2
thickness direction 1
hoop direction 3
Figure 2.17.12
Meridional stress
Stress directions.
The meridional stresses are computed using the following relations. The meridional force per unit circumferential length is
F2 =
where
Z2
t=
0t=2
2
R dx; Rc
2 is the stress in the meridional direction, R is the radius of the point being integrated, and Rc = 1 (RA + RB ) is the mean circumferential radius. 2
2.17.13
STRESS LINEARIZATION
F2 t
t=2 Z
0t=2
2
R dx: Rc
X
n j =1
( 2j R j +
2(j+1) Rj +1);
where
2j is the meridional stress component at point j along the path and Rj is the radius at point j along the path.
To compute the meridional bending stresses, we rst need to compute the distance from the center surface to the neutral surface for meridional bending. That distance, x2, is equal to
x2 =
t2 1 cos ; 12 1 Rc
where is the angle between the thickness direction and the radial direction. The meridional bending moment per unit circumferential length is dened as
M2 =
t= Z2
0t=2
(x
R 0 x2 )2 R dx;
c
I2 =
t= Z2
0t=2
(x
1 R 0 x2)2 R dx = 12 t3 0 x2t2 :
Hence, the meridional bending stresses at the end points A and B are obtained with
b 2A =
M 2 (x A 0 x 2 ) ; I2
n
b 2B
M 2 (x B 0 x 2 ) : I2
M2 =
2.17.14
STRESS LINEARIZATION
Hoop stress
m The hoop membrane stress 3 is obtained with m 3
F3 t
t= Z2
0t=2
3
1+
x dx;
Procedures
where
3 is the hoop stress and is the radius of curvature of the mid-surface of the section in the meridional plane.
The hoop bending stresses at the end points are obtained using the relations
b 3A
=
M 3 (x A 0 x 3 ) ; I3 x3
=
b 3B
M 3 (x B 0 x 3 ) ; I3
where
t2 12
is the distance from the center surface to the neutral surface for hoop bending,
t= Z2
=
M3
0t=2
(x
0 x 3 ) 3
1+
1+
x dx
I3
t= Z2
=
0t=2
(x
0 x3 )
x dx =
1 3 t 12
0 x2 t 3
is the moment of inertia for circumferential bending. The numerical scheme to compute the circumferential membrane stress is
3
1 2n
n X j =1
M3
2.17.15
STRESS LINEARIZATION
Thickness stress
The thickness stress does not transfer any forces or moments. Typically, the stress arises due to applied external pressures and thermal expansion effects, and there is no obvious preferred method for determining membrane and bending stresses. Hence, we choose the thickness membrane stress as the average thickness stress:
m 1
=t
t= Z2
0t=2
1dx =
1 2n
X
n j=1
We choose the thickness bending stress such that the sum of the thickness membrane and bending stresses at the end points A and B is equal to the total stress at these points:
b m 1A = 1A 0 1 ; b 1B
=
m 1B 0 1 ;
and interpolate the bending stress linearly between these values. This seems a reasonable assumption, since the thickness surface stresses are determined by the applied pressure and should not have a strong peak stress contribution.
Shear stress
The membrane shear stress in the meridional plane is computed in the same way as the meridional membrane stress:
21 =
t= Z2
0t=2
21
R dx; Rc
where 21 is the shear stress along the path. The shear stress distribution is assumed to be parabolic and equal to zero at the ends. Hence, the bending shear stresses are set to 0.0. The numerical scheme used to compute the membrane shear stress is shown below:
21 =
1 2nRc
n X j =1
Stress linearization requires the results to be transformed from the global coordinate system to a local coordinate system dened by the stress line.
2.17.16
STRESS LINEARIZATION
Axisymmetric case
The computation of the local coordinate system is a trivial procedure when performed for axisymmetric stress linearization. The transformation matrix in this case will be
2 cos 4 0 sin
sin 0 3 cos 0 5 : 0 1
Procedures
Three-dimensional case
When performing three-dimensional stress linearization, the local x-axis will be dened by the stress line. The local y- and z -axes are computed by a series of cross-products. This procedure is shown below. The vector between points (X1 , Y1 , Z1 ) and (X2 , Y2 , Z2 ) is
2 1 X 3 2 X2 0 X 1 3 x = 4 1 Y 5 = 4 Y2 0 Y 1 5 :
1Z
z
Z2
0 Z1
Assuming the local y-axis lies in the plane of the local x-axis and the global Y -axis, the local z -axis is dened by 3 2
1Z = 4 0 5: 01X
y = z2x
3 2 1X 1Y = 4 01Z 1Z 0 1X 1X 5 :
1Y 1Z
After normalization, the above three vectors can be combined to create the transformation matrix.
Reference
2.17.17
2.18.1
ABAQUS/Design supports design sensitivity analysis (DSA) for static stress and frequency problems. DSA provides derivatives of certain response quantities with respect to specied input quantities. These derivatives are known as sensitivities. The responses available for DSA are a subset of the list of ABAQUS output variables and are known as design responses; the specied input quantities are known as design parameters. Quantities that are functions of design parameters are referred to as being design dependent. The DSA theory is presented from the perspective of computing the required derivatives analytically, rst for static stress analysis and then for frequency analysis. At the end of each section an alternative numerical approach based on this theory is discussed.
DSA for static stress analysis
Procedures
The equations pertaining to DSA can be derived based on a total displacement formulation or an incremental displacement formulation. The total displacement formulation is intended for historyindependent problems, where the current state of the problem depends only on the total displacements. The incremental formulation is intended for history-dependent problems, where the current state of the problem depends on the state at the beginning of the increment and the incremental displacements.
Total displacement DSA formulation for nonlinear equilibrium problems
Let R and P be the numbers of design responses and design parameters, respectively. Let each response r , r = 1; . . . ; R, be a function of design parameters hp , p = 1; . . . ; P and depend on them both explicitly and via the displacement eld represented here by the nodal displacement vector uN (see the denition of nite element interpolation in Procedures: overview and basic equations, Section 2.1.1), The dependence uN (hp ) is only implicit; i.e., it is implied only by the design dependence of coefcients in the equilibrium equation system whose solution is uN . Assume that we have solved an equilibrium problem dened by Equation 2.1.12 at the end of an increment and that we have the converged solution uN as well as values of all responses. Sensitivity of a response r with respect to design parameter hp is dened as
r
r uN (hp ); hp :
0 1
dr dhp
@r @hp
(2.18.11)
All but one quantity in the above equation can be determined explicitly given the equilibrium solution. The only unknown is duN =dhp; to compute it, an additional system of equations has to be solved. Rewrite Equation 2.1.12 in the form
F N (u M ) = 0 ;
2.18.11
(2.18.12)
where
V0 All the quantities in the above equation are assumed to depend on design parameters hp explicitly or via displacement eld uN . Differentiation of the above two equations with respect to design
parameters leads to the following equation:
FN
=0
N : c dV 0 +
NT 1 t dS + NT 1 f dV : S N V N
M
KMN
in which
duN dhp
= 0 @F p @h = @F N @u
M
(2.18.13)
is the tangent stiffness (Jacobian) matrix dened in Equation 2.1.14 and @F M =@hp is an explicitly determinable quantity. Substituting Equation 2.18.13 into Equation 2.18.11, we obtain
KMN
dr dhp
(2.18.14)
which is the solution of the total displacement DSA problem. The DSA algorithm used in ABAQUS is known as the direct differentiation method (DDM) and consists of the following operations. After the converged equilibrium solution is obtained, the three arrays @r =@hp , @r =@uM , and @F M =@hp have to be computed in an element-by-element manner. @F M =@hp is often called the pseudoload since it becomes the right-hand side of the DSA problem. The nal DSA solution is obtained by solving the system of Equation 2.18.13 for each p ; ;P with respect to the unknown vectors of nodal displacement sensitivity duN =dhp . The displacement sensitivities are then substituted into Equation 2.18.11 to compute dr =dhp . The coefcient matrix KMN used in the DSA computations is simply the last tangent stiffness matrix used in the equilibrium iterative algorithm. At the stage of the DSA computations this matrix is still available in the decomposed form and can be retrieved easily to perform the back substitutions for the DSA right-hand-side vectors. This makes the DSA module a very efcient add-on to the equilibrium analysis enabling sensitivity computations at a relatively low cost.
= 1 ...
The formulation of DSA presented above provides a brief introduction to the way DSA is implemented in ABAQUS; however, due to some simplications, the discussion is not relevant to a large number of nonlinear mechanical problems, especially those involving history-dependent behavior of the structure modeled. The main difculty in such problems is that many quantities necessary to compute the residual F N in Equation 2.18.12 or to dene design responses do not lend themselves to be expressed as functions of total displacement uN . Rather, at each time increment they are functions of certain state variables at the beginning of the increment (referred to as the time instant t) and of the incremental displacements, uN : 0 1
see, for example, Kleiber et al. (1997). The notation t stands for a set of state variables that may include tensors (stress, back stress, etc.) as well as scalar quantities (equivalent plastic strain, etc.) dened for a particular material point at time t. Some responses may also depend directly on the displacement uN , and the beginning-of-the-increment value of uN will, generally, also enter into the set t. In such a case Equation 2.18.14 takes the following form:
(2.18.15)
The fundamental difference, from the point of view of the DSA solution algorithm, between the total and incremental approach is that in the latter case all state variables effectively become additional, or internal, design responses, whose sensitivities must be computed and updated at the end of each time increment to proceed with the DSA in the next increment. The number of such internal responses may be signicant with obvious effects both on the computational time and memory requirement. The DSA solution procedure is similar to that in the total displacement approach. After the equilibrium computations are complete, the arrays of explicit design derivatives Dr =Dhp , DF M =Dhp (the pseudoload), and the derivatives with respect to displacements @r =@ uM are ; ; R, includes in this assembled in the element loop. The set of design responses r , r case all the scalars and tensor components of . In the direct differentiation method the following system of equations is solved for each design parameter hp :
@( dt = @h1) + @(1) dh t p p @ denotes the explicit design derivative of a quantity (1). D (1 ) Dhp
def
Procedures
= 1 ...
KMN
d1 uN dhp
= 0 DF p Dh
The derivatives required for DSA can be computed analytically or numerically. In the analytical approach the nite element equations are differentiated exactly, following the theory described in the previous sections. This approach is difcult to implement, but it is efcient and yields exact sensitivities. In the numerical approach some or all of the required derivatives are computed using the nite difference technique. The numerical approach can be further subdivided into the overall or global nite difference approach and the semi-analytic approach. In the global nite difference approach the response sensitivities with respect to a particular design parameter are obtained by perturbing that design parameter a number of times (depending on the nite difference technique) and performing an entire equilibrium analysis for each perturbation. The responses are retained for each analysis and then differenced to obtain the response sensitivities. This approach is computationally
2.18.13
expensive since an entire equilibrium problem must be solved for each perturbation, but it is easily implemented. The semi-analytic approach is used in ABAQUS and can be viewed as a compromise between the analytic and global nite difference approaches. In the semi-analytic approach the DSA element vectors are obtained by differencing; but, like the analytic approach, the DSA solution is obtained by back-substitution against KMN . The advantage of the semi-analytic approach is that it is much easier to implement than the analytic approach and much more efcient than the global nite difference approach. The details of this method are described in the following paragraphs. The objective of the semi-analytic approach is to compute the DSA vectors DF M =Dhp and dr =dhp numerically by nite differencing. For simplicity, assume that the nite difference technique is central difference such that for a given function A x , the derivative of A with respect to x is
where x is the perturbation of x. For generality, consider the history-dependent case. To approximate the explicit design derivatives of F M , the incremental displacement is held constant while a positive perturbation hp is applied to each design parameter hp . In this way perturbed values of F M are obtained as
The change in the state corresponding to a perturbation in the design parameters is approximated by
The above process is repeated for a negative perturbation (0hp ), after which the results are differenced to arrive at the explicit design derivative DF M =Dhp . Once the (incremental) displacement sensitivities are found, the response sensitivities dr =dhp can be obtained using
The process is repeated for a negative perturbation of hp , and the results are differenced. The nite difference interval must be chosen carefully. If the interval is too small, round-off or cancellation errors occur due to loss of precision during the differencing operations. On the other hand, if the interval is too large, truncation errors may occur. Truncation errors arise from the fact that differencing formulas are based on truncated Taylor series expansions. ABAQUS will automatically choose a perturbation size that provides the best compromise between cancellation and truncation errors.
N h : p p
2.18.14
This discussion will build upon the concepts and terminology described above, so it is recommended that the previous section be read rst. The discussion below is divided into two sections depending on the characteristics of the eigenvalue problem: distinct eigenvalues and repeated eigenvalues.
Distinct eigenvalue case
The theory presented below assumes that all the eigenvalues are distinct (i.e., no repeated eigenvalues). If this is not the case, further manipulations are required to obtain the eigenvalue and eigenvector sensitivities corresponding to the repeated eigenvalues, and the following equations for the sensitivities will be incorrect. The repeated eigenvalue case is considered in the next section. Performing a frequency analysis means solving the following eigenvalue problem (see Eigenvalue extraction, Section 2.5.1):
(
Procedures
K N M 0 M NM )N
=0
; M NM
where represents the eigenvalues, N represents the eigenvectors, and In addition, the eigenvectors are scaled such that either
Nmax = 1
or
(2.18.17)
N M NM M = 1
(2.18.18)
for each mode. The default is the rst scaling scheme. To obtain eigenvalue and eigenvector sensitivities, rst differentiate Equation 2.18.16 with respect to design parameter hp to obtain the following equation:
0
K NM 0 M NM
dM dhp
=
(2.18.19)
where represents a particular mode number. Pre-multiplying by N , making use of Equation 2.18.16, and manipulating the result gives the eigenvalue sensitivities:
d dhp
NM NM N DK p 0 DM p M Dh Dh : N M NM M
(2.18.110)
Except for the mass and stiffness derivatives, all quantities in this equation are known once the eigenvalue problem has been solved.
2.18.15
This section outlines the formulation used to obtain eigenvalue sensitivities for repeated eigenvalues. Further information can be found in the papers by Mills-Curran (1988) and Shaw (1991). When an eigenvalue repeats R times, the eigenvectors 8 = b1 . . . R c associated with are linearly independent but not uniqueany linear combination of these eigenvectors is also an eigenvector. Because of this non-uniqueness, the eigenvectors may not be continuous or differentiable in the design parameter. However, a set of R eigenvectors Z = bz1 . . . zR c that are continuous and differentiable can be obtained by an orthogonal transformation:
where AT A = I, and A is to be determined. Replacing the eigenvectors with the eigenvectors z in Equation 2.18.19 and premultiplying by 8T yields in matrix notation:
Z = 8A;
(2.18.111)
B0
where
d3 C A = 0; dhp DK Dhp
(2.18.112)
0 DM 8; Dh
p
(2.18.113)
and
Equation 2.18.112 is recognized as an R 2 R reduced eigenvalue problem whose R eigenvalues d3=dhp are the eigenvalue sensitivities associated with the repeated eigenvalue and whose eigenvectors are A. N sensitivities are obtained for the single eigenvalue . The physical interpertation of this is that a perturbation in the design parameter may cause the original repeated mode to branch into as many as N distinct modes (imagine a beam with a circular cross-section perturbed into an elliptical section; the repeated modes associated with the original symmetry of the section now split into distinct modes associated with the minor and major axes of the ellipse).
Computational approach
C = 8T M8:
A semi-analytic approach is used to compute the eigenvalue sensitivities. The basic idea of this approach, as outlined in the section on static DSA, is to compute some of the required intermediate derivatives using nite differencing. In the context of DSA for frequency procedures this means that the derivatives of the mass and stiffness matrices are computed using nite differencing.
Choosing the nite difference interval
ABAQUS uses a heuristic algorithm to automatically determine the perturbation sizes to be used in the differencing scheme. The objective of the algorithm is to select the perturbation sizes that lead to accurately computed derivatives. This is done on an element to element basis, so it is possible
2.18.16
for a given design parameter that the perturbation size for one element will be different from that of another element. The selection of the perturbation sizes is based on the behavior of a scalar metric sp for each design parameter hp . This metric must be both convenient and representative of the terms that are numerically differenced. It is chosen as follows: Static steps. For static steps the scalar metric is chosen as the norm of the element pseudoload:
Procedures
sp = N
NM DK NM 0 DM M : Dhp Dhp
The choice of N depends on whether a given mode has a distinct eigenvalue or is associated with a repeated eigenvalue. If mode has a distinct eigenvalue, N is taken as N . Consequently, sp becomes simply the numerator of Equation 2.18.110. Therefore, sp is a direct measure of the magnitude of the eigenvalue sensitivity and is also convenient since this term already must be calculated to obtain the eigenvalue sensitivity. Unlike the distinct eigenvalue case, the sensitivities of a repeated eigenvalue cannot be treated independently. The sensitivities of a repeated eigenvalue are extracted simultaneously from the same reduced eigenvalue system, and this system is obtained by numerical differencing (recall Equation 2.18.112 and Equation 2.18.113). Consequently a single perturbation size (for each design parameter) must be used to obtain all sensitivities of a repeated eigenvalue. To calculate the single perturbation size, a single scalar metric is obtained by taking N as the sum of the eigenvectors associated with the repeated eigenvalue. The calculation of sp is similar to the calculation of the matrix (the only term in the reduced eigenvalue system obtained by numerical differencing); therefore, this choice sp is both representative of the repeated eigenvalue case and involves differencing calculations that are already being done to obtain the reduced eigenvalue system.
The basic idea of the algorithm is compute the scalar metric sp for a range of perturbation sizes hI , p I = 1; 2; . . ., which vary by orders of magnitude. For each hI , a corresponding sI is computed. The p p relative change in sp between consecutive perturbation sizes, calculated as eI = jsI 0 sI 01 j=sI , is used p p p p to measure how close a perturbation size is to optimum. In this range of perturbation sizes the one that yields the smallest relative change, denoted as ea , is identied as the best perturbation size, ha . p p If ea is not less than a specied tolerance, the range of perturbation sizes is expanded (up to a certain p limit), and the testing continues. It is important to realize that eI is not used directly in assessing p the accuracy of the numerical differentiation but rather is intended as a means for determining the
2.18.17
optimum perturbation size. Thus, a tighter tolerance on ea causes the algorithm to expend more effort p in nding an optimum perturbation size but does not guarantee more accurate sensitivities.
Reference
Design sensitivity analysis, Section 7.14.1 of the ABAQUS Analysis Users Manual
2.18.18
Chapter 3 Elements
Overview Continuum elements Innite elements Membrane and truss elements Beam elements Shell elements Rebar Hydrostatic uid elements Special-purpose elements 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9
3.1.1
The ABAQUS element library provides a complete geometric modeling capability. For this reason any combination of elements can be used to make up the model. Sometimes multi-point constraints are required for application of the necessary kinematic relations to form the model (for example, to model part of a shell surface with solid elements and part with shell elements or to model a pipe elbow with a mixture of beam and shell elements). All elements use numerical integration to allow complete generality in material behavior. Shell and beam element properties can be dened as general section behaviors, or each cross-section of the element can be integrated numerically, so that nonlinear response can be tracked accurately when needed. A composite layered section can be specied, with different materials at different heights through the section. Some special elements (such as line springs) use an approximate analytical solution to model nonlinear behavior. All of the elements in ABAQUS are formulated in a global Cartesian coordinate system except the axisymmetric elements, which are formulated in terms of rz coordinates. In almost all elements, primary vector quantities (such as displacements and rotations ) are dened in terms of nodal values with scalar interpolation functions. For example, in elements with a two-dimensional topology the interpolation can be written as
where the interpolation functions N N (g; h) are written in terms of the parametric coordinates g and h. In most element types the same parametric interpolation is used for the coordinate vector :
Elements
Such isoparametric elements are guaranteed to be able to represent all rigid body modes and homogeneous deformation modes exactly, a necessary condition for convergence to the exact solution as the mesh is rened. All elements in ABAQUS are integrated numerically. Hence, the virtual work integral as described in Nonlinear solution methods in ABAQUS/Standard, Section 2.2.1, will be replaced by a summation:
:
D dV !
n Xi i=1
:
D i Vi ;
where n is the number of integration points in the element and Vi is the volume associated with integration point i. ABAQUS will use either full or reduced integration. For full integration the number of integration points is sufcient to integrate the virtual work expression exactly, at least for linear material behavior. All triangular and tetrahedral elements in ABAQUS use full integration. Reduced integration can be used for quadrilateral and hexahedral elements; in this procedure the number of integration points is sufcient to integrate exactly the contributions of the strain eld that are one order less than the order of interpolation. The (incomplete) higher-order contributions to the strain eld present in these elements will not be integrated.
3.1.11
The advantage of the reduced integration elements is that the strains and stresses are calculated at the locations that provide optimal accuracy, the so-called Barlow points (Barlow, 1976). A second advantage is that the reduced number of integration points decreases CPU time and storage requirements. The disadvantage is that the reduced integration procedure can admit deformation modes that cause no straining at the integration points. These zero-energy modes make the element rank-decient and cause a phenomenon called hourglassing, where the zero energy mode starts propagating through the mesh, leading to inaccurate solutions. This problem is particularly severe in rst-order quadrilaterals and hexahedra. To prevent these excessive deformations, an additional articial stiffness is added to the element. In this so-called hourglass control procedure, a small articial stiffness is associated with the zero-energy deformation modes. This procedure is used in many of the solid and shell elements in ABAQUS. Most fully integrated solid elements are unsuitable for the analysis of (approximately) incompressible material behavior. The reason for this is that the material behavior forces the material to deform (approximately) without volume changes. Fully integrated solid element meshes, and in particular lowerorder element meshes, do not allow such deformations (other than purely homogeneous deformation). For that reason ABAQUS uses selectively reduced integration in these elements: reduced integration is used for the volume strain and full integration for the deviatoric strains. As a consequence the lower-order elements give an acceptable performance for approximately incompressible behavior. For fully incompressible material behavior, another complication occurs: the bulk modulus and, hence, the stiffness matrix becomes innitely large. For this case a mixed (hybrid) formulation is required, where the displacement eld is augmented with a hydrostatic pressure eld. In this formulation only the inverse of the bulk modulus appears, and, consequently, the contribution to the operator matrix vanishes. The hydrostatic pressure eld plays the role of a Lagrange multiplier eld enforcing the incompressibility constraints. ABAQUS/Standard also provides elements for multield problems. Examples are the pore pressure elements used for the analysis of porous solids with uid diffusion, coupled temperature-displacement elements that couple heat transfer with stress analysis, and piezoelectric elements that couple electrical conduction with stress analysis. In these multield elements the scalar variable (such as the temperature) is usually interpolated with different scalar functions as the displacement eld; i.e.,
T (g; h)
N (g; h) T N ;
where M N (g; h) may differ from N N (g; h). The coupling of the elds will generally occur at the integration points; for example, in coupled temperature-displacement elements the coupling is due to temperature-dependent mechanical properties and heat generation is due to inelastic work. Finally, ABAQUS offers a complete set of diffusion elements to analyze conductive and convective heat transfer. In these elements only temperatures appear as nodal degrees of freedom. The temperatures are interpolated with essentially the same interpolation function, M N (g; h), as used in the coupled temperature-displacement elements.
Reference
Element library: overview, Section 13.1.1 of the ABAQUS Analysis Users Manual
3.1.12
SOLID ELEMENTS
3.2.1
ABAQUS contains a library of solid elements for two-dimensional and three-dimensional applications. The two-dimensional elements allow modeling of plane and axisymmetric problems and include extensions to generalized plane strain (when the model exists between two planes that may move with respect to each other, providing thickness direction strain that may vary with position in the plane of the model but is constant with respect to thickness position). The material description of three-dimensional solid elements may include several layers of different materials, in different orientations, for the analysis of laminated composite solids. A set of nonlinear elements for asymmetric loading of axisymmetric models is also available, and linear innite elements in two and three dimensions can be used to model unbounded domains. The solid element library includes isoparametric elements: quadrilaterals in two dimensions and bricks (hexahedra) in three dimensions. These isoparametric elements are generally preferred for most cases because they are usually the more cost-effective of the elements that are provided in ABAQUS. They are offered with rst- and second-order interpolation and are described in detail in Solid isoparametric quadrilaterals and hexahedra, Section 3.2.4. For practical reasons it is sometimes not possible to use isoparametric elements throughout a model; for example, some commercial mesh generators use automatic meshing techniques that rely on triangulation to ll arbitrarily shaped regions. Because of these needs ABAQUS includes triangular, tetrahedron, and wedge elements. For most cases it is recommended that these elements be only used to ll in awkward parts of the mesh and, in particular, that well-shaped isoparametric elements be used in any critical region (such as an area where the strain must be predicted accurately). The isoparametric elements can also be degenerated to make simpler shapes. Generally the elements written for those particular geometries are preferred to this method. The exception to this rule occurs in cases where singularities are to be modeled (such as in fracture mechanics applications), since the degenerate second-order isoparametric elements can provide a 1= r singularity through the use of the quarter point technique (placing the midside nodes 1/4 of the distance along the side from the node at the singularity instead of at the middle point of the side). Solid elements are provided with rst-order (linear) and second-order (quadratic) interpolation, and the user must decide which approach is more appropriate for the application. Some guidelines are as follows. Standard rst-order elements are essentially constant strain elements: the isoparametric forms can provide more than constant strain response, but the higher-order content of the solutions they give is generally not accurate and, thus, of little value. The incompatible mode elements, described in Continuum elements with incompatible modes, Section 3.2.5, are from the users perspective lower-order elements but have internal degrees of freedom that enable the element to represent almost all linear strain patterns. These elements can represent certain important linear strain elds exactly: the most important eld is the one due to bending. The second-order elements are capable of representing all possible linear strain elds. Thus, in the case of elliptic problemsproblems for which the governing partial differential equations are elliptic in character, such as elasticity, heat conduction, acoustics, in which smoothness of the solution is assuredmuch higher solution accuracy per degree of freedom is usually available with the higher-order elements. Therefore, it is generally recommended that the highest-order elements available be used for such cases: in ABAQUS this means second-order elements. This observation logically leads to the use of the
Elements
3.2.11
SOLID ELEMENTS
hierarchical nite element technique or p-methodrening the model by increasing the interpolation order in the elements in critical regions: this approach is as yet not available in ABAQUS. A case where both incompatible mode elements and second-order elements can be used effectively is the stress analysis of relatively thin members subjected to bending: such problems are often encountered in practical applications. In such cases the strain variation through the thickness must be at least linear, and constant strain (rst-order) elements do a poor job of representing this variation. Fully integrated rst-order isoparametric elements also suffer from shear locking in these geometries: they cannot provide the pure bending solution because they must shear at the numerical integration points to respond with an appropriate kinematic behavior corresponding to the bending. This shearing then locks the elementthe response is far too stiff. For the isoparametric elements reduced integration provides a cure for these problems, but at the cost of allowing spurious singular modes (hourglassing). The use of second-order elements is a more reliable alternative, because the second-order interpolation naturally contains the linear strain eldone element through the thickness is enough to represent the behavior of a thin component subjected to bending loads quite accurately. Another alternative is formed by the incompatible mode elements: the linear strain eld in these elements contains the modes required to solve the bending problem exactly if the elements are rectangular in shape. For a detailed discussion of the performance of ABAQUS continuum elements in bending problems, see Performance of continuum and shell elements for linear analysis of bending problems, Section 2.3.5 of the ABAQUS Benchmarks Manual. (It should be remembered, however, that ABAQUS offers shell and beam elements that are specically written for thin geometries: the use of solid elements for such cases should only be considered when beam or shell elements are not practical.) For all of these reasons the second-order elements are preferred in elliptic applications. The argument is readily extended to higher-order interpolation (cubic, quartic, etc), but the rapid increase in cost per element for higher-order forms means thateven though the accuracy per degree of freedom is higher the accuracy per computational cost may not be increasing. Practical experience suggests thatexcept in special caseslittle is gained by going beyond the second-order elements, so ABAQUS does not offer any higher-order forms. Many problems of practical interest are not elliptic: localizations arise in one form or another. Plasticity applications are an exampleas the solution approaches the limit load, most plasticity models tend toward hyperbolic behavior. This allows discontinuities to occur in the solutionthe slip line solutions of classical perfect plasticity theory are plots of the characteristic lines of velocity discontinuities in the hyperbolic equations of the problem. If the nite element solution is to exhibit accuracy, these discontinuities in the gradient eld of the solution should be reasonably well modeled. With a xed mesh that does not use special elements that admit discontinuities in their formulation, this suggests that the lowest-order elementsthe rst-order elementsare likely to be the most successful, because, for a given number of nodes, they provide the most locations at which some component of the gradient of the solution can be discontinuous (the element edges). This argument is hardly rigorous, but it is, nevertheless, true that rst-order elements tend to be preferred for such cases. The incompatible mode elements can represent discontinuities particularly well. They are also able to represent strain localization such as occurs in shear bands. One should realize, however, that better dened shear localization increases the strain magnitude and, hence, tends to increase the number of increments and iterations required for the analysis. All of the solid elements in ABAQUS, except the innite elements, are written to include nite-strain effects. When these elements are used with a hyperelastic (elastomeric) material denition, the constitutive behavior is calculated directly from the deformation gradient matrix, F. When the elements are used for
3.2.12
SOLID ELEMENTS
geometrically nonlinear analysis with any other material denition (at nite strain this means the material has some inelastic behavior, since all of the elasticity denitions in ABAQUS except the hyperelasticity models assume that the elastic strains are small), the strains are calculated as the integral of the rate of deformation,
v D = sym x
@ @
with the effects of material rotation with respect to the coordinate system taken into consideration. In all cases the solid elements report stress as the true (Cauchy) stress. Unless a local orientation is specied for an element, stress and strain components are given as physical components referred to the global spatial directions. When a local orientation is dened for a solid element, the stress and strain components are given in the user-dened local system: this system rotates with the average material rotation calculated at each material point.
Reference
General-purpose continuum elements, Section 14.1 of the ABAQUS Analysis Users Manual
Elements
3.2.13
3.2.2
All the solid elements in ABAQUS allow for nite strain and rotation in large-displacement analysis. For kinematically linear analysis the strain is dened as
"
= sym
@ @
where is the total displacement and is the spatial position of the point under consideration in the original conguration. As discussed in Chapter 1, Introduction and Basic Equations, this measure of strain is useful only if the strains and rotations are small (all components of the strain and rotation matrices are negligible compared to unity). For cases where the strains and/or rotations are no longer small, two ways of measuring strain are used in the solid elements in ABAQUS. When the hyperelastic or hyperfoam material denition is used with an element, ABAQUS internally uses the stretch values calculated directly from the deformation gradient matrix, , to compute the material behavior. With any other material behavior it is assumed that any elastic strains are small compared to unity, so the appropriate reference conguration for the elasticity is only innitesimally different from the current conguration, and the appropriate stress measure is, therefore, the Cauchy (true) stress. (More precisely, the appropriate stress measure should be the Kirchhoff stress dened with respect to the elastic reference conguration but the assumption that this reference conguration and the current conguration are only innitesimally different makes the Kirchhoff and Cauchy stress measures almost the same: the differences are of the order of the elastic strains compared to unity). The conjugate strain rate to Cauchy stress is the rate of deformation:
u; X
Elements
where is the rate of deformation, is the velocity at a point, and are the current spatial coordinates of the point. The strain is, therefore, dened as the integral of the rate of deformation. This integration is nontrivial, particularly in the general case where the principal axes of strain rotate during the deformation. In ABAQUS the total strain is constructed by integrating the strain rate approximately over the increment by the central difference algorithm; and, when the strain components are referred to a xed coordinate basis, the strain at the start of the increment must also be rotated to account for the rigid body rotation that occurs in the increment. This is also done approximately, using the Hughes-Winget (1980) method. This integration algorithm denes the integration of a tensor associated with the material behavior as
@v D = sym @x
where is the tensor; is the increment in the tensor associated with the materials constitutive behavior, , dened by the central difference formula as and, therefore, dependent on the strain increment,
1 a
a +1 = 1R a 1RT + 1(1D); a
t t
1
t 1
1D
1D = sym @x t+1t=2
@
1u ;
3.2.21
where xt+1t=2 as
1! is the central difference integration of the rate of spin: @ 1u 1! = asym @x : t+1t=2 A somewhat different algorithm to calculate 1R is used for the Green-Naghdi rate in ABAQUS/Explicit.
where For example, the stress is integrated by this method as
t+1t
= (1=2)(xt + xt+1t ); and 1R is the increment in rotation, dened by Hughes and Winget 01 1R = I 0 1 1! 1 I + 1 1! ; 2 2
= 1R 1 t 1 1RT + 1(1D); +1
where is the stress increment caused by the straining of the material during this time increment t refer to the beginning and the end and is the Kirchhoff ( Cauchy) stress. The subscripts t and t of the increment, respectively. As shown in Procedures: overview and basic equations, Section 2.1.1, the contribution of the internal work terms to the Jacobian of the Newton method that is often used in ABAQUS/Standard is
1(1D)
Vo
(d : D + : dD) dV ;
(3.2.21)
where d and are evaluated at the end of the increment. Using the integration denition above, it can be shown that d t+1t
where
However, rather than computing the tangent matrix for the Newton method on this basis, we approximate this by using
= d
1 t+1t + t+1t 1 d
T + C : dD;
!!
dV :
Z
V
T 1 @v D : C : dD 0 : 2D 1 D 0 1 @v 2 @x @x
This Jacobian is the tangent stiffness of the rate form of the problem. Experience with practical cases suggests that this approximation provides an acceptable rate of convergence in the Newton iterations in most applications with real materials.
3.2.22
The strain and rotation measures described above are approximations. Probably the most limiting . While this measure does give aspect of these approximations is the denition of the rotation increment a representation of the rotation of the material at a point in some average sense (both in ABAQUS/Standard and ABAQUS/Explicit), it is clear that each of the individual material bers at a point has a different rotation (unless the material point undergoes rigid body motion only or, as an approximate extension, if the strains at the point are small). This suggests that the formulation described above will not be suitable for applications where the strains and rotations are large and where the material exhibits some form of anisotropic behavior. A common example of such cases is the induction of anisotropy through straining, as in kinematic hardening plasticity models. The integration methods described above are not suitable for such material models at large strains (for practical purposes with typical material parameters this means that the solutions will be quite wrong when the strains are greater than 20%30%). Therefore, the use of the kinematic hardening model in ABAQUS at such strain levels is not recommended. There is extensive literature on this subject; see Agah-Tehrani et al. (1986), for example.
1R
Reference
Solid (continuum) elements, Section 14.1.1 of the ABAQUS Analysis Users Manual
Elements
3.2.23
HYBRID FORMULATION
3.2.3
Many problems involve the prediction of the response of almost incompressible materials. This is especially true at large strains, since most solid materials show relatively incompressible behavior under large deformations. In this section we describe the augmented virtual work basis provided in ABAQUS/Standard for such cases. The method is described in the context of incompressible elasticity theory, since that is where it is most likely to be used. When the material response is incompressible, the solution to a problem cannot be obtained in terms of the displacement history only, since a purely hydrostatic pressure can be added without changing the displacements. The nearly incompressible case (that is, when the bulk modulus is much larger than the shear modulus or Poissons ratio, , is greater than 0.4999999) exhibits behavior approaching this limit, in that a very small change in displacement produces extremely large changes in pressure, so that a purely displacement-based solution is too sensitive to be useful numerically (for example, round-off on the computer may cause the method to fail). We remove this singular behavior in the system by treating the pressure stress as an independently interpolated basic solution variable, coupled to the displacement solution through the constitutive theory and the compatibility condition, with this coupling implemented by a Lagrange multiplier. This independent interpolation of pressure stress is the basis of these hybrid elements. More precisely, they are mixed formulation elements, using a mixture of displacement and stress variables with an augmented variational principle to approximate the equilibrium equations and compatibility conditions. The hybrid elements also remedy the problem of volume strain locking, which can occur at much lower values of (i.e., = 0.49). Volume strain locking occurs if the nite element mesh cannot properly represent incompressible deformations. Volume strain locking can be avoided in regular displacement elements by fully or selectively reduced integration, as described in Solid isoparametric quadrilaterals and hexahedra, Section 3.2.4. We begin by writing the internal virtual work:
W " where " is the virtual strain: " "
Elements
Z
V
" " dV ;
(3.2.31)
= sym
@ u
@
where is the virtual displacement eld; is the true (Cauchy) stress; V is the current volume; and W is the virtual work as dened by this equation. See Equilibrium and virtual work, Section 1.5.1, for a detailed discussion of the virtual work concept. In a displacement-based formulation the Cauchy stress, , is obtained with the constitutive equations from the deformation, usually in rate form:
" = C : d" + d
0 1 d
; (3.2.32) where C is the material stiffness matrix and d
is the rate of rotation (spin) of the material. ^ We modify the Cauchy stress by introducing an independent hydrostatic pressure eld p as follows: d
1
3.2.31
HYBRID FORMULATION
= + (1 0 )I(p 0 p); ^
p
(3.2.33)
where
1 = 0 3 trace( ) ^
is the hydrostatic pressure stress and is a small number. If was set equal to zero, the hydrostatic component in would be identical to the independent pressure eld p, corresponding to a pure mixed formulation. The small nonzero value ( 09) is chosen to avoid equation solver difculties. This relation is used in incremental form:
10
= 0 + 1 + (1 0 )I(1p 0 1^); p
Z
(3.2.34)
where 0 is the modied Cauchy stress at the start of the increment. We use the modied Cauchy stress in the virtual work expression and augment the expression with the Lagrange multiplier enforced constraint p0 p :
1^ = 0
W
(3.2.35)
with J the volume change ratio (Jacobian) and a Lagrange multiplier whose interpolation must still be determined. p will be interpolated over each element so that the constraint is satised in an integrated (average) sense. Since p is the value of the equivalent pressure stress increment computed from the kinematic solution, Equation 3.2.34 does not make sense if the material is fully incompressible because then p cannot be computed. For the purpose of development we regard the bulk modulus as nite, and we will be able to show that the nal formulation approaches a usable limit as we allow the bulk modulus to approach innity. For the formulation of the tangent stiffness (the Jacobian), we need to dene the rate of change of W . Therefore, we rewrite the virtual work equation in terms of the reference volume V 0 :
1^
W
V0
(3.2.36)
=
Z
V0
" " " [dJ : " + J d : " + J : d" + (dp 0 dp) + d(1p 0 1^)] dV 0: ^ p
(3.2.37)
dW
(3.2.38)
" where we used the identity J 01dJ I d". The rate of the modied stress follows from Equation 3.2.34 and the constitutive equations:
= :
3.2.32
HYBRID FORMULATION
d
(3.2.39)
where
part. Substituting these expressions into the expression for the rate of virtual work yields
dW
Z
V
" " " " " : C : d" + " : I : d" 0 1 (1 0 )" : I I : C : d"0 3 " " (1 0 )" : I dp 0 1 J 01 I : C : d" 0 J 01 dp + ^ 3 ^ o " " p J 01d(1p 0 1^) + : d" + " : (d
1 0 1 d
) dV: " "
(3.2.310)
It remains to choose . To get a symmetric expression for the rate of virtual work, we choose
= (1 0 )J
(3.2.311)
Elements
where
K
(3.2.312)
is the (instantaneous) bulk modulus. This is a suitable choice for , because the (independent) term proportional to p ensures that the modied incremental pressure eld, p, is properly constrained to the and K change slowly with strain incremental pressure, p. If we assume that the volumetric moduli and ignore changes in volume, we can write for the second variation d:
d
(3.2.313)
Z
V
(3.2.314)
where
~ = + (1 0 )(1p 0 1^) 31 I : C 0 I : p K
(3.2.315)
3.2.33
HYBRID FORMULATION
dW
Z
" " " " " : C : d" 0 91 (1 0 )" : CT : I I : C : d" + " : I : d"0 K V 1 (1 0 ) 0p I : C : d" + " : CT : I dp1 0 1 (1 0 ) p dp+ " " ^ ^ ^ ^ K 3K ~ " " " : d" + : (" d
0 d
1 ") dV: " "
1
(3.2.316)
~ : (d" + " 1 d
0 d
1 ") dV; " " "
@
~:
so that the nal expression for the rate of virtual work becomes
" " " " " : C : d" 0 91 (1 0 ) " : CT : I I : C : d" + " : I : d"0 K V 1 (1 0 ) 0p I : C : d" + " : CT : I dp1 0 1 (1 0 ) p dp+ " " ^ ^ ^ K (3.2.317) 3K T^ u u " " ~ @x 1 @dx 0 2" 1 d" dV: : @ @ " " The asymmetric term " : I : d" is only signicant if large volume changes occur. Hence, the term is ignored except for material models with volumetric plasticity, such as the (capped) Drucker-Prager model and the Cam-clay model. For these models the constitutive matrix C is usually asymmetric anyway dW
Z
" "
so that the addition of this nonsymmetric term does not affect the cost of the analysis. It was assumed in the expression for d that the (volumetric) moduli change only slowly with strain. This is not the case for material models with volumetric plasticity, in which these moduli can change abruptly. This may lead to slow convergence or even convergence failures. Failures usually occur only in higher-order elements, since in lower-order elements p 0 p approaches zero at every point and the error in d has no impact.
1 1^
Reference
Solid (continuum) elements, Section 14.1.1 of the ABAQUS Analysis Users Manual
3.2.34
3.2.4
The library of solid elements in ABAQUS contains rst- and second-order isoparametric elements. The rst-order elements are the 4-node quadrilateral for plane and axisymmetric analysis and the 8-node brick for three-dimensional cases. The library of second-order isoparametric elements includes serendipity elements: the 8-node quadrilateral and the 20-node brick, and a full Lagrange element, the 27-node (variable number of nodes) brick. The term serendipity refers to the interpolation, which is based on corner and midside nodes only. In contrast, the full Lagrange interpolation uses product forms of the one-dimensional Lagrange polynomials to provide the two- or three-dimensional interpolation functions. All these isoparametric elements are available with full or reduced integration. Gauss integration is almost always used with second-order isoparametric elements because it is efcient and the Gauss points corresponding to reduced integration are the Barlow points (Barlow, 1976) at which the strains are most accurately predicted if the elements are well-shaped. The three-dimensional brick elements can also be used for the analysis of laminated composite solids. Several layers of different material, in different orientations, can be specied in each solid element. The material layers or lamina can be stacked in any of the three isoparametric coordinates, parallel to opposite faces of the master element (Figure 3.2.41). These elements use the same interpolation functions as the homogeneous elements, but the integration takes the variation of material properties in the stacking direction into account. Hybrid pressure-displacement versions of these elements are provided for use with incompressible and nearly incompressible constitutive models (see Hybrid incompressible solid element formulation, Section 3.2.3, and Hyperelastic material behavior, Section 4.6.1, for a detailed discussion of the formulations used).
Interpolation
Elements
Isoparametric interpolation is dened in terms of the isoparametric element coordinates g, h, r shown in Figure 3.2.41. These are material coordinates, since ABAQUS is a Lagrangian code. They each span the range 01 to +1 in an element. The node numbering convention used in ABAQUS for isoparametric elements is also shown in Figure 3.2.41. Corner nodes are numbered rst, followed by the midside nodes for second-order elements. The interpolation functions are as follows. First-order quadrilateral:
u=
1 4 (1
3.2.41
Second-order quadrilateral:
u=0
1 4 1 1
(1
0 g)(1 0 h)(1 + g + h)u1 0 1 (1 + g)(1 0 h)(1 0 g + h)u2 4 0 g)(1 + g)(1 0 h)u5 + 1 (1 0 h)(1 + h)(1 + g)u6 2 0 g)(1 + g)(1 + h)u7 + 1 (1 0 h)(1 + h)(1 0 g)u8 2
(1
First-order brick:
u=
+
1 8 1 8 1 1
0 r)u3 + 1 (1 0 g)(1 + h)(1 0 r)u4 8 1 + (1 0 g )(1 0 h)(1 + r )u5 + (1 + g )(1 0 h)(1 + r )u6 8 8
(1 + g )(1 + h)(1 + 8 (1 + g )(1 + h)(1 + r )u7 + 1 8 (1
20-node brick:
u=0
1 8
(1
0 1 (1 + g)(1 + h)(1 0 r)(2 0 g 0 h + r)u3 0 1 (1 0 g)(1 + h)(1 0 r)(2 + g 0 h + r)u4 8 8 0 1 (1 + g)(1 + h)(1 + r)(2 0 g 0 h 0 r)u7 0 1 (1 0 g)(1 + h)(1 + r)(2 + g 0 h 0 r)u8 8 8 1 1 amp; + (1 0 g)(1 + g)(1 0 h)(1 0 r)u9 + (1 0 h)(1 + h)(1 + g)(1 0 r)u10 4 4
amp; + amp; + amp; +
1 4 1 4 1 4 1 1 (1 (1 (1
0 g)(1 + g)(1 + h)(1 0 r)u11 + 1 (1 0 h)(1 + h)(1 0 g)(1 0 r)u12 4 0 g)(1 + g)(1 0 h)(1 + r)u13 + 1 (1 0 h)(1 + h)(1 + g)(1 + r)u14 4
0 g)(1 + g)(1 + h)(1 + r)u15 + 1 (1 0 h)(1 + h)(1 0 g)(1 + r)u16 4 1 amp; + (1 0 r)(1 + r)(1 0 g)(1 0 h)u17 + (1 0 r)(1 + r)(1 + g)(1 0 h)u18 4 4
amp; +
4 (1
3.2.42
7 4 h 3
1 5
Elements
8 16 r 5 13 6 20
15
7 14
19
g 17 4 11 12 10 1 9 2 18 3
3.2.43
All the isoparametric solid elements are integrated numerically. Two schemes are offered: full integration and reduced integration. For the second-order elements Gauss integration is always used because it is efcient and it is especially suited to the polynomial product interpolations used in these elements. For the rst-order elements the single-point reduced-integration scheme is based on the uniform strain formulation: the strains are not obtained at the rst-order Gauss point but are obtained as the (analytically calculated) average strain over the element volume. The uniform strain method, rst published by Flanagan and Belytschko (1981), ensures that the rst-order reduced-integration elements pass the patch test and attain the accuracy when elements are skewed. Alternatively, the centroidal strain formulation, which uses 1-point Gauss integration to obtain the strains at the element center, is also available for the 8-node brick elements in ABAQUS/Explicit for improved computational efciency. The differences between the uniform strain formulation and the centroidal strain formulation can be shown as follows: For the 8-node brick elements the interpolation function given above can be rewritten as
NI
can be written as
3I = [01; 01; +1; +1; 01; 01; +1; +1]; 2 3I = [01; 01; 01; 01; +1; +1; +1; +1]; 3 0I = [+1; +1; 01; 01; 01; 01; +1; +1]; 1 0I = [+1; 01; +1; 01; +1; 01; +1; 01]; 3 0I = [01; +1; 01; +1; +1; 01; +1; 01]; 4 0I = [+1; 01; 01; +1; 01; +1; +1; 01]; 2
and the superscript I denotes the node of the element. The last four vectors, 0I ( has a range of four), are the hourglass base vectors, which are the deformation modes associated with no energy in the 1-point integration element but resulting in a nonconstant strain eld in the element. In the uniform strain formulation the gradient matrix I is dened by integrating over the element Z as
BiI =
Ni
(g; h; r) =
@N
@xi
where Ve` is the element volume and i has a range of three. In the centroidal strain formulation the gradient matrix
Bi
BI is simply given as
Ni
(0; 0; 0);
0 3 i = 0 2 i = 0 4 i = 0
1
=
Bi ; Bi ; Bi ; Bi :
7 5 8 6
It can be seen from the above that the centroidal strain formulation reduces the amount of effort required to compute the gradient matrix. This cost savings also extends to strain and element nodal force calculations because of the antisymmetric property of the gradient matrix. However, the centroidal strain formulation is less accurate when the elements are skewed. For two-dimensional plane elements and hexahedron elements in a parallelepiped conguration the uniform strain approach is identical to the centroidal strain approach. Full integration means that the Gauss scheme chosen will integrate the stiffness matrix of an element with uniform material behavior exactly if the Jacobian of the mapping from the isoparametric coordinates to the physical coordinates is constant throughout the element; this means that opposing element sides or faces in three-dimensional elements must be parallel and, in the case of the secondorder elements, that the midside nodes must be at the middle of the element sides. If the element does not satisfy these conditions, full integration is not exact because some of the terms in the stiffness are of higher order than those that are integrated exactly by the Gauss scheme chosen. Such inaccuracy in the integration does not appear to be detrimental to the elements performance. As will be discussed below, full integration in ABAQUS in rst-order elements includes a further approximation and is more accurately called selectively reduced integration. Reduced integration usually means that an integration scheme one order less than the full scheme is used to integrate the elements internal forces and stiffness. Supercially this appears to be a poor approximation, but it has proved to offer signicant advantages. For second-order elements in which the isoparametric coordinate lines remain orthogonal in the physical space, the reduced-integration points have the Barlow point property (Barlow, 1976): the strains are calculated from the interpolation functions with higher accuracy at these points than anywhere else in the element. For rst-order elements the uniform strain method yields the exact average strain over the element volume. Not only is this important with respect to the values available for output, it is also signicant when the constitutive model is nonlinear, since the strains passed into the constitutive routines are a better representation of the actual strains. Reduced integration decreases the number of constraints introduced by an element when there are internal constraints in the continuum theory being modeled, such as incompressibility, or the
Elements
3.2.45
Kirchhoff transverse shear constraints if solid elements are used to analyze bending problems. In such applications fully integrated elements will lockthey will exhibit response that is orders of magnitude too stiff, so the results they provide are quite unusable. The reduced-integration version of the same element will often work well in such cases. Finally, reduced integration lowers the cost of forming an element; for example, a fully integrated, second-order, 20-node three-dimensional element requires integration at 27 points, while the reducedintegration version of the same element only uses 8 points and, therefore, costs less than 30% of the fully integrated version. This cost savings is especially signicant in cases where the element formation costs dominate the overall costs, such as problems with a relatively small wavefront and problems in which the constitutive models require lengthy calculations. The deciency of reduced integration is that, except in one dimension and in axisymmetric geometries modeled with higher than rst-order elements, the element stiffness matrix will be rank decient. This most commonly exhibits itself in the appearance of singular modes (hourglass modes) in the response. These are nonphysical response modes that can grow in an unbounded way unless they are controlled. The reduced-integration second-order serendipity interpolation elements in two dimensionsthe 8-node quadrilateralshave one such mode, but it is benign because it cannot propagate in a mesh with more than one element. The second-order three-dimensional elements with reduced integration have modes that can propagate in a single stack of elements. Because these modes rarely cause trouble in the second-order elements, no special techniques are used in ABAQUS to control them. In contrast, when reduced integration is used in the rst-order elements (the 4-node quadrilateral and the 8-node brick), hourglassing can often make the elements unusable unless it is controlled. In ABAQUS the articial stiffness method and the articial damping method given in Flanagan and Belytschko (1981) are used to control the hourglass modes in these elements. The articial damping method is available only for the solid and membrane elements in ABAQUS/Explicit. To control the I hourglass modes, the hourglass shape vectors, , are dened:
= 0I 0 V1
which are different from the hourglass base vectors, 0I . It is essential to use the hourglass shape vectors rather than the hourglass base vectors to calculate the hourglass-resisting forces to ensure that these forces are orthogonal to the linear displacement eld and the rigid body eld (see Flanagan and Belytschko (1981) for details). However, using the hourglass base vectors to calculate the hourglassresisting forces may provide computational speed advantages. Therefore, for the 8-node brick elements ABAQUS/Explicit provides the option to use the hourglass base vectors in calculating the hourglassresisting forces. For hexahedron elements in a parallelepiped conguration the hourglass shape vectors are identical to the hourglass base vectors. The hourglass control methods of Flanagan and Belytschko (1981) are generally successful for linear and mildly nonlinear problems but may break down in strongly nonlinear problems and, therefore, may not yield reasonable results. Success in controlling hourglassing also depends on the loads applied to the structure. For example, a point load is much more likely to trigger hourglassing than a distributed load. Hourglassing can be particularly troublesome in eigenvalue extraction problems: the low stiffness of the hourglass modes may create many unrealistic modes with low eigenfrequencies.
e`
I Bi xJ 0J ; i
3.2.46
A renement of the Flanagan and Belytschko (1981) hourglass control method that replaces the articial stiffness coefcients with those derived from a three-eld variational principle is available in ABAQUS/Explicit. The approach is based on the enhanced assumed strain and physical hourglass control methods proposed in Engelmann and Whirley (1990), Belytschko and Bindeman (1992), and Puso (2000). It can provide increased resistance to hourglassing for nonlinear problems and coarse mesh displacement solution accuracy for linear elastic problems at a small additional computational cost. Experience suggests that the reduced-integration, second-order isoparametric elements are the most cost-effective elements in ABAQUS for problems in which the solution can be expected to be smooth. Note that in the case of incompressible material behavior, such as hyperelasticity at nite strain, the mixed formulation elements with reduced integration should be used (see Hybrid incompressible solid element formulation, Section 3.2.3, and Hyperelastic material behavior, Section 4.6.1). When large strain gradients or strain discontinuities are expected in the solution such as in plasticity analysis at large strains, limit load analysis, or analysis of severely loaded rubber componentsthe rst-order elements are usually recommended. Reduced integration can be used with such elements, but because the hourglass controls are not always effective in severely nonlinear problems, caution should be exercised. Fully integrated rst-order elements should not be used in cases where shear locking can occur, such as when the elements must exhibit bending behavior. The incompatible mode elements (Continuum elements with incompatible modes, Section 3.2.5) should be used for such applications.
Elements
For fully integrated rst-order isoparametric elements (4-node elements in two dimensions and 8-node elements in three dimensions) the actual volume changes at the Gauss points are replaced by the average volume change of the element. This is also known as the selectively reduced-integration technique, because the order of integration is reduced in selected terms, or as the technique, since the strain-displacement relation ( -matrix) is modied. This technique helps to prevent mesh locking and, thus, provides accurate solutions in incompressible or nearly incompressible cases: see Nagtegaal et al. (1974). In addition, ABAQUS uses the average strain in the third (out-of-plane) direction for axisymmetric and generalized plane strain problems. Hence, in the two-dimensional elements only the in-plane terms need to be modied. In the three-dimensional elements the complete volumetric terms are modied. This may cause slightly different behavior between plane strain elements and three-dimensional elements for which a plane strain condition is enforced by boundary conditions. In a nite-strain formulation the selectively reduced-integration procedure works as follows. Dene the modied deformation gradient
; J where F = @ x=@ X is the deformation gradient; n is the dimension of the element; J Jacobian at the Gauss point; and J is the average Jacobian over the element, J
= 1
F=F
J 1
= det(F) is the
Ve` Ve`
J dVe` :
3.2.47
For three-dimensional elements n = and J (J ) are the volume change; for two-dimensional elements n and J (J ) are the change in area. Note that in the last case J is the change in area averaged over the element volume, which is different from the actual element area for distorted elements with variable thickness. The modied rate of deformation tensor, , is obtained from the modied deformation gradient as !
=2
where f = for two-dimensional elements, f = for three-dimensional elements, and is the identity matrix in two or three dimensions. This expression can also be written directly in terms of the velocities: where
=1 2
=1 3
@ @
_ u D = sym x + I trace _ u= 1 Z x e` v
@ @ f @ @ v
@ @
e`
_ _ u 0 trace u x x _ u e` x
@ @ dv :
This expression is used in the virtual work equation, where it is used to obtain the nodal forces from the element stresses. In ABAQUS the central difference operator is used to dene an increment of strain from the rate of deformation tensor, so we can write
For axisymmetric and generalized plane strain elements, the out-of-plane component of the deformation gradient is obtained by averaging over the original element volume,
F33 dVe` ; Ve` and the out-of-plane strain increment is calculated by averaging over the current element volume, F V
1u + I trace 1u 0 trace 1u 1D = sym x x x In the above x = (1 2)(xt + xt+1t ) is the position of the point at the middle of the increment.
@ @ f @ @ @ @ : =
Z = 1 33 e` Z
1 33 = 1
D
33 dve` :
The composite solid elements are also integrated numerically to obtain the element matrices. Gauss quadrature is used in the layer plane, and Simpsons rule is used in the stacking direction. These integration positions are referred to as integration points and section points, respectively, for output purposes. The number of section points required for the integration through the thickness of each layer is specied by the user.
Reference
Solid (continuum) elements, Section 14.1.1 of the ABAQUS Analysis Users Manual
3.2.48
3.2.5
The lower-order quadrilateral continuum elements in ABAQUS/Standard of type CPS4I, CPE4I, CAX4I, CPEG4I, and C3D8I as well as the related hybrid, coupled temperature-displacement and pore pressure elements are enhanced by incompatible modes to improve the bending behavior. In addition to the displacement degrees of freedom, incompatible deformation modes are added internal to the elements. The primary effect of these degrees of freedom is to eliminate the so-called parasitic shear stresses that are observed in regular displacement elements if they are loaded in bending. In addition, these degrees of freedom eliminate articial stiffening due to Poissons effect in bending. In regular displacement elements the linear variation of the axial stress due to bending is accompanied by a linear variation of the stress perpendicular to the bending direction, which leads to incorrect stresses and an overestimation of the stiffness. The incompatible modes prevent such a stress from occurring. In the nonhybrid elements (except CPS4I) additional incompatible modes are added to prevent locking of the elements for approximately incompressible material behavior. For fully incompressible material behavior, hybrid elements must be used. In these elements pressure degrees of freedom are added to enforce a linear pressure variation inside the element. In the hybrid elements the additional incompatible modes used to prevent locking are not included. The incompatible mode elements perform almost as well as second-order elements in many situations if the elements have an approximately rectangular shape. The performance is considerably less if the elements have a parallelogram shape. For trapezoidal element shapes the performance is not much better than the performance of regular displacement elements. Because of the internal degrees of freedom (4 for CPS4I; 5 for CPE4I, CAX4I, and CPEG4I; and 13 for C3D8I) the elements are somewhat more expensive than regular displacement elements. However, the additional degrees of freedom do not substantially increase the wavefront size since they can be eliminated immediately. In addition, it is not necessary to use selectively reduced integration, which partially offsets the cost of the additional degrees of freedom. The geometrically linear incompatible mode formulation used in ABAQUS is related to the work presented by Simo and Rifai (1990). Simos formulation is very similar to much earlier work done by Wilson et al. (1973) and Taylor et al. (1976). The nonlinear formulation is based on work by Simo and Armero (1992).
Geometrically linear formulation
Elements
As discussed in the paper by Simo, the incompatible mode formulation can be derived in a rigorous way from the general Hu-Washizu variational principle. In this discussion we will not present this derivation but use only the key results of Simos work. In the incompatible mode formulation, the displacement gradient additional, incompatible displacement gradient eld
~ G: ~ ~ G = G+G = u +G x
@ @
G = u is augmented with an x
@ @
3.2.51
The incompatible displacement gradient ~ is chosen internal to an element. The eld cannot be selected arbitrarily. It must be independent of the regular displacement gradient.
~ G \ G = 0;
G 6= 0:
(3.2.51)
In addition, it must be orthogonal to any constant gradient eld, which yields the condition
Vel
~ G dVel = 0:
If these conditions are violated, the element does not pass the patch test. The last condition is used to obtain a suitable general form of the incompatible modes. We describe the incompatible eld as a transformation of a parametric gradient eld ~ ( ):
where
(3.2.52)
j ( ) is the Jacobian of the parametric transformation at the location , and j (0) is the Jacobian at the center of the element. For planar elements the Jacobian can be written as j ( ) = h( ) det @
@
j ( ) = 2r( ) det
@
j ( ) = det
@
:
~ g( ):
Substitution of Equation 3.2.52 in Equation 3.2.51 allows us to create a simple condition for
Z
Vel
~ G dVel =
j (0)
Vel j ( )
Vpar
3.2.52
Z +1 Z +1
01 01
and for three-dimensional elements,
g( ) d1d2 = 0; ~
~ g() d1d2d3 = 0:
Z +1 Z +1 Z +1
01 01 01
This makes it possible to write written in the form
i
where are vectorial degrees of freedom and i are vectors and the summation i extends over the parametric coordinates. In two-dimensional elements 1 and 2 are vectors of the form
1 1 = 0
and in three-dimensional elements
and
2 = 0 ; 2
Elements
j (0) 1 t 01 : j ( ) i
With the addition of these terms, parasitic shear and Poissons effect in bending are eliminated. Note that the vectors i in ~ appear in a similar form as the nodal displacement vectors i in ,
u G
G = u @N ; @x
i i
and can be treated similar to displacement degrees of freedom. For approximately incompressible material behavior, bilinear patterns in the hydrostatic stress can still be observed in all elements except CPS4I. These patterns can be eliminated by the introduction of additional incompatible modes of the form
~ G ( ) =
I j
j (0) # ( ) I; j ( ) j 1 is
where j are additional scalar degrees of freedom. For two-dimensional elements a single term added with
# 1 = 1 2 :
3.2.53
#1 = 1 2 ; #2 = 1 3 ; #3 = 2 3 ;
Thus, the incompatible displacement gradient
and
# 4 = 1 2 3 :
11 22 33 12 13 23
9 8 > > > > > > = < ~ > = >G > >G ; >G :~ ~
~ > G11 9 ~ > G22 > ~ G33 = : ~ 21 + G12 > ~ > 31 + G13 > ~ ; 32 + G23
Since we want to use a formulation that can be used for any material model, we want to express the incompatible modes as a modication of the deformation gradient F. The most obvious approach is to add the incompatible modes to the deformation gradient:
~ @x ~ F = F + F = @ X + F:
(3.2.53)
This approach has been used successfully by Simo and Armero. Elements formulated on this basis satisfy the large-strain patch test; i.e., any patch of elements will be able to represent homogeneous deformations exactly. However, once the elements become distorted due to deformation, the patch test will no longer be satised in an incremental sense; that is, subsequent homogeneous deformations will not be represented exactly. This turns out to be a fatal aw in the formulation for problems involving large distortions in compression. Satisfaction of the instantaneous patch test requires the addition of an incompatible deformation rate tensor to the standard rate of deformation:
3.2.54
Principal incompatible modes are then added to the incremental deformation gradient:
Similar to Equation 3.2.52 the principal incompatible modes are described as a transformation of the parametric gradient eld g :
where tt is the parametric transformation at the center of the element in the state at the start of the increment,
(3.2.54)
j ( ) is the Jacobian of the parametric transformation at the location ; and j (0) is the Jacobian at the centroid at the start of the increment. Note that j ( ) is evaluated based on the deformation caused
by the displacement degrees of freedom only and does not include the volume change due to the incompatible modes. The incremental parametric gradient eld g has exactly the same form as in the linear formulation,
x tt = @@t
=0
1~( ) 1~( ) = 1i i; g
Elements
Bilinear volumetric terms are added to the principal terms in a multiplicative way:
(3.2.55)
The variation in the gradient of the position with respect to the current state is obtained from the fundamental relation
L = F 1 F
with
01
~ = 1F 1F01 = (1F + 1F)1FI + 1Fp1FI 1F01 ~ = (1F + 1F) 1F01 + 1FI 1FI01; p
h i
1 1 1
3.2.55
L
The integral of the principal incompatible modes over the element volume at the start of the increment is, hence, equal to
Z
vel
~ p 1F 1 1F01 dvel =
Vpar
Note that the integral will vanish if the incremental deformation is homogeneous; that is,
1Fp( ) =
vel
01
Vpar
gp ( ) dVpar ~
!
1
t01 1F01 = 0: t 0
1
Hence, the incremental patch test will be satised. The rate of the gradient of the position with respect to the current state is obtained similar to the variation
~ dL = dF F01 = d1F 1F01 = (d1F + d1F)1FI + 1Fp d1FI 1F01 ~ = (d1F + d1F) 1F01 + d1FI 1FI01; p
i
1 1 1 1
with
d1F = @dx ; @ xt j (0) ~ d1F = j( ) d~( ) t01 = di j(0) i t01; g t t j ( ) t t d1FI = dj j(0) #j ( ) 1FI : j ( ) t
1
For a nite-strain increment we use the midincrement approach proposed by Hughes and Winget. This yields
1 1L 1F 1 1F0+1t=2: t
The second variation is obtained in the usual way. Since the regular deformation gradient and the principal incompatible modes are purely displacement based, the initial stress stiffness terms are readily Z obtained as where p 1 p is the gradient of the p p1 p is the gradient of the velocity and p T 1 p displacement variation including the primary incompatible modes. Further, p p is 2 T the rate of deformation, and p 1 p p is the variation in the deformation. 2
dL = dF 1F
01
vel
D = (L + L )
L = F 1F01 dD = (dL + dL )
3.2.56
The additional bilinear modes appear in the rst variation only as variations (the values i themselves do not appear); hence, the contributions to the second variation can be neglected.
Reference
Solid (continuum) elements, Section 14.1.1 of the ABAQUS Analysis Users Manual
Elements
3.2.57
3.2.6
The library of solid elements in ABAQUS includes rst- and second-order triangles, tetrahedra, and wedge elements for planar, axisymmetric, and three-dimensional analysis. Hybrid versions of these elements are provided for use with incompressible and nearly incompressible constitutive models (see Hybrid incompressible solid element formulation, Section 3.2.3, for a detailed discussion of the formulation used). However, these hybrid forms should be used only to ll in regions in meshes made of brick elements; otherwise, too many constraint variables may be introduced. Second-order tetrahedra are not suitable for the analysis of contact problems: a constant pressure on an element face produces zero equivalent loads at the corner nodes. In contact problems this makes the contact condition at the corners indeterminate, with failure of the solution likely because of excessive gap chatter. The same argument holds true for contact on triangular faces of a wedge element.
Interpolation
The interpolation is dened in terms of the element coordinates g, h, and r shown in Figure 3.2.61. Since ABAQUS is a Lagrangian code for most applications, these are also material coordinates. They each span a range from 0 to 1 in an element but satisfy the constraint that g + h 1 for triangles and wedges and g + h + r 1 for tetrahedra. The node numbering convention used in ABAQUS for these elements is also shown in Figure 3.2.61. Corner nodes are numbered rst, and then the midside nodes for second-order elements. The interpolation functions are as follows. First-order triangle (3 nodes):
u
= (1
Elements
0 g 0 h) u1 + g u2 + h u3
0 g 0 h 0 r) 0 1)(1 0 g 0 h 0 r) u1 + (2g 0 1)g u2 + (2h 0 1)h u3 + (2r 0 1)r u4 + 4(1 0 g 0 h 0 r )g u5 + 4gh u6 + 4(1 0 g 0 h 0 r )h u7 + 4(1 0 g 0 h 0 r )r u8 + 4gr u9 + 4hr u10
3.2.61
h 3 4
10 6 5 h 8 9 7 1 4 g 2 1 5 2 g 6 3
r 12 4 10 18 13 16 9 1 7 2 14 8 g 5 17 11
15 h
3.2.62
1 = 2 (1 0 0
g
h)(1
r )u 1
1 2
g (1
r )u 2
1 2
h(1
r )u 3
(1
0 0
g
h)(1
+ r )u 4 +
1 2
g (1
+ r )u5 +
1 2
h(1
+ r )u 6
0 0
g
h)(2(1
0 0 ) 0 1)(1 0 ) 0 (1 0 0
g h r g r g (1
h)(1
))u1
h(1
0 1)(1 0 ) 0
h r
))u2 +
1 2
(h(2h
0 1)(1 0 ) 0
r g h r r
))u3
0 0 )(2(1 0 0 ) 0 1)(1 + ) 0 (1 0 0 )(1 0 2)) 4 0 1)(1 + ) 0 (1 0 2)) 5 + 1 ( (2 0 1)(1 + ) 0 (1 0 2)) 2 + 2(1 0 0 ) (1 0 ) 7 + 2 (1 0 ) 8 + 2 (1 0 0 )(1 0 ) 9 + 2(1 0 0 ) (1 + ) 10 + 2 (1 + ) 11 + 2 (1 0 0 )(1 + ) 12 2 2 2 + (1 0 0 )(1 0 ) 13 + (1 0 ) 14 + (1 0 ) 15
g h r u g r u h h h r g g h g h g r u gh r u h g h r u r u u gh r u h g h r u g h r g r u h r u
u6
Elements
Second-order variable 1518 node wedge (assuming all 18 nodes are dened):
u
1 g h)(2(1 g =( ((1 2 1 1)(1 r) + ( (g (2g 2 1 1)(1 r) + ( (h(2h 2 1 g h)(2(1 + ( ((1 2 1 1)(1 + r ) + ( (g (2g 2 1 1)(1 + r ) + ( (h(2h 2
0 0
0 0 ) 0 1)(1 0 ) 0 (1 0 0
h r g g (1 h
h)(1
)) +
1 4
(N16 +
N18 ))u1
0 0 0
0 0 0 0 0
0 (1 0
h
0 0
0 0 ) 0 1)(1 + ) 0 (1 0 0
g r g g (1
)) +
1 4
(N16 +
N18 ))u4
(N16 + N17 ))u5 4 1 2 h(1 r )) + (N17 + N18 ))u6 4 1 1 N16 )u7 + (2gh(1 N17 )u8 + (2(1 g h)g (1 r) r) 2 2 1 1 N18 )u9 + (2(1 N16 )u10 + (2h(1 g h)(1 r) g h)g (1 + r ) + 2 2 1 1 N18 )u12 + (2gh(1 + r ) + N17 )u11 + (2h(1 g h)(1 + r ) + 2 2 1 1 2 2 (N16 + N18 ))u13 + (g (1 (N16 + N17 ))u14 + ((1 g h)(1 r ) r ) 2 2 1 2 (N17 + N18 ))u15 + +N16 u16 + N17 u17 + N18 u18 ; + (h(1 r ) 2
r
)) +
0 0 0
0 0 0 0 0 0 0
0 0
0 0
0 0
0 0
0 0 0 0 0
3.2.63
where
N16
=4
g (1
0 0
g
h)(1
N17 N18
= 4gh(1
= 4h(1
0 0
g
0
)
h)(1
):
Integration
The rst-order triangle and tetrahedron are constant stress elements and use a single integration point for the stiffness calculation when used in stress/displacement applications. A lumped mass matrix is used for both elements, with the total mass divided equally over the nodes. For heat transfer applications a three-point integration scheme is used for the conductivity and heat capacity matrices of the rst-order triangle, with the integration points midway between the vertices and the centroid of the element; and a four-point integration scheme is used for the rst-order tetrahedron. Distributed loads are integrated with two and three points for rst-order triangles and tetrahedrons, respectively. The three-point scheme is also used for the stiffness of the second-order triangle when it is used in stress/displacement applications. The mass matrix is integrated with a six-point scheme that integrates fourth-order polynomials exactly (Cowper, 1973). Distributed loads are integrated using three points. The heat transfer versions of the element use the six-point scheme for the conductivity and heat capacity matrices. For stress/displacement applications the second-order tetrahedron uses 4 integration points for its stiffness matrix and 15 integration points for its consistent mass matrix. For heat transfer applications the conductivity and heat capacity matrices are integrated using 15 integration points. The rst-order wedge uses 2 integration points for its stiffness matrix but 6 integration points for its lumped mass matrix. The second-order wedge uses 9 integration points for its stiffness matrix but 18 integration points for its consistent mass matrix. The integration schemes used for the second-order tetrahedra and wedge elements can be found in Stroud (1971).
Reference
Solid (continuum) elements, Section 14.1.1 of the ABAQUS Analysis Users Manual
3.2.64
3.2.7
The generalized plane strain theory used in ABAQUS assumes that the model lies between two bounding planes, which may move as rigid bodies with respect to each other, thus causing strain of the thickness direction bers of the model. It is assumed that the deformation of the model is independent of position with respect to this thickness direction, so the relative motion of the two planes causes a direct strain of the thickness direction bers only. This strain and its rst and second variations are dened as follows. Let P0 X0 ; Y0 be a xed point in one of the bounding planes, as shown in Figure 3.2.71. The length uz , where t0 is the length of of the ber between P0 and its image in the other bounding plane is t0 this ber in the initial conguration and uz is the change in length of this ber. uz is the value of degree of freedom 3 at the reference node of the element.
+1
Bounding planes
Elements
(x,y)
(X0 ,Y0 )
Length of line through the thickness at (x,y) is t0 + uz + x (y - Y0) - y (x - X0) where quantities are defined in the text.
3.2.71
The reference node should be the same for all elements in any given connected region so that the bounding planes are the same for that region. Different regions may have different reference nodes. Since the bounding planes are rigid, the length of a ber at any other point x; y in the element is
( ) t = t0 + 1uz + 1 (y 0 Y0 ) 0 1 (x 0 X0 );
x y x y x y x y
of freedom 4 and 5 at the reference node of the element. The thickness direction logarithmic strain is
"zz
1 = (1 )j0 + (1 )1; 1 = (1 )j0 + (1 )1; where (1 )j0; i = x; y; are the initial values of 1 , specied by the user; and (1 )j1 are the degrees
where
i i i
= ln
t
t0
= t ; t
y x z
where
t
= 01
x
+ 1
y
0 x 1 + y 1 + 1u ;
= dt 0 tt2dt ; t
x y x:
where
dt
References
Choosing the elements dimensionality, Section 13.1.2 of the ABAQUS Analysis Users Manual Two-dimensional solid element library, Section 14.1.3 of the ABAQUS Analysis Users Manual
3.2.72
AXISYMMETRIC ELEMENTS
3.2.8
AXISYMMETRIC ELEMENTS
ABAQUS includes two libraries of solid elements, CAX and CGAX, whose geometry is axisymmetric (bodies of revolution) and which can be subjected to axially symmetric loading conditions. In addition, CGAX elements support torsion loading. As a result, CGAX elements will be referred to as generalized axisymmetric elements, and CAX elements as torsionless axisymmetric elements. In both cases, the body of revolution is generated by revolving a plane cross-section about an axis (the symmetry axis) and is readily described in cylindrical polar coordinates r, z , and . The radial and axial coordinates of a point on this cross-section are denoted by r and z , respectively. At = 0, the radial and axial coordinates coincide with the global Cartesian X - and Y -coordinates. If the loading consists of radial and axial components that are independent of and the material is either isotropic or orthotropic, with being a principal material direction, the displacement at any point will only have radial (ur ) and axial (uz ) components and the only stress components that will be nonzero are rr , zz , , and rz . Moreover, the deformation of any rz plane completely denes the state of strain and stress in the body. Consequently, the geometric model is described by discretizing the reference cross-section at = 0. If one allows for a circumferential component of loading (which is independent of ) and for general material anisotropy, displacements and stress elds become three-dimensional, but the problem remains axisymmetric in the sense that the solution does not vary as a function of and the deformation of the reference rz cross-section still characterizes the deformation in the entire body. The motion at any point will have, in addition to the aforementioned radial and axial displacements, a twist (in radians) about the z -axis, which is independent of . This section describes the formulation of the generalized axisymmetric elements. The formulation of the torsionless axisymmetric elements is a subset of this formulation.
Kinematic description
Elements
The coordinate system used with both families of elements is the cylindrical system (r, z , ), where r measures the distance of a point from the axis of the cylindrical system, z measures its position along this axis, and measures the angle between the plane containing the point and the axis of the coordinate system and some xed reference plane that contains the coordinate system axis. The order in which the coordinates and displacements are taken in these elements is based on the convention that z is the second coordinate. This order is not the same as that used in three-dimensional elements in ABAQUS, in which z is the third coordinate, nor is it the order (r, , z ), usually taken in cylindrical systems. Let eR , eZ , and e2 be unit vectors in the radial, axial, and circumferential directions at a point of the point can be in the undeformed state, as shown in Figure 3.2.81. The reference position represented in terms of the original radius R and the axial position Z :
Likewise, let r , z , and be unit vectors in the radial, axial, and circumferential directions at a point in the deformed state. As shown in Figure 3.2.81, the radial and circumferential base vectors
e e
X = R e R + Z eZ :
3.2.81
AXISYMMETRIC ELEMENTS
er
x e eR X
e Z, e z
Figure 3.2.81 Cylindrical coordinate system and denition of position vectors. depend on the coordinate: er = er ( ) and e = e (). The current position x of the point can be represented in terms of the current radius r and the current axial position z :
x = r e r + z ez :
The general axisymmetric motion at a point can be described by
(3.2.81)
As the above description implies, the degrees of freedom ur , uz , and are independent of 2. Moreover, the reference cross-section of interest is at 2 = 0, but for the benet of the mathematical analysis to follow it is important that 2 be nonzero in the above expression for .
Parametric interpolation and integration
( (
(3.2.82)
3.2.82
AXISYMMETRIC ELEMENTS
where g, h are isoparametric coordinates in the reference rz cross-section at 2 = 0, and ur N , uz N , N are the nodal degrees of freedom. The interpolation functions N N (g; h) are those described in Solid isoparametric quadrilaterals and hexahedra, Section 3.2.4, where the integration scheme of isoparametric solid elements is also discussed.
Deformation gradient
For a material point in space, the deformation gradient position x with respect to the original position X:
@x F = @X :
The current position x is given by Equation 3.2.81, and the gradient operator can be described in terms of partial derivatives with respect to the cylindrical coordinates:
@ 1 @X
( ) = @( ) e + @( ) e + 1 @( ) e : R Z 2 @R @Z R @2
1 1 1
Since the radial and circumferential base vectors depend on the original circumferential coordinate , the partial derivatives of these base vectors with respect to are nonvanishing:
@ er @
= e ;
@ e @
= 0er :
e @ @ = @@r @Z = @Z e :
Elements
e @ @ = @@r @R = @R e and
@ er @Z
(3.2.83)
=4
1 + @ur =@R
@uz =@R r@=@R
1+
r=R
0 0
3 5 ;
where the motion given by Equation 3.2.82 has been used explicitly. Similarly, the inverse deformation gradient F01 is readily obtained as
F01 = @R eR er + @R eR ez + @Z eZ er + @Z eZ ez + @r @z @r @z
R @ @r
2 e e + R @2 e e + R @2 e e : 2 r 2 z r @ 2 @z
3.2.83
AXISYMMETRIC ELEMENTS
Virtual work
As discussed in Equilibrium and virtual work, Section 1.5.1, the formulation of equilibrium (virtual work) requires the virtual velocity gradient L, which is the variation in the gradient of the position with respect to the current state. This tensor is given by
where F is the linearized deformation gradient. ABAQUS formulates the nite element equations in terms of a xed spatial basis with respect to the axisymmetric twist degree of freedom. Therefore, the desired result for F in Equation 3.2.84 does not simply follow from the linearization of Equation 3.2.83. Namely, it is necessary to cancel out the contributions from the variations
r
L = F F 01
1
(3.2.84)
e = e and
e = 0 er :
To this end
where
or
e er + r @z e ez + r e e :
3.2.84
@
r
(3.2.85)
AXISYMMETRIC ELEMENTS
~ ~ ~ D = 1 L + LT : 2
As shown in Procedures: overview and basic equations, Section 2.1.1, the contribution of the internal work terms to the Jacobian of the Newton method that is used in ABAQUS/Standard for solid element formulations is Z
d
5=
~ ~ (d : D + : d D) dV:
~ ~ ~ ~ ~ ~ ~ D = dLT L 0 2dD 1 D + dF 1 F01; ~ ~ ~ where dL has the same form as L in Equation 3.2.85. Moreover, in this formulation d F is nonzero, ~ 1 F01 has the form ~ and it can be shown with the aid of Rotation variables, Section 1.3.1, that d F
d
1
d
Elements
In component form,
~ ~ d F 1 F01 =
Reference
r
+ dr @r e er +
@
@d @z
r
+ dr @z e ez :
@
Axisymmetric solid element library, Section 14.1.6 of the ABAQUS Analysis Users Manual
3.2.85
3.2.9
ABAQUS/Standard includes a library of solid elements whose geometry is initially axisymmetric and that allow for nonlinear analysis in which bending can occur about the plane = = in the (r, z , ) cylindrical coordinate system of the model. The geometric model is dened in the rz plane only. The displacements are the usual isoparametric interpolations with respect to r and z , augmented by Fourier expansions with respect to . Since the elements are written for bending about the plane = only, they cannot be used to model torsion of the structure about the original axis of symmetry. Because the elements are intended for nonlinear applications, the orthogonality properties associated with Fourier modes cannot be used to reduce the problem to a series of smaller, uncoupled, cases, since the stiffness before projection onto the Fourier modes is not necessarily constant. For this reason these elements are signicantly more expensive to use than the corresponding axisymmetric elements intended for axisymmetric deformations.
= 2
Interpolation
The coordinate system used with these elements is the cylindrical system (r, z , ), where r measures the distance of a point from the axis of the cylindrical system, z measures its position along this axis, and measures the angle between the plane containing the point and the axis of the coordinate system and some xed reference plane that contains the coordinate system axis. The order in which the coordinates and displacements are taken in these elements is based on the convention used in ABAQUS for axisymmetric elements, so that z is the second coordinate. This allows these elements to be used in conjunction with other elements in the library that allow only axisymmetric deformation. This order is not the same as that used in three-dimensional elements in ABAQUS in which z is the third coordinate, nor is it the order (r, , z ), usually taken in cylindrical systems. The original geometry of the elements is assumed to be axisymmetric with respect to the axis of the coordinate system and, thus, independent of . Let er , ez , and e be unit vectors in the radial, of axial, and circumferential directions at a point in the undeformed state. The reference position the point can be represented in terms of the original radius R and the axial position Z :
Elements
X = R e r + Z ez :
Similarly the displacement of the point can be represented in terms of the components ur , uz , and u with respect to these same vectors at the original position of the point:
u = u r e r + uz e z + u e :
For small radial and circumferential displacements the circumferential displacement is proportional to R ), but for large displacements this relation becomes the change in circumferential angle (u R ur ), as shown in Figure 3.2.91. The distinction is of importance nonlinear (u only in geometrically nonlinear analysis with radially applied concentrated loads and/or sliding radial boundary conditions.
= ( + ) tan(1 )
= 1
3.2.91
x u u X ur
Figure 3.2.91 Displacement and rotation in the r plane. This denition of the degrees of freedom is equivalent to applying transformations to the global (x, y, z ) degrees of freedom associated with a standard continuum element. Hence, the nonlinear equations associated with these elements have the same structure as the equations for standard continuum elements. A general interpolation scheme for u using Fourier terms with respect to is used:
XH u =
M m=1
m (g; h)
u m0 +
P X p=1
mp mp (cos p uc + sin p us )
= r; z; )
where g, h are isoparametric coordinates in the original RZ plane; H m are polynomial interpolation functions; and um0 , ump , and ump are solution amplitude values. M is the number of terms used for c s interpolation with respect to g, h; and P is the number of terms used in the Fourier interpolation with respect to . Purely axisymmetric deformation results when P = 0. We reduce the number of variables in such an element by assuming that bending is allowed only about one plane, = =2, so that the plane = 2n, n integer, is a plane of symmetry. The only terms that satisfy this condition are
8 mp 9 8 91 8 9 0 8 m0 9 P M P < ur 0 = X < urc = X < 0 = < ur = X m uz ; = H (g; h) @: um ; + cos p : ump ; + sin p : 0 ;A : z zc : u m=1 p=1 p=1 ump 0 0
s
(3.2.91)
For convenience we use the values of the ur and uz displacement components at specic locations around the model between = 0 and = instead of the Fourier amplitudes um0 and ump . The r rc main reason for this is to allow the elements to be used with interface elements, such as slide lines,
3.2.92
for which physical displacement values are required. This is accurate only if it is assumed that the relative displacements in the -direction are small so that the interface conditions are considered with respect to ur and uz only; that is, in planes of constant . In addition, we omit the subscript s in mp u . Equation 3.2.91 is, therefore, the expression for the circumferential displacement: ump s rewritten
8 9 P 8 0 91 8u 9 M 0P +1 < = < r= X m r X p < ump = X uz 0 = H (g; h) @ R () : ump ; + sin p : mp ;A ; z : u ; m=1 p=1 p=1 u 0
(3.2.92)
where Rp () are trigonometric interpolation functions and ump , ump are physical radial and axial r z displacement components at = (p 0 1)=P . The Rp interpolators at the associated positions p are taken as P = 1:
R1 =
1
R2 =
2 1 2
(1 + cos ) (1
0 cos )
Elements
and
1 = 0
2 =
P = 2:
R1 = R2 = R3 =
1 4 1 2 1 4
(1 + 2 cos + cos 2 ) (1 (1
and
3 =
P = 3:
R1 = R2 = R3 =
1 6 1 3 1 3 1
and
1
=0
1 3 2 3
2 = 3 =
4 = P = 4: R1 = R2 = R3 =
1 8 1 4 1 4 1 1
2 cos
2 cos 3
0 cos 4)
and
5 =
P = 4 is the highest-order interpolation offered with respect to in these elements: the elements become signicantly more expensive as higher-order interpolation is used; and it is assumed that, because of this, full three-dimensional modeling is less expensive than using these elements with P > 4.
Integration
The integration scheme used in these elements is a product of integration with respect to element coordinates in surfaces that were originally in the RZ plane and integration with respect to . For the former the same scheme is used as in the corresponding purely axisymmetric elements (for example, either full or reduced Gauss integration in the isoparametric quadrilaterals). For integration with respect to the trapezoidal rule is used, with the number of integration points set to 2(P + 1).
3.2.94
Deformation gradient
For a material point in space the deformation gradient position x with respect to the original position X:
@x F = @X :
x can be described in terms of the original position X and the displacement u; x = X + u = (R + ur )er + (Z + uz )ez + u e ;
and the gradient operator can be described in terms of partial derivatives with respect to the cylindrical coordinates:
@X @
=
@ @ er @R + ez @Z
@ e R@ :
Since the radial and circumferential base vectors depend on the original circumferential coordinate : er = er (), e = e (), the partial derivatives of these base vectors with respect to are nonvanishing:
@ er @
=
e ;
@ e @
0er :
Elements
F = e r er
+ +
1+
ez er @uz @R e er @u @R
er ez @ur + er e @Z
Alternatively, this can be written in matrix form with components relative to the local reference basis er , ez , e :
2 2 3
F =4
1 + @ur =@R
(3.2.93)
To be able to analyze approximately incompressible material behavior, the volume change in the fully integrated 4-node quadrilaterals is assumed to be independent of g and h in an RZ plane. 2 3 Hence, F is modied according to
2
=
J0 J
1
3
2 3
F ;
3.2.95
where J = F is the volume change at the integration point and J0 is the average volume change over the RZ plane of the element. In addition, the part of the axisymmetric hoop strain that does not depend on is made independent of g and h. Experience has shown this considerably improves the solution accuracy for axisymmetric problems. Thus, we use
F
det2
= 1 + ur =R + @u =R@;
# ump ; r
where
M X X (0; 0) P +1 C p + X H m(g; h) P +1 (Rp() 0 C p) ur =R = R(0; 0) R(g; h) m=1 p=1 m=1 p=1 p the leading (constant) term in Rp ( ). with C M X Hm
"
Strain and rotation increments are calculated from the integrated velocity gradient matrix, as
1L = @x@1u ; t+1t=2 1 1u = xt+1t 0 1 1u. This expression is not easily evaluated directly, since where xt+1t=2 = xt + 2 2 points that were in an RZ plane in the undeformed shape will no longer be located in the same plane after deformation. Instead, we calculate the gradient of 1u with respect to the reference state and obtain 1L with the transformation 01 @ xt+1t=2 @X 1 @1u 01 : @ 1u @ 1u @ 1u 1L = @ X @ x = @X = @X F 0 2 @X @X
1
1L, dened
+1t=2
1L3 = 2@1u=@X 3
2
2 3 F
0 1 @ 1u=@X 2
2
01
(3.2.94)
with
1 01 1 1 +1 The strain increments are approximated as the symmetric part of 1L: i h 2 3 1" = 1 21L3 + 21L3T : 2
2 @ u=@X
1 1 1
1 1 1
(3.2.95)
As was the case for the deformation gradient, we modify the volume strain increment in the fully integrated 4-node quadrilaterals to be independent of g and h in an RZ plane, which yields
2 i h 1"3 = 1 21L3 + 21L3T ; with 21L3 = 21L3 + 1 0tr(1L0) 0 tr(1L)12I 3; 2 3
3.2.96
where I is the unit matrix, L Lrr Lzz L is the volume strain increment at L0 is the average volume strain increment over the RZ plane of the the integration point, and element. In addition we use the approximation
tr(1 ) = 1 tr(1 )
+1
+1
1ur =R =
"
M X Hm m
=1
1
, are approximated as the antisymmetric part of 1L, which in matrix i h 2 1
3 = 1 21L3 0 21L3T : 2
Virtual work
The formulation of equilibrium (virtual work) requires linearization of the strain-displacement relation in the current state. For fully integrated 4-node elements the volume strain modication provides
" "
= 1 L + LT 2
2
where
Elements
and
u L = @x : @
As was the case for the strain increments, the linearized strain-displacement relation involves taking derivatives in the deformed shape (x xt+1t), which is troublesome since points that were in an RZ plane in the undeformed shape will no longer be located in the same plane. Hence, L is computed in a similar manner to L:
u L = @X
@
X = @ u @x @X
L
3
1
x 01 = @ u F01: @X @X
= 2@u=@X 3 2F 301;
@ur =@Z @uz =@Z @u =@Z @ur =R@ u =R 5: @uz =R@ @u =R@ ur =R
(3.2.96)
where
2 2
@u=@X
(3.2.97)
3.2.97
"
Hm
#
ump : r
The displacements and, hence, the displacement variations, are interpolated in terms of nodal displacement variations with Equation 3.2.92. The derivatives of the displacements with respect to R, Z , and are readily obtained from these expressions:
( ) : ump ; + sin p : z m=1 p=1 p=1 u 0 8 mp 9 8 91 0 M P +1 P ur = X < 0 = X @H m X p < @ @uz =@Z R () ump 0 = + sin p : mp ;A ; : @u =@Z ; m=1 @Z : z ; p=1 p=1 u 09 8 8 mp 8 91 9 0 M P +1 P < @ur =@ = X < 0 = X @Rp < ur = X m @u =@ H @ ump 0 = + p cos p : mp ;A : z : @uz =@ ; m=1 @ : p=1 u 0 ; p=1
: @uz =@R ; 8 9 < @ur =@Z =
@u =@R
M X
@H m @R
0 @
P +1 X
Rp
8 mp 9 < ur =
P X
8 <
0 = 0 A; mp ;
91
Since the elements are formulated in terms of Cartesian components of displacements, the equations presented in Solid element formulation, Section 3.2.2, apply. For the 4-node quadrilaterals, we can adapt Equation 3.2.21 to the averaged volume change formulation, which yields
d
5=
Z
V
where dL has the same form as L: This can be worked out in terms of nodal degrees of freedom " with the expressions for L and " obtained in the previous paragraph on virtual work.
Hourglass control
In the 4-node reduced integration element the hourglass modes must be controlled. These modes are similar to the ones in regular axisymmetric elements but have some additional features. 1. The hourglass pattern can vary along the circumference, which requires application of an hourglass stiffness at multiple points around the circumference. 2. Hourglassing can also occur in the circumferential direction.
3.2.98
Hence, at each integration point around the circumference, we calculate the hourglass strains
q ( ) = x
( ) F ( ) 0 X
:
1
Here F is the deformation gradient as given by Equation 3.2.93 and x and modes in the deformed and undeformed geometry respectively:
x ( ) =
m=1
X
M
m x m ( ) ;
X =
m=1
X
M
m Xm ;
where m is the same hourglass operator as used for the 4-node axisymmetric continuum elements and xm () and Xm are the nodal positions at angle in the deformed and undeformed states. Observe that since the initial geometry is axisymmetric, Xm is independent of :
8 Rm 9 < = Xm = : Z m ; : 0
8 R m + u m ( ) 9 < = r x m ( ) = X m + u m ( ) = : Z m + u m ( ) ; : z m u ()
Elements
8 Rm 9 P +1 8 9 P 8 0 9 < m = X p < ump = X < = r xm = : Z ; + R () : ump ; + sin p : 0 ; : z p=1 p=1 ump 0 0
The hourglass strain transforms into an hourglass force with the hourglass stiffness c:
Q = cq :
This hourglass stiffness can be obtained with the same procedure as used for the regular axisymmetric elements, where the only difference is the scaling factor required to reect the fact that each point reects only part of the circumference. The rst variation of q is readily obtained as
q = x
1 F + x 1 F = u 1 F + x 1 @ u : @X
8 9 P 8 91 0P +1 < 0 = r X p < ump = X A u
=
m um =
m @ R () ump + sin p : 0z ; p=1 : u0mp ; : m=1 m=1 p=1
3.2.99
q = x
dF + d x
F = u
1 1
@d @
u X + d u
@ @
u X;
For geometrically linear problems equivalent nodal loads due to applied surface pressures and body forces are readily calculated since the geometry is axisymmetric. For geometrically nonlinear problems the treatment of body forces does not change because of the xed direction of the forces and because the forces are proportional to the volume, which is assumed to change by a negligible amount. However, for surface pressures nonaxisymmetric deformations must be taken into consideration. The equivalent nodal loads associated with surface pressure p can be obtained by considering the virtual work contribution
n udA =
1
2
Z +1
01
x 2 @ x 1 x dd; @ @
x = r e r + z e z + u e ;
with r = R + ur and z = Z + uz . Hence, the current position of a point can be expressed in terms of the surface interpolator hm ( ) and the standard circumferential interpolators:
8 9 8 91 8 9 0 M P +1 P < 0 = <r= X m X p < rmp = X A h ( ) @ R () z mp sin p 0 z = + : 0 ; p=1 : ump ; : : u ; m=1 p=1
(3.2.98)
x = @r e + @z e + @u e ; r z @ @ @ @r @ @x = 0 u e + @z e + r + @u e
@ @
r
@
@
;
and, hence,
x 2 @ x = 0 @z r + @u + @u @z e + r @ @ @ @ @ @
@r 0 u + @ r + @u @r @z @z @r @ 0 @ @ + @ @ 0 u e :
0 @u @
@r
@
ez +
3.2.910
Z
A
p n 1 u dA
2
Z +1
01
0p
With use of the interpolation functions we, thus, obtain the equivalent nodal forces:
mp Fr =
Z 2 Z +1
0
mp Fz =
Z 2 Z +1
1 Z0 2 Z0+1 0
01
0ph
m
( )R ( )
@z @
r+
@u @
@u @z
0ph
( )R ( )
@u @r
F
mp
01
0 p hm ( ) sin p
Z 2 Z +1
1 Z0 2 Z0+1 0
0 u 0 @ r + @ @z @r @r @z 0 @ @ 0 u dd: @ @
@ @
@r
@ @
dd
@u
dd
For geometrically linear analysis this reduces to the standard axisymmetric equivalent nodal loads
Elements
mp Fr = mp Fz =
F
mp
= 0:
01
mpnq
3 8 nq 9 < du = nq 5 du r ; : du z ; nq
with
Z 2 Z +1
0
mpnq Kzr =
Z 2 Z +1
0
01
p hm ( )Rp ()
01
p hm ( )Rp ()
+
mpnq Kr =
Z 2 Z +1
0
@u n h ( ) @ d
m
r+
@u @
dhn
d
R q ( )
01
p h ( ) sin p
@z dhn @ d
R ( )
@z n dRq h ( ) @ d
dd
3.2.911
mpnq Krz
2 Z
+1
01
p h ( )R ( )
r+
@u @
dhn
d
R ( )
@u n dRq h ( ) @ d
dd
Z 2 Z +1
0
Z 2 Z +1
0
01
@r n dRq h ( ) p h ( ) sin p @ d
m m p
@r @
0 u
dhn d
R () dd
Kz
01 Z 2 Z +1
0
K
mpnq
Z 2 Z +1
0
@z n h ( ) cos q p h ( )R ( ) q @
@z dhn
@z @
h ( ) sin q dd:
In the case of hydrostatic pressure (p dependent on z ) some additional terms appear. These terms are readily obtained from the expression
Z
A
dp
n 1 u dA =
Z 2 Z +1
0
01
0 duz dz
dp
@
@z
@u
@u @z
With use of the interpolation functions we, thus, obtain the additional load stiffness contributions:
0 @ @ hn( )Rq () dd @u @u @r @r mpnq m p h ( )R ( ) 0 u 0 @ r + @ hn ( )Rq () dd = Kzz @ @ 0 Z01 dz @r @z @z @r Z 2 +1 dp mpnq hm ( ) sin p 0 @ @ 0 u hn ( )Rq () dd: = Kz dz @ @
mpnq Krz =
Z 2 Z +1 dp Z0
0
01 dz 2 Z +1 dp 01
h ( )R ( )
@z
@
r+
@u @
@u @z
Mass matrix
At each material point the displacement components in the three directions (radial, axial, circumferential) are dependent only on the corresponding nodal displacement components. Hence,
3.2.912
the mass matrix does not involve any coupling between the radial, axial, and circumferential degrees of freedom, and we can write the mass matrix in the form of three separate expressions:
mnpq Mrr
Z ZV ZV
V
mp nq Nr N r d V mp nq Nz N z d V
N
mp
N dV:
nq
Here the superscripts m and n refer to a particular node in the rz plane, and the superscripts p and mp mp q refer to a particular position along the circumference. The interpolation functions Nr , Nz , and mp m (g; h) in the r z plane and interpolation functions N are the product of interpolation functions H in the -direction: mp m p mp
Nr
= H (g; h)R ( ) = Nz mp N = H m (g; h) sin p:
The volume integral used to form the mass matrix can be split into an integral over the rz crosssection and an integral around the circumference. For the rr component of the mass matrix this yields Z 2 Z 1
mnpq Mrr = A
2
Elements
This matrix can be written in a convenient form by dening the primitive mass matrix,
mn Mprim =
Z
A
This primitive mass matrix is the same mass matrix that is used for the regular axisymmetric elements. We can also dene the circumferential distribution matrices
f1 = f2 =
pq pq
1 2 1 2
Z Z
0 0
2 2
Rp ()Rq () d
sin p sin q d:
and
The radial, axial, and circumferential components of the mass matrix then take the form
mnpq mn mnpq Mrr = Mprim f1 = Mzz mnpq mn pq M = Mprim f2 : pq
The circumferential distribution matrices can be evaluated for various values of the number of terms P in the Fourier series. After some calculations the following results are obtained: P = 1:
f1 =
pq
1 3 1 1 3 8
f2 =
pq
(1) :
3.2.913
= 2:
f1 =
pq
1 32
0 @
2 12 2
01
2 20
01
2 7
1 A;
f2 =
pq
1 2
1 0
0 1
:
P = 3:
0 11 1 B 2 pq f1 = @ 02 72
1
02
4 20 2
02
1 02 C ; A
1 2 11
f2 =
pq
1 2
0 1 @0
0
0 1 0
1 0A:
0 1
P = 4:
0 15 2 02 2 01 1 B 2 28 4 04 2 C 1 B pq f1 = C B 02 4 28 4 02 C ; 128 @ 2 04 4 28 2 A 01 2 02 2 15
01 1 B0 f pq = @
2
2 0 0
0 1 0 0
0 0 1 0
1 0C A:
0 0 1
For hybrid and pore pressure elements additional degrees of freedom p are added. In the hybrid elements these degrees of freedom are internal to the element and represent the hydrostatic pressure in the material. In the pore pressure elements the degrees of freedom represent the hydrostatic pressure in the uid as interpolated from the pressure variables at the external, user-dened nodes. Let the interpolation function for the (hydrostatic or pore) pressure in the rz plane be denoted by Gm (g; h). The interpolation functions are the same as for the regular axisymmetric hybrid and pore pressure elements, respectively. Along the circumference, we observe that in the geometrically linear formulation the volumetric strain only shows cosine dependence:
= @uz ur @u @ur + + + : @R @Z R R@
Gm (g; h)
P =1 p=1
Rp () pmp :
In the nonlinear case will exhibit higher-order variations in . For approximately incompressible materials, these higher-order terms are likely to lead to locking of the nite element mesh for nonaxisymmetric deformations. In the hybrid formulation, however, the higher-order terms in are not used for calculation of hydrostatic pressure: only the cosine terms as used in the interpolation for p are used. Hence, the hybrid elements prevent locking in the nonaxisymmetric modes as well as in the axisymmetric modes.
3.2.914
For a material point in space in the pore pressure element, the pore pressure gradient calculation involves taking derivatives of the pore pressure with respect to the current position x. Again, we do not evaluate the gradient directly but calculate it with respect to the original position X, with the following transformation: where F is the deformation gradient, and the cylindrical components of the scalar gradient of the pore pressure with respect to X are readily obtained from the following expressions:
@p @R @p @Z @x @p @p = @X
1
@X @x
@p = @ X F 01 ;
1
=
=
@p = @
R ( )p
mp ;
R ( )p
mp ; mp :
Elements
m=1
p=1
@R () p @
Reference
Axisymmetric solid elements with nonlinear, asymmetric deformation, Section 14.1.7 of the ABAQUS Analysis Users Manual
3.2.915
3.3.1
The stress analyst is often faced with problems dened in unbounded domains or problems in which the region of interest is small compared with the surrounding medium. The unbounded or innite medium can be approximated by extending the nite element mesh to a far distance, where the inuence of the surrounding medium on the region of interest is considered small enough to be neglected. This approach calls for experimentation with mesh sizes and assumed boundary conditions at the truncated edges of the mesh and is not always reliable. It is particularly of concern in dynamic analysis, when the boundary of the mesh may reect energy back into the region being modeled. A better approach is to use innite elements: elements dened over semi-innite domains with suitably chosen decay functions. ABAQUS provides rst- and second-order innite elements that are based on the work of Zienkiewicz et al. (1983) for static response and of Lysmer and Kuhlemeyer (1969) for dynamic response. The elements are used in conjunction with standard nite elements, which model the area around the region of interest, with the innite elements modeling the far-eld region.
Static analysis
The solution in the far eld is assumed to be linear, so only linear behavior is provided in the innite elements. The static behavior of the innite elements is based on modeling the basic solution variable, u (in stress analysis u is a displacement component) with respect to spatial distance r measured from 0 as r 1, and u ! 1 as r ! 0. The interpolation provides a pole of the solution, so that u terms of order 1=r, 1=r2 , and, when the solution variable is a stress-like variable (such as the pore liquid pressure in the analysis of ow through a porous medium), 1=r3 as r ! 0. The far-eld behavior of many common cases, such as a point load on a half-space, is thereby included. This modeling is achieved by using standard quadratic or cubic interpolation for u(s) in 01 s 1, where s is a mapped coordinate that is chosen such that the mapping r(s) causes r ! 1 as s ! 1. We obtain two- and three-dimensional models of domains that reach to innity by combining this interpolation in the s-direction in a product form with standard linear or quadratic interpolation in orthogonal directions in the mapped space. In using innite elements for static analysis, the pole must be located so as to provide a reasonable far-eld solution for the particular problem being modeled. The innite elements in ABAQUS are written with nodes on the interface between the nite and innite elements and, on each edge that stretches to innity, a node that must be placed in the innite direction such that the straight line from that node through the corresponding interface node passes through the pole for that ray at a distance on the other side of the interface from the innite element equal to the distance between these nodes (Figure 3.3.11). The one-dimensional concept is, thus, based on a node (node 1) on the interface between the nite and innite elements, distance r1 = a from the pole and at s = 01 in the mapped space, and node 2, at r2 = 2a from the pole (the pole is at r = 0) and at s = 0 is the mapped space. The r(s) mapping is chosen as
Elements
012 0
s s
r1
1+ 1
s s
r2 ;
3.3.11
r a 1 a 2 3
r0
r1
r2
r3
1 s = -1
2 s=0
3 s = +1
= 12 0
=1
a s
02
When an element with 1=r and 1=r2 behavior is required, we combine this geometric mapping with standard quadratic interpolation of u with respect to s, written in terms of its values at node 1 and at node 2:
u
=0
at
= 1,
where
u
s (s
0 1)
u1
+ (1
)u 2
u (r )
then
=(
u1
+ (2u1
04
u2 )
a 2
r
which provides the desired behavior. Likewise, when 1=r3 behavior is also required, we use cubic interpolation of u with respect to s, written in terms of its values at nodes 1 and 2 and at a third node, which we choose to place at s = 1 : 2
u
1 03
s (s
0 1)( 0 1 ) 1 0 2(1 0 2
s u a r
)(s
0 1) 2
u2
8 3
s(1
)u 3 :
=(
1 3
u1
04
u2
32 3
u3 )
+(
02
u1
+ 20u2
0 32
u3 )
a 2
r
+(
8 3
u1
0 16
u2
64 3
u3 )
a 3
r
3.3.12
The innite elements in ABAQUS consist of two- and three-dimensional elements for uncoupled stress analysis that use quadratic interpolation for displacement components and two- and three-dimensional elements for coupled stress-pore liquid pressure elements, in which the displacements use quadratic interpolation and the pore liquid pressure uses cubic interpolation in the innite direction. This higherorder interpolation is used for the pore liquid pressure for compatibility: since the displacement varies as 1=r2, the strain (and, therefore, the stress) may vary as 1=r3 .
Dynamic response
The dynamic response of the innite elements is based on consideration of plane body waves traveling orthogonally to the boundary. Again, we assume the response adjacent to the boundary is of small enough amplitude so that the medium responds in a linear elastic fashion. The equilibrium equation is
where is the materials density, u is the material particle acceleration, is the stress, and x is position. We assume the materials response is isotropic, linear elastic, andthuscan be written as
0u + @@x 1 = 0;
=
(1 + )(1
E E
0 2 )
and
) are Lams constants (E is Youngs modulus and is Poissons ratio). Introducing this material
2(1 +
G=
"=
1 2
T ) @u @u ; + @x @x
ui = G
@ 2 ui @xj @xj
+ ( + G)
@ 2 uj ; @xi @xj
where index notation has been used for simplicity. We consider plane waves traveling along the x-axis. Two body wave solutions of this form exist for this equation. One describes plane, longitudinal (push) waves, which have the form
u x = f (x 6 c p t);
uy = uz
= 0;
3.3.13
where, by substitution in the governing equation above, we nd that the wave speed, cp , is
cp =
+ 2G : ux = uz
u y = f (x 6 c s t);
or
=0
uz
f (x 6 c s t);
ux = u y = 0 ; G :
cs =
In each case the solution f (x 0 ct) represents waves moving in the direction of increasing x, while f (x + ct) represents waves moving in the direction of decreasing x. Now consider a boundary at x = L of a medium modeled by nite elements in x < L. We introduce distributed damping on this boundary, such that
xx = 0dp ux _ xy = 0ds uy _ xz = 0ds uz ; _ where we will now choose the damping constants dp and ds to avoid reection of longitudinal and shear wave energy back into the medium in x < L. Plane, longitudinal waves approaching the boundary have the form ux = f1(x 0 cp t), uy = uz = 0. If they are reected at all as plane, longitudinal waves, their reection will travel away from the boundary in some form ux = f2 (x + cp t), uy = uz = 0. Since the problem is linear, superposition provides the total displacement f1 + f2, with corresponding 0 0 0 0 stresses xx = ( + 2G)(f1 + f2 ), all other ij = 0, and velocity ux = 0cp (f1 0 f2 ). For this _ solution to satisfy the damping behavior introduced on the boundary at x = L requires
( + 2 G
and
0 d c )f 1 + ( + 2 G + d c )f 2 = 0 :
p p
0
p p
=0
0 (so that f2
= 0) =
dp =
+ 2G cp
cp :
ds = cs :
3.3.14
These values of boundary damping are built into the innite elements in ABAQUS. From the above discussion we see that they transmit all normally impinging plane body waves exactly (provided that the material behavior close to the boundary is linear elastic). General problems involve nonplane body waves that do not impinge on the boundary from an orthogonal direction and may also involve Rayleigh surface waves and Love waves. Nevertheless, these quiet boundaries work quite well even for such general cases, provided that they are arranged so that the dominant direction of wave propagation is orthogonal to the boundary or, at free surfaces and interfaces where Rayleigh or Love waves are of concern, they are orthogonal to the surface (see, for example, Cohen and Jennings, 1983). As the boundaries are quiet rather than silent (perfect transmitters of all waveforms), and because the boundaries rely on the solution adjacent to them being linear elastic, they should be placed some reasonable distance from the region of main interest. During dynamic response analysis following static preload (as is common in geotechnical applications), the traction provided by the innite elements to the boundary of the nite element mesh consists of the constant stress obtained from the static response with the quiet boundary damping stress added. Since the elements have no stiffness during dynamic analysis, they allow a net rigid body motion to occur, which is usually not a signicant effect.
Reference
Elements
3.3.15
3.3.2
Problems in unbounded domains, or problems in which the region of interest is small compared to the surrounding medium, are common in acoustic analysis. For example, a submarine deep underwater may experience loads due to the uid and radiate sound into the uid, as if the ocean were innitely large. The extent to which an exterior uid may be considered unbounded or innite depends on the number of wavelengths between the body or region of interest and the nearest boundary: the higher this number, the more likely that the inuence of these boundaries is small enough to be neglected. For example, the effect of the surrounding uid on a submarine in a relatively shallow harbor may be like an innite medium at high frequencies, but the harbor bottom and free surface may exert an effect at lower frequencies. Similarly, a loudspeaker in air may radiate sound from the high-frequency tweeter as if the surrounding air were innite, but the effects of the walls of the room may affect the radiation pattern of the low-frequency woofer. ABAQUS provides a range of features to model exterior uid effects. All of the surrounding uid may be modeled with nite elements; clearly, this is practical only if the extent of the surrounding medium is small. At the next level of sophistication, the user may model a small region of uid and apply a simple radiation boundary condition to the terminating surface. These radiation boundary conditions are derived using simple models of waves passing through a boundary. ABAQUS provides several alternative formulations of these boundary conditions (see Coupled acoustic-structural medium analysis, Section 2.9.1, and Acoustic, shock, and coupled acoustic-structural analysis, Section 6.9.1 of the ABAQUS Analysis Users Manual). These radiation conditions do not add degrees of freedom to the system and do not affect the symmetry of the matrix. Finally, acoustic innite elements are provided, which allow the retained nite element uid region to be even smaller, with similar accuracy. The acoustic innite element formulation differs from the radiation boundary condition formulation in several key respects. In the innite elements the innite exterior is subdivided into elements, and a method of weighted residuals statement is enforced on the elements in a manner entirely analogous to the usual nite element method. Degrees of freedom, corresponding to interpolation functions in the innite direction, are added to the overall matrix system. The method of weighted residuals used in ABAQUS results in nonsymmetric innite element matrices, so the relative cost of these elements is higher than that of a simple radiation boundary condition. However, the accuracy of the innite elements is sufciently high that the nite element region can be reduced considerably, offsetting the cost in many applications.
Transient formulation
Elements
The solution in the unbounded acoustic medium is assumed to be linear and governed by the same equations as the nite acoustic region:
Kf
p+
f K f
p0 _
@ @x
1 1
@p @x
= 0:
is used here to denote the volumetric drag parameter. Consider the innite exterior of a region of acoustic uid bounded by a convex surface and a conventional nite element mesh dened on this
3.3.21
surface. Each facet of this surface mesh, together with the normal vectors at the nodes, denes a subdivision of the innite exterior that will be referred to as the innite element. Application of the method of weighted residuals results in a weak form of this equation over the innite element volume:
Z
Vf
p
Kf
p+
f K f
p0 _
@ @x
1 1
@p f @x
dV
= 0:
This equation is formally identical to that used in the nite element region (see Coupled acousticstructural medium analysis, Section 2.9.1); however, the choice of basis functions for the weight, p, and for the solution eld, p, will be different. Selection of these functions has been the subject of considerable experimentation and analysis (for example, see Allik, 1991, and Burnett, 1994). In ABAQUS considerations of accuracy, numerical stability, time-domain well-posedness, and economy have led to the selection of a form proposed by Astley (Astley, 1994). Here, the Fourier transform is applied to the equilibrium equation, and the functions for the solution eld are chosen as tensor products of conventional nite element shape functions in the directions tangential to the terminating surface and polynomials in 1=r, where r is a coordinate in the innite direction, and a spatially oscillatory factor. The weights are chosen as complex conjugates of the solution eld functions times a 1=r2 factor. This combination has shown to result in an element with several desirable properties. First, if the material properties are constant with respect to frequency in steady-state analysis, the mass, damping, and stiffness element matrices are constant as well; equivalently, in transient analysis the element results in a second-order differential equation for the pressure. Moreover, the damping matrix has positive eigenvalues, a necessary condition for well-posedness in transient analysis. Analytical investigations of the formulation demonstrate that the element captures the exact solutions for radiation impedances for modes of a sphere; the order of the spherical mode modeled exactly is equal to the order of the polynomial used in the innite direction. The element integrals do not contain oscillatory kernels and can be evaluated using standard Gauss quadrature methods. Finally, the element can be formulated using the usual coordinate map due to Bettess (1984) so that arbitrary convex terminating surfaces can be used. To continue with the derivation, we transform the weighted residual statement into the frequency domain:
Vf
1 @ p dV = 0:
@ ~ 2 1 p 0
~ ~ ~ K p + i
K p 0 @ x 1 @ x
f f f f
(3.3.21)
3.3.22
1 @ @ 1 @F @i j j +F dV pj i @ x f @ x @ x f @ x Vf Z @ (r 0 r ) 1 @F 0 1 @ x i j dV pj 0ik @x Vf Z Ff
p +ik ij dV pj Vf f K f f Z F 0ik @ (r@0 r0 ) 1 @i j 0 i @j 1 @ (r@0 r0) dV pj x @x @x x Vf f Z F @ (r 0 r ) @ (r 0 r ) 0 0 1 dV p = 0: 0k 2 10
Z
1 1
(3.3.22)
Vf
Kf
@x
@x
i j
This equation is clearly nonsymmetric, due to the fact that the weight and trial functions are not identical. Nevertheless, if the material properties are constant as a function of frequency, each element matrix corresponding to the terms above is constant as well. Gradients of density inside the element volume have been ignored in the formulation. The element shape functions are dened as follows:
i (g; h; r) fb B ;
Elements
where
f
in two spatial dimensions and
r0 (g; h) r
f
r0 (g; h) r
in three. This function serves to specify the minimum rate of decay of the acoustic eld inside the innite element domain. The subindex ranges over the n nodes of the innite element on the terminating surface, while the subindex ranges over the number of functions used in the innite direction. The function index i is equal to the subindex for the rst n functions, i n + for the second n functions, and so on. The functions b (g; h) are conventional two-dimensional shape functions (in three dimensions) or, with h 0, one-dimensional shape functions for axisymmetric or two-dimensional elements. The role of these functions is to specify the variation of the acoustic eld in directions tangent to the terminating surface. The variation of the acoustic eld in the innite direction is given by the functions B , which are members of a set of ten ninth-order polynomials in 1 0 2r0 (g;h) . The members of this set are r constructed to correspond to the Legendre modes of a sphere. That is, if innite elements are placed on a sphere and if tangential renement is adequate, an i-th order acoustic innite element will absorb waves associated with the (i 0 1)th Legendre mode. The rst member of this set corresponds to the value of the acoustic pressures on the terminating surface; the other functions are generalized degrees of freedom. Specically,
B 1 1;
3.3.23
and where
^ B (1 + )B ; for > 1;
01
^ ^ B B (1 + )2 f 2 F d = 0; for 6= :
This last condition is enforced to improve conditioning of the element matrices at higher order and results in different functions in two and three spatial dimensions. All of the B are equal to zero at the terminating surface, except for B1 . The coordinate map is described in part by the element shape functions, in the usual isoparametric manner, and in part by a singular function. Together they map the true, semi-innite domain onto the parent element square or cube. To specify the map for a given element, we rst dene distances r between each innite element node on the terminating surface and the elements reference node, located at x0:
r j x 0 x0 j:
Intermediate points in the innite direction are dened as offset replicas of the nodes on the terminating surface:
y x + n r ;
where n is the node ray at the node. The node ray at a node is the unit vector in the direction of the line between the reference point and the node. We use the interpolated reference distance on the terminating surface,
r0(g; h)
rb (g; h);
10
2 0(
r g; h) ; r
where r is the distance between an arbitrary point in the innite element volume and the reference point, x0. With these denitions, the geometric map can be specied as
x(g; h; )
The innite elements are not isoparametric, since the map uses a lower-order function of the parent coordinates than the interpolation scheme does. However, this singular mapping is convenient and invertible. The phase factor (r 0 r0) appears in the element formulation. This factor models the oscillatory nature of the solution inside the innite element volume but must also satisfy some additional properties. First, it must be zero at the face of the innite element that connects to the nite element
3.3.24
mesh. Second, it must be C 0 continuous across innite element lateral boundaries. Third, it should be such that the mass-like term in the innite element equation,
Z
Vf
F Kf
10
@ (r 0 r 0 ) @ (r 0 r 0 ) 1 @ x i j dV; @x
be positive semi-denite. This last criterion is not essential for execution of a steady-state analysis, but it is essential for the well-posedness of a transient analysis. Dening
q minE (n 1 nE ) ;
where the subscript
rq b(g; h)
01
satises these requirements. The inclusion of the factor q is the only departure from the denition used in Astley (1994). In a transient analysis the element matrix equation is transformed back to the time domain, using constant material properties taken from the value at zero frequency. The resulting second-order ordinary differential equation for the pressure degrees of freedom is added to the overall system in the model and integrated in the usual manner.
Steady-state analysis
Elements
In steady-state analysis the formulation for the acoustic innite elements is consistent with that used for acoustic nite elements (see Coupled acoustic-structural medium analysis, Section 2.9.1, and Acoustic, shock, and coupled acoustic-structural analysis, Section 6.9.1 of the ABAQUS Analysis Users Manual). First, the complex density
f + ~
~ ~
2 p dV + Z 1 @p 1 @ p dV + Z p 1 @ p 1 n0 dS = 0 0 p K ~ @x @x ~ @x ~ f Vf Vf S
after integration by parts. The expression for the basis and weight functions is formally identical to the preceding transient development, but with the complex wave number
~ k
~ Kf
3.3.25
used in the denition of the spatially oscillatory term in the weight and basis functions:
e 6 k (r 0 r 0 ) :
~
After substitution and some simplication, the acoustic innite element expression becomes
Now
so the mass matrix Mfij is the same as for the transient case. It is real-valued. The remaining terms are complex; and, as in the case for acoustic nite elements, they are manipulated into real and imaginary parts. We use f
=
1= ~ 2 +
2 =
2 + i 2 +
2 =
2 R + iI f f to obtain
i
@j @x
(3.3.23)
~ k ~
where
= p
"
#"
2 f
Kf
+
2 =
2
#(1=4)
"
I tan = R :
i
Kf
[R2 + iI2] ;
Using also
1 @p _ @p =
@x ; @x we obtain real-valued element matrices as follows: Z I @ [F ] @ i j Cfij =
@ x 1 @ x dV; Z Vf R @F @ (r 0 r ) + p 2 0 @ x i 1 @ x 0 j + ZVf K f @ R2 @Fx i 1 @ xj dV; Kfij = @ Z Vf
I @F @ (r 0 r ) 0 p 2 0 @ x i 1 @ x 0 j + Kf Vf
3.3.26
@j @x
1 @ (r@0 r0 ) F i x 1
dV;
@j @x
@ (r 0 r 0 ) F i dV: @x
An innite element volume may have an impedance boundary condition dened on a face extending to innity. Such a denition can be used to model, for example, an innite half-space of water above a lossy seabed or under a layer of ice. The derivation of the contribution of the impedance condition to the innite element matrix is straightforward and follows the derivation for nite elements (see Coupled acoustic-structural medium analysis, Section 2.9.1). The contributing term in the transient case is Z
1 p + 1 p dS: p p+ + _ k f c1 f k1 c1 1 Slateral Applying the innite element basis and weight functions, the lateral face contributes to the mass, damping, and stiffness matrices of the acoustic innite element:
Z
Slateral
F i j
f c1
dS pj +
Slateral
F ij
Z
f k1
Slateral
F ij
+ c1 dS p_j 1 1 dS p :
k1
Elements
Lateral faces of acoustic and solid innite elements (that is, those that extend to innity) can be coupled to each other. For example, two innite half-spaces may interact at their interface. As in the case of lateral-face impedance terms, the derivation of the contribution of the coupling condition to the innite element matrix follows the derivation for nite elements to the greatest degree possible. We restrict the consideration to solid medium innite elements that share an edge with the acoustic medium innite elements and where the acoustic innite element is the slave in a tied contact condition. The contributing terms in the transient case are
0
on the acoustic innite element and
Slateral
Z
p
1 n0 1 um dS
Slateral
n0 1 um
p dS
on the solid innite element. In tied contact between uid and solid medium elements, a tributary area of the slave node is dened, and the conservation of momentum and continuity of displacement is enforced between the
3.3.27
slave node and the master surface. On a lateral face of an innite element, the area is unbounded, so an area measure based on the acoustic innite element formulation,
Alateral
Z
Slateral
F f 2 dS ;
For consistency the boundary traction on the solid medium innite element due to the acoustic pressure is dened as Z 0 0 1 0 1 + n 1 um p dS 7! Alateral n0 1 um : Slateral In steady-state and eigenvalue analysis the corresponding frequency-domain representation is used.
References
is dened. This area measure is used to compute a volumetric acceleration load on the acoustic innite element: Z 0 1 0 1 0 p n0 1 um dS 7! Alateral n0 1 um : Slateral The structural displacement, um , is evaluated at the shared edge to compute this coupling condition.
Acoustic, shock, and coupled acoustic-structural analysis, Section 6.9.1 of the ABAQUS Analysis Users Manual Innite elements, Section 14.2.1 of the ABAQUS Analysis Users Manual
3.3.28
MEMBRANE ELEMENTS
3.4.1
MEMBRANE ELEMENTS
Membrane elements are sheets in space that can carry membrane force but do not have any bending or transverse shear stiffness, so the only nonzero stress components in the membrane are those components parallel to the middle surface of the membrane: the membrane is in a state of plane stress. At any time we use a local orthonormal basis system ei , where e1 and e2 are in the surface of the membrane and e3 is normal to the membrane. The basis system is dened by the standard convention used in ABAQUS for a basis on a surface in space. In this section Greek indices take the range 1, 2, and Latin indices take the range 1, 2, 3. Greek indices are used to refer to components in the rst two directions of the local orthonormal basis (in the surface of the membrane).
Equilibrium
The virtual work contribution from the internal forces in a membrane element is
W I
Z
V
" : " dV;
(3.4.11)
" where is the Cauchy stress, " = sym( ) is the virtual rate of deformation ( = @ =@ , where is the virtual velocity eld), and V is the current volume of the membrane. We assume that only the membrane stress components in the surface of the membrane are nonzero: 3i = 0. Then Equation 3.4.11 simplies to
u x
Elements
W I =
=
ZV
V
" dV L dV
(since is symmetric);
where
L =
u e @x e = e @u @ @x
1
and dV
Jacobian
= t dA,
where t is the current thickness of the element and A is its current area.
" t " :
= 0,
3.4.11
MEMBRANE ELEMENTS
We also assume that there is no transverse shear strain of the element: "3i "i d"i = " d" . Thus, the second term in the integrand is
@u @x @
@u @x
@u 0 1 e
1 @x 2
12
@u e 1 @x
@u e 1 @x
e 1 @x
= @u
0,
and, hence,
u 1 @du +
@x1 @
22
0 211e1 1 @xu e1 1 @xu 0 222e2 1 @ u e2 1 @du @x2 @x2 1 1 1 @ u @du @ u @du @ u @du @ u @du 0 2 (11 + 22) e2 1 @x e2 1 @x + e1 1 @x e1 1 @x + e1 1 @x e2 1 @x + e2 1 @x e1 1 @x 1 2 2 2 1 1 2 @ u @du1 @ u @du @ u @du @ u @du 0 12 e1 1 @x e1 1 @x + e2 1 @x e2 1 @x + e1 1 @x e1 1 @x + e2 1 @x e2 1 @x 1 1 1 2 2 1 2 1 @ u @du @ u @du @ u @du @ u @du + e1 1 e2 1 @x + e2 1 @x e1 1 @x + e1 1 @x e2 1 @x + e2 1 @x e1 1 @x @x
2 2 1 1 1 1 2 2
@x2 @d
u 1 @du +
@x2
@ u
@x1
1 @x
@d
u + @ u 1 @du
2
@x2
@x1
Thickness change
In geometrically nonlinear analyses the cross-section thickness changes as a function of the membrane strain with a user-dened effective section Poissons ratio, . In plane stress 33 = 0; linear elasticity gives
33 = 0 1 0 (11 + 22): l1
t t0
ln
01 0
ln
+ ln
0 l1
0 l2
l2
01 0
ln
A0
where A is the area on the membranes reference surface. This nonlinear analogy with linear elasticity leads to the thickness change relationship:
t t0
=
0 10 A
A0
=0
For
= 0 :5
Total deformation
The deformation gradient is F = @ x=@ X. Since we take e3 normal to the current membrane surface and assume no transverse shear of the membrane,
F 3 = 3 @x e 1 @X
=0
and
F 3 =
@x e 1 @X
= 0:
3.4.12
MEMBRANE ELEMENTS
By the thickness change assumption above, the direct out-of-plane component of the deformation gradient is
F33
:
To calculate the deformation gradient at the end of the increment, rst we calculate the two tangent vectors at the end of the increment dened by the derivative of the position with respect to the reference coordinates: n+1 n+1
@x
where @ xn+1 =@ is obtained by interpolation with the shape function derivatives from the nodal coordinates and the change of coordinate transformation @ =@X is based on the reference geometry. The deformation gradient components are dened
F
@X
@X
x = en+1 1 @@X
n+1
To choose the element basis directions en+1, we do the following. Find any pair of in-plane such that orthonormal vectors en+1 (by the standard ABAQUS projection). Then nd the angle the element basis vectors en+1, dened
en+1 = cos(1 )^n+1 + sin(1 )^n+1 e1 e2 1 en+1 = 0 sin(1 )^n+1 + cos(1 )^n+1 ; e1 e2 2
F
Elements
x = en+1 1 @@X = F :
n+1
Using the denitions of en+1 in terms of en+1 in the above equation, the rotation angle to be " #
is found
where
The deformation gradient then follows immediately. F F and B33 F33 2 . For inelastic For elastomers we work directly in terms of B material models we need measures of incremental strain and average material rotation, which we F 1 Fn , where Fn is the deformation gradient at the start of the compute from F dened by F current increment (at increment n):
=( )
=1
@x (F n) = (en) 1 @X :
3.4.13
MEMBRANE ELEMENTS
1F01 by
n (1F 01) = (en) @x @x
1
and, hence, dene F by inversion. The incremental strain and rotation are then dened from the polar decomposition V 1 R, where R is a rotation matrix and V is a pure stretch:
1 1
1 1
1V =
X1
3
1F =
I =1
I
aI aI
1 F 1 FT = 1 V 1 V = 2 2
2
X(1
3
I =1
I 2
) aI aI :
Since we assume no transverse shear in the membrane, the normal direction (along e3) is always a principal direction, so the eigenproblem is the 2 problem
1F
F
=
The logarithmic strain increment is then
X(1 )
I =1
2 aI aI :
1" =
X ln(1
2
I =1
I a I aI ;
and the average material rotation increment is dened from the polar decomposition of the increment:
1R =
X
2
1 aI aI 1 F :
I =1 1I
Due to the choice of the element basis directions, we can assume that
1R
Reference
:
3.4.14
TRUSS ELEMENTS
3.4.2
TRUSS ELEMENTS
Truss elements are one-dimensional bars or rods that are assumed to deform by axial stretching only. They are pin jointed at their nodes, and so only translational displacements and the initial position vector at each node are used in the discretization. When the strains are large, the formulation is simplied by assuming that the trusses are made of incompressible material. There are two truss elements in ABAQUS: a 2-node linear interpolation truss and a 3-node quadratic interpolation truss. The quadratic interpolation version is in the library mainly for compatibility with the quadratic interpolation elements of other types, such as shell element S8R5. The same interpolation functions are used for both the Cartesian displacement components and for the Cartesian components of the initial position vector, so these elements are the simplest form of isoparametric elements. The elements are one-dimensional: a single material (isoparametric) coordinate, g, is dened along the element, with 01 g 1 in the element. In a 2-node element node 1 is at g = 01 and node 2 is at g = +1. In the 3-node version node 1 is at g = 01, node 2 is at g = 0, and node 3 is at g = +1.
Interpolation
u (g ) =
and for the 3-node element,
1 2
(1
0 g)u1 + 1 (1 + g)u2 ; 2
1
Elements
where u1 , u2 , and u3 are the values of a variable at the nodes and u(g) is the interpolated value of this variable.
Strain measure
These are one-dimensional elements, and the only strain considered is that along the axis of the element. The stretch ratio along the axis is
=
dl ; dL
where l measures length along the truss axis in the current conguration:
dl =
dx dx 1 dg; dg dg
3.4.21
TRUSS ELEMENTS
= ln
dl
dL
dg dl
t1
d x dg
where
t=
dg dx dl dg
2
dg dl
d x dg
[I 0 2t t] 1
d dx dg
Integration Stiffness
The linear truss is a constant strain element and so is integrated exactly. The quadratic truss is integrated numerically using two Gauss points.
Mass and consistent loads
A linear truss has two Gauss points. A quadratic truss has three Gauss points.
Virtual work contribution
a" dl;
where a is the current cross-sectional area of the truss, is the true (Cauchy) stress along the truss, " is the logarithmic strain, and l is the length of the element. Since we assume the truss is incompressible, a dl = A dL, where A is the original area and L the original length of the truss. So,
W =
A" dL:
3.4.22
TRUSS ELEMENTS
This is the form in which the internal virtual work contribution is used for truss elements.
Mixed (hybrid) forms
Hybrid truss elements are also available in ABAQUS/Standard. In those elements the axial force at the integration points is taken as an additional variable, with the compatibility condition introduced to dene these variables. The formulation is identical to that used for the hybrid beam elements (Hybrid beam elements, Section 3.5.4), without the bending terms.
Reference
Elements
3.4.23
AXISYMMETRIC MEMBRANES
3.4.3
AXISYMMETRIC MEMBRANES
ABAQUS includes two libraries of axisymmetric membrane elements, MAX and MGAX, whose geometry is axisymmetric (bodies of revolution) and that can be subjected to axially symmetric loading conditions. In addition, MGAX elements support torsion loading and general material anisotropy. Therefore, MGAX elements will be referred to as generalized axisymmetric membrane elements, and MAX elements will be referred to as regular axisymmetric membrane elements. In both cases the body of revolution is generated by revolving a line that represents the membrane surface (a membrane has negligible thickness) about an axis (the symmetry axis) and is described readily in cylindrical coordinates r, z , and . The radial and axial coordinates of a point on this cross-section are denoted by r and z , respectively. At = 0, the radial and axial coordinates coincide with the global Cartesian X - and Y -coordinates. If the loading consists of radial and axial components that are independent of and the material is either isotropic or orthotropic, with being a principal material direction, the displacement at any point will have only radial (ur ) and axial (uz ) components. The only nonzero stress components are ss and , where s denotes a length measuring coordinate along the line representing the membrane surface on any rz plane. The deformation of any rz plane (or, more precisely, any rz line) completely denes the state of stress and strain in the body. Consequently, the geometric model is described by discretizing the reference cross-section at = 0. If one allows for a circumferential component of loading (which is independent of ) and general material anisotropy, displacements and stress elds become three-dimensional. However, the problem remains axisymmetric in the sense that the solution does not vary as a function of , and the deformation of the reference rz cross-section characterizes the deformation in the entire body. The motion at any point will havein addition to the aforementioned radial and axial displacementsa twist (in radians) about the z -axis, which is independent of . There will also be a nonzero in-plane shear stress, s , as a result of the deformation. This section describes the formulation of the generalized axisymmetric membrane elements. The formulation of the regular axisymmetric membrane elements is a subset of this formulation.
Kinematic description
Elements
The coordinate system used with both families of elements is the cylindrical system (r, z , ), where r measures the distance of a point from the axis of the cylindrical system, z measures its position along this axis, and measures the angle between the plane containing the point and the axis of the coordinate system and some xed reference plane that contains the coordinate system axis. The order in which the coordinates and displacements are taken in these elements is based on the convention that z is the second coordinate. This order is not the same as that used in three-dimensional elements in ABAQUS, in which z is the third coordinate; nor is it the order (r, , z ) that is usually taken in cylindrical systems. Let eR , eZ , and e2 be unit vectors in the radial, axial, and circumferential directions at a point of the point can be in the undeformed state, as shown in Figure 3.4.31. The reference position
3.4.31
AXISYMMETRIC MEMBRANES
er
x e eR X
eZ, ez
Figure 3.4.31 Cylindrical coordinate system and denition of position vectors. represented in terms of the original radius, R, and the axial position, Z :
X = ReR + Z eZ ;
Likewise, let r , z , and be unit vectors in the radial, axial, and circumferential directions at a point in the deformed state. As shown in Figure 3.4.31, the radial and circumferential base vectors depend on the coordinate: r = r ( ) and = (). The current position, , of the point can be represented in terms of the current radius, r, and the current axial position, z , as
e e
x = r er + z ez :
The general axisymmetric motion at a point on the membrane surface can be described by
As this description implies, the degrees of freedom ur , uz , and are independent of 2. Moreover, the reference cross-section of interest is at 2 = 0; however, for the benet of the mathematical analysis to follow, it is important that 2 be considered an independent variable in the above expression for .
(3.4.31)
3.4.32
AXISYMMETRIC MEMBRANES
= N N (g )u r N ; u z = N N (g )u z N ; N (g ) N ; =N
ur
where g is the isoparametric coordinate in the reference rz cross-section at 2 = 0; and ur N , uz N , N are the nodal degrees of freedom. The interpolation functions are identical to those used for truss elements (see Truss elements, Section 3.4.2). All elements use reduced integration.
Deformation gradient
For a material point the deformation gradient F is dened as the gradient of the current position, x, with respect to the original position, X, and is given by
@x F = @X :
(3.4.32)
Elements
The components of the deformation gradient require that two sets of orthonormal basis vectors be dened. In the undeformed conguration the basis vectors are dened by
@X Ei = @S ;
where the Si denote length measuring coordinates in the reference conguration along the element length and the hoop direction, respectively. Thus,
@X E1 = @ X ; E2 = R@ 2 : @S
In the current conguration ABAQUS formulates the equations in terms of a xed spatial basis with respect to the axisymmetric twist degree of freedom. The basis vectors convect with the material. However, because of the axisymmetry of the model in the deformed conguration, these vectors can be dened at 2 = 0 as
@x ei = @s ;
where the si denote length measuring coordinates in the current conguration along the element length and the hoop direction, respectively. Thus, the basis vectors in the reference and current congurations can be written as
@X @x E1 = @ X ; E2 = R@ 2 ; e1 = @ x ; e2 = r@ ; @S @s
(3.4.33)
3.4.33
AXISYMMETRIC MEMBRANES
where S and s are length measuring coordinates along the element length in the reference and current congurations, respectively. The components of the deformation gradient in the two sets of basis vectors may be computed as
F
= e F E :
1 1
Using the denitions of the basis vectors in Equation 3.4.33, the components of the deformation gradient tensor are
F11 F21 @ = r @S ; F12 = 0 ; r F22 = : R @r @z = @r @S + @z @S ; @s @s
Virtual work
As discussed in Equilibrium and virtual work, Section 1.5.1, the formulation of equilibrium (virtual work) requires the virtual velocity gradient, which takes the form
L = F F 01 ;
1
where F represents the rst variation of the deformation gradient tensor. Alternatively, the virtual velocity gradient can be written as
x L = @x : @
Recall that ABAQUS formulates the nite element equations in terms of a xed spatial basis with respect to the axisymmetric twist degree of freedom. Therefore, the desired result for F does not simply follow from the linearization of Equation 3.4.32. Namely, the contributions from the variations
er = e and
e = 0 er
arising from the spin of the coordinate system must be canceled out. To this end, according to where
F can be modied
~ F = R1F;
c
0 0 = 4 0 0 0 5 ; 0 0 0
3.4.34
AXISYMMETRIC MEMBRANES
with respect to the basis er , ez , and e at = 0. With this modication, the corotational virtual deformation gradient is given by
~ =
c R F + R F = F + F ;
1 1 1
c ~ F F 01 = + L :
1
@r @r
@z @z
r ~ L32amp; = 0 :
~ L22amp; =
r
Elements
~ Moreover, there are additional contributions from d Lij , which are given by ~ d L11 = 0 ; ~ d L21 = r ~ d L12 = 0 ; ~ d L22 = 0 :
@d @s
+ dr
@ @s
= 0.
Axisymmetric membrane element library, Section 15.1.4 of the ABAQUS Analysis Users Manual
3.4.35
BEAM ELEMENTS
3.5.1
The element library in ABAQUS contains several types of beam elements. A beam in this context is an element in which assumptions are made so that the problem is reduced to one dimension mathematically: the primary solution variables are functions of position along the beam axis only. For such assumptions to be reasonable, it is intuitively clear that a beam must be a continuum in which we can dene an axis such that the shortest distance from the axis to any point in the continuum is small compared to typical lengths along the axis. This idea is made more precise in the detailed derivations in Beam element formulation, Section 3.5.2. There are several levels of complexity in the assumptions upon which the reduction to a one-dimensional problem can be made, and different beam elements in ABAQUS use different assumptions. The simplest approach to beam theory is the classical Euler-Bernoulli assumption, that plane crosssections initially normal to the beams axis remain plane, normal to the beam axis, and undistorted. The beam elements in ABAQUS that use cubic interpolation (element types B23, B33, etc.) all use this assumption, implemented in the context of arbitrarily large rotations but small strains. The Euler-Bernoulli beam elements are described in Euler-Bernoulli beam elements, Section 3.5.3. This approximation can also be used to formulate beams for large axial strains as well as large rotations. The beam elements in ABAQUS that use linear and quadratic interpolation (B21, B22, B31, B32, etc.) are based on such a formulation, with the addition that these elements also allow transverse shear strain; that is, the crosssection may not necessarily remain normal to the beam axis. This extension leads to Timoshenko beam theory (Timoshenko, 1956) and is generally considered useful for thicker beams, whose shear exibility may be important. (These elements in ABAQUS are formulated so that they are efcient for thin beams where Euler-Bernoulli theory is accurateas well as for thick beams: because of this they are the most effective beam elements in ABAQUS.) The large-strain formulation in these elements allows axial strains of arbitrary magnitude, but quadratic terms in the nominal torsional strain are neglected compared to unity, and the axial strain is assumed to be small in the calculation of the torsional shear strain: thus, while the axial strain may be arbitrarily large, only moderately large torsional strain is modeled correctly, and then only when the axial strain is not large. We assume that, throughout the motion, the radius of curvature of the beam is large compared to distances in the cross-section: the beam cannot fold into a tight hinge. A further assumption is that the strain in the beams cross-section is the same in any direction in the cross-section and throughout the section. Some additional assumptions are made in the derivation of these elements: these are introduced in the detailed derivation in Beam element formulation, Section 3.5.2. For certain important designs the beam is constructed from thin segments made up into an open section. The response of such open sections is strongly effected by warping, when material particles move out of the plane of the section along lines parallel to the beam axis so as to minimize the shearing between lines along the wall of the section and along the beam axis. The beam element formulation (Beam element formulation, Section 3.5.2) includes provision for such effects. Beam elements that allow for warping of open sections (B31OS, B32OS etc.) are also derived. The particular approach used for modeling open section warping in ABAQUS is based on the assumption that the warping amplitude is never large anywhere along the beam axis because the warping will be constrained at some points along the beam perhaps because one or both ends of the beam are built into a stiff structure or because some form of transverse stiffeners are added.
Elements
3.5.11
BEAM ELEMENTS
The regular beam elements can be used for slender and moderately thick beams. For extremely slender beams, for which the length to thickness ratio is O(103 ) or more and geometrically nonlinear analysis is required (such as pipelines), convergence may become very poor. For such cases use of the hybrid elements, in which the axial (and transverse) forces are treated as independent degrees of freedom, can be benecial. The hybrid beam formulation is described in Hybrid beam elements, Section 3.5.4. Distributed pressure loads applied to beams (for example, due to wind or current) will rotate with the beam, leading to follower force effects. The derivation of the load stiffness that accounts for this effect is presented in Load stiffness for beam elements, Section 6.5.2. In some piping applications thin-walled, circular, relatively straight pipes are subjected to relatively large magnitudes of internal pressure. This has the effect of creating high levels of hoop stress around the wall of the pipe section so that, if the section yields plastically, the axial yield stress will be different in tension and compression because of the interaction with this hoop stress. The PIPE elements allow for this effect by providing uniform radial expansion of the cross-section caused by internal pressure. In other piping cases thin-walled straight pipes might be subjected to large amounts of bending so that the section collapses (Brazier collapse); or a section of pipe may already be curved in its initial congurationit might be an elbow. In such cases the ovalization and, possibly, warping, of the crosssection may be important: these effects can reduce the bending stiffness of the member by a factor of ve or more in common piping designs. For materially linear analysis these effects can be incorporated by making suitable adjustments to the sections bending stiffness (by multiplying the bending stiffness calculated from beam theory by suitable exibility factors); but when nonlinear material response is a part of the problem it is necessary to model this ovalization and warping explicitly. Elbow elements are provided for that purpose; they are described in Elbow elements, Section 3.9.1. Elbow elements look like beam elements to the user, but they incorporate displacement variables that allow ovalization and warping and so are much more complex in their formulation. In particular, ovalization of the section implies a strong gradient of strain with respect to position through the wall of the pipe: this requires numerical integration through the pipe wall, on top of that used around the pipe section, to capture the material response. This makes the elbow elements computationally more expensive than beams. Since consideration of planar deformation only provides considerable simplication in formulating beam elements, for each beam element type in ABAQUS a corresponding beam element is provided that only moves in the (X; Y ) plane. However, the open section beams are provided only in three dimensions for reasons that are obvious.
Reference
Beam modeling: overview, Section 15.3.1 of the ABAQUS Analysis Users Manual
3.5.12
3.5.2
At a given stage in the deformation history of the beam, the position of a material point in the cross-section is given by the expression
^ x(S; S ) = x(S ) + f (S )S n(S ) + w(S ) (S )t(S ):
In this expression x(S ) is the position of a point on the centerline, n(S ) are unit orthogonal direction vectors in the plane of the beam section, t(S ) is the unit vector orthogonal to n1 and n2, (S ) is the warping function of the section, w(S ) is the warping amplitude, and f (S ) is a cross-sectional scaling factor depending on the stretch of the beam. These quantities are functions of the beam axis coordinate S and the cross-sectional coordinates S , which are assumed to be distances measured in the original (reference) conguration of the beam. The warping function is chosen such that the value at the origin of the section vanishes: (0) = 0. It is assumed that at the integration points along the beam, the beam section directions are approximately orthogonal to the beam axis tangent s given by
s = 01
where is the axial stretch given by
dx ; dS
Elements
d =
dS
The normality condition is enforced numerically by penalizing the transverse shear strains
s 1 n :
This condition is assumed to be satised exactly in the original conguration. In what follows, is the alternator
2 1
The curvature of the beam is dened by
01 = 1 ; 1 = 2 = 0: 2 1 2
b
=
t1
dn ; dS
b = n2 1
The bicurvature of the beam is dened by
dn1 dS
n 0n1 1 ddS2 :
=
dw : dS
3.5.21
The bicurvature denes the axial strain variation in the section due to the twist of the beam. The expression for the curvature and the twist can be combined to yield
dn dS
= (0 b t + b n ):
Before we derive strain measures from these expressions, we will consider in detail how the above quantities and their rst and second variations are obtained for a typical beam nite element.
Beam element in undeformed conguration
In the undeformed conguration, we use capital letters for all quantities. We assume that the undeformed state has no warping, so the position of a material point is given by
^ X(S; S) = X(S) + SN(S)
N T 1 ddS ; dN 1 dN 2 B = N2 1 = 0N1 1 ;
dS dS
In the element the position of a point on the axis is interpolated from nodal positions standard interpolation functions as
T is the unit vector orthogonal to N1 and N2; i.e., T = N 1 2 N2 : We assume that the section normal T coincides with the beam tangent S.
where
XN with
X = g N ( )X N ;
=
where is a parametric coordinate, typically varying between axis tangent is readily computed as
S= S
d
dS d
dgN d
dg M
d
M :
The section normals are interpolated from user-dened nodal normals N . However, we cannot use a simple interpolation since this would not create integration point normals orthogonal to the beam tangent . Hence, we use a two-step approach. First, we create approximate normals by the interpolation
(no summation)
3.5.22
dS d d dS 1 dN dg 1 : B = N 1 = N 1 N N N 2 dS 2 d d
dN dS
(0B T + B N );
Elements
dN B = T 1 ; dS dN 1 dN 2 = 0N1 1 : B = N2 1 dS dS dN The procedure followed above to derive N and is not unique but provides values that satisfy the dS proper orthonormality conditions. This procedure is followed only for the undeformed conguration. dn are obtained individually by forward integration of the For subsequent congurations n and dS
kinematic equations.
Change of position, warping, and normal direction
We assume that the position of the beam axis and the orientation of the normals can undergo (independent) changes. The change in position of the axis is described by the velocity vector = ( ), which can be obtained from nodal velocities N with the standard interpolation functions as
v ( ) = g N ( )v N :
The change in orientation of the normals is described by the spin vector ! = ! ( ), which is obtained from nodal spin vectors ! N with the same interpolation functions gN ( ), as
v v
n = ! ( ) 2 n ; _
! ( ) = gN ( )! N :
3.5.23
Rigid body motion is included since the original position is obtained by the same interpolation as . The rate of change of warping is also dened in terms of the rate of change of nodal warping wN with the standard interpolation functions as
v( )
X( )
w
_ ( ) = g N ( )w N : _
Z t+1t
t
The velocity and spin describe the rate of change of the position and orientation. Finite changes in position are obtained by integration of the velocities over a nite time increment as
1 x ( ) = 1 w ( ) = 2q q = ! ; _
y
Z t+1t
t
v()dt =
w dt
gN
() ()
vN dt = gN ()1XN :
wN dt
Z t+1t
t
_( ) =
gN
Z t+1t
t
= gN ()1wN :
q=
where is the total or Euler rotation. The relation between spin and quaternion can be integrated exactly if it is assumed that the spin is constant over the time increment t. We then dene
= n
cos 2 ; sin 2 n
= cos 2 ; 0 sin 2 n
;
This allows us to update the normal directions at the nodes and at stress points with
n = 1qN1qy: In this equation we use the notation that N is n at the beginning of the current increment; i.e., N def n(t): =
For the entire motion the new normal directions can be formally expressed as
n = q N q y ;
3.5.24
q=
Here i is an increment and is updated the same way.
n Y i=1
1q i :
( )
1q i
( )
The curvature and the twist involve the derivative of the normal vector with respect to update rule for ,
S . From the
dn dS
Hence, the scalar parts of the rst two terms cancel each other, and the vector parts are the same. Therefore, Because q is a rotation quaternion, its inverse is equal to its conjugate ( q01 we can write
Elements
dn dS
where
p = V(p) denotes the vector part of a quaternion. For the rst term we use the relation 1 sin 1 d1 ; 1 cos 1 n d1 + sin 1 dn : d1 q = 0 2 2 dS 2 2 dS dS 2 dS
This leads to
dn dn 1 1 2 ddSq 1qy = n ddS + sin 1 dS + (1 0 cos 1)n 2 dS = 1%; which is a vector. From the denitions of 1 and n it follows that 1 1 d1 = 10 1 1 ddS = n 1 ddS ; dS dn 0 d1 0 10 1 n 1 d1 = 10 d1 0 n n 1 d1 : = 1 dS dS dS dS dS
1 1 2 1
3.5.25
d in terms of the curvature and twist at the beginning of the For the second term we express dS increment, which yields
1 1 cos 1 1 1% = n n ddS + sin 1 ddS 0 sin 1 n n 1 ddS + (1 0 1 1) n 2 ddS 1 1 sin 1 d1 + (1 0 cos 1) n 2 d1 + (1 0 sin 1 )n n 1 d1 : = 1 dS 1 dS 1 dS d1 We see that 1% ) if 1 ! 0. dS
d % 2 0 B B : dS The current curvature and twist are, therefore, d b 1 % 1 B % 1 2 0 B dS and d 1 %1 12 2 b 2 B %1 B: 21 1 dS Hence, the current curvature and twist are updated by a summation over all increments as given by n (n) % (i ) 1 (i ) B ; b i=1 n (n ) %(i) 1 (i) B: b i=1
= t n = 1 (n =n
=1 n +
n = 1 (n n ) +
= 1 t+
X1 X1 =
=
;
n +
t +
First variations
The rst variations of the geometric quantities are readily obtained. Recall that
dx = s 1 n (with s = 0 dS ); dn b = t 1 ; dS dn b =n 1 ; dS dw : = dS
1 2 1
dx = dS
3.5.26
It follows that
dw ; dS where in the expression for
we have assumed that
' 0 and s curvature and twist, we note that t = 2 t. Hence, it follows that
=
01 n 1 d x 0 n 1 ; dS
=s
d x ; dS
' t.
dn dS
d 1 (n 2 t) = n 1 d ; dS dS dn 1 d n1 dn 1 dn1 d b = n2 1 + n2 1 = ( 2 n2 ) 1 + n2 1 ( 2 + 2 n1 ) dS dS dS dS dS d = 1 (n1 2 n2) = t 1 d : dS dS These terms will be used in the virtual work equation that will be discussed later. In using the above expressions the rotational quantities are obtained by interpolation from nodal variational quantities that are assumed to be valid for the velocity elds as x( ) = gN ( ) xN ; w( ) = gN ( )wN ; ( ) = gN ( )N :
=
dn b = t 1 dS
t1
d n dS
=
(
2 t) 1
dn dS
t 1 ( 2
d 2 n ) dS
Elements
Newtons algorithm involves linearization of the incremental equations. The equations must be linearized around the current (latest) state. At the integration points these equations simply take the form ddx ; d = s 1 dS ddx 01 0 n 1 d; d n 1 dS dd db = 1 n ; dS dd db = 1 t; dS ddw : d = dS
3.5.27
The corrections x and w are readily obtained from the nodal corrections with the interpolation functions g N as
()
The corrections are dened with respect to the end of the increment: we call such corrections compound corrections. However, it has been assumed that the total incremental rotation would be interpolated with gN as
()
d1 are corrections in 1 in an additive sense such that 1new = 1old + d1: To relate the additive correction d1 to the compound correction d, we use the formula obtained for 1 to nd cos d = 1qydq = sin 1 d1 + (1 0 1 1) n 2 d1 + (1 0 sin 1 )n n 1 d1: 1 1
where Similarly, at the nodes we nd
d1() = gN ()d1N ;
N cos 1 N dN = sin1N d1N + (1 0 1N ) nN 2 d1N 1 N + (1 0 sin 1N )nN nN 1 d1N (no summation): 1
Now we assume that the difference in incremental rotation along a beam element is small; i.e.,
j1( ) 0 1N j 1;
which implies that either
j1( )j 1 and j1N j 1: In the rst case we can use the approximations n( ) nN n and 1( ) 1N 1 which,
d1, gives sin 1 d1N + (1 0 cos 1) n 2 d1N + (1 0 sin 1 )n n 1 d1N ; d gN () 1 1 1 sin 1 d1N + (1 0 cos 1) n 2 d1N + (1 0 sin 1 )n n 1 d1N : dN 1 1 1
3.5.28
N dN N = cos d ; dN sin d ; (no summation) 2 2 ! N N N N = 01qN ; 1qN 1 = cos 1 ; 1 sin 1 ; (no summation) 1q 2 1 N 2 N 1qnew = dqN 1qN = 0dqN 1qN 0 dqN 1 1qN ; dqN 1qN + dqN 1qN + dqN 2 1qN 1 ; (no summation) N N 1N = 2 jqnewj tan01(jqN j; qnew) (no summation): new new qN
This approximate relationship is used only in the creation of the Jacobian, so the approximation will at worst result in reduced speed of convergence. Once a nodal update vector N is obtained, an exact update procedure is followed. This is achieved by a transformation into quaternions, use of exact quaternion update formula, and transformation of the results back into an incremental Euler rotation vector:
dq N
dq N ; dq N 1 =
The incremental rotation vector at the integration point is obtained by interpolation. Subsequently, we can calculate the updated integration point normals and the incremental curvature and twist.
Second variations
new
Elements
For the calculation of the Jacobian, we also need the second variation in the generalized quantities. These follow from the rst variations: d x d x 01 1 I 0 ss 1 d x ; s1 dS dS dS 02 0 01 0 01 0 02 d x 1 n dS 01 d x 1 2 n 0 01 s 1 2 n 01 d x 1 2 n dS dS
d d
d ( = ) = d = d d d d d ( (d + ) d ( )+ ) 2 3 + 1 s 1 2 (d 2 n) + d 2 ( 2 n) 2 0 01( d
+ d
) + 1 1 (nt + t n) 1 d 2 dx 0 01 ( d 1 n + d 1 n ) + 01 ( d x 1 n t 1 d + ddS 1 n t 1 ) dS dx d 0 02 d x 1 (nt + t n) 1 ddS + 01 d x 1 n t 1 d + 1 t n 1 ddSx dS dS 1 + 2 1 (nt + t n) 1 ; .
3.5.29
where we have again used 0. For the second variation of the curvature we nd
db =
ddn d n dd n dn + t 1 + dt 1 +t1 d t 1 dS dS dS dS 1 1 t dn + dn t 1 d 0 t 1 dn ( 1 d) = 2 dS dS dS dd n 2 n + d 2 ddS + (d 2 t) 1 d + ( 2 t) 1 dS 1 dd dS d dn 1 1 1 n + d 1 n + 2 d 1 dS +t1 2 dS 2 dS dd dn 1 1 1 d d 1 n + 1 n + 2 d 1 dS +t1 2 dS 2 dS d n 0 t 1 ( d 1 d n + 1 ddS n + 1 d ddS ) : dS
n 2 n + 2 ddS
Rewriting the second and third terms and combining the others yields
db =
1
1
dn dn dn t + t 1 d 0 2 t 1 ( 1 d)
dS dS dS
1
d 1 (n t + t n ) 1 d + 2 1 (n t + t n ) 1 ddS n n d 0 1 n t 1 ddS + t 1 ddS ( 1 d) 0 1 ddS t 1 d dn d n d 0 d 1 n t 1 dS + t 1 dS (d 1 ) 0 d 1 dS t 1 d d 1 1 (n t 0 t n ) 1 d 0 1 (n t 0 t n ) 1 ddS : = 2 dS +
d 2 dS
The second variation of twist is dn1 ddn 1 d n1 dd n 1 + n2 1 + d n2 1 + n2 1 db =d n2 1 dS dn dn dS dS dn dS 1 1 1 = 1 n2 + n2 1 d 0 n2 1 1 ( 1 d) 2 dS dS dS dd d dn 1 2 n1 + d 2 dS + (d 2 n2) 1 dS + ( 2 n2 ) 1 dS 1 dd d 1 1 n1 + d 1 n1 + 1 d 1 dn1 + n2 1 2 dS 2 dS 2 ddS dd dn 1 1 d 1 1 d 1 n1 + 1 n1 + 2 d 1 dS + n2 1 2 dS 2 dS d dd dn 1 0 n2 1 dS 1 d n1 + 1 dS n1 + 1 d dS :
2 n1 + 2
dn 1 dS
3.5.210
db = 1
+
n2
d 1 (n1 n2 + n2n1) 1 d + 1 1 (n1n2 + n2n1) 1 ddS 2 dn dd 0 1 n1n2 1 dS + n2 1 dS1 ( 1 d) 0 1 dn1 n2 1 d dS dn dn 1 dd 1 0 d 1 n1n2 1 dS + n2 1 dS (d 1 ) 0 d 1 dS n2 1 d 1 d 1 (n1n2 0 n2n1) 1 d 0 1 (n1n2 0 n2n1) 1 ddS : = 2 dS 1 2
d dS
dn1 dS
dn 1 n2 dS
0 2
n2 1
dn 1 dS
( d )
Finally, we note
d = 0:
01
01 d x 1 n t 1 d + 1 n t 1 ddx +
dS dS
Elements
+ db =
1 2 1
1 (n t + t n) 1 ; .
d d 1 (n t 0 t n ) 1 d 0 1 (n t 0 t n ) 1 ddS ; 2 dS d 1 d 1 (n1n2 0 n2n1 ) 1 d 0 1 (n1n2 0 n2n1) 1 ddS ; db = 2 dS d = 0:
Strains
The gradient of the current position of a point in the section with respect to the coordinate S is
df dt dn dw S n + fS + t+w : dS dS dS dS h df 1 , andhencethe second term We will keep only terms up to order . We assume that ` dS ` 1 h2 dt = O ( )we can neglect the can be neglected. We also assume that w = O( ); andsince ` dS ` last term as well. However, the warping function may vary rapidly near the ends where warping is h dw = O ( ) and should be preserved. With those approximations, constrained. Hence, dS ` dx @x ^ n dS + fS ddS + dw t = s + fS (0b t + bn ) + t: @S dS @x ^ @S
= +
dx dS
3.5.211
@ n + w @S t:
X = dX + S dN
@S @^
dS
=
dS
T + S ( 0 B T + B N );
N :
@^ @S @^
= (1 =
0 S B )01 T ;
N 0 (1 0 S B )01 S B T:
d 0 S B )f nN + (1 0 S B )w dS t N B f n T 0 S Bw @ t T : 0S @S
0 S B :
In the expression enclosed within square brackets above, terms of order h2 =`2 can again be neglected. However, it is assumed that the section may have low resistance to torsion and that, hence, the warping and twist may be large. This is particularly true for thin walled open sections. Hence, we obtain:
This
We again neglect all terms of order hn =`n for n 2, except the term involving Bw. The equations then simplify to
@ 0 S fb + Bw @S =
+ fS (b 0 B ); d =w ; dS = f :
=
0 B
;
Elements
Consistent with traditional shell and beam theories, we slightly adapt the term involving the initial curvature B instead of multiplying it with , we multiply it with f . Such a change does not signicantly increase the error in the curvature calculation, since we do not properly account for the initial curvature in the volume integration anyway. Hence, we nd for F11:
F11
@ 0 S f (b 0 B ) + Bw @S
:
We now make a multiplicative decomposition of F into a stretch part such that F = Fd 1 Fs . For Fs we choose
s F11 = ;
and, hence,
F(s +1)1
s F1(+1)
= 0;
and
f ;
Fd is
11
Fd
F(d +1)1
1( +1) d F(+1)(+1)
+
01 ;
Fd
3.5.213
Since
= ln ;
s (+1)1
= 0;
s 1(+1)
= 0;
es+1)(+1) (
ln(f ):
Since the distortional strains are small, we obtain them from Fd with the Green-Lagrange formula:
Ed = 1 Fd T Fd 0 I 2
1
02 2
1
1 2
+ f 2 0 2 S S (b 0 B )2 ;
1 2
E(+1)1
E(+1)(+1)
@ 01 +
+ f S (b 0 B ) @S @ @ @ 0 01 w @S
b
0 B
+ Bw @S
+ f 01 w @S 01 ; 1 02 2 @ @ = f w @S : 2 @S
= 2
f 01 w
Note that these strains are small. Since the various terms have different dependence on the crosssectional coordinates, this leads to the conditions
1;
1; @ f 01 w + f01 S (b 0 B ) 1: @S
The last condition will generally require that both w and b 0 B are small, since @ =@S will not be proportional to S . However, for thin walled open section beams the proportionality is approximately satised and, therefore, we obtain in that case
S f01 (b 0 B ) 1; 01
@ @S
0 f 2 0 1 S (b 0 B ):
3.5.214
Note that within the desired accuracy this equation even holds for other sections: in that case both the right-hand and left-hand side are very small. Substitution in the expression for the Green-Lagrange strain yields the (small) distortional strains:
ed 11
= 0f01S (b 0 B ) + 01
=
d (+1)1 d 1(+1)
@
+ f01 S (b 0 B ) + f 01 w ; @S d =
(+1)1 ; 1 2 02 2 = + f S S (b 0 B ) :
2
1 2 2
f 0 2 S S (b 2 0 B 2 );
0 f01 S (b 0 B ) + 01 + f 2 02 S S (b2 0 B 2 ); 2 f01 S (b 0 B ) + f 01 w @ ;
(+1)1 =
+ @S
1(+1) =
(+1)1 ; 1 e(+1)(+1) = ln(f ) + f 2 02 S S (b 0 B )2 :
= ln 1 2
We assume that there are no stresses in the ( + 1)( + 1) directions. Hence, the strains in these directions do not contribute to virtual work and do not need to be considered any further. It is useful to split the total warping w in two parts: a part due to free warping wf minus a part due to warping prevention wp : w = wf 0 wp . We assume that the warping function is chosen such that the free warping is related to the twist with the relation
Elements
w f = f 2 0 1 (b 0 B )
and
hence
; wp = f 2 01 (b 0 B ) 0 w:
@ (+1)1 = + f01 S + @S
(+1)1 as
b 0 B ) 0 f 01 w p
@ : @S
It is desirable to choose the cross-sectional resultants such that they are completely uncoupled. In addition, we assume that the axial strain variation across the section due to the second-order term in the twist is not signicant. Therefore, we consider only the average axial strain due to the secondorder twist term. Hence, we introduce the average axial strain
e = ln 0 f01 Sc (b 0 B ) + 01
1 2
I
S c S c f 2 0 2 (b 2 0 B 2 );
is the average
where Sc are the coordinates of the centroid, Ip is the polar moment of inertia, and value of the warping function: Z
=
A A
dA:
3.5.215
= + f 01
S c
1Z +A
@ dA A @S
1Z (b 0 B ) 0 f 0 1 w p A
@ dA: A @S
This last expression can be simplied by the introduction of the shear center coordinates Ss , which are related to the warping function by S s
This yields
Z 1 = Sc + A
@ dA; A @S
or
Ss
Z 1 = Sc + A
@ dA: @S
Note that the average value is in fact the value at the shear center if warping prevention is absent (wp = 0). However, for full warping prevention (wp = f 201 (b 0 B )) the average value corresponds to the value obtained at the centroid. Instead of the original warping function (S ) we now introduce a modied warping function ) related to by
(S
(S ) def (S ) 0 : = This function in fact represents the classical denition of a warping function with an area weighted average of zero. The average value can be obtained from the classical warping function with the condition that (0) = 0:
= (S ) 0
(S ) = (0) 0
(0) = 0
(0) = 0
:
The expression for the average axial strain then becomes
e = ln 0 f01 Sc (b 0 B ) 0 01
+
and the location of the shear center is
1 2
I
p A
+ S c Sc
f 2 0 2 (b 2 0 B 2 );
Ss
Z 1 = Sc + A
@
dA: @S
e11 (+1)1
= e 0 f01 (S 0 Sc ) (b 0 B ) + 01
; 0 1 (S 0 S ) + @
(b 0 B ) + f 0 1 w p (S 0 S ) 0 @
: =
+ f s s c @S @S
The second term in the expression for
(+1)1 is proportional to the shear strain eld for pure elastic torsion: @
def t
(S
) = (S 0 S s ) + : @S
3.5.216
We use this denition to eliminate the gradient of yields as nal expression for the strains
e11
(+1)1
= e 0 f01 (S 0 Sc ) (b 0 B ) + 01
; 0 1
t (b 0 B ) + f 0 1 w p 0 (S 0 S ) 0
t 1 : =
+ f c
Virtual work
5 =
ZV
e11 (+1)1
= e 0 f01 (S 0 Sc )b + 01
; 0 t t1 =
+ f01
b + f 01 (S 0 Sc ) 0
wp ;
Elements
where all terms of the order of the distortional strains have been neglected. From the expressions for the average axial strain and average shear strain, we obtain the average axial and shear strain variations as
e
where wp = f 2 01 b 0 w and again all terms of the order of the distortional strains have been neglected. We now introduce the generalized section forces as dened below: Axial force Shear forces Bending moments Twisting moment Warping moment Bimoment
F=
F = Mt =
A Z
11dA (+1)1dA
M = Mw = W=
A Z
ZA
0f (S 0 S c) 11dA
0 t1 f (S 0 S c ) 0
(+1)1 dA
A Z
t f (+1)1 dA
Z A
11dA
3.5.217
5 =
Observe that the total torque T relative to the centroid of the section is the sum of the twisting moment and the warping moment:
f (S
0 Sc )(+1)1 dA = Mt + Mw :
To obtain the rate of change of virtual work, we rst transform the integrations in the virtual work equation to the original volume such that
5 =
`
d`
A
f 2 e11 f 2 (+1)1
= f 2 e 0 f 3 (S 0 Sc ) b + f 2
; 0 2 3 t t1 = f
+ f
b + f (S 0 Sc ) 0
wp :
Ip f e = f 0 f Sc b 0 f 2
+ f 401 A f 2
= f 2
+ f 3 Ss b + f (Sc 0 Ss )wp :
2 2 3
+ Sc Sc b b;
d 5 =
`
d`
A
df + f 01 Ip + Sc Sc dbb
A @ + f @S d w
+ (+1)1(f 2 d
+ 2f df
@ + @S df
+ 3f 2 S b df
where we have neglected terms of order b. The changes in stress follow from the constitutive law
38
3.5.218
We approximate de11 and d 11 with the same relations as used for virtual work:
Ip A
+ Sc Sc
b db;
We now neglect all terms that involve the product of a stress tensor; a variation in curvature, twist, or warping; and a change in axial strain of the cross-section. Using the previously obtained expressions for the second variations in ; b ; b; , and and transforming the results into the current state, this provides
d 5 =
Z
+ 11(201 f 01 df + 01 d 0 01 fS db + f 2 02 + (+1)1 (d
+ 01 fS db + f 01 @ dw) dA:
@S
The incremental moments, forces, etc. are dened as
d`
Ip A
+ Sc Sc dbb)
Elements
dF = dF = dM = dM t = dM w = dW =
Z ZA ZA ZA ZA ZA
t 0 Sc ) 0 d(+1)1dA;
d11dA:
For the determination of the initial stress stiffness, we assume that the second variations of the warping function and its derivative vanish: dw = d = 0. Consequently,
and, therefore, only the torque relative to the centroid plays a role in the initial stress contribution to the rate of change of virtual work:
d 5 = (dF e + dF
+ dM 01b + dMt 01b + dMw f 02wp + dW 01 ` + F de + Fd
+ M 01 db + T 01 db)d`:
The second variations of e and
contain contributions of the second variation in curvature and twist. These can be separated out by dening the bending moments and the torque relative to the origin of the cross-sectional coordinate system:
d 5 = (dF e + dF
+ dM 01 b + dMt 01 b + dMw f 02 wp + dW 01 + ` Ip + Sc Sc dbb)+ F (01d + 201 f 01 df + f 2 02 A 01 01 db)d`: Fd
+ M db + T
We propose to make the cross-section size a function of the stretch . For the thermal stretch of the section we use isotropic expansion (f = ); and for the mechanical stretch of the section we assume an effective Poissons ratio (fm = 0 ). The total cross-sectional stretch is m
f
so that
Using this in the above expression for the rate of change of virtual work, we nd
Z d 5 = (dF e + dF + dM 01 b + dMt 01 b + dMw f 02 wp + dW 01 + ` 01d 0 202d + f 2 02 Ip + S S dbb)+ F ( c c A Fd + M 01 db + T 01 db)d`: [D] is symmetric.
3.5.220
Section integration
The formulation presented in the previous pages is valid for all possible beam types. However, different classes of beams will result in different nal formulations. We consider three different classes of beams: Beams in which warping may be constrained. These beams generally have an open, thin walled section reinforced with some relatively solid parts or some relatively small closed cells and have a torsional stiffness that is considerably smaller than the polar moment of inertia. Hence, in the elastic range, the warping can be large, and warping prevention at the ends can contribute signicantly to the torsional rigidity of the beam. In this case both the torsional shear stresses and the axial warping stresses can be of the same order of magnitude as the stresses due to axial forces and bending moments, and the complete theory must be used. Beams in which warping is unconstrained. These beams generally have a solid section or a closed, thin walled section and have a torsional stiffness that is of the same order of magnitude as the polar moment of inertia of the section. Hence, in the elastic range the warping is rather small, and it is assumed that warping prevention at the ends can be neglected. The axial warping stresses are assumed to be negligible, but the torsional shear stresses are assumed to be of the same order of magnitude as the stresses due to axial forces and bending moments. In this case the warping is dependent on the twist and can be eliminated as an independent variable, which leads to a considerably simplied formulation. Beams in which warping constraints dominate the torsional rigidity. These beams generally have an open, thin walled section and have a torsional stiffness that is much smaller than the polar moment of inertia. In the elastic range the warping is likely to be large, and warping constraints are essential to provide torsional stiffness for the beam. In this case the axial stresses may be of the same order of magnitude as the stresses due to axial forces and bending moments, but the torsional shear stresses are relatively small. Hence, the warping can be coupled to the twist with a relatively stiff elastic constraint but cannot be eliminated because it must be possible to prevent warping at the nodes. Examples of cross-sections for the last two classes will be derived after the general discussion of sections with and without warping prevention. In ABAQUS we neglect the effect of shear stresses due to transverse shear forces at individual material points. We will consequently always assume elastic behavior of the section in transverse shear, leading to the relations
Elements
= fp GA ;
dF = fp GA d ;
where F are the transverse shear forces, working at the shear center, and fp is the slenderness compensation factor used to prevent the shear stiffness from becoming too big in slender beams. The slenderness compensation factor is dened as
3.5.221
fp
12I
where ` is the length of the element and I is the larger of the moments of inertia I11 and I22. Hence, the transverse shear terms do not need to be considered in any further detail. The fact that the transverse shear forces are considered separately allows us to write
(+1)1
F M where
(+1)1 and (F +1)1 are the strains and stresses due to transverse shear forces and
(+1)1 and (M+1)1 are the strains and stresses due to a twisting around the shear center. Substitution in the expressions for the twisting and warping moments yields
Mt Mw
= =
Z ZA
A (S = s
t f (M+1)1 dA;
0 Sc ) F +
t 0 Sc ) 0
1 M
( +1)1
dA;
t where we used the fact that
was calculated based on application of a twisting moment around the shear center and, hence, does not do any work on the shear stresses due to transverse shear forces. The warping function
is assumed to be determined based on isotropic, homogeneous elastic behavior of the section in shear. For this case the elastic energy due to twist is
Ut
= 1 f 2 0 2 (b 0 B ) 2 G 2
J
A
Z
t t dA =
1 f 2 02 (b 0 B )2 GJ; 2
t t
dA:
f 2 01 (b 0 B ) and the energy per unit length of the
Ut
= 1 f 2 0 2 (b 0 B ) 2 G 2
(S
3.5.222
Ip
= (S 0 Sc )(S 0 Sc )dA = A
Z
A
(S
0 Sc ) (S 0 Sc )dA:
Mt
and since
f (S
0 Sc )(+1)1 dA = G
Z Z
f (S
t 0 Sc ) dA;
Mt
it follows that
Z
A
t f (+1)1 dA = G
Z
A
t (S 0 Sc ) dA =
Z
A
t t f dA;
t t dA = J:
For this beam type, warping prevention is not taken into consideration. Hence we assume wp addition we assume that axial strains due to warping can be neglected:
0. For the strains at a material point this yields
= 0. In
Elements
e11 M
(+1)1
and for the variations in the strains
e11 M (+1)1
5 =
Z
`
where the expressions derived earlier for F , F , M , and Mt apply. Although there is no warping prevention in the section, the warping moment Mw does not vanish. From the expression obtained earlier follows
Mw
= (S s 0 S c )F :
= (S s 0 S c )F + M t
3.5.223
= Ss F + Mt:
Consider the case that the shear stresses are dened from the shear strains by linear elastic response, with a constant shear modulus G:
M (+1)1
t = G
(M+1)1 = f01 G
(b 0 B ) + f 01 G
Z Z 0
0 S
t ( 0 Sc ) 0
1
Mt
dA
10 S
t ( 0 Sc ) 0
dA
Substitution of these expressions in the part of the virtual work equation related to torsion yields
5t
Hence, we can write virtual work in terms of the primary variables b and w:
5t
= =
Z`
20
3.5.224
5 =
Z
`
where F , F, M and W are dened as before. For the rate of change of virtual work we obtain similarly
d 5 = (dF e + dF
+ dM 01 b + f 2 02 GJ db b + f 02 G(Ip 0 J )dwp wp + ` dW 01 + F (01d 0 202d + f 2 02 Ip + Sc Sc dbb)+ A 01 01 db)d` F d
+ M db + T
with
= T + f Sc F :
Elements
@
= @S
= 0:
Solid noncircular sections
Solid sections such as rectangles or trapezoids are included in this category. The warping function is a harmonic function and is subject to the condition that no shear stress component can act normal to the boundary of the cross-section. Although it is possible to determine the warping function in this manner, we choose to work in terms of the Saint-Venants stress function because of its simplicity. Following standard procedures we normalize this function so that the (elastic) shear strains can be derived directly from it. We introduce the function '(S ), which is differentiable in the cross-section and has the property that @' t = : @S The stress function is determined by solving the differential equation of the form
3.5.225
For the solid noncircular sections this differential equation is solved numerically using a secondorder isoparametric nite element. The torsional constant of the bar is then equal to twice the volume under the normalized stress function surface.
Closed, thin walled cross-sections
In this case we assume that the shear strain perpendicular to the section must vanish so that
Since t
@ @
= @S
n = t S ; @n which can be identically satised anywhere. The normalized shear strain along the section is
@ = ( S + @S
) t = n S + @
; @s
= 0 n, this yields
@ ( S + @S
)n = 0:
where s is the distance along the section. Integrating this around the circumference yields I I I @
ds = 2AC ;
ds = nS ds + @s where AC is the area enclosed by the section. With the assumption that the shear modulus is constant along the section, the total torsional elastic strain energy is I 1 G
2tds; UT =
where t is the wall thickness. Minimization of the energy with the constraint enforced with a Lagrange multiplier yields I I P = (G h 0 ) ds 0 ( ds 0 2AC ) = 0: Hence,
Gh
and
G
2 = H AC 1
h
h ds
This allows calculation of
at any point in the section based on the section geometry.
Thin walled open sections
The most important sections that exhibit substantial warping are the thin walled open sections. For a single branch section we can conveniently express
as a function of the coordinate s along the section and the coordinate z perpendicular to the section. A suitable approximation for
is
(s; z ) =
c (s) + z (s);
3.5.226
where yields
Z s2 Z h=2
s1
0h=2
0 Ss ) + ddsc
+ z ds
d
d
c + z d + (n S + ) ds ds
dz ds
= 0:
We now introduce S0 , which is the position of the middle of the wall. With t = n and S = S0 + z n , and after carrying out the integration over z , the minimization condition simplies to
Z s2
s1
h
n S 0
0 Ss ) + ddsc
d
d
c + h 0t (S 0 S ) + + 1 h3 1 + d 0 s ds 12 ds
d ds
ds
= 0:
Clearly, the dominant terms are of order h. This yields the equations
c = 0 n (S 0 S ); 0 s ds
Zs
0
= t (S 0 0 S s );
so that
(s; z ) is
(s; z ) =
Elements
where
s is the value of
at the start of the integration over the section.
c (s ) =
Zs
0
0n (S0 0 Ss )ds +
s
represents the sectorial area between two points on the midsection relative to the shear center. Therefore, it readily follows that
Z send
0
c S ds = 0;
Z send
0
so there is no coupling between twist and bending in the section for linear elastic material behavior. As was discussed before,
s must be chosen such that
Z send
0
cds=
h ds
= 0;
which eliminates the coupling between twist and axial extension in the section for linear elasticity. Note that coupling terms still exist but that they are incorporated in the generalized strain displacement relations. The coupling between twist and extension is governed by
0, the value of the warping function at the origin of the cross-sectional coordinate system. If the origin is on the section, this value can be evaluated properly. If the origin is not on the section (which means that the node is not connected to the section), we assume that
0 = 0.
3.5.227
is readily obtained as
n S
=
=
Z s2 Z h=2
Z Z
s1
s1 s2
0h=2
h=2
0h=2
(2 )2
z
dz ds
2 2 0 s ) +
+ 0 ( 0 s ) +
Z s2 1
S d ds t S S d dz
dz ds
s1
ds;
Z s2
s1
h S
( 0
Sc
)(S
0 c)
S
ds:
The above derivations cover single branch open sections. Multibranch open sections can be transformed into single branch open sections by connecting the end of one branch with the beginning of the next branch with a section that has thickness h = 0. Such dummy sections do not yield any contribution to the area, the moments of inertia, or the torsion integral and, hence, have no inuence on the results.
Reference
Beam modeling: overview, Section 15.3.1 of the ABAQUS Analysis Users Manual
3.5.228
EULER-BERNOULLI BEAMS
3.5.3
In these elements it is assumed that the internal virtual work rate is associated with axial strain and torsional shear only. Further, it is assumed that the cross-section does not deform in its plane or warp out of its plane, and that this cross-sectional plane remains normal to the beam axis. These are the classical assumptions of the Euler-Bernoulli beam theory, which provides satisfactory results for slender beams. Let (S; g; h) be material coordinates such that S locates points on the beam axis and (g; h) measures distance in the cross-section. In addition, let n1 ; n2 be unit vectors normal to the beam axis in the current conguration: n1 = n1(S ); n2 = n2(S ). Then the position of a point of the beam in the current conguration is
xf
x + g n1 + hn2 ;
dn1 dS
where
Then
dx dS
+g
+h
dn 2 dS
Elements
d f 2 (dl ) =
=
xf
dS dx
(dS ) dS dn 1 dn 2 +g +h dS dS dS
1
dx f
1
dx dS
+g
dn 1 dS
+h
dn 2 dS
(dS ) :
Now since the beam is slender, we will neglect terms of second-order in g and h, the distance measuring material coordinates in the cross-section. Thus,
f 2 (dl ) =
dx
dS
dx dS
+ 2g
dx dS
dn 1 dS
+ 2h
dx dS
dn 2 dS
(dS ) :
(3.5.31)
Strain measures
I W1 =
Z Z
L
f
f "f dA dLf ;
where f and "f are any material stress and strain measures associated with axial deformation at the point (S; g; h) of the beam, since strains are assumed to be small. For this purpose we will use Greens strain so that
"f =
1 2 [( )
0 1];
3.5.31
EULER-BERNOULLI BEAMS
where (f )2 = (dlf )2=(dLf )2, the square of the ratio of current conguration length to reference conguration length in the axial direction on the ber. From Equation 3.5.31 and its equivalent in the reference conguration, we have
" dL =
f f
1 2
1
dx
dS d d
1
dS
dS
X 1 dX + 2 g dX 1 dN 1 + 2 h dX 1 dN 2
dS dS dS dS dS d
dx dS
+ 2g
dx dS
d 1 d + 2h dS dS dS
x dn 2
1
dS
dS
dS
dS
dS
1 2
dS:
Again, neglecting all but rst-order terms in g and h because of the slenderness assumption, this becomes
"f dLf " + gG01
+ hG
01
dx
dS
dL;
where
G= d
dS
X 1 dX
dS
and " =
1 2
(
0 1) = 2
1
01
#
is the Greens strain of the beam axis.
This simplication allows us to write the internal virtual work rate associated with axial stress as
I W1 =
where with
Z Z
Lf A
f "f dA dLf =
Z Z
L A
f "f dA dL; ~
"f = " ~
1
0 gK2 + hK1 ;
n 0 (1 + ") dX 1 dN2 : 1 dS dS dS Now the cross-sectional base vectors n1 and n2 are assumed to remain normal to the beam axis, so dx 1 n 1 = dx 1 n 2 = 0 :
and
K 1 = 0 G 01
dx
d 2 dS
dS
dS
3.5.32
EULER-BERNOULLI BEAMS
Hence,
dx d 1 1 dS dS
So we have
K 2 = G 01
d2 11 dS 2
x 0 (1 + ")N 1 d2X 1 dS 2
and
K1
This denes the generalized strains associated with axial stretch. For torsional strain the internal virtual work rate is Z Z
=
L A
t f e1 dAdL;
where
e1
t and is the proportionality constant between shear strains and torsional strains (see Beam element formulation, Section 3.5.2, for details). For computational simplicity the form of the torsional strains is taken to be d~ 1 0 2 1 ddS1 ; e1 = 2 1 dS
N n2 1 dn1 0 N2 1 ddS1 dS n N N
Elements
where
~ n 1 = C ( p 1! 1 + p 2 ! 2 ) 1 N 1 ; 1 2 1 2
and
p1 p2
= =
(1
0 g );
(1 + g );
0 1 g 1:
This assumes a linear interpolation of rotation ! along the beam. Thus, the generalized strain measures for these beams are
" K1 and K2 e1
axial strain, the beam curvature change measures, and the torsional strain.
With these measures, the internal virtual work rate can be written
W I
Z h Z
L
"
f dA 0 K2
gf dA + K1
hf dA + e1
Z
A
t f dA dL:
3.5.33
EULER-BERNOULLI BEAMS
For the Jacobian matrix of the overall Newton method, Equation 2.1.13, the variation of this internal work rate with respect to nodal displacement variations must be formed. The constitutive theory is written as f f d d" = [ H ] d
f ; d f and so
8 9 > d" > Z Z < = 5 dK1 f dA + dK1 hf dA " K1 K2 e1 [ B ] + d" > > A A : dK2 ; de1
dW I
Z 4
0dK2
gf dA + de
Z
1
f f dA dL;
where [ B ] is the section stiffness matrix obtained by integration over the cross-section. See Beam element formulation, Section 3.5.2, for more information on section integration.
First variations of strains
n n
x n x n
d2 dS 2 d2 dS 2
dS 2
In these expressions we need 1 and 2 as well as ~ 1; these are now derived. From the expression for ~ 1, namely 1 2 ~ 1 = (p 1! + p 2 ! ) 1 1 ;
n n C
n1 = n1 0 n1 1 tt ~ ~
so that
n1 = n1=jn1j:
3.5.34
EULER-BERNOULLI BEAMS
In addition,
t
dx 2 = g0 1 dS
so that
t
2 = g0 1 (I 0 tt) 1 du : dS
1 1 1 du 01 t; = n1 1 n2 n2 0 g 2 n1 1 dS
= n1 n2 n2 + n1 tt
(3.5.32)
So
n1 n2 =
Elements
and Thus,
2 n2 = n1t 0 n1 1 t n1 ~ ~
jn1j = n1 1 n1 = n1 1 n1: ~ 1
n t 1 0 n~ 111n ~
1
! ! n1 + g0 2 n2
1 du dS
:
0n ~
n1 t ~
1
3.5.35
EULER-BERNOULLI BEAMS
! ! n1 + g 0 2 n1
1 du dS
n1
2 0 g0 1 n2 1 du t: dS
1 du dS
1 d S2
d2 x
n1 t ~
2 K1 = G01 n1 1
and
d2 x
~ n1 t
e1 =
dn 1 ~
so that In addition,
n1 = n1 = n1 : t = g0 2
1
j j
dx dS
so that
t = g
0 1 (I 0 tt) 1 du ; 2
dS
EULER-BERNOULLI BEAMS
because
n1 1 t = 0 :
= G01 d u dS
dd
dS
u: n n
To compute the second variations of bending strain, we need expressions for d 1; d 2 . These are obtained by approximating
! ~ n n2t 1 ! 0 tn1 1 du dS d u ! ~ n2 0n1t 1 ! 0 tn1 1 dS ! ~ ! 1 n2 0 ! 1 n1 n2 t ! d u ! ! 0 t dn1 1 d u : d n1 = dn2t 1 ! + n2dt 1 ! 0 dtn1 1 ~ 1 dS dS
from which
Elements
Using these expressions, the second variations of bending strains are written as
dK1
+ d"N2 1 d X dS 2 2x d ! ! ! ! ! + n1 1 dS 2 n2 1 d! n1 1 ! 0 n1 1 d! n2 1 ! + n2 1 d u n2 1 d! dS d u d2 x ! 0 n2 1 dS 2 n2 1 dS n1 1 d! ddu d u d2 x 0 t 1 dS 2 n2 1 dS n2 t 1 dS
u u ! + n1 1 ddS 2 t 1 d! + t 1 d 2 n2 1 ddu dS dS
3.5.37
EULER-BERNOULLI BEAMS
and
dK2
dS 2
dS
For the torsional strain contribution to the initial stress matrix, we approximate
! ! ! dt 1 d! = d! 2 t 1 d! : dS dS
In the virtual work equation the strains include second derivatives of displacement. For this reason continuity of rotation as well as displacement is needed so that the Hermitian polynomial interpolation functions are the minimum interpolation order needed. These are used here. The Hermite cubic is written in terms of the function value and its derivative at the ends of the interval:
u (g ) = N 1 (g )u 1 + N 2 (g )u 2 + N 3 (g ) du1 dg
+ N 4 (g )
du2 dg
0 1 < g < 1;
with node 1 at g = 01 and node 2 at g = +1. These functions are used in ABAQUS to interpolate the components of displacement and the initial position vector , so that the elements are basically isoparametric. In addition, rotation of ( 1 2 ) about the beam axis, a , is interpolated linearly. This interpolation is unsatisfactory for the user, because the nodal variables are
n n
(ux uy uz
a ):
3.5.38
EULER-BERNOULLI BEAMS
The last four of these variables are difcult to work with; furthermore, making them the same in all elements sharing the same node causes excessive constraint of axial stretch in these elements, especially if the beam axis is not continuous through the node, as in a frame structure or T junction. To avoid this difculty, the following procedure is adopted. At a node the tangent to the beam axis is
t
so
dx = dx = dS dS ; dl dl dx dS dl = dS t:
where ! is the rotation denition introduced above. Since the initial geometry and hence , the initial is the rotation matrix dened by ! , and direction of the beam axis, is known, = 1 ; where hence
x ; uy ; uz ; !x; !y ; !z ;
dl=dS );
t C T
dx dS
=
is dened by these variables and the initial geometry. Furthermore, a is directly available from ! and . Thus, the above set is a satisfactory set of nodal variables. To eliminate the unwanted axial strain constraint, in ABAQUS the stretch dl=dS at the node of each such element is taken as an internal variable, local to the element (a third internal node is created for this purpose, and so it is not shared with neighboring elements.) It should be remarked that the above transformation (Equation 3.5.33) is nonlinear. This leads to some complicationsfor example, the dAlembert forces no longer have the simple form N MMN u ; rather, a matrix M MN = QMP MPQ QQN replaces MMN where QMP and QQN use the transformation (Equation 3.5.33) and its appropriate time derivatives. The resulting Jacobian is nonsymmetric; ABAQUS ignores the nonsymmetric terms.
dl C1T dS
(3.5.33)
Elements
Integration
The cross-section integration has already been discussed in the context of the shear beamsit is the same for these beams. Along the beam axis, the integration schemes are as described below.
Stiffness and internal forces
Three Gauss points are used. Two Gauss points are not sufcient because the torsional strain is not independent.
Mass and distributed loads
Three Gauss points are used. Rotary inertia is not included in the mass, except for rotation of the section about the beam axis, where it is included to avoid singularity in perfectly straight beams.
Reference
Beam modeling: overview, Section 15.3.1 of the ABAQUS Analysis Users Manual
3.5.39
3.5.4
The hybrid beam elements in ABAQUS/Standard are designed to handle very slender situations, where the axial stiffness of the beam is very large compared to the bending stiffness; and so a mixed method, where axial force is treated as an independent unknown, is required. For the shear beams mixed elements are provided where the transverse shear forces are also treated as independent unknowns. This section discusses the basis of these mixed methods.
Axial and bending behavior
I W2 =
~ 0 N )dL; ~ = N:
A linear combination
Elements
0 )W2I ;
Then
I WC
Z
=
(N + (1
dL:
L
=
(dN + (1
Z h
~ 0 )(N 0 N )
dL;
where
N = N + (1
~ 0 )N :
A01 A11
dM3
A03 > d" > < = A13 7 dK1 : 5 A23 > dK2 > : ; A33 de1
38
3.5.41
If L2 A00 < A11 ; A22 (where L is the element length), then the beam is exible axially and the mixed formulation is unnecessary. Otherwise, we assume that an inverse of the rst equation above ~ denes d" from dN :
d" =
1
A00
~ (d N
and so
dM1 = dM2 = dM3 =
A11 A12
0 A01 A
00
2
dK1 + A12
0 A01A02 A
00
dK1 + A22
00 A2 02
de1 + de1 +
~ dN ~ dN
A13
0 A01A03 A
00
dK1 + A23
0 A02A03 A
00
dK2 + A33
0 A03 A
00
2
de1 +
~ dN : 1
Now using the rst tangent section stiffness multiplied by and the second multiplied by Newton contribution of the element becomes
Z
0 , the
L
+
~ A
Z Z "
~ (N d" + M1 dK1 + M2 dK2 + M3 de1 )dL ~ N " + M1 K1 + M2 K2 + M3 e1 + A00 (1
0
3
L
is
0 )
~ 0N
!#
A00
dL;
where
2 6 6 6 6 6 6 6 6 6 6 6 4
~ A
A00 A11
A01
0 (1 0 ) A01 A00
symm
0
3 7 7 7 7 7 7 7: 7 7 7 7 5
A22
0 (1 0 ) A02 A00
A33
0 (1 0 ) A03 A00
~ The variable N is taken as an independent value at each integration point in the element. We choose ~ as =A00, where is a small value. With this choice, by ensuring that the variables N are eliminated ~ ~ after the displacement variables of each element, the Gaussian elimination scheme has no difculty with solving the equations.
3.5.42
Transverse shear
In the mixed elements that allow transverse shear (B21H, B22H, B31H, B32H), the transverse shear constraints are imposed by treating the shear forces as independent variables, using the following formulation. The internal virtual work associated with transverse shear is Z W T S = T
dL i = 1 ; 2;
1
where T1 and T2 are shear forces on the section, and 1 and 2 are variations of transverse shear ~ strain. The virtual work can also be written by introducing independent shear force variables T1 and ~ T2 , as Z
i i
TS W2
where the i are Lagrange multipliers. As in the axial case, we take a linear combination of these two forms, TS TS W TS = W1 + (1 0 )W2 ; where will be dened later. This gives
W TS
where
Ti
Elements
Z 8 9 ~ T i
i + i (1 0 )(Ti 0 Ti ) dL: =0 L ABAQUS treats transverse shear elastically, so Ti = GA
i , where GA is constant. Then the Newton contribution is Z 8 9 ~ ~ (GA d
i + (1 0 )dTi )
i + 1 (1 0 )(GA d
i 0 dTi ) + T i d
i dL L Z 8 9 T i
i + i (1 0 )(Ti 0 T i ) dL: =0 L ~ = GA and choose = =GA, where is a small value compared to GA, We now dene Ti ~ ~ i to give Z 8 9 1 ~ ~ dTi )Ti + T i d
i dL ~ ~ d
i + (1 0 =GA)dTi )
i + (1 0 =GA)(d
i 0 ~ GA L Z 8 9 ~ ~ T i
i + (1 0 =GA)(
i 0 Ti =GA) Ti dL: ~ =0 L
Reference
(dTi + (1
dL
Choosing a beam element, Section 15.3.3 of the ABAQUS Analysis Users Manual
3.5.43
3.5.5
Mass and inertia properties for Timoshenko beams (including PIPE elements) in ABAQUS may come from two separate sources. The rst source is the beams own density and the cross-section geometry. The second source comes from any additional mass and inertia properties per element length that may be applied at specied locations on the beam cross-section. When the mass and inertia are given only from the rst source, the user has the choice of requesting either an isotropic approximation or the exact rotary inertia formulation for the beam mass matrix. When the isotropic formulation is used, the mass of the beam per length is applied to the beams node that is located at the origin of the cross-section axis and the offset between the beams node and the center of mass for the cross-section (if it is nonzero) is neglected in the mass matrix formulation. Let B be the beam density. Mass and the rotary inertia about the center of mass in the beams cross-section coordinate system are dened as
= =
= =
ZV ZV
VZ
= I11 + I22 :
B x1x2 dV;
Here, x1 and x2 are measured relative to the center of mass of the cross-section. For the isotropic (approximate) formulation the mass matrix for the element takes the form
m
0
I331
In two dimensions I33 = I11. In all expressions in this section the mass matrix m that applies to the translational degrees of freedom is lumped for 2-node beams and consistent for 3-node beams. When the exact formulation is used, any offset between the beams node and the center of mass for the cross-section will produce coupling between the translational degrees of freedom and rotational degrees of freedom in the mass matrix for the element. Let mA (x) dene the added mass per beam length. The combined beam mass is dened as
m = mB
mA dl:
Let r be the vector between the center of mass C and some point with current coordinates x,
r = xC 0 x:
3.5.51
For a rigid body the velocity of any point in the body is given by
u = u 0 ! 2 r; _ _
C
_ where ! = is the angular velocity of the body. Taking the time derivative of this expression, the acceleration is _ u = uC 0 ! 2 r 0 ! 2 (! 2 r):
The local or strong form of the equilibrium equations represents the balance of linear momentum and balance of angular momentum; these two equilibrium equations are
m dt d dt
( u + ! 2 r) = F ; _
3
d2
_ I 1 ! + m r 2 (u + ! 2 r ) = M + r 2 F ;
where F and M are external forces acting at the center of mass and external moment and I is the rotary inertia tensor. The variational or weak form of equilibrium is
Wint + Wext = 0 :
Taking the time derivative in the equilibrium equations, the internal or dAlembert force contribution is
_ 0W = u 1 u dV = m u + ! 2 r + ! 2 (! 2 r) 1 u 1 13 2 0 0 2 _ _ _ + I 1 ! + ! 2 I 1 ! + m (! 2 r) 2 u + r 2 u + r 1 0 rr 1 ! 0 (! 1 r) (! 2 r) 1 ;
int V
where u is the variation of the position of a point in the body and is the variation of the rotation of the rigid body reference node. The external loading contribution is
Wext = F u +
Z
M+r2F
1 :
For linear theory all nonlinear terms are neglected, so the internal force contribution simplies to
0W
1 2 r 1 u 1 13 2 0 0 _ _ + I 1 ! + m r 2 u + r2 1 0 rr 1 ! 1
int
_ u u dV = m u + !
and leads to the following mass matrix for all linear and linear perturbation analyses:
m m1^ r
0m 1 ^ r
Ie
;
3.5.52
When the inertia of a rigid body is used with implicit time integration, the Jacobian contribution of
Wint is required. It can be written in the form
0dW = m u
int
dU + 1 (dK1 + mdK2 ) ;
! _ _ 1 2 I 1 ! + ! 2 I 1 d! + d 2 I 1 ! + I 1 (! 2 d ) + (! 1 I 1 !)d 0 (! 1 d )I 1 ! + ! 2 I 1 (! 2 d ); 2 3 _ _ r ! dK2 = ru 0 (r 1 u)1 0 (! 2 r)r + (! 1 r)^ 1 d! 2 ^ ^ _ r _ _ _ + 0(! 1 r)u 0 (r 2 u)! + ru 0 (r 1 u)1 + (r 1 ! )^ 0 rr 1 ! + (! 2 r) ! 1 ^ r 3 2 _ _ 0 (! 1 r) ((! 1 r)1 0 r! ) 1 d + (r! 0 ! r) 1 du + ^ 1 du + (r 1 0 rr) 1 d! : r
_ ! dK1 =I d! + d!
In the Jacobian formulation for 3-node beams, a consistent mass matrix is used for translational degrees of freedom and a lumped mass matrix is used for rotational degrees of freedom and the terms that couple translational and rotational degrees of freedom.
Elements
3.5.53
MESHED SECTIONS
3.5.6
Overview
The beam theory introduced in Beam element formulation, Section 3.5.2, applies to homogenous beams (made out of a single material) and assumes that the shear center of the beam cross-section is either known or can be easily calculated. However, if the beam cross-section is arbitrarily shaped and/or the beam is made of more than one material layer, nding the cross-section shear center and warping function are no longer trivial tasks. To perform these tasks, the cross-section has to be numerically integrated using a nite element discretization over the two-dimensional cross-section region. The nodal degrees of freedom of the nite element cross-section model represent warping displacements (in general, in and out-of-plane warping degrees of freedom) that allow the shear center and beam torsional rigidity to be determined. Numerical integration of this meshed section also EiAi , integrated bending provides the stiffness and inertia statistics: integrated axial stiffness i stiffness ( i EiIi ) , integrated shear stiffness ( i Gi Ai ) , total mass per unit length i i Ai , and rotary inertia ( i i Ii ) . The warping function and the shear center are derived for shear exible Timoshenko beams under the following assumptions:
Cross-sections are solid or closed and thin-walled and have a torsional stiffness that is of the same order of magnitude as the polar moment of inertia of the section. Hence, in the elastic range the warping is small, and it is assumed that warping prevention at the ends can be neglected. The axial warping stresses are assumed to be negligible, but the torsional shear stresses are assumed to be of the same order of magnitude as the stresses due to axial forces and bending moments. In this case the warping is dependent on the twist and can be eliminated as an independent variable, which leads to a considerably simplied formulation. Hence, the theory is based on a solid cross-section with unconstrained warping. Using the notation from Beam element formulation, Section 3.5.2, we assume that wp = 0 and the axial strains due to warping can be neglected:
0. Beams can be made out of linear elastic materials either with isotropic properties or orthotropic shear properties dened by two shear moduli G1 and G2 given in two perpendicular directions. The stress-strain relationship for the elastic orthotropic material in the beam cross-section axis yields
Elements
8 9 2 38 9 E 0 0 < e11 = < 11 = G1(cos )2 + G2(sin )2 (G1 0 G2 ) cos sin 5
21 ; 21 = 4 : ; sym G1 (sin )2 + G2(cos )2 :
31 ; 31 where 11 represents the beams axial stress, 21 and 31 represent two shear stresses, and the angle is a user-dened material orientation. For the isotropic material properties the above relationship becomes 8 9 2 38 9 E 0 0 < e11 = < 11 = 4 05 : 21 ; = sym G G :
21 ; : 31
31
3.5.61
MESHED SECTIONS
Material bers are aligned with or perpendicular to the beam axis; hence, in-plane warping can be neglected and the out-of-plane degree of freedom is the only unknown warping value. This assumption can be inaccurate if the beam consists of materials with very different stiffness properties.
Dening the shear center and warping function
At a given stage in the deformation history of the beam, the position of a material point in the crosssection is given by the expression
e11 (+1)1
= e 0 S b +
@ = + b(S + @S ): ;
Express these strain components relative to the centroid and the shear center strains, respectively, as
e11 (+1)1
= e c 0 (S 0 S c ) b +
where Sc is a section centroid, Ss is a section shear center, ec is the axial strain at the centroid, and s is the shear strain at the shear center. The elastic energy in the beam is
@ = + b (S 0 Ss ) + @S s
;
condition that requires the warping function to be orthogonal to the axial and bending energies:
fE (ec 0 (S 0 Sc ) b + )2 + ` A s s G ( + b ; + b(S 0 Ss ) )( + b ; + b(S 0 Ss ) )gd`dA: Although we assume no warping prevention (i.e., = 0), the above energy leads to the following
1Z 5= 2
Z
A
8 E e c 0 b (S
0 Sc )
dA = 0:
3.5.62
MESHED SECTIONS
The cross-section centroid is dened as the point about which the coupling between axial and bending vanishes. Hence, the centroid location follows from
e c b
E(S
0 Sc )dA = 0;
and
S EdA R : Sc = A A EdA
The shear center is dened as the point about which the coupling between twist and transverse shear vanishes. Hence, the following term is zero in the elastic energy:
Z
A
G b( ; + (S 0 Ss ) )dA = 0: s
Let us express the warping function as a sum of three parts: a warping function 0 superimposed on the unknown rigid translation C0 and rigid rotation about the yet unknown shear center Ss . This assumption can be written as
= 0 + C0 0 Ss (S 0 Sc ):
Elements
Substituting the above into the expression for elastic energy, using the property of the shear center, and minimizing the energy with respect to the warping function, we get
@ G A @S
@ 0 + S dA = 0: @S
This equation is solved numerically over the meshed section and gives the value of 0(S ). Recall that the warping function satises the orthogonality condition
Z 8
A
Substituting
ec 0 b (S
0 Sc )
E dA = 0:
, grouping axial and bending terms, and using the centroid denition, we get
Z 8
A
ec E( 0 + C0) 0 b (S
0 Sc )E[ 0 0 Ss (S 0 Sc )]
dA = 0:
This expression must be true for any value of axial strain ec and curvatures b , so we can write two separate equations that provide constant C0 and shear center components Ss :
Z
A
A
(S
E ( 0 + C0) dA = 0;
0 Sc )E
0 Ss (S 0 Sc )
dA = 0:
3.5.63
MESHED SECTIONS
Hence,
C0 Ss
=0
Z
A E 0 dA ; R A EdA 0 S c )E
0 dA:
= (EI )01
(S
Finally, the section integrated stiffness properties are dened as Z (EA) = EdA;
Z
(EI ) =
E (S
0 Sc ) (S 0 Sc )dA;
Z
(GA) =
Z
G dA;
(GJ ) =
G (
; + (S
0 Ss ))(
Z
; + (S
0 Ss ))dA:
(A) =
Z
dA;
(I ) =
(S
0 Sm ) (S 0 Sm )dA;
K
fp
ts = f K ts ; p
l2 (EA)
= 1=(1 +
12(EI )av
);
ts = k(GA) ;
where k equals 1.0 for meshed cross-sections and depends on the nite element interpolation.
3.5.64
SHELL ELEMENTS
3.6.1
The ABAQUS shell element library provides elements that allow the modeling of curved, intersecting shells that can exhibit nonlinear material response and undergo large overall motions (translations and rotations). ABAQUS shell elements can also model the bending behavior of composites. The library is divided into three categories consisting of general-purpose, thin, and thick shell elements. Thin shell elements provide solutions to shell problems that are adequately described by classical (Kirchhoff) shell theory, thick shell elements yield solutions for structures that are best modeled by shear exible (Mindlin) shell theory, and general-purpose shell elements can provide solutions to both thin and thick shell problems. All shell elements use bending strain measures that are approximations to those of KoiterSanders shell theory (Budiansky and Sanders, 1963). While ABAQUS/Standard provides shell elements in all three categories, ABAQUS/Explicit provides only general-purpose shell elements. For most applications the general-purpose shell elements should be the users rst choice from the element library. However, for specic applications it may be possible to obtain enhanced performance by choosing one of the thin or thick shell elements. It should also be noted that not all ABAQUS shell elements are formulated for large-strain analysis. The general-purpose shell elements are axisymmetric elements SAX1, SAX2, and SAX2T and threedimensional elements S3, S4, S3R, S4R, S4RS, S3RS, and S4RSW, where S4RS, S3RS, and S4RSW are small-strain elements that are available only in ABAQUS/Explicit. The general-purpose elements provide robust and accurate solutions in all loading conditions for thin and thick shell problems. Thickness change as a function of in-plane deformation is allowed in their formulation. They do not suffer from transverse shear locking, nor do they have any unconstrained hourglass modes. With the exception of the small-strain elements, all of these elements consider nite membrane strains. No hourglass control is required for the axisymmetric general-purpose shells, nor in the bending and membrane response of the fully integrated element S4. The membrane kinematics of S4 are based on an assumed-strain formulation that provides accurate solutions for in-plane bending behavior. The ABAQUS/Explicit elements S3RS, S4RS, and S4RSW are well-suited for many impact dynamics problems, including structures undergoing large-scale buckling behavior, which involve small-strains but large rotations and severe bending. These elements use simplied methods for strain calculation and hourglass control and offer signicant advantages in computational speed. Thin shell elements are available only in ABAQUS/Standard. STRI3 and STRI65 are triangular small-strain, thin shell elements; S4R5, S8R5, and S9R5 comprise the quadrilateral small-strain, thin shell elements, while SAXA is a nite-strain, thin shell element suitable for modeling axisymmetric geometries subjected to arbitrary loadings. Thin shell elements may provide enhanced performance for large problems where reducing the number of degrees of freedom through the use of ve degree of freedom shells is desirable. However, they should be used only for the modeling of thin structures that exhibit at most weak nonlinearities in problems where rotation degree of freedom output is not required and for situations where the shell surface and the displacement eld are smooth so that higher accuracy can be achieved with the use of second-order shells. SAXA elements very effectively model axisymmetric structures undergoing asymmetric deformation when only a few circumferential Fourier modes describe the circumferential variation of the deformation accurately.
Elements
3.6.11
SHELL ELEMENTS
The Discrete Kirchhoff (DK) constraint, which refers to the satisfaction of the Kirchhoff constraint at discrete points on the shell surface, is imposed in all thin shell elements in ABAQUS. For element type STRI3 the constraint is imposed analytically and involves no transverse shear strain energy calculation. Solutions obtained with these elements converge to those corresponding to classical shell theory. For element types STRI65, S4R5, S8R5, S9R5, and SAXA the discrete Kirchhoff constraint is imposed numerically where the transverse shear stiffness acts as a penalty that enforces the constraint. Shell behavior that can be properly described with shear exible shell theory and results in smooth displacement elds can be analyzed accurately with the second-order ABAQUS/Standard thick shell element S8R. Nonnegligible transverse shear exibility is required for this element to function properly; hence, the element is suitable for the analysis of composite and sandwich shells. Irregular meshes of S8R elements converge very poorly because of severe transverse shear locking; therefore, this element is recommended for use in regular mesh geometries for thick shell applications.
Thickness change
In geometrically nonlinear analyses in ABAQUS/Standard the cross-section thickness of nite-strain shell elements changes as a function of the membrane strain based on a user-dened effective section Poissons ratio, . In ABAQUS/Explicit the thickness change is based on the effective section Poissons ratio for all shell elements in large-deformation analyses, unless the user has specied that the thickness change should be based on the element material denition. The thickness change based on the effective section Poissons ratio is calculated as follows. In plane stress 33 = 0; linear elasticity gives
33 = 0 1 0 (11 + 22):
ln
t0
01 0
ln
l1 0 l1
+ ln
l2 0 l2
01 0
ln
A0
where A is the area on the shells reference surface. This nonlinear analogy with linear elasticity leads to the thickness change relationship:
t t0
A 0 10
A0
= 0 :0
For
= 0 :5
Reference
Shell elements: overview, Section 15.6.1 of the ABAQUS Analysis Users Manual
3.6.12
AXISYMMETRIC SHELLS
3.6.2
These two shell elements are axisymmetric versions of the shells described in the previous section and use the reduced-integration penalty method of Hughes et al. (1977). While these are shell elements, they are also simple extensions of the two-dimensional beam elements B21 and B22. The extension is the inclusion of the hoop terms. These elements are thus one-dimensional, deforming in a radial plane. The Cartesian coordinates in this plane are r (radius) and z (axial position). Distance along the shell reference surface in such a plane is measured by the material coordinate S (see Figure 3.6.21).
z
n Shell normal
Elements
The 2-node element (SAX1) uses one-point integration of the linear interpolation function for the distribution of loads. The mass matrix is lumped. The 3-node element (SAX2) uses two-point integration of a quadratic interpolation function for the stiffness and three-point integration of a quadratic interpolation function for the distribution of loads. SAX2 uses a consistent mass matrix. All integrations use the Gauss method. The integration through the thickness follows the usual numerical or exact scheme used in ABAQUS.
3.6.21
AXISYMMETRIC SHELLS
Theory
This shell theory allows for nite strains and rotations of the shell. The strain measure used is chosen to give a close approximation (accurate to second-order terms) to log strain. Thus, the theory is intended for direct application to cases involving inelastic or hypoelastic deformation where the stress-strain behavior is given in terms of Kirchhoff stress (true stress in the usual engineering literature) and log strain, such as metal plasticity. The theory is approximate, but the approximations are not rigorously justied: they are introduced for simplicity and seem reasonable. These approximations are as follows: a. A thinness assumption is made. This means that, at all times, only terms up to rst order with respect to the thickness direction coordinate are included. b. The thinning of the shell caused by stretching parallel to its reference surface is assumed to be uniform through the thickness and dened by an incompressibility condition on the reference surface of the shell. Obviously this is a relatively coarse approximation, especially in the case where a shell is subjected to pure bending. It is adopted because it is simple and models the effect of thinning associated with membrane straining: this is considered to be of primary importance in the type of applications envisioned, such as the failure of pipes and vessels subjected to overpressurization. c. The thinning of the shell is assumed to occur smoothlythat is to say, gradients of the thinning with respect to position on the reference surface are assumed to be negligible. This means that localization effects, such as necking of the shell, are only modeled in a very coarse way. Again, the reason for adopting this approximation is simplicitydetails of localization effects are not important to the type of application for which the elements are designed. d. All stresses except those parallel to the reference surface are neglected; and, for the nonnegligible stresses, plane stress theory is assumed. As with (c) above, this precludes detailed localization studies, but introduces considerable simplication into the formulation. e. Plane sections remain plane. This has been shown to be consistent with the thinness assumption, (a) above, for most material models. Here it is simply assumed without further justication. f. Transverse shears are assumed to be small, and the material response to such deformation is assumed to be linear elastic. Transverse shear is introduced because the elements used are of the reduced integration, penalty type (see Hughes et al., 1977, for example). In these elements position on the reference surface and rotation of lines initially orthogonal to the reference surface are interpolated independently: the transverse shear stiffness is then viewed as a penalty term imposing the necessary constraint at selected (reduced integration) points. This transverse shear stiffness is the actual elastic value for relatively thick shells. For thinner cases the penalty must be reduced for numerical reasonsthis is done in ABAQUS in the manner described in Hughes et al. (1977). The theory is now described in detail. The concepts are taken from various sources, most especially Budiansky and Sanders (1963) and Rodal and Witmer (1979). The position of a material point in the shell is given by
3.6.22
AXISYMMETRIC SHELLS
x1 = x + t n;
where
(3.6.21)
x ( 1 ; 2 ) n ( 1 ; 2 )
t
( 1 ;
2 )
is the position of a point on the reference surface of the shell; is a unit vector in the thickness direction, this direction being initially orthogonal to the reference surface; is the stretch of the shell in the thickness direction; measures position with respect to the thickness direction, in the reference conguration; and are material coordinates in the reference surface.
The assumptions listed above imply that t = t (1 ; 2 ) only and that @t =@1 ; @t =@2 are small quantities. Equation 3.6.21 is written at the end of an increment, and at the start of an increment the same equation is written as
X1 = X + o N : t
The metric at the end of an increment is
(3.6.22)
n x n
n
(3.6.23)
Elements
where
g
and
@ @
@ @ @ @ 1 @ + @ 1 @ @ is an approximation to the curvature tensor (second fundamental form) of the reference surface. b would be precisely the curvature tensor as it is usually dened if b @ n @x = 0:
1
This is only approximately true for these elements, because a small transverse shear is allowed. At the start of the increment the same quantities are
@ 1 @ 1 o 1 = G + t B : (3.6.24) @ @ Axisymmetric shells undergoing axisymmetric deformations have the great simplication that principal directions do not rotate. Thus, by assuming that 1 and 2 are oriented in these principal
3.6.23
AXISYMMETRIC SHELLS
directions ( 1 is meridional and 2 is circumferential), the stretch ratios that occur within the increment in these directions are written as
1 =
1
1 @ x1 @ x1 @ X1 @ X 1 0 2 1 = 1 ; @ @ @ @
where from this point onward the summation convention has been dropped for indexes and . Using Equation 3.6.23 and Equation 3.6.24 and truncating to rst order in then gives
11 1(1 + t1k);
where and
2 1 = (g = G) 1
(3.6.25)
1 , is dened as
"1
1
b 1k = t g 0 1t B : G
t
1" = 2 1 + 1 1
1 1 0 1
Because this expression approximates the increment of log strain correctly to second-order terms, it can be thought of as a central difference approximation for the rate of deformation. This expression is used because we anticipate that strain increments of a maximum of 20 percent per increment will be used: at that magnitude the difference between this denition of incremental strain and the increment of log strain is about 1%, which seems to be acceptable (4 % of the increment). At lowerand probably more typicalvalues of strain increment, the error is very much less. Again expanding to rst order in the thickness direction coordinate, , we obtain
1 = 1+
41 (1 + 1)2 :
AXISYMMETRIC SHELLS
41 (1 + 1)2 1:
This then gives
1 1" 1" + t1k:
(3.6.26)
The stretch ratio in the thickness direction is assumed to be dened by the following relation on the reference surface:
where th is the thickness stretch ratio caused by thermal expansion. t From the denition of k;
b 1B 1k = g 0 G :
t
Elements
(3.6.27)
@ @X 1 = @x 1 n 0 @ 1 N:
(3.6.28)
This simple form is used because these strains are always assumed to be small. This completes the statement of the incremental strain denitions, and sotogether with a virtual work statement to represent equilibriuma theory is available. However, it is necessary to satisfy the minimum requirement that the theory provide constant strain under appropriate motions. This is essential if the theory is to be suitable for many practical cases, most especially those involving thermal loading. Interestingly, the theory in Rodal and Witmer (1979) appears to violate this requirement. To achieve this, a modied incremental curvature change measure is dened as
~ 1k = 1k + sym(
1"
);
is a tensor, dened as follows. We know that the radii of curvature of the -line at the end and at the beginning of an increment are
given by where
1 = b r g
3.6.25
AXISYMMETRIC SHELLS
and
In these expressions, as in the following development, no summation is implied by a repeated index . If the -line is stretched uniformly by during the increment, we require that
and, further, such uniform stretch of the shell must give constant strain so that since we assume
1 ~ 1" = 1" + t1k;
~ 1k = 0 10 1 = 1 1 0 1 1k = r 1 R R 1 1 t t 1 1 0 1 : = R 1 1
t
Dening
~ 1k = 1k + 11 0 11 to B G t
(3.6.29)
and assuming
(3.6.210)
satises the requirement. Equation 3.6.29 may be simplied by substituting in the denition of in Equation 3.6.27 to give
1k
B ; G
and so
~ 1k = g 1 (b 0 1B):
(3.6.211)
The formulation is completed by the assumption that the virtual work equation can be written
Z
o t
1 ;2
where
(3.6.212)
3.6.26
AXISYMMETRIC SHELLS
( 1 ; 2 )
"1 T
in the shell, dened by plane stress theory using the summation of the strain increments in Equation 3.6.210 to dene the strain at this point; are the variations of the strain increments in Equation 3.6.210; are the transverse shear forces per unit area, dened by T
=
kGh ; where
h kG
are the transverse shear strains from Equation 3.6.28, is the original thickness of the shell, is the elastic transverse shear stiffness (reduced according to the suggestions of Hughes et al. (1977) if the shell is too thin, to avoid numerical problems);
W E
Axisymmetric shell element library, Section 15.6.9 of the ABAQUS Analysis Users Manual
Elements
3.6.27
3.6.3
This section discusses the formulation of the small-strain shear exible elements in ABAQUS/Standard, which are quadrilaterals (S4R5, S8R5, S9R5, and S8R), except for the 6-node triangle STRI65. The essential idea of these elements is that the position of a point in the shell reference surfacexand the components of a vector nwhich is approximately normal to the reference surfaceare interpolated independently. The kinematics of the shell theory then consist of measuring membrane strain on the reference surface from the derivatives of x with respect to position on the surface and bending strain from the derivatives of n; the strain measures that are used for this purpose are approximations to Koiter-Sanders theory strains (Budiansky and Sanders, 1963). The transverse shear strains are measured as the changes in the projections of n onto tangents to the shells reference surface. For these element types the strain measures are suitable for large rotations but small strains, and the change in the shells thickness caused by deformation is neglected.
Notation
Elements
z y x
3.6.31
Let (1 , 2) be a set of Gaussian surface coordinates on the shell reference surface. Since these coordinates are only needed locally at an integration point, we use the elements isoparametric coordinates as these coordinates. x(1 ; 2 ) is the current position of a point on the interpolated reference surface, and (1 ; 2 ) is the initial position of the same point. The unit vector
N N N
@ = @1
@ @2
,r
@ @1
@ @2
is the unit normal to the interpolated reference surface in the initial conguration. This vector gives a sidedness to the surfaceone surface of the shell is the top surface (in the positive direction along from the shells reference surface) and the other is the bottom surface. The vector corresponding in the current conguration, , will be made approximately normal to the reference surface in to the current conguration by imposing the Kirchhoff constraint discretely. In the rest of this section Greek indices will be used to indicate values associated with the (twodimensional) reference surface and so will sum over the range 1, 2 under the summation convention. First, we establish convenient directions for stress and strain output. These will be local material directions, indistinguishable (to the order of approximation) from corotational directions, since we assume strains are small. The standard convention used throughout ABAQUS for such local directions on a surface is as follows. It is most convenient to choose orthogonal directions. Dene
so long as N1 < cos 0:1 , where is a unit vector in the global X -direction; otherwise,
T2 = N i i N
p 2 2
where
T2 = N kk ; N
p 2 2
Let
dS =
so that the dS are locally dened distance measuring coordinates at each material point. The transformation
@S @ @
=
@X T @ d;
1
@S @
@
=
@S
3.6.32
dS 2) directions.
The following surface measures are dened. The metric of the deformed surface is
g =
@x @x 1 ; @S @S
1 @ n 1 @x + @n 1 @x b = 0 2 @S @S @S @S
(this is only an approximation because is not exactly normal to the surface in the current conguration). The corresponding measures associated with the original reference surface are the metric
G =
and the approximation to the curvature
@ X @X 1 @S @S
Elements
B = 0
The vectors @ =@S are dened from the derivatives of the interpolation functions and the normals at the nodes. These nodal normals are calculated as average values of the normals to the surfaces of all elements abutting the node. ABAQUS determines if the surface is intended to be smooth at the node (the criterion is that the angle between the normals at the node should be less than 20). If the surface is not calculated as smooth, separate normals are set up in the different surface branches at the node. Thus, B should be a reasonable approximation to the second fundamental form of the original reference surface.
Displacements
1 @ N 1 @X + @N 1 @X : 2 @S @S @S @S
The nodal variables for shell elements are the displacements of the shells reference surface, = 0 , and the normal direction, . Since is dened to be a unit vector, only two independent values are needed to dene , so that this type of shell element needs only ve degrees of freedom per node. In ABAQUS this issue is addressed in two ways. At nodes in a smooth shell surface in those elements that naturally have ve degrees of freedom per node, ABAQUS stores the values of the projections of the change in projected onto two orthogonal directions in the shell surface at the start of the increment to dene . Otherwise, ABAQUS stores the usual rotation triplet, ! , at the node. This latter method leaves a redundant degree of freedom if the node is on a smooth surface. A small stiffness is introduced locally at the node to constrain this extra degree of freedom to a measure of the same rotation of the shells reference surface.
u x X
3.6.33
Interpolation
The same bipolynomial interpolation functions are used for all components of u, , , and . The shear exible shell elements in the library use bilinear interpolation (four nodes), fully biquadratic interpolation (nine nodes), and serendipity quadratics (eight nodes).
Strains
XN
=
The curvature change is
1 2
g 0 G ):
= B 0 b +
The transverse shears are where
1 2
B + B ):
t =
@x @S
,r @ x
3 = n 1 t ; @S @ 1 @Sx
(no sum on )
is a unit vector, tangent to the dS line in the current surface. In addition to these strains, when six degrees of freedom are used at the nodes of the elements, the extra rotation degree of freedom is constrained with a penalty, as follows. When such a node is the corner node of an element, dene 1 , 2 , 1 , dS 1 , and dS 2 in the element as above. Notice that these will be different in each element at the node, since the interpolated surface is not generally continuous. Then the strain to be penalized is dened as
T T N
SRC =
where
1 2
( 1
t 1 t 2 0 t2 1 t 1 );
t = C 1 T t =
@x @S
is the rotated tangent direction, as dened by the rotation values at the node, and
,r @ x
1 @x @S @S
(no sum on )
is the rotated tangent direction dened by the motion of the interpolated reference surface at the node.
3.6.34
At each midside node in the original conguration, dene as the average surface normal for the elements of this surface branch at the nodes (there will be at most two such elements) and as the tangent to the edge. Then dene
t = C T and n = C N
1 1
as rotated values of
T and N, as dened by the rotation values at the node. The vector p=t n
2
is then normal to and to the edge. The strain to be penalized at these midside nodes is then dened as
SRM
where
t= x
@ @S
,r x
@ @S
= t p;
1 1
@ @S
x
Elements
is the tangent to the edge of the element in the current position of the reference surface.
Penalties
The transverse shear strains are calculated at a set of reduced integration points and have the following stiffness associated with them:
K3
= G3h1(area)=(1 + q1(area)=h2); 1 4 10
where the G3 are the elastic moduli associated with transverse shear. The G3 are dened directly by the user or are computed from the elastic moduli given for the layers of the shell. h is the shell is the value of reference surface area associated with this integration point in the thickness; numerical integration scheme; q is a numerical factor, currently set to = 2 04 . (See Hughes et al., 1977, for a discussion of such factors.) Transverse shears are always treated elastically: nonlinear material calculations in shells are based on plane stress theory, using the membrane and bending strains to dene the strain on the surface parallel to the shells reference surface at each integration point through the shells thickness. When rotation constraints are required at nodes that use six degrees of freedom, the penalty used is K kGh = q =h2 :
1(area)
to the node and the factor k is introduced. This (small) factor has been chosen based on numerical experiments, to be large enough to avoid singularities yet small enough to avoid adding signicantly to the stiffness of the model.
1(area) (1 + 1(area) ) This is the same as the transverse shear constraint, except that 1(area) is here an area assigned
3.6.35
These strain measures, with the interpolation specied above, give zero strain for any general rigid body motion
where 1,
x = X + u 1 + ( C 1 0 I ) 1 (X 0 X 1 ) n = C1 1 N;
= g
= 0b + (B
+ B
) 2
3 = n 1 t + n 1 t
1 1 2
g =
In forming the initial stress matrix we approximate by neglecting d , d , etc, to simplify the expressions and reduce the cost of forming the matrix. Numerical experiments have suggested that, at least for the problems tested, this does not signicantly affect the convergence rate. With this approximation,
3.6.36
d =
@d @ @d ! 2 n 1 @Su + ! 2 @Sn 1 @Su @S ! @! @du @ n @du ! amp; + 2 n 1 + ! 2 1 @S @S @S @S ! @! @ u ! @ n @ u amp; + 2 n 1 + d! 2 1 @S @S @S @S @! ! @ u ! @ n @ u amp; + 2 n 1 @S + d! 2 @S 1 @S @S 1
amp; + (B d
+ B d
)
1 2 2
@! !
d 3
@u (no sum on ) @S @du 01 ! d
SRC =g1 2 ! 2 t1 1 (I 0 t2 t2 ) 1 2 @S 0 1 ! 2 t 1 (I 0 t t ) 1 @du 2 ! 0 g2 2 1 1 @S 1 1 @ u 0 ! amp; +g1 2 d! 2 t1 1 (I 0 t2 t2 ) 1 @S 2 1 @ u 0 ! amp; 0g2 2 d! 2 t2 1 (I 0 t1 t1) 1 @S 1
+
01 ! =g 2 ! 2 n 1 (I 0 t
1
t ) 1
0 ! g 2 d! 2 n 1 (I 0 t t ) 1
@du @S
Elements
and
! d
SRM =g0 2 d! 2 p 1 (I 0 t t) 1
1
! g0 2 ! 2 p 1 (I 0 t t) 1
1
@ u @S 2
@du : @S 1
For these shell elements the internal virtual work rate is assumed to be
W I =
+ + +
Z Z
A X r X nc X n
m
"f dzdA
where KI , KII , and KIII are the transverse shear stiffness and the penalties dened above and r indicates the integration points at which transverse shears are calculated, nc indicates corner nodes
( ) ( ) ( )
3.6.37
at which six degrees of freedom are used, and nm indicates midside nodes at which six degrees of freedom are used. Here "f and are the strain and stress in the (dS , dS ) material directions in a surface offset by a distance z from the reference surface. The usual Kirchhoff assumption is adopted:
"
= + z ;
Z
h
Z
A
Z
dz +
z dz dA:
The thickness direction integrations are performed numerically in ABAQUS. The integration scheme is a Simpsons rule, of user-chosen order. The shell can also be considered layered, with different properties at each layer and a different integration scheme assigned (by the user) to each layer.
Pressure load stiffness
The load stiffness associated with pressure loading is often important in shells, especially in eigenvalue buckling estimates on elastic shells. In ABAQUS/Standard the pressure load stiffness is implemented as a symmetric form, thus assuming that the pressure magnitude is constant over the surface and neglecting free edge effects. See Hibbitt (1979) and Mang (1980) for details. The load stiffness is obtained in such a form as follows. The external virtual work associated with pressure is Z
W e
where p is the pressure load per unit area, given in terms of the (externally prescribed) pressure magnitude, p, as
p 1 u dA;
p dA = p
Z
A
p
@ 1 @x
@x
Thus,
W e
The change in this term caused by change of displacement of the shell (the load stiffness) is
dW e
p u
@x
d1 d2 ;
since we assume that pressure magnitude, p, is externally prescribed and has no dependence on position, x. Neglecting free edge effects, and assuming the magnitude p is uniform, results in the symmetric form Z
dW e
1 p @ x 1 @du 2 u + @ u 2 du @2 @2 A 2 @1 @du @ u @x 0 @2 1 @1 2 u + @1 2 du d1 d2 : =
3.6.38
Shell elements: overview, Section 15.6.1 of the ABAQUS Analysis Users Manual
Elements
3.6.39
3.6.4
Element type STRI3 in ABAQUS/Standard is a facet shella plate element used to approximate a shell. The element has three nodes, each with six degrees of freedom. The strains are based on thin plate theory, using a small-strain approximation. Arbitrary rigid body rotations are accounted for exactly by formulating the deformation of the element in a local coordinate system that rotates with the element. The element also satises the patch test, so that it will produce reliable results with appropriate meshes. The bending of the element is based on a discrete Kirchhoff approach to plate bending, using Batozs interpolation functions (Batoz et al., 1980). This formulation satises the Kirchhoff constraints all around the boundary of the triangle and provides linear variation of curvature throughout the element. However, the membrane strains are assumed constant within the element. In addition, a curved shell is approximated by this element as a set of facets formed by the planes dened by the three nodes of each element. For these reasons it is necessary to use a reasonably well rened mesh in most applications.
Kinematics
A local orthonormal basis system, T1 and T2, is dened in the plane of each element in the reference conguration, using the standard ABAQUS convention. S 1 and S 2 measure distance along T1 and T2 in the reference conguration.
3
Elements
T2 N
T1
3 2 2 1
3.6.41
" =
where
1 (g 0 G); 2
g = G =
and = 1; 2
@x @x 1 @S @S @ X @X 1 @S @S
is the metric in the reference conguration. are the spatial coordinates of a point in the current and reference congurations, Here and respectively. Curvature changes are dened incrementally. To account for large rigid body rotations we use a local coordinate system that rotates with the plane dened by the three nodes of the element. The basis vectors chosen for this local system are 1 @ =@S 1 and 2 @ =@S 2 . Since the membrane strains are assumed to be small, these vectors will be approximately orthonormal. The components of incremental rotation of the normal to the plate are dened as !1 about 1 and !2 about 2 . The incremental displacement of the reference surface of the plate along the normal to the plane of its nodes is dened as w. (Note that w will be zero at the nodes at all times because the plane containing 1 and 2 always passes through the nodes.) The Kirchhoff constraints are, approximately,
t = x
t = x 1
1 01!2 + @@Sw = 0 1
and
dened independently along each of the three sides of the element as a cubic function. The Kirchhoff constraints are then imposed at the corners and at the middle of each element edge along the direction of the edge to give
1 1!1 + @@Sw = 0: 2 Batoz (1980) assumes that 1!1 and 1!2 vary quadratically over the element and that 1w is
P 1 !1 = H 1 1 P
and where
1 1 2 2 3 3 1 1 ; 1 2 ; 1 1 ; 1 2 ; 1 1 ; 1 2 5 : P P In the above expressions H1 (g; h) and H2 (g; h) are interpolation functions that are dened by N Batoz (1980), and the incremental rotation components at the nodes, 1 , are dened as N 1 = n N 1 t ;
1 P is the array
P 1 !2 = H 2 1 P ;
4
3.6.42
where
and N are the increments of the rotational degrees of freedom at the node N , is the rotation i is the normal to the plane of the elements nodes at the beginning of matrix dened by N , and i the increment. Finally, the incremental curvature change measures are dened as @ !2 k11 @S 1 @ !1 k22 0 @S 2 and @ !1 @ !2 k12 0 : @S 1 @S 2 The three membrane strains and three curvature strains complete the basic kinematic description of the element, except that the use of six degrees of freedom per node introduces a spurious rotation at each node (only two incremental rotations at each node appear in the above equationsthe rotation about the normal to the plane of the elements nodes does not enter). To deal with this problem, we dene a generalized strain to be penalized with a small stiffness at each node as
N nN = C(1i ) 1 N;
1 = 1 1 = 1 1 =
1 + 1
i
where
= b 1 1 a 1 0 b2 1 a 2 ;
and j , k are the node numbers in cyclic order forming the two sides of the triangle at the node i.
First variations of strain
a1 = C 1 A 1 a2 = C 1 A 2 A1 = N 2 A1 A2 = A2 2 N A 1 = ( X j 0 X i )= j X j 0 X i j A 2 = ( X k 0 X i )= j X k 0 X i j b 1 = ( x j 0 x i )= j x j 0 x i j b 2 = ( x k 0 x i )= j x k 0 x i j
@x
Elements
K11
"
=1 2
@ x @ x @ x 1 @ S + @S 1 @S @S
3.6.43
where
and in
!1 !2 P ,
P 1 = H1 1 P 1 = H P 1 P ;
2
N
1 = N 2 nN 1 t + nN 1 t:
Also, for the strain used to introduce the extra stiffness at the nodes to avoid singularity caused by the component of rotation about the normal,
i
! ! = b 1 a 0 b 1 a + b 1 ! 2 a 0 b 1 ! 2 a :
1 1 2 2 1 1 2 2
d"
=1 2
@ x
@dx @ x @dx 1 @S + @S 1 @S @S
N d
0
and
@g1 x 2 0 x1 @S 1
)+
@g2 x 3 0 x1 @S 1
! ) 2 nN 1 d!N
d i
Here g1 and g2 are coordinates in the plane of the element, normalized so that the nodes of the element are at (0,0), (1,0) and (0,1).
Internal virtual work rate
W I
Z Z
A h
"f dz dA
X
nodes(i)
K i i i ;
3.6.44
where "f = " + zk is the strain at a point, f , away from the reference surface; are the stress components at f ; h is the shell thickness; and K i is the penalty stiffness used to constrain the spurious rotation. The formulation now proceeds as for the shell elements described in Shear exible small-strain shell elements, Section 3.6.3, using a 3-point integration scheme in the plane of the element.
Reference
Shell elements: overview, Section 15.6.1 of the ABAQUS Analysis Users Manual
Elements
3.6.45
3.6.5
This section describes the formulation of the quadrilateral nite-membrane-strain element S4R, the triangular element S3R and S3 obtained through degeneration of S4R, and the fully integrated nitemembrane-strain element S4.
Geometric description
At a given stage in the deformation history of the shell, the position of a material point in the shell is dened by
where the subscript i and other Roman subscripts range from 1 to 3. Subscripts and other lowercase Greek subscripts which describe the quantities in the reference surface of the shell range from 1 to 2. In the above equation t3 is the normal to the reference surface of the shell. The gradient of the position is
@x @S
@x @S
f 33
@ t3 S; @S 3
@x @S3
f 33 t3;
Elements
where we have neglected derivatives of f 33 with respect to S . Note that in the above S are local surface coordinates that are assumed to be orthogonal and distance measuring in the reference state. S3 is the coordinate in the thickness direction, distance measuring and orthogonal to S in the reference state. The thickness increase factor f 33 is assumed to be independent of S3 . In the deformed state we dene local, orthonormal shell directions ti such that
ti 1 tj =
ij ;
t i t i = I;
where ij is the Kronecker delta and I is the identity tensor of rank 2. Summation convention is used for repeated subscripts. The in-plane components of the gradient of the position are obtained as
f = t 1
@x @S
f + B f 33S3 ;
def
=
@x @x t 1 = t 1 @S S3 =0 @S @ t3 : @S
B def t 1 =
3.6.51
In the original (reference) conguration we denote the position by and the direction vectors by Ti , which yields
X (S ) = X (S ) + T 3 (S )S 3 :
i
@S
@S
@ 3 S3 ; @S
@S3
X =T ; 3
@S
= + B S3 ;
where we have assumed that the in-plane direction vectors follow from the surface coordinates with
T = @S
@
S3 =0
@S
T @ T3 : @S
1
The original reference surface normal gradient is obtained in the nite element formulation from the interpolation of the nodal normals with the shape functions. In the deformed conguration it is not derived from the nodal normals but is updated independently based on the gradient of the incremental rotations.
Parametric interpolation
The position of the points in the shell reference surface is described in terms of discrete nodal positions with parametric interpolation functions N I ( ). The functions are C continuous, and are nonorthogonal, nondistance measuring parametric coordinates. For the reference surface positions one, thus, obtains
x ( ) = N I ( ) x I ;
@
X ( ) = N I ( ) X I :
@
@N I @
xI ;
@
X = @N I XI :
@
Note that uppercase Roman superscripts such as I denote nodes of an element and that repeated superscripts imply summation over all nodes of an element.
3.6.52
Now consider the original conguration. The unit normal to the shell reference surface is readily obtained as . @X @X @X @X
T3 =
Subsequently, we dene two orthonormal tangent vectors T and distance measuring coordinates S along these vectors. The derivatives of these coordinates with respect to follow from
@S @
@1
@2
@
@2
I @X = T @ = T XI @N : @
1
@S @
01
= @N @
I @ : @S
The original reference surface normal gradient is obtained from the nodal normals
B
TI with 3
Elements
= T TI @N 3 @S
1
Since the original reference surface normal gradient is obtained by taking derivatives with respect to orthogonal distance measuring coordinates, we will call B = b the original curvature of the reference surface.
Membrane deformation and curvature
3 01
With this expression we can dene the gradient operator in the current state:
@ @s
def
@ = h @S
or inverted
@ @S
@ = f @s
The gradient operator in the current state can also be dened as the derivative with respect to distance measuring coordinates s along the base vectors t, since
x =t @s
@
x h =f h =
@S
@
3.6.53
and, hence,
t
@ = @sx :
= T
@s
since
f
h
= t
x T T @X t = t @ x @X t = t t = :
@x @X @X @ x
@
1 1 1 1 1 1
t+1t t 1f = t+1t @x t @s
1
1h = tt
def t =
@s
t+1t :
xt
With a local coordinate system dened in the current state, the current gradient of the normal can be transformed into the curvature of the surface:
b
t3 = B h :
@s
@
Orientation update
The equations given in the earlier sections are valid for any local coordinate system dened in the current state. The t vectors at the beginning of the analysis are determined following the standard ABAQUS conventions. In this section, we outline the way in which the in-plane coordinates are made corotational. To obtain the updated version of t, we follow a two-step approach. First, we construct orthogonal vectors t tangential to the surface (following ABAQUS conventions). Subsequently, we calculate
^ = ^ @ xt+1t : t @S
1
where
Curvature change
We assume that the nodal spin will be interpolated with the interpolation functions N I . During an increment the nodal spin is assumed to be constant; consequently, the value of the spin at each material point will be constant. Hence, we can use the same interpolation functions for the incremental nite rotation vector :
( )
1 = !1t = N I ()!I 1t = N I ()1I : The nite rotation vector can be split in a rotation amplitude 1 and a rotation axis p: 1 = 1 p; with 1 def j1j and p = 1=1: =
To rotate the shell normal, we use quaternion algebra. The incremental nodal rotation is represented by the rotation quaternion q , which is dened by
Elements
~3+1t = 1q tt31qy: tt
This updated shell normal does not actually have to be calculated: it is used only for the derivation of the expression for the curvature change. It is not equal to the shell normal used at the start of t the next increment t3+1t, which will again be chosen perpendicular to the reference surface. The updated normal used here will be approximately orthogonal to the reference surface, depending upon the amount of transverse shear deformation. The gradient of the updated shell normal can be obtained by differentiation:
t @ t3+1t
@S
The second term on the right-hand side can be written in the form
1y 1q t3 @@Sq
t
@ 1 q 0 t 1y y 1q t @S 3
y
=0
@ 1q t y y t 1q : @S 3
1 t = 2V @S tt31qy + 1q @S3 1qy: @S The inverse of a rotation quaternion such as 1q is equal to its conjugate (1q01 = 1q y). Hence, we can write @~3+1t tt = 2V @1q 1qy1q tt 1qy + 1q @tt3 1qy
@S
Hence, the scalar parts of the rst two terms cancel each other and the vector parts reinforce each other, leading to @ q @ tt+1t @ t
~3
1R
def V =
2 @ 1q 1q y @S
which must be expressed in terms of the gradient of the incremental rotation. From the denition of the incremental quaternion q follows
thus, for
1R, again with use of the incremental quaternion denition 1R = p @1 + sin 1 @p + (1 0 cos 1) p 2 @p :
@S @S @S
1R and some algebra one obtains sin 1 @1 + 1 0 cos 1 p 2 @1 + 1 0 sin 1 pp 1 @1 : 1R = 1 @S 1 @S 1 @S
3.6.56
Note that
1R ! @1 when 1 ! 0. @S
t+ B1t
~t+1t t t t t = t+1t 1 @t3 = t+1t 21R 2 ~3+1t + 1q t
B 1qy3 tt @S t t t t t t tt = 1R 1 0~3+1t 2 t+1t1 + t+1t 1 t
+1tB
=
t
+1t 1 1R + B ;
1 1
= 2 = 0 ; 2
2 1
= 0 1 = 1 : 2
t Note that the change in B is independent of B . Calculation of B involves taking the gradient with respect to the reference conguration. It is more convenient to use the reference surface curvature tensor
b
def B =
@t = t 1 @s3 :
Elements
t+1 b t
t = t +1t 1 1r + bt 1h :
This expression makes it feasible to calculate the update in the reference surface curvature by taking gradients in the latest updated state only.
Deformation gradient
We already have obtained an expression for the deformation gradient in the reference surface, and we have assumed that the thickness change is constant:
F
= f ;
F 33
= f 33:
( )
= F + S3(f 33B 0 f
B
):
3.6.57
We can write this as the product of a nite-membrane deformation and a bending perturbation:
F
+ S3(f 33B h
0 f B"3h"
)3 f
2 =
+ S3(f 33b
0 f b h"
) f
: "
It will be assumed that the deformation (strain and rotation) due to bending is small and, therefore,
S3 f 33b
0 f b h" ) 1: "
The membrane strain increment follows from the incremental stretch tensor , whose components follow from the incremental deformation gradient f by the polar decomposition f V R . t t+1t Let f and f be the deformation gradient at the beginning and the end of the increment,
1V
respectively. By denition f
t+1t
Since R are the components of an orthogonal matrix, the square of the incremental stretch tensor can be obtained by
1f
1f
= 1V
1V
= 1 = 1R =
2 X
I =1
2 X
I =1
(1I )2aI aI
ln(1I )aI aI
and the average material rotation increment is dened from the polar decomposition:
2 X
1 I I 1I aa 1f : I =1
1R :
Curvature increment
Following Koiter-Sanders shell theory, and compensating for the rotation of the base vectors relative to the material, we dene the physical curvature increment as
1 = sym
t+1t 0 bt 1R
+ bt 1R 1"
3.6.58
(1" )2 relative to 1" , this expression can be rewritten as i h i h t+1 t+1 t 1 = sym b t 0 bt 1R 1V 01 = sym b t 0 bt 1h = sym 2 t +1t 1 1r 3 ; =0 F = f + S3f 33b f :
where use was made of the curvature update formula. Observe that the curvature at the beginning of the increment, bt , does not appear in this equation. Hence, there is no need to calculate the initial . The deformation gradient can, hence, also be simplied curvature b , and we can assume b to For the material strain increment at a point through the shell thickness Koiter-Sanders theory thus yields
Virtual work
5=
" dV:
Elements
We assume that the variations in the strain can be expressed in terms of variation in membrane strain and curvature with the same relations as apply to the increment in strain:
"
Z
5=
"
+ f 33S3 )dV:
M
def =
f 33 dS3;
Z
def =
f 33S3 S3 ;
5 = (N " + M ) dA:
A
The variation in the curvature is obtained by taking variations in the incremental curvature, which yields
= sym(
1 1 h +
1 1 h +
1 1 h ): We neglect the terms of order 1 and also terms of order 3 1 1 , which yields
= sym(
1 1 h ):
We evaluate
= sym( t 1r + t 1r )
1 r
with respect to the current state (at the end of the increment). Hence for the
1 = 0 .
@
1
t
1 r t
@ @S
= sym
@ @S
h
= sym
@ @s
:
To obtain an expression for the rate of virtual work, we rst write the virtual work equation in terms of the reference volume
5=
V
" dV
Z Z
A
" dS3 dA ;
where is the Kirchhoff stress tensor, related to the Cauchy or true stress tensor via
= J :
d 5 = d
Z Z
A h
Here r indicates that the rates are taken in a material, corotational coordinate system. The terms involving stress rates are related to the material behavior. We assume constitutive equations of the form Substituted in the expression d
d 5 =
Z Z
5 and transformed back to the current conguration, this yields (" C
d"
+ dr" )f 33dS3dA:
3.6.510
dr = J C d" :
A h
Consistent with the derivation of the virtual work equation itself, we neglect terms of the order df 33S3 . Hence, the rate of virtual work can be written as
d 5 =
hZ
It remains to determine
dr " and dr .
dr h = 0h
dr f
h :
Substitution in the expression for the second variation yields
dr " = sym dr t 1 @ x 0 t 1 @ x dr t
1
dr x 0 t 1 @s t
1 @s @s @s
@s
@ x @ x @ dr x @ x r t 1 r t
1 t t 1 0d 0 @s 1 t
t 1 @s : = sym d @s @s
@x @ x @
Elements
@s
@s
The in-plane components of the corotational rate of the base vectors can also be expressed in terms of the in-plane material spin in the reference surface:
dr t
= d
t = 1
d 0 t 1 @ dx t : 2 t 1 @s
@s
@ x
3.6.511
d "
1
yields
@ dx @s
@ x @s
0 2d" "
:
This expression is identical to the one obtained with standard continuum elements.
Second variation of the curvature
We need to calculate the second variation of the curvature to calculate the initial stress contribution from the curvature:
m dr :
To simplify the computation, we rely on the intrinsic denition of curvature and express the curvature in derivatives with respect to the isoparametric coordinates. Accordingly,
m dr
ab r = m d
ab
ab r =m d
@ x
@ a
1 @ b + @ a 1
@t3
@x
@t3 @ b
where the bending resultant components mab are the components expressed in the orthonormal coordinate system (m ) transformed by @ a =@s . Denoting derivatives with respect to the isoparametric coordinates as @=@ a ;a , the second variation of the curvature is
1 dr t3;b + dx;a 1 t3;b + x;a 1 (dr t3);b : Using the fact that t3 = 2 t3 and dr t3 = 2 t3, we nd that .
d ab = x;a d ab = x;a
t 1 [0^3] 1 ;b + x;a 1 [0^3;b] 1 + ;b 1 [^3] 1 dx;a + 1 [^3;b] 1 dx;a t t t . . + ;b 1 (t3 x;a ) 1 + 1 (t3;b x;a ) 1 + 1 (t3 x;a ) 1 . . . ;b 0 ;b 1 (t3 1 x;a I) 1 0 1 (t3 1 x;a I) 1 ;b 0 1 (t3;b 1 x;a I) 1 : . . .
Here [^3] indicates the skew-symmetric tensor with axial vector t3. t
Transverse shear treatment
Several interpolation schemes have been proposed to avoid shear-locking, which typically arises as the thickness of a plate or shell goes to zero. Here we employ an assumed strain method based on the Hu-Washizu principle. This scheme derives from that by MacNeal (1978), subsequently extended and reformulated in Hughes and Tezduyar (1981) and MacNeal (1982) and revisited in Bathe and Dvorkin (1984). Computational aspects of the nonlinear theory are investigated in Simo, Fox, and Rifai (1989) for fully integrated quadrilateral shell elements. For reduced integration quadrilateral and triangular shell elements that can be used for both implicit and explicit integration, this assumed strain method needs to be modied. We summarize below the assumed strain method used with fully integrated
3.6.512
elements, followed by the modications required for the one-point integration plus stabilization used in ABAQUS.
Construction of the assumed strain eld
Consider a typical isoparametric nite element, as depicted in Figure 3.6.51, and denote by A; B; C; D the set of midpoints of the element boundaries.
Elements
Figure 3.6.51 Notation for the assumed strain eld on the standard isoparametric element. The following assumed transverse shear strain eld is used:
1
2
where
A
2 C
2
=1 2
= 2
(1 (1
0 ) 1 0 ) 2
+ (1 + ) 1 + (1 + ) 2
12
;
B ; B B ;
= tA =
C
1x2 0T 1X2;
A ; A A ; C ; C C ;
B 1 D 1
= =
t 1x10T 1X1;
B
t 1x 20T 1X2;
t 1x1 0T 1X1
D D ; D D ;
are the covariant transverse shear strains evaluated at the midpoints of the element boundaries. In the above transverse shear strain denitions, the use of uppercase letters indicates quantities in the
3.6.513
reference conguration and the use of lowercase letters indicates the deformed conguration. For readability we have omitted the subscript 3 from the director eld. Making use of the bilinear element interpolation, it follows that
x A = 1 (x 4 0 x 1 ) ; ;2 2
where xI , for I ; ; ; , are the reference surface position vectors of the element nodes. By making use of the assumed strain eld along with the update formulae for the director eld, the assumed covariant transverse shear eld can be written concisely in matrix notation. Recall the director eld update equation and the corresponding linearized director eld:
tt+1t
=1 2 3 4
x C = 1 (x 3 0 x 2 ) ; ;2 2
x B = 1 (x 2 0 x 1 ) ; ;1 2 xD ;1
= 1 ( x3 0 x4 ) ; 2
^ = exp[1] 1 t
A
)
tB tD
t
= 2 t :
tA tC
; ;
; :
1 2
;
where
1 2
=B
sm
x
+B
sb
;
BT AT
(1 0 )t 0(1 + )t
A ;
BT CT
(1 + )t (1 + )t
B 1 D 1
B
DT CT
DT
:
AT
A 2 C 2
= t 2x2; = t 2x2;
A C C ; BT AT
=t 2x1; =t 2x1:
D D ; DT CT
= 1 (1 0 )1 B 4 (1 0 )2
sb
(1 0 ) 1 (1 + )2
BT
CT
(1 + )1 (1 + )2
(1 + ) 1 (1 0 )2
DT AT
:
3.6.514
Constitutive relations
A St. Venant-Kirchhoff constitutive model for the Kirchhoff curvilinear components of the resultant transverse shear force is written in terms of the transverse shear strains as
Q1 Q2
= Cs
1 2
where
Cs is the transverse shear stiffness in curvilinear coordinates. For a single isotropic layer, Cs = 5 Gs h 6
A11 A21 A12 A22
The matrix A is the inverse of the metric A , where metric components in the reference conguration A are dened by the inner product
= X; X; :
1
The Cauchy or true transverse shear force components in 1 2 T are calculated with the coordinate transformation fq ; q g
q=
Initial stress stiffness
q1 q2
=A a
1 f1 2 f1
1 f2 2 f2
Elements
Q1 Q2
The calculation of the initial stress stiffness matrix requires the second variation of the assumed transverse strain eld. This calculation can be summarized in matrix notation as follows. Dene vectors of variations of the nodal displacement quantities:
8 9 > x1 > > > > 1 > > > > > > > x2 > > > > > > < = 2 u = ; > x3 > > > > > > 3 > > > > > > > x4 > > > > : ; 4
Then the initial stress contribution is written
9 8 > d x1 > > d > > 1> > > > > > > > dx2 > > > > > < d2 = : du = > dx3 > > > > > > d3 > > > > > > > > dx4 > > > : d4 ; Z
q d
=
1
u Ks 1u da ;
1 1
3.6.515
where da is the area measure in the current conguration and s is the (symmetric) transverse shear contribution to the initial stress, dened as follows. Let I be the 3 2 3 identity matrix; then dene the symmetric matrices
A B A = Q2(1 0 ) symftA (xA )T g 0
2 I ; B = Q1 (1 0 ) symftB (xB )T g 0
1 I ; ;2 ;1 C D C = Q2 (1 + ) symftC (xC )T g 0
2 I ; D = Q1(1 + ) symftD (xD )T g 0
1 I : ;2 ;1
Also dene the skew-symmetric matrices
^ t a = Q2 (1 0 )^A ; ^ b = Q1 (1 0 )^B ; t ^ d = Q1(1 + )^D : t
3 3
c = Q2(1 + )^C ; ^ t
Also, let 0 be the
2
3
Ks =
For reduced-integration elements the transverse shear force components need to be evaluated at the center of the elements. Consider s the transverse shear contribution to the internal energy:
s
1 2
1 Cs 1 dA :
The reference area measure dA is written in terms of the isoparametric coordinates as dA = Jdd, p where J = A11A22 0 (A12 )2 and A are the components of the reference surface metric in the undeformed conguration. This transverse shear energy can be approximated in many ways to produce a one point integration at the center of the element plus hourglass stabilization. It is important that this treatment yield accurate representation of transverse shear deformation in thick shell problems and provide robust performance for skewed elements. The treatment should collapse smoothly to a triangle, which should be insensitive to the node numbering during collapse; that is, the triangles response should not depend on the nodal connectivity. For an entire mesh of triangular elements, the treatment should give convergent results (that is, the element should not lock). Furthermore, the high frequency response of the transverse shear treatment should be controlled so that transverse shear response does not dominate the stable time
3.6.516
increment for explicit dynamic analysis (including for skewed triangular or quadrilateral geometries). All of these requirements are embodied in the following transverse shear treatment. Dene the transverse shear strain at the center of the element (the homogeneous part) and the hourglass transverse shear strain vectors as
=1 2
B D 1 + 1 A + C 2 2
c + cc c
and hg
= bf
cc
The element distortion coefcients c and c are constants determined by the element reference geometry. For geometries with constant Jacobian transformation, c = c = 0. The components of the hourglass strain vector hg are dened in terms of the edge strains as
bf
A B C D = 0 0A 2 + 0 B 1 + 0 C 2 0 0 D 1
and cc
A B C D = 0 2 + 1 + 2 0 1 :
The coefcients 0A , 0B , 0C , and 0D are constants determined from the reference geometry of the element. For rectangular elements 0A = 1 , 0B = 0 1 , 0C = 1 , 0D = 0 1 ; and bf can be identied 2 2 2 2 as the strain associated with the rotational buttery deformation pattern. We call cc the crop circle mode strain since it corresponds to a deformation pattern that resembles the sweeping over the element normals in a circular pattern. The inclusion of the crop circle strain cc in the homogeneous part of the transverse shear strain 0 has two important consequences. First, it makes the transverse shear response insensitive to the nodal connectivity for a triangular element. That is, when a side of a quadrilateral element is collapsed to form a triangle, the elements response is independent of the choice of node numbering on the element. Second, for explicit dynamic analyses the coefcients c and c are chosen to minimize the highest frequencies associated with the homogeneous part of the transverse shear response. To illustrate the crop circle and buttery transverse shear patterns, consider a square, initially at element. Furthermore, consider plate theory kinematics; that is, two rotations and a vertical deection at the nodes. The crop circle pattern has zero vertical deection at the nodes and a nodal rotation vector pattern as illustrated in Figure 3.6.52.
Elements
x
Figure 3.6.52 Crop circle pattern: zero deection and circularly symmetric rotations.
3.6.517
The buttery pattern has vertical deections that correspond to cross-diagonal bending; that is, two equal deections at two nodes across a diagonal, with equal and opposite deections at the remaining two nodes. The nodal rotations develop in a way that opposes the bending motion of the reference surface; that is, the rotations are opposite the rotations that would develop for this displacement pattern to produce pure bending. The buttery modes nodal vertical deection and rotation vector pattern are illustrated in Figure 3.6.53.
+ y
Figure 3.6.53 Buttery pattern: vertical deection and rotation vectors. Let the reference element area be A0 = 4J0. The transverse shear energy can be approximated as a center point value plus a stabilization term:
s
1 2
A0 0 1
Cs 0
0 + A0
hg 1
2
hg ;
where s 0 is the transverse shear stiffness evaluated at the center of the element and the hourglass is the diagonal matrix stiffness
H=
eff C s0
12
1 0
0:001
:
eff The effective stiffness Cs0 is the average direct component of the transverse shear stiffness, eff 11 + C 22)=2. C s0 = ( C s0 s0 The formulation of the homogeneous part of the transverse shear has two contributions: the average edge strain across the element, plus the element distortion term. The average strain treatment is essentially the same as that for the assumed strain formulation of MacNeal and others presented earlier, with expressions evaluated at the center of the element ( = 0 and = 0). The details of
3.6.518
this part are omitted; only the element distortion term is presented in detail. The variation of the homogeneous transverse shear strain can be written
0 Bccd
and
where Bsm0 and Bsb0 are Bsm and Bsb evaluated at the center of the element, = 1 c (tA 0 tB )T c (tB 0 tC )T c (tC 0 tD )T c (tD 0 tA)T 2 c (t A 0 t B ) T c (t B 0 t C ) T c (t C 0 tD ) T c (t D 0 t A ) T
Bccr
1 c ( B 0 A )T c ( B + C )T c ( C 0 D )T c (0 A 0 D )T 1 2 1 2 2 1 2 1 : 2 c ( B 0 A )T c ( B + C )T c ( C 0 D )T c (0 A 0 D )T 1 2 1 2 2 1 2 1
hg
The stabilization term has a similar formulation. The variation of the hourglass strain is
= Bd x + Br ;
Elements
where
h1
12
bf
and h2
= 0:001
12
cc :
Comments on stabilization
(1) The buttery mode bf is applied with a large or physical hourglass stiffness. For a reference geometry with constant Jacobian, the buttery stabilization term can be derived from an exact integration of the assumed strain formulation of the transverse shear energy. It is important to apply this constraint with a high stiffness to prevent overly exible response for quadrilateral elements. The crop circle mode is applied with a small or weak stiffness. Although this mode can propagate, it is rarely problematic and is often prevented with boundary conditions. (2) As the quadrilateral element is degenerated to a triangle, the two hourglass constraints converge into a single constraint: the crop circle constraint. However, as is well-known, for a constant strain triangle the element will lock for certain meshes with three transverse shear constraints per element. Therefore, in the case of a triangular element, the (strong) buttery mode stabilization is not applied.
3.6.519
Only the (weak) crop circle mode stabilization is applied. Thus, in addition to the two homogeneous transverse shear strains, the triangle has a weak constraint to prevent spurious zero energy modes, yet avoids locking in most situations. The initial stress contribution from the stabilization terms takes the following form:
A0 (h1 d
bf + h2 d
cc ) = A0 u
2 2 3
K h 1 du ;
2 3
where Kh is the (symmetric) transverse shear stabilization contribution to the initial stress. Dene the symmetric matrices
^ t ah = (h1 0A + h2)^A ;
c ^h = (h1 0C + h2)^C ; t Kh =
2
Then Kh is written
^ bh 0 0 0 6a 0 b ^h ^ HB 0 0 0 ah 6 ^h 6 ^ ^ ^h 0 0bh + ch 0 c 0 6 ^ ^ c 1 6 0bh c ^h HC 0 HB b h 0 ^h H B + H C 6 6 ^ 0 0 0 ^h c 0 0 ^h + d h c 0 46 0 6 ^ ^ ^h 0 d h H C 0 H D dh 0 0 0 ^h c HC c 6 ^ 4 ^ 0 ah 0 0 0 0 dh 0 A ^h ^ ^ ^h 0H 0 0 0d 0HD dh 0 ah a Note that once the matrix entries in Kh are dened, Kh is lled just as Ks .
0 0 ^ bh 0
^ ^ 0 ah + b h 0HA + HB ^ 0 bh
0 0 ^ dh 0HD ^ h + ah 0d ^ 0HD 0 HA
The initial stress contribution from the homogeneous part consists of two terms, one from the assumed strain formulation (evaluated at = = 0) as detailed earlier, and the other from the crop circle mode addition. These two terms can be written where qj0 and Ks j0 are the shear force and matrix Ks evaluated at the element center. The matrix expression for (q1 c + q2c )d cc is analogous to Kh from the stabilization terms.
A 0 q j0 1 d 0
The in-plane displacement hourglass control is applied in the same way as in the ABAQUS membrane elements. The hourglass strains are dened by
@x 1 xI I @S
@X @ 0 @S 1 XI I = @Sx 1 xI I 0 T 1 XI I ;
3.6.520
where I is the hourglass mode. This mode is obtained by making the regular hourglass mode 0I = f1; 01; 1; 01g orthogonal to the homogeneous deformation mode in the undeformed shape of the element. This last condition can be written as
I
Observe that
= 0 I 0 0J XJ 1 T
@N I : @S
@X T 1 XI I = T 1 XI 0I 0 0J XJ 1 T @S 1 T = T 1 XI 0I 0 0J XJ 1 T = 0;
and consequently
@x 1 xI I : @S This expression can be worked out further. We dene the projected nodal coordinates z
I S
and the projected element area
def
T 1 XI
Elements
A=
1
2
3
= = = =
A A A A
Kh
= ( rF G )
@N I @N I hA; @S @S
where G is the shear modulus and rF is a small number chosen to be 0.005 in ABAQUS/Standard and 0.05 in ABAQUS/Explicit. When the hourglass control is based on assumed enhanced strain, the articial stiffness factor is replaced by coefcients derived from a three-eld variational principle. The hourglass force Z conjugate to z is then equal to
= K h z :
3.6.521
I I Note that the second term vanishes in the initial conguration since is needed for the Jacobian. From the rst variation follows right away
For virtual work we need the rst variation of the hourglass strain. From the expression for the strain follows immediately @x I I @ x 1 xI I : z = 1 x + @S @S
= 0.
dz =
@d I I 1 @S
x x
@ @S
x dxI
I :
1
The second variation does not contribute in the initial conguration since initially Z
Rotational hourglass control
= 0.
The expressions for the curvature change, the transverse shear constraints, and the drilling mode constraints still leave three nonhomogeneous rotational modes unconstrained. These modes correspond to zero rotation at the midedges and zero gradient at the centroid. Hence, they correspond to the familiar 0I = f1; 01; 1; 01g hourglass pattern. To pass curvature patch tests exactly, it is necessary to use orthogonalized hourglass patterns as derived for in-plane hourglass control. This last aspect implies that the rotational hourglass mode corresponds to the mixed derivative of the rotation at the centroid: @ 2 1 I 1I = 4 2 : @1@2 We cannot use the above formulation directly in a formulation suitable for multiple nite rotation increments. Hence, we use the same approach as for the calculation of the curvature change. For the purpose of the calculation we dene the updated shell direction vectors
~t+1t = 1q tt 1qy : ti i
The updated shell direction vectors do not actually have to be stored: they are used only for the derivation of the expression for the hourglass strain. We now formally dene the hourglass strain tensor as @2 j #ij = i 1 : @1 @2
Observe that
t t
= ti 1
@2 j @1 @2
@2 i @1 @2
For the purpose of hourglass strain calculation we assume that all products of rst-order derivatives with respect to 1 and 2 can be neglected. Consequently,
t t
= ti 1
@2 j @1 @2
t 1 tj = 0;
3.6.522
and, hence,
#ij
= 0#
j
ji
is skew-symmetric. Observe that the mixed derivative of ti can be expressed in terms of the hourglass strain tensor with @2t @2t incrementally. From the expression for tt+1t we obtain i
@ 2 ti
=tt
@1 @2
~ +1
t
ij
ij
are obtained
@1 @2
@1
@2 @1
1q
t i
@ tt i
@2
@2
t i
t i
@1
@1 @2 @ tt @ i
In this expression we also ignore all terms with products of derivatives with respect to 1 and 2 . Hence, the above expression simplies to
@2 @ qy = @11q2 t 1qy + 1q t @1@2 + 1q @@1t@2 1qy: @1 @2 @ 1 @ ti
t t t i t i t i
~ +1
Elements
The second term on the right-hand side can be written in the form
1q y = @ 1q 1q @ @ @ @
tt i
@
0 t 1y ti
1q
y
1 q t 1 q y y : = 0 @ @
@
t i
1 t 1q + 1 q t 1q : @1 @2 The inverse of a rotation quaternion such as 1q is equal to its conjugate (1q01 = 1q y); hence, we can write @~ t +1 = 2V @1q 1qy1q t 1qy + 1q @t 1qy
@1 @2
Hence, the scalar parts of the rst two terms cancel each other and the vector parts reinforce each other, leading to t+1t @ ti @ q t @ t y y i
= 2V
@1 @2
t i
@1 @2
@1 @2 where we have formally dened the incremental hourglass update vector
@1 @2
t i
t i
t i
t i
@1 @2
t i
t i
1q 1q y ; 1R0 = V 2 @ @
def
@
3.6.523
which must be expressed in terms of the incremental rotation hourglass mode. From the denition of the incremental quaternion q follows, while neglecting the products of 1 and 2 derivatives:
1q = 1 sin 1 @1 ; 1 cos 1 p @1 + sin 1 @p ; @1 @2 2 2 @1@2 2 2 @1@2 2 @1@2 thus, for 1R0 again with use of the incremental quaternion denition 1R0 = p @1 + sin1 @p + (1 0 cos 1) p 2 @p :
@ @1 @2 @1 @2 @1 @2
1 and p follows, again neglecting the products of rst derivatives @ 1 1 @ 1 = 1 1 1 @1@2 ; @1 @2 @p 1 @1 0 1 1 1 1 @1 = 1 @1 0 pp 1 @1 : = 1 @ @ 13 @1 @2 @1 @2 1 @1@2 @1 @2 1 2 After substitution in the expression for 1R0 and some algebra one obtains @ @ @ cos 1R0 = sin 1 @1 + 1 0 1 1 p 2 @1 + 1 0 sin 1 pp 1 @1 : 1 1@2 1 1 @2 1 @2 @ 1 Note that 1R ! when 1 ! 0. For the updated hourglass tensor one readily obtains
From the denition of
@1 @2
+1t #
t
ij
=~ t t = 1R (~ = 1R (~ t
t i
~ t
+1t 1
t t
"
t j
+1t
@2
t j
t j t j
t i t i
t t
t k
t kj
t k
t kj
t i
t j
t ij :
This expression simplies further with the introduction of the hourglass vector
#i
= 01e 2
ijk #jk ;
= 1R0 1 ~ +1 + # : t
t i t t i
The rst and second variation are obtained in entirely the same way as the rst and second variation of the curvature change. For the rst variation we neglect terms of order 0 and obtain
1R
2
:
#i
3.6.524
For the second variation we ignore in addition the terms of order d 0 and 0 with as nal result
d#i = 0:
Degenerate elements
In general meshes it will be desirable to collapse at least some of the quadrilateral elements to triangles or to use the triangular element S3 or S3R, which is in fact an internally collapsed S4R element. For this case the calculation of the membrane strains and the curvature changes proceeds along the same lines as before. The transverse shears will now be zero at the degenerate edges. Finally, calculation of all hourglass constraints will be omitted.
Rotary inertia scaling for explicit dynamics
For numerical efciency in explicit dynamic analysis, it is desirable to have the stable time increment determined by the membrane response of the structure. For this reason scaling of the rotary inertia based on the elements reference geometry is included in ABAQUS/Explicit. In explicit dynamic analyses the stable time increment is proportional to the inverse of the highest frequency of the element. Therefore, we must ensure that the highest frequency associated with the transverse shear response does not exceed the highest frequency associated with the membrane response. For thick elements (that is, for elements whose thickness is order unity relative to a characteristic length in the element), the membrane frequencies are dominant. The primary consideration in choosing appropriate scalings is that in the limit as the elements thickness goes to zero, the transverse shear frequencies remain below the membrane ones. Recall that for a onedimensional spring-mass oscillator, the natural frequency ! can be written in terms of the stiffness K and the mass M as
!=
Elements
rK
For the transverse shear response the rotary inertia, which is proportional to the cube of the thickness, plays the role of the mass of the system. All other quantitiesthe membrane stiffness, the mass associated with membrane deformation, and the transverse shear stiffnessare proportional to the thickness. Hence as the thickness of the element goes to zero, the frequencies associated with transverse shear go to innity proportional to the inverse of the thickness, while the membrane frequencies remain constant. Without scaling, the stable time increment would go to zero as the thickness becomes small.
Rotary inertia scaling
For thin elements the rotary inertia is small (negligible) relative to the rotational inertia of the mass at the nodes rotating about an axis through the element center. Therefore, we choose a scaling on the rotary inertia such that it never becomes smaller than a xed (small) percentage of the rotational inertia of the mass at the nodes rotating about an axis through the center of the element. Let R be the nondimensional rotary inertia scaling, where R 1. When R = 1, the true rotary inertia is used. Consider a lumped mass matrix for a 4-node element, and let the element be at and
3.6.525
square. For rotations about an axis in the plane of the element, parallel to an element edge, passing through the center, the contribution to the rotational inertia of the mass at the nodes is
1
4
2 AhL ;
where A is the area of the element, L is the edge length, and is the mass density. The sum of the rotary inertia at the four nodes is
RA h3
12
For the rotary inertia to remain a xed fractionsay of the mass contribution as the thickness goes to zero, asymptotically R must be proportional to L2=h2 ; that is,
R
! hL2
3 2
For planar geometries with element directors along the normal direction, closed-form expressions are possible for the highest membrane and transverse shear frequencies. In such cases the length parameter L is interpreted as a characteristic element length that depends on the element distortion. To handle arbitrarily shaped curved elements, exact calculation of the element frequencies becomes difcult. However, we can safely bound the frequencies by an appropriate choice of L in the following scaling:
"
R=
16 +
3L2 4 # 1 4
h2
where = 1=4. For triangular elements the characteristic length, L; is the minimum element edge length; and for quadrilateral elements,
L=
The factor 16 in the denition of R is used to protect against bending frequencies determining the stable time increment in very ne meshes subjected to loads that cause an increase in thickness of the shell.
Rotational bulk viscosity for explicit dynamics
For the displacement degrees of freedom, bulk viscosity introduces damping associated with volumetric straining. Linear bulk viscosity or truncation frequency damping is used to damp the high frequency ringing that leads to unwanted noise in the solution or spurious overshoot in the response amplitude.
3.6.526
For the same reason, in shells we need to damp the high frequency ringing in the rotational degrees of freedom with linear bulk viscosity acting on the mean curvature strain rate. This damping generates a bulk viscosity pressure moment, m, which is linear in the mean curvature strain rate:
m=b
h2 0
12
cd L
1 ; 1t
where b is a damping coefcient (default = 0.06), h0 is the original thickness, is the mass density, cd is the current dilatational wave speed, L is the characteristic length used for rotary inertia scaling, 11 22 is twice the increment in mean curvature. The dilatational wave speed is and given in terms of the effective Lams constants as
1 = 1 +1
cd =
( + 2 ) :
The resultant pressure moment mh, where h is the current thickness, is added to the direct components of the moment resultant.
Fully integrated nite-membrane-strain shell formulation
Elements
Element S4 is a fully integrated nite-membrane-strain shell element. Since the elements stiffness is fully integrated, no spurious membrane or bending zero energy modes exist and no membrane or bending mode hourglass stabilization is used. Drill rotation control, however, is required. Element S4 uses the same drill stiffness formulation as used for element S4R. Similarly, element S4 assumes that the transverse shear strain (and force, since the transverse shear treatment is elastic based on the initial elastic modulus of the material) is constant over the element. Therefore, all four stiffness integration locations will have the same transverse shear strain, transverse shear section force, and transverse shear stress distribution. The transverse shear treatment for S4 is identical to that for S4R. It is well known that a standard displacement formulation will exhibit shear locking for applications dominated by in-plane bending deformation. However, a standard displacement formulation for the out-of-plane bending stiffness is not subject to similar locking response. Hence, S4 uses a standard displacement formulation for the elements bending stiffness, and the theory presented above for the rotation kinematics and bending strain measures applies to S4. The primary difference between the element formulations for S4 and S4R is the treatment of the membrane strain eld. This formulation is the topic of the following discussion. The membrane formulation used for S4 does not rely on the fact that S4 is a shell element. Hence, the discussion below details the formulation from the point of view that the membrane response is governed by the equilibrium for a three-dimensional body in a state of plane stress. to the rate of deformation tensor . We introduce the enhanced Consider an enhancement rate of deformation tensor, , as
D
~ D
where
~ D is dened subsequently.
~ = D def D + D ;
3.6.527
where
~ = D def D + D ;
@ x @x
D def sym =
Z
:
V
Z
3 D 0 D : dV =
D
0 D 0
Z
: dV = 0
so that the modied virtual work statement can be written in the form
V
(D) : D dV +
b 1 x dV
ST
t 1 x dST = 0;
3
where t is the specied traction on ST and x = 0 on Su . is an arbitrary stress eld, and the constitutive equation = (D) is enforced pointwise. In the modied virtual work statement all kinematic quantities and corresponding variations (D, D, D, and D) are known functions of x, x, and the reference conguration. A fundamental requirement for the validity of the formulation is that the modied virtual work statement leads to the 3 proper equilibrium equations. If is arbitrary, the constraint equations can be rewritten as
D = D and
D = D :
Substituting these two relations in the modied virtual work equation yields
Z
: D dV +
b 1 x dV
ST
t 1 x dST = 0
where we have used the constitutive equation = (D). We recognize this variational statement as the usual virtual work equation, and a straightforward application of the divergence theorem leads to the standard equilibrium equations. In the actual implementation we choose to satisfy the constraints only for piecewise constant 3 stress elds . Hence, over the element domain Ve we require
Z
Ve
~ D dV =
Ve
~ D dV = 0:
~ ~ The enhancements D and D are chosen such that they eliminate the shear locking for in-plane bending. In addition, the direct strain eld is enhanced to approximate the strains caused by Poissons effect in bending.
3.6.528
Patch test
To pass the patch test, the choice of enhancements ~ cannot be arbitrary. A sufcient condition or for the satisfaction of the patch test is that for homogeneous deformations we have = ~ = 0 pointwise. In that case (D) = (D) and the stress is homogeneous. Since the stress is homogeneous, it can be moved outside the volume integral in the modied virtual work statement. ~ The volume integral condition on D implies that the expression is independent of the enhancement and leads to the standard displacement formulation, which is known to satisfy the patch test.
References
Shell elements: overview, Section 15.6.1 of the ABAQUS Analysis Users Manual Shell section behavior, Section 15.6.4 of the ABAQUS Analysis Users Manual
Elements
3.6.529
3.6.6
The small-strain shell elements in ABAQUS/Explicit use a Mindlin-Reissner type of exural theory that includes transverse shear and are based on a corotational velocity-strain formulation described by Belytschko et al. (1984, 1992). A corotational nite element formulation reduces the complexities of nonlinear mechanics by embedding a local coordinate system in each element at the sampling point of that element. By expressing the element kinematics in a local coordinate frame, the number of computations is reduced substantially. Therefore, the corotational velocity-strain formulation provides signicant speed advantages in explicit time integration software, where element computations can dominate during the overall solution process.
Corotational coordinate system
The geometry of the shell is dened by its reference surface, which is determined by the nodal ^ coordinates of the element. The embedded element corotational coordinate system, x, is tangent to the reference surface and rotates with the element. This embedded corotational coordinate system serves as a local coordinate system and is constructed as follows:
^ 1. For the quadrilateral element the local coordinate x1 is coincident with the line connecting the ^ ^ midpoints of sides, rac , as shown in Figure 3.6.61. The x1 x2 plane is dened to pass through this line normal to the cross product rac 2 rbd . ^ 2. For the triangular element the local coordinate x1 is coincident with the side connecting nodes 1 ^ ^ and 2 as shown in Figure 3.6.61. The x1x2 plane coincides with the plane of the element. ^ For notational purposes the corotational coordinate system is dened by a triad ei , and any vector or tensor whose components are expressed in this system will bear a superposed hat. Although the corotational coordinate system described here is used in the actual element computations, this system is transparent to the user. All reported stresses, strains, and other tensorial quantities for these shell elements are dened with respect to the coordinate system described in Finite-strain shell element formulation, Section 3.6.5.
Elements
The velocity of any point in the shell reference surface is given in terms of the discrete nodal velocity with the bilinear isoparametric shape functions N I () as
vm
= N ( )v ;
= N I ( ) I ;
where vI and I are the nodal translation and rotation velocity, respectively. The functions N I ( ) are C continuous, and are nonorthogonal, nondistance measuring parametric coordinates. Here Greek subscripts range from 1 to 2, and uppercase Roman superscripts denote the nodes of an element.
3.6.61
^ x3 ^ e2 ^ e3 4 a 1 ^ x3 ^ e3 ^ e2
^ x2 3 d ^ e1 c ^ x1
b ^ x2 3
2 ^ e1
^ x1
Figure 3.6.61 Local coordinate system for small-strain quadrilateral and triangular shell elements. A standard summation convention is used for repeated superscripts and subscripts except where noted otherwise. In the Mindlin-Reissner theory of plates and shells, the velocity of any point in the shell is dened ^ ^ by the velocity of the reference surface, vm , and the angular velocity vector, , as
v = v m 0 x 3 e3 2 ; ^ ^ ^^ ^
where 2 denotes the vector cross product and x3 is the distance in the normal direction through the ^ thickness of the shell element. The corotational components of the velocity strain (rate of deformation) are given by
dij
^ =1 2
d1
@ vi
@ x1 j
^ ^ @v ^ + @ x1 i ^
j
which allows us to write each velocity strain component in terms of the nodal translational and rotational velocities:
^ ^ ^ ^ = @ v 1 + x3 @ 1 ; @ x1 ^ @ x1 ^
m
3.6.62
^ d2 = ^ d12 =
@ vm ^ @ x2 ^
1
@ x2 ^
@ 0 x3 @ x2 ; ^ ^
2
+ x3 ( ^ ^ @
^ d23 = ^ d13 =
@ vm ^ @ v3 ^m @ x1 ^ @ x2 ^
3
@ x2 ^
@ 0 @ x1 ); ^
1
^ 0 1 ; ^ + 2 :
The S4RS element is based on Belytschko et al. (1984). By using one-point quadrature at the center of the elementi.e., at =0we obtain the gradient operator
I B
1 I B2
1 2A
y2 ^ x4 ^
0 y4 ^ 0 x2 ^
y3 ^ x1 ^
0 y1 ^ 0 x3 ^
y4 ^ x2 ^
0 y2 ^ 0 x4 ^
y1 ^ x3 ^
0 y3 ^ 0 x1 ^
I ^I ^ I ^I ^ d 1 = B 1 v1 + x 3 B 1 2 ;
^ ^ ^ ^ ^ 2d12 = B2 v1 + B1 v2 + x3 (B2 2 ^ 2d13 = B1 v3 + ^ ^ ^ 2d23 = B2 v3
Elements
I I
I ^I ^ d 2 = B 2 v2
I I
I ^I 0 x 3 B 2 1 ; ^
1 4
I I
I ^I 0 B 1 1 );
I I I I
^I S I 2 ;
^I 0 1 S I 1 ; 4
where
S I = [1; 1; 1; 1]:
The local nodal forces and moments are computed in terms of the section force and moment resultants by
I^ I^ ^I f1 = A B1 f1 + B2 f12 ; I^ I^ ^I f2 = A B2 f2 + B1 f12 ;
0A
0 S I f^23 4
I^ I^ B1 m1 + B2 m12
0 S I f^13 4
; ;
3.6.63
mI ^3
= 0:
m ^
h Z
dz ; ^ ^ z dz; ^^ ^
where A is the area of the element, h is the thickness, and are the Cauchy stresses computed in ^ the corotational system from the velocity strain and the applicable constitutive model. Although is the shear factor in classical Mindlin-Reissner plate theory, it is used here as a penalty parameter to enforce the Kirchhoff normality condition as the shell becomes thin.
Small-strain element S4RSW
The major objective in the development of the S4RS element was to obtain a convergent, stable element with the minimum number of computations. Because of the emphasis on speed, a few simplications were made in formulating the equations for the S4RS element. Although the S4RS element performs very well in most practical applications, it has two known shortcomings: 1. It can perform poorly when warped, and in particular, it does not solve the twisted beam problem correctly. 2. It does not pass the bending patch test in the thin plate limit. In the S4RSW element additional terms are added to the strain-displacement equations to eliminate the rst shortcoming, and a shear projection is used in the calculation of the transverse shear to address the second shortcoming. The components of velocity strain in the S4RSW element are given in Belytschko et al. (1992) as
^ d1 ^ d2 = =
I ^I B 1 v1
+ x 3 (B 1 v 1 + B 1 p 1 ); ^ ^ _
cI I
I I
z
I hI
= 1 4
I xI ; ^3
[h
I 0 B I x J hJ ]; ^
= [+1;
3.6.64
The pseudonormal pI represents a nodal normal local to a particular element found by taking the vector cross product of the adjacent element sides. The components of the transverse shear velocity strain are given by
I 13 = N I ( )2 ; ^ I I where nodal rotational components 1 and 2 are based on a projection and a transformation. Consider three adjacent local element nodes K , I , and J as shown in Figure 3.6.62. Outward facing vectors eK and eI are constructed perpendicular to element sides KI and IJ , respectively. In addition, they n n are tangent to the reference surface at the midsides. I 23 = 0N I ( )1 ; ^
^ x2 K LK LI
K en I en
Elements
J I side I
^ x1
side K
Figure 3.6.62 Vector and edge denition for shear projection in the element S4RSW.
I The angular velocity n about outward facing vector eI is then given by a nodal projection n I n =
1 I 1 J J ( + nI ) + I (^3 v 2 nI L
0 v3 ) ; no sum on I; ^I
I J where nI is the rotational velocity at node I about eI , nI is the rotational velocity at node J about n I I eI , and LI is the length of side I . Finally, the nodal rotational components 1 and 2 required for n the transverse shear velocity strain are given by the transformation I K I 1 = (eI 1 e1 )n + (eK 1 e1 )n ; n ^ n ^ I K I 2 = (eI 1 e2 )n + (eK 1 e2 )n : n ^ n ^
Evaluating the resulting forms for the transverse shear at the centroidal quadrature point gives
3.6.65
sI B
1i sI B 2i
where
1 2(JI x1 x2 4 2(JI
0 xIK ) 1 0 xIK ) 2
no sum on I; J; and K;
x = xJ ^JI ^
and
xJI =
(L I )
xJI ^
2;
no sum on I:
The local nodal forces and moments are then given in terms of the section resultant forces and moments by I^ I^ cI ^ cI ^ ^I f1 = A B1 f1 + B2 f12 + B1 m1 + B2 m12 ;
I^ I^ cI ^ cI ^ ^I f2 = A B2 f2 + B1 f12 + B2 m2 + B1 m12 ;
0A
The triangular shell element formulation is similar to that of the S4RS element and is based on Kennedy et al. (1986). This element is not subject to zero energy modes inherent in quadrilateral element formulations. The velocity strain is computed as in the S4RS element except that the gradient operator is given by I 1 B1 ^ 0 y3 y3 0 : ^ = I B2 ^ ^ ^ ^ 2A x 3 0 x 2 0 x3 x 2 The local nodal forces and moments for the triangular shell can be expressed in terms of section resultant forces and moments as
0 f^23 3
3.6.66
mI ^2
= 0A
I I B1 m1 + B2 m12 0 ^ ^
mI ^3
= 0:
^ f13 ;
The x3 -components of the nodal forces are obtained by successively solving the following ^ equations: m 1 + m 2 + m 3 + x 3 f3 = 0 ; ^ 1 ^ 1 ^ 1 ^ ^3
m 1 + m 2 + m 3 0 x 3 f3 0 x 2 f3 ^ 2 ^ 2 ^ 2 ^ ^3 ^ ^2
which represent the equations of moment equilibrium about the x1-axis, moment equilibrium about ^ ^ ^ the x2-axis, and force equilibrium in the x3-direction.
Hourglass control
^1 ^2 ^3 f3 + f3 + f3
= 0;
= 0;
Since the one-point quadrature is used, several spurious modes, often known as hourglass modes, are possible for the quadrilateral elements. To suppress the hourglass modes, a consistent spurious mode control as described by Belytschko et al. (1984) is used. The hourglass shape vector I is dened as
I = hI 0 B xJ hJ : ^
Elements
q _B q3 _B
^I = I ;
=
I v3 ; ^I ^I q =
I v ; _M
where the superscripts B and M denote hourglass modes associated with bending and in-plane (membrane) response, respectively. The corresponding generalized hourglass stresses for the element S4RS are given by h3EA I I B _ Q B =
r sr B B q ; _
192
3
_3 QB
GA I I B = w sw h12 B B q_3 ;
_ QM
where h is the thickness of the shell and E and G are Youngs modulus and shear modulus, respectively. The default hourglass control parameters are s = r =0.050 and w =0.005. The scaling factors sr , sr , and sw (by default sr =sr =sw =1) are used to change the corresponding default hourglass control
I I M = sss hEA B B q_ ; 8
3.6.67
_3 _ parameters by the user. For the S4RSW element the generalized hourglass stresses QB and QM are _ the same as those in the element S4RS, but the generalized hourglass stress QB is modied to
3 I I M _ B =
r sr h EA 1 + 2A B B q_ : Q 192 3h2
mhI ^
The nodal hourglass forces and moments corresponding to the generalized hourglass stresses are
= I QB ;
=
I QB ; 3 hI ^ =
I QM : f
These hourglass forces and moments are added directly to the local nodal forces and moments described previously.
Reference
^hI f3
Choosing a shell element, Section 15.6.2 of the ABAQUS Analysis Users Manual
3.6.68
3.6.7
The ABAQUS/Standard element library includes a family of nonlinear thin shell elements with axisymmetric reference geometry that allow asymmetric loading and deformation (SAXA1N and SAXA2N). This section provides their theoretical formulation. These elements encompass a broad range of practical applications from the bending/ovalization of variable diameter pipes to the bending of circular plates. The theoretical formulation of these elements is similar to the general nite-strain shell element described in Finitestrain shell element formulation, Section 3.6.5. Furthermore, this formulation is the shell counterpart to the continuum axisymmetric bending elements described in Axisymmetric elements allowing nonlinear bending, Section 3.2.9. As with the continuum axisymmetric bending formulation, the restriction is made that a plane of symmetry exists in the rz plane at = 0. Hence in-plane bending of the model is permitted, while deformations such as torsion about the axis of symmetry are precluded. The symmetries of the undeformed conguration and of the deformation are exploited through the assumption of particular displacement and rotation interpolations around the circumference of the shell. Specically, Fourier series expansions are used in the or circumferential direction that preserve the plane of symmetry.
Geometric description
Let (S; ) be coordinate functions parametrizing the reference surface of the shell and let S3 2 0h=2; h=2] be the coordinate function in the thickness direction, where h is the shells initial thickness. (For a detailed account of the geometric description of the nite-strain shell formulation, see Finite-strain shell element formulation, Section 3.6.5.) Then points in the reference or undeformed conguration are identied by the normal coordinates mapping
[
Elements
where X is the three-dimensional position of a material point, X is the shell reference surface mapping, and T3 is the unit normal to the shell reference surface. The fact that T3 is a unit vector assumes that the reference conguration is (locally) of constant thickness. Owing to the axisymmetric reference conguration, X can be given relative to a global Cartesian coordinate system as
8 X 1(S; ) 9 8 r(S) cos 9 < = < = X(S; ) = : X 2 (S; ) ; = : r(S ) sin ; ; X 3 (S; ) z (S )
where r(S ) is the radius, z (S ) is the axial position, and (r; z; ) are the cylindrical coordinates. (Note that the usual convention for cylindrical coordinates (r; ; z ) has been changed, which is consistent with the axisymmetric shell elements and the axisymmetric elements allowing nonlinear bending.) By denition the normal eld to the shell reference surface is T3 = X;S 2 X; =kX;S 2 X; k, which by direct computation yields
with
3.6.71
where r0
0
01 [(r )2 + (z 0 )2 ] 2 ( z 0 er + r 0 ez ).
@r=@S and z 0
The basic kinematic assumption is that for any deformed conguration, the position of a point in the body can be identied by
@z=@S .
T3 =
(3.6.71)
where x is the deformed position of the material point, x is the deformed shell reference surface mapping, t3 is the deformed unit director eld, and f 33 is the thickness change parameter. Of critical importance for any shell formulation is the treatment of the rotation eld; that is, the treatment of the director eld t3. The geometric description and the incremental update procedure for the director eld are given in detail below. Under the kinematic assumption above, the deformed conguration of the shell is completely determined by the reference surface mapping x, the deformed director eld t3 and the thickness parameter f 33. We dene the following displacement quantities. Since x is an element of a (linear) vector space, we can dene the reference surface displacement vector u by the difference between the deformed reference surface and the undeformed reference surface; i.e.,
0 and
^
and
^ exp[]
k ^ k kI + sink k + (1 0kcos2kk) : k k
Alternatively, quaternion algebra can be used to specify the orientation of the deformed director eld ^ t3. In this case the orthogonal matrix exp[] is replaced by the quaternion parameter q = (q0 ; q), where sin k=2k q0 = cos k=2k and q = :
kk
3.6.72
^ exp[]
^ exp[ ] = (2q0
Interpolations
Displacement and rotation components are given relative to the cylindrical coordinate system (r; z; ) with orthonormal basis vectors fer ; ez ; e g that are xed in the reference or undeformed conguration. A general interpolation scheme for u = ur er + uz ez + u e and = r er + z ez + e using a Fourier expansion in the variable is
ui (S; ) = i (S; ) =
X
M
H (S ) H m (S )
m=1 M
m=1
!9 > > > = p=1 !> P X mp mp > m0 + (cos(p )ic + sin(p )is ) > i ; p=1
u m0 + i
P X mp mp (cos(p )uic + sin(p )uis )
(i = r; z; ) :
Here H m (S ) are the polynomial interpolation functions along the generator lines of the axisymmetric reference conguration; um0 , ump , ump , m0 , mp , mp are the solution amplitude values (Fourier i i ic is ic is coefcients); M is the number of terms used in the interpolation along the generator lines; and P is the number of Fourier interpolation terms used around the circumference of the reference shell. Note that an axisymmetric deformation is obtained for the choice P = 0. The symmetry requirement in the rz plane at = 2n, n = 0; 1; . . ., eliminates many of the above Fourier coefcients. For the displacement vector the only admissible terms are
Elements
8 mp 9 8 91 8 9 0 8 m0 9 P M P < ur = X < urc = X < 0 = < ur (S; ) = X m A uz (S; ) H (S ) @ u m 0 + cos(p ) ump + sin(p ) 0 = : z ; p=1 : zc ; p=1 : ump ; : : u (S; ) ; m=1 0 0
s
For the rotation components, symmetry requirements switch the role of the r and z components with the components:
8 8 9 8 mp 91 9 08 9 P M P < r (S; ) = X m < 0 = X < 0 = X < rs = mp A z (S; ) = H (S ) @ 0 cos(p ) 0 sin(p ) + + : (S; ) ; m=1 : m0 ; p=1 : mp ; p=1 : zs ; : 0
c
For practical reasons the values of ur , uz , and are often required at specic locations around the circumference of the shell. Therefore, displacement and rotation components ump , ump , and mp are r z used instead of the Fourier coefcients ump , ump , and mp . Furthermore, a negative sign is introduced rc zc c in the interpolation for for the following reason: The ABAQUS convention for axisymmetric shell elements is that the axial tangent direction is drawn between nodes in ascending node number (the shell local 1-direction). The normal to the shell is then obtained by a 90 counter-clockwise rotation of the tangent (the shell local 3-direction). However, a positive rotation of the normal eld (about
3.6.73
the shell local 2-direction) is counterclockwise. This convention implies a left-handed shell local coordinate system. For the axisymmetric shell bending elements, a right-handed shell local coordinate system is required at the integration points; thus, the direction of positive rotation is reversed there. Rearranging the Fourier series expansions and making the substitution ump 7! ump , the s interpolations for the displacement components are
8 9 8 91 8 9 0 M P +1 P < 0 = < ur (S; ) = X r X p < ump = X A uz (S; ) H m (S ) @ R () ump + sin(p ) 0 = z : 0 ; p=1 : ump ; : : u (S; ) ; m=1 p=1
(3.6.72)
Similarly, replacing mp and mp with mp and mp respectively, the interpolation for the rotation rs zs r z components becomes
8 8 mp 91 8 9 9 0 M P +1 P < r = < r (S; ) = X m X p < 0 = X mp A z (S; ) = H (S ) @ R ( ) 0 sin(p ) + : 0mp ; p=1 : z ; : : (S; ) ; m=1 p=1 0
(3.6.73)
and are physical displacement and rotation components trigonometric interpolation functions with the property that
mp
P = 1: 1 = 0, 2 = ,
dened by:
R 1 ( ) = R 2 ( ) =
1 2 1 2 (1 + cos ) (1
0 cos )
P = 2: 1 = 0, 2 = =2, 3 = , R 1 ( ) = R 2 ( ) = R 3 ( ) =
1 4 1 2 1 4
(1 + 2 cos + cos 2 ) (1 (1
P = 3: 1 = 0, 2 = =3, 3 = 2=3, 4 = , R 1 ( ) = R 2 ( ) = R 3 ( ) =
1 6 1 3 1 3 1 (1 + 2 cos + 2 cos 2 + cos 3 ) (1 + cos (1
with respect to . The element becomes signicantly more expensive as higher-order interpolations are used, and it is assumed that the general-purpose nite-strain shell is less expensive than using this element with P > 4.
Virtual work
= 4: 1 = 0, 2 = =4, 3 = =2, 4 = 3=4, 5 = , 1 R1 ( ) = (1 + 2 cos + 2 cos 2 + 2 cos 3 + cos 4) 8 1 (1 + p2 cos 0 p2 cos 3 0 cos 4) 2 R ( ) = 4 1 (1 0 2 cos 2 + cos 4) 3 R ( ) = 4 1 (1 0 p2 cos + p2 cos 3 0 cos 4) 4 R ( ) = 4 1 (1 0 2 cos + 2 cos 2 0 2 cos 3 + cos 4) 5 R ( ) = 8 As with the continuum axisymmetric bending element, P = 4 is the highest-order interpolation offered
P
5=
Z
V
Elements
ij Eij dV ;
where V is the current volume of the deformed body, ij are the curvilinear components of the Cauchy stress tensor, Eij are the components of the Lagrange strain tensor, and Eij are the variational or linearized strain measure components. By denition the Lagrange strain tensor components are given by
Eij
def 1 2
@ i
@x
@ @X 1 @xj 0 @ X 1 @ j @ i
Note that in the statement of virtual work, no choice has thus far been made regarding the curvilinear coordinate functions i (i = 1; 2; 3). Furthermore, the current volume measure dV is given by the parametric relationship
dV
= jd 1 d 2d 3 ;
where
def
We now introduce the kinematic assumption Equation 3.6.71 into the denition of Eij to nd
E
= 1 (x; 1 x; 0 X; 1 X; ) + 1 S3 [x; 1 (f 33t3); + x; 1 (f 33t3); 2 2 0 X; 1 T3; 0 X; 1 T3;] + 1 (S3 )2 [(f 33t3); 1 (f 33t3); 0 T3; 1 T3; ] ; 2 E3 = 1 (f 33x; 1 t3 + S3 f 33f 33;) ; 2 E33 = 1 [(f 33)2 0 1] ; 2
3.6.75
where differentiation is now with respect to the parametric coordinates, so that ; 2 fS; g and 3 = S3. Dene the following shell strain measure components and kinematic relationships:
def 1 (x; 1 x; 0 X; 1 X; ) =2 def 1 = 2 (x; 1 t3; + x; 1 t3;) 0 b0 def
= x; 1 t3 def ln(f 33) =
def
In the above, b0 = X; 1 T3; = b0 are the components of the second fundamental form of the undeformed reference surface. Substituting the above denitions into the virtual work expression, we nd (after some manipulation) that the volume integral reduces to the following integral over the deformed reference surface
5 =
where = f 33=f 33 and the current reference surface Jacobian determinant is j = kx;1 2 x;2k. In the above virtual work expression, the (S3 )2 term in E has been neglected. This term is O(h2=R2) where h is the thickness and R is some characteristic radius of curvatureand is negligible in light of the kinematic assumption Equation 3.6.71. The shell stress resultant components are dened by the following integrals through the thickness of the shell:
N + M + Q + L3 + M 3(); jdSd ;
1 Z S 03 f +
1 jdS : M 3 = f 33 3 3 33
j h
33
j h
33
33;
+ S3 ( + ; ) jdS3 ;
plane stress assumption 33 = 0. Consistent with these approximations, we neglect all O(h=R) terms and terms proportional to the gradient of the thickness parameter. Accordingly, we set
t3 is (approximately) the normal eld to the deformed reference surface, is introduced along with the
= 0 ; j = f 33 j
and
For thin shells the Kirchhoff-Love approximation, which states that the deformed director eld
f 33; = ; = 0 :
We can now summarize the virtual work expression for thin (Kirchhoff-Love) shells:
3.6.76
5 =
Z
A
N + M
+ Q
jdSd ;
(3.6.74)
where the shell resultant components are dened in terms of the Cauchy stress tensor components by the integrals
= f 33
dS3
R
and M
= (f 33 )2
S3 dS3 :
In the expression for 5, Q = (f 33 )2 h 3 dS3 is interpreted as a constraint stress that enforces that the director eld remain normal to the reference surface. Two other contributions to the virtual R work expression M and M 3();, where M 3 = (f 33 )2 h S3 3 dS3 are O(h=R) and, thus, neglected.
Orthonormal surface coordinate system and coordinate transformation
It is desirable to dene stress resultant quantities relative to an orthonormal basis in the deformed conguration. To do this, we dene a normal coordinate system fn1 ; n2; n3g, where n1 and n2 are tangent to the deformed reference surface and n3 is the unit normal eld. a Dene the following notation. Let f (a = 1; 2) be the components of x; relative to the basis fn1; n2g; that is
Elements
@x @
= f a na :
a
and ha f
(3.6.75)
Furthermore, let the inverse of the matrix of components [f ] be given by [ha ], such that
a f hb
a = b
= :
Note that the basis vectors n1 and n2 induce distance measuring coordinates s1 and s2 such that
a f
@s = @
and
ha
= @ a : @s
It follows from Equation 3.6.75 that the orthonormal tangent vectors are given by
na = ha x; =
@x : @sa
For the material calculations, it is important to express both the strain and stress quantities relative to the local orthonormal frame fn1; n2g. Accordingly, let N ab = N ba and M ab = M ba be the membrane and bending stress resultant components relative to this local orthonormal basis. Thus, we can write N = N abha hb and M = M ab ha hb ;
3.6.77
where we recall that ha = @ =@sa . Then the stress resultant contributions to the virtual work expression can be transformed as follows. First, the membrane contribution:
def
N
= N x;
na. Thus,
na 1
@ ( x ) 3 @sb
and e2 1
3
0 e1 2
= 1.
Then
=M
@ t3 @sb
( 1 ( @@sx) 0 2 na ) a
Since the second term in the brackets is proportional to the bending curvature, we neglect this term relative to the rst, yielding
c M = M ab sym ea nc
Strain displacement operators
( 1 @@s) b
We now write the virtual work expression Equation 3.6.74 in matrix operator notation. Dene the following stress resultant component vectors:
def
8 9 11 <N = = N 22 : 12 ; N
and
def
8 9 11 <M = = M 22 : : ; M 12
5=
Z
A
N 1 BN
+ M 1 BM
x
jdSd :
Thus far n1 and n2 are any two orthonormal vectors in the tangent plane to the reference surface. We can uniquely choose these two vectors by requiring the matrix components of the incremental a reference surface deformation gradient f b , where
[1 ]
(t+1t)a
1 1=1
(t+1t)a
=1
a (t)b
Elements
(3.6.76)
(t+1t)
x x (cos(1 )^1 + sin(1 )^2) 1 @@s(t)2 = (0 sin(1 )^1 + cos(1 )^2) 1 @@s(t)1 t t t t
From this it can be determined that
t+1t
= tan01
"
1f a 1f ab = n(at+1t) 1 @ x (t)b b @s 1 =1
ha
(t+1t)
and
(t+1t)a
( = 1f af t)b : b
Note that due to the symmetry of the incremental deformation gradient components, there is no a f ab . We can now calculate the inverse components ambiguity in writing f b
(t+1t)
(t+1t)a i01
Consistent linearization
The iterative solution procedure requires the calculation of the consistent tangent stiffness. This stiffness has two parts, one resulting from the material model and the other resulting from the changing geometry. We denote the second variational quantities with . Returning to Equation 3.6.74, the virtual work expression can be written as
5=
J ab Eab dV 0 =
Z Z
A h
where J j=j 0 , j 0 is the Jacobian determinant of the reference conguration parametrization given r , and components are relative to the corotated frame with a; b 2 f ; g. We assume by j 0 constitutive behavior such that
= = det[ X]
12
d(5) =
Z hZ j0
The rst term in the integrand forms the material stiffness, while the second two terms form the geometric stiffness. The second variation of the strain measure components are calculated in Finitestrain shell element formulation, Section 3.6.5. The second variation of the membrane strain is
h ( i (d d(ab) = sym @@sx) 1 @@sx) 0 2acdcb : a b
d(ab) = 0 :
3.6.710
k At the beginning of an increment we have the conguration at iteration k, denoted xk ; 3 . In the incremental solution procedure we solve for the incremental displacement eld and the incremental rotation eld in components relative to the reference cylindrical coordinate system: mp mp (1ump; 1uz ; 1u ; 1mp; 1mp; 1mp) : r r z
We use these incremental elds to update the conguration to iteration k . 1. Reference surface update: The displacement increments are interpolated using the same interpolation scheme as the total displacement vector in Equation 3.6.72:
( t)
+1
8 8 8 91 9 M 9 P 0 < 1ur (S; ) = X m P +1 p < 1ump = X < 0 = r X 1u(S; ) = : 1uz(S; ) ; = H (S) @ R () : 1ump ; + sin(p) : 0mp ;A : z p=1 p=1 1u(S; ) m=1 1u 0 xk+1(S; ) = xk (S; ) + 1u(S; ) :
The reference surface position map is updated from the displacement increment by
Elements
2. Rotation eld update: The incremental rotation eld is updated with the same interpolation scheme as the total rotation eld in Equation 3.6.73:
8 1 (S; ) 9 M 8 8 1mp 91 9 P < r = X m 0P +1 p < 0 = X < r = X 1(S; ) = : 1z (S; ) ; = H (S) @ R () : 0 mp ; + sin(p) : 1mp ;A : z p=1 p=1 1(S; ) m=1 01 0
1 = cos 1 2 1 1q = sink1k 2 1
This incremental rotation vector corresponds to a nite rotation, characterized by quaternion parameters as k = k : q0 k = k and The total rotation quaternion parameter can be updated by the update formula
Similarly, the deformed unit director eld can be updated from the incremental rotation eld as
tk+1 = (2(1q0)2 0 1)tk + 21q01q 2 tk + 2(1q 1 tk)1q def (1q0; 1q) tk : = 3 3 3 3 3 Here we have used the notation (q0; q) v to denote the rotation of the vector v by the quaternion (q0; q). The curvatures are calculated from the gradient of the director eld, which is updated by +1 tk; = (1q0; 1q) tk; + 1R 2 tk+1 ; 3 3 3
3.6.711
where
1R = H(1) 1; ; sin k1k I + 1 1 0 sin k1k 11 + (1 0 cos k1k) 1 : c H(1) = k1k k1k2 k1k k1k2
1
1;. First, along the generator lines we have 8 8 9 0 mp 91 M P +1 P 0 = X < < 0 1 r = X X m 1;S = H;S (S) @ Rp () : 0mp ; + sin(p) : 1mp ;A : z m=1 p=1 p=1 1 0
For the circumferential derivative we must account for the derivative of the basis vectors in the r; 0 r z; z ; r . direction: r; and ; 0 r . Hence, ; Introducing the interpolation function, we have
e =e
e = e
1 = (1
8 <
1 )e +1
e +(1 +1 )e
8 9
mp 1 P 0 = X < 1 r = 1; = : 0 ; + H m (S) @ Rp () : 0 mp ; + p cos(p) : 1mp ;A ; z ; m=1 p=1 p=1 1 r 01 0 where Rp () are computed from the denitions of the interpolation functions Rp () as given above ; and 1 and 1r are given by the interpolation for the incremental rotation eld as given in the M X
0
8 <
01 =
P +1 X
Following the formulation of the nite-strain shell element formulation in Finite-strain shell element formulation, Section 3.6.5, the three-dimensional (nite) strain increment is calculated as
t+1 1Eab = 1ab + S3f 33 t1ab : Here the increment in the membrane strain components 1ab is given by the Hughes-Winget second-
1ab = 2(1f ac 0 ac)(1f cb + cb)01 ; and the increment in the bending strain components 1ab is given by the expression t 1ab = sym2ec nc+1t 1 1rb3 ; a
where
1 rb = h 1 R : b
(t+1t)ab
As an example of the stress update procedure, consider the simple case of a Saint Venant-Kirchhoff material model. In this case,
(1
0 2)
ef gh
(1
0 ) 0 eg fh + eh fg 1 :
2
Note that a; b; c; d; e; f; h 2 f1; 2g, and the components (t+1t)ab are relative to the current t t orthonormal basis f1+1t ; n2+1tg.
Pressure loads and load stiffness
For geometrically linear problems, equivalent nodal loads due to applied surface pressure are readily calculated since the geometry is axisymmetric. For geometrically nonlinear problems, however, asymmetric deformations must be taken into consideration. The equivalent nodal loads associated with surface pressure p can be obtained by considering the virtual work contribution to the external loading
Z
A
pn xdA =
Z 2 Z
+1
01
p
@x
@x 2 @S 1 x dSd; @
where S is the parametric surface coordinate in the RZ plane and the reference surface position is
Elements
x = r e r + z e z + u e ;
with r = R + ur and z = Z + uz . Recall that the current position of a point can be expressed in terms of the axial interpolator H m (S ) and the circumferential interpolators Rp () by
8 9 < r = :
H m (S ) @ Rp () z mp z = + : ; ; m =1 p=1 p=1 u 0
M X
P X
+1
8 9 < rmp =
P X
8 91 < 0 = A: sin p 0 : mp ;
u
(3.6.77)
= =
0 u @
er +
@z @S @u @S
@r
er + ez +
@z
@r @S
@z @S
@ @u @S
ez + r + e ;
@u @
e ;
and, hence,
@x
@x 2 @S = @
+ +
r+
@u @
@r 0 u 0 @S @
@r
0 @u @z @S @
er
@u @
r+
@z 0 @S @S @
@r @z
@r @
0 u
ez
e :
3.6.713
The variations x are written x = xr er + xz ez + u e , where the components have interpolations similar to those in Equation 3.6.77. Therefore, the virtual work contribution becomes
Z
A
p n x dA =
2 Z
+1
01
@z
@u
@u @z
With the introduction of the interpolation functions, we obtain the equivalent nodal forces:
Z 2 Z +1
p H m (S )R p ( )
@z
r+
@u
01
For geometrically linear analysis, the equivalent nodal forces reduce to the standard axisymmetric expressions Z Z
mp Fr = mp Fz =
2
+1
Z0
0
01 2 Z +1 01
p H m (S )R p ( )
@Z @S
R dSd
F
mp
= 0:
The linearization of the pressure loading term leads to the following pressure load stiffness matrix:
mpnq
with
mpnq Krr = mpnq = Kzr
Z 2 Z +1 Z0
0
01 2 Z +1 01
p H m (S )R p ( ) p H m (S )R p ( )
+
@z
@u @S
H n (S )
@S
r+
@u @
dH n dS
R q ( )
3.6.714
mpnq Kr
2 Z
+1
mpnq = Krz
1 Z02 Z0+1 0
p H (S ) sin p p H (S )R ( )
m p
@z dH n
@ dS r+
R ( )
01
@u @
0 @S H
q
@z
(S )
dRq d
dSd
n
dH n dS
R ( )
@u @S
H (S )
dRq d
dSd
Z 2 Z +1
1 Z02 Z0+1 0
p H (S ) sin p
@r @S
H (S )
dRq d
@r @
0 u
dH n dS
Kr
mpnq
Kz
mpnq
Z 2 Z +1
0
01
p H m (S )R p ( ) q p H m (S )R p ( )
@z @S
H n (S ) cos q
01
K
mpnq
Z 2 Z +1
0
@r 0q @S H n (S ) cos q p H m (S ) sin p @z @S
0 @u H n (S ) sin q + @S
dSd
0 @z dH @ dS @r
R () dSd
sin q dSd
0 u @
dH n
dS
sin q
01
In the case of hydrostatic pressure (p dependent on z ), additional terms must be included in the pressure load stiffness. These terms appear due to the variation of the pressure magnitude and are readily obtained from the expression
Elements
dp
n 1 u dA =
Z 2 Z +1
0
01
du z
dp dz
@z
With use of the interpolation functions and denoting the additional contributions with an over-bar, we obtain the additional load stiffness contributions:
0 @S @ H n(S )Rq () dSd dp @u @r @r mpnq 0 u 0 @S r + @u H n (S )Rq () dSd K zz = H m (S )R p ( ) dz @S @ @ 1 @r Z0 2 Z0+1 @z dp @r @z mpnq 0 @S @ 0 u H n (S )Rq () dSd: K z = H m (S ) sin p dz @S @
mpnq K rz =
Z 2 Z +1 Z0
0
01 2 Z +1
H (S )R ( )
dp dz
@z
@S
r+
@u @z
01
It is necessary to enforce rotation constraints at selected points on the element surface to prevent singular modes of deformation. One axial transverse shear constraint is enforced on the rotation
3.6.715
eld between each pair of nodes within each nodal plane. One circumferential transverse shear constraint and one drill rotation constraint on the rotation elds r and z are enforced between each pair of nodes on a circumferential line of nodes. In each instance the rotation eld is constrained to follow the nodal displacements. To summarize, for element SAXAMN: Axial transverse shear: M (N + 1) Circumferential transverse shear: (M Drill rotation: (M + 1)N
+ 1)N
constraints are enforced, where M 2 f1; 2g is the order of integration in the axial direction and N 2 f1; 2; 3; 4g is the number of Fourier modes.
Transverse shear
The transverse shear strain is the measure of the amount the director eld t3 has rotated relative to the normal to the shell surface. We dene the transverse shear strain as
3c
def
= t3 tc ;
where
tc = x;c = x;c ;
def
with no sum on c and force this quantity to be zero with a penalty constraint. Note that tc is a unit vector tangent to the shell surface dened by the displacement eld along a parametric coordinate line. Recall that 1 = S and 2 = . Hence, 31 is the axial transverse shear strain and 32 is the circumferential transverse shear strain. For convenience, record the variation of the unit vector tc :
tc
tc
kx;c k
3c
=
x;c
t 3 1 t c + t3 1 t c :
Since by denition t3
2 t3 , it follows that
=
3c
1 (t3 2 tc ) +
with no sum on c. For completeness, the second variation of the transverse shear strain is
d(
3c ) = +
x;c
1 kx 1c k2 [t3tc + tc t3 ] 1 dx;c
;
3.6.716
where terms proportional to
3c have been neglected and the coupling between and symmetrized.
Drill rotation
1 has been
Mathematically, the equations governing the deformation of the shell are invariant with respect to drill rotations; that is, a rotation of the director eld with axis of rotation parallel to the director. It is necessary then to assign a kinematic denition to this rotation. We dene the drill strain as the difference between the rotation of the circumferential tangent vector as measured by the displacement eld and that measured by the rotation eld. Accordingly, we dene the drill strain by
= a b:
1
Here, is the rotated reference axial tangent vector and tangent vector, dened by and
a = exp[^]A X
In the above is a linear approximation to the axial tangent vector in the reference conguration K 0 N =k K 0 N k, where K and N are the position vectors to two given by 1 adjacent nodes in an axial plane. The drill rotation constraint then requires that the component of rotation about the normal to the surface match the in-plane rotation of the surface as measured by the displacement eld. The linearized drill strain calculation is similar to the transverse shear linearized strain calculation. Without repeating the calculation,
A A = (X
X ) X
Elements
d
= 1 (a 2 b) + kx1 k (a 0
d b) 1 x; :
;
symmetrized.
2 d( d ) = 1 1 [ab + ba] 1 d 0 x; 1 kx1 k2 [ab + ba] 1 dx; ; 1 [^ 0 (a 2 b)b] 1 dx + x 1 1 [0a 0 b(a 2 b)] 1 d ; + 1 kx; k a ; ; kx; k ^ where terms proportional to d have been neglected and the coupling between and 1 has been
For the case when the reference radius of any point on the shell surface goes to zero, all of the offset nodes collapse to the same point and the edge constraints along that circumferential edge become redundant. It is, therefore, necessary to treat the zero radius case separately. For the zero radius case all redundant degrees of freedom are constrained to follow the average motion of the nodes at 0 and 180. The edge constraints are broken into two parts: First, a
3.6.717
circumferential transverse shear strain is dened that requires that the rotation in the radial direction follow the circumferential rotations at the rst and last nodal plane:
0
32
= r + 1 (1 + P +1 ) sin : 2
p sin pr +
0 32 =
X
P p=1
0 Note that d(
32 ) = 0. Second, a drill rotation strain is dened that requires that the rotation about the Z -axis be zero: 0
d = z :
This leads to a linearized strain given by
0
d
Note that
Mass matrix
X
P p=1
p sin pz
0 d( d ) = 0.
At each material point the displacement components in the three directions (radial, axial, circumferential) are dependent only on the corresponding nodal displacement components. Hence, the mass matrix does not involve any coupling between the radial, axial, and circumferential degrees of freedom, and we can write the mass matrix in the form of three separate expressions:
mnpq M ur = mnpq M uz mnpq M u
= =
Z Z Z Z Z Z
z z z
dz dz dz
A A A
mp nq N r N r dA mp nq N z N z dA mp nq N N dA:
Similarly, we can write the terms associated with the rotational degrees of freedom as:
mnpq M r mnpq M z
= =
mnpq M =
Z Z Z
z z z
z 2 dz z 2 dz z 2 dz
Z Z Z
A A A
mp nq N N dA mp nq N N dA mp nq Nr Nr dA:
3.6.718
Here the superscripts m and n refer to a particular node in the rz plane, and the superscripts p and mp mp q refer to a particular position along the circumference. The interpolation functions Nr , Nz , and mp m ( S ) in the r z plane and interpolation functions N are the product of interpolation functions H in the -direction: mp m p mp
Nr
= H (S ) R ( ) = N z mp N = H m (S ) sin p:
The area integral used to form the mass matrix can be split then into an integral along the length of the element in the rz plane and an integral around the circumference. For the rr component of the mass matrix this yields
mnpq Mur =
Z
z
Z
dz
S
H m (S )H n (S ) 2R dS
1 2
Z
0
2
z dz
H (S )H (S ) 2R dS
1 2
Z
0
2
The matrix can be written in a convenient form by dening the primitive mass matrix as
mn Mprim =
dz
H m (S )H n (S ) 2R dS;
Elements
Z
z
z dz
Z
S
H m (S )H n (S ) 2R dS:
These primitive mass matrices are the same mass matrices that are used for the regular axisymmetric shell elements. We can also dene the circumferential distribution matrices
f1 = f2 =
pq pq
1 2 1 2
Z Z
2 0 0 2
Rp ()Rq () d
sin p sin q d:
and
The various components of the mass matrix then take the form
mnpq mn M ur = Mprim f1 mnpq mn pq Muz = Mprim f1 mnpq mn Mu = Mprim f2 pq mn mnpq pq M r = M prim f2 mn mnpq pq M z = M prim f2 mn mnpq pq M = M prim f1 : pq
3.6.719
The circumferential distribution matrices can be evaluated for various values of the number of terms P in the Fourier series. After some calculations the following results are obtained: P = 1:
f1 P
pq
3 1
1 3
f2
pq
(1) :
= 2:
pq f1 =
P
1 32
0 @
2 12 2
01
2 20
01
2 7
1 A
1 pq f2 =
1 0
0 1
:
= 3:
0 11 1 B 2 pq @ 02 1 = 72
1
02
4 20 2
02
1 02 C A
1 2 11
f2
pq
1 2
0 1 @0
0
0 1 0
1 0A
0 1
= 4:
0 15 2 02 2 01 1 B 2 28 4 04 2 C 1 B pq C B 02 4 28 4 02 C 1 = 128 @ 2 04 4 28 2 A 01 2 02 2 15
f2
01 1 B0 pq = @
2 0 0
0 1 0 0
0 0 1 0
1 0C A
0 0 1
Reference
Axisymmetric shell elements with nonlinear, asymmetric deformation, Section 15.6.10 of the ABAQUS Analysis Users Manual
3.6.720
3.6.8
TRANSVERSE SHEAR STIFFNESS IN COMPOSITE SHELLS AND OFFSETS FROM THE MIDSURFACE
ABAQUS supports element types S3R, S3RS, S4R, S4RS, S4RSW, and S8R for the analysis of laminated composite shells. These elements are based on rst-order transverse shear exible theory in which the transverse shear strain is assumed to be constant through the thickness of the shell. This assumption necessitates the use of shear correction factors. The development of these factors also provides a basis for estimating interlaminar shear stresses in a composite section. This section describes the development of the transverse shear stiffness for element types S3R, S4R, and S8R.
Finite-strain shells
The transverse shear stiffness correction factors are easily shown to be 5=6 for isotropic plates. We want to establish the equivalent factors for laminated plates and sandwich constructions. For this purpose we calculate the distribution of transverse shear stress through the thickness of the shell, for the case of unidirectional bending and assuming linear elastic response. Then the shear strain energy, expressed in terms of section forces and strains, is equated to the strain energy of this distribution of transverse shear stresses. This method, outlined below, provides an approximate method for calculating interlaminar shear stresses and supplies reasonable estimates of transverse shear stiffness. In this calculation ABAQUS assumes that the shell section directions are the principal bending directions (bending about one principal direction does not require a restraining moment about the other direction). For composite shells with orthotropic layers that are not symmetric about the shell midsurface, the shell section directions may not be the principal bending directions. In such cases the transverse shear stiffness and interlaminar shear stresses are less accurate approximations and will change if different shell section directions are used. Consider a plate in the (x; y ) plane. Assume only bending and shear in the x-direction, without gradients in the y-direction. Then the membrane forces in the shell are zero: Nxx = Nyy = Nxy = 0, and @=@y = 0 for all response variables. In this case equilibrium within the section in the x-direction is
Elements
@xx @x
@xz @z
= 0:
(3.6.81)
Vx +
@Mxx @x
= 0;
(3.6.82)
where Vx is the transverse shear force per unit width in the plate and Mxx is the bending moment per unit width for bending about the y-axis.
3.6.81
For the bending behavior we assume the strain varies linearly across the section:
= 0 (z 0 z0 ) ;
where is the membrane strain of the reference surface z = z0 and is the curvature of that surface. If the response of the shell is linear elastic, any in-plane component of stress at a point through the shell section is given by
= D
(
0 (z 0 z0 ) );
(3.6.83)
where the plane stress elastic stiffness, D , is dened from the elasticity and orientation of the material at the particular layer of the shell. Greek subscripts take the range (1; 2). Integrating through the thickness and inverting the resultant section stiffness provides the 6 2 6 section exibility matrix, [H ]:
= [H ]
N M
We have already assumed that Nxx = Nyy = Nxy . We now also assume that Myy = Mxy = 0; that is, that it is possible to have no bending in the y-direction without any restraining moments associated with the y-direction. This is clearly not the case for an unbalanced composite section, but we still use it as a simplifying assumption to obtain the shear correction factors. Thus,
where fHi4g is the fourth column of [H ]. Combining this result with the elastic stiffness at a point through the shell thickness provides the in-plane stress components in terms of Mxx as
xx = (Bx1
fHi4 g Mxx ;
(3.6.84)
0 (z 0 z0)Bx2 ) Mxx ;
(3.6.85)
where
Bx1 = DxxxxH14 + Dxxyy H24 + Dxxxy H34
and
Bx2 = Dxxxx H44 + Dxxyy H54 + Dxxxy H64:
Combining the gradient of this equation in the x-direction with the equilibrium equations Equation 3.6.81 and Equation 3.6.82 yields a description of the variation of the transverse shear stress through the thickness of the plate:
@xz @z
= ( B x1
0 ( z 0 z 0 ) B x2 ) V x :
(3.6.86)
In calculating @xx =@x we have assumed that the elasticity and thickness of the composite section do not vary (or vary slowly) with position along the shell.
3.6.82
A laminated composite shell section consists of N layers 1; 2; 3; . . . with different values of 1 1 2 2 N N (Bx1 ,Bx2 ) at layer 1, (Bx1 ,Bx2 ) at layer 2, . . . (Bx1 ,Bx2 ) at layer N . Layer i extends from zi to zi+1 and its thickness is ti = zi+1 0 zi . Integrating Equation 3.6.86 through the shell, using the boundary i i+1 conditions xz = 0 at z = 0, xz = xz at z = zi+1 and xz = 0 at z = zN +1 , gives the transverse shear stress in layer i as
i xz =
where
i B x1 ( z 0 z i ) 0 i B x0 =
10 2
1 z 2 0 z i 2 0 z x0 ( z 0 z i )
i i B x2 + B x0
B x2
Vx ;
(3.6.87)
i01 X j=1
t j B x1
1 2
(z
j +1
+z )
j 0 z x0
and
The subscript zx0 is result is associated with pure bending in the x-direction. The variation of yz through the shell thickness is obtained using a similar procedure, based on pure bending in the y-direction. These results provide the estimates of interlaminar shear stresses. We dene the shear exibility of the section by matching the shear strain energy obtained by integrating the elastic strain energy density associated with transverse shear stress distribution obtained above: N Z zi+1 2 3 1 1X V
2
01 i i 1 i=1 ti 2 (zi+1 + zi )Bx2 0 Bx1 : z x0 = PN i i=1 ti Bx2 used instead of z0 in this case because the
PN
Elements
b Vx
Vy [F s ]
Vy
where [F s ] is the shear exibility of the section and F i is the continuum transverse shear exibility within layer i. Here we introduce the assumption that the transverse shear exibility within a layer is not coupled to the in-plane exibility. This is usually the case for shell constructions. i i Substituting the relations for xz and yz into the above equation and integrating denes the shear exibility of the section as
N s = X F i t i ( B i ) 2 + t i B i 0 B i 0 ( z i 0 z x0 ) B i 1 + 1 t 2 0 B i 0 ( z i 0 z x0 ) B i 1 2 0 Fxx xx x0 x0 x1 x2 3 i x1 x2 i=1 1 1 i 0 i 1 i 0 4 t3 Bx2 Bx1 0 (zi 0 zx0 )Bx2 2 + 20 t4 (Bx2 )2 i i
i=1 zi
b xz yz c F i
3
yz
xz
dz;
s Fyy =
N X i=1
i 0 (z i 0 zy 0 )B y 2
1 3
i t2 B y 1 i
1
i 0 (z i 0 zy 0 )B y 2 2 0
1
i i t 3 B y 2 (B y 1 0 (z i 0 zy 0 )B y 2 ) 2 + i 4
3.6.83
i t 4 (B y 2 ) 2 i 20
s Fxy =
XF
N i=1
i xy ti
i i i i i B x0 B y 0 + t i B x0 B y 1 0 ( z i 0 z y 0 ) B y 2 2
1 + 1 2 i( 3 1 32 i 8 1 4 20
i i B y 0 ( B x1 0 ( z i 0 z x0 ) B x2
i i i i i i 0 t B x2 ( B y 1 0 ( z 0 0 z y 0 ) B y 2 ) + B y 2 ( B x1 0 ( z 0 0 z x0 ) B x2 )
+
i i t i B x2 B y 2 :
h i01
s s The transverse shear stiffness of the section is then available as F . Notice that F12 will be i nonzero if any layer is anisotropic or orthotropic in a local system (since then Fxy will be nonzero).
Small-strain shells
When the shell resultant forces at each increment are computed for pre-integrated sections, the transverse shear forces for small-strain shell elements S3RS, S4RS, and S4RSW are computed using the transverse shear stiffness derived for nite-strain shells. For numerically integrated sections the transverse shear behavior is based on a simplied stiffness for improved computational performance. For single or multilayer isotropic sections and single layer orthotropic sections, the transverse shear force converges to the proper thin and thick shell transverse shear solution and the transverse shear stress is assumed to have a constant distribution. The transverse shear stiffness is approximate for multilayer orthotropic sections, where the transverse shear stress distribution is assumed piecewise constant. Convergence to the proper transverse shear behavior for this case may not be obtained as shells become thick and principal material directions deviate from the principal section directions.
Offset: reference surface versus midsurface
In ABAQUS the geometry of the shell is dened by kinematic variables that exist at the nodes on the shell reference surface. The kinematics of the shell theory consist of measuring membrane strain on the reference surface and bending strain from the derivatives of the unit normal vector on the reference surface. The default reference surface is the shell midsurface. However, many situations arise in which it is more convenient to dene the reference surface as offset from the midsurface. In this case we assume that the in-plane strain at any material point varies linearly across the section:
= 0 (z 0 z0 ) ;
where and represent the two orthogonal axes on the reference surface, is the membrane strain of the reference surface, z0 is the distance to the reference surface measured from the midsurface, and is the curvature of that surface. The positive values of z0 are in the positive normal direction. When z0 = t=2, the top surface of the shell is the reference surface, where t is the shell thickness. The bottom surface of the shell becomes the reference surface when z0 = 0t=2. When z0 = 0, the midsurface represents the reference surface.
3.6.84
If the response of the shell is linear elastic, any in-plane component of stress at a point through the shell section, , is given by
D
(
0 (z 0 z0 ) );
where the plane stress elastic stiffness, D
, is dened from the elasticity and orientation of the material at the particular layer of the shell. Greek subscripts take the range (1; 2). The section force and moment resultants per unit length can then be dened as
N = M =
Z 0z0 +t=2
dz; zdz:
Integrating the above equations through the thickness leads to the resultant section stiffness, [A]:
N M
= [ A]
Elements
Reference
Shell section behavior, Section 15.6.4 of the ABAQUS Analysis Users Manual
3.6.85
3.6.9
Some of the shell elements in ABAQUS/Standard (S4R5, S8R5, S9R5, and STRI65) use two variables at a node to dene the change in the shell normal at the node, nN , during an increment. At some nodes in these elements and in other elements we use the three components of the rotation triplet as the rotation degrees of freedom. To provide inertia for all of these nodal variables, at a node N with two rotation variables we dene where @ p denotes any time derivative or variation of the following quantity and the eN are the basis vectors used to dene the rotational variables at the node during this increment. Barred superscripts are not summed. This expression neglects any rotation of the basis system that occurs during the increment. This is an approximation in large-displacement analysis: it is adopted for the sake of simplicity, based on the argument that we are not attempting to model the rotary inertia accurately. At a node at which we use the three global rotation components we dene where in this case the ei are the global Cartesian basis vectors. This denition is articial and serves simply to associate a reasonable measure of inertia with the rotational degrees of freedom. The rst-order elements in ABAQUS use a lumped mass matrix. In this case M NM is diagonal, so that the rotary inertia contribution at node N is where sums over the number of rotation components used at the node (2 or 3). For a consistent mass element the rotary inertia contribution is
@ p nN = @ p N eN 0 @ p N eN ; 2 1 1 2
@ p nN = @ p N eN ; i i
Elements
M NN N N ;
nodes nodes
M =1 N =1
X XM
NM N M eN 1 eM ;
where and sum over the number of rotation components used at each node (2 or 3). The time integration algorithms require initial conditions for each increment. For implicit integration _ these are the velocities and accelerations of the variables at the start of the increment, uj0 and uj0. At a node where three global rotation components are used, these initial values are directly available from the solution to the previous increment. At a node where only two variables dene the rotation, we convert variables from the basis of one increment to that of the next through the approximation where j0 indicates a variable associated with the previous increment and j+ indicates a variable associated with the current increment. The justications for this approximation are that it is simple, the incremental rotation will not be large anyway, and we are not trying to model the rotary inertia effect with high accuracy.
Reference
@ p j+ = @ p j0 e j+ 1 e j0 ;
Choosing a shell element, Section 15.6.2 of the ABAQUS Analysis Users Manual
3.6.91
3.7.1
Let gi ; i = ; be the elements usual isoparametric coordinates. Let r be an isoparametric coordinate along the line where the face of the element intersects the plane of reinforcement, with 0 r in an element (see Figure 3.7.11).
12
r g2
Elements
g1
Figure 3.7.11 Rebar in a solid, two-dimensional element. The plane of reinforcement is always perpendicular to the element face. The rebar will be integrated at one or two points, depending on the order of interpolation in underlying V , position, rebar strain " , and rst and second variations of elements. The volume of integration
(1 )
()
3.7.11
rebar strain
1 2
t0 W N ;
where
t0 Ar Sr WN
x = x (g i )
is the original thickness for plane elements and x1 for axisymmetric elements; is the rebar cross-sectional area; 0 is the spacing of rebar (for axisymmetric elements Sr x1=x0 Sr , where x0 is 1 1 0 is given); the radius where the spacing Sr is the Gauss weight associated with the integration point along the r line; is position; and
=(
()
Strain is
where dl and dlo measure length along the rebar in the current and initial congurations, respectively. For the deformations allowed in these elements,
@x @r
@x = @gi @gi : @r
dl dlo
2
where
@ x @ x @ xo @ xo 1 = 1 @r @r @r @r is the squared stretch ratio in the r-direction, and t is the stretch ratio in the thickness direction: t = 1 for plane stress or plane strain; t = t=to for generalized plane strain, where t is given in Generalized plane strain elements, t = x1=x1o
Section 3.2.7; and for axisymmetric elements. From these results the rst variation of strain is
" =
where
dl 2
o dl
(cos2 @x @r
@ x @ xo @ xo = 1 @r @r @r
+ pt);
pt = 0
3.7.12
x x dl 2 d" = 02 o cos2 @ x 1 @ x = @@ro 1 @@ro + pt dl @r @r x x 1 cos2 @ x 1 @dx = @@ro 1 @@ro + dpt @r @r dl 2 x x amp; + o cos2 @dx 1 @ x = @@ro 1 @@ro + dpt : dl @r @r
Reference
Dening rebar as an element property, Section 2.2.4 of the ABAQUS Analysis Users Manual
Elements
3.7.13
3.7.2
Let gi ; i = ; ; ; be the isoparametric coordinates of the basic nite element in which the rebar are placed. ; , be isoparametric coordinates on the surface of reinforcement, with 0 r . Let t Let r; be a material coordinate along the rebar direction. See Figure 3.7.21.
123 =1 2
r2 r1 t
g3 g2 g1
Elements
3 + 5
r2 r1
4 +
+ 1
+ 2
Figure 3.7.21
The rebar is integrated using 2 or 2 Gauss points, depending on the order of the underlying element. The volume of integration at a Gauss point is
2 2 1 1 1V =
Ar @ X @ X 2 W ; Sr @r1 @r2 N
3.7.21
where Ar is the cross-sectional area of each rebar, Sr is the rebar spacing, WN is the Gauss weighting is the position of the Gauss point, and associated with the integration point,
@r
X = @X @gi :
@gi @r
In these expressions all quantities are taken in the reference conguration, and so ABAQUS ignores changes in rebar cross-sectional area due to straining of the rebar and changes in the rebar spacing due to straining of the nite element in which the rebar is placed. The strain in the rebar is 1 g ; " = ln 2 G where @x @ x @ri @ x @x 1 and = ; g=
@t @t @t @ri @t
and G is the value of g in the original conguration. For convenience we dene s, a material coordinate that is distance measuring along the rebar in the current conguration: p ds = g dt: The rst variation of strain is
" = @x
1 @ u ; @s @s
1 @du 0 2 @ u 1 @ x @ x 1 @du : @s @s @s @s @s
Reference
Dening rebar as an element property, Section 2.2.4 of the ABAQUS Analysis Users Manual
3.7.22
3.7.3
The denition of rebar in shell, membrane, or surface elements is based on three geometric properties: the cross-sectional area of each individual rebar, the spacing between the rebars, and the orientation of the rebar with respect to the local coordinate system of the element. For shell elements the rebar denition also requires the distance from the midsurface to the rebar. In ABAQUS an equivalent smeared orthotropic layer is created based on these geometric properties and the elastic modulus of the rebar material. The equivalent rebar layer lies parallel to the midsurface of the element. For membrane and surface elements this layer coincides with the plane of the element, and for shell elements this layer can be offset by an amount up to half of the shells thickness. In geometrically linear analyses the geometric properties of the equivalent rebar layer remain constant. However, in geometrically nonlinear analyses each of these properties can change as a result of nite-strain effects. The user has many options for dening which direction the rebar acts in the element. In each case an angle 2 is determined between the reinforcement and one of the elements isoparametric coordinate directions (selected by the user), measured positive with the axis of rotation along the normal to the element. Let the unit vector T dene the rebar direction at a point in the element. The isoparametric directions are given by the tangent vectors dened
A @ A = @X = @N XA ; @
where is the reference midsurface position, are the isoparametric coordinate functions ( = 1 or 2), A are the elements shape functions, and N A are the elements reference nodal positions. The reference or initial rebar angle, 2, is calculated from the inner product between the rebar unit vector, , and the isoparametric direction, , where is specied by the user.
Elements
X A
no sum on .
Both the rebar direction, , and the user-selected isoparametric direction, , lie in a tangent plane parallel to the midsurface. The in-plane unit vector perpendicular to the rebar direction, , is dened by rotating through 90 around the normal to the midsurface, . The normal direction, , is found as
N = A1 A2 A1 A2
2 2 k
P N
As the rebar-reinforced element deforms, the rebars change in length and spacing. The smeared rebar layer assumption implies that the deformation of the rebar layer is determined from the deformation gradient of the underlying element. Following from this assumption, the rebar stretch r is
r = kF 1 T k = ktk ;
3.7.31
where t = F 1 T is the deformed rebar material ber. Since the deformation gradient maps material lines that are etched into the reference body into the deformed conguration, the length change of these material lines denes the stretch. The rebar logarithmic strain, "r , is
"r = ln r = ln ktk :
The rebar spacing stretch p is the stretch in the plane of the rebar in the direction perpendicular to the rebar. To determine the spacing stretch p , use the fact that the unit normal, p, perpendicular to the deformed rebar direction, t, in the plane of the rebar is
0T p = kF0T 11 Pk : F P
It is easily veried that
p is a unit vector.
< p; t >=
The spacing stretch, p , can be dened as the component along p of the deformation of the direction P perpendicular to the reference rebar direction. Since the deformation of P is F 1 P, the spacing stretch follows from
p = < p; F 1 P > =
Since
P is a unit vector,
The nal angle that the rebar direction makes with respect to the user-selected isoparametric direction is
kF 0T 1 P k :
each integration location of the element. The equivalent thickness of the smeared layer is equal to the area of the rebar divided by the rebars spacing; ABAQUS assumes that the volume of the rebar remains constant throughout the analysis. This assumption implies that the area and spacing of the rebar may change as a result of nite-strain effects. The rebars area and spacing in the deformed conguration are dened as follows:
1, is the difference between the nal angle and the original 1 = 0 2 : ABAQUS reports the current angle, , and the change in the rebar angle, 1, for each rebar denition at
ktkkak
no sum on .
A0 Ar = r r
and
0 Sr = S r p ;
3.7.32
where
A0 = r 0 Sr =
In shell elements the rebar layer can be dened initially at a distance above or below the midsurface. In shell elements that permit nite strain, the shells thickness can change as a function of the in-plane deformation. In ABAQUS/Standard the thickness change is dened by the section Poissons ratio, which can be specied by the user. In ABAQUS/Explicit this behavior is based on the actual material properties through the shells thickness. To account for the change in the shells thickness, the rebar layers distance from the midsurface is scaled by the thickness stretch.
Reference
Elements
3.7.33
3.8.1
ABAQUS includes a family of elements that can be used to represent uid-lled cavities under hydrostatic conditions. These elements provide the coupling between the deformation of the uid-lled structure and the pressure exerted by the contained uid on the boundary of the cavity. In ABAQUS/Explicit the uid must be compressible and the pressure is calculated from the cavity volume. In ABAQUS/Standard the uid inside the cavity can be compressible or incompressible, with the uid volume given as a function of the uid pressure, p; the uid temperature, ; and the uid mass, m, in the cavity:
V
= V (p; ; m):
We refer to the incompressible case as a hydraulic uid and to the compressible case as a pneumatic uid. The volume, V , derived from the uid pressure and temperature should equal the actual volume, V , of the cavity. In ABAQUS/Standard this is achieved by augmenting the virtual work expression for the structure with the constraint equation
V
0V = 0
without the cavity. The negative signs imply that an increase in the cavity volume releases energy from the uid. This represents a mixed formulation in which the structural displacements and uid pressure are primary variables. The rate of the augmented virtual work expression is obtained as
d 3
5 = 5 0 pV 0 p(V 0 V ); where 53 is the augmented virtual work expression and 5 is the virtual work expression for the structure
3
Elements
Here, 0pdV represents the pressure load stiffness, and dV =dp is the volume-pressure compliance of the uid. Since the pressure is the same for all elements in the cavity, the augmented virtual work expression can be written as the sum of the expressions for the individual elements:
3
5 = 5 0 p =
Xh
e
X
e e e
V e
0 p
e
X
e
Ve
e
X
e e
Moreover, since the temperature is the same for all elements in the cavity, the uid volume can be calculated for each element individually:
V
where me is the element mass. Note that in the solution, the actual volume of the element may be different from the element volume: V e 0 V e = 0: 6 The total uid volume will match the volume of the cavity, however.
Hydraulic uid with thermal expansion
In ABAQUS/Standard the uid is incompressible by default and the uid volume, upon temperature but independent of the uid pressure:
V,
is dependent
V (; m);
dV dp dV dp
=0
If compressibility is introduced, the uid volume depends upon both the temperature and pressure:
V (p; ; m);
0 m ; K
R
where K is the uid bulk modulus and R is the reference uid density at zero pressure and the initial temperature. The total uid mass in the cavity is the sum of the uid masses of the elements making up the cavity:
m= m
Xm :
e e
The mass of a uid element in the cavity, and the initial element volume, VIe :
(pI ; I ),
me = (pI ; I ) VIe;
where pI is the initial uid pressure and I is the initial temperature. The initial uid density follows from the user-dened reference density, R :
(pI ; I ) =
(1
0 pI =K ) :
R
(p; ), is obtained as 3 2 (p; ) = R 1 + 3()( 0 0 ) 0 3(I )(I 0 0 ) 0 p=K 01; () is the mean
where 0 is the reference temperature for the coefcient of thermal expansion and (secant) coefcient of thermal expansion, and it is assumed that j=R 0 1j << 1. Thus, the uid volume at the current pressure and temperature is
( )= 1
X
e
(p; ) =
X
e
me = p;
( ) = m=(p; ):
Fluid can be added to or removed from the cavity. The amount of uid added is given as the change in (uid) mass m. Consequently, the change of the uid volume at the current cavity temperature is
Ideal gas
In this case the uid is compressible, and the volume is a function of the pressure and the temperature in the cavity:
V
= V (p; ; m);
where, as before, the total uid mass in the cavity is the sum of the masses of the elements in the cavity. The uid is assumed to behave like an ideal gas; hence, the density of the uid in the cavity can be calculated as
p;
Elements
where R and pR are the temperature and pressure at the reference density R , A is the temperature at absolute zero, and pA is the ambient pressure. The current uid volume can again be calculated on an element by element basis:
V p;
( )=
dV dp
X
e
(p; ) =
X
e
me = p;
( ) = m=(p; ):
Fluid can be added to or removed from the cavity. The amount of uid added is again given as the change in (uid) mass. Consequently, the change of the uid volume at the current cavity temperature is
Volume calculation
The hydrostatic uid elements appear as surface elements that cover the cavity boundary, but they are actually volume elements when the cavity reference node is accounted for. Figure 3.8.11 depicts the 4-node hydrostatic uid volume element, F3D4. The dashed lines indicate that the element is actually pyramidal in shape.
3.8.13
4 3
Figure 3.8.11 F3D4 hydrostatic uid element. The volume, V e , of each element must be calculated. The coordinates of any point on the base of the pyramid element can be found by
x=
X
N
N (g; h)xN ;
where N N are the interpolation functions for the base of the pyramid (Solid isoparametric quadrilaterals and hexahedra, Section 3.2.4), expressed in terms of parametric coordinates g and N are the nodal coordinates; and the summation extends over all nodes on the base. For threeh; x dimensional elements the Jacobian on the surface is calculated as
@
@g
N face, n,
@N
@g
xN ;
@h
X
N
@N
@h
xN :
dA
n dA =
The innitesimal volume,
dV
@g
2 x
@ @h
dg dh:
1 3
xR 0 x) 1 n dA;
3.8.14
where xR is the position of the cavity reference node. The volume of the element, V e , is then obtained by integration. For a quadrilateral base this yields
Ve =
Ve
dV
Z +1 Z +1
1 3
01 01
(x R
0 x)
@x
@g
2 @h
@x
dg dh:
The integration boundaries will be different for a triangular base. Introducing the relative position, x = x 0 xR , this becomes
Ve =
Z +1 Z +1
01 01
1 03 x 1
@x
2 @x @g @h
@ x
@g
dg dh:
Z +1 Z +1
01 01
1 03
x
@x
2 @x @g @h
x1
2 @ x + @ x 2 @ x @h @g @h
dg dh:
This expression includes contributions to the volume change due to variation in the sides of the pyramid element. Hence, the equivalent forces will also include the effect of the pressure on these sides. The pressure on the sides will be balanced by the pressure on the side of the adjacent pyramid element; hence, we only need to calculate the contribution from the pressure on the base of the pyramid. This can be done by separating the contributions by using partial integration, which yields
Vbe =
Z +1 Z +1
01 01
0 x 1
@x @g
2 @h
@x
Elements
The last two integrals represent the contributions on the sides of the pyramid; hence, the resulting expression is Z Z
Vbe =
+1
+1
01 01
0 x 1
@x @g
2 @x @h
dg dh:
This equation can readily be obtained directly (see Pressure load stiffness, Section 6.5.1). The second variation of the expression for the volume is
dVbe =
Z +1 Z +1
01 01
03
"
x
Similar expressions can be obtained for planar and axisymmetric uid elements. For a planar element the volume is readily obtained as
1 e V = 0 e3 1 2
x2
@x @g
dg;
3.8.15
where e3 is the unit vector in the direction perpendicular to the plane. Similarly, for axisymmetric elements Z +1
Ve
= 0 2 e 3
01
x2
@x @g
(x 1 er ) dg:
The rst and second variations are obtained in the same way as for three-dimensional elements. The integrations can be carried out analytically. For instance, for element type F3D4 the above expressions yield, after some manipulation,
Ve
1h = 12 x1 1 [(x2 0 x4) 2 (2x1 0 x2 0 x3)] + x2 1 [(x3 0 x1) 2 (2x2 0 x3 0 x4)]+ i x3 1 [(x4 0 x2 ) 2 (2x3 0 x4 0 x1 )] + x4 1 [(x1 0 x3 ) 2 (2x4 0 x1 0 x2 )] ; 1h = 12 ( x2 2 dx1 0 x1 2 dx2) 1 (x3 + x4) + (x3 2 dx2 0 x2 2 dx3) 1 (x4 + x1)+ ( x4 2 dx3 0 x3 2 dx4) 1 (x1 + x2) + (x1 2 dx4 0 x4 2 dx1) 1 (x2 + x3)+ i ( x1 2 dx3 0 x3 2 dx1) 1 (x2 0 x4) + (x2 2 dx4 0 x4 2 dx2) 1 (x3 0 x1) :
In addition to the uid cavity elements ABAQUS also offers a 2-node uid link element that can be used to model uid ow between two cavities or between a cavity and the outside world. This is typically used when the uid has to ow through a narrow orice. It is assumed that the mass ow rate, q, through the link is a homogeneous function of the pressure difference, p p1 0 p2 . In 1 2 = addition, it is assumed that the ow rate may depend on the average temperature p1 p 2 = : and, for a compressible uid, on the average pressurep
=( + ) 2
1 = =( + ) 2
= q(1p; p; ): 1
(3.8.11)
The ow rate needs to be integrated over a nite increment. We assume that the dependence on the average pressure, p, is weak. Hence, we use a semi-implicit method: we use p at the end of the m increment and p0 at the start of the increment. For the temperature we choose to be the average of at the start and end of the increment, because this is likely to be the most accurate. Hence, we obtain the mass ow through the link:
This mass change needs to be converted to a volume change in each cavity. It is assumed that the uid in both cavities is the same, but the pressures (and possibly the temperatures) may be different. With use of the pressure and temperature-dependent density i p; for each cavity i, the relations become
( )
1V 1 = 01m=1;
1V 2 = 1m=2;
and
@ V1 @p1
1 = 0 1 @1m + 1m @1 ; @1V 1 = 1 @1m ; 1 @ 1 p 2 @p1 @p2 1 @ 1 p 1 1 @ 1V 2 = 1 @@1m ; @@pV 2 = 0 1 @@1m 0 1m @2 : @p1 2 @p2 2 1p 2 2 1p 2
(3.8.12)
Observe that, except for isothermal incompressible uids, the resulting matrix is nonsymmetric. For an incompressible uid with thermal expansion, the nonsymmetric contribution is small and the equations can be symmetrized without signicant degradation in convergence rate. For an ideal gas, the nonsymmetric terms can have a signicant effect if the drop in pressure is large. Many applications of uid ow through a uid link involve dynamic loading in the form of steady-state vibration; and often in such cases the dissipative losses in the uid link must be modeled to obtain useful results. In most problems of this class the uids on either side of the uid link are rst pressurized statically. In the implementation in ABAQUS/Standard, it is assumed that the vibration amplitude is sufciently small that the uid link response in the dynamic phase of the problem can be treated as linear perturbations about the prepressurized state. For small vibrations about a prepressurized state we linearize Equation 3.8.11 to give
Elements
(1p; p; ) 1p: q= @ 1p 0
@q
1m = =
Zt
01
Zt
qdt
Zt
01
@q @ p 0
1 1pdt
i @q
@q
= 0 ! @1p 1p: 0
i @q
01
@ p 0
Substituting the above expression for mass ow into Equation 3.8.12 and noting that yields
@ V 1 @p1 0 @ V 2 @p1 0
1 m j0 = 0
@ V1 @p2 0
2
3.8.17
In ABAQUS this model is provided only for the direct-solution steady-state dynamic analysis procedure.
Negative eigenvalues
It is possible that negative eigenvalues will be encountered in the solution of certain hydrostatic uid element problems. With standard elements this can indicate that a bifurcation or buckling load has been exceeded. However, this is not necessarily true with hydrostatic uid elements; negative eigenvalues can result solely from the numerical implementation. Consider the simple hydrostatic uid model depicted in Figure 3.8.12.
F 1
k1
fluid
Figure 3.8.12
If the uid is considered incompressible, application of the downward force causes the uid to compress vertically and expand horizontally, while maintaining the original uid volume. Thus, the model can be adequately discretized as a three degree of freedom system: the horizontal displacement of the right platen, u; the vertical displacement of the top platen, v; and the uid pressure, p. The corresponding system of equations in matrix form is
The equation solver processes the equations in sequence. Hence, it will process the 2 2 2 submatrix relating the displacements to the forces prior to processing the constraints. This leads to a
3.8.18
,, ,, ,, ,, ,, ,,
,,,,,, ,,,,,,
F 2 k2
, , , , , , , , , ,,,,,,,,, ,,,,,,,,,
v
negative eigenvalue if p2 > k1 k2. However since the mode associated with the negative eigenvalue is subsequently constrained by the continuity equation, no instability occurs.
Reference
Hydrostatic uid elements, Section 18.3 of the ABAQUS Analysis Users Manual
Elements
3.8.19
ELBOW ELEMENTS
3.9.1
ELBOW ELEMENTS
The design of piping systems relies heavily on the use of elbows (pipe bends) to provide exibility in response to thermal strains. For this reason the elbows themselves are the critical components in the design, and it is essential to predict their response accurately to qualify a pipeline structurally. This is especially true in accident conditions, where inelastic behaviorplasticity or viscoplasticitydominates the response. It is well known that piping elbows achieve their exibility through a shell-type behavior responding to bending loads with signicant ovalization of the pipe cross-sectionin contrast to the beam response of straight pipes, where the cross-section does not deform to any signicant extent until Brazier buckling occurs. Ovalization gives the elbows elastic exibility that is 520 times that of the same size straight pipe in bending (for typical primary piping loop elbows in nuclear reactors). This exibility is accompanied by stresses and strains that are typically 312 times those of a straight pipe under the same loadwith of coursevery high strain gradients around the pipe and through its thickness, caused by the ovalization of the cross-section. For the purpose of linear elastic analysis these effects present little difculty because the exibility factors and stress intensity factors have been accurately calculated and veried with experiment and are, thus, used directly in response prediction using three-dimensional beam theory (see Dodge and Moore, 1972, for a survey of such work).
Elements
For nonlinear analysis or pipes of noncircular section, these solutions are not useful. Two possible approaches suggest themselves for the nonlinear case: nonlinear equivalents to the linear elastic exibility and strain intensity factors can be sought or detailed analysis based on the shell-type behavior of the pipe bends can be undertaken. The concept of nonlinear equivalents to the linear elastic exibility and stress intensity factors has been pursued by a number of workers (e.g., Spence, 1972), although the success of such methods does not yet warrant their use in design (Scheller and Mallett, 1979). Detailed inelastic analysis remains, therefore, as the obvious approach to qualifying piping systems that cannot be designed on an elastic basis. The main difculty with such analysis is its cost in both engineering time and computer time. The structure is a shell exhibiting severe stress gradients, so that the numerical models required to provide detailed analysis of adequate accuracy are large and complicated. Many attempts have been made to provide such models. Hibbitt (1974) described the formulation of a shell-type nite element that uses piecewise cubic interpolation of displacement around the pipe and piecewise constant strain along the pipe. The latter simplication gives rise to undesirable displacement discontinuities along the pipe. Nevertheless, the approach has documented accuracy (Sobel, 1977) and has been used in design evaluation (Chen and Weiner, 1978). Ohtsubo and Watanabe (1976) develop shell elements of ring geometry, based on the virtual work statement for small strains of a thin walled torus section. They interpolate the normal and two tangential displacement components of the middle surface of the shell each by a complete Fourier series truncated after M termsaround the pipe and a piecewise cubic along the pipe. Their model allows both ovalization and warping of the section, thus potentially describing the behavior of the pipe to high accuracy. Their results show that, with such an approach, for typical primary piping elbows four to six Fourier modes are needed around the pipe to obtain a convergent solution, with fewer modes necessary in less exible elbows. However, their formulation is not well-suited to practical application in multi-elbow pipelines because strains result from rigid body motions of the elbows. This is due to the use of cylindrical shell
3.9.11
ELBOW ELEMENTS
theory with polynomial and Fourier interpolation of displacement components in a curvilinear coordinate system. Takeda et al. (1979) have developed an extension of the Ohtsubo and Watanabe shell element, using both Fourier and piecewise cubic interpolation along the pipe. The element has been applied to some welldocumented, nonlinear benchmark problems (e.g., Kwata, 1978) and shown to be accurate but relatively expensive. In Kwata (1978), Takeda et al. compare the element to the single point integration, bilinear interpolation shell element of Hughes et al. (1977) and conclude that the latter element is more economical for given accuracy. To a large extent this conclusion would appear to depend on the architecture of the computer code in which the element is implemented: unless the code architecture allows economic calculation of such elements, they will be very expensive. In Kwata (1978) the simple bilinear shell is shown to need about 40 elements (and hence 40 constitutive points) around the pipe to provide accurate stresses (the paper actually shows 20 elements around the half-pipe in a case of in-plane bending), thus introducing about 240 degrees of freedom per segment; while in Ohtsubo and Wanatabe (1976) six Fourier modes (and hence 12 degrees of freedom for in-plane modeling, 24 degrees of freedom for out-of-plane modeling, per segment) give the same level of accuracywith again, about 40 constitutive calculation points. With the formulation in ABAQUS, we also conclude that 3248 constitutive calculation stations around the pipe are needed. The contrast in number of degrees of freedom per segment suggests that in spite of the relative complexity of elements like that of Ohtsubo and Watanabegiven suitable code architecture designed to handle such elements efciently, elements that take advantage of the particular geometry and behavior of elbows should be signicantly cheaper than most simple shell elements such as that of Hughes et al. (1977). The element of Bathe and Almeida (1980) is a direct nite element implementation of the classical von Karmann approach. Warping of the section is neglected, and in the plane of the cross-section only pure bending of the pipe wall is allowed. This latter restriction precludes the direct application of thermal loads to the element. In addition, it does not allow for the circumferential membrane force correction added by Gross to the von Karmann theory (see Dodge and Moore, 1972). As an additional simplication, the gradients of axial and shear strain through the thickness of the pipe wall are ignored. Around the pipe a Fourier series is used for the tangential displacement component of the pipe wall middle surface in the plane of the pipe section. In this element only the even numbered Fourier terms (sin 2m; cos 2m; m = 1; 2:::M ) are used, as in von Karmanns analysis. This implies that these displacements are symmetric about the crown of the elbow (about = =2; = 3=2 in Figure 3.9.11). This is satisfactory for cases where the pipe radius is small compared to the elbow torus radius but is likely to be inadequate when the pipe radius is of the same order as the torus radius. The interpolation along the pipe is a piecewise Lagrange cubic, in contrast to the Hermite cubic of Takeda et al. (1979), so that considerably more degrees of freedom are introduced per segment. The reason for this choice is not made obvious in Bathe and Almeida (1980). The authors restrict themselves to 3 Fourier modes and 24 integration stations (constitutive calculation points) around the pipe; the element would probably require some extension to allow more modes and constitutive calculation points around the pipe for application to more exible elbows. In ABAQUS/Standard we build on the experience summarized above and provide a capability for detailed inelastic analysis of pipelines at reasonable cost. The achievement of such complex response prediction in a reasonable amount of computer time requires that the geometric modeling be as efcient as possiblethat is, that the model be tuned to the characteristics of the specic structural geometry and
3.9.12
ELBOW ELEMENTS
response. Pipe bends behave as shells; however, they have the simplifying characteristic that the strain gradients along the pipe are usually mild compared to those around the pipe. The choices of geometric modeling used for the elbow elements in ABAQUS use this simplication to minimize the computer costs associated with the nonlinear analysis of pipelines.
Geometric denitions
We assume that the undeformed pipe is of circular or nearly circular section. Let (; S ) be the material coordinates of a point in the pipe wall middle surface, dened as follows (see Figure 3.9.11). In the undeformed conguration measures the angular position of the point from the crown of the section such that increases towards the extrados (on a straight segment the crown and extrados are arbitrarily dened as xed positions around the pipe section) and S measures the distance along the pipe axis from some arbitrary origin. Let x(; S ) be the current and (; S ) the reference positions of a point. Let (S ) and (S ) be the current and reference positions of a point on the pipe axis, and dene = 0 as the offset of a midwall point from the pipe axis. Let = [ 1 ; 2 ; 3] be a right-handed director set of orthogonal, unit vectors, with 3 approximately tangent to the pipe axis. = [ 1; so that the 2; 3 ] are the same directors in the reference conguration: we choose point = 0 (the crown of an elbow section) lies on 1 and 2 points to the extrados ( = =2): For simplicity we rewrite
y x x A A A A
a a a A
y=ya
i
= (r + u ) +
o r
ut
t + y 3 a3 ;
Elements
where and
Finally, we dene a unit vector, , which will be made approximately outwardly normal to the pipe wall middle surface, as follows. Let
t
x x x 0 a 3 0 a3 1 t t :
t t
1 2
=constant,
and let
t
Then write where
n = n (n 1 n )0 ; n = @x 2 a3 +
3tn;
1 2
@
with
3 (; S ):
3.9.13
ELBOW ELEMENTS
n x
x=@ = 0:
Interpolation
The concept is that of a beam with deforming section, so that we rst interpolate of S and then interpolate and as functions of S and .
x and a as functions
x=
n=1
H n (S ) x n ;
(N
where Hn(S ) are the usual polynomial interpolation functions of order are used to interpolate a rotation triplet,
! (S ) =
which then gives where
N X n=1
H n (S ) ! n ;
a = C (! ) 1 A ;
Numerical experiments with standard shell models have shown that warping effects are important in the geometries and loadings of interest. We, therefore, introduce the Fourier/polynomial interpolation
y3 (; S ) =
+
in the plane
out-of-plane
where
cos p sin p
sin p
cos p
3.9.14
ELBOW ELEMENTS
The number P gives the order of Fourier interpolation, and the terms in-plane and out-ofplane refer to symmetries that occur if the motion is symmetric about the plane of the pipes initial curvature or antisymmetric about that plane. To model ovalization we assume that
ur
=
+
m =1 M P
m=1 p=1 M P
X XX XX
M M P M P
H m (S )(ur )m U
(out-of-plane)
m=1 p=1
and
ut
m=1 p=2
XX XX
M
(in-plane)
(out-of-plane)
m=1 p=2
Elements
n
=
m=1 M P
m=1 p=1 M P
X XX XX
(out-of-plane).
m=1 p=2
The p = 0; 1 terms in y3 and ut and the p = 1 term in 3 are omitted since these relative quantities should not include rigid body motion. The implementation of the formulation has been based on choosing linear polynomials for the Hn (S ) and H m (S ) (referred to as element type ELBOW31) and quadratic polynomials for these functions (element type ELBOW32). Obviously element type ELBOW31 is the lowest order possible for such interpolation. However, we also provide an element in which the Hn(S ) are linear, while M = 1 and H 1 (S ) = 1; that is, y and 3 are constant within an element and, thus, discontinuous from element to element. This is the level of approximation used in routine linear design formula analyses (Dodge and Moore, 1972) and that has been used previously in nonlinear cases (Hibbitt, 1974). We refer to this as element type ELBOW31B. For the linear elements (ELBOW31, ELBOW31B) one-point integration is used in each element with respect to S ; the quadratic element uses a two-point Gauss rule.
3.9.15
ELBOW ELEMENTS
The choice of P the number of terms taken in the Fourier serieshas been based on numerical experiments. Ohtsubo and Watanabe (1976) documented a detailed numerical investigation of this choice for their original element, and their results should be relevant to our formulation. For most practical thin-walled piping system applications, we nd that six modes are sufcient to predict strains within a few percent.
Strains and constraint
The formulation is discrete Kirchhoff, using the Koiter-Sanders generalized section strains at each point of the pipe wall middle surface:
" (; S ) =
and
1 2
(g
0 G )
(; S ) = B
where
0 b + sym(B
"
);
@x @
g (; S ) =
is the metric of the pipe wall middle surface and
@ 1 @x
b (; S ) = 0sym
@x @
@n @
where h is the material coordinate measuring position through the pipe wall: 0t=2 h t=2. In applications numerical integration is used through the wall. The integration rule varies from case to case but is usually a ve, seven or nine point Simpson rule for inelastic material response involving creep and plasticity. The discrete Kirchhoff constraint is imposed by associating a penalty with the strain
is the curvature of the surface, where n is made normal to the surface and G and B are the same measures in the reference conguration1 = S and 2 = and Greek indices span the range (1, 2). The strain components at any point through the pipe wall are then dened by the Kirchhoff assumption as e (h) = e + h ;
@x : @S The penalty is obtained from the transverse shear stiffness of the pipe wall, suitably modied for thin walls in the manner of Hughes et al. (1977).
t
=
n1
3.9.16
ELBOW ELEMENTS
n = @ x=@ 2 a3 +
3 a3 :
@n @S
= n n
1 1 @n 1 0 2 [I 0 nn] 1 @S ;
and we impose a penalty to make n 1 @ x=@S small. Then in the computation of the curvatures, b , we approximate n 1 @ x=@S = 0, so that
1 @ x = 0; @
@
@n : @
This completes the basic denition of the element formulation. The formulation allows arbitrary rigid body motions to occur with no strain and uniform thermal expansion with constant strain. The detailed derivation of the strain variations, initial stress matrix, etc., follow directly from the denitions given above, although the forms are complicated. In the actual implementation the nonsymmetric terms in the initial stress matrix caused by our choice of denition of rotation! as well as nonsymmetric terms in some load stiffnesses, are neglected. A major implementation approximation is in the modeling of inertia terms in the form of a mass matrix dened by
Z
Elements
t x 1 x + y 1 y dA +
! tr3! 1 ! dS;
where is the mass density of the material. This is a coarse approximation locally, but the intended applications concern analysis of pipelines, where you are mainly interested in inertia effects associated with locally low-mode response. Table 3.9.11 shows a comparison of free vibration frequencies for a single 90 bend restrained against rigid body motion on its axis at one end and prevented from warping (but allowed to ovalize) at both ends. The agreement between the various models and the standard shell element is not very close, but we feel it is adequate for the intended use. More accurate modeling of inertia would incur a large penalty in computational cost. Dimensions:
90 bend
3.9.17
ELBOW ELEMENTS
Material properties: Youngs modulus: Poissons ratio: Density: 2.165 2 1011 N/m2 (31.4 106 lb/in2 ) 0.3 7822.8 kg/m3 (7.32 Boundary conditions:
2 10
lb sec2 /in4 )
One end restrained on its axis against rigid body motion; both ends restrained against warping.
Table 3.9.11 Free vibration frequencies for a 90 elbow. Model Mode 1 Standard shell elements Element type ELBOW31B 2 P=4 elements P=6 3 P=4 elements P=6 Element type ELBOW31 2 P=4 elements P=6 3 P=4 elements P=6 79.3 73.5 73.0 75.6 75.1 80.0 79.2 82.9 82.0 Free vibration frequencies, Hertz Mode 2 83.0 79.0 78.3 81.7 81.0 85.2 84.4 89.0 88.1 Mode 3 193 171 169.2 168.7 167.2 170.2 164.7 183.5 178.1 Mode 4 195 172 170.6 169.3 167.8 172 165.5 185 179.6 Mode 5 473 474 473 184.7 183 418.7 415.5 427.5 415.3
3.9.18
ELBOW ELEMENTS
Extrados
ro Pipe axis
Crown
Intrados
Elements
(,S) a2 a3
ro a1
_ x
Figure 3.9.11
Reference
Elbow geometry.
Pipes and pipebends with deforming cross-sections: elbow elements, Section 15.5.1 of the ABAQUS Analysis Users Manual
3.9.19
FRAME ELEMENTS
3.9.2
Frame elements are designed for the analysis of initially straight, slender beams. The elements are implemented for large displacements and large rotations but small strains. The elastic response of the frame elements follows Euler-Bernoulli beam theory. Plasticity is included in the elements response through a lumped plasticity model with kinematic hardening, which permits yielding only at the ends of the beam. The hardening data are given as a relationship between the generalized force and generalized displacement. Hence, the plastic response of the element is length dependent. The elastic-plastic frame elements are designed to represent plastic hinge formation in frame-like structures, where a single frame element can be used as a member between the structures nodes.
Degrees of freedom on the element
Frame elements are formulated in terms of the solution variables at user-dened end nodes and extra internal degrees of freedom associated with an internal node. The three-dimensional version of the element is discussed here. The two-dimensional version is found by appropriate reduction of the threedimensional degrees of freedom. The element has three nodes (two user-dened and one internal), 12 external degrees of freedom, and three internal degrees of freedom. Each of the two end nodes has six external degrees of freedom: three displacements and three rotations. An internal node (at the center of the element) has three displacement degrees of freedom only, as shown in Figure 3.9.21.
z
Elements
Ny2
Nx1
Nx2
Mx2
Mz1
x=- L 2
x=
L 2
Mz2
Figure 3.9.21 Frame element in space. The element is formulated in a local system with the x-direction representing the axial direction and the y- and z -directions representing the directions transverse to the frame element axis. In this
3.9.21
FRAME ELEMENTS
q1 = fux1; uy1 ; uz1 ; x1; y1; z1gT ; q2 = fux2; uy2; uz2; x2; y2; z2gT ; q3 = fux3; uy3; uz3gT ; and 8 9T q = qT ; qT ; qT : 1 2 3
The elastic formulation
The elastic response of the element is governed by Euler-Bernoulli beam theory. The displacement interpolation for the deections transverse to the frame element axis (the y- and z -directions) uses fourth-order polynomials, allowing for quadratic variation of the curvature along the beam axis. Let 2 [01; 1] be the isoparametric coordinate along the length of the beam. Then,
2 3 4 u y = a 0 + a1 + a2 + a3 + a 4 ; 2 3 4 u z = b 0 + b 1 + b2 + b 3 + b 4 :
The transverse displacement interpolations incorporate exact solutions to force and moment loading at the ends and constant distributed loads along the beam axis (such as gravity loading). The displacement interpolation function along the frame element axis (the x-direction) is a secondorder polynomial, allowing for linear variation of the axial strain along the frame element axis:
2 ux = c0 + c1 + c2 :
The twist rotation degree of freedom interpolation along the beam axis (rotation about the x-axis) is linear, allowing for constant twist strain:
x = d0 + d1:
L d
z =
L 2 d 2
4 d2 uy
y =
4 0 L2 ddu2z ;
t =
2 dx
L d
where x is the axial strain, z and y are the beam curvatures, and t is the twist strain. The 15 undetermined constants in the interpolation equations for the displacements are determined by introducing the nodal degrees of freedom; that is,
u y ( =
@uz @
( = 1) =
def
0 L y 2 ; 2
def ux ( = 0) = ux3;
etc:
The interpolations in terms of the nodal degrees of freedom are described below in the section discussing the large-displacement formulation.
3.9.22
FRAME ELEMENTS
" = Bq;
where
B is a 4 2 15 matrix and
" = fx ; z ; y ; t gT :
The elastic stiffness matrix is integrated numerically and used to calculate 15 nodal forces and moments12 forces/moments (also called generalized forces) associated with the two end nodes,
F1 = fNx1; Ny1; Nz1; Mx1; My1; Mz1gT ; F2 = fNx2; Ny2; Nz2; Mx2; My2; Mz2gT ;
and three forces associated with the internal node,
9T
Elements
15
F = K e 1 q:
Material properties of frame elements can, in general, be temperature dependent. Let us dene the elastic strain vector as where " denotes the strain is nonzero and is given by
"elastic = " 0 "th + " initial ; total strain and " th denotes the thermal expansion th = ( )( 0 0 ) 0 (I )(I 0 0 ); x
where
( ) 0 I
is the thermal expansion coefcient, is the current temperature at the frame element section, is the reference temperature for , and is the user-dened initial temperature at this point (Initial conditions, Section 19.2.1 of the ABAQUS Analysis Users Manual).
The temperature eld is dened by the user at the elements ends and is assumed to be linear along the element axis but constant within the element cross-section. If the thermal expansion coefcient
3.9.23
FRAME ELEMENTS
is temperature dependent, it is evaluated at the nodes. Thermal strains are calculated at the elements end nodes, and thermal strains at the integration points are interpolated from the nodal points using appropriate interpolation schemes: axial strains are interpolated linearly, curvatures are interpolated quadratically, and twist strain is constant along the frame element axis. Initial generalized strains, " initial , are calculated from the initial generalized forces given by the user, using the relationship " initial = 01 1 initial; where
0I12 E ()
0
I22 E ()
0I12 E ()
I11 E ()
0
0 0 0
3 7 5;
J p G ( )
and A is the cross-section area; E is Youngs modulus; G is the shear modulus; I11 , I12 , and I22 are cross-section moments of inertia; and Jp G is the torsional stiffness. The vector of generalized initial forces includes the following components:
Ke =
01
BT D()B L d; 2
where temperature-dependent material properties are evaluated at the integration points, assuming a linear variation of temperature along the element axis. For the simplest case of a temperature-independent material and a pipe cross-section, the elastic stiffness matrix can be integrated analytically to give: Twisting moments at the end nodes:
M x1 M x2
=
GJp L
01
7 1 1 7
01
1
x 1 x2
:
8 9 < N x1 =
N x3
2
AE
3L
N = : x2 ;
08 08
3.9.24
0 8 < u x1 = 0 8 5 : u x2 ; :
16
38
u x3
FRAME ELEMENTS
Bending moments at the end nodes and transverse forces for all three nodes:
8 9 > Ny 1 > > > Nz1 > > > > > > > > My 1 > > > > > > > > > > > Mz 1 > > > < = > Nz 2 > > > > > My 2 > > > > > > > > > > Mz 2 > > > > > > > > > Ny 3 > : ; 9 8 > uy 1 > > > > uz 1 > > > > > > > > y 1 > > > > > > > > > > z 1 > > > = <
y
Ny 2
where
Ke is the bending part of the elastic stiffness matrix and takes the following form: m
79 0 0 47 2 79 0 0 017 2 6 79 047 2 0 0 79 17 22 0 6 6 9L2 0 0 017 2 03 2 0 2 6 6 6 0 03 2 9L2 17 2 0 6 6 79 0 0 047 2 6 6 79 47 2 0 6 6 9L 2 0 2 6 6 9L 6
2
L L L L L L L L L
Nz 3
uz 3
sym
0128
Elements
We assume that the displacement and rotation increments admit an additive decomposition into elastic and plastic parts. Hence,
1q = 1q + 1q
e
pl
The total forces and moments result from the elastic constitutive relation
F = Ke 1 q 0 qpl :
Introduce the lumped plasticity concept such that plastic deformation can develop at the beam external (end) nodes only and develops through plastic rotations (hinges) and plastic axial displacement at one or both nodes. Further assume that the plastic deformation at the external nodes can be caused by the interaction of all three moments and the axial force. Therefore, the vector of plastic deformation at those end nodes has the following form:
oT
3.9.25
FRAME ELEMENTS
The yield function , called here the plastic interaction surface, is written in terms of the nodal forces and moments. To calculate the increment of plastic deformation, the plastic interaction surface, , is checked at each of the two ends of the frame element during the loading history. In general, the plastic interaction surface is a function of the sectional forces, its plastic cross-sectional capacities, and the hardening parameters. The frame element is elastic if the following conditions are fullled at both frame ends:
oT n pl pl qpl = ux1 ; 0; 0; pl1; y1; pl1 ; 1 x z oT n qpl = upl2; 0; 0; pl2 ; pl2; pl2 : 2 x x y z
81 < 0
and
8 2 < 0:
The frame element is elastic-plastic if the plastic interaction surface is exceeded at one or both frame ends:
(81 < 0 and 82 0) or (81 0 and 82 < 0) or (81 0 and 82 0): 1qpl = 1I VI ; I
Assuming associated plasticity with the direction of the increment of plastic deformation along the outward normal to the plastic interaction surface, the following relationship holds:
where I denotes the magnitude of the plastic deformation at end of the plastic ow at that end.
The hardening model
and
The hardening model follows a nonlinear kinematic hardening rule generalized from the linear Ziegler hardening law and the relaxation term (the recall term), which introduces the nonlinearity. For details on the hardening model, see Models for metals subjected to cyclic loading, Section 4.3.5. Now introduce the generalized backstress vector, I , which denes the origin of the moving plastic interaction surface, and dene it as
9T
I = 1 ; 2: SI ,
Thus, the plastic interaction surface can be expressed as a function of the generalized force I I SI at the end I , where for each component i,
8 =8 ( )
SiI = FiI 0 iI :
The nonlinear kinematic hardening evolution law for the increment of the backstress takes the form
3.9.26
FRAME ELEMENTS
1 I = (CiViI 0
iiI )1I ; i = 1; 4; I = 1; 2;
i
(3.9.21)
where
ViI =
Ci and
i are sectional parameters for the ith plastic component, which must be calibrated from the test data dening the hardening response of the cross-section. The parameters Ci are the initial kinematic hardening moduli, and the parameters
i determine the rate at which the kinematic hardening modulus decreases with increasing plastic deformation. The test data required for this implementation are the values of the sectional force and moment components as a function of generalized plastic displacements. This will be either the axial force versus plastic axial displacement or a moment versus plastic rotation of a hinge. The data can be given as pairs of valuesgeneralized sectional force versus conjugate generalized plastic displacementor by specifying a yield stress for the section. The curve tting algorithm will determine the parameters Ci and
i for each component i, since the hardening law is written separately for each sectional force component, Fi, where the range of i depends on the number of forces and moments entering the plastic interaction surface. Integrating the hardening rule, Equation 3.9.21, the following evolution law for the backstress is obtained: iI =
1 Ci 0 ViI 1 0 e0
i 1I + 0 e0
i 1I ; iI
i
@ 8I : @SiI
Elements
where the backstress 0 indicates the value of the backstress at the beginning of the increment. iI
Plastic interaction surface
Plastic interaction surfaces formulated in the generalized sectional variables depend on the crosssection prole. Frame elements with lumped plasticity are valid for tubular cross-sections only, and in the simplest form the interaction surface can be expressed as an ellipsoid in a space of four sectional components: axial force and three moments. Normalized with the ultimate forces and moments for each of the sectional components, the plastic interaction surface, I , can be written for each end I of the frame element as
(3.9.22)
where Nx0 , Mx0, My0 , and Mz0 represent the cross-sectional capacities at initial yield: the axial force and three moments, respectively. Any other cross-sectional prole for which plastic interaction can be approximated well enough by the above ellipsoidal surface can be used within the lumped plasticity concept for frame elements. For two-dimensional problems modeled with frame elements the plastic interaction condition becomes an ellipsoid in the axial force and bending moment plane. By checking the plastic interaction condition at any time of the deformation at both frame element ends, it is determined that if
3.9.27
FRAME ELEMENTS
1. 2.
end J is exceeded, an iterative procedure is needed to nd the nal deformation state at the end of the increment. Either end I stays elastic and end J becomes plastic, or both ends become plastic.
81 < 0 and 82 < 0, the frame element stays elastic. 8I < 0 and 8J 0(I 6= J ), the frame element is elastic-plastic. If the plastic condition at the 8 0 8 0
3.
I and J , the frame element is elastic-plastic. If the plastic condition is exceeded at one or both nodes, an iterative procedure is needed to nd the nal deformation state at the end of the increment. Depending upon the ratio of plastic deformation at both ends, one or both ends will become plastic.
The integration of the plasticity model for frame elements follows the same general rule as described in Integration of plasticity models, Section 4.2.2. To solve for the value of the deformation and sectional forces at the end of the increment for an arbitrary load increment, an iterative process is required. To set up an appropriate Newton loop, the following relationships are used, with some of them linearized: Elastic equilibrium equation:
FiI
2 6 XXK =
J =1 j =1
e iIjJ
pl qjJ 0 qjJ ;
where F stands for the generalized force at the end of the increment. The associated ow rule:
(3.9.23)
iI =
The backstress denition:
(3.9.24)
SiI = FiI 0 iI :
The plastic interaction surfaces at both frame ends:
(3.9.25)
8I (SI) = 0:
Large-displacement and large-rotation formulation
(3.9.26)
The frame elements admit large overall displacements and rotations; however, it is assumed that the strains are small. Accordingly, the nonlinear geometric formulation corresponds to Euler-Bernoulli beam theory superposed on a rotating reference frame. An Euler-Bernoulli displacement eld, qeb , is
3.9.28
FRAME ELEMENTS
dened relative to this rotation reference conguration, which causes straining. The Euler-Bernoulli displacement eld is dened as follows. Let x and be the average position and average rotation of the frame element in the deformed conguration: 1 1 = and = (1 + 2): x def (x1 + x2 ) 2 2 The motion of the rotating reference system is dened by the rigid body motion, where the translational x part of the motion is the displacement vector ( 0 3), since 3 initially corresponds to the element centroid. The rotation part of the motion is the rotation matrix created from the average rotation vector: ^ = exp[]:
The average rotation denes the rotation of the elements local directions from the reference values ( ; 1 ; 2) to the current values ( ; 1; 2) through
TN N
To dene the strain-inducing rotation contributions, we multiplicatively decompose the rotation at each node,
Elements
where I is the strain-inducing part of the nodal rotation. Since the strains are assumed small, dene ^ eb the Euler-Bernoulli rotations at node I as the axial vector I , such that
^ RI def exp[I ] = R rI ; =
1
(3.9.27)
^ rI = exp[eb]: I
(3.9.28)
^ Using Equation 3.9.28 in Equation 3.9.27, we can solve for I by quaternion extraction. In components relative to the reference element coordinate directions ^ eb = I yI
eb
eb
N1 ;
eb ^ zI = I
N2 ; ^ eb eb = I T: xI
1 1
eb
The Euler-Bernoulli displacement eld is the difference between the position of the node relative to the element center and the reference position of the node relative to the reference center rotated to the current conguration by the average rotation:
3.9.29
FRAME ELEMENTS
Once the equivalent Euler-Bernoulli displacements and rotations are determined from the nonlinear displacements and rotations, standard expressions following Euler-Bernoulli beam theory are used. The element interpolations are
ueb y
L L (0 + 2 + 3 0 4)eb + (0 0 2 + 3 + 4 )eb ; z1 z2 8 8 1 1 ueb = (03 + 4 2 + 3 0 2 4 )ueb + (3 + 4 2 0 3 0 2 4 )ueb + (1 0 2 2 + 4)ueb z z1 z2 z3 4 4 0 L (0 + 2 + 3 0 4)eb 0 L (0 0 2 + 3 + 4 )eb ; y1 y2 8 8 1 1 ueb = (0 + 2 )ueb + ( + 2 )ueb + (1 0 2 )ueb ; x x1 x2 x3 2 2 1 1 eb = (1 0 )eb + (1 + )eb : x x1 x2 2 2 +
The strain increments, following Euler-Bernoulli beam theory, are
x =
2 dueb x ; L d
z =
4 d2ueb y ; L2 d 2
4 d2ueb z y = 0 2 ; L d 2
t =
2 deb x ; L d
where x is the axial strain, z and y are the bending strains, and t is the twist strain.
Additional data
The user can supply hardening data for sectional forces in the form of pairs of values relating the axial nodal force to the plastic extension and the bending and twisting moments to the plastic rotations. If both the hardening data and the yield stress value for the frame section are omitted, ABAQUS assumes that the frame element will remain elastic. The curve tting algorithm is used for the evolution hardening equation. At least three pairs of data are required for ABAQUS to t the curve and solve for the constants Ci and i for each generalized force component, i, as shown in Figure 3.9.22. The parameters i are scaled according to the following formula:
i =
i Vi0;
where Vi0 denotes the component i of the plastic direction at the beginning of plastic deformation. For the plastic interaction surface in the form of Equation 3.9.22, the scaling factor is equal to 0 Vi0 = 2=Si .
3.9.210
FRAME ELEMENTS
K2 F K1 C
0
FU
F = F0+
F0
s =
qpl
Elements
3.9.211
3.9.3
Frame elements admit an optional force response in which axial force only is supported by the element. Furthermore, the axial force is constant along the element; all transverse forces and all moments in the element are zero. The axial forces in the element may be linear elastic or may admit a buckling strut response where the force versus axial strain is characterized by a buckling envelope with hysteresis, as described below. For details on the standard frame element response, see Frame elements with lumped plasticity, Section 3.9.2. In compression the buckling strut response models, in a simple way, the highly nonlinear buckling and postbuckling damage of slender members when loaded monotonically or cyclically. In tension the response is modeled by isotropic hardening plasticity. The buckling strut envelope is phenomenological, derived from experiments with pipe-like members. Since the description of the buckling envelope includes the outer pipe diameter and the pipe thickness, only pipe cross-section types are permitted with buckling strut response. The buckling strut response is linear elastic until the compressive loading exceeds Pcr , the critical load to cause buckling. The value of Pcr is determined with the ISO (International Organization for Standardization) equation, as described below.
Buckling prediction and the ISO equation
Elements
The ISO equation is used to predict the onset of buckling in slender members with pipe-like crosssections. All quantities with dimensions have dimensions of stress. We dene I , which is a function of the axial compressive stress, fc , and the maximum bending stresses about the local 1 and 2 axes, fb1 and fb2 , by the expression
Fb
v" # " #2 u u cm 1 f b1 2 c m 2 f b2 t + :
1
0 Ffec1
0 Ffec2
Here, Fc is a characteristic axial compressive stress, Fb is a characteristic bending stress, cm1 and cm2 are reduction factors corresponding to the cross-section directions 1 and 2, and Fe1 and Fe2 are the Euler buckling stresses corresponding to the 1- and 2-directions. The ISO equation states that buckling does not occur as long as
0 E A k1; k2
is the yield stress, is Youngs modulus of elasticity, is the cross-sectional area, are the effective length factors in the 1- and 2-directions (user-dened),
3.9.31
L1 ; L2 I11; I22 Ze Zp r
11
are the unbraced lengths for the 1- and 2-directions (user-dened), are the bending moments of inertia about the local cross-section directions, is the elastic section modulus, is the plastic section modulus, is the radius of gyration.
For pipe sections if D is the outside diameter and t is the pipe wall thickness, I = I = (D4 0 (D 0 2t)4 ),
Ze Zp r
22
64
pI
22=A
=1 4
pD2 + (D 0 2t)2.
fc f b1 ; f b2 Fyc
is the axial compressive stress, fc = P=A with P the axial force in the element. are the maximum bending stresses about the 1 or 2 cross-section axis, fb1 = M1 =Ze, fb2 = M2 =Ze with M1 and M2 the bending moments about the 1 and 2 direction. is the characteristic local buckling stress, 0 Fyc = 0 for Fe 0:170, 0 0 Fyc = (c2 0 c3 e )0 for e 1:911, where c2 = 1:04654873 and F F c3 = 0:27381606, 0 Fyc = Fe for e > 1:911, F Fe = 2CE (t=D), C = 0:3 is a critical buckling coefcient. is the characteristic axial compressive stress, Fc = [1:0 0 0:282 ]Fyc for 1:34, c Fc = 1 Fyc for > 1:34, where c1 = 0:89282978, 2 with = max(1 ; 2 ), yc yc 1 = k1 L1 FE ; 2 = k2 L2 FE . r r is the characteristic bending stress (for pipe sections = 0 D=Et), 0 for 0:0517, Fb = (Zp =Ze ) Fb = (c4 0 2:58)(Zp =Ze)0 for 0:0517 < 0:1034, )(Zp =Ze)0 for 0:1034 < 1200=E , Fb = (c5 0 0:76 where c4 = 1:133386 and c5 = 0:945198.
Fc
Fb
are Euler buckling stresses corresponding to the 1 or 2 directions, Fe1 = Fyc and Fe2 = Fyc . 2 2 1 2 are reduction factors corresponding to the cross-section directions 1 and 2, respectively. These factors are user-dened as functions of the end moments,
3.9.32
compression stress, and Euler buckling stresses. The default value for each factor is 0:85. If switching between standard frame element response and buckling strut response is permitted, the one-time-only switch to buckling strut response occurs when I (fc ; bb1; fb2) = 1:0. The ISO equation provides the value of critical load, Pcr , which is dened as the value of axial force fc A in the element when the ISO equation is satised. To prevent switching in cases where negligible axial force exists with large bending moments, an additional inequality is used. This additional check, called the strength equation, takes the following form:
S=
f F
yc
1
b
q2
b
f 1 + f 22:
b
For a frame element to switch to buckling strut behavior, both the ISO equation and the strength equation must be satised, I = 1:0 and S 1:0. If buckling strut response is requested for the element from the beginning of the analysis, bending moments cannot be supported by the element. In this case the ISO equation becomes the simplied statement that no buckling occurs for
f <F
c
and
cr
F A:
c
Elements
The Marshall strut envelope denes the postbuckling damaged elasticity model and the hysteretic loop response. To dene the Marshall strut envelope, the value of Pcr and the following seven constants are needed:
0 1
0A ( = 0:95),
, (0 = 0:03), L , (1 = 0:004), D
L D
5 8( D )0:7
t
= )).
The values in parentheses are the default values supplied by ABAQUS, and the value of Pcr is found from the ISO equation as explained above. The Marshall envelope governs the compressive and tensile response of the strut as shown in Figure 3.9.31. The dotted lines in the interior of the envelope indicate the damaged-elastic modulus dening the loading-unloading force versus strain path.
3.9.33
EA
Pcr
Frame elements, Section 15.4.1 of the ABAQUS Analysis Users Manual Frame section behavior, Section 15.4.2 of the ABAQUS Analysis Users Manual
3.9.34
3.9.4
These elements are provided for the specic case of modeling the interaction between a tube and a support that is not always in contact with the tube during dynamic events. The tube is assumed to have a circular section and can interact with one of two tube support geometries: a circular hole and an egg-crate support. Two interface elements of this type are provided, one for each geometry, as shown in Figure 3.9.41 and Figure 3.9.42. As indicated in Figure 3.9.42, one cylindrical geometry interface is needed to model the interaction of the tube with a circular hole, while Figure 3.9.41 shows that several unidirectional geometry elements are needed to model the interaction with an egg-crateone element perpendicular to each pair of egg-crate faces.
Tube
ITSUNI elements 1
Elements
n2
2 2
n1
Figure 3.9.41 ITSUNI elements for tube/egg-crate support interaction. The interface elements themselves consist of a spring and friction link and a dashpot, as shown in Figure 3.9.43. The spring is assumed to behave as shown in Figure 3.9.44: when there is no contact between the tube and the support, no force is transmitted by the spring; when the tube is in contact with the support, the force increases as the tube wall is deformed. This force can be modeled as a linear or a
3.9.41
C of tube L 1
Center of hole
ITSCYL element
Figure 3.9.42 ITSCYL element for tube/drilled hole support interaction. nonlinear function of the relative displacement between the axis of the tube and the center of the hole in the support. The frictional part of the spring and friction link uses the Coulomb friction model in ABAQUS: that model is described in Coulomb friction, Section 5.2.3. The dashpot is provided to model uid effects in the annulus between the tube and the support plate. Its behavior can be linear or nonlinear. The model assumes that shear forces created by the uid are negligible, so that the only shear forces transmitted by one of these interface elements are the frictional forces caused by direct contact between the tube and the support. A major simplication in these elements that saves considerable computational effort in dynamic applications is the assumption that impacts between the tube and its support plates involve no instantaneous transfer of momentum or energy loss: the standard impact algorithm of ABAQUS/Standard used with gap and other interface elements (and described in Intermittent contact/impact, Section 2.4.2) is not needed. This simplication derives from the assumption that these elements will be used in conjunction with beam element models of the tube, so the tube section is dened by the position and orientation of its axis and local deformation of the cross-section of the tube is neglected. In reality, when the tube hits a support, initially only a small part of the tube wall loses momentum so that there isinstantaneouslyonly a small
3.9.42
P3 Q 2
Friction
Elements
1 Q P3
Figure 3.9.43 Tube support element behavior. loss of kinetic energy. This instantaneous energy loss is neglected when these elements are used. The subsequent attening of the tube wall is modeled by the spring link in the element, acting between the node on the tube axis and the node representing the center of the hole. Thus, the modeling of this local attening behavior as an equivalent spring provides the simplication that instantaneous impact calculations are not needed. In cases where this approach is not reasonable, gap elements can be used instead of these special interface elements, at the cost of more computational effort. The remainder of this section discusses the kinematic denitions used in these elements and their contributions to the overall equilibrium equations and to the Jacobian (stiffness) matrix needed in the Newton solution of those equations.
Geometry and kinematics
Each tube support element has two nodes. One node represents the axis of the tube, the other is the center of the hole in the support plate (or midway between a pair of parallel sides of an egg-crate).
3.9.43
ITSUNI
P3
-c0 c0 u3
P3
ITSCYL
c0
u3 c0 = difference between support plate hole radius and tube outside radius
Figure 3.9.44 Nonlinear spring behavior in ITS elements to model clearance and tube attening. Let a2 be a unit vector along the axis of the tube, and let a3 be a unit vector along the axis of the interface element (that is, perpendicular to the parallel sides of the support in the unidirectional interface that is used with egg-crate supports, and parallel to the line joining the two nodes of the element in the cylindrical interface that is used with circular holes). It is assumed that a3 is in the cross-section of the tube and, hence, orthogonal to a2 . We dene a third basis vector as
a 1 = a 2 2 a3 :
3.9.44
Let xN be the current position of node N of the element at any point in time (here N is 1 or 2). Relative displacements in the element are measured from the position when the tube and the hole in the support plate are exactly aligned; that is, when the nodes of the element are at the same location. They are dened as follows: axial to the interface element,
u3
= (x 1 0 x 2 ) 1 a 3 ;
2 = ( x 1 0 x2 ) 1 a 2 ;
and tangential, in the plane of the tubes cross-section, for the unidirectional case:
u
1 = ( x 1 0 x2 ) 1 a 1 ;
Z
1=
x 1 0 dx2 ) 1 a 1 :
Elements
The basis vectora2 , along the axis of the tube and of the hole in the support plateis assumed to be xed. In the unidirectional element the a1 and a3 vectors are also xed. In the cylindrical interface a3 is parallel to the line joining the nodes of the element, so
a3
where
l
= ( x1 0 x2 )
=l;
= (x 1 0 x 2 ) 1 (x 1 0 x 2 )
p
a1
= a 2 2 a 3 = a 2 2 (x 1 0 x 2 )
Z
=l:
1=
1 a 2 2 (x 1 0 x 2 ) 1 (
l
x 1 0 dx2 )
and
du
1 = l a 2 2 (x 1 0 x 2 ) 1 (d x 1 0 d x 2 ):
For simplicity we replace the integral with the backward difference approximation
1 1 = [ 1 a2 2 (x1 0 x2)]
u l
t+1t
1 (1x1 0 1x2 )
3.9.45
The element generates an axial forceP3, parallel to a3 and two shear forcesQ1, parallel to a1; and Q2, parallel to a2 . In addition, because the nodes of the element are at the center of the tube and at the center of the hole in the support, while the interaction forces between the tube and its support are transmitted at the point of contact of the tube with the support, these forces also cause moments at the nodes of the element. It is assumed that the moments caused by P3 and by Q2 are not signicant and can be neglected. The only moments considered are Q1 d1=2 at node 1, the center of the tube, and Q1 d2=2 at node 2, the center of the hole. Here d1 is the outside diameter of the tube and d2 is the diameter of the hole for the cylindrical interface or is the distance between the parallel support plates for the uniaxial interface (see Figure 3.9.45). The virtual work contribution of the element is, then,
W I = P3 u3 + Q1 u1 + Q2 u2 + Q1 d1
2
1 2
0 Q1 d22 2; 2
where N is the virtual rate of rotation about the a2 -axis at node N . 2 From this expression the contribution of the element to the Jacobian (stiffness) matrix of the equilibrium equations is immediately available as
dW I = u3 dP3 + (u1 + d1
2
1 + 2
d2
2
2) dQ1 2
p x 1 0 x2
(
) (x 1
0 x2 ) = l;
so that and so
du3 =
u3 = a3 (x1
1 (x
0 x2 );
u3
u3
1
a2
and so
du1 =
u3
(x
3.9.46
Q1 P3 P3 d1/2 d2/2 1 2
Q1
Q1
Elements
P3
P3 1 Q1 d1/2 d2/2 2
Figure 3.9.45 Contact forces in the cross-section. The initial stress terms for the cylindrical interface are, therefore,
P3 du3 + Q1 du1 =
( x
0 x ) 1 u
2
P 3 [I
The other terms in the stiffness matrix are associated with changes in the forces in the element, We assume P3 is made up of a spring force that is a function of u3 and a dashpot force that is a function of u3:
s d P 3 = P 3 + P3 ; s s d d _ P 3 = P 3 ( u 3 ) ; P3 = P 3 ( u 3 ) ;
so that
dP3 =
For the implicit integration operator used for nonlinear dynamic analysis in ABAQUS/Standard,
du 3 = _
where
du 1
_= ; du 1t where 1t is the time increment and
and are the parameters of the integration operator.
Thus,
du
du3;
dP3 =
s 0 dP3
The values of dQ1 and dQ2 come from the friction theory and are dened from du3 by that theory (see Coulomb friction, Section 5.2.3). In summary,
du3
_ + du dP3 1 du3: du du _
d
3
and
dQ = K du + K3 du3; ; = 1; 2;
2 3
K33 0 0 0 0 8 du3 9 > > 6 K13 > > K11 K12 0 0 7 > du1 > = 7< 4 56 I = u u u 1 2 6 K23 7 du2 K21 K22 0 0 7 dW 3 1 2 2 2 6 > 4 d1 K13 d1 K11 d1 K12 0 0 5 > d1 > > 2> > 2 2 2 : ; d2 K d2 K d2 K d2 13 11 12 0 0 2 2 2 2 amp; +P3 du3 + Q1 du1 : This matrix is not symmetric if Q1, dQ1, or dQ2 is nonzero. Without friction they are zero, and the terms in K33 and P3 are the only nonzero terms. With relatively small friction coefcients in dynamic applications the terms K3 andif the tube diameter is not very large(d1 =2)K1, (d1=2)K13, (d2=2)K1; and (d2=2)K13 can be neglected and, thus, a symmetric approximation to the
Jacobian matrix used without serious degradation of the convergence rate of the Newton solution of the equilibrium equations.
Reference
Tube support elements, Section 18.4.1 of the ABAQUS Analysis Users Manual
3.9.48
3.9.5
The line spring elements in ABAQUS/Standard provide a computationally inexpensive tool for the analysis of part-through cracks in plates and shells. The basic concept was rst proposed by Rice (1972) and has been further discussed by Parks and White (1982). The line spring is a series of one-dimensional nite elements placed along the part-through aw, which allows local exibility of one side of the aw with respect to the other (points A and B in Figure 3.9.51). This local exibility is calculated from existing solutions for single edge notch specimens in plane strain (Figure 3.9.52). The approach is computationally inexpensive compared to fully three-dimensional models of the vicinity of the aw; it is also approximate because of the use of two-dimensional solutions embedded in the shell model. Practical experience with the method on typical geometries has shown that, for several important geometries, the method provides acceptable accuracy.
n q t B
Elements
Figure 3.9.51 Surface geometry; line spring modeling. Side B of the element contains nodes 1, 2, and 3; and for LS6 elements side A contains nodes 4, 5, and 6.
3.9.51
M N
Figure 3.9.52 Line spring compliance calibration model. This section discusses the geometric and kinematic basis of the elements as well as the equilibrium statement and the development of the local solutions that dene the constitutive relationships. The constitutive relations are expressed in terms of the forces and moments carried across the crack and the relative displacements and rotations of points on opposite sides of the crack (A and B in Figure 3.9.51), and are derived from local solutions to single edge cracked plane strain specimens. Elastic and fully plastic (limit analysis) solutions are used to construct an approximate elastic-plastic model. At each point along the aw a local orthonormal basis system is dened (t; n; q), with t the tangent to the shell along the aw, n the normal to the shell, and q dened as
q = t 2 n:
3.9.52
We use the shell normal, n, to determine the side of the shell on which the aw occurs; aws that open on the positive n side are given positive aw depths to indicate this, and those on the negative n side are given negative aw depths. The relative motion between two pointsA and B in Figure 3.9.51on opposite sides of the aw but otherwise at the same place, then denes a set of six generalized strains as follows. Side B of the element contains nodes 1, , ; and for LS6 elements side A contains nodes , , and . Mode I:
23
45
These relative motions form the kinematic basis of the element. Since the line spring elements are written for geometrically linear analysis only, the rst variations of these relative motions are identical to the above denitions, with the total values replaced by their variations.
Virtual work contribution
1III = (B 0 A) 1 q: The relative rotation 1II = (B 0 A ) 1 n plays no role in the deformation.
Elements
= (NI 1uI + MI 1I + NII 1uII + L MII 1II + NIII 1uIII + MIII 1III ) dL;
where NI ; MI , etc., are forces and moments per unit length of aw that are conjugate to the corresponding relative displacement and rotation values. In the above expression the integration is taken along the entire aw.
Interpolation
The elements use quadratic interpolation of displacement and rotation components along the crack, so they are compatible with the second-order shell elements (S8R, S8R5, S9R5, STRI65).
3.9.53
Two line spring elements are providedLS6 is a general element for use with arbitrary aws in a shell, while LS3S is provided for Mode I use in cases when the crack lies on a plane of symmetry and the deformation will be symmetric about the same plane, so that only one-half of the geometry must be modeled.
Elasticity
The Mode I line spring compliance is based on a single edge notched specimen subject to far-eld tension and bending, as shown in Figure 3.9.52. This compliance is
u
N M
where the 2 3 matrix G can be obtained from the energy compliance calibrations of Rice (1972). The inverse of G provides the Mode I stiffness per unit length of aw, relating NI and MI to uI and I . Similar results in Mode II and Mode III complete the elastic stiffness. Stress intensity factors are calculated as
2 3
KI
= (at) 2
1=2
F1
a NI t t
+ F2
a MI ; t
where approximate expressions for F1 and F2 are given by Tada et al. (1973), and similar expressions apply for Mode III and (without F2 ) for Mode II. The J -integral is then calculated by combining these stress intensities as 2 2 2
J
where
E0
= KI + KII + KIII ; E E 2G
0 0
= E=(1 0 2)
and
= E=2(1 + );
Plasticity
The elastic-plastic line spring model in ABAQUS is based on Mode I response only, since no theory is available for an elastic-plastic line spring model in mixed mode loading. For convenience we dene a generalized strain vector as
= 1 uI q2 = 1 I q = q1 q2
q1
= NI Q2 = M I Q1 Q = Q2 :
Q1
3.9.54
The formalism of a simple associated ow, isotropic hardening plasticity model is used as follows. The generalized strain rate is decomposed into elastic and plastic response as
dq = dqel + dqpl ;
and the Mode I elasticity relationship described above is used to dene the generalized stresses:
Q = Del qel:
1
8:
dqpl
where 8 = 8( ; qpl ) is the yield function, whose denition is discussed in detail below, and qpl is a scalar hardening parameter used to provide isotropic hardening. The hardening is calculated from the basic stress-strain data for the material by a work equivalency argument. The plastic work per unit length of aw is
@8 = dqpl @ Q ;
dW pl = Q 1 dqpl ;
Z Z
Elements
dW pl =
0d"pl dzdy;
where 0("pl ) is the uniaxial stress-strain behavior of the material and z and y measure position through the thickness and along the length of the single edge notch specimen. We approximate this second expression by where 0 ("pl ) is a representative value of the yield stress (at an equivalent plastic strain of "pl ); t is the shell thickness; a is the aw depth; and f is a constant, introduced to provide a matching to numerical results for the specimen. Parks and White (1982) suggest choosing f = 0:4, and this value is used in ABAQUS. The yield surface is dened with respect to the generalized stress variables NI and MI as follows. Following Rice (1972), we dene p
dW pl = 0d"pl (t 0 a)2 f;
3 = 2 (t 0 a)2 (MI + t=2NI ): 0 Then for X 2Y the yield function, 8, is taken as an envelope to limit analysis results, as proposed by Rice: X 2 2 81 = (X 0 0:3) + 4:41 Y 0 2 0 0:49 = 0:
Y
3.9.55
and
= 2 (3NI a) 0 t0
Otherwise, we use
82 = 1 X + 2 Y + 3 X 3 Y + 4XY 3 0 1 = 0;
with
This surface is chosen to blend continuously with 81 at X = 2Y and as a reasonable estimate of the behavior for X > 0; Y < 2X . It is, otherwise, arbitrary. Rice (1972) points out that at X = 1:0; Y = 0:5, the yield surface will have a vertex. The smooth surface used in ABAQUS has been adopted for numerical reasons. This smoothness restricts the possible ow behavior at X = 1; Y = 0:5, but we assume that this is not a critical issue. These surfaces are shown in Figure 3.9.53. The gure also indicates a region where the model is not appropriate (because the crack will close). Warning messages are provided if the generalized stress point enters this region at any integration point.
Y= Nt 3 (MI + I ) 2 2 0 (t-a)2 Rice's [1972] envelope limit analysis solutions
0.5
Figure 3.9.53 Generalized stress yield surface assumed for line springs.
3.9.56
The plasticity model is integrated by the usual backward Euler method (see Integration of plasticity models, Section 4.2.2, for details). Once the plastic strain increments are known, from the kinematics of the slip line eld proposed by Rice (1972) for an edge cracked strip, the increment of plastic crack-tip opening is given by
d tip
pl
= dq 1 + (t=2 0 a)dq2 :
pl pl
The increment in the plastic part of the J -integral is related to the increment of plastic crack-tip opening by
dJ
pl
= m0 d
;a
pl tip
= f8 + (8 2 0 a)8 g ; t= 0 1 2
; ;
where ;a indicates differentiation with respect to crack depth and ;1 and ;2 indicate differentiation by Q1 and Q2 , respectively. 8 is either 81 or 82 depending on the state of deformation. The elastic part of the J -integral is obtained from the generalized stresses by calculating the stress intensity factors as described in the previous section (ignoring plasticity effects). The total J -integral is the sum of the plastic and elastic J -integral contributions. Although the method used for computing the elastic contribution is obviously approximate, it is reasonably accurate if J pl dominates J el , which is the case once a signicant amount of plasticity develops (Parks and White, 1982).
Reference
Elements
Line spring elements for modeling part-through cracks in shells, Section 18.5.1 of the ABAQUS Analysis Users Manual
3.9.57
3.9.6
The JOINTC elements in ABAQUS/Standard provide for exible joints between two nodes. This section denes the kinematic variables used in these elements.
Kinematics
A JOINTC element consists of two nodes, referred to here as nodes 1 and 2. Each node has six degrees of freedom: displacements u and rotations . A local orientation is dened for the element by the user. In a large-displacement analysis that local system rotates with the rst node of the element.
z y 2
Elements
= 1 ; 2; 3 .
where ( ) is the rotation matrix dened by the rotation at the rst node of the element. The relative displacements in the element are then
1
e = C e 0 ;
1
u = ( u2 0 u 1 ) 1 e ;
with rst variations
u = ( u2 0 u1 ) 1 e + 1 1 e 2 (u2 0 u1);
3.9.61
where is a linearized rotation eld (see Rotation variables, Section 1.3.1), and second variations du =( u2 0 u1) 1 d 1 2 e + (du2 0 du1) 1 1 2 e 0 d 1 1 1 u + 1 1 1 (u2 0 u1 ) d 1 1 e + 1 1 1 e d 1 1 (u2 0 u1): 2 2
0 e 1 ) 1 (e 2 + e 1 ); 1 2 2
with 1 and 2 dened by cyclic permutation of the local direction indices. These rotation measures dene only relative angular rotations for small relative rotations. They are simple to compute, increase monotonically for relative rotations up to 180, and are taken as suitable for use in the elements for these reasons. The rst variation of 3 is
3=
1 2
(
1 1 2 e2 + d 2 1 e2 2 + 2 1 e2 d 2 1 1 1 1 1 1 1 1 2 1 1 1 1 + 2d 1 e1 0 d 1 e1 0 1 e1 d ) 1 (e2 + e2 ) 1 2 2 2 2 2 2 2 2 2 1 2 + (e1 0 e1 ) 1 (02d 1 e2 + d 1 e2 + 1 e2 d 4 2 0 2d 1 1 1 e1 + d 1 1 e1 1 + 1 1 e1 d 1 ) 2 2 1 1 1 2 2 2 2 1 1 1 1 1 2 2 2 + (0 1 e2 d 1 e1 + 1 e2 d 1 e1 + 1 d (e1 1 e2 0 e1 1 e2 )) 2 1 1 2 2 2 1 1 1 1 1 1 2 2 2 2 + (0d 1 e2 1 e1 + d 1 e2 1 e1 + d 1 (e1 1 e2 0 e1 1 e2 )): 2
and the relative angular velocity about the local 3-axis is taken as
_
3=
1 2
_ (
Virtual work
3.9.62
F = F (u ; u ) _
and
M = M ( ; _ )
(no sum on ):
dW
X u @F
@u
@F @ u _ _ @M @M @ u + + du + d + Fdu + M d ; @ u @u _ @ @ _ @u
where
Reference
Flexible joint elements, Section 17.4 of the ABAQUS Analysis Users Manual
Elements
3.9.63
3.9.7
The MASS and ROTARYI elements allow the inertia of a rigid body to be introduced at a node. In this section the formulation used with these elements is dened. It is assumed that the node at which the mass and rotary inertia are introduced is the center of mass of the body. We refer to the node as the rigid body reference node, C . Let the local principal axes of inertia of the body be e, = 1; 2; 3. Let r be the vector between C and some point in the rigid body with current coordinates x, so that
r=x
0 x C = x e ;
dV:
say
where x are local coordinates in the rigid body. The mass of the rigid body is the integral of the mass density (x ) over the body Z
m=
Since
Z
V
x dV
=0
:
Elements
xx dV
0
=0
for
6= :
1 1 1
Let I11, I22, and I33 be the second moments of inertia of the body about its principal axes e1 , e2 , and e3; Z then
I11 =
I22 = I33 =
The rotary inertia tensor is written
ZV ZV
I=
=1
I e e :
For a rigid body the velocity of any point in the body is given by
_ _ u = uC + !
2 r;
where
2 r + ! 2 (! 2 r):
3.9.71
The local or strong form of the equilibrium equations represents the balance of linear momentum and balance of angular momentum; these two equilibrium equations are
_ I 1! +!
The variational or weak form of equilibrium is
muC = ; f
2 I1! = m:
WA + Wext = 0 :
u u dV =
_ 0muC 1 uC 0 (I 1 ! + ! 2 I 1 ! ) 1 ;
where u = uC + 2 r is the variation of the position of a point in the body. Here uC is the variation of the position of the rigid body reference node, and is the variation of the rotation of the rigid body reference node. The external loading contribution is Wext = uC + m : f
Introducing component expressions relative to the principal axes of inertia, the rotational contribution to the weak form becomes
WA =
0m uC 1 uC 0 10I11 !1 + (I33 0 I22)!2 !3 1 _ 2 _ 0 0I22 !2 + (I11 0 I33)!3 !1 1 3 _ 0 I33 !3 + (I22 0 I11)!1 !2 :
2
When the inertia of a rigid body is used with implicit time integration, the Jacobian contribution of
_ ! ! _ _ 0dWA = m uC 1 duC + 1 I 1 d! + d! 2 I 1 ! + ! 2 I 1 d! + d 2 I 1 ! + I31 (! 2 d ) ! 1 I 1 !)d 0 (! 1 d )I 1 ! + ! 2 I 1 (! 2 d ) : +(
Reference
3.9.72
DISTRIBUTING COUPLING
3.9.8
The distributing coupling elements in ABAQUS/Standard provide a means to connect a reference node to a group of coupling nodes in a way that distributes loads according to weight factors that are prescribed individually at each coupling node. The element distributes forces and moments at the reference node as a coupling nodeforce distribution only. The formulation presented here is also used for the surface-based distributing coupling. This section denes this load distribution relationship and the resultant element development. R The reference node has displacement (u ) and rotation (R ) degrees of freedom. The coupling nodes n ) degrees of freedom active in this element. Each coupling node has a weight have only displacement (u factor wn assigned, which determines the proportion of load carried by the element that is transmitted through the coupling node. Weight factors are dimensionless, and their magnitude is signicant only in a relative sense. Hereafter, normalized weights are used:
wn = ^
wn P wn :
Elements
Load distribution
Let FR and MR be the load and moment applied to the reference node. The statically admissible force distribution Fn among the coupling nodes satises
X n F = FR n X n x 2 Fn = M R + xR 2 F R ; n
(3.9.81)
where xR and xn are the positions of the reference and coupling nodes, respectively. For an arbitrary number of coupling nodes there is no unique solution to Equation 3.9.81. The force distribution adopted in ABAQUS has the property that the linearized motion of the reference node is compatible with the coupling node group motion in an average sense. This compatibility can be described by considering the momentum of a moving coupling node group in a case where weight factors are considered as masses. In this case the reference node motion is identical to that of a point on a rigid body occupying the position of the reference node, where the center of mass of the rigid body is the center of mass of the coupling nodes and the rigid body moves with the same linear and angular momentum as the coupling node group. Since the element mass is distributed this way, the dynamic behavior of the element also has this property.
^ Fn def wn FR + T01 MR = ^
1
2
rn ;
3.9.81
DISTRIBUTING COUPLING
where
^ M R = M R + rR 2 F R ; rn = xn 0 x; Pn wnxn X n n ^ x = P wn = w x ;
n n
T=
X
n
wn
where I is the second-order identity tensor. This force distribution is recognized to be equivalent to the classic bolt-pattern force distribution when the weight factors are interpreted as bolt cross-section areas.
Constraint expression
The load distribution results in the following linearized constraint on node motions:
X n n X n n n xR = w x + T 01 1 ^ w (r 2 x ) ^ n X n ! ! R = T01 1 w n (r n 2 x n ) ; ^ n
where
2 rR
rR = x R 0 x .
Finite motion
Finite displacement and rotation terms take the form of a constraint on the motion of the reference node as a function of the coupling-node nite incremental motions. A measure of the nite rotation of the coupling node arrangement is developed rst and is based on the mid-increment position of the coupling nodes, dened as
n def r0 + 1 1xn; r = n 2
wn
J=
The mid-increment spin is then
X
n
1 R = J01 1
!
:
3.9.82
DISTRIBUTING COUPLING
The nite incremental rotation tensor Hughes and Winget (1980) formula,
1 1 R01 1 I + 1 1R : c 1R = I 0 2 c 2
n R = 1R : 1
This orthogonal tensor yields an incremental nite-rotation vector . From this rotation description comes the constraint expressions for nite displacement and rotation:
1xR =
X n n w 1 x + 1 R 1 r R 0 rR ; ^ 0 0
The virtual work expression for the attached structure is augmented with the contribution of the constraint
5 = 5
3
"
+
(3.9.82)
where 3 is the augmented virtual work expression, is the virtual work expression for the attached structure, and f and m are the respective Lagrange multiplier variables for force and moment.
Initial stress stiffness terms
Elements
The initial stress stiffness terms are derived from a suitable approximation of the exact virtual work expression shown in Equation 3.9.82. This approximation is based on an assumption of innitesimal incremental motions, rn and R , that implies
R = ! R ! R 01
=T
"
X 1 wn (rn 2 xn) : ^
n
and
f3 5 = 5 + f 1
DISTRIBUTING COUPLING
" #
X R f 1 1xR 0 wn1xn 0 1R 1 r0 + rR ^ 0 n i h R R : m 1 1 0 1
! ! R
m 1 T 01 1
"
! ! R
1 d! R rR 0 1 2
1
! ! R
d!R
Mass
n n
wn
1 rR d! R 0 1 d! R 1 rR 2
1 0
! ! R
+
#
The distributing coupling elements also distribute masses to each coupling node according to the weight distribution. A prescribed element mass of M is distributed to the cloud nodes according to
mn
= wnM; ^
where mn is the cloud node mass. Masses are distributed only to the cloud nodes; no mass is associated with the reference node.
References
Distributing coupling elements, Section 17.5 of the ABAQUS Analysis Users Manual Coupling constraints, Section 20.3.2 of the ABAQUS Analysis Users Manual
3.9.84
CONSTITUTIVE MODELS
4.1.1
A wide variety of materials is encountered in stress analysis problems, and for any one of these materials a range of constitutive models is available to describe the materials behavior. For example, a component made from a standard structural steel can be modeled as an isotropic, linear elastic, material with no temperature dependence. This simple model would probably sufce for routine design, so long as the component is not in any critical situation. However, if the component might be subjected to a severe overload, it is important to determine how it might deform under that load and if it has sufcient ductility to withstand the overload without catastrophic failure. The rst of these questions might be answered by modeling the material as a rate-independent elastic, perfectly plastic material, orif the ultimate stress in a tension test of a specimen of the material is very much above the initial yield stressisotropic work hardening might be included in the plasticity model. A nonlinear analysis (with or without consideration of geometric nonlinearity, depending on whether the analyst judges that the structure might buckle or undergo large geometry changes during the event) is then done to determine the response. But the severe overload might be applied suddenly, thus causing rapid straining of the material. In such circumstances the inelastic response of metals usually exhibits rate dependence: the ow stress increases as the strain rate increases. A viscoplastic (rate-dependent) material model might, therefore, be required. (Arguing that it is conservative to ignore this effect because it is a strengthening effect is not necessarily acceptablethe strengthening of one part of a structure might cause load to be shed to another part, which proves to be weaker in the event.) So far the analyst can manage with relatively simple (but nonlinear) constitutive models. But if the failure is associated with localizationtearing of a sheet of material or plastic bucklinga more sophisticated material model might be required because such localizations depend on details of the constitutive behavior that are usually ignored because of their complexity (see, for example, Needleman, 1977). Or if the concern is not gross overload, but gradual failure of the component because of creep at high temperature or because of low cycle fatigue, or perhaps a combination of these effects, then the response of the material during several cycles of loading, in each of which a small amount of inelastic deformation might occur, must be predicted: a circumstance in which we need to model much more of the detail of the materials response. So far the discussion has considered a conventional structural material. We can broadly classify the materials of interest as those that exhibit almost purely elastic response, possibly with some energy dissipation during rapid loading by viscoelastic response (the elastomers, such as rubber or solid propellant); materials that yield and exhibit considerable ductility beyond yield (such as mild steel and other commonly used metals, ice at low strain rates, and clay); materials that ow by rearrangement of particles that interact generally through some dominantly frictional mechanism (such as sand); and brittle materials (rocks, concrete, ceramics). The constitutive library provided in ABAQUS contains a range of linear and nonlinear material models for all of these categories of materials. In general the library has been developed to provide those models that are most usually required for practical applications. There are several distinct models in the library; and for the more commonly encountered materials (metals, in particular), several ways of modeling the material are provided, each suitable to a particular type of analysis application. But the library is far from comprehensive: the range of physical material behavior is far too broad for this ever to be possible. The analyst must review the material denitions provided in ABAQUS in the context of each particular application. If there is no model in the library that is useful for a particular
Constitutive Theories
4.1.11
CONSTITUTIVE MODELS
case, ABAQUS/Standard contains a user subroutineUMATand, similarly, ABAQUS/Explicit contains a user subroutineVUMAT. In these routines the user can code a material model (or call other routines that perform that task). This user subroutine capability is a powerful resource for the sophisticated analyst who is able to cope with the demands of programming a complex material model. Theoretical aspects of the material models that are provided in ABAQUS are described in this chapter, which is intended as a denition of the details of the material models that are provided: it also provides useful guidance to analysts who might have to code their own models in UMAT or VUMAT. From a numerical viewpoint the implementation of a constitutive model involves the integration of the state of the material at an integration point over a time increment during a nonlinear analysis. (The implementation of constitutive models in ABAQUS assumes that the material behavior is entirely dened by local effects, so each spatial integration point can be treated independently.) Since ABAQUS/Standard is most commonly used with implicit time integration, the implementation must also provide an accurate material stiffness matrix for use in forming the Jacobian of the nonlinear equilibrium equations; this is not necessary for ABAQUS/Explicit. The mechanical constitutive models that are provided in ABAQUS often consider elastic and inelastic response. The inelastic response is most commonly modeled with plasticity models. Several plasticity models are described in this chapter. Some of the constitutive models in ABAQUS also use damage mechanics concepts, the distinction being that in plasticity theory the elasticity is not affected by the inelastic deformation (the Youngs modulus of a metal specimen is not changed by loading it beyond yield, until the specimen is very close to failure), while damage models include the degradation of the elasticity caused by severe loading (such as the loss of elastic stiffness suffered by a concrete specimen after it has been subjected to large uniaxial compressive loading). In the inelastic response models that are provided in ABAQUS, the elastic and inelastic responses are distinguished by separating the deformation into recoverable (elastic) and nonrecoverable (inelastic) parts. This separation is based on the assumption that there is an additive relationship between strain rates:
el pl _ _ _ " = " +" ; el pl
(4.1.11)
_ _ _ where " is the total strain rate, " is the rate of change of the elastic strain, and " is the rate of change of inelastic strain. A more general assumption is that the total deformation, F, is made up of inelastic deformation followed by purely elastic deformation (with the rigid body rotation added in at any stage in the process):
F = Fel Fpl :
1
(4.1.12)
In The additive strain rate decomposition, Section 1.4.4, the circumstances are discussed under which Equation 4.1.11 is a legitimate approximation to Equation 4.1.12. We conclude that, if 1. the total strain rate measure used in Equation 4.1.11 is the rate of deformation:
_ " = sym(L) = sym
F 1 F 01
_
= sym
@v @x
where v is the velocity and x is the current spatial position of a material point; and 2. the elastic strains are small,
4.1.12
CONSTITUTIVE MODELS
then the approximation is consistent. ABAQUS uses the rate of deformation as the strain rate measure in nite-strain problems for this reason. (The distinction between different strain measures matters only when the strains are not negligible compared to unity; that is, in nite-strain problems.) The elastic strains always remain small for many materials of practical interest; for example, the yield stress of a metal is typically three orders of magnitude smaller than its elastic modulus, implying elastic strains of order 1003 . However, some materials (polymers, for example) can undergo large elastic straining and also ow inelastically, in which case the additive strain rate decomposition is no longer a consistent approximation. Various elastic response models are provided in ABAQUS. The simplest of these is linear elasticity:
= Del : " el ;
where el is a matrix that may depend on temperature but does not depend on the deformation (except when such dependency is introduced in damage models). This elasticity model is intended to be used for small-strain problems or to model the elasticity in an elastic-plastic model in which the elastic strains are always small. An extension of the elastic type of behavior is the hypoelastic model:
_ _ = Del : " el ;
where now el may depend on the deformation. In this case the elasticity may be nonlinear, but the implementation in ABAQUS still assumes that the elastic strains will always be small. In porous and granular media, the elastic properties are strongly dependent on the volume strain; porous elastic behavior is described in Porous elasticity, Section 4.4.1. The most general type of nonlinear elastic behavior is the hyperelastic model, in which we assume that there is a strain energy density potentialU from which the stresses are dened (to within a hydrostatic stress value if the material is fully incompressible) by
=
@U ; " @"
Constitutive Theories
where and " are any work conjugate stress and strain measures. This form of elasticity model is generally used to model elastomers: materials whose long-term response to large deformations is fully recoverable (elastic). The hyperelasticity modeling provided in ABAQUS is described in Large-strain elasticity, Section 4.6. The hyperelasticity models cannot be used with the plastic deformation models in the program, but can be combined with viscoelastic behavior, as described in Finite-strain viscoelasticity, Section 4.8.2. The plasticity models offered in ABAQUS are discussed in general terms in Plasticity overview, Section 4.2. Both rate-independent and rate-dependent models, with and without yield surfaces, are offered. Models are included in the program that are intended for applications to metals (Metal plasticity, Section 4.3) as well as some nonmetallic materials such as soils, polymers, and crushable foams (Plasticity for non-metals, Section 4.4). The jointed material model (Constitutive model for jointed materials, Section 4.5.4) and the concrete model (An inelastic constitutive model for concrete, Section 4.5.1) also include plasticity modeling.
4.1.13
CONSTITUTIVE MODELS
The constitutive routines in ABAQUS exist in a library that can be accessed by any of the solid or structural elements. This access is made independently at each constitutive calculation point. These points are the numerical integration points in the elements. Thus, the constitutive routines are concerned only with a single calculation point. The element provides an estimate of the kinematic solution to the problem at the point under consideration. These kinematic data are passed to the constitutive routines as the deformation gradientFor, more typically, as the strain and rotation increments " and R. The constitutive routines obtain the state at the point under consideration at the start of the increment from the material point data base. The state variables include the stress and any state variables used in the constitutive modelsplastic strains, for example. The constitutive routines also look up the constitutive denition. Their function then is to update the state to the end of the increment and, if the procedure uses implicit time integration and if Newtons method is being used to solve the equations, to dene the " material contribution to the Jacobian matrix, @ =@" . For material models that are dened in rate form and, therefore, require integration (such as incremental plasticity models), this Jacobian contribution depends on the model and also on the integration method used for the model. Its derivation is, therefore, discussed in some detail in the sections that dene such models.
Reference
Material library: overview, Section 9.1.1 of the ABAQUS Analysis Users Manual
4.1.14
PLASTICITY MODELS
4.2.1
Incremental plasticity theory is based on a few fundamental postulates, which means that all of the elasticplastic response models provided in ABAQUS (except the deformation theory model in ABAQUS/Standard, which is primarily provided for fracture mechanics applications) have the same general form. The basic equations of the models are dened in their general form in this section. Plasticity models are written as rate-independent models or as rate-dependent models. A rateindependent model is one in which the constitutive response does not depend on the rate of deformation the response of many metals at low temperatures relative to their melting temperature and at low strain rates is effectively rate independent. In a rate-dependent model the response does depend on the rate at which the material is strained. Examples of such models are the simple metal creep models provided in ABAQUS/Standard and the rate-dependent plasticity model that is used to describe the behavior of metals at higher strain rates. Because these models have similar forms, their numerical treatment is based on the same technique. A basic assumption of elastic-plastic models is that the deformation can be divided into an elastic part and an inelastic (plastic) part. In its most general form this statement is written as
where F is the total deformation gradient, F is the fully recoverable part of the deformation at the point under consideration ([Fel ]01 is the deformation that would occur if, after the deformation F, inelastic response were somehow prevented but at the same time the stress at the point reduced to zero), and Fpl is dened by Fpl = [Fel ]01 1 F. The rigid body rotation at the point can be included in the denition of either Fel or Fpl or can be considered separately before or after either part of the decomposition: this makes no difference except in the convenience of the basis for writing the parts of the deformation. This decomposition can be used directly to formulate the plasticity model. Historically, an additive strain rate decomposition,
el
F = Fel Fpl ;
1
(4.2.11)
Constitutive Theories
"
_ _ _ has been used in its place. Here " is the total (mechanical) strain rate, " is the elastic strain rate, and "pl is the plastic strain rate. It is shown in The additive strain rate decomposition, Section 1.4.4, that Equation 4.2.12 is a consistent approximation to Equation 4.2.11 when the elastic strains are innitesimal (negligible compared to unity) and when the strain rate measure used in Equation 4.2.12 is the rate of deformation:
el
_ = "el + " pl ; _ _
(4.2.12)
"
_ = sym @@x
:
Equation 4.2.12, with the rate of deformation used as the denition of total strain rate, is used in all of the plasticity models that are implemented in ABAQUS. Rices argument implies that the elastic response must always be small in problems in which these models are used. In practice this is the case: plasticity models are provided for metals, soils, polymers, crushable foams, and concrete; and in each of these materials it is
4.2.11
PLASTICITY MODELS
very unlikely that the elastic strain would ever be larger than a few percent (and even this would be quite unusual in a metal). Thus, the use of Equation 4.2.12 does not appear to be objectionable for the models in question, at least from a formal point of view. However, the user who needs to develop user subroutine UMAT or VUMAT for a different material model in which the elastic strains and the inelastic strains may both be large should consider using Equation 4.2.11 directly. The elastic part of the response is assumed to be derivable from an elastic strain energy density potential, so the stress is dened by
where U is the strain energy density potential. Since we assume that, in the absence of plastic straining, the variation of elastic strain is the same as the variation in the rate of deformation, conjugacy arguments dene the stress measure as the true (Cauchy) stress. All stress output in ABAQUS is given in this form. In some materials the elastic response is essentially incompressible. But this is not usually the case for the materials whose inelastic deformation is considered with the models provided in ABAQUS, so Equation 4.2.13 can be taken to dene the stress completely. However, the inelastic response is often assumed to be approximately incompressible (in metals, for example, or in soils undergoing large plastic ows), so the user must be careful to ensure that the elements chosen can accommodate incompressible response without exhibiting locking problems when the model is three-dimensional, plane strain, or axisymmetric. This requires the use of hybrid or fully or selectively reduced integration elements. For several of the plasticity models provided in ABAQUS the elasticity is linear, so the strain energy density potential has a very simple form. For the soils model the volumetric elastic strain is proportional to the logarithm of the equivalent pressure stress. For the concrete model damaged elasticity is used to account for crack opening after the concrete has cracked: in that case the elasticity model is more complex. The rate-independent plasticity models in ABAQUS and one of the rate-dependent models all have a region of purely elastic response. The yield function, f , denes the limit to this region of purely elastic response and is written so that for purely elastic response. Here is the temperature, and H are a set of hardening parameters. The subscript is introduced simply to indicate that there may be several hardening parameters, H: the range of is not specied until we dene a particular plasticity model. The hardening parameters are state variables that are introduced to allow the models to describe some of the complexity of the inelastic response of real materials. In the simplest plasticity model (perfect plasticity) the yield surface acts as a limit surface and there are no hardening parameters at all: no part of the model evolves during the deformation. Complex plasticity models usually include a large number of hardening parameters. The models provided in ABAQUS are generally not the most complex models and use only a few such parameters (only one is used in the isotropic hardening metal model and in the Cam-clay model; six are used in the simple kinematic hardening model). In the concrete and the jointed material plasticity models in ABAQUS the yield behavior is modeled with several independent inelastic ow systems. For this case Equation 4.2.14 is generalized to
=
@U ; " @"el
(4.2.13)
f ( ; ; H ) < 0
(4.2.14)
fi < 0
4.2.12
PLASTICITY MODELS
for purely elastic response in system i, where fi ( ; Hi;) is one of the yield functions and Hi; are the hardening parameters for the ith system. For generality in this discussion we assume the model uses such a system of independent yield functions. In the simpler systems with a single yield function i can only take the value 1. Stress states that cause the yield function to have a positive value cannot occur in rate-independent plasticity models, although this is possible in a rate-dependent model. Thus, in the rate-independent models we have the yield constraints
fi = 0
during inelastic ow. When the material is owing inelastically the inelastic part of the deformation is dened by the ow rule, which we can write as
" d"pl =
X
i
di
@gi ; @
(4.2.15)
where gi ( ; ; Hi;) is the ow potential for the ith system and di is the rate of change of time, dt, for a rate-dependent model or is a scalar measuring the amount of the plastic ow rate on the ith system, whose value is determined by the requirement to satisfy the consistency condition fi = 0, for plastic ow of a rate-independent model. The summation is over only the actively yielding systems: di = 0 for those systems for which fi < 0. The form in which the ow rule is written above assumes that there is a single ow potential, gi , in the ith system. More general plasticity models might have several active ow potentials at a point. This is, for instance, the case in the concrete and jointed material models built into ABAQUS. For some rate-independent plasticity models the direction of ow is the same as the direction of the outward normal to the yield surface:
@gi @fi = ci ; @ @
where ci is a scalar. Such models are called associated ow plasticity models. Associated ow models are useful for materials in which dislocation motion provides the fundamental mechanisms of plastic ow when there are no sudden changes in the direction of the plastic strain rate at a point. They are generally not accurate for materials in which the inelastic deformation is primarily caused by frictional mechanisms. The metal plasticity models in ABAQUS (except cast iron) and the Cam-clay soil model use associated ow. The cast iron, granular/polymer, crushable foam, Mohr-Coulomb, Drucker-Prager/Cap, and jointed material models provide nonassociated ow with respect to volumetric straining and equivalent pressure stress. The concrete model uses associated ow. The rate form of the ow rule is essential to incremental plasticity theory, because it allows the history dependence of the response to be modeled. The nal ingredient in plasticity models is the set of evolution equations for the hardening parameters. We write these equations as
dHi; = di hi;( ; ; Hi; );
Constitutive Theories
(4.2.16)
where hi; is the (rate form) hardening law for Hi;. In complex plasticity modelsfor example, models used to describe the cyclic behavior of metals used for high temperature applicationsthese evolution
4.2.13
PLASTICITY MODELS
laws have complicated forms, since such complexity is required to match the experimentally observed behavior. The plasticity models offered in ABAQUS use simple evolution equations: isotropic hardening, Prager-Ziegler kinematic hardening, and the location of the center of the yield surface along the equivalent pressure stress axis in the Cam-clay model. The independence of the yield systems designated by the subscript is implicit in the assumption in Equation 4.2.16 above that the evolution of the i; depends only on other hardening parameters, i; , in the same ( th) system. Equation 4.2.11 to Equation 4.2.16 dene the general structure of all of the plasticity models in ABAQUS. Since the ow rule and the hardening evolution rules are written in rate form, they must be integrated. The general technique of integration is discussed in Integration of plasticity models, Section 4.2.2. The sections immediately following that discussion describe the details of the specic plasticity models that are provided in ABAQUS.
Reference
4.2.14
PLASTICITY INTEGRATION
4.2.2
The plasticity models provided in ABAQUS have been described in general terms in Plasticity models: general discussion, Section 4.2.1. The only rate equations are the evolutionary rule for the hardening, the ow rule, and the strain rate decomposition. The simplest operator that provides unconditional stability for integration of rate equations is the backward Euler method: applying this method to the ow rule (Equation 4.2.15) gives
1"pl =
X1
i
i
@gi ; @
(4.2.21)
1Hi; = 1ihi;: +1
(4.2.22)
In these equations and throughout the remainder of this section any quantity not specically associated t). The strain rate decomposition, with a time point is taken at the end of the increment (at time t Equation 4.2.12, is integrated over a time increment as
@ 1x
We integrate the total values of each strain measure as the sum of the value of that strain at the start of the increment, rotated to account for rigid body motion during the increment, and the strain increment. The rotation to account for rigid body motion during the increment is dened approximately using the algorithm of Hughes and Winget (1980). This integration allows the strain rate decomposition to be integrated into
Constitutive Theories
(4.2.23)
From a computational viewpoint the problem is now algebraic: we must solve the integrated equations of the constitutive model for the state at the end of the increment. The set of equations that dene the algebraic problem are the strain decomposition, Equation 4.2.23; the elasticity, Equation 4.2.13; the integrated ow rule, Equation 4.2.21; the integrated hardening laws, Equation 4.2.22; and for rate independent models, the yield constraints
). for active systems (systems in which fi < have i We assume that the ow surface is sufciently smooth so that its (second) derivatives with respect to stress and the hardening parameters are well-dened. This is generally true for the models in ABAQUS:
fi = 0 ;
(4.2.24)
1 =0
4.2.21
PLASTICITY INTEGRATION
the exceptions occur at corners or vertices of the surfaces. These special cases are handled individually when they arise. For some plasticity models the algebraic problem can be solved in closed form. For other models it is possible to reduce the problem to a one variable or a two variable problem that can then be solved to give the entire solution. For example, the Mises yield surfacewhich is generally used for isotropic metals, together with linear, isotropic elasticityis a case for which the integrated problem can be solved exactly or in one variable (see Isotropic elasto-plasticity, Section 4.3.2). For other rate-independent models with a single yield system the algebraic problem is considered to be a problem in the components of "pl . Once these have been foundthe elasticitytogether with the integrated strain rate decompositiondene the stress. The ow rule then denes and the hardening laws dene the increments in the hardening variables. We now derive the equations for the Newton solution of the integrated problem for the case of rateindependent plasticity with a single yield system. The rate-dependent problem with a single yield system is solved in a similar way. For the particular cases of multiple, independent, yield systems (concrete and jointed material) particular techniques are used for this algebraic solution, taking advantage of the simplications available in those particular models. The concrete model and its integration are described in An inelastic constitutive model for concrete, Section 4.5.1, and the jointed material model is described in Constitutive model for jointed materials, Section 4.5.4. During the solution, the elasticity relationship and the integrated strain rate decomposition are satised exactly, so that
where is the correction to the stress, " is the correction to the plastic strain increments, and
U Del = @"@el@"el " "
2
c = 0Del : c";
(4.2.25)
is the tangent elasticity matrix. The hardening laws are also satised exactly (because the increments of the hardening parameters are dened from these laws) so that
c
= h c + 1
@h @
where c is the correction to H and c is the correction to This set of equations can be rewritten
c
: c + @H c
@h
1 .
where
^ =
^ = H c + w : c ; ^
@h 0 1 @H
(4.2.26)
h
01
and
w = 1 ^
0 1 @H
@h
01 @h
@
4.2.22
PLASTICITY INTEGRATION
The ow rule is not satised exactly until the solution has been found, so it gives the Newton equations
@g c" 0 c @ 0 1
@ 2g @ @
g @g : c + @@@H c = 1 @ 0 1"pl:
Using Equation 4.2.25 and Equation 4.2.26 allows these equations to be rewritten as
(4.2.27)
where
and
g ^ n = @ + 1 @@@H H: ^ @g
2
Likewise, the yield condition is not satised exactly during the Newton iteration, so
@f @
@f : c + @H c = 0f:
@f ^ m : Del : c" 0 @H Hc = f; ^
where
(4.2.28)
We now eliminate c between Equation 4.2.27 and Equation 4.2.28. Taking Equation 4.2.27 along
^ @f @f ^ m = @ + @H w:
Constitutive Theories
^^ Z = I 0 1 n m : Del; d which is a set of linear equations solved for the c" . The solution is then updated and the Newton loop
where
4.2.23
PLASTICITY INTEGRATION
The solution for rate-dependent plasticity models with a single yield function is developed in the same way, the only differences being the lack of a yield constraint and the identication of with time.
Tangent matrix
" The tangent matrix for the material, @ =@" , is required when ABAQUS/Standard is being used for implicit time integration and Newtons method is being used to solve the equilibrium equations. The matrix is obtained directly by taking variations of the integrated equations with respect to all solution parameters, and then solving for the relationship between and ". The procedure closely follows the derivation used above for the Newton solution: the result is the tangent matrix = D : ";
where
^ D = I + 1 Del : Z : N
i01
: Del : Z:
Reference
4.2.24
METAL PLASTICITY
4.3.1
ABAQUS offers several models for metal plasticity analysis. The main options are a choice between rate-independent and rate-dependent plasticity, a choice between the Mises yield surface for isotropic materials and Hills yield surface for anisotropic materials, and for rate-independent modeling a choice between isotropic and kinematic hardening. Special plasticity theories are the cast iron model (Cast iron plasticity, Section 4.3.7), the ORNL model for types 304 and 316 stainless steel in nuclear applications (ORNL constitutive theory, Section 4.3.8), and deformation plasticity for fracture mechanics applications (Deformation plasticity, Section 4.3.9). Rate-independent plasticity is used mostly in modeling the response of metals and some other materials at low temperature (typically below half the melting temperature on an absolute scale) and low strain rates. The rate-independent metal plasticity model uses associated ow. Two types of rate-dependent models are offered. In the rst type a rate-dependent yield strength is introduced in the material model. This is intended for relatively high strain rate applications, such as dynamic events or metal forming process simulations. This type of rate dependence can be introduced in different ways. One way is to use an overstress power law,
_ " =D
pl pl
01 0
n
for
0 ;
_ where " is the equivalent plastic strain rate; is the yield stress at nonzero plastic strain rate; 0 ("pl ; ; fi ) is the static yield stress (which may depend on the plastic strain"pl via isotropic hardening, on the temperatureand on other eld variables, fi ); and D(; fi ), n(; fi ) are material parameters that can be functions of temperature and, possibly, of other predened state variables. Another way is to dene _ pl a yield stress ratio, =0 , as a function of the equivalent plastic strain rate, " . Both of these options assume that the shapes of the hardening curves at different strain rates are identical. If the shapes of the hardening curves at different strain rates are different, the static and rate-dependent stress-strain relations can be specied directly. The yield stress at a given strain rate is interpolated directly from these relations. Finally, the user can describe general rate-dependent isotropic hardening with user subroutine UHARD. See Symonds (1967), Lindholm and Besseny (1969), and Eleiche (1972) for collections of material response measurements or bibliographies of such measurements at high strain dependents. For high temperature creep problems ABAQUS/Standard offers some simple built-in creep laws. But for many practical problems the user must write the uniaxial creep behavior into user subroutine CREEP because of the complexity of the experimentally measured material response. Creep response under cyclic loading shows signicant Bauschinger effects, which cannot be modeled except by introducing sophisticated hardening models. The only capability in ABAQUS for such cases is the ORNL option. This option uses simple rules to model the Bauschinger effect and is intended primarily as a design evaluation model for the high temperature response of stainless steel. It does not model the materials response in detail. User subroutine UMAT must be used if that hardening model is not adequate. Isotropic hardening means that the yield function is written
f ( ) = 0 ("pl ; );
Constitutive Theories
4.3.11
METAL PLASTICITY
where 0 is the equivalent (uniaxial) stress, "pl is the work equivalent plastic strain, dened by
0"pl _
=
pl _ :" ;
and is temperature. Isotropic hardening is generally considered to be a suitable model for problems in which the plastic straining goes well beyond the incipient yield state where the Bauschinger effect is noticeable (Rice, 1975). Therefore, this hardening theory is used for such applications as dynamic problems involving nite strains and manufacturing processesany process involving large plastic strain and in which the plastic strain does not continuously reverse direction sharply. Some cases, such as low-cycle fatigue situations, involve relatively low amplitude strain cycling. In these cases it becomes important to model the Bauschinger effect. Kinematic hardening is the simplest theory that does this. ABAQUS offers a linear kinematic and a nonlinear isotropic/kinematic hardening model for such cases. These models are described in Models for metals subjected to cyclic loading, Section 4.3.5.
Reference
4.3.12
ISOTROPIC ELASTO-PLASTICITY
4.3.2
ISOTROPIC ELASTO-PLASTICITY
This material model is very commonly used for metal plasticity calculations, either as a rate-dependent or as a rate-independent model, and has a particularly simple form. Because of this simplicity the algebraic equations associated with integrating the model are easily developed in terms of a single variable, and the material stiffness matrix can be written explicitly. This results in particularly efcient code. In this section these equations are developed. For simplicity of notation all quantities not explicitly associated with a time point are assumed to be evaluated at the end of the increment. The Mises yield function with associated ow means that there is no volumetric plastic strain; since the elastic bulk modulus is quite large, the volume change will be small. Thus, we can dene the volume strain as "vol = trace("); and, hence, the deviatoric strain is
e = " 0 "vol I:
1 3
Using the standard denition of corotational measures, this can be written in integrated form as
(4.3.21)
Constitutive Theories
The elasticity is linear and isotropic and, therefore, can be written in terms of two temperaturedependent material parameters. For the purpose of this development it is most appropriate to choose these parameters as the bulk modulus, K , and the shear modulus, G. These are computed readily from the users input of Youngs modulus, E , and Poissons ratio, , as
K
and
3(1
0 2 )
:
G=
E
2(1 + )
The elasticity can be written in volumetric and deviatoric components as follows. Volumetric:
p = 0K"vol ;
(4.3.22)
4.3.21
ISOTROPIC ELASTO-PLASTICITY
where
p
1 = 0 3 trace()
S
= 2G eel ; = + p I: = depl n;
(4.3.23)
The ow rule is
depl
(4.3.24)
where
n
and depl is the (scalar) equivalent plastic strain rate. The plasticity requires that the material satisfy a uniaxial-stress plastic-strain strain-rate relationship. If the material is rate independent, this is the yield condition:
q
= 2 S : S;
r3
3 = 2 S; q
where 0 epl ; is the yield stress and is dened by the user as a function of equivalent plastic strain (epl ) and temperature (). If the material is rate dependent, the relationship is the uniaxial ow rate denition:
( )
= 0 ;
(4.3.25)
where h is a known function. For example, the rate-dependent material model offers an overstress power law model of the form
e
_ pl = D
01
where D and n are user-dened temperature-dependent material parameters and 0 epl ; is the static yield stress. Integrating this relationship by the backward Euler method gives
()
()
( )
(4.3.26)
This equation can be inverted (numerically, if necessary) to give q as a function of epl at the end of the increment.
4.3.22
ISOTROPIC ELASTO-PLASTICITY
Thus, both the rate-independent model and the integrated rate-dependent model give the general uniaxial form
where 0 for the rate-independent model and is obtained by inversion of Equation 4.3.26 for the rate-dependent model. Equation 4.3.21 to Equation 4.3.27 dene the material behavior. In any increment when plastic ow is occurring (which is determined by evaluating q based on purely elastic response and nding that its value exceeds 0 ), these equations must be integrated and solved for the state at the end of the increment. As in the general discussion in Metal plasticity models, Section 4.3.1, the integration is done by applying the backward Euler method to the ow rule (Equation 4.3.24), giving
q = (pl ); e
(4.3.27)
1epl = 1pl n: e
S = 2G(eel t + 1e 0 1pl n): e
(4.3.28)
Combining this with the deviatoric elasticity (Equation 4.3.23) and the integrated strain rate decomposition (Equation 4.3.21) gives (4.3.29)
Using the integrated ow rule (Equation 4.3.28), together with the Mises denition of the ow direction, n (in Equation 4.3.24), this becomes
^ = eelt + 1e; e
Constitutive Theories
G ^ e (1 + 3q 1pl )S = 2Ge:
Taking the inner product of this equation with itself gives
(4.3.210)
q + 3G1pl = 3G~; e e
where
(4.3.211)
e= ~
The Mises equivalent stress, q, must satisfy the uniaxial form dened in Equation 4.3.27, so that from Equation 4.3.211,
2e : ^ : 3^ e
3G(~ 0 1pl ) 0 = 0: e e
4.3.23
(4.3.212)
ISOTROPIC ELASTO-PLASTICITY
This is a nonlinear equation for epl in the general case when depends on the equivalent plastic strain (that is, when the material is rate-dependent, or when there is nonzero work hardening). (It is linear in epl for rate-independent perfect plasticity.) We solve it by Newtons method:
1
cpl
where
= dd ; epl
1
q
and so, from Equation 4.3.210,
= ;
S=
2G e 3G 1pl ^: e 1+
q
n=
3 S; 2q
1epl = 1pl n: e
p
For cases where three direct strain components are provided by the kinematic solution (that is, all but plane stress and uniaxial stress cases), Equation 4.3.22 denes
= 0K"vol ;
For plane stress "33 is not dened by the kinematics but by the plane stress constraint
= 0: This additional equation (or equivalently p = S33 ) must be solved along with the yield condition
33 "33 " ^ = 0 1 0 (^11 + "22); ^ "11
where
^ = "el t + 1"11 11
4.3.24
ISOTROPIC ELASTO-PLASTICITY
and
serves as an initial guess toward the nal value of "33 that enables (with the correct plastic straining) the plane stress condition to be satised.
Uniaxial stress
For cases with only one direct strain component dened by the kinematic solution (uniaxial deformation), we require
1 2
so that
3 2
(4.3.213)
Material stiffness
For this simple plasticity model the material stiffness matrix can be derived without the need for matrix inversion (as was needed in the general case described in Integration of plasticity models, Section 4.2.2), as follows. Taking the variation of Equation 4.3.210 with respect to all quantities at the end of the increment gives
G G e e (1 + 3q 1pl )@ S + S 3q (@epl 0 1 q
Now, from Equation 4.3.25, and, from Equation 4.3.211,
pl
@q) = 2G @e: ^
(4.3.214)
Constitutive Theories
= 3G @e: ~
1 ~ = 1 + B @e;
B= H 3G :
where
@e = ~
2^ ^ 3~e : @e; e
4.3.25
ISOTROPIC ELASTO-PLASTICITY
and, hence,
@ epl =
2 1 e : @ e: 3 e(1 + B ) ^ ^ ~
@S = Q = 0 R S S
where Q
^ : @ e;
For all cases where three direct strains are dened by the kinematic solution, the material stiffness is completed by
" @p = 0K I : @"; = S0 pI
so since and
^ @e =
" = 0 1 I I : @" ; 3
1
we have
0 1 1 @ = Q = + K 0 Q I I 0 R S S 3
" : @":
Plane stress
For the plane stress case the material stiffness matrix is found by imposing
@33 = 0
on the general material stiffness matrix obtained for the plane strain case.
Uniaxial stress
For the uniaxial stress case the material stiffness matrix is available directly from the variation of Equation 4.3.213 as
2 @11 = [ Q 0 R 11] @"11:
3 2
Reference
Classical metal plasticity, Section 11.2.1 of the ABAQUS Analysis Users Manual
4.3.26
STRESS POTENTIALS
4.3.3
The metal plasticity models in ABAQUS use the Mises stress potential for isotropic behavior and the Hill stress potentials for anisotropic behavior. Both of these potentials depend only on the deviatoric stress, so the plastic part of the response is incompressible. This means that, in cases where the plastic ow dominates the response (such as limit load calculations or metal forming problems), except for plane stress problems, the nite elements should be chosen so that they can accommodate the incompressible ow. Usually the reduced integration elements are used for this purpose: in ABAQUS/Standard the hybrid elements can also be used, at higher cost. The fully integrated rst-order continuum elements in ABAQUS/Standard use selectively reduced integration, whereby the volumetric strain is calculated at the centroid of the element only. Those elements that are described in Solid isoparametric quadrilaterals and hexahedra, Section 3.2.4, are also suitable for such problems. The Mises stress potential is f ( ) = q; where
q=
in which S is the deviatoric stress:
S=
r3
2
S : S;
0 1 trace( )I = 0 1 I I : ( ): 3 3
in which = is the fourth-order unit tensor. Hills stress function is a simple extension of the Mises function to allow anisotropic behavior. The function is
The potential is a circle in the plane normal to the hydrostatic axis in principal stress space. For this function, @f 1 3 = S; @ q 2 and 1 3 1 @2f = = 0 2 I I 0 @f @f @ @ q 2 @ @
Constitutive Theories
f ( ) =
in terms of rectangular Cartesian stress components, where F; G; H;L; M; N are constants obtained by tests of the material in different orientations. They are dened as
F= G=
0 1 2 0 1 2
11
); );
22
4.3.31
STRESS POTENTIALS
2 0 1 1 1 ( 2 + 2 0 2 ); 2 11 22 33 3 2 L= ( 0) ; 2 23 3 0 2 M= ( ); 2 13 3 2 N= ( 0); 2 12
H=
where 0; 11; 22; 33; 12; 23; 13 are specied by the user and 0 = 0 = 3. and are the values of stress that make the potential equal to 0 if only one stress component is nonzero. For this function 1 @f = b;
@
where
In addition,
@b
@
2Lyz
0 f2 b b
;
0 0 0 0 0 0 7 7 0 7 7: 0 7 5 0 2L
where
2G + H 6 0H @ b 6 0G 6 =6 @ 6 0 4 0
0
F +H 0F
0 0 0
0H
0 0 F +G 0 0 2N 0 0 0 0
0G 0F
2M 0
Reference
4.3.32
4.3.4
The rate-dependent plasticity (creep) models provided in ABAQUS/Standard are used to model inelastic straining of materials that are rate sensitive. High-temperature creep in structures is one important class of examples of the application of such a material model. Because such problems generally involve relatively small amounts of inelastic straining (otherwise the structure is not a suitable design), the explicit, forward Euler method is often satisfactory as an integrator for the ow rule. This method is only conditionally stable, but the stability limit is usually sufciently large compared to the time history of interest in such cases that the explicit method is very economical. Cormeau (1975) has developed formulae for the stability limit for most common cases of stress induced creep, and these results are used to monitor stability. For this explicit approach the integration is trivial. Combining the integrated ow rule
1"pl = 1t @Gcr jt @
with the integrated strain rate decomposition and the (linear) elasticity gives
jt+1t = D
el
(4.3.41)
All of the terms on the right-hand side of this set of equations are known when the constitutive integration is done, so these equations dene t+1t explicitly. There also exist many problems involving rate-dependent plastic response in which the characteristic relaxation times for the material under the stress states to which it is subjected are very short compared to the time period of interest in the analysis, so the conditional stability of the explicit operator will only allow very short time increments. For such cases it can be more economical to use the backward Euler method because of its unconditional stability. ABAQUS always uses the implicit method for high strain rate applications to avoid time increment restrictions being introduced by considerations of stability in the integration of the constitutive model. ABAQUS will also use the implicit method in all geometrically nonlinear problems and in problems for which rate-independent plasticity is active simultaneously. The backward Euler method is implicit; and because the plastic strain rate is usually a strong function of stress, some care must be taken to develop an effective algorithm to solve the nonlinear algebraic equations that result from the use of this operator. The problem has been posed formally in Integration of plasticity models, Section 4.2.2. The main difculty is to obtain a reasonable starting guess for "pl . For this we proceed as follows. For simplicity, we consider rate-dependent behavior only and the particular form of ow rule dened by
Constitutive Theories
1 I + "cr n; _ 3 _ sw _ cr where " is the equivalent swelling strain rate, " is the equivalent creep strain rate, and n is the gradient of the deviatoric stress potential, ~ n = @q ; _ _ "pl = "
sw
@
4.3.41
where q is the Mises or Hill stress potential (dened in Stress potentials for anisotropic metal plasticity, Section 4.3.3). The equivalent strain rates are part of the stress potential for the plastic response and, therefore, are assumed to have evolution laws of the form
"
and
"
(4.3.42)
and
1"
and " are usually dened in user subroutine CREEP. The solution to the algebraic problem is obtained by rst nding reasonable initial guesses for and "sw and then solving the full problem. The Mises and Hill equivalent stress denitions (q) both have the property that
cr
(4.3.43)
1"cr
~ n : = q: ~
: = 03p:
The initial estimates for "cr and "sw are obtained by projecting the problem onto nel and I, where n is @ q =@ dened at el , the stress state that would arise at the end of the increment if there were no inelastic deformation during the increment. The projections are
el
(4.3.44)
and
p
(4.3.45)
where
~ = n : el ;
4.3.42
and
Equation 4.3.42 to Equation 4.3.45 are a set of nonlinear equations that can be solved for "cr and " . We solve these equations by Newtons method and then use this solution as the starting estimate for solving the complete problem. When the Mises stress potential is used and the problem is not plane stress, this starting estimate is the solution to the complete problem because the Mises stress potential is a circle in the deviatoric plane.
1 nel : Del : I: 3
sw
ABAQUS/Standard provides a very general capability for implementing viscoplastic models such as creep and swelling in which the strain rate potential can be written as a function of equivalent pressure stress, p; the Mises or Hills equivalent deviatoric stress, q; and any number of solution-dependent state variables. The purpose of this section is to provide an overview of the operations that need to be performed in user subroutine CREEP. To illustrate the main ideas, the creep law is assumed to be of a strain hardening type and of the form
_ "
cr
= Bqnf ("cr ); ~
Constitutive Theories
where B and n are material constants, and f is a nonlinear function of its argument. In user subroutine CREEP the user needs to dene the creep strain increment based on the above creep law. Given that the creep law is in rate form, an integration scheme is needed to convert it to an incremental form dening the creep strain increment. This conversion can be accomplished by using either an explicit (forward Euler) or an implicit (backward Euler) integration scheme. In the explicit scheme the creep strain rate during any time increment is dened in terms of (known) quantities at the beginning of the increment. Thus, an explicit integration would lead to the following incremental form of the creep law:
In an implicit scheme the creep strain rate during any time increment is dened in terms of (unknown) quantities at the end of the increment. Thus, an implicit scheme would lead to the following incremental form of the creep law:
+1
4.3.43
Recognizing that "cr 1t = "cr "cr and also that qt+1t is a function of "cr t, the implicit t+ t t+1 integration scheme leads to a nonlinear equation in the creep strain increment "cr that is solved by ABAQUS iteratively at each material point (these are local material point iterations and are not the same as the global equilibrium iterations). The user subroutine gets called at each integration point during each local iteration. The Newton Raphson scheme used to solve the above nonlinear equation iteratively is given by
+1
@h = 0[ @1"cr ]01hc(~t+1t; "cr t)i; q t+1 i cr cr cr 1"i+1 = 1"i + c1" ; where the subscripts i and i + 1 are iteration counters, and c1"cr represents the correction to the creep strain increment. As illustrated by the two equations above, the Jacobian of the implicit scheme requires the partial derivative of the function hc with respect to 1"cr . For the example considered c "cr i+1
There are additional terms if the function hc also depends on the hydrostatic pressure. In an implicit scheme the user also needs to dene the appropriate derivatives that enter the Jacobian of the nonlinear @h c @hc equation for "cr . In the example above the user needs to dene the quantities @~t+1t and @"cr . q t+1t The incremental creep strain and the Jacobian contributions require the values of the equivalent Mises or Hills stress (and the pressure stress, if relevant) and the equivalent creep strain at both the beginning and the end of the increment, which are available in the user subroutine. On the other hand, in the explicit scheme the creep strain increment is dened in terms of quantities known at the beginning of the increment and, hence, no local iterations are needed. However, as discussed in Rate-dependent plasticity: creep and swelling, Section 11.2.4 of the ABAQUS Analysis Users Manual, the explicit scheme has limitations related to stability. Irrespective of the integration scheme used to integrate the rate form of the creep equation, user subroutine CREEP is called at each material point once at the beginning and once at the end of each increment. These calls are for the purpose of getting the creep strain increment based on the creep strain rate at the beginning and at the end of the increment, respectively. The difference between these two creep strain increment values measures the accuracy of the integration scheme and must be less than the value specied with the CETOL parameter on the relevant analysis step option.
Reference
Rate-dependent plasticity: creep and swelling, Section 11.2.4 of the ABAQUS Analysis Users Manual
4.3.44
4.3.5
The kinematic hardening models in ABAQUS are intended to simulate the behavior of metals that are subjected to cyclic loading. These models are typically applied to studies of low-cycle fatigue and ratchetting. The basic concept of these models is that the yield surface shifts in stress space so that straining in one direction reduces the yield stress in the opposite direction, thus simulating the Bauschinger effect and anisotropy induced by work hardening. Two kinematic hardening models are available in ABAQUS. The simplest model provides linear kinematic hardening and is, thus, mainly used for low-cycle fatigue evaluations. This model yields physically reasonable results if the uniaxial behavior is linearized in the plastic range (a constant workhardening slope). This is usually best accomplished by guessing the strain levels that will be attained in the problem and linearizing the actual material behavior accordingly. It is important to recognize this restriction on the theorys ability to provide reasonable results and to provide material data accordingly. This model is available with the Mises or Hill yield surface. The combined isotropic/kinematic hardening model is an extension of the linear model. It provides a more accurate approximation to the stress-strain relation than the linear model. It also models other phenomenasuch as ratchetting, relaxation of the mean stress, and cyclic hardeningthat are typical of materials subjected to cyclic loading. This model is available only with the Mises yield surface. This section rst describes those aspects of the formulation that are common to both models; the specic formulation of each model is presented subsequently.
Strain rate decomposition
_ " is written in terms of the elastic and plastic strain rates as _ _ _ " = "el + " pl :
(4.3.51)
Constitutive Theories
Elastic behavior
where el represents the fourth-order elasticity tensor and strain tensors, respectively.
Plastic behavior
The models are pressure-independent plasticity models. For both models the yield surface is dened by the function
f ( 0 ) = 0 ;
4.3.51
(4.3.53)
where f ( 0 ) is the equivalent Mises stress or Hills potential with respect to the backstress or kinematic shift , and 0 is the size of the yield surface. For instance, the equivalent Mises stress is dened as
f ( 0 ) =
r3
2
(S 0 dev ) : (S 0 dev );
(4.3.54)
where dev is the deviatoric part of the backstress and These models assume associated plastic ow:
This model is the simpler of the two kinematic hardening models available in ABAQUS. The size of the yield surface, 0 (), can be a function of temperature only for this model. The evolution of is dened by Zieglers hardening rule, generalized to the nonisothermal case as
_ ( 0 ) + C; (4.3.56) 0 C where C () is the hardening parameter (C () is the work-hardening slope of the isothermal uniaxial _ stress-strain response, d=d"pl , taken at different temperatures) and C is the rate of change of C with respect to temperature. This form of evolution law for denes the rate of due to plastic straining to be in the direction of the current radius vector from the center of the yield surface, 0 , and _ _ = C "pl
the rate due to temperature changes to be toward the origin of stress space. Rice (1975) writes this concept quite generally as
_ = ( 0 ) + h : _ _
pl
(4.3.57)
_ _ The particular identication of = C " =0 and h = (dC=d)=C in Equation 4.3.56 above is assumed, so the material behavior is dened by the isothermal, uniaxial work hardening data, C () only.
Nonlinear isotropic/kinematic hardening model
This model is based on the work of Lemaitre and Chaboche (1990). The size of the yield surface, 0 (pl ; ; fi ), is dened as a function of equivalent plastic strain, "pl ; temperature, ; and eld " variables, fi . This dependency can be provided directly, can be coded in user subroutine UHARD, or can be modeled with a simple exponential law for materials that either cyclically harden or soften as
4.3.52
" 0 = j0 + Q1 (1 0 e0b );
pl
(4.3.58)
where j0(; fi) is the yield surface size at zero plastic strain, and Q1 (; fi ) and b(; fi ) are additional material parameters that must be calibrated from cyclic test data. The evolution of the kinematic component of the model is dened as
_ _ = C "pl
0
_ pl ( 0 ) 0 " +
_ C;
(4.3.59)
_ where C and are material parameters, with C representing the rate of change of C with respect to temperature and eld variables. The rate of change of with respect to temperature and eld variables is not accounted for in the model. This equation is the basic Ziegler law, generalized to account for _ pl temperature and eld variable dependency of C and to which a recall term, " , has been added. The recall term introduces the nonlinearity in the evolution law. The evolution of the backstress and the isotropic hardening are illustrated in Figure 4.3.51 for unidirectional loading and in Figure 4.3.52 for multiaxial loading.
max
0 0
0
Constitutive Theories
s + 0 0 s = C
pl
Figure 4.3.51 One-dimensional representation of the nonlinear model.
4.3.53
s3
2 max 3
limit surface
2 C 3 limiting location of
s
0
dev
2 0 3
s1
yield surface
s2
Figure 4.3.52 Three-dimensional representation of the nonlinear model. The center of the yield surface is contained within a cylinder of radius 2=3 C= , which follows directly from Equation 4.3.59. Therefore, the yield surface is contained within the limiting surface of radius 2=3 max , as shown in the gures. This model can be degenerated into the linear kinematic model described above by setting 0 = j0 and = 0. The physical behavior that can be captured by this model, as well as its limitations, is described in detail in the ABAQUS Analysis Users Manual.
Reference
Models for metals subjected to cyclic loading, Section 11.2.2 of the ABAQUS Analysis Users Manual
4.3.54
4.3.6
The porous metal plasticity model is intended for use with mildly voided metals. Even though the material that contains the voids (also known as the matrix material) is assumed to be plastically incompressible, the plastic behavior of the bulk material is pressure-dependent due to the presence of voids. The model is described in the following paragraphs, followed by a brief description of the material point calculations.
Yield condition
For a metal containing a dilute concentration of voids, based on a rigid-plastic upper bound solution for spherically symmetric deformations of a single spherical void, Gurson (1977) proposed a yield condition of the form
=
where
q 2
y
3 q2 p + 2q1 f cosh 0 2 y
0 (1 + q3 f 2 ) = 0;
S = pI +
q=
is the Mises stress;
3 S:S 2 1 3
p=0 :I
Constitutive Theories
is the hydrostatic pressure; f is the volume fraction of the voids in the material; and y (pl ) is the "m m yield stress of the fully dense matrix material as a function of "pl , the equivalent plastic strain in the 2 matrix. Tvergaard (1981) introduced the constants q1, q2 , and q3 = q1 (as coefcients of the void volume fraction and pressure terms) to make the predictions of the Gurson model agree with numerical studies of ordered voided materials in plane strain tensile elds; one can recover the original Gurson model by setting q1 = q2 = q3 = 1. It should be noted that f = 0 implies that the material is fully dense, and the Gurson yield condition reduces to that of von Mises; f = 1 implies that the material is fully voided and has no stress carrying capacity. This is illustrated in Figure 4.3.61, where the yield surfaces for different levels of porosity are shown in the pq plane. Figure 4.3.62 compares the behavior of a porous metal (which has an initial void volume fraction of f0) in tension and compression against the perfectly plastic matrix material; the initial yield stress of the porous metal is y0 . In compression the porous material hardens due to closing of the voids, and in tension it softens due to growth and nucleation of the voids.
4.3.61
q y f = 0 (Mises)
|p| y
f 0 = 0 (Mises)
tension (f 0 )
compression (f 0 )
4.3.62
Flow rule
The plastic strains are derived from the yield potential; the presence of the rst invariant of the stress tensor in the yield condition results in nondeviatoric plastic strains:
_ "pl = _
@ _ = @
0 1 @ I + 23q @ S ; 3 @p @q
The hardening of the (fully dense) matrix material is described through y = y (pl ). The evolution "m m of "pl is assumed to be governed by the equivalent plastic work expression; i.e.,
pl
Constitutive Theories
A=
sN
fN
1 exp 0 2 2
"
The normal distribution of the nucleation strain has a mean value "N , a standard deviation sN , and nucleates voids with volume fraction fN . The total rate of change of f is given as
f_ = f_gr + f_nucl :
Voids are nucleated only in tension; ABAQUS will not consider the nucleation term at a material point if the stress state is compressive. The nucleation function A=fN , which is assumed to have a normal distribution, is shown in Figure 4.3.63 for different values of the parameter sN . Figure 4.3.64 shows the extent of softening in a uniaxial tension test of a porous material for different values of fN .
4.3.63
A fN 1 sN2
1
s N< s N
1
1 sN2
2
Material 1 Material 2
N
Figure 4.3.63 Nucleation function
pl
A=fN .
fN
fN < f N
1 1 1
N = N fN
2 1
sN = sN
f 01 = f 0 2
Figure 4.3.64 Softening (in uniaxial tension) as a function of fN .
Integration of the elastoplastic equations
The integration of the elastoplastic equations for the porous plasticity model is carried out using the backward Euler scheme proposed by Aravas (1987). This method is briey discussed in the following paragraphs; the user can refer to the paper for further details.
4.3.64
During the constitutive calculations in an increment, the stress and state variables are known at time t (beginning of the increment). Given a total incremental strain " , the stress and state variables t (end of the increment) so that they satisfy the yield condition, ow rules, need to be updated at t and evolution equation of the state variables. To do this, consider the elasticity equations
+1
(4.3.61)
is the linear isotropic elasticity tensor with G and K being the shear and bulk modulus, respectively, and = and being the fourth- and second-order identity tensors, respectively. Also, in the above, the additive decomposition of strain is used to write the total incremental strain as the sum of the t unless indicated elastic and plastic parts. All of the stress and state variables are evaluated at t otherwise. The yield condition, the ow rule, and the evolution of the state variables are rewritten as
+1
( ; H ) = 0;
(4.3.62)
where
1"q = 1 @ : @q , = 1; 2 are the state variables "pl and f , respectively. In the above H m
is eliminated from the last two equations to give
(4.3.63)
Constitutive Theories
1
1"p @ + 1"q @ = 0: @q @p
Equation 4.3.62 is used in the elasticity Equation 4.3.61 to yield
= el 0 K 1"p I 0 2G1"q n:
4.3.65
(4.3.64)
Sel
Once n is known, it is easily seen that consistent determination of the scalars "p and "q completes the solution. Therefore, the problem of integrating the pressure-dependent elastoplastic constitutive equations is reduced to solving the following two nonlinear equations for the scalars "p and "q :
3 = 2qel Sel :
(4.3.65)
1 1
(p; q; H ) = 0;
(4.3.66) (4.3.67)
1"p @ + 1"q @ = 0: @q @p
p = pel + K 1"p ; q = qel 0 3G1"q ;
1H = h(1"p ; 1"q; p; q; H ); 1 1
where Equation 4.3.68 and Equation 4.3.69 are obtained by projecting Equation 4.3.64 onto I and n, respectively, and Equation 4.3.610 is an alternate form of Equation 4.3.63. Solving the above system of equations for the unknowns p; q; "p; "q ; and H completes the integration algorithm for the porous plasticity model. Equation 4.3.66 and Equation 4.3.67 are solved for "p and "q using Newtons method; p and q are updated using Equation 4.3.68 and Equation 4.3.69; the state variables are updated using Equation 4.3.610.
In the implicit nite element method of solving large-deformation problems, the discretized equilibrium equations result in a set of nonlinear equations for the nodal unknowns at the end of the increment. ABAQUS/Standard uses Newtons method to solve these equations, which requires the computation of linearization moduli To compute J (also known as the Jacobian), we start with the elasticity equations (Equation 4.3.64), which can be rewritten as
J = @ 1 = @ 1"
= " 0 (1=3)"kk I is the deviatoric part of ". From the above equation you nd that
4.3.66
(4.3.611)
Once again Equation 4.3.66 and Equation 4.3.67 are used in computing the variations @ "p and @ "q . After some lengthy algebraic calculations, a set of linear equations is obtained that can be solved for @ "p and @ "q . These derivatives are substituted into Equation 4.3.611 to obtain the linearization moduli. In general this linearization moduli is not symmetric. Further details of the derivation of the Jacobian can be found in Aravas (1987).
Reference
Porous metal plasticity, Section 11.2.9 of the ABAQUS Analysis Users Manual
Constitutive Theories
4.3.67
4.3.7
The cast iron plasticity constitutive model is intended for modeling the elastoplastic behavior of gray cast iron. In tension gray cast iron is more brittle than most metals. This brittleness is attributed to the microstructure of the material, which consists of a distribution of graphite akes in a steel matrix. In tension the graphite akes act as stress concentrators, leading to an overall decrease in mechanical properties (such as yield strength). In compression, on the other hand, the graphite akes serve to transmit stresses, and the overall response is governed by the response of the steel matrix alone. The above differences manifest themselves in the following macroscopic properties: (i) different yield strengths in tension and compression, with the yield stress in compression being a factor of three or more higher than the yield stress in tension; (ii) inelastic volume change in tension, but little or no inelastic volume change in compression; and (iii) different hardening behavior in tension and compression. It is commonly accepted (Hjelm, 1992, 1994) that a Mises-type yield condition along with an associated ow rule models the material response sufciently accurately under compressive loading conditions. This assumption is not true for tensile loading conditions: a pressure-dependent yield surface with nonassociated ow is required to model the brittle behavior in tension. The model is described in detail in the remainder of this section.
Strain rate decomposition
_ = "el + " pl ; _ _
_ _ _ where " is the total strain rate, "el is the elastic part of the strain rate, and " pl is the inelastic (plastic) part of the strain rate.
Constitutive Theories
Elastic behavior
In compression the elastic behavior of gray cast iron is similar to that of many steels. It shows a welldened elastic stiffness. In uniaxial tension the slope of the stress/strain curve decreases continuously, and it is difcult to estimate the elastic modulus from experimental results. The model in ABAQUS assumes that the elastic behavior of gray cast iron can be represented by linear isotropic elasticity, with the same stiffness in tension and compression.
Yield condition
The model makes use of a composite yield surface to describe the different behavior in tension and compression. In tension yielding is assumed to be governed by the maximum principal stress, while in compression yielding is assumed to be pressure-independent and governed by the deviatoric stresses alone. In principal stress space the composite yield surface consists of the Rankine cube in tension and the Mises cylinder in compression.
4.3.71
The material is assumed to be isotropic; hence, the yield surface can be expressed as a function of three invariant measures of the stress tensor: the equivalent pressure stress,
1 p = 0 : I;
the Mises equivalent stress,
q= r=(
where S is the stress deviator, dened as
S
r3
2 S : S;
9 1 3 2 S 1 S : S) ; = pI + ;
It is convenient to combine
is the Cauchy stress tensor, and I is the second-order identity tensor. the invariants q and r to dene a nondimensional quantity, 2, where cos(32) = ( r )3 : q
In principal stress space the variable 2 identies the meridional plane for a given stress state. On any given meridional plane the yield surface consists of two distinct line segments given by
Ft = RR (2)q 0 p 0 t = 0 Fc = q 0 c = 0 : In the expressions above RR (2) = (2=3) cos 2; t(pl ; ; f ) is the yield stress in uniaxial tension, "t t which may depend on the equivalent plastic strain in uniaxial tension "pl , temperature , and eld variables f ( = 1; 2; ::); and c(pl ; ; f ) is the yield stress in uniaxial compression, which may "c c depend on the equivalent plastic strain in uniaxial compression "pl , temperature , and eld variables f ( = 1; 2; ::). The composite yield surface is illustrated in the meridional plane in Figure 4.3.71,
in the deviatoric plane in Figure 4.3.72, and in the principal stress space in Figure 4.3.73.
Flow rule
and
For the purposes of discussing the ow and hardening behavior, it is useful to divide a meridional plane into two regions as shown in Figure 4.3.74. The region to the left of the uniaxial compression line (labeled UC) is referred to as the tensile region, and the region to the right of the uniaxial compression line is referred to as the compressive region.
4.3.72
= 60 o (compression)
c 3 Ft 3 t 1.5 t p
Figure 4.3.71 Schematic of the yield surface in the meridional plane.
= 0 S1
= 0 (tension)
o
Fc
1.5 Ft
= 60
Constitutive Theories
Rankine triangle
Rankine (tension)
Mises (compression)
Figure 4.3.73 Schematic of the yield surface in the principal stress space.
UC 3 -1 1 3
c t tensile region
Gc compressive region
Gt -t 3 c 3 p
4.3.74
_ where is the nonnegative plastic multiplier. The ow potential, G(p; q; ; f ), is assumed to be independent of the third stress invariant (independent of 2). It consists of the Mises cylinder in compression with an ellipsoidal cap in tension. The transition between the two surfaces is smooth. The projection of the ow potential onto the deviatoric plane is a circle. On the meridional plane the potential G can take one of two values, Gt and Gc , dened by the relations (p 0 Gt)2 + q2 = 9G 2 when t a2
q p< c
The plastic strains are dened to be normal to a family of self-similar ow potentials parametrized by the value of the potential G: _ _ @G ; " pl = @
= 3Gc when
p
c
3;
Gt is the value of the potential in the tensile region, and Gc is the value of the potential in the compressive region. The potential in the tensile region is of the form of an ellipse, where a is the ratio of the horizontal (p) to the vertical (q) axis of the ellipse. The shape of the ellipse is controlled by a, which is chosen such that it passes through the two points (0
t =3,
t ) and (c =3, c), where
is a material parameter that controls plastic dilatation. The above requirement determines a2 to be equal to (c +
t )=9(c 0
t ). The value of the potential, G, depends on the current stress point (p, q). In the compressive region the ow potential consists of the Mises straight line. A consequence of the above choice of the ow potential is that plastic ow results in inelastic volume change in the tensile region and no inelastic volume change in the compressive region. Figure 4.3.74 illustrates two potentials in the pq plane. It can be shown that for uniaxial tension,
= 2rct(1 + pl )=3, where rct = c =t and pl (referred to as the plastic Poissons ratio) is equal to the absolute value of the ratio of the transverse to the longitudinal plastic strain under uniaxial tension. The quantities c, t, and pl are material properties that must be provided by the user. The plastic Poissons ratio, pl , is expected to be less than 0:5 since experiments suggest that there is a permanent increase in the volume of gray cast iron when it is loaded in uniaxial tension beyond yield. In addition, pl must be greater than 01 for the potential to be well-dened. For the gray cast iron described in the work of Cofn (1950), the value of pl is approximately 0:04. This estimate is based on the reported value of the permanent volume change close to the ultimate tensile stress. The value of pl may change with plastic deformation. However, the model in ABAQUS assumes that it is constant with respect to plastic deformation. It can depend on temperature and eld variables. Since the ow potential is different from the yield surface (nonassociated ow), the material Jacobian matrix is unsymmetric.
Hardening
3:
Constitutive Theories
ABAQUS assumes that the graphite akes do not inuence the hardening behavior in the compressive region, where the matrix behavior (Mises plasticity) characterizes the overall response. In the tensile region the plastic strain consists of both a volumetric part (corresponding to opening up of the graphite
4.3.75
akes) and a deviatoric part (corresponding to inelastic shearing of the matrix). In the limiting case of pure hydrostatic tension it can be expected that the matrix material does not respond plastically; therefore, the entire plastic strain is volumetric, corresponding to the opening up of the graphite akes. Thus, in the tensile region, we expect that the deviatoric part of the plastic strain decreases as the loading path approaches hydrostatic tension, eventually becoming zero at hydrostatic tension, while the volumetric part of the plastic strain decreases as the loading path approaches uniaxial compression and is zero for uniaxial compression and conning pressures higher than c =3. "t The hardening of the model is controlled by the evolution of t = t (pl ; ; f ) and c = pl ; ; f ). To develop the hardening model, we need to nd expressions for the equivalent plastic c (c " t c strains, "pl and "pl , as functions of the plastic strain tensor " pl . To simplify the discussion, we assume that the temperature and the eld variables f ( = 1; 2; ::) are constants, so that the yield stresses are taken to be functions of the corresponding equivalent strains only. Furthermore, it is convenient to decompose the plastic strain rate into volumetric and deviatoric components:
_ _ c "pl = : "pl : c
Using the denition of the ow potential, it then follows (given that "pl = 0 and _vol compression) that _ _ "pl = epl : c
c
in
This expression is used to update c (pl ) for all loading conditions, including tensile loading "c t conditions. Next, we nd an expression relating "pl to " pl . Using the equivalent plastic work expression _ _ t"pl = : "pl t and the denition of the ow potential, it follows that
_ "pl = t
t
This expression is used to update the yield stress t(pl ) for all loading conditions, including "t compressive loading conditions. _ pl _ pl _vol _ _ Under predominantly compressive loading conditions, "pl = 0, "c = epl , and "t = (c =t)epl . pl , alone. This behavior Thus, both c and t are updated based on the deviatoric plastic strain, e is depicted in Figure 4.3.75. At the other extreme, under hydrostatic tension (see Figure 4.3.76) _ _ _vol c t epl = 0. Hence, "pl = 0, and "pl = "pl . Consequently, c does not evolve, but t evolves based _ _vol _ _vol on "pl . For all other loading paths in the tensile region, both "pl and epl are nonzero. In such
4.3.76
q stress evolution
Fc hardening Ft p
Figure 4.3.75 Hardening during compression loading.
hardening
Fc
Constitutive Theories
q = 0).
4.3.77
cases c is updated based on the deviatoric plastic strain, epl , alone, while t is updated based on both "pl and epl (see Figure 4.3.75). Given the micromechanics of deformation of gray cast iron, vol the behavior seems reasonable, although there appears to be no experimental evidence conrming the above hardening behavior. Since the matrix behavior of gray cast iron is similar to that of steel, it is reasonable to expect that under reversal of loading a combined kinematic and quasi-isotropic hardening scheme may be more appropriate to model, for example, the Bauschinger effect. Also, under compressive loading following tensile loading, we expect that some of the voids that had opened up at the graphite akes may close. Thus, a cap in the yield surface may be appropriate under high conning pressures. However, only limited information is available in the literature. The model in ABAQUS does not address these issues related to loading reversals and, therefore, should be used only for essentially monotonic loading conditions.
Reference
Cast iron plasticity, Section 11.2.10 of the ABAQUS Analysis Users Manual
4.3.78
4.3.8
This section describes the constitutive theory for Types 304 and 316 stainless steel in ABAQUS/Standard, as set forth in nuclear standard NE F95T(1981). The constitutive theory is uncoupled into a rate-independent plasticity response and a rate-dependent creep response, each of which is governed by a separate constitutive law. The plasticity theory uses a Mises yield surface that can expand isotropically and translate kinematically in stress space. Nuclear standard NE F95T provides for some coupling between the plasticity and creep responses by allowing prior creep strain to expand and translate the subsequent yield surface in stress space. For Types 304 and 316 stainless steel, however, prior plasticity does not change the subsequent creep response. A set of auxiliary creep and load reversal detection rules, using modied strain hardening creep theory, overcomes the inconsistencies usually encountered with standard strain hardening theories under a stress reversal. In particular, creep theories based on strain hardening assumptions predict creep rates that are too small under conditions of stress reversal, so that the amount of creep that occurs under cyclic loading conditions will generally be underestimated.
ORNL plasticity theory of stainless steel
The plasticity theory for stainless steels, as set forth in nuclear standard NE F95T, employs a von-Mises yield surface with kinematic hardening. Zieglers hardening rule, generalized to the nonisothermal case, is used in ABAQUS. Normally, when combined isotropic and kinematic hardening is considered, the center of the yield surface is assumed to translate linearly with plastic strain according to a Prager or Ziegler kinematic hardening rule. Incorporation of isotropic hardening into the constitutive formulation then changes the form of the stress-strain relation but leaves the Prager or Ziegler kinematic shift rule for determining the motion of the center of the yield surface intact. In the ORNL plasticity formulation the form of the stress-strain law is left intact (a bilinear representation in one dimension), and modications to the kinematic shift rule are made to accommodate isotropic hardening. The Mises yield surface is used:
Constitutive Theories
f (S 0 ) =
where
q3
2 (S
(4.3.81)
S=0
II :
and is the kinematic shift (the position of the center of the yield surface in stress space) that is taken to be purely deviatoric, so The quantity 0("pl ; "H ; ) is the equivalent uniaxial stress and denotes the distance in stress space from the center of the yield surface to any point on its surface. In the general case this yield surface
I : = 0:
4.3.81
radius is assumed to depend on the equivalent plastic strain, "pl ; the ORNL dened equivalent creep strain, "H ; and temperature, . The model uses associated plastic ow, which means that the plastic strain rate is dened by
where
d is a scalar that can be identied as the work equivalent plastic strain rate, d"pl , dened by " 0 d"pl = ( 0 ) : d"pl :
@f " d"pl = d ; @
(4.3.82)
This is so because
@f @ 0 d"pl = d
3 1 2
0
S 0 ) 0 d:
0
@f 0 ) : @
For continued satisfaction of the yield condition (Equation 4.3.81) during active plastic straining, the consistency condition is
0 0 @f @pl d"pl 0 @ d = 0: : (d 0 d) 0 @ @ @" For the evolution of we use the modied form of Zieglers hardening rule,
(4.3.83)
d = C
0
0 )d"pl 0
0
(4.3.84)
The second term in this equation accounts for isotropic hardening, so when this equation is combined with the consistency condition (Equation 4.3.83), the terms containing @0 =@"pl cancel. This produces a uniaxial stress-strain relation that is bilinear. The strain rate decomposition during active plastic loading is
" d"el
is the
= Del : " el ;
where
Del() is the elasticity matrix. This gives the rate form " d = Del : d"el + gel d;
4.3.82
where
gel def =
Combining this with the ow rule (Equation 4.3.82) allows the incremental stress-strain relation to be written in the form
@ Del el :" : @
@f " d = Del d" 0 d"pl + gel d: @ evolution equation for (Equation 4.3.84)
0 1 @f @f : d = C d"pl + C dC @ : d + @ d: @ d @ gives Projecting Equation 4.3.85 onto @f=@ @f @f " : d = @f : Del : d"pl 0 @f : Del : @ d"pl + @f : gel d: @ @ @ @ Combining these two denitions then provides d"pl from the total strain rate
and temperature
change rate as
where and
n def @ = @f
d def n : Del : n + C: =
Constitutive Theories
The incremental stress-strain relationship during active plastic deformation is now obtained directly from Equation 4.3.85 as
(4.3.86)
Reduction of Equation 4.3.83 and Equation 4.3.84 to one dimension, with @0 =@"pl = 0 at def , shows that C = d11=d"pl is the slope of the plastic portion of large strain values where "pl 11 the uniaxial stress versus plastic strain relation. The incremental stress-strain relation (Equation 4.3.86) has the same form as that obtained with pure kinematic hardening. As mentioned previously, this is due to the cancellation of the terms @0 =@"pl in the expressions for the evolution of and the consistency relation. The bilinear stressstrain law and the translation of the yield surface center are depicted in Figure 4.3.81. In pure kinematic hardening under isothermal conditions the yield center follows the path OCOGOC for two stress reversals under fully reversed strain controlled conditions, while the stress-strain curve follows the path OABCDEFGHAB. In the ORNL theory the translation of the center of the yield surface is governed by Equation 4.3.86. The term d0="pl attains its largest value when "pl = 0 and tends to . zero under continued cumulative straining as "pl
!1
!1
4.3.83
B A H
0(p = 0)
O G K E F M N J
0(p = 0)
0 p ( ) C I 0 p ( ) D L
Figure 4.3.81 Bilinear stress-strain behavior and movement of the yield surface center with the ORNL plasticity theory. The yield surface center, thus, follows the path OI under monotonic loading. If point I is pl 0 at point I, then as " continues sufciently distant from the origin, point O, so that @0 =@"pl to grow under continued stress reversals, the continued growth of is governed by pure kinematic hardening, with @0 =@"pl = 0. The yield surface center then translates along the path IJKJI. Unlike pure kinematic hardening the ORNL theory, which allows incremental changes in 0 due to cumulative plastic strain, produces a stress-strain curve that is asymmetric about the stress-strain origin. The expansion of the yield surface is approximated by a step change in the value of 0 . 0 Figure 4.3.82 shows the value of 0 appropriate to the virgin (initial) stress-strain curve, 0 (), 0 appropriate to the 10th cycle curve, 0 (). and the value of 10
4.3.84
Point 1 has coordinates (1,1) where 1 =E1. Point 2 has coordinates (2,2).
C=
2 - 1
(2 -2 /E)
Parallel slopes
2 1
100(T)
0(T) 0
Monotonic curve
Constitutive Theories
Figure 4.3.82 The virgin and 10th cycle ORNL stress-strain curves are assumed to have equal slopes in the plastic portion of the bilinear representation. The value of 0 is assumed to undergo a step change according to the relations
4.3.85
0 " 0 = 10( ) if : d"pl 0 or if "H > 0:002; 0 = 0 () if : d"pl > 0 and "H 0:002. " 0
(4.3.87)
These conditions state that a step change in the size of the yield surface occurs when the value of the ORNL equivalent creep strain, "H , reaches 0.2% or when yielding rst occurs after stress reversal. Nuclear standard NE F95T recommends that the yield surface center remain xed during the step change in 0 . In this case the term involving @0 =@"pl in Equation 4.3.86 remains identically equal to zero and the yield surface center translates according to Zieglers kinematic hardening rule. The nuclear standard recommends that the constant C in Equation 4.3.86 remain xed on changing the value of 0 from the virgin to the 10th cycle values. Since C is the slope of the stress versus plastic strain in a uniaxial tensile test, this implies that the plastic tangent modulus of the 10th cycle stress-strain curve is the same as the plastic tangent modulus of the virgin stress-strain curve. This requirement is also necessary so that the 10th cycle stress-strain curve under fully reversed strain controlled loading is symmetric about the stress-strain origin.
ORNL creep theory for stainless steel
The ow rule for ORNL creep theory can be written in the form
3S _ _ "cr = "; 2
(4.3.88)
where
_ "cr (; "cr ; ) is the uniaxial equivalent creep strain rate and is the Mises equivalent stress,
def =
cr
q3
2 S : S:
_ The functional dependence of " on stress, creep strain, and temperature can be dened either in a user subroutine or by data lines. If the data line option is chosen, ABAQUS assumes that the effective creep strain rate can be written as _ "cr = Antm :
"cr =
Elimination of
2
m+1
0
Antm+1 :
1m 31=m+1
t between these equations gives the strain hardening form of the creep law,
_ "cr = An (m + 1)"cr
which is now assumed to be valid even when the stress changes with time. The ORNL creep theory now replaces the total effective creep strain, "cr , in this equation by an effective creep strain, "H , determined according to the following algorithm. At any time during the creep response the equivalent total creep strain used to dene the equivalent creep strain rate is dened as the distance from an origin that is either "+ or "0 . L indicates at any
4.3.86
time which origin is active. If the origin is " + , we set L = 0, while if the origin is "0 , L = 1. The ~ quantity " denes the distance in total creep strain space between the current origin and the previous origin. The effective creep strain, "H , is then determined by the following steps: 0. 1. 2. 3.
~ Initially set " + = "0 = 0; set " = 0 and set L = 0. Set the current origin ag M = L. Set X = 1006 . Compute
S + = (" cr
0 " + ) : S;
S 0 = ("cr 02
0 " 0 ) : S;
G+ =
02
If the current origin is "+ , the effective creep strain, G+ , is decreasing if S + < 0. Similarly, if the current origin is "0 , the effective creep strain, G0 , is decreasing if S 0 < 0. The rules for choosing a new origin follow in Steps 4 through 14, based on the concept of a decreasing effective creep strain (a load reversal). If G+ < X and G0 < X , set N = 1 and go to Step 13. Otherwise, set N = 0 and continue to Step 5. If S + 0 and S 0 < 0, set N = 1 and go to Step 13. Otherwise, set N = 0 and continue to Step 6. If S + > 0 and S 0 0, set N = 1 and go to Step 13. Otherwise, set N = 0 and continue to Step 7. If S + < 0 and S 0 0 , set N = 2 and go to Step 13. Otherwise, set N = 0 and continue to Step 8. If S + 0 and S 0 0, set N = 2 and go to Step 13. Otherwise, set N = 0 and continue to Step 9. If S + > 0 and S 0 > 0 and G0 > (G+ + X ), set N = 2 and go to Step 13. Otherwise, set N = 0 and continue to Step 10. If S + > 0 and S 0 > 0 and G0 (G+ + X ), set N = 1 and go to Step 13. Otherwise set N = 0 and continue to Step 11. If S + < 0 and S 0 0 and G > (G+ + X ), set N = 1 and go to Step 13. Otherwise, set N = 0 and continue to Step 12. If S + < 0 and S 0 < 0 and G0 (G+ + X ); set N = 1 and go to Step 13. Otherwise, set N = 0 and continue to Step 13. If N = 0, go to Step 18. (In this case the load did not reverse during the current creep increment, and no updating of the origins is required.) If N = 1 or 2, continue to Step 14. If N = 1, set L = 0. (In this case the new origin is " + and the origin ag is set equal to zero.) If N = 2, set L = 1. (In this case the new origin is " 0 and the origin ag is set equal to one.) If L = M , go to Step 18. (In this case the new origin is the same as the current origin. No load reversal has, therefore, taken place; and updating of the origins is not required.)
3 ("
G0 =
3 ("
Constitutive Theories
4.3.87
~ ~ 16. If G0 > ", leave "+ at its current value, set "0 = " cr , set " = G+ , set L = 1, and go to Step 18. (New origin is now "0 .) ~ ~ If G+ ", leave "0 ; " + , and " at their current values; set L = 1; and go to Step 18. (New origin is now "+ .)
(In this case a load reversal has taken place.) 15. If M = 0, go to Step 16. (Current origin is "+ and load reversal has taken place.) If M = 1, go to Step 17. (Current origin is "0 and load reversal has taken place.)
If
~ ~ 17. If G0 > ", leave "0 at its current value, set " + = "cr , set " = G+ , set L = 0, and continue to Step 18. (New origin is now "+ .) ~ ~ If G0 ", leave " 0 ; "+ , and " at their current values; set L = 1; and continue to Step 18. New origin is now "+ .)
18. If 19.
In addition to the preceding algorithm for the determination of the effective creep strain, "H , in the strain hardening creep formulation, the ORNL procedures also allow for a translation of the center of the yield surface during creep. Nuclear standard NE F95T recommends that the center of the yield surface be shifted during creep according to the relation
_ _ = H "cr ;
2 31 L = 0, set "H = 2=3("cr 0 " +) : ("cr 0 "+ ) 2 : 2 31 If L = 1, set "H = 2=3("cr 0 "0 ) : ("cr 0 "0 ) 2 . Go to Step 1 to determine "H for the next creep increment.
where
H=C H=0
if if
3 : 0:3 0 2 3 : > 0 : 3 0; 2
the constant C being the slope of the uniaxial stress versus plastic strain curve and effective yield stress.
Reference
0
the current
ORNL Oak Ridge National Laboratory constitutive model, Section 11.2.12 of the ABAQUS Analysis Users Manual
4.3.88
DEFORMATION PLASTICITY
4.3.9
DEFORMATION PLASTICITY
Deformation plasticity theory is provided in ABAQUS/Standard to allow fully plastic analysis of ductile metals, usually under small-displacement conditions, for fracture mechanics applications. The model is based on the Ramberg-Osgood relationship. In this section the detailed constitutive model is dened. The procedure for obtaining fully plastic solutions generally consists of incremental loading until the response is fully plastic. The model is termed deformation plasticity because the stress is dened by the total mechanical strain with no history dependence. There is no unloading criterion (to allow recovery of the initial elastic stiffness immediately after a strain reversal), so that the model is only useful as a plasticity model in cases of continuous ow. It is, in fact, a nonlinear elastic model; but at a limit state when all of a specimen or structure is responding plastically, this model is a useful equivalent representation of the plastic response because it has such a simple form.
One-dimensional model
E" = +
is Youngs modulus (dened as the slope of where is the stress, " is the mechanical the stress-strain curve at zero stress), is the yield offset (in the sense that, when = 0 , " = (1 + )0 =E ), and n is the hardening exponent for the plastic (nonlinear) term: n > 1. The material behavior described by this model is nonlinear at all stress levels, but for commonly used values of the hardening exponent (n 5 or more) the nonlinearity becomes signicant only at stress magnitudes approaching or exceeding 0 .
Multiaxial generalization
0 strain, E
n01
j j
;
(4.3.91)
Constitutive Theories
A linear elastic relation is used to generalize the rst term of Equation 4.3.91; the nonlinear term is generalized to multiaxial stress states through the use of the Mises stress potential and associated ow law, giving the multiaxial model
E " = (1 + ) S 0 (1 0 2 )p I +
3 2
q n01
0
S;
(4.3.92)
where
" p = q1 : I 03 q = 3S : S 2
is the strain tensor, is the stress tensor, is the equivalent hydrostatic stress, is the Mises equivalent stress,
4.3.91
DEFORMATION PLASTICITY
S = + pI
The linear part of the behavior may be compressible or incompressible depending on the value of Poissons ratio, but the nonlinear part of the behavior is incompressible (because the ow is normal to the Mises stress potential). Since the model will generally be used for cases when the deformation is dominated by plastic ow, the use of selectively reduced integration elements or hybrid (mixed formulation) elements is recommended with this material model (except in plane stress).
Strain energy density
The model is often used to obtain fully plastic solutions for fracture mechanics when the is needed. For evaluation of the J -integral the strain energy density is required. This is
J -integral
W=
" : d":
W=
Stress solution
n+1 1 + q 2 + 3 1 0 2 p2 + n : n q 3E 2 E n + 1 E0 01
During an analysis at each integration point the latest estimate of the kinematic solution is provided to the constitutive routines, which must provide the corresponding stress tensor calculated for the material model being used. Since this material model is nonlinear, we solve for the stresses by using the methods described below.
The uniaxial case (the only nonzero stress component is one direct stress)
In this case
E" = +
q n0 1
0
;
(4.3.93)
where now q = jj. We solve Equation 4.3.93 for using Newtons method. Writing c as the correction to , the Newton equations for Equation 4.3.93 are
"
1 + n 0
q n01#
c = E " 0 0
q n01
0
;
= + c :
4.3.92
DEFORMATION PLASTICITY
As an initial guess we use = E j"j if E j"j 0 and as the same sign as ") if E j"j > 0. In this case the material stiffness matrix is
@ @"
E n01 : 1 + n (q=0)
When all strain components are dened kinematically (that is, all cases except uniaxial and plane stress), projecting Equation 4.3.92 onto I gives
p=0
3(1 0 2 )
I : ";
a linear relationship for the volumetric behavior. Dening the deviatoric strain as
e=
1 " 0 II : " 3
Ee =
3 1++ 2
n0 1 ! q
0
S:
(4.3.94)
e=
(4.3.95)
Constitutive Theories
Ee =
This equation is solved for
2 (1 + )q + 3
q n01
0
q:
(4.3.96)
"
2 (1 + ) + n 3
n01# q
0
cq = E e 0
2 (1 + )q 0 3
n01 q
0
q;
q = q + cq :
As for the uniaxial case we will use the starting guesses:
q=
3 E e 21+
if
3 E e 0 ; 21+
4.3.93
DEFORMATION PLASTICITY
and
q=
n01 E e 1=n
0
if
3 E e > 0 : 21+ 3 e E ; 2 q
1++
so that Equation 4.3.94 becomes
3 2
n0 1
q 0
Thus, once
S=
(4.3.97)
= S 0 p I:
The material stiffness is dened as follows. From Equation 4.3.97 we have
@S =
and Equation 4.3.96 gives
2q 3e
@e +
@q @ e 0 e q
e ;
(4.3.98)
@q =
and from Equation 4.3.95,
2 (1 + ) + n (q= )n01 0 3
@ e;
@e =
2q @S = 3e
Now
"
=0
2 3 e
1 1 0 e e + (n 0 1) (q=0)n01 (q=E )
) #
e e : @ e:
and
1 II 3
" : @" :
2q @ = 3e
"
=0
2 3 e
1 1 0 e e + (n 0 1) (q=0)n01 (q=E )
) !
ee
1 + 3
1 0 2
2q " I I : @" : 3e
#
4.3.94
DEFORMATION PLASTICITY
Plane stress
For this case we obtain a solution for assuming that the material is fully incompressible. We then use this solution as a starting guess for a Newton loop to nd . Finally, we compute the corresponding material stiffness matrix. Incompressible approximation: In this case = 1=2 and "33 = 0("11 + "22), so that e is known and Equation 4.3.96 can be solved as before for q. Equation 4.3.97 then denes S, and the plane stress constraint requires that 33 = 0, so 2q p = S33 = 0 ("11 + "22 ); 3e and, hence, we obtain the initial estimate of the solution
11 = 22 =
and
12 =
12 :
Newton solution for the actual stresses: Equation 4.3.92 denes the stresses, where, for this case,
p = 0 (11 + 22); S = + p I;
and
1 3
q=
Therefore, Equation 4.3.92 becomes
r3
2
Constitutive Theories
S : S:
E" =
3 1++ 2
q n01 !
0
1 +3 + 2
q n01 !
0
p I:
(4.3.99)
"
3 1++ 2
n01! q
0
= + 9 (n 0 1) 4
q 0
n01 q S S
0 q q
= E" 0
3 1+ + 2
n01!
1 03 + 2
n01!
q 0
1 + 2
n01! # q
0
II
: c
p I;
4.3.95
DEFORMATION PLASTICITY
= + c :
The material stiffness is directly available from Equation 4.3.99 as
"
1+
+
2
n01 ! q
0
= + 4 (n 0 1) 0
9
n01 q S S
q q
0 + 2 0
1
n01! #01 q
II
Reference
4.3.96
4.3.10
ABAQUS allows the introduction of the inelastic heat fraction, , which denes heat generation caused by mechanical dissipation associated with plastic straining. This term can be introduced as a source of coupling for thermal-mechanical analysis. Such coupling might be important in a simulation in which extensive inelastic deformation is occurring fairly rapidly in a material whose mechanical properties are temperature dependent. If the process is very slow, the heat generated by the plastic deformation has time to dissipate; and uncoupled, isothermal, analysis is sufcient to model the process. If the process is extremely rapid, the heat has no time to diffuse; and uncoupled, adiabatic, analysis (in which each integration point is treated as if it is thermally insulated from its neighbors) is sufcient. Fully coupled thermal-stress analysis is required for cases that lie far enough from both extremes. This section denes the heat generation term caused by inelastic straining and describes how this term contributes to the overall Newton solution scheme. The model assumes that plastic straining gives rise to a heat ux per unit volume of
rpl
_ = : "
pl
where rpl is the heat ux that is added into the thermal energy balance; is the inelastic heat fraction, which is assumed to be a constant; is the stress; and "pl is the rate of plastic straining. For all the plasticity models in ABAQUS, the plastic strain increment is written from the ow potential as
is a scalar measure of plastic straining that is used as a hardening parameter in the yield surface and ow potential denitions in some of the plasticity models. In this case we consider isotropic hardening theories only: ABAQUS provides only thermo-mechanical coupling for such models. ABAQUS generally uses a backward Euler scheme to integrate the plastic strain, so rpl at the end of the increment is approximated as
_ = "_ n; where n is the ow direction (we assume n = n( ; " ; ), where is the temperature) and "
"pl
pl pl
pl
Constitutive Theories
rpl
1 = 21t 1" n : ( + );
pl t
where all quantities are evaluated at the end of the increment (at time t t) except t. This notation is adopted throughout the remainder of this section. This term is used as the contribution to the thermal energy balance equation. When Newtons method is used to solve the nonlinear equations, the coupling term gives rise to three contributions to the Jacobian matrix for the Newton method:
+1
@rpl ; @
from the thermal energy balance equation, and
@ @
4.3.101
from the mechanical equilibrium equation. The general form for these terms is now derived. The mechanical constitutive model has the following general form. The elasticity denes the stresses by
=
where W = W " el ; is the strain energy density potential and "el is the mechanical elastic strain. We implicitly assume that the elasticity is not fully incompressible, although the derivation is not signicantly different if this is not the case, since the pressure stress will do no work in a fully incompressible material and so makes no contribution to the terms under discussion. We assume that there is an additive strain rate decomposition that can be integrated to give
"el
@W ; " @"el
(4.3.101)
(4.3.102)
where " is the total strain and "th is the strain caused by thermal expansion. In the constitutive models in ABAQUS "th "th only. This form of decomposition of the deformation depends on " being measured as the integrated rate of deformation and on the elastic and thermal strains being small: this is true for the standard plasticity models provided in the program. The plastic ow denition is integrated by the backward Euler method to give
= ()
1"pl = 1"pln:
(4.3.103)
Finally, assuming there is a single active yield surface or a single active ow surface, rate-independent models introduce a yield surface constraint, while rate-dependent models provide an integrated ow rate constraint, both of which are incorporated in the general form
f
where f is a scalar stress function (for example, the Mises or Hill stress function for simple metal plasticity models) and 0 "pl ; is the yield stress for a rate-independent model, while B0 for a rate-dependent model, where
()
( ) = ;
pl
(4.3.104)
= (
= B( 1"t ; ) 1
q ~
denes the rate effect from the average plastic strain rate over the increment. For example, by default, the rate-dependent plasticity model denes
"
_ =D
pl
0
01
n
where q is the Mises or Hill equivalent stress, and D and n are material parameters. Using the average plastic strain rate over the increment in this expression denes
B
()
()
=1+
1"pl 1=n
D t
4.3.102
Equation 4.3.101 to Equation 4.3.104 are a general denition of all of the standard isotropic hardening plasticity models integrated by the backward Euler method. We now take variations of these equations with respect to all quantities at the end of the increment:
@ @" " @ @ = @" elW @ + @" elW el : " @ " @" " @n
2 2
" @"
" 0 @" @
th
@
" 0 @" pl
"pl
= @"
pl
n + 1"
pl
@
: @ +
@n @"pl @"pl
@n @ @
;
and where
H = I + 1"pl Del : Z : @n ; @
@ g = @"elW 0 Del : " @
2
^ = m : Del : n + ^ Z = I 0 1 n m : Del;
d
@"pl @ ; @"pl
~ n = n + ( + t) : @n ; @
g
" @" th @
+ 1"pl @n @
;
Constitutive Theories
~ = m : g 0 @ ; g @ h = H01 : g 0 ~ Del : n ;
d
D = H01 : Del : Z:
@ " = D : @" + h @;
4.3.103
and
@r
pl
1 1"pl n : D + 1 ( + t) : n m : Del : I 0 1"pl @n : D : @" " ~ ^ = 21t d @ 1 ^ ~ + 21t 1 ( + t) : n g 0 1"pl m : Del : @n : h d @ + 1"pl( + t) : @n + 1"pl n + ( + t) : @n : h @ @
@:
References
Fully coupled thermal-stress analysis, Section 6.5.4 of the ABAQUS Analysis Users Manual Adiabatic analysis, Section 6.5.5 of the ABAQUS Analysis Users Manual
4.3.104
POROUS ELASTICITY
4.4.1
POROUS ELASTICITY
The porous elasticity model in ABAQUS/Standard is designed to be used in conjunction with plasticity models that allow plastic volume changes as described in Models for granular or polymer behavior, Section 4.4.2, to Models for crushable foams, Section 4.4.6. Use of the porous elasticity model without one of these plasticity options is not recommended. The model is based on the experimental observation that in porous materials during elastic (recoverable) straining, the change in the void ratioeand the change in the logarithm of the equivalent pressure stresspdened as
p=0
are linearly related, so that in rate form,
1 3
trace =
1 0 3 : I;
deel = 0 d(ln(p));
where is a material parameter. In this form the material has zero tensile strength. If the tensile strength pel is nonzero, the equivalent relation is t
deel = 0 d
el ln(p + pt )
(4.4.11)
which includes the special case of zero tensile strength (pel = 0). It can be shown that, if the compressibility t of the solid material is neglected, the volume change of a material sample is
1 + e0
1+e
(4.4.12)
where e0 is the initial void ratio. If we dene the elastic void ratio from the elastic volume change according to the relationship
Constitutive Theories
J el =
ln
1 + eel 1 + e0
(4.4.13)
and then integrate the linear relation, the volumetric elasticity relationship is
1 + e0
p + pel t p0 + pel t
exp
=1
0 J el ;
(4.4.14)
where p0 is the initial pressure stress, prescribed by initial conditions. Note that for a zero tensile strength material it is required that p0 > 0. This equation can be inverted to yield
p = 0pel + p0 + pel t t
as illustrated in Figure 4.4.11.
1 + e0
(1
(4.4.15)
4.4.11
POROUS ELASTICITY
p0 p0 elol v p el t -p el t
Figure 4.4.11 Porous elastic volumetric behavior. The deviatoric elastic behavior is dened either by choosing a constant shear modulus, G, so that the deviatoric elastic stiffness is independent of the equivalent pressure stress or choosing a constant Poissons ratio, , so that the deviatoric elastic stiffness increases as the equivalent pressure stress increases. If a constant shear modulus is given, the deviatoric relationship is
S = 2Geel ;
whereas when Poissons ratio is given, the relationship has the rate form
^ dS = 2G deel ;
where
^ G= 3(1
)
vol )
(4.4.16)
for a material with nonzero tensile strength. In these equations S is the deviatoric stress:
S=
+ p I;
and
eel
0 1 "el I; vol 3
Elastic behavior of porous materials, Section 10.3.1 of the ABAQUS Analysis Users Manual
4.4.12
4.4.2
The behavior of granular and polymeric materials is complex. However, under essentially monotonic loading conditions rather simple constitutive models provide useful design information. These constitutive models are essentially pressure-dependent plasticity models that have historically been popular in the geotechnical engineering eld. However, more recently they have also been found to be useful for the modeling of some polymeric and composite materials that exhibit signicantly different yield behavior in tension and compression. The models described here are extensions of the original Drucker-Prager model (Drucker and Prager, 1952). In the context of geotechnical materials the extensions of interest include the use of curved yield surfaces in the meridional plane, the use of noncircular yield surfaces in the deviatoric stress plane, and the use of nonassociated ow laws. In the context of polymeric and composite materials, the extensions of interest are mainly the use of nonassociated ow laws and the inclusion of rate-dependent effects. In both contexts the models have been extended to include creep.
Available yield criteria
Three yield criteria are provided in this set of models. They offer differently shaped yield surfaces in the meridional plane (pq plane): a linear form, a hyperbolic form, and a general exponent form (see Figure 4.4.21). The stress invariants used in the formulation are dened in Conventions, Section 1.2.2 of the ABAQUS Analysis Users Manual. The choice of model depends largely on the material, the experimental data available for calibration of the model parameters, and on the range of pressure stress values likely to be encountered. The linear model (available in ABAQUS/Standard and ABAQUS/Explicit) provides a noncircular section in the deviatoric (5) plane, associated inelastic ow in the deviatoric plane, and separate dilation and friction angles. The smoothed surface used in the deviatoric plane differs from a true Mohr-Coulomb surface that exhibits vertices. This has restrictive implications, especially with respect to ow localization studies for granular materials, but this may not be of major signicance in many routine design applications. Input data parameters dene the shape of the yield and ow surfaces in the deviatoric plane as well as the friction and dilation angles, so that a range of simple theories is provided; for example, the original Drucker-Prager model (Drucker and Prager, 1952) is available within this model. The hyperbolic and general exponent models (available in ABAQUS/Standard only) use a von Mises (circular) section in the deviatoric stress plane with associated plastic ow. A hyperbolic ow potential is used in the meridional plane, whichin generalmeans nonassociated ow.
Hardening, rate dependence, and creep
Constitutive Theories
Perfect plasticity as well as isotropic hardening are offered with these models. Isotropic hardening is generally considered to be a suitable model for problems in which the plastic straining goes well beyond the incipient yield state where the Bauschinger effect is noticeable (Rice, 1975). This hardening theory is, therefore, used for processes involving large plastic strain and in which the plastic strain
4.4.21
d p
a) Linear Drucker-Prager: F = t p tan d = 0
d
d /tan pt
2
pt c) Exponent form: F = aq p pt = 0
b
Figure 4.4.21 Yield criteria in the meridional plane. rate does not continuously reverse direction sharply; that is, the models are intended for problems involving essentially monotonic loading, as distinct from cyclic loading. The isotropic hardening models can be used for rate-dependent as well as rate-independent behavior. The rate-dependent version is intended for relatively high strain rate applications.
4.4.22
_ = c (pl ; "pl ; ; f ) if hardening is dened by the uniaxial compression yield stress ; c; " _ = t (pl ; " ; ; f ) if hardening is dened by the uniaxial tension yield stress ; t; " pl _ = d(pl ; " ; ; f ) if hardening is dened by the pure shear (cohesion) yield stress ; " pl _ where is the shear stress; K is a material parameter; "pl is the equivalent plastic strain; " is the equivalent plastic strain rate; is temperature; and f ; = 1; 2::: are other predened eld variables. _ pl The equivalent plastic strain rate, " , is dened for the linear Drucker-Prager model as _ _pl "pl =j11 j if hardening is dened in uniaxial compression; =pl if hardening is dened in uniaxial tension; _11
pl _ =p 3 if hardening is dened in pure shear,
where
pl is the engineering shear plastic strain rate and is dened for the hyperbolic and exponential _ Drucker-Prager models by the plastic work expression
pl
hardening as well as rate-dependent The functional dependence effects. If the shapes of the stress-strain curves are different at different strain rates, the test data are entered as tables of yield stress values versus equivalent plastic strain at different equivalent plastic strain rates: one table per strain rate. The yield stress at a given strain and strain rate is interpolated directly from these tables. Alternatively, when it can be assumed that the shapes of the hardening curves at different strain rates are similar, the hardening and rate dependence are specied separately. In this case we assume that the rate dependence can be written in a separable form:
(4.4.21)
Constitutive Theories
= R 0 ;
_ where 0 (pl ; ; f ) is the static yield stress for the Drucker-Prager hardening model and R(" ; ; f ) " _ pl = 0:0). The rate-dependent yield ratio R is scales this value at nonzero strain rate (R = 1:0 at " dened either in a tabular form or using the standard power law form _ "pl R= 1+ D
1 !n
pl
4.4.23
where D (; f ) and n(; f ) are material parameters. Creep models are most suitable for applications that exhibit time-dependent inelastic deformation at low deformation rates. Such inelastic deformation, which can coexist with rate-independent plastic deformation, is described later in this section. However, the existence of creep in an ABAQUS material denition precludes the use of rate dependence as described above.
Strain rate decomposition
(4.4.22)
" " " where d" is the total strain rate, d"el is the elastic strain rate, d"pl is the inelastic (plastic) strain rate, " " and d"cr is the inelastic creep strain rate. The term d"pl is omitted if the stress point is inside the " yield surface, and the term d"cr is omitted if creep has not been dened or is not active.
Elastic behavior
The elastic behavior can be modeled as linear or with the porous elasticity model including tensile strength described in Porous elasticity, Section 4.4.1. If creep has been dened, the elastic behavior must be modeled as linear.
Linear Drucker-Prager model
t=
q
2
"
1+
1
0K
1
3 #
r q
where K (; f ) is a material parameter. To ensure convexity of the yield surface, 0:778 K 1:0. This measure of deviatoric stress is used because it allows the matching of different stress values in tension and compression in the deviatoric plane, thereby providing exibility in tting experimental results when the material exhibits different yield values in triaxial tension and compression tests. This function is sketched in Figure 4.4.22. It only provides a coarse match to Mohr-Coulomb behavior (where the yield is indepe7ndent of the intermediate principal stress). Since (r=q)3 = 1 in uniaxial tension, t = q=K in this case; since (r=q)3 = 01 in uniaxial compression, t = q in that case. When K = 1, the dependence on the third deviatoric stress invariant is removed; the Mises circle is recovered in the deviatoric plane: t = q. With this expression for the deviatoric stress measure, the yield surface is dened as
t 0 p tan 0 d = 0;
(4.4.23)
4.4.24
S3
Curve a b S1 a S2 b
K 1.0 0.8
Figure 4.4.22 Typical yield surfaces for the linear model in the deviatoric plane. where
1 d = (1 0 tan )c if hardening is dened by the uniaxial compression yield stress; c; 3 1 1 = ( + tan )t if hardening is dened by the uniaxial tension yield stress; t; K 3 = d if hardening is dened by the shear (cohesion) yield stress; d;
and (; f ) is the friction angle of the material in the meridional stress plane. In the case of hardening dened in uniaxial compression, the linear yield criterion precludes friction angles > 71.5 (tan >3). This is not seen as a limitation since it is unlikely this will be the case for real materials. The hardening parameter d( ) measures the cohesion of the material and represents isotropic hardening, as illustrated in Figure 4.4.23. The formulation treats as constant with respect to stress, although it is straightforward to extend the theory to provide for the functional dependence of on quantities such as p. A method for converting Mohr-Coulomb data (, the angle of Coulomb friction, and c, the cohesion) to appropriate values of and d is described in the ABAQUS Analysis Users Manual.
Flow rule
4.4.25
) ) )
1 r _ _ t = 1 q 1+ 1 - 1- _ _ 2 K K q
Constitutive Theories
d t
pl
hardening d p
Figure 4.4.23 Schematic of hardening and ow for the linear model in the pt plane.
" d"pl
where
" = d c
pl @G
@
(4.4.24)
1 c =(1 0 tan ) if hardening is dened in uniaxial compression; 3 1 1 =( + tan ) if hardening is dened in uniaxial tension; K 3 1 1 = (1 + ) if hardening is dened in pure shear (cohesion); 2 K
and
dpl =jdpl j in the uniaxial compression case; " 11 =dpl in the uniaxial tension case; 11 d
pl in the pure shear case, where
pl is the engineering shear plastic strain: =p 3 G is the ow potential, chosen in this model as G = t 0 p tan ; (4.4.25) ) is the dilation angle in the pt plane. A geometrical interpretation of is shown in where (; f the tp diagram of Figure 4.4.23. In the case of hardening dened in uniaxial compression, this ow rule denition precludes dilation angles > 71.5 (tan >3). This is not seen as a limitation since
it is unlikely this will be the case for real materials. Comparison of Equation 4.4.23 and Equation 4.4.25 shows that the ow is associated in the deviatoric plane, because the yield surface and the ow potential both have the same functional dependence on t. However, the dilation angle, , and the material friction angle, , may be different, so the model may not be associated in the pt plane. For = 0 the material is nondilational; and if
4.4.26
= , the model is fully associatedthe model is then of the type rst introduced by Drucker and Prager (1952). For = and K = 1 the original Drucker-Prager model is recovered.
Hyperbolic and general exponent models
The hyperbolic and general exponent models, which are only available in ABAQUS/Standard, are written in terms of the rst two stress invariants only. The hyperbolic yield criterion is a continuous combination of the maximum tensile stress condition of Rankine (tensile cut-off) and the linear Drucker-Prager condition at high conning stress. It is written as
F=
pl 2 + q2 0 p tan 0 d = 0;
0
0
(4.4.26)
where l0 = (d0 j0 0 pt j0 tan ), pt j0 is the initial hydrostatic tension strength of the material, d0j0 is the initial value of d0, and (; f ) is the friction angle measured at high conning pressure, as shown in Figure 4.4.21(b). d0( ) is the hardening parameter, which is obtained from test data:
d0 =
if hardening is dened by the uniaxial compression yield stress; c; if hardening is dened by the uniaxial tension yield stress; t;
The isotropic hardening assumed in this model treats as constant with respect to stress and is depicted in Figure 4.4.24. Calibration of this model is described in the ABAQUS Analysis Users Manual.
hardening
Constitutive Theories
p
l0/tan l0/tan l0/tan
Figure 4.4.24 Schematic diagram of hardening for the hyperbolic model in the pq plane.
4.4.27
The general exponent form provides the most general yield criterion available in this class of models. The yield function is written as
F
= aqb 0 p 0 pt = 0;
(4.4.27)
where a(; f ) and b(; f ) are material parameters that are independent of plastic deformation and pt ( ) is the hardening parameter that represents the hydrostatic tension strength of the material, as shown in Figure 4.4.21(c). pt( ) is related to test data as
pt = ac b 0 c if hardening is dened by the uniaxial compression yield stress ; c; 3 b + t if hardening is dened by the uniaxial tension yield stress ; ; = at t 3 b if hardening is dened by the shear (cohesion) yield stress ; d: = ad
The isotropic hardening assumed in this model treats a and b as constant with respect to stress and is depicted in Figure 4.4.25.
hardening
(p ) a
t
1/b
pt
Figure 4.4.25 Schematic diagram of hardening for the general exponent model in the pq plane. The material parameters a, b, and pt can be given directly; or, if triaxial test data at different levels of conning pressure are available, ABAQUS will determine the material parameters from the triaxial test data. A least squares t, which minimizes the relative error in stress, is used to obtain the best t values for a, b, and pt .
Flow rule
d"pl @G ; f @
(4.4.28)
4.4.28
where f depends on how the hardening is dened (by uniaxial compression, uniaxial tension, or pure shear data) but can be written in general as
f
and
1 = : @G ; @
d"pl
= dpl in the uniaxial compression case; 11 pl in the uniaxial tension case; =d11 pl p = d
in the pure shear case, where
pl is the engineering shear plastic strain: 3
j j
where (; f ) is the dilation angle measured in the pq plane at high conning pressure; j0 = jpl =0;pl =0 is the initial equivalent yield stress; and is a parameter, referred to as the eccentricity, " " _ that denes the rate at which the function approaches the asymptote (the ow potential tends to a straight line as the eccentricity tends to zero). This ow potential, which is continuous and smooth, ensures that the ow direction is dened uniquely. The function asymptotically approaches the linear Drucker-Prager ow potential at high conning pressure stress and intersects the hydrostatic pressure axis at 90. A family of hyperbolic potentials in the meridional stress plane is shown in Figure 4.4.26. The ow potential is a von Mises circle in the deviatoric stress plane (the 5-plane).
d
pl
G=
p(j tan
0
)2 + q2 0 p tan
(4.4.29)
Constitutive Theories
|0
Figure 4.4.26 Family of hyperbolic ow potentials in the pq plane. In both models ow is associated in the deviatoric stress plane. In the general exponent model, ow is always nonassociated in the meridional pq plane. In the hyperbolic model comparison of
4.4.29
Equation 4.4.26 and Equation 4.4.29 shows that the ow is nonassociated in the pq plane if the dilation angle, , and the material friction angle, , are different. The hyperbolic model provides associated ow in the pq plane only when = and d0 j0 = tan 0 ptj0 = j0.
Creep models
Classical creep behavior of materials that exhibit plastic behavior according to the extended DruckerPrager models can be dened. The creep behavior in such materials is intimately tied to the plasticity behavior (through the denition of the creep ow potential and test data), so it is necessary to dene the Drucker-Prager plasticity and hardening behavior as well. The elastic part of the behavior must be linear. The rate-independent part of the plastic behavior is limited to the linear Drucker-Prager model with a von Mises (circular) section in the deviatoric stress plane (K =1). The plastic potential is the hyperbolic ow potential described in conjunction with the hyperbolic and general exponent models (Equation 4.4.29).
Creep behavior
We adopt the notion of creep isosurfaces (or equivalent creep surfaces) of stress points that share the same creep intensity, as measured by an equivalent creep stress. When the material plasties, the equivalent creep surface should coincide with the yield surface; therefore, we dene the equivalent creep surfaces by homogeneously scaling down the yield surface. In the pq plane that translates into parallels to the yield surface, as depicted in Figure 4.4.27.
cr no creep
4.4.210
ABAQUS requires that creep properties be dened through the same type of test data used to dene work hardening properties. The equivalent creep stress, cr , is determined as the intersection of the equivalent creep surface with the appropriate stress path. As a result,
cr =
(q 0 p tan ) if creep is dened in terms of the uniaxial compression stress; c; (1 0 1 tan ) 3 (q 0 p tan ) if creep is dened in terms of the uniaxial tension stress; t; = (1 + 1 tan ) 3 = (q 0 p tan ) if creep is dened in terms of the shear (cohesion) stress; d;
where (; f ) is the material angle of friction. Figure 4.4.27 shows how the equivalent creep stress is determined when the material properties are dened via a shear test: a parallel to the yield surface is drawn, such that it passes by the material point; the intersection of such a line with the test stress path (p = 0) produces cr . This approach has the consequence that the creep strain rate is a function of both q and p and allows realistic material properties to be determined in cases in which, due to high hydrostatic pressures, q is very high. If one looks at the yield strength of this material to be a composite of cohesion strength and friction strength, this model corresponds to cohesion-determined creep. Thus, there is a cone in pq space inside which there is no creep. The built-in ABAQUS creep laws or the uniaxial laws dened through user subroutine CREEP can be used. The integration of the creep strain rate is rst attempted explicitly, as described in Ratedependent metal plasticity (creep), Section 4.3.4. If the stability limit is exceeded, a geometrically nonlinear analysis is being performed, or plasticity becomes active, the integration is done by the backward Euler method, as described in Rate-dependent metal plasticity (creep), Section 4.3.4.
Creep ow rule
The creep ow rule is derived from a creep potential, Gcr , in such a way that
Constitutive Theories
" d"cr =
" where dcr is the equivalent creep strain rate, which must be work conjugate to the equivalent creep stress: dcr =jd"cr j in the uniaxial compression case; " 11 =d"cr in the uniaxial tension case; 11 d
cr in the pure shear case, where
cr is the engineering shear creep strain: =p 3 " Since d"cr is obviously work conjugate to , f cr is dened by 1 @Gcr : f cr = cr : @
4.4.211
(4.4.210)
The equivalent creep strain rate is then determined from the uniaxial creep law:
d"
The creep strain rate is assumed to follow from the same hyperbolic potential as the plastic strain rate
G
cr
where
jpl =0;pl =0 is the initial yield stress; and is a parameter, referred to as the eccentricity, that denes _ " " the rate at which the function approaches the asymptote (the creep potential tends to a straight line as the eccentricity tends to zero). This creep potential, which is continuous and smooth, ensures that the creep ow direction is always uniquely dened. The function approaches the linear Drucker-Prager creep potential asymptotically at high conning pressure stress and intersects the hydrostatic pressure axis at 90. A family of hyperbolic potentials in the meridional stress plane is shown in Figure 4.4.26. The creep potential is the von Mises circle in the deviatoric stress plane (the -plane). Equation 4.4.210 and Equation 4.4.211 produce the complete ow rule
(; f ) is the dilation angle measured in the pq plane at high conning pressure; j0 =
(4.4.211)
where
(4.4.212)
and
f
cr
0 1 tan if creep is dened via the uniaxial compression data; ( 0 tan )2 + (cr )2 3 cr + 1 tan if creep is dened via the uniaxial tension data; =q 2 ( 0 tan )2 + (cr ) 3 cr if creep is dened via the shear data: =q ( 0 tan )2 + (cr )2 =q
cr
j j j
The expressions for f cr indicate that when creep properties are dened in terms of uniaxial compression data, f cr will become negative if
Thus, below this stress level, which for typical materials will be very low, the stress vector and the normal to the creep potential are pointing in opposite directions:
: @G @
cr
< ;
4.4.212
which is equivalent to
q 0 p tan < q 0 p
(j0 tan )2 + q2
q2
Therefore, if = , there is a small zone just outside the no creep cone for which this is the case. Consequently, creep data obtained within this zone (such as data obtained in uniaxial compression) should show a creep strain rate in the opposite direction from the applied stress at very low stress levels, which will usually not be the case. To overcome this difculty, ABAQUS will modify the creep data entered such that f cr 0:1. Thus, one would not expect correspondence between calculated creep strains and measured creep properties in a region dened by
where
= 0:1 + tan
1 if the model is dened in terms of uniaxial compression data; 3 1 if the model is dened in terms of uniaxial tension data; = 0:1 0 tan 3 = 0:1 if the model is dened in terms of shear data:
The exact size of this region depends on the value of tan and the type of test data entered. This modication is usually not signicant since typical creep analyses have loads that are applied quickly, followed by long-term creep. Hence, the stress level for most of the analysis will usually be well beyond the modied zone. An example of slow loading in which the approximation is visible is included in Verication of creep integration, Section 3.2.6 of the ABAQUS Benchmarks Manual. As is clear in the example, the effect of the approximation is small in spite of the fact that the load is ramped up over the step. Although creep ow is associated in the deviatoric stress plane, the use of a creep potential different from the equivalent creep surface implies that creep ow is nonassociated.
Reference
Constitutive Theories
Extended Drucker-Prager models, Section 11.3.1 of the ABAQUS Analysis Users Manual
4.4.213
4.4.3
The inelastic constitutive theory provided in ABAQUS/Standard for modeling cohesionless materials is based on the critical state plasticity theory developed by Roscoe and his colleagues at Cambridge (Schoeld et al., 1968, and Parry, 1972). The specic model implemented is an extension of the modied Cam-clay theory. The discussion is entirely in terms of effective stress: the soil may be saturated with a permeating uid that carries a pressure stress and is assumed to ow according to Darcys law. The continuum theory of this two phase material is described in Continuity statement for the wetting liquid phase in a porous medium, Section 2.8.4. The modied Cam-clay theory is a classical plasticity model. It uses a strain rate decomposition in which the rate of mechanical deformation of the soil is decomposed into an elastic and a plastic part; an elasticity theory; a yield surface; a ow rule; and a hardening rule. These various parts of the theory are dened in this section. The model is implemented numerically using backward Euler integration of the ow rule and hardening rule: this approach is used throughout ABAQUS for plasticity models. The basic ideas of the Cam-clay model are shown geometrically in Figure 4.4.31 to Figure 4.4.37. The main features of the model are the use of an elastic model (either linear elasticity or the porous elasticity model, which exhibits an increasing bulk elastic stiffness as the material undergoes compression) and for the inelastic part of the deformation a particular form of yield surface with associated ow and a hardening rule that allows the yield surface to grow or shrink. A key feature of the model is the hardening/softening concept, which is developed around the introduction of a critical state surface: the locus of effective stress states where unrestricted, purely deviatoric, plastic ow of the soil skeleton occurs under constant effective stress. This critical state surface is assumed to be a cone in the space of principal effective stress (Figure 4.4.31), whose vertex is the origin (zero effective stress) and whose axis is the equivalent pressure stress, p. The section of the surface in the 5-plane (the plane in principal stress space orthogonal to the equivalent pressure stress axis) is circular in the original form of the critical state model: in ABAQUS this has been extended to the more general shape shown in Figure 4.4.35. In the section of effective stress space dened by the equivalent pressure stresspand a measure of equivalent deviatoric stresst (the denition of t is given later in this section)the critical state surface appears as a straight line, passing through the origin, with slope M (see Figure 4.4.32 and Figure 4.4.33). The modied Cam-clay yield surface has the same shape in the 5-plane as the critical state surface, but in the pt plane it is assumed to be made up of two elliptic arcs: one arc passes through the origin with its tangent at right angles to the pressure stress axis and intersects the critical state line where its tangent is parallel to the pressure stress axis, while the other arc is a smooth continuation of the rst arc through the critical state line and intersects the pressure stress axis at some nonzero value of pressure stress, again with its tangent at right angles to that axis (see Figure 4.4.34). Plastic ow is assumed to occur normal to this surface. The hardening/softening assumption controls the size of the yield surface in effective stress space. The hardening/softening is assumed to depend only on the volumetric plastic strain component and is such that, when the volumetric plastic strain is compressive (that is, when the soil skeleton is compacted), the yield surface grows in size, while inelastic increase in the volume of the soil skeleton causes the yield surface to shrink. The choice of elliptical arcs for the yield surface in the (p; t) plane, together with the
Constitutive Theories
4.4.31
_ -3
_ -2 Yield surface _ -1
Figure 4.4.31 Cam-clay yield and critical state surfaces in principal stress space. associated ow assumption, thus causes softening of the material for yielding states where t > Mp (to the left of the critical state line in Figure 4.4.32, the dry side of critical state) and hardening of the material for yielding states where t < Mp (to the right of the critical state line in Figure 4.4.33, the wet side of critical state). The resulting stress-strain behavior under states of constant effective pressure stress but increasing shear (deviatoric) strain is then as shown in Figure 4.4.32 and Figure 4.4.33: following initial yield (which is governed by the initially assumed yield surface size; that is, by the extent of initial overconsolidation) strain softening or strain hardening occurs until the stress state lies on the critical state surface when unrestricted deviatoric plastic ow (perfect plasticity) occurs. The terms wet and dry come from the idea of working a specimen of soil by hand. On the wet side of critical state the soil skeleton is too loosely compacted to support pressure stresssuch stress, if applied (such as by squeezing the soil by hand) passes immediately into the pore water and thus causes this water to bleed out of the specimen and wet the hands. The opposite effect occurs when the soil is on the dry side of critical state. The preceding discussion describes the concepts of the theory. These are now formalized, as they are implemented in ABAQUS/Standard.
The strain rate decomposition
4.4.32
d ~
pl
stress trajectory
t t
y
tl
Constitutive Theories
Figure 4.4.32 Shear test response on the dry side of critical state (t > M p).
J
= Jg
J el 1 J pl ; Jg
(4.4.31)
is the ratio of current to original volume where J is the ratio of current volume to original volume, of the soil grain particles, J el is the elastic (recoverable) part of the ratio of current to original volume of the soil volume, and J pl is the plastic (nonrecoverable) part of the ratio of current to original volume of the soil volume. Volumetric strains are dened as
"vol "el vol "vol
pl
= ln J; = ln J el ; = ln J pl :
4.4.33
ty stress trajectory p
tr(d ) ~
pl
t tl
ty
Figure 4.4.33
Shear test response on the wet side of critical state (t < M p).
These denitions and Equation 4.4.31 result in the usual additive strain rate decomposition for volumetric strain rates:
d"vol
(4.4.32)
The model also assumes the deviatoric strain rates decompose in an additive manner, so that the total strain rates decompose as
" d" " " = d"g I + d"el + d"pl ;
vol
The elastic behavior can be modeled as linear or by using the porous elasticity model, typically with a zero tensile strength, as described in Porous elasticity, Section 4.4.1.
4.4.34
M = 0.8 K = 0.8
Compression ( = 0.5)
Compression ( = 1.0)
The modied Cam-clay yield function is dened in terms of the equivalent effective pressure stress, p, and the Mises equivalent stress and third stress invariant, dened as
p=0 q=
9
r
2
1 3 3 2
trace
=0 :I
1 3
Constitutive Theories
S:S
r3 = S : S 1 S:
The surface is
f (p; q; r) =
In this equation = (; f ) is a user-specied constant that can be a function of temperature and other predened eld variables f ; = 1; 2:::. This constant is used to modify the shape of the yield surface on the wet side of critical state, so the elliptic arc on the wet side of critical state has a different curvature from the elliptic arc used on the dry side: = 1 on the dry side of critical state, while < 1 in most cases on the wet side, as shown in Figure 4.4.34. a(; f ) denes the hardening of the plasticity model, and is the point on the p-axis at which the elliptic arcs of the yield surface intersect the critical state line, as indicated in Figure 4.4.34. M (; f ) is the
2 a
p
01
2
+
t Ma
2
0 1 = 0:
(4.4.33)
4.4.35
slope of the critical state line in the pt plane (the ratio of t to p at critical state); and t = q=g, where g is used to shape the yield surface in the 5 plane, and is dened as
form of the Cam-clay model. The effect of different values of K on the shape of the yield surface in the 5-plane is shown in Figure 4.4.35. To ensure convexity of the yield surface, 0:778 K 1:0.
S3
2K ; 1 + K + (1 0 K ) (r=q)3 where K (; f ) is a user-dened constant. If K = 1, the yield surface does not depend on the third stress invariant, and the 5-plane section of the yield surface is a circle: this choice gives the original
g
Curve a b S1 a S2 b
K 1.0 0.8
Figure 4.4.35 Cam-clay surfaces in the deviatoric plane. Associated ow is used with the modied Cam-clay plasticity model. The size of the yield surface is dened by a: the evolution of this variable, therefore, characterizes the hardening or softening of the material. It is observed experimentally that, during plastic deformation,
de
= 0 d(ln p);
;
where is a constant. Integrating this equation, and using Equation 4.4.31, Equation 4.4.12, and Equation 4.4.14, we obtain
10J a = a0 exp (1 + e0 ) 0 J
4.4.36
pl pl
) ) )
1 1 1 r t = _ q 1+ _ - 1- _ _ 2 K K q
(4.4.34)
where a0 denes the position of a at the beginning of the analysisthe initial overconsolidation of the material. The value of a0 can be specied directly by the user or can be computed as
a0 =
1 2
exp
e1 0 e0 0 ln p0 ; 0
where p0 is the initial value of the equivalent pressure stress, and e1 is the intercept of the virgin consolidation line with the void ratio axis in a plot of void ratio versus equivalent pressure stress, shown in Figure 4.4.36.
e1 - locates initial consolidation state, by the intercept of the plastic line with In p = 0.
e, voids ratio
elastic slope
de = - d( In p)
plastic slope
de = - d( In p)
Constitutive Theories
Figure 4.4.36 Assumed soil response in pure compression (exponential hardening/softening case). The evolution of the yield surface can alternatively be dened as a piecewise linear function relating the yield stress in hydrostatic compression, pc , and the corresponding volumetric plastic strain "pl (Figure 4.4.37): vol
pc = pc ("pl ): vol pc
(1 +
a=
4.4.37
p C
p 0 C
-vol
pl
-(vol 0 + vol )
pl
pl
Figure 4.4.37 Piecewise linear hardening/softening curve. Note that the volumetric plastic strain axis has an arbitrary origin: "pl j0 is the position on this axis vol corresponding to the initial state of the material, thus dening the initial hydrostatic pressure, pc j0; and, hence, the initial yield surface size, a0 . ABAQUS checks that the initial effective stress state lies inside or on the initial yield surface. At any material point where the yield function is violated, a0 is adjusted so that Equation 4.4.33 is satised exactly (and, hence, the initial stress state lies on the yield surface).
Reference
Critical state (clay) plasticity model, Section 11.3.4 of the ABAQUS Analysis Users Manual
4.4.38
CAP MODEL
4.4.4
The modied Drucker-Prager/Cap plasticity model in ABAQUS is intended for geological materials that exhibit pressure-dependent yield. The yield surface includes two main segments: a shear failure surface, providing dominantly shearing ow, and a cap, which intersects the equivalent pressure stress axis (Figure 4.4.41).
Transition surface, Ft t Shear failure, FS (d+patan)
Cap, Fc
d+patan
d pa pb p
R(d+patan)
Constitutive Theories
There is a transition region between these segments, introduced to provide a smooth surface. The cap serves two main purposes: it bounds the yield surface in hydrostatic compression, thus providing an inelastic hardening mechanism to represent plastic compaction, and it helps to control volume dilatancy when the material yields in shear by providing softening as a function of the inelastic volume increase created as the material yields on the Drucker-Prager shear failure and transition yield surfaces. The model uses associated ow in the cap region and nonassociated ow in the shear failure and transition regions. The model has been extended to include creep, with certain limitations that are outlined in this section. The creep behavior is envisaged as arising out of two possible mechanisms: one dominated by shear behavior and the other dominated by hydrostatic compression.
Strain rate decomposition
4.4.41
CAP MODEL
" " " where d" is the total strain rate, d"el is the elastic strain rate, d"pl is the inelastic (plastic) time" independent strain rate, and d"cr is the inelastic (creep) time-dependent strain rate.
Elastic behavior
The elastic behavior can be modeled as linear elastic or by using the porous elasticity model including tensile strength, described in Porous elasticity, Section 4.4.1. If creep has been dened, the elastic behavior must be modeled as linear.
Plastic behavior
The yield/failure surfaces used with this model are written in terms of the three stress invariants: the equivalent pressure stress,
p=0
the Mises equivalent stress,
1 3
trace( );
q=
and the third invariant of deviatoric stress,
r
9
3 2
S : S);
r = ( S 1 S : S) 3 ;
1
where
t=
q
2
"
1+
1
0K
1
3 #
r q
where K (; f ) is a material parameter that may depend on temperature, , and other predened elds f ; = 1; 2; . . .. This measure of deviatoric stress is used because it allows matching of different stress values in tension and compression in the deviatoric plane, thereby providing exibility in tting experimental results and a smooth approximation to the Mohr-Coulomb surface. Since r=q = 1 in uniaxial tension, t = q=K in this case; since r=q = 01 in uniaxial compression, t = q in that case. When K = 1, the dependence on the third deviatoric stress invariant is removed; and the Mises circle is recovered in the deviatoric plane: t = q. Figure 4.4.42 shows the dependence of t on K . To ensure convexity of the yield surface, 0:778 K 1:0.
4.4.42
CAP MODEL
S3
Curve a b S1 a S2 b
K 1.0 0.8
Figure 4.4.42
With this expression for the deviatoric stress measure, the Drucker-Prager failure surface is written as
Fs = t 0 p tan 0 d = 0;
Constitutive Theories
where (; f ) is the materials angle of friction and d(; f ) is its cohesion (see Figure 4.4.41). The cap yield surface has an elliptical shape with constant eccentricity in the meridional (pt) plane (Figure 4.4.41) and also includes dependence on the third stress invariant in the deviatoric plane (Figure 4.4.42). The cap surface hardens or softens as a function of the volumetric plastic strain: volumetric plastic compaction (when yielding on the cap) causes hardening, while volumetric plastic dilation (when yielding on the shear failure surface) causes softening. The cap yield surface is written as
Fc =
p0
p a )2 +
Rt (1 + 0 =cos )
2
0 R(d + pa tan ) = 0;
where R(; f ) is a material parameter that controls the shape of the cap, (; f ) is a small number that is dened below, and pa is an evolution parameter that represents the volumetric plastic strain driven hardening/softening. The hardening/softening law is a user-dened piecewise linear function relating the hydrostatic compression yield stress, pb , and the corresponding volumetric inelastic (plastic and/or creep) strain, pb = pb("in ; ; f ) (Figure 4.4.43), where "in = "pl + "cr . vol vol vol vol
4.4.43
) ) )
1 r _ _ t = 1 q 1+ 1 - 1- _ _ 2 K K q
CAP MODEL
pb
in
pl
cr
pa =
(1 +
pb 0 Rd : Rtan )
The parameter is a small number (typically 0.01 to 0.05) used to dene a smooth transition surface between the shear failure surface and the cap:
Ft =
p 0 pa
)2 +
t 0 (1 0
cos
)( + a tan )
d p
2
0 (d + pa tan ) = 0:
Flow rule
Plastic ow is dened by a ow potential that is associated on the cap and nonassociated on the failure yield surface and transition yield surfaces. The nonassociated nature of these surfaces stems from the shape of the ow potential in the meridional plane. The ow potential surface in the meridional plane is shown in Figure 4.4.44.
4.4.44
CAP MODEL
Gc (cap) d+patan
Figure 4.4.44 Modied Drucker-Prager/Cap model: ow potential in the pt plane. It is made up of an elliptical portion in the cap region that is identical to the cap yield surface:
Gc =
p0
p a )2 +
Rt 1 + 0 =cos
2
and another elliptical portion in the failure and transition regions that provides the nonassociated ow component in the model:
Constitutive Theories
Gs =
[( a
p 0 p)tan ]
+ 1+
0 =cos
2
The two elliptical portions, Gc and Gs, form a continuous and smooth potential surface. Nonassociated ow implies that the material stiffness matrix is not symmetric, so the unsymmetric solver should be invoked by the user. However, if the region of the model in which nonassociated inelastic deformation is occurring is conned, it is possible that a symmetric approximation to the material stiffness matrix will give an acceptable convergence rate: in such cases the unsymmetric solver may not be needed.
Creep model
Classical creep behavior of materials that also exhibit plastic behavior according to the modied Drucker-Prager/Cap model can be dened.
4.4.45
CAP MODEL
The creep behavior in such materials is intimately tied to the plasticity behavior (through the denition of creep ow potentials and test data), so it is necessary to dene the modied DruckerPrager/Cap plasticity and hardening behavior as well. The elastic part of the behavior must be linear. The rate-independent part of the plastic behavior is limited by the following restrictions:
The built-in ABAQUS creep laws or uniaxial laws dened through user subroutine CREEP can be used. The integration of the creep strain rate is rst attempted explicitly, as described in Rate-dependent metal plasticity (creep), Section 4.3.4. The integration is done by the backward Euler method (as described in Rate-dependent metal plasticity (creep), Section 4.3.4) if the stability limit is exceeded, a geometrically nonlinear analysis is being performed, or plasticity becomes active. In this model we assume the existence of two separate and independent creep mechanisms. One is a cohesion mechanism, which operates similarly to the Drucker-Prager creep model described in Models for granular or polymer behavior, Section 4.4.2. The other is a consolidation mechanism, which operates similarly to the cap zone plasticity. We then have
" "s "c d" cr = d"cr + d"cr ; "s "c where d"cr is the creep strain rate due to the cohesion mechanism and d"cr is the creep strain rate due to the consolidation mechanism. As described above, the cap surface hardens or softens as a function of the volumetric plastic strain and volumetric creep strain: volumetric inelastic compaction (when yielding on the cap or creeping through the consolidation mechanism) causes hardening, while volumetric plastic dilation (when yielding on the shear failure surface or creeping through the cohesion mechanism) causes softening. The separation between the two yield surfaces and the dominant regions for the two creep mechanisms are dened by the evolution parameter, pa , which relates to the user-dened hydrostatic compression yield stress, pb = pb("pl + "cr ; ; f ) (Figure 4.4.43). vol vol The cohesion mechanism is active for all stress states that have a positive equivalent creep stress as explained below. The consolidation mechanism is active for all stress states in which the pressure is larger than pa . Figure 4.4.45 illustrates the active regions in this formulation. We adopt the notion of the existence of creep isosurfaces (or equivalent creep surfaces) of stress points that share the same creep intensity, as measured by an equivalent creep stress. Consider the cohesion creep mechanism rst. When the material plasties, the equivalent creep surface should coincide with the yield surface; therefore, we dene the equivalent creep surfaces by homogeneously scaling down the yield surface. In the pq plane that translates into parallels to the yield surface, as depicted in Figure 4.4.46. ABAQUS requires that cohesion creep properties be measured in a uniaxial compression test.
4.4.46
CAP MODEL
p ree nc sio he co
no creep consolidation creep pa R(d+patan)
(d+patan)
Constitutive Theories
cr no creep
The equivalent creep stress, cr , is determined as the intersection of the equivalent creep surface with the uniaxial compression curve. As a result,
4.4.47
CAP MODEL
cr =
where (; f ) is the material angle of friction. Figure 4.4.46 shows how the equivalent creep stress was determined. In uniaxial compression p = 1 cr ; therefore, the uniaxial compression test line has 3 a slope of 1/3. This approach has several consequences. One is that the cohesion creep strain rate is a function of both q and p. This allows the determination of realistic material properties in cases in which, due to high hydrostatic pressures, q is very high. If one looks at the yield strength of the material in this region to be a composite of cohesion strength and friction strength, this model corresponds to cohesion-determined creep. As a result, there is a cone in pq space inside which there is no cohesion creep. Next consider the consolidation creep mechanism. In this case we wish to make creep dependent on the hydrostatic pressure above a threshold value of pa , with a smooth transition to the areas in which the mechanism is not active (p < pa ). Therefore, we dene equivalent creep surfaces as constant pressure surfaces. In the pq plane that translates into vertical lines. ABAQUS requires that consolidation creep properties be measured in a hydrostatic compression test. The effective creep pressure, pcr , is then the point on the p-axis with a relative pressure
(q 0 p tan ) ; (1 0 1 tan ) 3
(4.4.41)
pcr = p 0 pa :
(4.4.42)
This value is used in the uniaxial creep law. The equivalent volumetric creep strain rate produced by this type of law is dened as positive for a positive equivalent pressure. The internal tensor calculations in ABAQUS will account for the fact that a positive pressure will produce negative (that is, compressive) volumetric creep components.
Creep ow rule
The creep ow rules are derived from creep potentials, Gcr , in such a way that
" d"cr =
where d"cr is the equivalent creep strain rate, which must be work conjugate to the equivalent creep stress: d"cr =jd"cr j in the uniaxial compression case; 11 cr =jd"vol j in the volumetric compression case:
d"cr @Gcr ; f cr @
(4.4.43)
(4.4.44)
with
cr =j11j in the uniaxial compression case; = p in the volumetric compression case:
4.4.48
CAP MODEL
Cohesion creep
For the cohesion mechanism the creep potential is assumed to follow the same potential as the creep strain rate in the Drucker-Prager creep model (Models for granular or polymer behavior, Section 4.4.2); that is, a hyperbolic function. This creep ow potential, which is continuous and smooth, ensures that the ow direction is always uniquely dened. The function approaches a parallel to the shear-failure yield surface asymptotically at high conning pressure stress and intersects the hydrostatic pressure axis at a right angle. A family of hyperbolic potentials in the meridional stress plane is shown in Figure 4.4.47:
Gcr s
(4.4.45)
material point
cr
similar hyperboles
Constitutive Theories
pa
Figure 4.4.47 Creep potentials: cohesion mechanism. The equivalent cohesion creep strain rate is then determined from the uniaxial law:
(0:1
(1
0 1 tan ) tan ) 3
d
4.4.49
+q
n+
1 tan I); 3
CAP MODEL
where
n=
f cr
@q @
= 3S; 2q
and
The proportionality factor, f cr , is not a constant in this model. Its expression indicates that it will become negative if
(1 0
1 3
q tan )
(tan )2 : 1 0 ( 1 tan )2 3
1 3
It turns out that below this stress level, which for typical materials will be very low, the stress vector and the normal to the creep potential are pointing in opposite directions:
:
which is equivalent to
@Gcr s < 0; @
q 0 p tan < q 0
q2
Thus, there is a small zone just outside the no creep cone for which this is the case. Consequently, creep data obtained within this zone should show a creep strain rate in the opposite direction from the applied stress at very low stress levels, which will usually not be the case. To overcome this difculty, ABAQUS will modify the creep data entered such that f cr 0:1. Therefore, you would not expect correspondence between calculated creep strains and measured creep properties in a region dened by
(1 0
1 3
(0 q :1 + tan )
This modication is usually not signicant, since typical creep analyses have loads that are applied quickly, followed by long-term creep. Hence, the stress level for most of the analysis will usually be well beyond the modied zone. An example of slow loading in which the approximation is visible is included in Verication of creep integration, Section 3.2.6 of the ABAQUS Benchmarks Manual. As is clear in the example, the effect of the approximation is small in spite of the fact that the load is ramped up over the step. The equivalent cohesion creep strain rate is a function of both q and p through cr . The creep potential is the von Mises circle in the deviatoric stress plane (the 5-plane). Although creep ow is associated in the deviatoric stress plane, the use of a creep potential different from the equivalent creep surface implies that creep ow is nonassociated.
4.4.410
CAP MODEL
Consolidation creep
For the consolidation mechanism the creep potential is derived from the plastic potential of the cap zone (Figure 4.4.48):
(4.4.46)
q material point
cr
similar ellipses pa
cr
Constitutive Theories
The equivalent consolidation creep strain rate is then determined from the uniaxial law
= 1
Note that there is an equivalent pressure stress, p, work conjugate of the equivalent consolidation creep strain, which is different from the effective creep pressure, pcr . Such equivalent pressure stress is given by
= R
2 q2
+ p (p 0 p a )
Gcr c
4.4.411
CAP MODEL
and has the characteristic that it reduces to the pressure in a hydrostatic compression test. The creep potential is the von Mises circle in the deviatoric stress plane (the 5-plane). Creep ow is nonassociated in this mechanism. This formulation is quite simplistic and ignores the effects of q on the creep function, hc . The two creep mechanisms operate independently from each other. This implies that hs does not depend c s on "cr and that hc does not depend on "cr . The only cross effects between both mechanisms are obtained through the dependency of pa on the volumetric creep from any of them.
Reference
Modied Drucker-Prager/Cap model, Section 11.3.2 of the ABAQUS Analysis Users Manual
4.4.412
MOHR-COULOMB MODEL
4.4.5
MOHR-COULOMB MODEL
The Mohr-Coulomb failure or strength criterion has been widely used for geotechnical applications. Indeed, a large number of the routine design calculations in the geotechnical area are still performed using the Mohr-Coulomb criterion. The Mohr-Coulomb criterion assumes that failure is controlled by the maximum shear stress and that this failure shear stress depends on the normal stress. This can be represented by plotting Mohrs circle for states of stress at failure in terms of the maximum and minimum principal stresses. The Mohr-Coulomb failure line is the best straight line that touch es these Mohrs circles (Figure 4.4.51). Thus, the MohrCoulomb criterion can be written as
= c 0 tan ;
where is the shear stress, is the normal stress (negative in compression), c is the cohesion of the material, and is the material angle of friction.
(,) 1 - 3 2 1 1+3 2 3 3
s= c 1 m=
Constitutive Theories
= s cos ; = m + s sin :
s + m sin 0 c cos = 0;
4.4.51
MOHR-COULOMB MODEL
where
s
= 1 ( 1 0 3 ) 2
1 2 ( 1 + 3 )
is half of the difference between the maximum and minimum principal stresses (and is, therefore, the maximum shear stress) and
m =
is the average of the maximum and minimum principal stresses (the normal stress). Thus, unlike the Drucker-Prager criterion, the Mohr-Coulomb criterion assumes that failure is independent of the value of the intermediate principal stress. The failure of typical geotechnical materials generally includes some small dependence on the intermediate principal stress, but the Mohr-Coulomb model is generally considered to be sufciently accurate for most applications. This failure model has vertices in the deviatoric stress plane (see Figure 4.4.52).
S3 Mohr-Coulomb
S1
S2 Drucker-Prager (Mises)
Figure 4.4.52 Mohr-Coulomb model in the deviatoric plane. The constitutive model described here is an extension of the classical Mohr-Coulomb failure criterion. It is an elastoplastic model that uses a yield function of the Mohr-Coulomb form; this yield function includes isotropic cohesion hardening/softening. However, the model uses a ow potential that has a hyperbolic shape in the meridional stress plane and has no corners in the deviatoric stress space. This ow potential is then completely smooth and, therefore, provides a unique denition of the direction of plastic ow.
Strain rate decomposition
4.4.52
MOHR-COULOMB MODEL
(4.4.51)
" d"pl
Elastic behavior
The Mohr-Coulomb criterion written above in terms of the maximum and minimum principal stresses can be written for general states of stress in terms of three stress invariants. These invariants are the equivalent pressure stress,
p = 0 trace ( ); q=
r3
1 3
where
2 (S : S );
S = + pI;
and the third invariant of deviatoric stress,
r=(
9S 2
S S 3:
: )1
Constitutive Theories
F = Rmc q 0 p tan 0 c = 0;
(4.4.52)
where (; f ) is the friction angle of the material in the meridional stress plane, where is the " temperature and f ; = 1; 2::: are other predened eld variables; c(pl ; ; f ) represents the evolution of the cohesion of the material in the form of isotropic hardening (or softening); "pl is the equivalent plastic strain, its rate dened by the plastic work expression
_ _ c "pl = : "pl ;
and
1 Rmc (2; ) = p
MOHR-COULOMB MODEL
where
2 is the deviatoric polar angle (Chen and Han, 1988) dened as r 3 cos (32) = q :
The friction angle of the material, , also controls the shape of the yield surface in the deviatoric plane as shown in Figure 4.4.53. The range of values the friction angle can have is 0 < 90. In the case of = 0 the Mohr-Coulomb model reduces to the pressure-independent Tresca model with a perfectly hexagonal deviatoric section. In the case of = 90 the Mohr-Coulomb model would reduce to the tension cut-off Rankine model with a triangular deviatoric section and Rmc = 1 (this limiting case is not permitted within the Mohr-Coulomb model described here).
=0
Mohr-Coulomb ( = 20) Rmcq = /3 Tresca ( = 0) c p Rankine ( = 90) = 2/3 = 4/3 Drucker-Prager (Mises)
Potential ow is assumed, so
" d"pl
where g can be written as
= d" g
c
pl
@G ; @ @
(4.4.53)
g=
1 : @G
4.4.54
MOHR-COULOMB MODEL
and G is the ow potential, chosen as a hyperbolic function in the meridional stress plane and a smooth elliptic function in the deviatoric stress plane:
G
where
cj0 = cjpl =0 is the initial cohesion yield stress; and is a parameter, referred to as the eccentricity, that " denes the rate at which the function approaches the asymptote (the ow potential tends to a straight line as the eccentricity tends to zero). This ow potential, which is continuous and smooth in the meridional stress plane, ensures that the ow direction is dened uniquely in this plane. The function asymptotically approaches a linear ow potential at high conning pressure stress and intersects the hydrostatic pressure axis at 90. A family of hyperbolic potentials in the meridional stress plane is shown in Figure 4.4.54.
pl
(; f ) is the dilation angle measured in the pRmw q plane at high conning pressure;
(4.4.54)
d Rmwq
c |0
p
Constitutive Theories
Figure 4.4.54 Family of hyperbolic ow potentials in the meridional plane. The ow potential is also continuous and smooth in the deviatoric stress plane (the 5-plane); we adopt the deviatoric elliptic function used by Mentrey and Willam (1995):
Rmw
(2; e) =
where 2 is the and e is a parameter that describes the out-of-roundedness of the deviatoric section in terms of the ratio between the shear stress along the extension meridian (2 = 0) and the shear stress along the compression meridian (2 = ). The elliptic function has the value Rmw (2 = 0; e) = Rmc ( ; )=e 3 3 along the extension meridian and has the value Rmw (2 = ; e) = Rmc ( ; ) along the compression 3 3 meridian; this ensures that the ow potential matches the yield surface at the triaxial compression and
2 4(1 0 e2 ) cosp2 + (2e 0 1)2 Rmc ( ; ); 2 ) cos 2 + (2e 0 1) 4(1 0 e2 ) cos2 2 + 5e2 0 4e 3 2(1 0 e deviatoric polar angle dened previously, Rmc ( ; ) = (3 0 sin )=6 cos , 3
4.4.55
MOHR-COULOMB MODEL
extension in the deviatoric plane provided that e is dened appropriately (see further discussion later). Although the elliptic function is dened only in the sector 0 2 =3, the polar radius Rmw (2; e) extends to all polar directions 0 2 2 using the three-fold symmetry shown in Figure 4.4.55.
=0
Rankine (e = 1/2) = /3
= 2/3
= 4/3 Mises (e = 1)
Figure 4.4.55 Mentrey-Willam ow potential in the deviatoric plane. By default, the out-of-roundedness parameter, e, is dependent on the friction angle ; it is calculated by matching the ow potential to the yield surface in both triaxial tension and compression in the deviatoric plane:
e
Alternatively, e can also be considered to be an independent material parameter; in this case the user can provide its value directly. Convexity and smoothness of the elliptic function requires that 1=2 < e 1. The upper limit, e = 1 (or = 0), leads to Rmw (2; e = 1) = Rmc ( ), which 3; describes the Mises circle in the deviatoric plane. The lower limit, e = 1=2 (or = 90 ), leads to Rmw (2; e = 1=2) = 2Rmc( ; )cos 2 and would describe the Rankine triangle in the deviatoric 3 plane (this limiting case is not permitted within the Mohr-Coulomb model described here). Flow in the meridional stress plane can be close to associated when the angle of friction, , and the angle of dilation, , are equal and the eccentricity parameter, , is very small; however, ow in this plane is, in general, nonassociated. Flow in the deviatoric stress plane is always nonassociated. Therefore, the use of this Mohr-Coulomb model generally requires the solution of nonsymmetric equations.
Reference
sin = 3 0 sin : 3+
4.4.56
4.4.6
The constitutive models described here are available in ABAQUS for the analysis of crushable foams typically used in energy absorption structures. Two phenomenological constitutive models are presented: the volumetric hardening model and the isotropic hardening model. Both models use a yield surface with an elliptical dependence of deviatoric stress on pressure stress in the meridional plane. The crushable foam model with volumetric hardening is available in ABAQUS/Standard as well as in ABAQUS/Explicit. The crushable foam model with isotropic hardening is available only in ABAQUS/Explicit. The volumetric hardening model is motivated by the experimental observation that foam structures usually experience a different response in compression and tension. In compression the ability of the material to deform volumetrically is enhanced by cell wall buckling processes as described by Gibson et al. (1982), Gibson and Ashby (1982), and Maiti et al. (1984). It is assumed that the foam cell deformation is not recoverable instantaneously and can, thus, be idealized as being plastic for short duration events. In tension, on the other hand, cell walls break readily; and as a result the tensile load bearing capacity of crushable foams may be considerably smaller than its compressive load bearing capacity. The volumetric hardening model assumes a perfectly plastic behavior in hydrostatic tension, and the evolution of the yield surface is controlled by the volumetric plastic strain experienced by the material: compactive inelastic strains produce hardening while dilatant inelastic strains lead to softening. The isotropic hardening model was originally developed for metallic foams by Deshpande and Fleck (2000). It assumes symmetric behavior in tension and compression, and the evolution of the yield surface is governed by an equivalent plastic strain, which has contributions from both the volumetric plastic strain and the deviatoric plastic strain. The mechanical behavior of crushable foams is known to be sensitive to the rate of straining. This effect can be introduced by a piecewise linear law or by the overstress power law model.
The strain rate decomposition
Constitutive Theories
= J el
J pl ;
(4.4.61)
where J is the ratio of current volume to original volume, J el is the elastic (recoverable) part of the ratio of current to original foam volume, and J pl is the plastic (nonrecoverable) part of the ratio of current to original foam volume. Volumetric strains are dened as
"vol "vol "vol
pl el
= ln J; = ln J el ; = ln J pl :
vol vol
These denitions and Equation 4.4.61 result in the usual additive strain rate decomposition for volumetric strains:
"vol
_ = "_el + "_pl :
4.4.61
The model also assumes the deviatoric strain rates decompose additively, so that the total strain rates decompose as _ _ _ " = "el + " pl :
Elastic behavior
= Del : " el ;
where el represents the fourth-order elasticity tensor and elastic strain tensors, respectively.
Plastic behavior
and " el
The yield surface and the ow potential for the crushable foam models are dened in terms of the pressure stress 1 1 p = 0 trace = 0 : 3 3 and the Mises stress 3 : : q= 2 The yield surface is dened as
S S
0
pq2 + 2 (p 0 p )2 0 B = 0 p
(4.4.62)
G = q 2 + 2 p2 : (4.4.63) F and G can each be represented as an ellipse in the pq stress plane with and representing the shape of the yield ellipse and the ellipse for the ow potential, respectively; p0 is the center of the yield ellipse, and B is the length of the (vertical) qaxis of the yield ellipse. The ow potential is an
ellipse centered in the origin. The yield surface and the ow potential are depicted in Figure 4.4.61. The parameters p0 and B of the yield ellipse (Equation 4.4.62) are related to the yield strength in hydrostatic compression, pc , and to the yield strength in hydrostatic tension, pt , by
p0 =
p c 0 pt
2
and
B =A=
p c + pt
2
where pc and pt are positive quantities and A is the length of the (horizontal) p-axis of the yield ellipse. The shape factor, , remains as a constant during any plastic deformation process. The evolution of the yield ellipse is controlled by a plastic strain measure, ", which is the volumetric compacting
4.4.62
q flow potential
1
yield surface B
-pt
p0
pc
Figure 4.4.61 Typical yield surface and ow potential for the crushable foam model.
plastic strain, 0"pl , for the volumetric hardening model, and the equivalent plastic strain, "pl (to be vol dened later), for the isotropic hardening model. To dene the hardening behavior, uniaxial compression test data are required. A piecewise linear hardening curve of uniaxial Cauchy stress versus axial (logarithmic) plastic strain must be entered in a tabular form.
Crushable foam model with volumetric hardening
The volumetric hardening model, available in both ABAQUS/Standard and ABAQUS/Explicit, assumes that the hydrostatic tension strength, pt, remains constant throughout any plastic deformation process. By contrast, the hydrostatic compression strength evolves as a result of compaction (increase in density) or dilation (reduction in density) of the material:
Constitutive Theories
pc = pc() "
Yield surface
where
The yield surface for the crushable foam model, depicted in Figure 4.4.62, is dened by
pq2 + 2 (p 0 p )2 0 B = 0;
0
where the parameter represents the shape of the yield ellipse in the pq stress plane and can be 0 calculated from the initial yield strength in uniaxial compression, c , taken as a positive value; the 0; and the yield strength in hydrostatic tension, p ; initial yield strength in hydrostatic compression, pc t as 0 3k c pt = with k = 0 and kt = 0 ; pc pc (3kt + k)(3 0 k)
4.4.63
uniaxial compression
o c
q
3 1
flow potential
-pt
0 c 3
0 pc -pt 2
0 pc
pc
Figure 4.4.62 Yield surfaces and ow potential for the volumetric hardening model. where the yield stress ratios, k(; fi) and kt(; fi ), are provided by the user and can be functions of temperature and other eld variables. For a valid yield surface the choice of yield stress ratios must be such that 0 < k < 3 and kt 0. The yield surface is the Mises circle in the deviatoric stress plane.
Flow potential
The plastic strain rate for the volumetric hardening model is assumed to be
_ pl _ "pl = "
@G ; @
_ : "pl ; G
G=
q2 +
9 2 p : 2
(4.4.64)
This potential is a particular case of Equation 4.4.63 with = 3= 2 2:12. A geometrical representation of the ow potential in the pq stress plane is shown in Figure 4.4.62. Equation 4.4.64 gives a direction of ow that is identical to the stress direction for radial paths. This is motivated by simple laboratory experiments performed by Bilkhu (1987), which suggest that loading in any principal direction causes insignicant deformation in the other directions. As a result, the plastic ow is nonassociative. Therefore, the use of this foam model generally requires the solution of nonsymmetric equations.
4.4.64
Hardening
The yield surface intersects the p-axis at 0pt and pc . We assume that pt remains xed throughout any plastic deformation process. By contrast, the compressive strength, pc, evolves as a result of compaction (increase in density) or dilation (reduction in density) of the material. The evolution of the yield surface can be expressed through the evolution of the yield surface size on the hydrostatic stress axis, pc + pt , as a function of the value of volumetric compacting plastic strain, 0"pl . With pt vol constant, this relation can be obtained from a user-provided uniaxial compression test data using
pc ("pl ) = vol
c ("axial ) c("axial )
pl pt
pt +
c ("pl ) axial
3
1 9
+
pt
3
along with the fact that "pl = "pl in uniaxial compression (due to zero plastic Poissons ratio). vol axial Thus, the user provides input to the hardening law by only specifying, in the usual tabular form, the value of the yield stress in uniaxial compression as a function of the absolute value of the axial plastic strain. The table must start with a zero plastic strain (corresponding to the virgin state of the material), and the tabular entries must be given in ascending magnitude of "pl . If desired, the yield stress can axial also be a function of temperature and other predened eld variables.
Crushable foam model with isotropic hardening
The isotropic hardening model, available in ABAQUS/Explicit only, was originally developed for metallic foams by Deshpande and Fleck (2000). The model assumes similar behaviors in tension and compression. The yield surface is an ellipse centered at the origin in the pq stress plane and evolves in a self-similar manner governed by the equivalent plastic strain.
Yield surface
Constitutive Theories
F
with
q 2 + 2 p2 0 B = 0
r
1+
(4.4.65)
B = p c = c
2
3
where represents the shape of the yield ellipse in the pq stress plane, B denes the size of the yield ellipse, pc is the yield strength in hydrostatic compression, and c is the absolute value of the yield strength in uniaxial compression. The yield surface is the Mises circle in the deviatoric stress plane, and an ellipse in the meridional plane as depicted in Figure 4.4.63.
4.4.65
uniaxial compression
0 c
q
1 3 1
flow potential
yield surface
original surface
0 -pc -p c
0 c 3
pc
pc
Figure 4.4.63 Yield surfaces and ow potential for the isotropic hardening model.
0 The parameter can be calculated using the initial yield stress in uniaxial compression, c , and 0, as the initial yield stress in hydrostatic compression, pc
= p
3k 9 0 k2
with
k=
0 c : p0 c
The strength ratio k is provided by the user and must be in the range of 0 and 3. For many low-density foams the initial yield surface is close to a circle in the pq stress plane, which indicates that the value of is approximately one. The special case of k = 0 corresponds to the Mises yield surface.
Flow potential
G = q 2 + 2 p2 ;
where represents the shape of the ow potential in the pq stress plane and is related to the plastic Poissons ratio, p , by
3 =p 2
1 0 2 p : 1 + p
4.4.66
The plastic Poissons ratio, which is the ratio of the transverse to the longitudinal plastic strain under uniaxial compression, should be dened by the user; and it must be in the range of 01 and 0:5. The upper limit, p = 0:5, corresponds to an incompressible plastic ow. The plastic strains are dened to be normal to a family of self-similar ow potentials parametrized by the value of the potential G _ @G ; _ " pl =
_ where is the nonnegative plastic ow multiplier. The hardening of the foam is described through c = c (pl ), where "pl is the equivalent plastic strain. The evolution of "pl is assumed to be governed " by the equivalent plastic work expression; i.e., _ _ c "pl = : "pl :
The equivalent plastic strain is equal to the absolute value of the axial plastic strain in uniaxial tension or compression. The plastic ow is associative when the value of is the same as that of . In general, the plastic ow is nonassociated to allow for the independent calibrations of the shape of the yield surface and the plastic Poissons ratio. For many low-density foams the plastic Poissons ratio is nearly zero, which corresponds to a value of 2:12.
Hardening
@
A simple uniaxial compression test is sufcient to dene the evolution of the yield surface. The hardening law denes the value of the yield stress in uniaxial compression as a function of the absolute value of the axial plastic strain. The piecewise linear relationship is entered in tabular form. The table must start with a zero plastic strain (corresponding to the virgin state of the materials) and must be given in ascending magnitude of "pl . If desired, the yield stress can also be a function of axial temperature and other predened eld variables.
Reference
Constitutive Theories
Crushable foam plasticity models, Section 11.3.5 of the ABAQUS Analysis Users Manual
4.4.67
CONCRETE
4.5.1
This section describes the smeared crack model provided in ABAQUS/Standard for plain concrete. The material library in ABAQUS also includes a constitutive model for concrete based on theories of scalar plastic damage, described in Damaged plasticity model for concrete and other quasi-brittle materials, Section 4.5.2, which is available both in ABAQUS/Standard and ABAQUS/Explicit. In ABAQUS/Explicit plain concrete can also be analyzed with the cracking model described in A cracking model for concrete and other brittle materials, Section 4.5.3. It is intended that reinforced concrete modeling be accomplished by combining standard elements, using this plain concrete model, with rebar elementsrods, dened singly or embedded in oriented surfaces, that use a one-dimensional strain theory and that may be used to model the reinforcing itself. These elements are superposed on the mesh of plain concrete elements and are used with standard metal plasticity models that describe the behavior of the rebar material. This modeling approach allows the concrete behavior to be considered independently of the rebar, so this section discusses the plain concrete model only. Effects associated with the rebar/concrete interface, such as bond slip and dowel action, cannot be considered in this approach, except by modifying some aspects of the plain concrete behavior to mimic them, such as the use of tension stiffening to simulate load transfer across cracks through the rebar. The theory described in this section is intended as a model of concrete behavior for relatively monotonic loadings under fairly low conning pressures (less than four to ve times the largest compressive stress that can be carried by the concrete in uniaxial compression). Cracking is assumed to be the most important aspect of the behavior, and it dominates the modeling. Cracking is assumed to occur when the stresses reach a failure surface, which we call the crack detection surface. This failure surface is taken to be a simple Coulomb line written in terms of the rst and second stress invariants, p and q, that are dened below. The anisotropy introduced by cracking is assumed to be important in the simulations for which the model is intended, so the model includes consideration of this anisotropy. The model is a smeared crack model, in the sense that it does not track individual macro cracks: rather, constitutive calculations are performed independently at each integration point of the nite element model, and the presence of cracks enters into these calculations by the way the cracks affect the stress and material stiffness associated with the integration point. Various objections have been raised against such smeared crack models. The principal concern is that this modeling approach inherently introduces mesh sensitivity in the solutions, in the sense that the nite element results do not converge to a unique result. For example, since cracking is associated with strain softening, mesh renement will lead to narrower crack bands. Criseld (1986) discusses this concern in detail and concludes that Hilleborgs (1976) approach, based on brittle fracture concepts, is adequate to deal with this issue for practical purposes. This aspect of the model is discussed below in the section on cracking. For simplicity of discussion in what follows, the term crack is used to mean a direction in which cracking has been detected at the single constitutive calculation point in question: the closest physical concept is that there exists a continuum of micro-cracks at the point, oriented as determined by the model. When the principal stress components are dominantly compressive, the response of the concrete is modeled by an elastic-plastic theory, using a simple form of yield surface written in terms of the rst two stress invariants. Associated ow and isotropic hardening are used. This model signicantly simplies
Constitutive Theories
4.5.11
CONCRETE
the actual behavior: the associated ow assumption generally overpredicts the inelastic volume strain; the simple yield surface used does not match all data very accurately (the third stress invariant would be needed to improve this aspect of the model); and, especially when the concrete is strained beyond the ultimate stress point, the assumption of constant elastic stiffness does not reproduce the observation that the unloading response is signicantly weakened (the elastic response of the material appears to be damaged). In addition, when concrete is subjected to very high pressure stress, it exhibits inelastic response: no attempt has been made to build this behavior into the model. In spite of these limitations the model provides useful predictions for a variety of problems involving inelastic loading of concrete. The limitations are introduced for the sake of computational efciency. In particular, assuming associated ow leads to enough symmetry in the Jacobian matrix of the constitutive model (the material stiffness matrix) that the overall equilibrium equation solution usually does not require nonsymmetric equation solution for this reason. All of these limitations could be removed at some sacrice in computational cost. The cracking and compression responses of concrete that are incorporated in the model are illustrated by the uniaxial response of a specimen shown in Figure 4.5.11.
Stress Failure point in compression (peak stress)
Unload/reload response Idealized (elastic) unload/reload response Strain Cracking failure Softening
Figure 4.5.11 Uniaxial behavior of plain concrete. When concrete is loaded in compression, it initially exhibits elastic response. As the stress is increased, some nonrecoverable (inelastic) straining occurs, and the response of the material softens. An ultimate
4.5.12
CONCRETE
stress is reached, after which the material softens until it can no longer carry any stress. If the load is removed at some point after inelastic straining has occurred, the unloading response is softer than the initial elastic response: this effect is ignored in the model. When a uniaxial specimen is loaded into tension, it responds elastically until, at a stress that is typically 710% of the ultimate compressive stress, cracks form so quickly thateven on the stiffest testing machines availableit is very difcult to observe the actual behavior. For the purpose of developing the model, we assume that the material loses strength through a softening mechanism and that this is dominantly a damage effect, in the sense that open cracks can be represented by loss of elastic stiffness (as distinct from the nonrecoverable straining that is associated with classical plasticity effects, such as what we are using for the compressive behavior model). The model neglects any permanent strain associated with cracking; that is, we assume that the cracks can close completely when the stress across them becomes compressive. In multiaxial stress states these observations can be generalized through the concept of surfaces of failure and of ultimate strength in stress space. These surfaces are dened below and are tted to experimental data. Typical surfaces are shown in Figure 4.5.12 and Figure 4.5.13.
"crack detection" surface uniaxial tension 2
Constitutive Theories
"compression" surface
biaxial compression
4.5.13
CONCRETE
u c
2
u c
Figure 4.5.13 Concrete failure surfaces in the (pq) plane. This model makes no attempt to include prediction of cyclic response or of the reduction in the elastic stiffness caused by inelastic straining because the model is intended for application to relatively monotonic loading cases. Nevertheless, it is likely thateven in such casesthe stress trajectories will not be entirely radial and the model must predict the response in such cases in a reasonable way. An isotropically hardening compressive yield surface forms the basis of the model for the inelastic response when the principal stresses are dominantly compressive. In tension once cracking is dened to occur (by the crack detection surface of the model), the orientation of the cracks is stored and oriented, damaged elasticity is then used to model the existing cracks. Stress components associated with an open crack are not included in the denition of the crack detection surface for detecting additional cracks at the same point, and we only allow cracks to form in orthogonal directions at a point. Since ABAQUS/Standard is an implicit, stiffness method code and the material calculations used to dene the behavior of the concrete are carried out independently at each integration point in that part of the model that is made of concrete, the solution is known at the start of the time increment. The constitutive calculations must provide values of stress and material stiffness at the end of the increment, based on the current estimate of the kinematic solution for the response at the spatial integration point during the increment that provides the (logarithmic) strain, ", at the end of the increment. Once cracks exist at a point, the component forms of all vector and tensor valued quantities are rotated so that they lie in the local system dened by the crack orientation vectors (the normals to the crack faces). The model ensures that these crack face normal vectors will be orthogonal, so that this local system is rectangular Cartesian. This use of a local system simplies the computation of the damaged elasticity used for the components associated with existing cracks.
4.5.14
CONCRETE
The model, thus, consists of a compressive yield/ow surface to model the concrete response in predominantly compressive states of stress, together with damaged elasticity to represent cracks that have occurred at a material calculation point, the occurrence of cracks being dened by a crack detection failure surface that is considered to be part of the elasticity. The details of this model are now presented.
Elastic-plastic model for concrete
The model uses the classical concepts of plasticity theory: a strain rate decomposition into elastic and inelastic strain rates, elasticity, yield, ow, and hardening.
Strain rate decomposition
(4.5.11)
" " where d" is the total mechanical strain rate, d" is the elastic strain rate (which includes crack detection strainsthis elastic strain will be further decomposed when we describe the elasticity), and "c d"pl is the plastic strain rate associated with the compression surface. We assume that the elastic part of the strain is always small, so this equation can be integrated as " = "el + " pl : c
Compression yield
(4.5.12)
fc = q 0
3a 0 p
3c = 0 ;
(4.5.13)
Constitutive Theories
p = 0 trace( );
1 3
q=
3 2
S : S;
where S = + p I are the deviatoric stress components; a0 is a constant, which is chosen from the ratio of the ultimate stress reached in biaxial compression to the ultimate stress reached in uniaxial compression; and c (c ) is a hardening parameter (c is the size of the yield surface on the q-axis at p = 0, so that c is the yield stress in a state of pure shear stress when all components of are zero except 12 = 21 = c ). The hardening is measured by the value of c : the c (c ) relationship is user-specied.
4.5.15
CONCRETE
This simple surface is a straight line in (pq) space and provides a good match to experimental data over a fairly wide range of pressure stress values (up to four to ve times the maximum compressive stress that can be carried by the concrete in uniaxial compression). This form of the surface means that, as the hardening (c ) changes, the surfaces in (pq) space are similar, sofor examplethe ratio of ow stress in biaxial loading to ow stress in uniaxial loading is the same at all ow stress levels. This does not appear to be contradicted by any experimental data, and it means that only one constant (a0 ) is needed to dene the shape of the surface. The value of a0 is established from the users data as follows. In uniaxial compression p = 1 c 3 and q = c , where c is the stress magnitude. Therefore, on fc = 0,
c c
In biaxial compression p = 2=3 bc and q stress. Therefore, on fc = 0,
= =
p 0 3
1
a0
3
(4.5.14)
c bc
=
1 p 0 2a 0 3 3
(4.5.15)
u u The value of bc =c = rbc is specied by the user as part of the failure surface data (typically 1:16). a can be calculated from Equation 4.5.14 and Equation 4.5.15 as rbc 0
a0 =
0 rbc : 1 0 2rbc
1
The user denes the hardening by specifying the magnitude of the stress, j11j, in a uniaxial compression test as a function of the inelastic strain magnitude, j"11 j. These data are used to dene the c (c ) relationship, as follows. In uniaxial compression p = 1 c and q = c , where c is the stress magnitude. During active 3 plastic loading fc = 0, so by using the denition of fc (Equation 4.5.14), we obtain c immediately as
c =
Flow
p 0
1 3
a0
3
c :
(4.5.16)
=0
"c d"pl =
dc
1 + c0
2 !
p c
@fc ; @
(4.5.17)
4.5.16
CONCRETE
"c otherwise, d" pl = 0. "c In the denition of d"pl , c0 is a constant that is chosen so that the ratio of "pl in a monotonically 11 " loaded biaxial compression test to "pl in a monotonically loaded uniaxial compression test is rbc , a 11 " 1:28). The equation value specied by the user as part of the failure surface data (typically rbc " dening c0 from rbc and the other constants in the compression surface is derived next. The gradient of the ow potential for the compressive surface is @fc @
Since
=
@q @
3 a0
@p : @
@p @ @q @ @fc @
= =
01 I 3
3 S 2
and
;
I:
then
3 S 2
a p0
In uniaxial compression p = 1 c , q 3
0
1c d"pl 11 = c
c0
9
p 01
3
a0
(4.5.18)
p 01
3
(4.5.19)
Constitutive Theories
so c is known from "pl 11 and the constants a0 and c0 . Equation 4.5.16 and Equation 4.5.19, c therefore, dene the c (c ) relationship from the concrete model input data once c0 is known. 0 1bc 0 1c " The constant c0 is calculated from the users denition of rbc, the ratio of "pl 11 to "pl 11 , c c the total plastic strain components that would occur in monotonically loaded biaxial and uniaxial compression tests. In biaxial compression when both nonzero principal stresses have the magnitude bc, p = 2 bc = 2 rbc c , q = bc = rbc c , and S11 = 0 1 rbc c, so the ow rule gives 3 3 3
0
d"pl c
1bc
=
11
dc
1+
4 9
(rbc )
2c
0
a p0 0 1 2
3
" Using this equation and Equation 4.5.18 then denes c0 from rbc and the other constants as
c0 = 9
4.5.17
CONCRETE
Cracking dominates the material behavior when the state of stress is predominantly tensile. The model uses a crack detection plasticity surface in stress space to determine when cracking takes place and the orientation of the cracking. Damaged elasticity is then used to describe the postfailure behavior of the concrete with open cracks. Numerically we use the crack detection plasticity model for the increment in which cracking takes place and subsequently use damaged elasticity once the cracks presence and orientation have been detected. As a result there is at least one increment in which we calculate crack detection plastic strains. Since these are really just the outcome of a numerical device to treat cracking, they are recast as elastic strains in the direction of cracking and as plastic strains in the other directions. (This means that we retain the stresses calculated for equilibrium purposes, as well as the strain decomposition of Equation 4.5.11.) The basis of the postcracked behavior is the brittle fracture concept of Hilleborg (1976). We assume that the fracture energy required to form a unit area of crack surface, Gf , is a material property. This value can be calculated from measuring the tensile stress as a function of the crack opening displacement (Figure 4.5.14), as
Gf
t
t du:
u t
ucr u u
el
u, displacement
4.5.18
CONCRETE
5700 lb/in2 ). The implication of assuming that Gf is a material property is that, when the elastic part of the displacement, uel ; is eliminated, the relationship between the stress and the remaining part of the displacement, ucr = u 0 uel ; is xed, regardless of the specimen size. For example, consider a specimen developing a single crack across its section as tensile displacement is applied to it: ucr is the displacement across the crack and is not changed by using a longer or shorter specimen in the test (so long as the specimen is signicantly longer than the width of the crack band, which will typically be of the order of the aggregate size). Thus, one important part of the cracked concretes tensile behavior is dened in terms of a stress/displacement relationship. In the nite element implementation of this model we must, therefore, compute the relative displacement at an integration point to provide ucr . We do this in ABAQUS by multiplying the strain by a characteristic length associated with the integration point. The characteristic crack length is based on the element geometry: for beams and trusses we use the integration point length; for shell and planar elements we use the square root of the integration point area; for solid elements we use the cube root of the integration point volume. This denition of the characteristic length is used because we do not necessarily know in which direction the concrete will crack and so cannot choose the length measure in any particular direction. Thus, if there are elements in the model that have large aspect ratios, the model will likely provide different results if it is loaded in different directions and cracking occurs in such elements. This effect should be considered by the user in dening values for the material properties. In reinforced concrete the ucr relationship must also represent the action of the bond between the concrete and the rebar as the concrete cracks. We assume this is accommodated by increasing the value of Gf based on comparisons with experiments on reinforced material. We rst describe the crack detection plasticity model and then discuss the damaged elasticity.
Strain rate decomposition
Typical values of Gf range from 40 N/m (0.22 lb/in) for a typical construction concrete (with u u c 20 MPa, 2850 lb/in2 ) to 120 N/m (0.67 lb/in) for a high strength concrete (with c 40 MPa,
Constitutive Theories
(4.5.110)
" "d where d"el is the total mechanical strain rate for the crack detection problem, d"el is the elastic strain "t rate, and d"pl is the plastic strain rate associated with the crack detection surface.
Yield
b t ft = q 0 ^ (4.5.111) p 0 2 0 0 u t = 0 ; ^ 3 t u where t is the failure stress in uniaxial tension and b0 is a constant that is dened from the value of the tensile failure stress, I , in a state of biaxial stress when the other nonzero principal stress, u II , is at the uniaxial compression ultimate stress value, c . t(t) is a hardening parameter (t is t 3 0 b0 u t
4.5.19
CONCRETE
the equivalent uniaxial tensile stress). The hardening is measured by t, with the dened from the user-specied tension stiffening data (see Figure 4.5.15).
t(t ) relationship
Stress,
Failure point
u = t
t
E
Strain,
4.5.110
CONCRETE
u u = 0c (c is the magnitude of the ultimate stress in uniaxial compression) and the other nonzero principal stress has the value II = fu . Another value provided as part of the failure surface data t
I
is r , which denes t
u u t = rt c :
Cracking would, therefore, occur at the point with principal stresses values
For these
p=
and
q
u Therefore, with t = t , u ft = c
u = c
(1
u 0 frt ) c
2
1 + (frt ) +
frt :
1 + ( frt ) + frt
u 0 1 (3 0 b0 ) (1 0 frt ) c 0 3
2
0 b30
:
u rt c = 0 ;
so
b0 = 3
1 + (2
f )r t
1 + rt (1
1 + (frt ) + frt
0 f)
The crack detection surface is shown in Figure 4.5.12 and Figure 4.5.13.
Flow
=0
@ft ; @
Constitutive Theories
The user denes the tension stiffening behavior by specifying the magnitude of the stress, t , in a uniaxial tension test as a function of the inelastic strain. (When the fracture energy concept is used to dene the postfailure behavior, strain is now dened as ucr =c, where c is the characteristic length associated with the integration point.) The t(t ) relationship is dened as follows. Using the denition of ft , Equation 4.5.111, in the ow rule above we write
"t d"pl =
In uniaxial tension S11
= 3 t
dt
=
3 2
S
q
b0 t 10 u 3 t
I
and q
CONCRETE
and, hence,
dt = d"pl t
11
20
Upon integration Equation 4.5.113 gives t from obtained from the tension stiffening input data.
Damaged elasticity
b0 t u 3 t
11
(4.5.113)
d"pl t
Following crack detection we use damaged elasticity to model the failed material. The elasticity is written in the form
is the elastic stiffness matrix for the concrete. where Let represent a cracked direction, with corresponding direct stress and direct elastic strain "el . In these expressions and in the remainder of this section, no summation is implied by repeated indices with a bar over them. If the fracture energy concept is used, the strains are related to the user-specied stress/displacement denition for the tension stiffening behavior by " = u=c, where c is the characteristic length associated with the integration point. Then, in , D is the usual elasticity of the concrete if " 0. If "open > " > 0,
= D : " el ;
(4.5.114)
D =
open
"open ;
open where is the stress corresponding to "open (as dened in the tension stiffening data), and
"open =
If " = "open ,
over history
max
0 el 1 " :
open D = d d"open ;
which follows from the tension stiffening data. We also assume no Poisson effect for open cracks: for "el > 0, D = 0 for The shear terms in the elasticity associated with existing crack directions are
6= , 6= .
^ D = G; 6= ; ^ ^ where G = %close G for " 0; and G = %open G for " > 0: In these expressions G is the elastic shear modulus, %close is a constant dened by the user as part of the shear retention data (see Figure 4.5.16), and %open is a linear function of "el , %open = (1 0 "el ="max ) and is also dened by
4.5.112
CONCRETE
1.0
close
max
Constitutive Theories
As soon as the crack detection surface (ft ) has been activated, we assume that cracking has occurred. The crack direction, n, is taken to be the direction of that part of the maximum principal plastic strain increment conjugate to the crack detection surface, "pl , that is orthogonal to the directions t of any existing cracks at the same point. This crack orientation is stored for subsequent calculations, which are done for convenience in a local coordinate system oriented so that one of the coordinate
4.5.113
CONCRETE
directions is the crack direction, n . Cracking is irrecoverable in the sense that, once a crack has occurred at a point, it remains throughout the rest of the calculation. Following crack detection, the crack affects the calculations by damaging the elasticity, as dened above. Also, if the elastic strain across a crack is tensile, the invariants used in the crack detection surface are dened in the stress sub-space in which all stress components associated with the open crack direction are neglected, as described in the section above on yield. This implies that no more than three cracks can occur at any point (two in a plane stress case, one in a uniaxial stress case).
Integration of the model
The model is integrated using the backward Euler method generally used with the plasticity models in ABAQUS. A material Jacobian consistent with this integration operator is used for the equilibrium iterations.
Reference
Concrete smeared cracking, Section 11.5.1 of the ABAQUS Analysis Users Manual
4.5.114
4.5.2
This section describes the concrete damaged plasticity model provided in ABAQUS for the analysis of concrete and other quasi-brittle materials. The material library in ABAQUS also includes other constitutive models for concrete based on the smeared crack approach. These are the smeared crack model in ABAQUS/Standard, described in An inelastic constitutive model for concrete, Section 4.5.1, and the brittle cracking model in ABAQUS/Explicit, described in A cracking model for concrete and other brittle materials, Section 4.5.3. The concrete damaged plasticity model is primarily intended to provide a general capability for the analysis of concrete structures under cyclic and/or dynamic loading. The model is also suitable for the analysis of other quasi-brittle materials, such as rock, mortar and ceramics; but it is the behavior of concrete that is used in the remainder of this section to motivate different aspects of the constitutive theory. Under low conning pressures, concrete behaves in a brittle manner; the main failure mechanisms are cracking in tension and crushing in compression. The brittle behavior of concrete disappears when the conning pressure is sufciently large to prevent crack propagation. In these circumstances failure is driven by the consolidation and collapse of the concrete microporous microstructure, leading to a macroscopic response that resembles that of a ductile material with work hardening. Modeling the behavior of concrete under large hydrostatic pressures is out of the scope of the plasticdamage model considered here. The constitutive theory in this section aims to capture the effects of irreversible damage associated with the failure mechanisms that occur in concrete and other quasi-brittle materials under fairly low conning pressures (less than four or ve times the ultimate compressive stress in uniaxial compression loading). These effects manifest themselves in the following macroscopic properties: different yield strengths in tension and compression, with the initial yield stress in compression being a factor of 10 or more higher than the initial yield stress in tension; softening behavior in tension as opposed to initial hardening followed by softening in compression; different degradation of the elastic stiffness in tension and compression; stiffness recovery effects during cyclic loading; and rate sensitivity, especially an increase in the peak strength with strain rate. The plastic-damage model in ABAQUS is based on the models proposed by Lubliner et al. (1989) and by Lee and Fenves (1998). The model is described in the remainder of this section. An overview of the main ingredients of the model is given rst, followed by a more detailed discussion of the different aspects of the constitutive model.
Overview
Constitutive Theories
The main ingredients of the inviscid concrete damaged plasticity model are summarized below.
4.5.21
_ _ _ " = "el + " pl ; _ _ _ where " is the total strain rate, "el is the elastic part of the strain rate, and " pl is the plastic part of the strain rate.
Stress-strain relations
where el is the initial (undamaged) elastic stiffness of the material; el = (1 0 d) el is the degraded 0 0 elastic stiffness; and d is the scalar stiffness degradation variable, which can take values in the range from zero (undamaged material) to one (fully damaged material). Damage associated with the failure mechanisms of the concrete (cracking and crushing) therefore results in a reduction in the elastic stiffness. Within the context of the scalar-damage theory, the stiffness degradation is isotropic and characterized by a single degradation variable, d. Following the usual notions of continuum damage mechanics, the effective stress is dened as
= def
The Cauchy stress is related to the effective stress through the scalar degradation relation:
= (1 0 d) :
For any given cross-section of the material, the factor (1 0 d) represents the ratio of the effective load-carrying area (i.e., the overall area minus the damaged area) to the overall section area. In the absence of damage, d = 0, the effective stress is equivalent to the Cauchy stress, . When damage occurs, however, the effective stress is more representative than the Cauchy stress because it is the effective stress area that is resisting the external loads. It is, therefore, convenient to formulate the plasticity problem in terms of the effective stress. As discussed later, the evolution of the degradation ~ ~ variable is governed by a set of hardening variables, " pl , and the effective stress; that is, d = d( ; "pl ).
Hardening variables
Damaged states in tension and compression are characterized independently by two hardening ~t ~c variables, "pl and "pl , which are referred to as equivalent plastic strains in tension and compression, respectively. The evolution of the hardening variables is given by an expression of the form
4.5.22
as described later in this section. Microcracking and crushing in the concrete are represented by increasing values of the hardening variables. These variables control the evolution of the yield surface and the degradation of the elastic stiffness. They are also intimately related to the dissipated fracture energy required to generate microcracks.
Yield function
The yield function, F ( ; ~pl ), represents a surface in effective stress space, which determines the states " of failure or damage. For the inviscid plastic-damage model
F ( ; ~pl ) 0: "
The specic form of the yield function is described later in this section.
Flow rule
_ _ "pl =
@G( ) ; @
_ where is the nonnegative plastic multiplier. The plastic potential is dened in the effective stress space. The specic form of the ow potential for the concrete damaged plasticity model is discussed later in this section. The model uses nonassociated plasticity, therefore requiring the solution of nonsymmetric equations.
Summary
Constitutive Theories
In summary, the elastic-plastic response of the concrete damaged plasticity model is described in terms of the effective stress and the hardening variables:
_ _ _ where and F obey the Kuhn-Tucker conditions: F = 0; 0; F 0. The Cauchy stress is " calculated in terms of the stiffness degradation variable, d( ; ~pl ), and the effective stress as
= (1 0 d) :
(4.5.22)
The constitutive relations for the elastic-plastic response, Equation 4.5.21, are decoupled from the stiffness degradation response, Equation 4.5.22, which makes the model attractive for an effective numerical implementation. The inviscid model summarized here can be extended easily to account
4.5.23
for viscoplastic effects through the use of a viscoplastic regularization by permitting stresses to be outside the yield surface.
Damage and stiffness degradation
The evolution equations of the hardening variables "pl and "pl are conveniently formulated by ~t ~c considering uniaxial loading conditions rst and then extended to multiaxial conditions.
Uniaxial conditions
It is assumed that the uniaxial stress-strain curves can be converted into stress versus plastic strain curves of the form
_ t = t (~t ; "t ; ; fi ); " ~ _ c = c (~pl ; "c ; ; fi ); "c ~
pl pl pl pl
(4.5.23)
pl
_ _ ~ ~ where the subscripts t and c refer to tension and compression, respectively; "t and "c are the R t pl R pl pl _ dt and "pl = t " dt are the equivalent plastic strains, is _ ~ ~c equivalent plastic strain rates, "t = 0 "t ~ 0 ~c the temperature, and fi ; (i = 1; 2; :::) are other predened eld variables. Under uniaxial loading conditions the effective plastic strain rates are given as _ _ "t = "11 ; in uniaxial tension and ~ _ "c = ~
pl pl pl
0"_pl ; 11
(4.5.24)
in uniaxial compression.
In the remainder of this section we adopt the convention that c is a positive quantity representing the magnitude of the uniaxial compression stress; that is, c = 011. As shown in Figure 4.5.21, when the concrete specimen is unloaded from any point on the strain softening branch of the stress-strain curves, the unloading response is observed to be weakened: the elastic stiffness of the material appears to be damaged (or degraded). The degradation of the elastic stiffness is signicantly different between tension and compression tests; in either case, the effect is more pronounced as the plastic strain increases. The degraded response of concrete is characterized by two independent uniaxial damage variables, dt and dc , which are assumed to be functions of the plastic strains, temperature, and eld variables:
"c dc = dc(~pl ; ; fi); dt = dt(~t ; ; fi ); (0 "
pl
dt 1); (0 dc 1):
(4.5.25)
The uniaxial degradation variables are increasing functions of the equivalent plastic strains. They can take values ranging from zero, for the undamaged material, to one, for the fully damaged material. If E0 is the initial (undamaged) elastic stiffness of the material, the stress-strain relations under uniaxial tension and compression loading are, respectively:
4.5.24
t0
(a)
E0
(1 dt )E 0
~ pl
tel
(b)
c u c 0
E0
(1_ dc )E 0
~ pl
el c
Constitutive Theories
Figure 4.5.21 Response of concrete to uniaxial loading in tension (a) and compression (b). Under uniaxial loading cracks propagate in a direction transverse to the stress direction. The nucleation and propagation of cracks, therefore, causes a reduction of the available load-carrying area, which in turn leads to an increase in the effective stress. The effect is less pronounced under compressive loading since cracks run parallel to the loading direction; however, after a signicant amount of crushing, the effective load-carrying area is also signicantly reduced. The effective uniaxial cohesion stresses, t and c , are given as
The effective uniaxial cohesion stresses determine the size of the yield (or failure) surface.
Uniaxial cyclic conditions
Under uniaxial cyclic loading conditions the degradation mechanisms are quite complex, involving the opening and closing of previously formed micro-cracks, as well as their interaction. Experimentally, it is observed that there is some recovery of the elastic stiffness as the load changes sign during a uniaxial cyclic test. The stiffness recovery effect, also known as the unilateral effect, is an important aspect of the concrete behavior under cyclic loading. The effect is usually more pronounced as the load changes from tension to compression, causing tensile cracks to close, which results in the recovery of the compressive stiffness. The concrete damaged plasticity model assumes that the reduction of the elastic modulus is given in terms of a scalar degradation variable, d, as
E = (1 0 d)E0 ;
where E0 is the initial (undamaged) modulus of the material. This expression holds both in the tensile (11 > 0) and compressive (11 < 0) sides of the cycle. The stiffness reduction variable, d, is a function of the stress state and the uniaxial damage variables, dt and dc . For the uniaxial cyclic conditions, ABAQUS assumes that
(1 0 d) = (1 0 st dc )(1 0 sc dt);
0 s t ; sc
1;
(4.5.26)
where st and sc are functions of the stress state that are introduced to represent stiffness recovery effects associated with stress reversals. They are dened according to
r (11) = H (11) =
1 if 0 if
The weight factors wt and wc, which are assumed to be material properties, control the recovery of the tensile and compressive stiffness upon load reversal. To illustrate this, consider the example in Figure 4.5.22, where the load changes from tension to compression. Assume that there was no ~c previous compressive damage (crushing) in the material; that is, "pl = 0 and dc = 0. Then
(1 0 d) = (1 0 sc dt) = (1 0 (1 0 wc (1 0 r3 ))dt ):
In tension (11 > 0), r3 = 1; thus, d = dt as expected. In compression (11 < 0), r3 = 0, and d = (1 0 wc)dt. If wc = 1, then d = 0; therefore, the material fully recovers the compressive stiffness (which in this case is the initial undamaged stiffness, E = E0). If, on the other hand, wc = 0, then d = dt and there is no stiffness recovery. Intermediate values of wc result in partial recovery of the stiffness.
4.5.26
t0
E0
(1 dt )E 0
t wc = 1 wc = 0
Figure 4.5.22 Illustration of the effect of the compression stiffness recovery parameter wc . The evolution equations of the equivalent plastic strains are also generalized to the uniaxial cyclic conditions as
pl
(4.5.27)
which clearly reduces to Equation 4.5.24 during the tensile and compressive phases of the cycle.
Constitutive Theories
Multiaxial conditions
The evolution equations for the hardening variables must be extended for the general multiaxial conditions. Based on Lee and Fenves (1998) we assume that the equivalent plastic strain rates are evaluated according to the expressions
pl
pl
(4.5.28)
pl pl " _ " _ where ^max and ^min are, respectively, the maximum and minimum eigenvalues of the plastic strain _ rate tensor " pl and 3 ^ h i ^ ^ def r( ) = i=1 i ; 0 r( ) 1 3 ^ ji j i=1
P P
^ is a stress weight factor that is equal to one if all principal stresses i ; (i = 1; 2; 3), are positive and equal to zero if they are negative. The Macauley bracket h1i is dened by hxi = 1 (jxj + x). In 2
4.5.27
uniaxial loading conditions Equation 4.5.28 reduces to the uniaxial denitions Equation 4.5.24 and pl pl _11 _11 Equation 4.5.27, since ^max = "pl in tension, and ^min = "pl in compression. " _ " _ " _ If the eigenvalues of the plastic strain rate tensor (^i; i = 1; 2; 3) are ordered such that ^ ^pl = ^1 "2 ^3 = ^pl , the evolution equation for general multiaxial stress conditions can be _ _ " _ "min _ "max " _ expressed in the following matrix form:
"
pl
where
0 0
^ 0(1 0 r( ))
and
The plastic-damage concrete model assumes that the elastic stiffness degradation is isotropic and characterized by a single scalar variable, d:
el Del = (1 0 d)D0 ;
0 d 1:
(4.5.29)
The denition of the scalar degradation variable d must be consistent with the uniaxial monotonic responses (dt and dc), and it should also should capture the complexity associated with the degradation mechanisms under cyclic loading. For the general multiaxial stress conditions ABAQUS assumes that
(1 0 d) = (1 0 st dc )(1 0 sc dt);
0 s t ; sc
1;
(4.5.210)
^ similar to the uniaxial cyclic case, only that st and sc are now given in terms of the function r( ) as ^ st = 1 0 wtr( ); 0 wt 1; ^ sc = 1 0 wc(1 0 r()); 0 wc 1:
It can be easily veried that Equation 4.5.210 for the scalar degradation variable is consistent with the uniaxial response. The experimental observation in most quasi-brittle materials, including concrete, is that the compressive stiffness is recovered upon crack closure as the load changes from tension to compression. On the other hand, the tensile stiffness is not recovered as the load changes from compression to tension once crushing micro-cracks have developed. This behavior, which corresponds to wt = 0 and wc = 1, is the default used by ABAQUS. Figure 4.5.23 illustrates a uniaxial load cycle assuming the default behavior.
4.5.28
t t 0
E0
wt = 1
wt = 0 wc = 0
(1-d t )E 0
(1-d c)E 0
(1-d t ) (1-dc )E 0
wc = 1
E0
Figure 4.5.23 Uniaxial load cycle (tension-compression-tension) assuming default values for the stiffness recovery factors: wt = 0 and wc = 1.
Yield condition
Constitutive Theories
The plastic-damage concrete model uses a yield condition based on the yield function proposed by Lubliner et al. (1989) and incorporates the modications proposed by Lee and Fenves (1998) to account for different evolution of strength under tension and compression. In terms of effective stresses the yield function takes the form
F ( ; ~pl ) = "
where
10
(4.5.211)
q=
3 S:S 2
4.5.29
S = pI + ; ^ is the deviatoric part of the effective stress tensor ; and max is the algebraically maximum eigenvalue . The function (~pl ) is given as " of
(~pl ) = "
where t and c are the effective tensile and compressive cohesion stresses, respectively. ^ In biaxial compression, with max = 0, Equation 4.5.211 reduces to the well-known DruckerPrager yield condition. The coefcient can be determined from the initial equibiaxial and uniaxial compressive yield stress, b0 and c0, as
b0 0 c0 : 2 b0 0 c0
Typical experimental values of the ratio b0=c0 for concrete are in the range from 1.10 to 1.16, yielding values of between 0.08 and 0.12 (Lubliner et al., 1989). The coefcient enters the yield function only for stress states of triaxial compression, when ^ max < 0: This coefcient can be determined by comparing the yield conditions along the tensile and compressive meridians. By denition, the tensile meridian (TM) is the locus of stress states satisfying ^ ^ ^ ^ the condition max = 1 > 2 = 3 and the compressive meridian (CM) is the locus of stress states ^ ^ ^ ^ ^ ^ ^ max = 1 = 2 > 3, where 1, 2 , and 3 are the eigenvalues of the effective stress such that ^ max )TM = 2 q 0 p and (max )CM = 1 q 0 p, along the tensile ^ tensor. It can be easily shown that ( 3 3 ^ max < 0 the corresponding yield conditions are and compressive meridians, respectively. With
1
3
Let
2 + 1 q 0 ( + 3) = (1 0 )c; p 3
(TM) (CM)
+ 1 q 0 ( + 3) = (1 0 )c : p +3 : 2 + 3
^ Kc = q(TM) =q(CM) for any given value of the hydrostatic pressure p with max < 0; then
Kc =
The fact that Kc is constant does not seem to be contradicted by experimental evidence (Lubliner et al., 1989). The coefcient is, therefore, evaluated as
=
A value of
3(1 0 Kc ) : 2K c 0 1
^ If max
> 0, the yield conditions along the tensile and compressive meridians reduce to
1
3
Let
2 + 1 q 0 ( + 3) = (1 0 )c ; p 3
(TM) (CM)
+ 1 q 0 ( + 3) = (1 0 )c : p +3 : 2 + 3
^ Kt = q(TM) =q(CM) for any given value of the hydrostatic pressure p with max > 0; then
Kt =
Typical yield surfaces are shown in Figure 4.5.24 in the deviatoric plane and in Figure 4.5.25 for plane-stress conditions.
_S
K c = 2/3
_S
Kc = 1
(T.M.)
Constitutive Theories
(C.M.)
_
S3
Figure 4.5.24 Yield surfaces in the deviatoric plane, corresponding to different values of
Flow rule
Kc .
_ _ "pl =
@G( ) : @
The ow potential G chosen for this model is the Drucker-Prager hyperbolic function:
4.5.211
1 (q - 3 p + 2 ) = c0 1-
uniaxial tension
2 t0 1
1 (q - 3 p + 1 ) = c0 1-
(b0 ,b0 )
c0
biaxial compression
1 (q - 3 p ) = c0 1-
p(
t0
tan )2 + q2 0 p tan ;
(4.5.212)
where is the dilation angle measured in the pq plane at high conning pressure; t0 is the uniaxial tensile stress at failure; and is a parameter, referred to as the eccentricity, that denes the rate at which the function approaches the asymptote (the ow potential tends to a straight line as the eccentricity tends to zero). This ow potential, which is continuous and smooth, ensures that the ow direction is dened uniquely. The function asymptotically approaches the linear Drucker-Prager ow potential at high conning pressure stress and intersects the hydrostatic pressure axis at 90. See Models for granular or polymer behavior, Section 4.4.2, for further discussion of this potential. Because plastic ow is nonassociated, the use of the plastic-damage concrete model requires the solution of nonsymmetric equations.
Viscoplastic regularization
Material models exhibiting softening behavior and stiffness degradation often lead to severe convergence difculties in implicit analysis programs. Some of these convergence difculties can be overcome by using a viscoplastic regularization of the constitutive equations. The concrete damaged plasticity model can be regularized using viscoplasticity, therefore permitting stresses to be outside of
4.5.212
the yield surface. We use a generalization of the Duvaut-Lions regularization, according to which the _v viscoplastic strain rate tensor, "pl , is dened as
"pl = _v
("
pl
0 "pl ): v
(4.5.213)
Here is the viscosity parameter representing the relaxation time of the viscoplastic system and "pl is the plastic strain evaluated in the inviscid backbone model. Similarly, a viscous stiffness degradation variable, dv , for the viscoplastic system is dened as
d_v =
(d
0 d v );
(4.5.214)
where d is the degradation variable evaluated in the inviscid backbone model. The stress-strain relation of the viscoplastic model is given as
, where The solution of the viscoplastic system relaxes to that of the inviscid case as t= t represents time. Using the viscoplastic regularization with a small value for the viscosity parameter (small compared to the characteristic time increment) usually helps improve the rate of convergence of the model in the softening regime, without compromising results.
Integration of the model
!1
(4.5.215)
The model is integrated using the backward Euler method generally used with the plasticity models in ABAQUS. A material Jacobian consistent with this integration operator is used for the equilibrium iterations.
Constitutive Theories
4.5.213
CRACKING MODEL
4.5.3
This section describes the cracking constitutive model provided in ABAQUS/Explicit for concrete and other brittle materials. The material library in ABAQUS also includes a constitutive model for concrete based on theories of scalar plastic damage, described in Damaged plasticity model for concrete and other quasi-brittle materials, Section 4.5.2, which is available in ABAQUS/Standard and ABAQUS/Explicit. In ABAQUS/Standard plain concrete can also be analyzed with the smeared crack concrete model described in An inelastic constitutive model for concrete, Section 4.5.1. Although this brittle cracking model can also be useful for other materials, such as ceramics and brittle rocks, it is primarily intended to model plain concrete. Therefore, in the remainder of this section, the physical behavior of concrete is used to motivate the different aspects of the constitutive model. Reinforced concrete modeling in ABAQUS is accomplished by combining standard elements, using this plain concrete cracking model, with rebar elementsrods, dened singly or embedded in oriented surfaces, that use a one-dimensional strain theory and that can be used to model the reinforcing itself. The rebar elements are superposed on the mesh of plain concrete elements and are used with standard metal plasticity models that describe the behavior of the rebar material. This modeling approach allows the concrete behavior to be considered independently of the rebar, so this section discusses the plain concrete cracking model only. Effects associated with the rebar/concrete interface, such as bond slip and dowel action, cannot be considered in this approach except by modifying some aspects of the plain concrete behavior to mimic them (such as the use of tension stiffening to simulate load transfer across cracks through the rebar). It is generally accepted that concrete exhibits two primary modes of behavior: a brittle mode in which microcracks coalesce to form discrete macrocracks representing regions of highly localized deformation, and a ductile mode where microcracks develop more or less uniformly throughout the material, leading to nonlocalized deformation. The brittle behavior is associated with cleavage, shear and mixed mode fracture mechanisms that are observed under tension and tension-compression states of stress. It almost always involves softening of the material. The ductile behavior is associated with distributed microcracking mechanisms that are primarily observed under compression states of stress. It almost always involves hardening of the material, although subsequent softening is possible at low conning pressures. The cracking model described here models only the brittle aspects of concrete behavior. Although this is a major simplication, there are many applications where only the brittle behavior of the concrete is signicant; and, therefore, the assumption that the material is linear elastic in compression is justied in those cases.
Smeared cracking assumption
Constitutive Theories
A smeared model is chosen to represent the discontinuous macrocrack brittle behavior. In this approach we do not track individual macro cracks: rather, the presence of cracks enters into the calculations by the way the cracks affect the stress and material stiffness associated with each material calculation point. Here, for simplicity, the term crack is used to mean a direction in which cracking has been detected at the material calculation point in question. The closest physical concept is that there
4.5.31
CRACKING MODEL
exists a continuum of microcracks at the point, oriented as determined by the model. The anisotropy introduced by cracking is included in the model since it is assumed to be important in the simulations for which the model is intended. Some objections have been raised against smeared crack models. The principal concern is that this modeling approach inherently introduces mesh sensitivity in the solutions, in the sense that the nite element results do not converge to a unique result. For example, since cracking is associated with strain softening, mesh renement will lead to narrower crack bands. Many researchers have addressed this concern, and the general consensus is that Hilleborgs (1976) approachbased on brittle fracture conceptsis adequate to deal with this issue for practical purposes. A length scale, typically in the form of a characteristic length, is introduced to regularize the smeared continuum models and attenuate the sensitivity of the results to mesh density. This aspect of the model is discussed in detail later.
Crack direction assumptions
Various researchers have proposed three basic crack direction models (Rots and Blaauwendraad, 1989): xed, orthogonal cracks; the rotating crack model; and xed, multidirectional (nonorthogonal) cracks. In the xed, orthogonal crack model the direction normal to the rst crack is aligned with the direction of maximum tensile principal stress at the time of crack initiation. The model has memory of this crack direction, and subsequent cracks at the point under consideration can only form in directions orthogonal to the rst crack. In the rotating crack concept only a single crack can form at any point (aligned with the direction of maximum tensile principal stress). Thus, the single crack direction rotates with the direction of the principal stress axes. This model has no memory of crack direction. Finally, the multidirectional crack model allows the formation of any number of cracks at a point as the direction of the principal stress axes changes with loading. In practice, some limitation is imposed on the number of cracks allowed to form at a point. The model has memory of all crack directions. The multidirectional crack model is the least popular, mainly because the criterion used to decide when subsequent cracks form (to limit the number of cracks at a point) is somewhat arbitrary: the concept of a threshold angle is introduced to prevent new cracks from forming at angles less than this threshold value to existing cracks. The xed orthogonal and rotating crack models have both been used extensively, even though objections can be raised against both. In the rotating crack model the concept of crack closing and reopening is not well-dened because the orientation of the crack can vary continuously. The xed orthogonal crack model has been criticized mainly because the traditional treatment of shear retention employed in the model tends to make the response of the model too stiff. This problem can be resolved by formulating the shear retention in a way that ensures that the shear stresses tend to zero as deformation on the crack interfaces takes place (this is done in the ABAQUS model, as described later). Finally, although the xed orthogonal crack model has the orthogonality limitation, it is considered superior to the rotating crack model in cases where the effect of multiple cracks is important (the rotating crack model is restricted to a single crack at any point). The xed orthogonal cracks model is used in ABAQUS so that the maximum number of cracks at a material point is limited by the number of direct stress components present at that material point of the nite element model (for example, a maximum of three cracks in three-dimensional, axisymmetric, and plane strain problems or a maximum of two cracks in plane stress problems). Once cracks exist at
4.5.32
CRACKING MODEL
a point, the component forms of all vector and tensor valued quantities are rotated so that they lie in the local system dened by the crack orientation vectors (the normals to the crack faces). The model ensures that these crack face normal vectors are orthogonal so that this local system is rectangular Cartesian. Crack closing and reopening can take place along the directions of the crack surface normals. The model neglects any permanent strain associated with cracking; that is, we assume that the cracks can close completely when the stress across them becomes compressive.
Elastic-cracking model for concrete
The main ingredients of the model are a strain rate decomposition into elastic (concrete) and cracking strain rates, elasticity, a set of cracking conditions, and a cracking relation (the evolution law for the cracking behavior). The main advantage of the strain decomposition is that it allows the eventual addition of other effects, such as plasticity and creep, in a consistent manner. The elastic-cracking strain decomposition also allows the separate identication of a cracking strain that represents the state of a crack; this contrasts with the classical smeared cracking models where a single strain quantity is used to represent the state of a cracked solid in a homogenized form leading to a modied (damaged) elasticity formulation.
Strain rate decomposition
" " " d" = d"el + d"ck ; (4.5.31) " " where d" is the total mechanical strain rate, d"el is the elastic strain rate representing the uncracked " concrete (the continuum between the cracks), and d"ck is the cracking strain rate associated with any
existing cracks.
Crack direction transformations
Constitutive Theories
The strains in Equation 4.5.31 are referred to the global Cartesian coordinate system and can be written in vector form (in a three-dimensional setting) as
" = ["11
"22
"33
12
13
23 ]T :
For incorporating the cracking relations it is convenient to dene a local Cartesian coordinate system n; t; s that is aligned with the crack directions. In the local system, shown in Figure 4.5.31, the strains are e = [enn ett ess gnt gns gts]T : The transformation between global and local strains is written in matrix form as where T is a transformation matrix constructed from the direction cosines of the local cracking coordinate system. T is constant in our xed crack model.
" = T e;
(4.5.32)
4.5.33
CRACKING MODEL
t n
Figure 4.5.31 Global and local cracking coordinate systems. The conjugate stress quantities can be written in the global coordinate system as = [ ]T
11 22 33 12 13 23
= [tnn
= TT :
(4.5.33)
Elasticity
The intact continuum between the cracks is modeled with isotropic, linear elasticity. The orthotropic nature of the cracked material is introduced in the cracking component of the model. As stated earlier, the approach of decomposing the strains into elastic, intact concrete, strains, and cracking strains has the advantage that this smeared model can be generalized to include other effects such as plasticity and creep (although such generalizations are not yet included in ABAQUS/Explicit).
Crack detection
A simple Rankine criterion is used to detect crack initiation. This states that a crack forms when the maximum principal tensile stress exceeds the tensile strength of the brittle material. The Rankine
4.5.34
CRACKING MODEL
crack detection surface is shown in Figure 4.5.32 in the deviatoric plane, in Figure 4.5.33 in the meridional plane, and in Figure 4.5.34 in plane stress. Although crack detection is based purely on Mode I fracture considerations, ensuing cracked behavior includes both Mode I (tension softening) and Mode II (shear softening/retention) behavior, as described later.
S1 o =0 = 60
o
S2
S3
Constitutive Theories
q = 3/2 S:S 3
= 60 (compression) = 0 (tension)
o
1.5 3 It 1.5 t t
I I
p = -1/3 trace ()
4.5.35
CRACKING MODEL
2 / t
1 1 1 /
I t
Figure 4.5.34 Rankine criterion in plane stress. As soon as the Rankine criterion for crack formation has been met, we assume that a rst crack has formed. The crack surface is taken to be normal to the direction of the maximum tensile principal stress. Subsequent cracks can form with crack surface normals in the direction of maximum principal tensile stress that is orthogonal to the directions of any existing crack surface normals at the same point. The crack orientations are stored for subsequent calculations, which are done for convenience in a local coordinate system oriented in the crack directions. Cracking is irrecoverable in the sense that, once a crack has occurred at a point, it remains throughout the rest of the calculation. However, a crack may subsequently close and reopen.
Cracking conditions
We introduce a consistency condition for cracking (analogous to the yield condition in classical plasticity) written in the crack direction coordinate system in the form of the tensor
C = C(t; I ;I I ) = 0;
where
(4.5.34)
and I ;I I represents a tension softening model (Mode I fracture) in the case of the direct components of stress and a shear softening/retention model (Mode II fracture) in the case of the shear components
4.5.36
CRACKING MODEL
of stress. The matrices @ =@ and @ =@ I ;I I are assumed to be diagonal, implying the usual assumption that there is no coupling between cracks in the cracking conditions. Each cracking condition is more complex than a classical yield condition in the sense that two cracking states are possible (an actively opening crack state and a closing/reopening crack state), contrasting with a single plastic state in classical plasticity. This can be illustrated by writing the cracking conditions for a particular crack normal direction n explicitly:
I I Cnn = Cnn(tnn ; t ) = tnn 0 t (eck ) = 0 (4.5.35) nn I ck for an actively opening crack, where t (enn ) is the tension softening evolution (dened by the user),
C t
and
Cnn = Cnn(tnn;
I ) c
tnn 0 e
e
I ck ( nn ) c
open
=0
(4.5.36)
nn
open
nn
eopen = nn
over history
max
eck : nn
These conditions are illustrated in Figure 4.5.35 and represent the tension softening model adopted for the cracking behavior normal to crack surfaces. Similar conditions can be written for the other two possible crack normal directions, s and t. It must be emphasized that, although the cracking condition of Equation 4.5.34 has been written for the most general case of all possible cracks existing, only the components of that refer to existing cracks are considered in the computations with this model.
t nn
Constitutive Theories
It
D nn
e nn cI e nn
open
ck
4.5.37
CRACKING MODEL
The cracking conditions for the shear components in the crack coordinate system are activated when the associated normal directions are cracked. We now present the shear cracking conditions by writing the conditions for shear component nt explicitly. The crack opening dependent shear model (shear retention model) is written as
II II ck Cnt = Cnt(tnt ; s ) = tnt 0 s (gnt ; eck ; eck ) = 0 nn tt
(4.5.37)
II ck for shear loading or unloading of the crack, where s (gnt ; eck ; eck ) is the shear evolution that nn tt depends linearly on the shear strain and also depends on the crack opening strain (this dependency being dened by the user). Figure 4.5.36 illustrates the model. Although this model is inspired by the traditional shear retention models, it differs from those models in one important aspect: the shear stress tends to zero as the crack develops. This is discussed in more detail later.
t nt
D nt
II
g nt s
II
ck
e ck , e ck nn tt
Figure 4.5.36
Cracking relation
The relation between the local stresses and the cracking strains at the crack interfaces is written in rate form as
is a diagonal cracking matrix that depends on the state of the existing cracks. The where I I I II II II denition of these diagonal components (Dnn; Dtt; Dss; Dnt ; Dns; Dts ) is given in Figure 4.5.35 and Figure 4.5.36.
dt = Dck deck ;
ck
(4.5.38)
4.5.38
CRACKING MODEL
Using the strain rate decomposition (Equation 4.5.33) and the elasticity relations, we can write the rate of stress as
is the isotropic linear elasticity matrix. where Premultiplying Equation 4.5.39 by T and substituting Equation 4.5.35 and Equation 4.5.38 into the resulting left-hand side yields
el
(4.5.39)
deck =
h
(4.5.310)
Finally, substituting Equation 4.5.310 into Equation 4.5.39 results in the stress-strain rate equations:
d =
(4.5.311)
The brittle fracture concept of Hilleborg (1976) forms the basis of the postcracked behavior in the direction normal to the crack surface (commonly referred to as tension softening). We assume that the fracture energy required to form a unit area of crack surface in Mode I, GI , is a material property. f This value can be calculated from measuring the tensile stress as a function of the crack opening displacement (Figure 4.5.37), as Z
GI f
I t dun :
Constitutive Theories
Gf
u ck n un
u el n
un
4.5.39
CRACKING MODEL
Typical values of GI range from 40 N/m (0.22 lb/in) for a typical construction concrete (with f a compressive strength of approximately 20 MPa, 2850 lb/in2 ) to 120 N/m (0.67 lb/in) for a high strength concrete (with a compressive strength of approximately 40 MPa, 5700 lb/in2 ). The implication of assuming that GI is a material property is that, when the elastic part of the f displacement, uel , is eliminated, the relationship between the stress and the remaining part of the n displacement, uck = un 0 uel , is xed, regardless of the specimen size. For example, consider a n n specimen developing a single crack across its section as tensile displacement is applied to it: uck is n the displacement across the crack and is not changed by using a longer or shorter specimen in the test (so long as the specimen is signicantly longer than the width of the crack band, which will typically be of the order of the aggregate size). Thus, this important part of the cracked concretes tensile behavior is dened in terms of a stress/displacement relationship. In the nite element implementation of this model we must, therefore, compute the relative displacement at a material point to provide uck . We do this in ABAQUS by multiplying the strain by n a characteristic length associated with the material point (the cracking strain in local crack direction n is used as an example):
uck = eck h; n nn
where h is the characteristic length. This characteristic crack length is based on the element geometry: for beams and trusses we use the length associated with the material calculation point; for shell and planar elements we use the square root of the area associated with the material calculation point; for solid elements we use the cube root of the volume associated with the material calculation point. This denition of the characteristic length is used because we do not necessarily know in which direction the concrete will crack; and, hence, we cannot choose the length measure a priori in any particular direction. These characteristic length estimates are only appropriate for well-shaped elements (elements that do not have large aspect ratios). This should be considered by the user in dening values for the material properties. For reinforced concrete, since ABAQUS provides no direct modeling of the bond between rebar and concrete, the effect of this bond on the concrete cracks must be smeared into the plain concrete part of the model. This is generally done by increasing the value of GI based on comparisons with f experiments on reinforced material. This increased ductility is commonly refered to as the tension stiffening effect. In reinforced concrete applications the softening behavior of the concrete tends to have less inuence on the overall response of the structure because of the stabilizing presence of the rebar. I Therefore, it is often appropriate to dene tension stiffening as a t eck relationship directly. This nn option is also offered in ABAQUS.
Cracked shear models
An important feature of the cracking model is that, whereas crack initiation is based on Mode I fracture only, postcracked behavior includes Mode II as well as Mode I. The Mode II shear behavior is described next. The Mode II model is based on the common observation that the shear behavior is dependent on the amount of crack opening. Therefore, ABAQUS offers a shear retention model in which the
4.5.310
CRACKING MODEL
postcracked shear stiffness is dependent on crack opening. This model denes the total shear stress as a function of the total shear strain (shear direction nt is used as an example):
II ck tnt = Dnt (eck ; eck ) gnt ; nn tt
where
II II Dnt (eck ; eck ) is a stiffness that depends on crack opening. Dnt nn tt II Dnt = (eck ; eck ) G; nn tt
where G is the shear modulus of the uncracked concrete and (eck ; eck ) is a user-dened dependence nn tt of the form shown in Figure 4.5.38.
tends to infinity
e max
ck
e nn
ck
Figure 4.5.38 Shear retention factor dependence on crack opening. A commonly used mathematical form for this dependence when there is only one crack, associated with direction n, is the power law proposed by Rots and Blaauwendraad (1989):
nn 0 eeck ck max p ; (enn) = eck nn 1 0 1 0 eck max
1
Constitutive Theories
ck
p
(4.5.313)
where
ck nn
and eck are material parameters. This form satises the requirements that as max e 0 (corresponding to the state before crack initiation) and 0 as eck eck (corresponding nn max to complete loss of aggregate interlock). Note that the bounds of , as dened in our model using and zero. This contrasts with some of the traditional the elastic-cracking strain decomposition, are shear retention models where the intact concrete and cracking strains are not separated; the shear retention in these models is dened using a shear retention factor, , which can have values between one and zero. The relationship between these two shear retention parameters is
!1
4.5.311
CRACKING MODEL
=
( + 1) :
(4.5.314)
The shear retention power law form given in Equation 4.5.313 can then be written in terms of
( e
ck nn
)=
1 0 ee
ck nn
as
ck max
Since users are more accustomed to specifying shear retention factors in the traditional way (with values between one and zero), the ABAQUS input requests eck data. Using Equation 4.5.314, nn these data are then converted to eck data for computation purposes. nn When the shear component under consideration is associated with only one open crack direction (n or t), the crack opening dependence is obtained directly from Figure 4.5.38. However, when the shear direction is associated with two open crack directions (n and t), then
g
with
ck nt
=g
ck;n nt
+g
ck;t nt
t D D
nt
I I ;n
nt
t D
nt
I I ;t
nt
I I ;n nt
= (e ) G;
ck nn
I I ;t nt
= (e ) G;
ck tt I I ;t I I ;t nt nt
and, therefore,
II nt
t =g =
nt ck nt
D D
I I ;n nt
I I ;n
nt
D +D
This total stress-strain shear retention model differs from the traditional shear retention models in which the stress-strain relations are written in incremental form (again, shear direction nt is used as an example):
II where Dnt eck ; eck is an incremental stiffness that depends on crack opening. The difference nn tt between the total model used in ABAQUS (Equation 4.5.312) and the traditional incremental model (Equation 4.5.315) is best illustrated by considering the shear response of the two models in the case when a crack is simultaneously opening and shearing. This is shown in Figure 4.5.39 for the total model and in Figure 4.5.310 for the incremental model. It is apparent that, in the total model, the shear stress tends to zero as the crack opens and shears; whereas, in the incremental model the shear stress tends to a nite value. This may explain why overly stiff responses are usually obtained with the traditional shear retention models.
1 t = D (e
nt II nt
ck nn
;e
ck tt
) 1g
ck nt
(4.5.315)
Reference
Cracking model for concrete, Section 11.5.2 of the ABAQUS Analysis Users Manual
4.5.312
CRACKING MODEL
t nt
D nt
II
( t nt )i = D nt ( e ck , e ck ) 1 nn tt
II
(e , e )
ck nn
ck tt i
( g nt )i
ck
D nt
II
( e ck , e ck ) i nn tt
( g nt )1
ck
( g nt )i
ck
g nt
ck
( t nt )i = D nt
II
(e , e )
ck nn
ck tt i
( g nt )i
ck
Constitutive Theories
D nt D nt
II
II
( e ck , e ck )i nn tt
( e ck , e ck )1 nn tt g nt
ck
Figure 4.5.310
4.5.313
JOINTED MATERIAL
4.5.4
The jointed material model is intended to provide a simple, continuum model for materials containing a high density of parallel joint surfaces in different orientations. The spacing of the joints of a particular orientation is assumed to be sufciently close compared to characteristic dimensions in the domain of the model that the joints can be smeared into a continuum of slip systems. An obvious application is the modeling of geotechnical problems where the medium of interest is composed of signicantly faulted rock. In this context, models similar to the one described next have been proposed in the past; see, for example, the model formulated by Zienkiewicz and Pande (1977). The model implemented in ABAQUS/Standard provides for opening of the joints, or frictional sliding of the joints, in each of these systems (a system in this context is a joint orientation in a particular direction at a material calculation point). In addition to the joint systems, the model includes a bulk material failure mechanism. This is based on the Drucker-Prager failure criterion.
Joint system denitions
We consider a particular joint a oriented by the normal to the joint surface na . We dene ta ; = 1; 2 as two unit, orthogonal vectors in the joint surface. The local stress components are the pressure stress across the joint
pa def na 1 1 na; =
a = na 1 1 ta;
where
Constitutive Theories
a = aa :
p
The local strain components are the normal strain across the joint
a = na 1 " 1 ta + ta 1 " 1 na ;
where
4.5.41
JOINTED MATERIAL
" " " where d" is the total strain rate, d"el is the elastic strain rate, and d"pl is the inelastic (plastic) strain rate. Supposing that several systems are active (we designate an active system by i, where i = b indicates the bulk material system and i = a is a joint system a), we write " d"pl
Elasticity and joint opening/closing
=
(4.5.41)
X
i
"i d"pl :
(4.5.42)
When all joints at a point are closed, the elastic behavior of the material is assumed to be isotropic and linear. The material cannot be elastically incompressible (Poissons ratio must be less than 0.5). We use a stress-based joint opening criterion whereas joint closing is monitored based on strain. Joint system a opens when the estimated pressure stress across the joint (normal to the joint surface) is no longer positive:
pa 0:
In this case the material is assumed to have no elastic stiffness with respect to direct strain across the joint system. Open joints, thus, create anisotropic elastic response at a point. The joint system remains open as long as where "el is the component of direct elastic strain across the joint and "el (ps) is the component of an an direct elastic strain across the joint calculated in plane stress as
"el (ps) = 0 an
( + a 2 ); E a1
where E is the Youngs modulus of the material, is the Poissons ratio, and
a = ta 1 1 ta;
are the direct stresses in the plane of the joint. The shear response of open joints is governed by the shear retention parameter, fsr , which represents the fraction of the elastic shear modulus retained when the joints are open (fsr =0 means no shear stiffness associated with open joints, while fsr =1 corresponds to elastic shear stiffness in open joints; any value between these two extremes can be used).
Plastic behavior of joint systems
fa
a 0 pa tan a 0 da = 0;
(4.5.43)
where a is the friction angle for system a, and da is the cohesion for system a (see Figure 4.5.41).
4.5.42
JOINTED MATERIAL
dpl a
a a
fa
da
pa
Figure 4.5.41 Joint system material model. As long as fa < 0, joint system a does not slip. When fa (plastic) strain on the system is then given by
= 0
joint system
a slips.
The inelastic
pl where d
a is the rate of inelastic shear strain in direction on the joint surface, d"pl is the magnitude a of the inelastic strain rate, a is the dilation angle for this joint system (choosing a = 0 provides pure shear ow on the joint, while a > 0 causes dilation of the joint as it slips), and d"pl is the an inelastic strain normal to the joint surface. In order to add the plastic ow contributions from different systems we write the tensorial plastic strain rate for joint a as pl "a d"pl = d"pl na na + d
a (na ta + tana ): an
(4.5.44)
Constitutive Theories
The sliding of the different joint systems at a point is independent, in the sense that sliding on one system does not change the failure criterion or the dilation angle for any other joint system at the same point. The model provides for up to three joint systems at a point.
Plastic behavior of bulk material
In addition to the joint systems, the model includes a bulk material failure mechanism. This is based on the Drucker-Prager failure criterion,
4.5.43
JOINTED MATERIAL
where q
def
=
q3
q 0 p tan b 0 db = 0;
is the Mises equivalent deviatoric stress (here
=
(4.5.45)
: is the equivalent pressure stress, b is the friction angle for the bulk material, and db is the cohesion for the bulk material (see Figure 4.5.42). =
def
+ pI), p
2S : S def
01I 3
dpl b q b fb
db p
Figure 4.5.42 Bulk material model. If this failure criterion is reached, the bulk inelastic ow is dened by
"b d" pl
where
d"pl b
0 1 tan 3
@gb ; b @
(4.5.46)
gb = q 0 p tan
(4.5.47)
is the ow potential. Here, d"pl is the magnitude of the inelastic ow rate (chosen so that b d"pl = j(d"pl )11j in uniaxial compression in the 1-direction) and b is the dilation angle for the b b bulk material. This bulk failure model is a simplied version of the extended Drucker-Prager model described in Models for granular or polymer behavior, Section 4.4.2. As with the joint systems, this bulk failure system is independent of the joint systems, in that bulk inelastic ow does not change the behavior of any joint system.
4.5.44
JOINTED MATERIAL
If 6= in any system the ow in that system is nonassociated. This implies that the material stiffness matrix is not symmetric, so that the unsymmetric matrix solution scheme should be invoked by the user. If the difference between and is not large, a symmetric approximation to the matrix can provide an acceptable rate of convergence of the equilibrium equations, and hence a lower overall solution cost. For this reason the unsymmetric solver is not automatically invoked for this material behavior. However, it is recommended for all cases where and are very different on any joint system.
Integration of the model
The constitutive equations described above are integrated using the backward Euler method generally used with the plasticity models in ABAQUS. A material Jacobian consistent with this integration operator is used for the overall equilibrium iterations.
Reference
Jointed material model, Section 11.4.1 of the ABAQUS Analysis Users Manual
Constitutive Theories
4.5.45
HYPERELASTICITY
4.6.1
The constitutive behavior of a hyperelastic material is dened as a total stresstotal strain relationship, rather than as the rate formulation that has been discussed in the context of history-dependent materials in previous sections of this chapter. Therefore, the basic development of the formulation for hyperelasticity is somewhat different. Furthermore, hyperelastic materials are often incompressible or very nearly so; hence, mixed (hybrid) formulations can be used effectively. In this section the hyperelastic model provided in ABAQUS is dened, and the mixed variational principles used in ABAQUS/Standard to treat the fully incompressible case and the almost incompressible case are introduced.
Denitions and basic kinematic results
We rst introduce some denitions and basic kinematic results that will be used in this section. Some of these items have already been discussed in Chapter 1, Introduction and Basic Equations: they are repeated here for convenience. Writing the current position of a material point as x and the reference position of the same point as , the deformation gradient is
F def X = x
@ @
def
= det(F)
J
F def =
01 F 3
Constitutive Theories
as the deformation gradient with the volume change eliminated. We then introduce the deviatoric stretch matrix (the left Cauchy-Green strain tensor) of
F as
B def F FT =
1
def
= trace
where
B=I:B
= 1 2
(4.6.11)
def 1
= 2
I1
0 trace (B 1 B)
I1
0I :B1B
:
(4.6.12)
The variations of , 1 , I 1, I 2 , and J will be required during the remainder of the development. We rst dene some variations of basic kinematic quantities that will be needed to write these results.
BB B
4.6.11
HYPERELASTICITY
The gradient of the displacement variation with respect to current position is written as
L def @x : = @u
0
which we decompose into the virtual rate of change of volume per current volume (the virtual volumetric strain rate),
= I : D;
e def D 0 1 "vol I: = 3
0 1
def asym( L) = 1 L 0 LT : = 2
B = e 1 B + B 1 e +
1 B 0 B 1
= H 1 : e +
1 B 0 B 1
;
where
where
0 1 (H2 )ijkl def 2 ik B jp B pl + B ip B pk jl + il B jp B pk + B ip B pl jk + B ik B jl + B il B jk ; = 1
I 1 I 2
= 2 B : e;
(4.6.13) (4.6.14)
= 2(I 1 B 0 B 1 B) : e;
and
4.6.12
HYPERELASTICITY
J
= J "vol :
Z
(4.6.15)
The Cauchy (true) stress components are dened from the strain energy potential as follows. From the virtual work principal the internal energy variation is
WI
:
D dV =
V0
J :
D dV 0;
where are the components of the Cauchy (true) stress, V is the current volume, and V 0 is the reference volume. We decompose the stress into the equivalent pressure stress,
p =
def
1 0 3 I : ;
S def + p I; =
Z
WI =
V0
S : e 0 p "vol ) dV 0:
so that
U =
@U @I 1
I 1 +
@U @I 2
I 2 +
@U @J
J:
Constitutive Theories
@U Z
@I 1
+ I1
@U @I 2
B 0 @I B 1 B
@U
2 vol 0
:
e + J @U "vol : @J
0
(4.6.16)
Since the variation of the strain energy potential is, by denition, the internal virtual work per reference volume, WI , we have
WI =
V0
J(
S : e 0 p "
@U @I 1
) dV
V0
U dV :
(4.6.17)
For a compressible material the strain variations are arbitrary, so this equation denes the stress components for such a material as
S= J
and
DEV
+ I1
@U @I 2
B 0 @I B 1 B
@U
2
(4.6.18)
4.6.13
HYPERELASTICITY
= 0 @U : @J
(4.6.19)
When the material response is almost incompressible, the pure displacement formulation, in which the strain invariants are computed from the kinematic variables of the nite element model, can behave poorly. One difculty is that from a numerical point of view the stiffness matrix is almost singular because the effective bulk modulus of the material is so large compared to its effective shear modulus, thus causing difculties with the solution of the discretized equilibrium equations. Similarly, in ABAQUS/Explicit the high bulk modulus increases the dilatational wave speed, thus reducing the stable time increment substantially. Another problem is that, unless reduced-integration techniques are used, the stresses calculated at the numerical integration points show large oscillations in the pressure stress values, becausein generalthe elements cannot respond accurately and still have small volume changes at all numerical integration points. To avoid such problems, ABAQUS/Standard ^ offers a mixed formulation for such cases. The concept is to introduce a variable, J , that is used in place of the volume change, J , in the denition of the strain energy potential. The internal energy ^ integral, WI , is augmented with the constraint that J 0 J = 0, imposed by the use of a Lagrange ^ multiplier, p, and integrated over the volume:
(W I )
A def =
Z
V0
^ U (I 1 ; I 2 ; J )
^ 0 p (J 0 J ) ^
dV :
A (W I ) =
Z
V0
JS : +
@U
^ @J
+p ^ ^ J
0 J p " 0 (J 0 ^
vol
0 ^ ^ J ) p dV :
p= ^
0 @U ; ^
@J @J
^ ^ ^ 0Q : e + @ p J ; ^
where
@S ^ = Q def J @J^:
Z
V0
S + (J 0 Q
^ J) ^
:
e 0 J p " 0 (J 0 ^
vol
^ J)
@p ^
^ @J
^ J dV 0:
(4.6.110)
This augmented formulation can be used for any value of compressibility except fully ^ ^ incompressible behavior. J is interpolated independently in each element: ABAQUS uses constant J ^ in rst-order elements and linear variation of J with respect to position in second-order elements, which ^ implies that J is discontinuous between elements: continuity in such variables is not a requirement.
4.6.14
HYPERELASTICITY
When the material is fully incompressible, U is a function of the rst and second strain invariantsI 1 and I 2only, and we write the internal energy in the augmented form,
(WI )A def =
Z
where p is again a Lagrange multiplier introduced to impose the constraint J 0 1 = 0 in such a way ^ that the variation of (WI )A can be taken with respect to all kinematic variables, thus giving
(W I ) A =
V0
0 p(J 0 1) ^
dV 0;
p is interpolated in the same way as J is interpolated in the augmented formulation for almost incompressible behavior; that is, p is assumed to be constant in rst-order elements and to vary
V0
JS
^ : e 0 J p " 0 (J 0 1) p ^
vol
dV 0:
(4.6.111)
The rate of change of the internal virtual work is required for use in the Newton method, which is generally used in ABAQUS/Standard to solve the nonlinear equilibrium equations (after discretization by nite elements). It will also be used when we extend the elasticity model to viscoelastic behavior for small (linearized) vibrations about a predeformed state. When the pure displacement formulation is used for the compressible case, the deviatoric stress components, , are dened by Equation 4.6.18, from which we can show that
d J
( S) = J (CS : de + Q d"vol + d
S 0 S 1 d
);
1
CS def 2 =
J
@U @I 1 J
+ I1
@2U
@U @I 2
CS , is dened as
@U @I 2
2 4 H1 0 J @U H2 + J
@2U @I 2
@2U @I 1
04
2 2 2 0 34 @U + 2I 1 @U + I 1 @ U2 + I 12 + 2I 2 @ U + 2I 1 I 2 @ U2 J @I 1 @I 2 @I 1 @I 2 @I 1 @I 2 ! ! 2U 2U + 2I 2 @ 2 I B 1 B + B 1 B I ; + 34 2 @U + I 1 @ J @I 2 @I 1@I 2 @I 2
"
@I 1@I 2
+ I1
@I 2
+
! !
+ 2I 1
2U
@2U @I 1 @I 2
4 B 1BB+ BB1B + J @
2 @2U I1 2 @I 2
Constitutive Theories
BB
!
@I 2
B 1BB1B
#
IB+ BI
Q = (@JS) = 2
def @ J
@2U @I 1 @J
+ I1
@2U @I 2 @J
Q, is
I1 @2U @I 1 @J
B 0 2 @I @J B 1 B 0 2 3 2
@2U
+ 2I 2
@2U @I 2@J
I:
4.6.15
HYPERELASTICITY
( ) = 0J (Q : de + Kd"vol );
=0
@p @J
+ p = J @ U + @U : @J 2 @J
Thus,
dWI
Z b e
V
"vol
(4.6.112)
since
" " e : (d
1 S 0 S 1 d
) + S : de 0 pd"vol = 0 : (2" 1 d" 0 LT 1 dL):
For the mixed formulation introduced for almost incompressible materials, the rate of change of the augmented variation of internal energy, Equation 4.6.110, is
2 ~S ^ ^ Q A 3 8 dvol 9 C < e = A= vol J c 4 Q ^ ^ d (WI ) b e " 0p 0@ p=@ J 5 : d" ; ^ ^ ^ V ^ ~ A 0@p=@J^ 0K dJ^ ^ 1 ^^ " " 0 + J (J 0 J )Q : (2" 1 d" 0 LT 1 dL) dV; Z
S ~ = CS def CS + (J 0 J^) @@C^ ; J ^ = Q def J @S ;
where
^ ^ = 1 A def J (J 0 J^) @Q ; ^ @J
and
@J
^ 1 ^ ~ ^ @K K = 2 K 0 (J 0 J ) ^ J
def
!
;
@J
in which
^ ^ def 0J @ p : = ^
@J
4.6.16
HYPERELASTICITY
For the case of incompressible materials the rate of change of the augmented variation of internal energy is similarly obtained from Equation 4.6.111 as
d (WI )
A=
Z
b e
"vol
2 S C p c 4 0 ^
0
9 38 < de = 0p 01 5 : d"vol ; ^ 0 0 1 dp ^
0 0
dV:
Several particular forms of the strain energy potential are available in ABAQUS. The incompressible or almost incompressible models make up: the polynomial form and its particular casesthe reduced polynomial form, the neo-Hookean form, the Mooney-Rivlin form, and the Yeoh form; the Ogden form; the Arruda-Boyce form; and the Van der Waals form. In addition, a hyperelastic model for highly compressible, elastic materials is offered.
Polynomial form and particular cases
Given isotropy and additive decomposition of the deviatoric and volumetric strain energy contributions in the presence of incompressible or almost incompressible behavior, we can write the potential, in general, as Setting g
=
PN
Constitutive Theories
U =
(4.6.113)
This form is the polynomial representation of the strain energy in ABAQUS. The parameter N can take values up to six; however, values of N greater than 2 are rarely used when both the rst and second invariants are taken into account. Cij and Di are temperature-dependent material parameters. The value of N and tables giving the Cij and Di values as functions of temperature are specied by the user. The elastic volume strain, Je` , follows from the total volume strain, J , and the thermal volume strain, Jth , with the relation
Je` = J
and Jth follows from the linear thermal expansion, "th , with
Jth = (1 + "th )3;
Jth
4.6.17
HYPERELASTICITY
where "th follows from the temperature and the isotropic thermal expansion coefcient dened by the user. The Di values determine the compressibility of the material: if all the Di are zero, the material is taken as fully incompressible. If D1 = 0, all Di must be zero. Regardless of the value of N , the initial shear modulus, 0 ; and the bulk modulus, k0; depend only on the polynomial coefcients of order N = 1:
0 = 2(C10 + C01);
k0 =
D1
If N = 1, so that only the linear terms in the deviatoric strain energy are retained, the MooneyRivlin form is recovered:
C10(I 1 0 3) + C01(I 2 0 3) +
2 (J 0 1) : D1 e`
1
The Mooney-Rivlin form can be viewed as an extension of the neo-Hookean form (discussed below) in that it adds a term that depends on the second invariant of the left Cauchy-Green tensor. In some cases this form will give a more accurate t to the experimental data than the neo-Hookean form; in general, however, both models give similar accuracy since they use only linear functions of the invariants. These functions do not allow representation of the upturn at higher strain levels in the stress-strain curve. Particular forms of the polynomial model can also be obtained by setting specic coefcients to 6 zero. If all Cij with j = 0 are set to zero, the reduced polynomial form is obtained:
N X i=1
Ci0(I 1 0 3)i +
N X i=1
2i (J 0 1) : Di e`
1
Following Yeoh (1993) the justication for reducing the general polynomial series expansion by omitting the dependence on the second invariant arises from the following observations. The sensitivity of the strain energy function to changes in the second invariant is generally much smaller than the sensitivity to changes in the rst invariant. In addition, the I 2 -dependence is difcult to measure, so it might be preferable to neglect it rather than to calculate it based on potentially inaccurate measurements. Finally, it appears that omitting the dependence on the second invariant if only data for a particular mode of deformation are known might enhance the prediction for other deformation states. This conjecture is supported by investigating the so-called reduced stresses in the presence of almost incompressible behavior:
1 0 2 2 0 2 1 2 1 0 3 2 0 2 1 3 2 0 3 2 0 2 2 3
@U @I 1 @U =2 @I 1 @U =2 @I 1
=2
@U ; @I 2 2 @U + 2 ; @I 2 2 @U + 1 ; @I 2
+
2 3
4.6.18
HYPERELASTICITY
where the i , i = 1 . . . 3 represent the principal Cauchy (true) stresses. If the derivatives with respect to I 2 are omitted and different stress statesuniaxial, biaxial, and planarare considered, the reduced stresses have the same form regardless of the stress state. Measurements of the I 2 -dependence of carbon-black reinforced rubber vulcanizates conrming these ndings can be found in Kawabata, Yamashita, et al. (1995). The paper of Kaliske and Rothert (1997) also supports the notion that often the prediction of general deformation states based on a uniaxial measurement can be enhanced only by ignoring the I 2 -dependence. In this context it is worth noting that the mathematical structure of the Arruda-Boyce model can be viewed as a fth-order reduced polynomial, where the ve coefcients C10 . . . C50 are implicit nonlinear functions of the two parameters and m in the Arruda-Boyce form. However, the ArrudaBoyce model offers a physical interpretation of the parameters, which the general fth-order reduced polynomial fails to provide. The Yeoh form (Yeoh, 1993) can be viewed as a special case of the reduced polynomial with N = 3:
U=
XC
3
Typically, if C10 = O(1), the second coefcient will be negative and one to two orders of magnitude smaller [i.e., C20 is 0O(0:1) to 0O(0:01)], while the third coefcient C30 is again one to two orders of magnitude smaller but positive [i.e., C30 is +O(1:E 0 2) to +O(1:E 0 4)]. These magnitudes will create the typical S-shape of the stress-strain behavior of rubber; at low strains C10 represents the initial shear modulus, which softens at moderate strains due to the effect of the negative second coefcient C20 and is followed by an upturn at large strains due to the positive third coefcient C30. Thus, the Yeoh model often provides an accurate t over a large strain range. If the reduced-polynomial strain-energy function is simplied further by setting N = 1, the neoHookean form is obtained:
i=1
i0(I 1 0 3)
i+
X
3
2i (J 0 1) : Di e` i=1
1
C10(I 1 0 3) +
2 (J 0 1) : D1 e`
1
Constitutive Theories
This form is the simplest hyperelastic model and often serves as a prototype for elastomeric materials in the absence of accurate material data. It also has some theoretical relevance since the mathematical representation is analogous to that of an ideal gas: the neo-Hookean potential represents the Helmholtz free energy of a molecular network with Gaussian chain-length distribution (see Treloar, 1975). The user can request that ABAQUS calculate the Cij and Di values from measurements of nominal stress and strain in simple experiments. The basis of this calculation is described in Fitting of hyperelastic and hyperfoam constants, Section 4.6.2.
Ogden form
The Ogden strain energy potential is expressed in terms of the principal stretches. In ABAQUS the following formulation is used:
U def =
X
N
2
i=1
2 ( 1 +
2 i + 3 i 0 3) +
X
N
2i (J 0 1) ; Di e` i=1
1
(4.6.114)
4.6.19
HYPERELASTICITY
where
Hence, the rst part of Ogdens strain energy function depends only on I 1 and I 2 . Ogdens energy function cannot be written explicitly in terms of I 1 and I 2. It is, however, possible to obtain closedform expressions for the derivatives of U with respect to I 1 and I 2 . The value of N and tables giving the i and i values as functions of temperature are specied by the user. If N = 2, 1 = 2, and 2 = 02, the Mooney-Rivlin model is obtained. If N = 1 and 1 = 2, Ogdens model degenerates to the neo-Hookean material model. In the Ogden form the initial shear modulus, 0 , depends on all coefcients:
i = J 0 3 i ! 1 2 3 = 1 :
1
0 =
N X i ; i=1
and the initial bulk modulus, k0, depends on D1 as before. The user can request that ABAQUS calculate the i and i values from measurements of nominal stress and strain.
Arruda-Boyce form
X U =
5
where
C1 = ;
1 2
C2 =
1; 20
C3 =
11 ; 1050
C4 =
19 ; 7000
and
C5 =
519 : 673750
The deviatoric part of the strain energy density comes from Arruda and Boyce (1993). This model is also known as the eight-chain model, since it was developed starting out from a representative volume element where eight springs emanate from the center of a cube to its corners. The values of the coefcients C1 . . . C5 arise from a series expansion of the inverse Langevin function, which arises in the statistical treatment of non-Gaussian chains. The series expansion is truncated after the fth term. The coefcient represents the initial shear modulus, and the coefcient m is referred to as the locking stretch. Approximately at this stretch the slope of the stress-strain curve will rise signicantly. The initial bulk modulus is obtained as K0 = 2=D. To the deviatoric part of the strain energy density we add a simplied representation of the volumetric strain energy density, which requires only one material parameter, so that all material parameters can be estimated easily even with limited knowledge of the material behavior. This volumetric representation has been used successfully by Kaliske and Rothert (1997) and provides sufcient accuracy for most engineering elastomeric materials. The Arruda-Boyce potential depends on the rst invariant only. The physical interpretation is that the eight chains are stretched equally under the action of a general deformation state. The rst invariant, I 1 = 2 + 2 + 2 , directly represents this elongation. 1 2 3
4.6.110
HYPERELASTICITY
When the Arruda-Boyce form is chosen, the user can specify the coefcients as functions of temperature; alternatively, ABAQUS can perform a t of the test data specied by the user to determine the coefcients.
Van der Waals form
The hyperelastic Van der Waals potential, also known as the Kilian model, has the following form:
U
where
= 0(m 0 3)
2
ln(1
0 ) + 0 3 a
2
~ I03 2
!3 2 s
J2 0 1 e` 0 ln Je` ; + D 2
1 ~ I 03 : 2 0 3 m
~ I = (1 0 )I 1 + I 2 and =
The name Van der Waals draws on the analogy in the thermodynamic interpretation of the equations of state for rubber and gas. While the neo-Hookean model can be compared with an ideal gas in that it starts out from a Gaussian network with no mutual interaction between the quasi-particles (Kilian, 1981), the Van der Waals strain energy potential is analogous to the equations of state of a real gas. This introduces two additional material parameters: the locking stretch, m , and the global interaction parameter, a. (Similarly, the Van der Waals equation for a real gas introduces two parameters to account for excluded volume and modied exchange of momentum between the particles.) The locking stretch, m , accounts for nite extendability of the non-Gaussian chain network. In contrast to the Arruda-Boyce model the mathematical structure of the Van der Waals potential is such that the strain energy tends to innity as the locking stretch, m , is reached; more precisely, as ~ I 2 . Thus, the Van der Waals potential cannot be used at stretches larger than the locking stretch. m The global interaction parameter, a, models the interaction between the chains; it is difcult to estimate. Kilian et al. (1986) point out that, given Mooney-Rivlin coefcients and a locking stretch m , a suitable value for the global interaction parameter is
Constitutive Theories
a=
2C01 3
3 m 0 1
2 m
where is the initial shear modulus at low strains and C01 is the second Mooney-Rivlin parameter. Given a positive initial shear modulus, , and locking stretch, m , too large a positive interaction parameter, a, will lead to Drucker instability in the tensile range. Realistic values of the global interaction parameter, a, will contribute to the characteristic S-shape of tensile stress-strain curves of rubber in the middle strain range before the nal upturn as the locking stretch is approached, without causing instability. The parameter represents a linear mixture parameter combining both invariants I 1 and I 2 into ~ I ; for = 0:0, the Van der Waals potential will be dependent on the rst invariant only. Admissible values for this parameter are 0:0 1:0. When the Van der Waals potential is chosen, the user can specify the coefcients as functions of temperature; alternatively, the parameters can be tted from user-dened test data. The data tting
4.6.111
HYPERELASTICITY
procedure will not necessarily yield a value of between zero and one. If during the curve tting procedure the parameter leaves the admissible range, the curve tting procedure is aborted and restarted with a xed value of = 0:0. Alternatively, the curve tting procedure can be used with a user-dened value of .
Strain energy potential for highly compressible elastomers
While the previous forms are intended for incompressible or almost incompressible materials, the elastic foam energy function is designed for describing highly compressible elastomers (Storkers, 1986). This energy function has the form
U
where
def
=
N X
2
i=1
2 i
^ i + ^ i + ^ i 1 2 3
03+
0 (Je` i i 0 1) ;
(4.6.115)
01 ^ ^ ^ ^ i = Jth 3 i ! 1 2 3 = Je` :
The volumetric and the deviatoric contributions are coupled in this expression, which can be demonstrated clearly by writing the expression in the form
N X 2 i i=1
2 i
0 Je`i i 0 1) :
(4.6.116)
Series expansion of the last two terms in terms of Je` 0 1 shows that the rst-order terms vanish and that the coefcients of the second-order terms are equal to 1 2( 1 + i )(Je` 0 1)2 . Hence, a stable 2 i 3 material is obtained if i > 0 1 . For each term in the energy function, the coefcient i determines 3 the degree of compressibility. i is related to the Poissons ratio, i , by the expressions
i =
0 2 i
i
i =
i
1+2
Thus, if i is the same for all terms, we have a single effective Poissons ratio, . The value of N and tables giving the i , i, and i values as functions of temperature are specied by the user for the hyperfoam material model. The Poissons ratio is valid for nite values of the logarithmic principal strains e1 ; e2 ; e3 ; e2 = e3 = 0e1 in uniaxial tension. For i = i = 0 there is no Poissons effect. The initial shear modulus, 0, again follows from
0 =
and the initial bulk modulus follows from
N X i=1
i ;
k0 =
N X i=1
i (
1 3
i ):
4.6.112
HYPERELASTICITY
If Poissons ratio is constant and known, ABAQUS can calculate the i and i from measurements of nominal stress and stretch as before. If Poissons ratio depends on the level of straining, ABAQUS can also calculate the i from the nominal lateral strains.
Subroutine UHYPER
ABAQUS/Standard also allows other forms of strain energy potentials to be dened for isotropic materials via user subroutine UHYPER by programming the rst and second derivatives of U with respect to I 1 , I 2 , and J in that subroutine.
References
Hyperelastic behavior, Section 10.5.1 of the ABAQUS Analysis Users Manual Elastomeric foam behavior, Section 10.5.2 of the ABAQUS Analysis Users Manual
Constitutive Theories
4.6.113
4.6.2
In this section we will derive the equations needed for tting the hyperelastic (polynomial, Ogden, ArrudaBoyce, and Van der Waals form) and hyperfoam constants to experimental test data. In addition, the procedures for checking the material stability using the Drucker criterion will be described. For the hyperelastic models full incompressibility is assumed in tting the hyperelastic constants to the test data, except in the volumetric test.
Stress-strain relations for the polynomial strain energy potential
The hyperelastic polynomial form can be tted by ABAQUS up to order N = 2. Since the MooneyRivlin potential corresponds to the case N = 1, these remarks also apply by setting the higher-order coefcients to zero. The energy potential is as follows:
U =C10(I 1
2
+
X
i=1
Di
(Je`
0 1)2i:
The deformation modes are characterized in terms of the principal stretches. The nominal stressstrain relations are now derived for the polynomial form with N = 2.
Uniaxial mode
1 = U ; 2 = 3 = U 2 ;
01
U = 1 + U
Constitutive Theories
01
I 2 = U + 2 U :
02
We invoke the principle of virtual work to derive the nominal stress-strain relationship,
U = TU U = @U @U U ;
@U @I 1
=2(1
03 =2(1 0 )
U
@U @I 2
0 3) + C11
I1
0 3 + U (I 2 0 3)
+ 2C02(I 2
0 3)
:
4.6.21
Equibiaxial mode
1
= 2 = B ;
3
= 02 ; B
B
= 1 + B
B ;
05 1 B 0 B
@U @I 1
2
+ B
2 @U
05 =2(B 0 B )
@I 2
0 3)
0 3) + C11
:
2 (I 1 B
0 3) + (I 2 0 3)
02
@U @S
S ;
03 =2(S 0 S )
0 03 ) S
@U @I 1
@U @I 2
:
0 3)
4.6.22
Volumetric mode
1
= 2 = 3 = V ;
U
= 3 V
= 0p J = @U J; @J
N X i=1
p
= 0 @U ; @J
2i 1
=0
Di
(J
0 1)2i01 :
The hyperelastic reduced polynomial form can be tted by ABAQUS up to order N = 6. For N = 3 the reduced polynomial is identical to the Yeoh model, and for N = 1 the neo-Hookean model is retained; hence, the following also applies to these forms. The reduced polynomial energy potential is as follows:
U =
N X i=1
Ci0 (I 1
0 3)i +
N X i=1
Di
(J e`
0 1)2i:
Constitutive Theories
Following the arguments in the previous section, we derive the nominal stress-strain relations for the reduced polynomial.
Uniaxial mode
1 = U ; 2 = 3 = U 2 ;
TU = 2(U
0 02 ) U
N X i=1
01
U = 1 + U
iCi0(I 1
0 3)i01:
Equibiaxial mode
1 = 2 = B ; 3 = B ; TB = 2(B
0 05 ) B
N X i=1
02
B = 1 + B
iCi0(I 1
0 3)i01:
4.6.23
X iC
N i=1
i0(I 1 0 3)
i0 1 :
1 = 2 = 3 = V ; J = 3 V p=0
Xi
N i=1
2
Di
J 0 1)2i01:
X
N
2
i=1
N i i i i ( 1 + 2 + 3 0 3) + 2 i i=1
= 6:
2i (J 0 1) : Di e`
1
Following the same approach as for the polynomial form, we can derive the nominal stress-strain equations for the Ogden form.
Uniaxial mode
1 1 = U ; 2 = 3 = 0 2 ; U = 1 + U U
TU
Equibiaxial mode
X 0 0 0
N
2
i=1
i ( i i U
1 2 i
0 1 ):
1 = 2 = B ; 3 = 02 ; B = 1 + B B TB =
X 0 0 0
N
2
i=1
i B
2 i 0 1
4.6.24
1 = S ; 2 = 1; 3 = 01 ; S = 1 + S S TS =
Volumetric mode
X 0 0 0 0 :
N
2
i=1
i ( i S
1 = 2 = 3 = V ; J = 3 V p=0
Xi
N i=1
2
Di
J 0 1)2i01:
U=
where
X
5
i=1
Ci i i 2i0 2 I 1 0 3 m C3 =
11
J2 0 1 e` 0 ln Je` ; + D 2
1
C1 = ;
2
C2 =
1 20
1050
C4 =
19 7000
and
C5 =
519 673750
Constitutive Theories
Following the same approach as for the polynomial form, we can derive the nominal stress-strain equations for the Arruda-Boyce potential.
Uniaxial mode
1 1 = U ; 2 = 3 = 0 2 ; U = 1 + U U
Equibiaxial mode
1 = 2 = B ; 3 = 02 ; B = 1 + B B
4.6.25
TB
Planar (pure shear) mode
= 2 ( B 0 0 5 ) B
X
5
i=1
1
S ; 2
= 1;
3 = 01 ; S
TS
Volumetric mode
= 2 ( S
0 03 ) S
X
5
S
= 1+
S
i=1
1 = 2 = 3 = V ; p=0
1
J0
3 V
Stress-strain relations for the hyperelastic Van der Waals energy potential
The hyperelastic Van der Waals potential, also known as the Kilian model, has the following form:
U
where
0(m 0 3)
2
ln(1
0 ) + 0 3 a
2
~ I03 2
!3 2 s
J2 0 1 e` 0 ln Je` ; D 2
1
~ I = (1 0 )I 1 + I 2 and =
~ I 03 : 2 0 3 m
Following the same approach as for the polynomial form, we can derive the nominal stress-strain relations for the Van der Waals form.
Uniaxial mode
0 TU = (1 0 03 ) @ U
Equibiaxial mode
1 = U ;
2 = 3
1 1
U 2 ;
01
0 0a
1 ~0 3 I A U (1 0 ) + :
2
U
= 1+
U
1
2
B ; 3 = 02 ; B
4.6.26
B
= 1 + B
0 T B = ( B 0 0 5 ) @ B
Planar (pure shear) mode
s
1
0 0a
1
~ I 0 3A 2
1
1
0 + B :
2
1 = S ; 2 = 1; 3 = 01 ; S = 1 + S S
0 T S = ( S 0 0 3 ) @ S
0 0a
1
~ I 0 3A 2
Volumetric mode
1 = 2 = 3 = V ; J = 3 V p=0 D
1
J0
The hyperfoam potential is a modied form of the Hill strain energy potential and can be tted up to order N = 6:
N X 2 i i=1
2
i
^ ^ ^ i + i + i 0 3 + 1 2 3
0i i 0 1) : (J i e`
1
The deformation modes are characterized in terms of the principal stretches and the volume ratio J . The elastomeric foams are not incompressible: J = 1 2 3 6= 1. The transverse stretches 2 and/ or 3 are independently specied in the test data either as individual values depending on the lateral deformations or through the denition of an effective Poissons ratio.
Uniaxial mode
Constitutive Theories
1 = U ; 2 = 3 ; J = U 2 ; U = 1 + U 2
Equibiaxial mode
1 = 2 = B ; J = 2 2 ; B = 1 + B B 3
Planar mode
1 = P ; 2 = 1; J = P 3 ; P
4.6.27
= 1+
P
The common nominal stress-strain relation for the three deformation modes above is
Tj =
N i i @U 2 0 i i ) ; = ( j 0 J @j j i=1 i
where Tj is the nominal stress and j is the stretch in the direction of loading.
Simple shear mode
2 1
F = 40 1
0 0
3 05;
0 1
where is the shear strain. Note also that for this deformation, stress TS is
TS =
@U @
2 X
J = det(F) = 1.
j =1
2( 2 j
0 1) 0
2 i=1 i
2
2
N X i
( j i
0 1) ;
as follows:
where j are the principal stretches in the plane of shearing, related to the shear strain
1;2 =
1+
r
1+
2
4
The stretch in the direction perpendicular to the plane of shearing is 3 = 1. The transverse stress TT developed during simple shear deformation (as a result of the Poynting effect) is ( )
TT
@U @"
2 X
j =1
Volumetric mode
1 = 2 = 3 = V ; J = 3 : V
The pressure
through
N X i
i=1 i
J 3 i 0 J 0 i i ) :
1
4.6.28
Least squares t
Given experimental data, the material constants are determined through a least-squares-t procedure, which minimizes the relative error in stress. For the n nominal stressnominal strain data pairs, the relative error measure E is minimized,
E=
X 0 ith
n
i=1
T =Titest
2
where Titest is a stress value from the test data and Tith comes from one of the nominal stress expressions derived above. The polynomial potential is linear in terms of the constants Cij ; therefore, a linear least-squares procedure can be used. The Ogden, the Arruda-Boyce, and the Van der Waals potential are nonlinear in some of their coefcients, thus necessitating the use of a nonlinear least-squares procedure.
Linear least squares t for the polynomial model
th For the full polynomial model we can rewrite the expressions for the Tk derived above as th Tk =
N X
i+j =1
k = 1 . . . n;
where the Xij (k ) are functions that depend on the stress state (uniaxial, biaxial, or planar), as explained above. N = 1 for the rst-order polynomial (or Mooney-Rivlin form), and N = 2 for the second-order polynomial. To minimize the relative error, we need to set
@E @Cij
which leads to the following set of M
=
= 0; + 3)
Constitutive Theories
n N X X
1 N (N 2
equations:
X Xlm (k ) Xij (k )Xlm (k ) Cij = test )2 test ; (T k k=1 i+j =1 k=1 Tk
n N X X X i (J k )X j (J k ) k=1 i=1
where
(ptest )2
l + m = 1 . . . N:
This linear set of M equations can be solved readily to dene the coefcients Cij . To t the volumetric coefcients, one needs to solve the system of N equations
Di
n X X j (J k )
test k=1 pk
j = 1 . . . N;
and
Jk
= Vk V
is given by the user. This system of equations can be solved readily for Di.
Linear least squares t for the reduced polynomial model
th For the reduced polynomial model we can rewrite the expressions for Tk derived above as follows:
N X k =
th
i=1
C i0 X i ( k ) ;
where again the functions Xi (k ) depend on the stress state and the stretch, as outlined above, and N is the order of the reduced polynomial, which can take values up to N = 6. The following also applies to the Yeoh and neo-Hookean forms since these models are special cases of the reduced polynomial, with N = 3 and N = 1, respectively. Following the same arguments as for the full polynomial, we arrive at the system of N equations
n N XX k=1 i=1
X i ( k )X j ( k ) C i0 = test (T k ) 2
n X
k=1
X j ( k ) ; test Tk
= 1 . . . N:
This system of equations can be solved readily for the coefcients Ci0. The volumetric coefcients are tted with the same procedure as used for the general polynomial models.
Nonlinear least squares t
The Ogden, Arruda-Boyce, and Van der Waals potentials are nonlinear in some of their coefcients; hence, a nonlinear least-squares-t procedure is required. We use the Marquard-Levenberg algorithm in the formulation by Twizell and Ogden (1986). Let ai, i = 1 . . . m be the coefcients of these hyperelastic models, where m is the number of coefcients contributing to the deviatoric behavior. Specically, m = 2N for the Ogden model, m = 2 for the Arruda-Boyce model, and m = 4 for the Van der Waals model. The coefcients are found by iterating the equation
r ai
( +1)
r = ai
( )
m n XX 0
j =1 k=1
Ek
Pik
@Ek @ai
1 0 T test @Tk @a
th
4.6.210
is the derivative of the vector of relative errors with respect to the coefcients ai . For
= 0, the Newton algorithm is obtained; for very large values of
, the steepest descent method is obtained. Thus, the Marquard-Levenberg algorithm represents a compromise between these two approaches: the value of
is increased if the error grows and is reduced otherwise.
Nonlinear least squares t for the Ogden model
After initializing the i , the parameters i are found with a linear least squares t. In the iterative procedure outlined above, the following derivatives are used:
(0) (0) th @Tk @i th @Tk @i
= =
i 01 0 ci 01 );
i i
0 22i ( 01 0 c 01 ) 0 2 i ( 01 0 cc 01 ) ln ;
i
where
8 <0 ; c = 0 2; : 0 1;
1 2
The Arruda-Boyce model is linear in the shear modulus but nonlinear in the locking stretch m . (0) The locking stretch is initialized as m = max(7:0; 3:0 2 max ), where max is the maximum stretch in the user-specied test data. Given this locking stretch, the initial shear modulus, (0), is obtained with a linear least squares t. In the iterative procedure outlined above, the following derivatives are used:
8 iC > 2(U 0 0 ) Pi 0 I 0 ; if uniaxial; U > < P @Tk 0 iC 2(B 0 B ) = i 0 I 0 ; if biaxial; @ > iC > 2(S 0 0 ) Pi 0 0 ; if planar. : S I 8 P iC > 2(U 0 0 ) Pi (2 0 2i) 0 I 0 ; if uniaxial; U < @Tk 0 iC 2 ( B 0 B ) = i (2 0 2i) 0 I 0 ; if biaxial; @m > iC > 2(S 0 0 ) Pi (2 0 2i) 0 0 ; if planar. : S I
2 th 5 =1
2i 2
m m
Constitutive Theories
5 =1
1
i
2i 2
5 =1
2i 2
m
th
5 =2
2i 1
m m
5 =2
1
i
2i 1
5 =2
2i 1
m
The Van der Waals model is linear in the shear modulus but nonlinear in the locking stretch m , the global interaction parameter a, and the mixture parameter . The locking stretch is initialized as (0) = max(10:0; 3:0 2 max ), where max is the maximum stretch in the user-specied test data. m
4.6.211
Given this guess for the locking stretch, we make use of an expression proposed by Kilian et al. (1986) to initialize the global interaction parameter
a(0)
( (
(0) 2
(0) 3 )
m)
01
The invariant mixture parameter is initialized to (0) = 0. Given these initial values, the shear modulus, (0) , is initialized using a linear least-squares-t procedure. In the iterative procedure outlined above, the following derivatives are used:
th @Tk @
8 > 0(1 0 03 ) (10)(m2 03) U (1 0 ) + ; if uniaxial; 2 > U m > < th @Tk = 0(B 0 05 ) (10)(m2 03) 1 0 + 2 ; if biaxial; 2 @m > B B m > : 0(S 0 03 ) m2 ; if planar. S (10)2(m 03) 8 q > I > 0(1 0 03 ) ~03 U (1 0 ) + ; if uniaxial; U 2 > < th q~ @Tk 05 ) I03 1 0 + 2 ; if biaxial; = B 2 > 0 ( B 0 B @a > q~ > : 0 ( S 0 0 3 ) I 0 3 ; if planar. S 2 8 > 2(1 0 03 ) @U U (1 0 ) + + U (1 0 U ) ; if uniaxial; > U @ < th @Tk = > 2(B 0 05 ) @U 1 0 + 2 + U 0(2 0 1) ; if biaxial; @ > B B B @ :
0
8 q ~ > (1 0 03 ) 1 0 a I 03 U (1 0 ) + ; > U 10 2 > > q ~ < 05 1 I 03 2 = ( B 0 B ) 10 0 a 1 0 + B ; 2 > > q~ > 1 > : ( S 0 0 3 ) 10 0 a I 0 3 ; S 2
0;
if planar.
U0 =
and
@U ~ @I
1
0 @ =
2 1
s
1
0 0a
1
~ I
1 0 3A ;
2
@U 0 @
"
=
1 4
y 0 3 (1 0 2 )
08
~ I 03
#
(I 1
0 I 2 ):
4.6.212
ABAQUS checks the Drucker stability of the material for the rst three modes of deformation described above. The Drucker stability condition requires that the change in the Kirchhoff stress d following " from an innitesimal change in the logarithmic strain d" satises the inequality
=D
satised. For the isotropic elastic formulation considered here, the inequality can be represented in terms of the principal stresses and strains:
" D : d" > 0; thus requiring the tangential material stiffness D to be positive denite for material stability to be
With the incompressibility assumption for the two hyperelastic models, the Kirchhoff stress is equal to the Cauchy stress: = and, thus,
d"1 =
d1 ; 1
d"2 =
d2 : 2
The stresses follow from the strain energy, which in turn follow from the changes in the strain invariants or in the stretches. The relation between changes in the stress and changes in strain are described by the matrix equation
d1 d2
d"1 d"2
4.6.213
where
@U 2 @U + 2 @I 1 @I 2 @I 1
2 + 1
+ 4( 1
2 0 2 )2 3
@U
@2U @I 2 1 @2U @I 2 1
2 2 2 @ U 4@ U ; + 2 2 + 2 @I 1@I 2 @I 2 2 2 + 2 1 2 @ 2U 4@ U ; + 1 @I 1@I 2 @I 2 2
! !
@U @I 2
2 + 4(2
2 )2 3
2 @U 02 @U + 4(2 0 2 )(2 0 2 ) @ U + 4 3 1 3 2 3 @I 1 @I 2 @I 2 1
2 2 + (1 + 2 )
2 @ 2U 2 2@ U : + 1 2 @I 1@I 2 @I 2 2
For the Ogden form we follow the same approach as the polynomial form. Using 3 have
0101 , we 1 2
N X 2 i
i=1
2 i
( 1i +
i + 0i 0i 0 3); 2 1 2 i
( 1i
1 = 1 2 = 2
and the material stiffness
@U @1 @U @2
N X 2 i i=1 N X i=1
2
0 0 i 0 i ) ; 1 2
i i 0 i 0 i ( 0 1 2 ); i 2
i=1
2i i 2 i 0i 0i 1 1 + 1 i 21 i + 1 : 1 2 1 2
Arruda-Boyce form
For positive values of the shear modulus, , and the locking stretch, m , the Arruda-Boyce form is always stable. Hence, it sufces to check the coefcients to determine whether the material satises Drucker stability.
4.6.214
@U @I 1
= (1
0 ) @U ~
@I
and
@U @I 2
@U : ~ @I
To determine the admissible stretch range, we need to nd the two positive real-valued roots neighboring = 1 of the equation
~ I ( ) 0 2 = 0 m
for each of the three stress statesuniaxial, biaxial, and planarby using a simple bisection method.
Hyperfoam
The Kirchhoff stress-strain relation for the uniaxial, biaxial, planar, and volumetric deformation modes is
j = j
N X i=1
@U @j
=2
X
N i=1
i i
i 0 J 0 i i : j
3
dj = 2
Since we cannot use the incompressibility assumption, we have to use all three principal stress and matrix, strain components and a 3 2 3
8 9 < d1 =
D11 D12 D13 < d"1 = d2 = 4 D21 D22 D23 5 d"2 : : d3 ; D31 D32 D33 : d"3 ;
Specically,
8 9 < d1 =
38
38
4.6.215
+ D 22 +
D33 >
0;
D11 D22
D)
D23
D13
D12 >
0;
>
0:
References
Hyperelastic behavior, Section 10.5.1 of the ABAQUS Analysis Users Manual Elastomeric foam behavior, Section 10.5.2 of the ABAQUS Analysis Users Manual
4.6.216
MULLINS EFFECT
4.7.1
MULLINS EFFECT
The Mullins effect material model provided in ABAQUS is intended for modeling the phenomenon of stress softening, commonly observed in lled rubber elastomers as a result of damage associated with straining. When an elastomeric test specimen is subjected to simple tension from its virgin state, unloaded, and then reloaded, the stress required on reloading is less than that on the initial loading for stretches up to the maximum stretch achieved during the initial loading. This stress-softening phenomenon is known as the Mullins effect (Mullins, 1947).
Material behavior
The following sections describe the Mullins effect model in ABAQUS in detail.
Idealized material behavior
' c
stress c
' b
Constitutive Theories
a stretch
Figure 4.7.11 Idealized response with Mullins effect. This gure and the accompanying description are based on work by Ogden and Roxburgh (Ogden and Roxburgh, 1999), which forms the basis of the model implemented in ABAQUS. Consider the 0 0 primary loading path abb of a previously unstressed material, with loading until an arbitrary point b . 0 0 On unloading from b , the path b Ba is followed. When the material is loaded again, the softened 0 0 0 0 0 path is retraced as aBb . If further loading is then applied, the path b cc is followed, where b cc is a
4.7.11
MULLINS EFFECT
continuation of the primary loading path abb cc d (which is the path that would be followed if there 0 0 was no unloading). If loading is now stopped at c , the path c Ca is followed on unloading and then 0 0 0 retraced back to c on reloading. If no further loading beyond c is applied, the curve aCc represents 0 the subsequent material response, which is then elastic. For loading beyond c , the primary path is again followed and the pattern described is repeated. This is an ideal representation of Mullins effect. More details regarding the actual behavior and how test data that display such behavior can be used to calibrate the ABAQUS model for Mullins effect are discussed in Mullins effect in elastomers, Section 10.6.1 of the ABAQUS Analysis Users Manual, and in Analysis of a solid disc with Mullins effect, Section 3.1.7 of the ABAQUS Example Problems Manual. 0 0 The loading path abb cc d will henceforth be referred to as the primary material response, and the constitutive behavior of the primary response can be dened using the standard energy potentials of the hyperelasticity models in ABAQUS. Stress softening is interpreted as being due to damage at the microscopic level. As the material is loaded, the damage occurs by the severing of bonds between ller particles and the rubber molecular chains. Different chain links break at different deformation levels, thereby leading to continuous damage with macroscopic deformation. An equivalent interpretation is that the energy required to cause the damage is not recoverable.
Basic framework for the model
Hyperelastic materials are described in terms of a strain energy potential function, U (F), which denes the strain energy stored in the material per unit reference volume (volume in the initial conguration). The quantity F may be the deformation gradient tensor or some other appropriate strain measure. To account for Mullins effect, Ogden and Roxburgh proposed a material description that is based on an energy function of the form U (F; ), where is a scalar variable. This scalar variable controls the material properties in the sense that it enables the material response to be governed by an energy function on unloading and subsequent submaximal reloading different from that on the primary (initial) loading path from a virgin state. Because of the interpretation and inuence of , it is no longer appropriate to think of U as the stored elastic energy potential. In fact, part of the energy is stored as strain energy, while the rest is dissipated due to damage. The shaded area in Figure 4.7.11 represents the energy dissipated by damage, while the unshaded portion represents the recoverable part of the strain energy. As pointed out by Ogden and Roxburgh, the inclusion of results in the following additional equation, due to equilibrium:
@U (F ; ) = 0 : @
The above equation determines the evolution of during the course of the deformation. During a deformation process, the variable may be either active or inactive and may switch from inactive to active or vice versa such that it always varies continously. When it is not active, the material behaves as an elastic material with strain energy U (F; ), with held constant. However, when is active, it is determined implicitly in terms of F by the preceding equation. The material again behaves as an elastic material but with a different strain energy function U (F; (F)). When the parameter switches from being active to inactive and vice versa, the energy as well as the associated stresses
4.7.12
MULLINS EFFECT
remain continuous. When is inactive, it is assumed to have a value of unity without any loss of generality.
Primary hyperelastic behavior
In preparation for writing the constitutive equations for the Mullins effect, it is useful to separate the deviatoric and the volumetric parts of the total strain energy density of the primary material response as U = Udev + Uvol : In the above equation U , Udev , and Uvol are the total, deviatoric, and volumetric parts of the strain energy density, respectively. All the hyperelasticity models in ABAQUS use strain energy potential functions that are already separated into deviatoric and volumetric parts. For example, the polynomial models use a strain energy potential of the form
i+j =1
N X
N X
i=1 Di
(J
e` 0 1)2i;
where the symbols have the usual interpretations. The rst term on the right represents the deviatoric part, and the second term represents the volumetric part of the elastic strain energy density function.
Augmented strain energy density function
The Mullins effect is accounted for by using an augmented energy function of the form
~ ~ U (i ; ) = Udev (i ) + () + Uvol (J e` ); ~ where Udev (i ) is the deviatoric part of the strain energy density of the primary material response, dened, for example, by the rst term on the right-hand-side of the polynomial strain energy function ~ given above; Uvol (J e` ) is the volumetric part of the strain energy density, dened, for example, by the second term on the right-hand-side of the polynomial strain energy function given above; i e` (i = 1; 2) represent the deviatoric principal stretches; and J represents the elastic volume ratio. The function () is a continuous function of the damage variable and is referred to as the damage function. As the above expression suggests, the deviatoric part of the augmented energy function is related to the deviatoric part of the energy corresponding to the primary response by the scaling factor . The volumetric part of the augmented energy function is the same as that for the primary response. A consequence of the above form of the augmented energy function is that the Mullins effect is associated only with the deviatoric part of the deformation. When = 1, it is required that ~ ~ () = 0, such that U (i ; 1) = Udev (i) + Uvol (J e` ), and the augmented energy function reduces to the strain energy density function of the primary material response. As pointed out earlier, this situation corresponds to being inactive and physically represents the energy of a material point that is on the primary curve. The value of varies continuously as the deformation proceeds. The above form of the energy function is an extension of the form proposed by Ogden and Roxburgh to account for material compressibility.
Constitutive Theories
4.7.13
MULLINS EFFECT
Stress computation
With the above modication to the energy function, the stresses, which are derived from the energy function in the usual manner, are given by
~ (i; ; J e` ) = S(i) 0 p(J e` )I; ~ ~ where S is the deviatoric stress corresponding to the primary material response at the current deviatoric ~ deformation level, i , and p is the hydrostatic pressure of the primary material response at the current volumetric deformation level, J e` . Thus, the deviatoric stress as a result of Mullins effect is obtained by simply scaling the deviatoric stress of the primary material response with the damage variable . For the above equation to predict stress softening, it is necessary that 1 on the unloading (and submaximal reloading) path. It is also assumed that > 0, such that the deviatoric stresses remain positive until 1 = 2 = 1 is reached. The pressure stress of the augmented response is the same as that of the primary response. The model predicts unloading-reloading along a single curve (that is different, in general, from the primary loading curve) from any given strain level, that passes through the origin of the stress-strain plot. It cannot capture permanent strains upon removal of load. The model also predicts that a purely volumetric deformation will not have any damage or Mullins effect associated with it.
As a result of the assumed form of U and the evolution equation for , the function () satises
~ 0 () = Udev (1 ; 2 );
0
which implicitly denes the damage parameter in terms of the deformation as long as the function 0 () is a monotonic function of its argument. The value of derived from the above equation will m m depend on the maximum values of the principal stretches 1 and 2 attained on a primary loading ~ path and also on the specic forms of the functions Udev (1; 2 ) and () used. On the primary curve at any point where the unloading is initiated, = 1, and the above equation specializes to
m ~ 0 (1) = Udev (m ; m ) Udev ; 1 2
0
m which denes the notation Udev . On the other hand, when the material is fully unloaded with 1 = 2 = 1 and attains its minimum value, m :
From the preceding two equations it follows that both the function () and the quantity m depend m m on the point (1 ; 2 ) at which unloading is initiated. ~ ~ When the material is in a fully unloaded state (Udev (1; 1) = Uvol (1) = 0 ), the augmented energy function has the residual value U (1; m) = (m );
4.7.14
MULLINS EFFECT
which may be interpreted as the amount of energy dissipated by damage in the material. In a uniaxial test such as simple tension, this quantity denotes the area between the primary loading curve and 0 the relevant unloading-reloading curve (the shaded area in Figure 4.7.11 for deformation until c ). m , on the point at which unloading starts, the Recalling that the function depends, through Udev function may be constructed, following Ogden and Roxburgh, using
m ( ) + Udev
0
m m m where the function f (; Udev ) must satisfy f (1; Udev ) = 0 and f (m ; Udev ) m Udev , the above equation can be integrated to obtain () as
( ) =
1
On the primary path = 1, and the above equation satises the requirement (1) 0 uses the following specic choice for (), which has the required properties
m m () = (m + Udev )erf 01 (r(1 0 )) 0 Udev :
0
ABAQUS
The quantities r, m, and are material constants, and where the error function is dened as
2 erf (x) = p
0
erf
01
x 0
exp(
0w2 )dw:
The above choice of () reduces to the form proposed by Ogden and Roxburgh when = 0. 0 0 ~ On substituting the above form of () in the equation () = 0Udev (1 ; 2 ), it can be shown that the damage variable varies with the deformation according to
= 10
erf (
Constitutive Theories
~ The above equation denes the evolution of the damage variable with deformation Udev (1; 2), the maximum value of the deviatoric strain energy density experienced by the material point during m its deformation history, Udev , and material constants r, m, and . While the parameters r and are dimensionless, the parameter m has the dimensions of energy. m ~ When Udev = Udev , corresponding to a point on the primary curve, = 1:0. On the other hand, ~ attains its minimum value, m upon removal of deformation, when Udev = 0. For all intermediate ~ values of Udev , varies monotonically between m and 1:0.
The nonlinear solution procedure in ABAQUS/Standard requires the material tangent stiffness or the material Jacobian. The derivation for the material tangent stiffness follows the procedure outlined in Hyperelastic material behavior, Section 4.6.1. As a result of the Mullins effect, the expression for
4.7.15
MULLINS EFFECT
The remaining terms are identical. To determine the form of S , it may be noted that the deviatoric stress with Mullins effect can be expressed in terms of the deviatoric stress of the primary material response and the damage variable as
CS is modied.
~ S = S:
@S = @S @e
~ ~ @ @ Udev @S ~ : @e + S : @ e: ~ @e @ Udev @ e ~ @S @e +J
which determines
Damage dissipation
CS .
As pointed out in the earlier discussion, the quantity (m ) determines the energy dissipated due 0 to damage. The equation for () can be integrated analytically through a change of variable of ~ integration from to Udev to yield the following expression for the function ():
( ) =
m (m + Udev )
The damage dissipation can be obtained from the above expression by substituting ~ obtained by setting Udev = 0 in the expression for ) and is given by
p r
[exp(
dev
m (with m
m m m Udev (m + Udev ) Um pU m (1 0 exp(0[ m + dev m ]2)] Udev : )0 m m + Udev Udev r dev The recoverable part of the strain energy, Ure , can then be obtained by subtracting the damage
(m ) = [erf(
~m ~m When the material is fully or almost incompressible, we may assume that Uvol << Udev , and the ~m total energy corresponding to the maximum deformation along the primary curve is given by Udev . The discussion below pertains to the above limit. From the expression for the damage dissipation, m (m ), it may be noted that when m = 0, the ratio of dissipation to Udev is a constant that depends ~m only on the material parameters r and . In particular, this ratio is independent of Udev , suggesting that the ratio of the dissipation to the total energy is a constant that is independent of the level of ~m deformation. For nonzero m the above is true only when Udev >> m, which correponds to relatively m ~ large levels of deformation. In that case when Udev << m, the energy dissipation is zero. Thus, the value of m (both zero and nonzero) may be thought of as providing a scale of the strain energy level for the total deformation beyond which dissipation takes place. In particular, when m = 0, dissipation occurs at low strain levels. The behavior described above is depicted in Figure 4.7.12, which shows a plot of the normalized m m dissipation, (m )=Udev , versus the ratio Udev =m, for nonzero m and for xed values of the parameters m r(= 2:0) and (= 0:1). For small values of the ratio Udev =m the dissipation tends to zero, while for larger values of the ratio, the normalized dissipation approaches a constant asymptotic value.
4.7.16
MULLINS EFFECT
(m)
Udev
m
Udev m
Figure 4.7.12 Normalized dissipation versus deformation.
Constitutive Theories
Reference
Mullins effect in elastomers, Section 10.6.1 of the ABAQUS Analysis Users Manual
4.7.17
VISCOELASTICITY
4.8.1
VISCOELASTICITY
(t) =
Z
0
2 G (
0 )e dt _
0
Z
0
+I
K (
_ 0 ) dt :
0 0
Here e and are the mechanical deviatoric and volumetric strains; K is the bulk modulus and G is the shear modulus, which are functions of the reduced time ; and _ denotes differentiation with respect to t0 . The reduced time is related to the actual time through the integral differential equation
Z
=
dt0
0
0 A ((t ))
d dt
A ((t))
where is the temperature and A is the shift function. (Hence, if A = 1, = t.) A commonly used shift function is the Williams-Landell-Ferry (WLF) equation, which has the following form:
g ( 0 0 log A = h() = CC1+( 0 g) ) ; g g
where C1 and C2 are constants and g is the glass transition temperature. This is the temperature g at which, in principle, the behavior of the material changes from glassy to rubbery. If g 0 C2 , g g deformation changes will be elastic. C1 and C2 were once thought to be universal constants whose values were obtained at g , but these constants have been shown to vary slightly from polymer to polymer. ABAQUS allows the WLF equation to be used with any convenient temperature, other than the glass transition temperature, as the reference temperature. The form of the equation remains the same, but the constants are different. Namely,
Constitutive Theories
where 0 is the reference temperature at which the relaxation data are given, and C1 and C2 are the g g calibration constants at the reference temperature. The universal constants C1 and C2 are related to C1 and C2 as follows: g
g 0 g )=C2 ; 0 g : Other forms of h() are also used, such as a power series in 0 0 . ABAQUS allows a general denition
1 + ( 0 g C 2 = C 2 + 0
C1 =
C1
of the shift function with user subroutine UTRS. The relaxation functions K (t) and G(t) can be dened individually in terms of a series of exponentials known as the Prony series:
K ( ) = K 1 +
X
nK i=1
Ki e0 =i
G ( ) = G 1 +
X
nG i=1
Gi e0=i ;
4.8.11
VISCOELASTICITY
where K1 and G1 represent the long-term bulk and shear moduli. In general, the relaxation times iK and iG need not equal each other; however, ABAQUS assumes that i = iK = iG . On the other hand, the number of terms in bulk and shear, nK and nG , need not equal each other. In fact, in many practical cases it can be assumed that nK = . Hence, we now concentrate on the deviatoric behavior. The equations for the volumetric terms can be derived in an analogous way. The deviatoric integral equation is
S=
0
2 2
G1 + G1 +
nG X i nG i
=1
=1
Gi e( 0 )=i
0
de 0 d : d 0
S = 2 G0 e 0
where G0 term i, and
n X i
=1
i ei ;
(4.8.11)
1 0 e( 0 )=i
0
de d 0 d 0
(4.8.12)
is the viscous (creep) strain in each term of the series. For nite element analysis this equation must be integrated over a nite increment of time. To perform this integration, we will assume that during e= . To use this relation, we break up the increment e varies linearly with ; hence, de=d 0 Equation 4.8.12 into two parts:
=1 1
e +1 =
n i
Z n
1 0 e( 0 n+1)=i
0 0
de=d 0 during the increment yields Z n de n+1 01=i ei = 1 0 e d 0 0 n d 0 Z de 01=i 1 0 e( 0 n)=i d 0 d 0 +e 0 1e Z n+1 1 0 e( 0 n+1 )=i d 0 : + 1 n
0 0
4.8.12
VISCOELASTICITY
The rst and last integrals in this expression are readily evaluated, whereas from Equation 4.8.12 follows that the second integral represents the viscous strain in the ith term at the beginning of the increment. Hence, the change in the ith viscous strain is
1=i approaches zero, this expression can be approximated by 1 1ei = 1i 2 1e + en 0 en : i The last form is used in the computations if 1=i < 1007 .
If
nG X i
1e 1
(4.8.13)
(4.8.14)
Hence, in an increment, Equation 4.8.13 or Equation 4.8.14 is used to calculate the new value of the viscous strains. Equation 4.8.11 is then used subsequently to obtain the new value of the stresses. The tangent modulus is readily derived from these equations by differentiating the deviatoric stress increment, which is !
1S = 2G0 1e 0 1 e.
=1
i(en+1 0 en ) i i
with respect to the deviatoric strain increment only on the reduced time step:
GT
h 8 < G0 h :G
Constitutive Theories
PD =
P PE =
= 1 (Sn+1 + Sn) : 1e 2 1
0
4G 0
Sn+1 : Sn+1 0 Sn : Sn :
4.8.13
VISCOELASTICITY
Finally, one needs a relation between the reduced time increment, , and the actual time increment, To do this, we observe that A varies very nonlinearly with temperature; hence, any direct approximation of A is likely to lead to large errors. On the other hand, h will generally be a smoothly varying function of temperature that is well approximated by a linear function of temperature over an increment. If we further assume that incrementally the temperature is a linear function of time t, one nds the relation
1 t.
()
h
( ) = 0 ln A ((t)) = a + bt
A 01 t
or
( ( )) = ea+bt
with
=1 b
ea+bt dt
n+1
ea+bt
0 ea+btn
So far, we have discussed full triaxial stress states. If the stress state is reduced (i.e., plane stress or uniaxial stress), the equations derived here cannot be used directly because only the total stress state is reduced, not the individual terms in the series. Therefore, we use the following procedure. For plane stress let the third component be the zero stress component. At the beginning of the increment we presumably know the volumetric elastic strain n , the volumetric viscous strain n , and e c the volumetric viscous strains n associated with the Prony series. The total volumetric strain can be i obtained by adding together the elastic volumetric strain and the volumetric viscous strain
n
= n + n : e c
(4.8.15)
VISCOELASTICITY
= 2G0 3
en+1 0
nG X
i
=1
e3 +1 :
G n i i
s3 +1 = 2GT 1e3+2G0 en 3
"
10
nG X i
0 2G 0
n
nG X i
=1
=1 G e01=i eni : i 3
nK X i
(1 0 e01=i )
G i
(4.8.16)
0p +1 = K T 1+K0 n 0K0
We can write these equations in the form
i
"
nK X
=1
=1 K e01=i n :
i i
10
(1 0 e01=i )
K i
#
(4.8.17)
2 G T 1 e 3 + K T 1 + s3 0 p = 0 : (4.8.18) Since 1e3 = 2 1 0 1"1 0 1"2, it follows that 3 (K T + 4 GT )1 = 2GT (1"1 + 1"2) 0 s3 + p; 3 from which 1 can be solved. One can then also calculate 1e1 and 1e2, and with Equation 4.8.13 n+1 n+1
or Equation 4.8.14 one can update the deviatoric viscous strains ei . The volumetric strains i are obtained with a relation similar to Equation 4.8.13. For uniaxial stress states a similar procedure is used. As before, n follows from Equation 4.8.15 "3 : and en and en follow from "1 3 2
Constitutive Theories
Equation 4.8.16 and Equation 4.8.17 can be used to calculate Equation 4.8.18. Applying Equation 4.8.19 for e3 ;
+2 = 1 1 1 en = e n = " n 0 n = n 0 "n : 3 2 3 3 6 21 1
(4.8.19)
s3
(K T + 1 GT )1 = GT 1"1 0 s3 + p: 3
4.8.15
VISCOELASTICITY
After this, one can follow the same procedure as for plane stress.
Automatic time stepping procedure
To create an automatic time stepping procedure in ABAQUS/Standard, we want to compare viscous strain rates at the beginning and the end of the increment. The strain rates in the individual terms at the beginning of the increment can be obtained directly by taking the limit of the incremental strain:
_i en =
= A ( n )
+1
i
en 0 en i
:
en+1 0 en+1 i
A similar expression can be derived for the strain rate at the end of the increment:
_i en+1 = 1V = 1t e = 1t
G X
i
A (n+1 )i
If we use these expressions to calculate a difference in estimated total viscous strain increment, one nds n
nG X G en+1 0 en+1 i i i
=1
_i G (_ n+1 0 en) i ei
=1 i
A ( n )
n en i 0 e 0n ) : A(
1V , and from these two quantities a suitable scalar measure can be constructed; for example, r est = 2 1V : 1V + 1 12 : 1" 3 e e 3 V
This expression is readily evaluated. A similar expression can be calculated for volumetric strain
Comparison with the user-specied strain increment tolerance allows construction of an automatic time stepping scheme.
Reference
Time domain viscoelasticity, Section 10.7.1 of the ABAQUS Analysis Users Manual
4.8.16
FINITE-STRAIN VISCOELASTICITY
4.8.2
FINITE-STRAIN VISCOELASTICITY
Integral formulation
The nite-strain viscoelasticity theory implemented in ABAQUS is a time domain generalization of either the hyperelastic or hyperfoam constitutive models. It is assumed that the instantaneous response of the material follows from the hyperelastic constitutive equations:
0(t) = D (F(t)) + H (J (t)) 0 0
for an incompressible material. In the above, D and H are, respectively, the deviatoric and the 0 0 hydrostatic parts of the instantaneous Kirchhoff stress 0. F is the distortion gradient related to the deformation gradient F by
F = F1 ;
J3
where
J = det(F)
is the volume change. Using integration by parts and a variable transformation, the basic hereditary integral formulation for linear isotropic viscoelasticity can be written in the form
(t) = 2 G 0 e (t) +
Z
0
_ 0 2G( ) (t
e 0 t )d
0
K 0 (t) +
Z
0
_ K ( 0 ) (t
0 t )d
0
Constitutive Theories
S 0 (t) +
Z
0
_ 0 G ( )
G0
S0(t 0 t )d
0
p0 (t) +
Z K ( ) _
0
K0
p (t
0 t )d
0
!
0
_ _ where is the reduced time, G( 0 ) = dG( 0)=d 0, and K ( 0) = dK ( 0)=d 0. G0 and K0 are the instantaneous small-strain shear and bulk moduli, and G(t) and K (t) are the time-dependent smallstrain shear and bulk relaxation moduli. Recall that the reduced time represents a shift in time with temperature and is related to the actual time through the differential equation
d 0 =
dt0 A ((t0 ))
4.8.21
FINITE-STRAIN VISCOELASTICITY
A suitable generalization to nite strain of the hereditary integral formulation is obtained as follows:
"Z
F 0 1 (t 0 t 0 ) 1
t
_ G( 0 ) D 0 (t G0
_ 0 t 0 ) + K (
0)
K0
0
0 (t 0 t )
1 Ft (t 0 t0)d 0
where Ft(t 0 t0) is the deformation gradient of the state at t 0 t0 relative to the state at t. A transformation is performed on the stress relating the state at time t 0 t0 to the state at time t. We also ensure the symmetry of the transformed stress. Observe that because H is proportional to the 0 identity tensor, we have
F 0 1 (t 0 t 0 ) 1 H (t 0 t 0 ) 1 F t (t 0 t 0 ) = H (t 0 t 0 ): 0 0 t
It can also be observed that
tr
F 0 1 (t 0 t 0 ) 1 D (t 0 t 0 ) 1 F t (t 0 t 0 ) = F 0 1 (t 0 t 0 ) : D (t 0 t 0 ) 1 F t (t 0 t 0 ) 0 0 t t 0 1 01 D (t 0 t 0 ) : F t (t 0 t 0 ) 1 F t (t 0 t 0 ) = D (t 0 t 0 ) : I = 0 ; 0 0
since D is deviatoric. Hence the deviatoric and volumetric parts can be separated into two hereditary 0 integrals:
"Z Z
0
_ G ( 0 )
_ K ( 0 ) H H (t) = H (t) + 0 (t 0 K0 0
G0
F01 (t 0 t0 ) 1 D (t 0 t0 ) 1 Ft (t 0 t0)d 0
t
#
;
0 t0 )d 0:
(4.8.21)
Implementation
As in small-strain viscoelasticity, we represent the relaxation moduli in terms of the Prony series
G ( ) = G 0 g1 + k1 +
NG X i=1 NK X i=1
G gi e0 =i
!
;
(4.8.22)
K ( ) = K 0
K kie0=i
!
;
(4.8.23)
PN PN
G K where gi and ki are the relative moduli of terms i. Note that g1 + i=1 gi = k1 + i=1 ki = 1. K G ABAQUS assumes that the relaxation times i = i = i are the same so that from here on, we will sum on N terms for both bulk and shear behavior. In reality, the number of nonzero terms in bulk and shear, NK and NG , need not be equal, unless the instantaneous behavior is based on the hyperfoam model. In the latter case, the two deformation modes are closely related and are then assumed to relax equally and simultaneously.
4.8.22
FINITE-STRAIN VISCOELASTICITY
"
N X gi Z i
(t) = 0 (t) 0
H
N X ki Z i
=1 i 0
#
;
=1
i
0
(t 0 t0)e0 i d 0:
(4.8.24)
Next, we introduce the internal stresses, associated with each term of the series
D t i
( ) SYM
gi i
;
(4.8.25)
H t i
( ) i
ki
H t 0
( 0 t0)e0 i d 0:
0
(4.8.26)
These stresses are stored at each material point and are integrated forward in time. We will assume t. that the solution is known at time t, and we need to construct the solution at time t
+1
+ 1t follow from
H t 0
( + 1t 0 t0)e0 i d 0:
0
( + 1 t) =
Z Z
ki
i
+1
With
= ki e
i
0 1
i
0 1 0 0 1
H 0 t
( 0 t)e0 i d
( 0 t)e0 i d + ki e0 1i
H t 0
H t 0
( 0 t)e0 i d ;
( + 1t) =
ki
i
0 1 e i
0
0 1
H t 0
(4.8.27)
To integrate the rst integral in Equation 4.8.27, we assume that H t 0 t varies linearly with 0 the reduced time over the increment
H t 0 ( 0 t) = (1 + 1 ) H (t) 0 1 H (t + 1t) 0 0
0 1 0:
(4.8.28)
4.8.23
FINITE-STRAIN VISCOELASTICITY
H (t + 1t) = i
The integrals are readily evaluated, providing the solution at the end of the increment
H (t + 1 t) = i
or, in a slightly different form
i = e0 i ; i = 1 0
1
= =0
1 t = 1 = 0, i
t + 1t follow from
(4.8.210)
FINITE-STRAIN VISCOELASTICITY
With
D t i
gi i gi i gi i
e e e
0 1i 1F 1 0 1
i
Z
0 1
i
1F 1
01Z
0 1
F01(t 0 t) 1 D (t 0 t) 1 Ft (t 0 t) e0 i d 1 1F01 0 t
F01(t 0 t) 1 D (t 0 t) 1 Ft (t 0 t) e0 i d 1 1F01 0 t
(4.8.211)
(4.8.212)
Note that
0 t
^D ( ) = SYM 21F 1 F01(t) 1 D (t) 1 Ft(t) 1 1F013 = SYM 21F 1 D (t) 1 1F013 0 0 t
(4.8.213)
and
0 t
^D ( + 1t) = SYM 21F 1 F01(t + 1t) 1 D (t + 1t) 1 Ft(t + 1t) 1 1F013 = D (t + 1t): 0 0 t
Then we can also introduce
i t
(4.8.214)
Constitutive Theories
( + 1 t) = i
0 1 e i
0
0 1
0 t
(4.8.216)
To integrate the rst integral in Equation 4.8.216, we assume that 0 t 0 t varies linearly with the reduced time over the increment:
0 t ^0 ^D ( 0 t) = (1 + 1 )^D (t) 0 1 D (t + 1t) 0
^D (
0 1 0; 0 1 0:
(4.8.217)
4.8.25
FINITE-STRAIN VISCOELASTICITY
Equation 4.8.216 and Equation 4.8.217 for the deviatoric stress have exactly the same form as Equation 4.8.27 and Equation 4.8.28 for the hydrostatic stress. Hence, after integration we obtain
(4.8.218)
Equation 4.8.213, Equation 4.8.215, and Equation 4.8.218, thus, provide a straightforward integration scheme. The total stress at the end of the increment becomes
1 i = e0 i ; i = 1 0 i (1 0 i ); i = i (1 0 i ) 0 i : 1 1
(4.8.219)
which with Equation 4.8.29 and Equation 4.8.218 can also be written as
(t + 1 t) =
10
0
N X i=1
N X i=1
i g i D (t + 1 t) + 0
10
iki H (t + 1t) 0
(4.8.220)
^0 i g i D (t) 0
N X i=1
Rate equation
To solve the system of nonlinear equations generated by the constitutive equations, we need to generate the corotational constitutive rate equations. From Equation 4.8.220 it follows
(t + 1t) =
0
10
N X i=1
N X i=1
0 i gi D (t + 1t) +
i=1
10
N X i=1
0 i k i H (t + 1 t)
(4.8.221)
N D X D ^ ^ i g i 0 (t) 0 i i (t);
where t is the corotational (Jaumann) stress rate. Since H t and H t in Equation 4.8.220 i 0 t and are independent of the increment size, their derivatives vanish. The derivatives D t 0 t follow from the hyperelastic equations being used and, thus, do not need to be considered H t 0 here. With Equation 4.8.213 it follows that
( ) ( +1 )
()
()
( +1 )
(4.8.222)
;
4.8.26
FINITE-STRAIN VISCOELASTICITY
where
is the velocity gradient. Using the denition of the corotational (Jaumann) rate, it follows that (4.8.223)
where
hence, substitution of Equation 4.8.223 and Equation 4.8.224 into Equation 4.8.222 yields
h i D ^0 ^0 ^ 0 (t) = SYM D 1 D (t) 0 D (t) 1 D = 0 D ^0 ^ since both D and D (t) are symmetric. Similarly for i (t), h i D ^i ^i ^ i (t) = SYM D 1 D (t) 0 D (t) 1 D = 0:
(4.8.225)
(4.8.226)
(t + 1 t) =
10
N X i=1
0 i g i D (t + 1 t) +
10
N X i=1
0 i k i H (t + 1 t):
(4.8.227)
All equations have been worked out in terms of the Kirchhoff stress. However, the implementation in ABAQUS uses the Cauchy stress. To transform to Cauchy stress, we use the relations
Constitutive Theories
With J J t t =J t , this allows us to write Equation 4.8.29, Equation 4.8.213, Equation 4.8.215, Equation 4.8.218, Equation 4.8.219, and Equation 4.8.227 in the following form:
( +1 ) ( )
k p (t) + i p i (t ) pi (t + 1t) = ikip0 (t + 1t) + i i 0 ; 1J 2 3 ^ S0(t) = SYM 1F 1 S0(t) 1 1F01 ; 2 3 ^ Si(t) = SYM 1F 1 Si(t) 1 1F01 ;
FINITE-STRAIN VISCOELASTICITY
(t + 1t) = 0 (t + 1t) 0
N N X S i (t + 1 t) + X p i (t + 1 t)I ;
(t + 1 t) =
10
N X i=1
i=1 !
i gi
i=1
i=1
The virtual work and rate of virtual work equations are written with respect to the current volume. Therefore, the corotational stress rates are rates of Kirchhoff stress mapped into the current conguration and transformed in the same way as the stresses themselves. This set of equationscombined with the expressions for i , i , and i describe the full implementation of the hyper-viscoelasticity model in a displacement formulation. The rate equations can be written in a form similar to Hyperelastic material behavior, Section 4.6.1. Introduce !
CS = 1 0 v
N X i=1
i gi
!
CS 0
and
Kv =
10
N X i=1
i ki K0 ;
where S and K0 are the instantaneous moduli, corresponding to S and K of Hyperelastic material 0 behavior, Section 4.6.1. Thus, all rate equations can be obtained by substitution of S by S and 0 v K0 by Kv .
The in-plane deformation produces u1 and u2, from which we can calculate only F11, F12, F21, and F22. F13, F23, F31 and F32 are zero. The deformation in the third direction, characterized by F33, is derived from the plane stress condition
33 = 0
or
33 = 0:
(t + 1t)j33 =
i=1 i=1 N N X X ^i ^0 0 i gi D (t)j33 0 i ki H (t)j33 0 i D (t) + H (t) j33 = 0; 0 i i=1 i=1 i=1 N X
10
N X
igi D (t + 1t)j33 + 0
10
N X
iki H (t + 1t)j33 0
(4.8.228)
where j33 stands for the projection along the 33 component. In the derivations it is convenient to and J . express kinematic variables in terms of incremental values, such as
1F
4.8.28
FINITE-STRAIN VISCOELASTICITY
Incompressible materials
1J = det1F = 1 or 1F 33 = (1F 111F 22 0 1F 121F 21)01; ^D ^D where 1F = F (t + 1t), from which 0 (t) and i (t) can be derived.
In this case
t
F, produce
j
D (t + 1t); 0
and then we can solve Equation 4.8.228 directly for H 0
H (t + 1t)j33 = 0 0
10
N X i=1
igi D (t + 1t)j33 0
(t + 1t) 33:
N X N X ^0 ^i i gi D (t)j33 +
i D (t)j33: i=1 i=1
( +1 )
Compressible materials
In this case Equation 4.8.228 becomes an implicit equation in We use the Newton method, for which the rst variation of to be calculated
(t + 1t)j33 =
10
N X
N X i=1
0 i gi D (t + 1t)j33 +
10
^0 i gi D (t)j33 0
N X
^i i D (t)j33;
Similar to Equation 4.8.227, the last two terms vanish, which yields
10
N X i=1
0 igi D (t + 1t)j33 +
10
N X i=1
0 iki H (t + 1t)j33;
(4.8.229)
0 (t + 1t) and H (t + 1t) are obtained directly from the rate-independent constitutive
4.8.29
FINITE-STRAIN VISCOELASTICITY
In the ABAQUS implementation we use Cauchy stresses instead of Kirchhoff stresses. The stresses can easily be mapped by dividing by J . Equation 4.8.228 and Equation 4.8.229 transform into
(t + 1t)j33 =
i k i p 0 (t + 1 t)0 i=1 # N N X X 1 ^ ^ 1J i=1 igiS0(t)j33 0 i=1 i kip0(t) + i=1
i Si(t)j33 0 pi(t) = 0; ! ! N N X X (t + 1t)j33 = 1 0 igi S0 (t + 1t)j33 0 1 0 iki p0(t + 1t): i=1 i=1
=1 " iN X
10
N X
i gi S0(t + 1t)j33 0
10
N X
_ (t + 1t)j33 =
10
N X i=1
i gi
1_ 1F_ 33, 1F_ 33, 1F_ 33, and 1F_ 33, which is again used in Equation 4.8.227 ( +1 )
i=1
Time domain viscoelasticity, Section 10.7.1 of the ABAQUS Analysis Users Manual
4.8.210
4.8.3
Many applications of elastomers involve dynamic loading in the form of steady-state vibration, and often in such cases the dissipative losses in the material (the viscous part of the materials viscoelastic behavior) must be modeled to obtain useful results. In most problems of this class the structure is rst preloaded statically, and this preloading generally involves large deformation of the elastomers. The response to that preloading is computed on the basis of purely elastic behavior in the elastomeric parts of the modelthat is, we assume that the preloading is applied for a sufciently long time so that any viscous response in the material has time to decay away. The dynamic analysis problem in this case is, therefore, to investigate the dynamic, viscoelastic response about a predeformed elastic state. In some such cases we can reasonably assume that the vibration amplitude is sufciently small that both the kinematic and material response in the dynamic phase of the problem can be treated as linear perturbations about the predeformed state. The small amplitude viscoelastic vibration capability provided in ABAQUS/Standard, which is described in Morman and Nagtegaal (1983) and uses the procedure described in Direct steady-state dynamic analysis, Section 2.6.1, is based on such a linearization. Its appropriateness to a particular application will depend on the magnitude of the vibration with respect to possible kinematic nonlinearities (the additional strains and rotations that occur during the dynamic loading must be small enough so that the linearization of the kinematics is reasonable) and with respect to possible nonlinearities in the material response, and on the particular constitutive assumptions incorporated in the viscoelastic model described in this sectionin particular, the assumption of separation of prestrain and time effects described below. In Hyperelastic material behavior, Section 4.6.1, it is shown that the rate of change of the true (Cauchy) stress in an elastomeric material with a strain energy potential is given by
d JS
( ) = J (CS : de + Q d"vol + d
S 0 S 1 d
)
1
(4.8.31)
Constitutive Theories
= 0Q : de 0 K d"vol
(4.8.32)
for the equivalent pressure stress in a compressible material. The various quantities in these equations are dened in Hyperelastic material behavior, Section 4.6.1. For a fully incompressible material all components of are zero and the equivalent pressure stress is dened only by the loading of the structure, so that the second equation is not applicable. For small viscoelastic vibrations about a predeformed state we linearize the additional motions that occur during the vibration so that the differential of a quantity in Equation 4.8.31 and Equation 4.8.32 can be interpreted as the additional incremental value,
d f
for any quantity f , where f jt is the current value of f at some time during the vibration and f j0 is the reference value of f ; that is, f j0 is the value of f at the end of the static (long term) preloading, about which f is uctuating during the vibration.
4.8.31
The incremental elastic constitutive behavior for small added motions dened by this interpretation of Equation 4.8.31 and Equation 4.8.32 is now generalized to include viscous dissipation as well as elastic response in the material, following Lianis (1965), to give
1(J S) = J C 0 : 1e + Q 01" + 1
1 Sj0 0 Sj0 1 1
+
Sj
j
vol
Z t
0
_ 8(j0; t 0 ) : e( ) d
1p = 0Qj0 : 1e 0 K j01" 0
vol
Z t
0
In these expressions f j0 ; t 0 is meant to indicate that f depends on the elastic predeformation ) and is evaluated at time t 0 , that has occurred prior to the small dynamic vibrations (the state at t between the start of the vibrations and the current time, t. and are the functions that dene the viscous part of the materials response: the notation is intended to imply that these are functions of the elastic def df=dt is the time rate of change of a quantity. predeformation and time. f The denitions of the viscous parts of the behavior, and , provided in ABAQUS are simplied by assuming that this viscous behavior exhibits separation of time and prestrain effects; that is, that
=0
_=
8 (j 0 ; t 0 ) = g (t 0 )C S j0
and where S j0 and K j0 are the effective elasticity of the material in its predeformed state, prior to the vibration. This assumption simply means that measurements of the viscous behavior during small motions of the material about a predeformed state depend only on the predeformation to the extent that the effective elasticity of the material also depends on that predeformation. There is experimental evidence that this simplication is appropriate for some practical materials (see Mormans (1979) discussion). With this assumption the denition of the viscous part of the materials behavior is reduced to nding the scalar functions of time, g and k (only g for fully incompressible materials), and the constitutive response to small perturbations is simplied to
(j 0 ; t 0 ) = k (t 0 )K j 0 ;
1(J S) = J C 0 : 1e +
Sj
Z t
0
g(t 0 )_ ( ) d e
+ Qj01" + 1
1 Sj0 0 Sj0 1 1
vol
1p = 0Qj0 : 1e 0 K j0 1" +
vol
Z t
0
k (t 0 )" _
vol
( ) d
In ABAQUS this model is provided only for the direct-solution and subspace-based steady-state dynamic analysis procedures, in which we assume that the dynamic response is steady-state harmonic vibration, so that we can write
f j t 0 f j0 =
where < f i= f is the complex amplitude of a variable f . Dening the Fourier transforms of the viscous relaxation functions g t and k t as
(1 ) + (1 )
Z 1
()
()
01
Z 1
g t
01
k t
allows the constitutive model to be written for such harmonic motions in the linear form
+ Qj0<(1" ) + <(1
) 1 Sj0 0 Sj0 1 <(1
)
and
9
+ Qj0=(1" ) + =(1
) 1 Sj0 0 Sj0 1 =(1
) ;
and, for compressible materials,
80
9 9
functions of frequency. When the pure displacement formulation is used for a compressible material, the virtual work equation for dynamic response is
WI
=(1p) = 0Qj0 : =(1e) 0 K j0 : !<(k)<(1"vol ) + 1 0 !=(k) =(1"vol ) : The viscous behavior of the material is, thus, reduced to dening <(g), =(g), <(k), and =(k) as 0 WD 0 WE = 0;
Constitutive Theories
(4.8.33)
where
WI
V
Z
: D dV u 1 u dV 0
=0
is the virtual work of the dAlembert forces (0 is the mass density of the material in the original conguration), and Z Z
WE
0
u 1 p dS +
4.8.33
u 1 f dV
is the virtual work of externally prescribed surface tractions per current surface area and body forces per current volume. For the linearized perturbations considered here we recast Equation 4.8.33 in incremental form, giving
where
1WI is obtained from Equation 4.6.112 with the interpretation d(f ) ! 1(f ); 1WD def WD = 1WE =
Z
(4.8.34)
and
+
where
SZ
u 1 (Pu 1 1u + pp1p) dS
Pu = J1 @(JA p) ; A @u
JA
u 1 (Fu 1 1u + ff 1f ) dV;
in which
dA dA0
pp = @ p ; @p 1 Fu = J @(Juf ) ; @ @f ff = @f ; and p and f are the externally prescribed tractions, so that 1p and 1f are externally prescribed traction increments. Note that 1WE includes terms dependent on 1u: these give rise to the load stiffness
matrix when the nite element interpolations are introduced. When the motion is harmonic we can recast these quantities as
1WI =
b <( e) =( e) <("vol ) =("vol ) c 2 3 e e < C = C 0 7 8 Qj0 6 <(1e) 9 6 7 > e e = < 0< C 0 0 j0 7 =(1e) > Q 7 6= C 6 :6 : > <(1"vol) > e e 6 Q j0 0 < K = K 7 : ; 7 4 5 =(1"vol ) e e 0 0Qj0 = K 0< K <(1e) 0 b <( L) =( L) c : <(1L) dV; 0 S0 : 2 b <( e) =( e) c : =(1e) =(1L)
4.8.34
where
0 1 e < C = 1 0 ! =(g) CS j0 ;
1WD = 0!2
and
Z
V0
0
dV 0 ;
Pu 0 1 <(1u) dS 1WE = b <(u) =(u) c 1 0 Pu =(1u) SZ + b <(u) =(u) c 1 F0u F0u 1 <(1u) dV =(1u) ZV <(1p) dS + Z b <( u) =( u) c 1 f <(1f ) dV: + b <(u) =(u) c 1 pp =(1p) f =(1f )
S
In these expressions < and = are understood to be independent variations. Thus, when the nite element displacement interpolations are introduced into Equation 4.8.34, we obtain a linear, frequency-dependent system that can be solved at each frequency for the real and imaginary parts of the nodal degrees of freedom of the model. In like fashion, the augmented variational principles for almost incompressible behavior and for fully incompressible behavior described in Hyperelastic material behavior, Section 4.6.1, can be used to obtain linear, frequency dependent systems for harmonic viscoelastic vibration problems. The steady-state dynamic analysis procedure based on modal superposition cannot be used because the viscous behavior assumed does not correspond to a small amount of Rayleigh damping, which is a requirement for steady-state harmonic response based on modal superposition.
Reference
( u)
( u)
Constitutive Theories
Frequency domain viscoelasticity, Section 10.7.2 of the ABAQUS Analysis Users Manual
4.8.35
HYSTERESIS
4.9.1
HYSTERESIS
The hysteretic, elastomeric material model provided in ABAQUS/Standard (Bergstrm and Boyce, 1997) is intended for modeling large-strain, rate-dependent behavior of elastomers. A primary aspect of such behavior is a pronounced hysteresis in stress-strain curves under cyclic loading. The material model is intended to model response under many load cycles. It does not model the Mullins effect, which refers to the signicant softening that is observed in the material response of elastomers in the rst few load cycles. The model decomposes the mechanical response of elastomers into an equilibrium network, A, corresponding to the state that is approached in long-time stress relaxation tests, and a time-dependent network, B, that captures the nonlinear rate-dependent deviation from the equilibrium state. The time dependence of the second network is assumed to be governed by the reptational motion of molecules having the ability to change conformation signicantly, thereby relaxing the overall stress state. The total stress is assumed to be the sum of the stresses in the two networks. The deformation gradient, F, is assumed to act on both networks and is decomposed into elastic and inelastic parts in network B according to the multiplicative decomposition
F = Fe Fcr B B
1
The material model is dened completely by specifying an isotropic, hyperelastic material model that characterizes the response of network A; a stress-scaling factor, S , that denes the ratio of the stress carried by network B to that carried by network A under identical elastic stretching; a positive exponent, m, generally greater than 1, characterizing the effective stress dependence of the effective creep strain rate in network B; an exponent C , restricted to lie in [01; 0], characterizing the creep strain dependence of the effective creep strain rate in network B; and a scaling constant A in the expression for the effective creep strain rate to maintain dimensional consistency in the equation. The mechanical response of network A under the imposed deformation gradient F is governed by standard isotropic hyperelasticity. The stress response of network B is dependent only on Fe and is B governed by the same hyperelastic potential as network A, up to the stress-scaling factor, S . Given F, the determination of Fe requires a constitutive specication for Fcr , which is provided through an evolution B B equation for Fcr given by B
_B where cr is the effective creep strain rate in network B,
Constitutive Theories
B, and B
q3
2
(4.9.11)
where cr 0 1 is the nominal creep strain in network B. cr , the chain stretch in network B, is dened as B B
cr B =
1 3
I : Ccr ; B
4.9.11
HYSTERESIS
where cr = cr T 1 cr . The correlation of the microscopic chain stretch to the principal macroscopic B B B stretch state through the above formula comes from Arruda and Boyce (1993). The numerical implementation of the material response of network A is identical to that for hyperelastic models in ABAQUS/Standard, documented in Hyperelastic material behavior, Section 4.6.1. The implementation of the mechanics of network B is based on an implicit exponential approximation to the ow rule, Equation 4.9.11, which preserves the property of inelastic incompressibility of the ow rule exactly (Weber and Anand, 1990). The choice of the exponential integrator along with the isotropy of the stress-response function result in the principal directions of the right elastic stretch tensor being equal to those of the right stretch tensor for a predictor to the elastic deformation gradient. This predictor is a known quantity in the discrete integration that is dependent only on the deformation gradient at the end of the increment and the plastic deformation gradient at the end of the previous increment. The complete state at the end of the increment can be obtained from the calculation of the elastic stretches and the effective creep strain increment. All other quantities required for the state update are known. This material update formula is linearized exactly in closed form and leads to an unsymmetric tangent stiffness matrix caused by the nominal creep strain dependence of the constitutive equation for the effective creep strain increment.
Reference
4.9.12
SMALL-SLIDING CONTACT
5.1.1
In ABAQUS/Standard a capability is included to model small-sliding contact of two bodies with respect to each other. With this formulation the contacting surfaces can undergo only relatively small sliding relative to each other, but arbitrary rotation of the bodies is permitted. Small-sliding contact is computationally less expensive than nite-sliding contact, which is described in Finite-sliding interaction between deformable bodies, Section 5.1.2. The small-sliding capability can be used to model the interaction between two deformable bodies or between a deformable body and a rigid body in two and three dimensions. With this approach one surface denition provides the master surface and the other surface denition provides the slave surface. A kinematic constraint that the slave surface nodes do not penetrate the master surface is then enforced. The contacting surfaces need not have matching meshes; however, the best accuracy is obtained when the meshes are initially matching. For initially nonmatching meshes, accuracy can be improved by judiciously specifying initial adjustments to ensure that all slave nodes that should initially be in contact are located on the master surface. The small-sliding contact capability is implemented by means of four internal contact elements designed to handle the following kinematic constraints: 1. two-dimensional contact between a slave node and a deformable master surface, 2. two-dimensional contact between a slave node and a rigid master surface, 3. three-dimensional contact between a slave node and a deformable master surface, and 4. three-dimensional contact between a slave node and a rigid master surface. These elements are not accessible to the user, and ABAQUS will automatically cover a slave surface with the appropriate element type, based on the nature of the corresponding master surface.
Identication of tangent plane based on nearest neighbor interaction
Although four internal element types are used to model the various types of small-sliding contact interactions supported by ABAQUS/Standard, all four formulations are based on the notion that a given slave node always interacts with the same subset of master surface nodes. This nodal subset is initially determined by the ABAQUS analysis input le processor from the undeformed model denition, thus avoiding the need to track the slave node during the course of the analysis. This set of nearest-neighbor nodes to the point on the master surface closest to the slave node is used to parametrize a contact plane with which the slave node will interact during the analysis. This concept is illustrated next for the case of a two-dimensional slave node interacting with a rst-order master surface. This formulation can be generalized to second order as well as three-dimensional situations, but this generalization will not be discussed here. Consider the contact interaction of three nodes102, 103, and 104on the slave surface with a master surface made up of rst-order element faces described by nodes 1, 2, and 3, as shown in Figure 5.1.11.
Interface Modeling
5.1.11
SMALL-SLIDING CONTACT
Figure 5.1.11 Slave nodes interacting with a two-dimensional master surface. Before initiating the search for the nodal subset of the master surface nodes that will interact with each node on the slave surface, unit normal vectors are computed for all the nodes on the master surface. For example, the unit normal vector 2 is computed by averaging the unit normal vectors of segments 12 and 23. The user can also specify the normal vector for each node on the master surface. Additional unit normal vectors are computed for each segment a distance "L from each end of the segment, where " is a fraction and L is the length of the segment; e.g., " . Currently the value of " is set to 0.5, and the user cannot change this value. The unit normals computed are then used to dene a smooth varying normal vector, ( ), at any point, , on the master surface. An anchor point on the master surface, 0 , is computed for each slave node so that the vector formed by the slave node and 0 coincides with the normal vector ( 0 ). Suppose that a search for the anchor point, 0 , of slave node 103 reveals that 0 is on segment 12. Then, we nd that
NX
X1 and X2 are the coordinates of nodes 1 and 2, respectively, and u0 is calculated so that X0 0 X103 coincides with N(X0). Moreover, the contact plane tangent direction, v0, at X0 is chosen so that it is perpendicular to N(X0 ); i.e., v 0 = N (X 0 ) 2 e z = T 1 @ X (u 0 ) = T 1 (X 2 0 X 1 ) ;
where where
NX
@u
5.1.12
SMALL-SLIDING CONTACT
The small-sliding contact constraint is achieved by requiring that slave node 103 interact with the tangent plane whose current anchor point coordinates are, at any time, given by
x 0 = N 1 (u 0 )x 1 + N 2 (u 0 )x 2 ;
where
N 1 (u 0 )
=1
u0
and
N 2 (u 0 )
u0
v = T1(
u N 1 (u 0 )
x1 +
u N 2 (u 0 )
x2 ) x
u u where N1 (u0) = 01 and N2 (u0 ) = 1. Since the above expressions for the point 0 and the vector resulted from barycentric (afne) combinations of the points 1 and 2 that is,
N 1 (u 0 ) N1
N 2 (u 0 )
= 1;
(u 0 ) + N 2 (u 0 ) = 0 ;
the contact plane will be mapped properly under afne transformations such as translation, scaling (stretching), and rotation. Next, suppose that a search for the anchor point of slave node 104 reveals that the anchor point is coincident to the master node 2. In this case the anchor point is chosen to be 2 , or in terms of the coordinates of the three master nodes 1, 2, and 3,
X0 =
where
N 1 (u 0 )
N 1 (u 0 )
X1 +
N 2 (u 0 )
X2 +
z :
N 3 (u 0 )
X3
= 0,
N 2 (u 0 )
= 1,
and
N 3 (u 0 )
= 0.
However, we want to express 0 in terms of the coordinates of the three nodes 1, 2, and 3 to be able to track the evolution of the tangent plane. To this end, we solve for the Niu from the equation
u N 1 (u 0 )
v0 = N 2 2 e X1 +
u N 2 (u 0 )
X2 is simply
X2 +
+
u N 3 (u 0 )
X 3 = v0
u N 2 (u 0 )
u N 3 (u 0 )
= 0:
The barycentric constraint ensures that the resulting expression for the contact plane tangent direction behaves properly under afne transformations such as translations and rotations.
Contact formulation as a constrained variational principle
Interface Modeling
At each slave node that can come into contact with a master surface we construct a measure of overclosure h (penetration of the node into the master surface) and measures of relative slip si . These kinematic measures are then used, together with appropriate Lagrange multiplier techniques, to introduce surface interaction theories for contact and friction, as described in Contact pressure denition, Section 5.2.1, and Coulomb friction, Section 5.2.3.
5.1.13
SMALL-SLIDING CONTACT
In two dimensions the overclosure h along the unit contact normal n between a slave point N +1 and a master line ( ), where parametrizes the line, is determined by nding the vector ( 0 N +1 ) from the slave node to the line that is perpendicular to the tangent vector at . Mathematically, we express the required condition as
v p
x p x
n = p() 0 xN +1 ; n
(5.1.11)
when
v 1 (p() 0 xN +1 ) = 0 : x v p x p p
Similarly, in three dimensions the overclosure h along the unit contact normal between a slave point N +1 and a master plane (1; 2), where i parametrize the plane, is determined by nding the vector ( 0 N +1 ) from the slave node to the plane that is perpendicular to the tangent vectors 1 and 2 at . Mathematically, we express the required condition as
n = p(1; 2) 0 xN +1 ;
(5.1.12)
when
If at a given slave node h < 0, there is no contact between the surfaces at that node, and no further surface interaction calculations are needed. If h 0, the surfaces are in contact. The ~ contact constraint h = 0 is enforced by introducing a Lagrange multiplier, p, whose value provides the contact pressure at the point. To enforce the contact constraint, we need the rst variation h; and for the Newton iterations, we need the second variation, dh. Likewise, if frictional forces are to be transmitted across the contacting surfaces, the rst variations of relative slip, si , and the second variations, dsi , are needed in the formulation. The derivation of some of these quantities is described next for all four small-sliding contact formulations.
Formulation for two-dimensional small-sliding deformable contact
For the case of two-dimensional small-sliding deformable contact, a point on the contact line associated with a slave node nN +1 is represented by the vector
p ( ) = x 0 + v ; (5.1.13) where, as described previously, the lines anchor pointx0 and its tangent vectorvare functions of the current master node coordinates, x1 . . . xN . Hence, the vector v is, in general, a nonunit vector.
Linearization of Equation 5.1.11 yields
h
n + hn = x0 + v + v 0 uN +1 (5.1.14) where uN +1 = (XN +1 + uN +1 ) = xN +1 , x0 = Ni (u0 ) xi , and v = R 1 Niu(u0 ) xi . Taking the dot product of Equation 5.1.14 with n results in the following expression for h:
5.1.14
SMALL-SLIDING CONTACT
h
= 0n 1 ( uN +1 0 x0 0 v) :
(5.1.15)
t = v=kvk and setting h = 0 results in the following expression for the variation in slip: def s = v 1 t = t 1 ( uN +1 0 x0 0 v) :
Likewise, taking the dot product of Equation 5.1.14 with the normalized tangent vector
(5.1.16)
Suitable expressions for dh and ds can be derived by linearizing Equation 5.1.14 and applying the techniques highlighted above. Since the resulting expressions do not provide additional insight into the understanding of this capability, they will not be presented here.
Formulation for three-dimensional small-sliding deformable contact
The three-dimensional small-sliding deformable contact formulation is a straightforward generalization of the previous two-dimensional formulation. A point on the contact plane associated with a slave node nN +1 is represented by the vector
p(1; 2) = x0 + 1v1 + 2v2 ; where the planes anchor pointx0 and its two tangent vectorsv1 and v2 are functions of the current master node coordinates x1; . . . ; xN . Linearization of Equation 5.1.12 yields hn + h n = x0 + 1 v1 + 1 v1 + 2v2 + 2 v2 0 uN +1 ; (5.1.17) where x0 = f ( x1 ; . . . ; xN ) and vi = gi( x1 ; . . . ; xN ). Taking the dot product of Equation 5.1.17 with n results in the following expression for h: h = 0n 1 ( uN +1 0 x0 0 1 v1 0 2 v2) : (5.1.18) Likewise, taking the dot product of Equation 5.1.17 with t1 = v1 =kv1 k and setting h = 0
results in the following expression for the variation of the rst slip component:
s1 = 1
(5.1.19) v1 1 t1 = t1 1 (uN +1 0 x0 0 1v1 0 2v2) : Similarly, taking the dot product of Equation 5.1.17 with t2 = v2 =kv2 k and setting h = 0 results in the following expression for the variation of the second slip component:
s2 = 2
def
def
(5.1.110)
Interface Modeling
The formulation for two-dimensional small-sliding rigid contact follows from its deformable counterpart by exploiting the fact that the evolution of the contact plane is fully determined by the motion of the rigid bodys reference node. Figure 5.1.12 shows how the undeformed coordinates 0
5.1.15
SMALL-SLIDING CONTACT
u rs rs R
X rs
X0 t0
n0
Figure 5.1.12
of the contact planes anchor point are related vectorially to the undeformed coordinates of the rigid reference node, r s, and the relative position vector .We can express this relationship as
R X0 = Xrs + R
Suppose the rigid reference node undergoes a motion described by the displacement vector rs and the rotation vector rs z , then the current coordinates of the contact planes anchor point are given by
is the orthogonal matrix that produces the rotation rs z (see Rotation variables, where Section 1.3.1) and is the current position vector from the rigid reference node to the anchor point. The rotation matrix is also used to obtain the contact planes current tangent and current normal as follows:
(5.1.111)
where 0 and 0 are the initial contact tangent and normal at 0 , respectively, as shown in Figure 5.1.12. By denition, a rigid surface cannot stretch; therefore, a point on the rigid contact line associated with a slave node nN +1 is represented by the vector
t = C(rsez ) t0 ; n = C(rsez ) n0 ;
1 1
(5.1.112)
p ( ) = x 0 + t ;
5.1.16
SMALL-SLIDING CONTACT
whereunlike the vector v in Equation 5.1.13the tangent t is always a unit vector. Linearization of Equation 5.1.111 and Equation 5.1.112 results in the following expressions for the rst variations of the anchor coordinates and the contact planes tangent:
= urs + (0ry ex + rx ey )rs ; (5.1.113) t = (0ty ex + tx ey )rs : Replacing v by t in Equation 5.1.15, noting that t 1 t = 0, and substituting for x0 and t
x0
0n 1
uN +1
0 urs +
2
(r + t )e
rs :
(5.1.114)
Similar manipulation of Equation 5.1.16 yields the following expression for s:
s = =
def
(5.1.115)
The three-dimensional small-sliding rigid contact formulation can also be derived from its deformable counterpart by generalizing some of the expressions that were introduced in the previous section. In particular, in this formulation the rigid reference node can undergo an arbitrary nite rotation described by the rotation vector rs . Consequently, Equation 5.1.113 for the variations of the anchor point coordinates and the contact planes tangent generalize to
b x0 = urs + rs b ti = rs ti ;
1r;
i = 1 ; 2;
b where rs is the skew-symmetric matrix associated with the linearized rotation rs , as explained in Rotation variables, Section 1.3.1. Replacing vi by ti in Equation 5.1.18 through Equation 5.1.110 and substituting for x0 and ti from above results in the following expressions for h, s1 , and s2:
h =
0n 1
uN +1
b 0 urs 0 rs 1 (r + 1 t1 + 2 t2)
def def
Interface Modeling
Reference
Contact formulation for ABAQUS/Standard contact pairs, Section 21.2.2 of the ABAQUS Analysis Users Manual
5.1.17
5.1.2
ABAQUS/Standard provides two formulations for modeling the interaction between two deformable bodies. The rst is a small-sliding formulation in which the contacting surfaces can undergo only relatively small sliding relative to each other but arbitrary rotation of the surfaces is permitted. This formulation is discussed in Small-sliding interaction between bodies, Section 5.1.1. The second is a nite-sliding formulation where separation and sliding of nite amplitude and arbitrary rotation of the surfaces may arise. The formulation for two-dimensional and axisymmetric analysis, as well as for tube-in-tube analysis, is discussed in this section. Depending on the type of contact problem, two approaches are available to the user for specifying the nite-sliding capability: (1) dening possible contact conditions by identifying and pairing potential contact surfaces and (2) using contact elements. With the rst approach ABAQUS automatically generates the appropriate contact elements. In axisymmetric problems with asymmetric deformations, ISL21A and ISL22A elements can be used to model contact with CAXA or SAXA elements. Sliding of tubes inside each other can be modeled with ITT21 and ITT31 elements. To dene a sliding interface between two surfaces, one of the surfaces (the slave surface) is covered with ISL or ITT elements. The other surface (the master surface) is dened as a slide line surface composed of a series of nodes ordered in sequence. The slide line itself can consist of linear or quadratic segments. If smoothing is used, these segments are connected with quadratic or cubic segments such that full slope continuity is achieved. The smoothing procedure is described later in this section.
Two-dimensional and axisymmetric slide line elements
Consider contact of a node on the slave surface n1 with a segment of the master surface described by nodes n2 , n3, . . ., where the number of nodes depends on the order of the segment. For a linear segment the number of nodes is 2, whereas for a quadratic segment the number of nodes is 3. For a smoothed section of a linear slide line, the number of nodes is also 3; and for a smoothed section of a quadratic slide line, the number of nodes is 5. If the contact occurs at the (convex) vertex of two segments, only a single node will enter the equations. A typical linear segment is shown in Figure 5.1.21, and a quadratic segment is displayed in Figure 5.1.22. Smoothed segments are shown later in this section. To derive the equations governing these elements, we consider the coordinates in the plane of the slide line. For the axisymmetric elements, this plane coincides with the two-dimensional space. First, we determine the point x on the segment closest to the point x1 on the slave surface. We also determine the normal n and tangent t to the segment at that point. The point x and the normal n can be related to the overclosure h with the relation
Interface Modeling
n h = x 0 x1 :
5.1.21
(5.1.21)
X1 n t X2 X X3
Figure 5.1.22 Quadratic slide line segment. Since x is on the segment, its position is dened completely by the interpolation function Ni for the segment, the position g, and the position xi of the nodes ni that are part of the segment. That allows us to write for Equation 5.1.21
where
N2
N1
tangent
= 2 (1 g ), N3 = 2 (1 + g). For a quadratic segment you use 1 g(g + 1). Similar expressions are obtained for smoothed N4 = 2
are functions of g. For instance, for a linear segment you obtain 1 g(g 0 1), N = 1 0 g2 , N2 = 3 2 segments of the slide line. The to the slide line at point x follows with
01 and
n h = N i (g )x i ;
N2
(5.1.22)
N3
...
t def =
where
ds d
dg dN i dg
dg
(5.1.23)
The position of point x is determined from the condition that the normal and tangent must be orthogonal, which leads to the following equation for g:
dg
xi :
n 1 t = N i (g )
d N j (g ) dg
xi 1 xj = 0 :
5.1.22
For linear segments this yields a linear equation, which can be solved directly. For quadratic and cubic segments it leads to third- or fth-order equations, which must be solved iteratively. The equation is solved using Newtons method preceded by a number of bisections to nd the true minimum distance solution. To obtain the contact/slip equation, the position equation (Equation 5.1.22) is linearized. This linearization yields
nh + nh =
dN i dg
xi g + Ni xi = ts + Ni xi ;
n 1 xi ;
(5.1.24)
0N t 1 x 0 h t 1 n:
i i
It is assumed that slip is relevant only if the node x1 is on the slide line; hence, it will be assumed that h = 0. From this follows
s =
0N t 1 x :
i i
(5.1.25)
To obtain the initial stress stiffness terms, the second variations of h and s must be calculated. From Equation 5.1.24 follows
dh =
xi 1 n
dN i dg d
dg + N i
x i 1 dn :
(5.1.26)
dg =
N 0 dds N t 1 dx :
i j j
(5.1.27)
n = (n n + t t) 1 dn = t t 1 dn = 0t n 1 dt:
(5.1.28)
In this equation
n 1 dt = n 1 d
=
dg
d
dg
=
n 1 xi
d2 N i dg 2
dg +
dN i dg ds
xi
d
dg
x
(5.1.29)
Interface Modeling
0
n ds
dN i ds
n 1 dx i = n N i t 1 dx i +
dN i
n 1 dxi ;
where use was made of Equation 5.1.25 and n (the segment curvature) is dened as
5.1.23
n
def
= 0n
d2
dg 2
dg
x 2
(5.1.210)
For straight segments n obviously vanishes. Substitution of Equation 5.1.27 to Equation 5.1.29 in Equation 5.1.26 yields the nal result:
dh =
0 x 1 n ds
i
dN i
N j t + tN i
dN j ds
n + tN i n N j t 1 d x j :
(5.1.211)
This expression is symmetric, as should be expected. The second variation in s can be derived along similar lines:
ds =
N 0 x 1 t ddg dg 0 N x 1 dt:
i i i i
(5.1.212)
Note that
dt = ( n
n + t t ) 1 d t = n n 1 d t:
(5.1.213)
Substitution of Equation 5.1.27, Equation 5.1.29, and Equation 5.1.213 in Equation 5.1.212 yields
ds =
xi 1 t
dN i ds
Nj t
0 nN
dN j ds
n 0 n Ni n N j t 1 dx j :
(5.1.214)
This expression is nonsymmetric. The second variations of h and s vanish if no slip in the slide plane occurs (ds = s = 0):
Tube-tube interface elements
In the case of tube-tube interface elements it is assumed that the inner tube can be considered the slave surface and the outer tube the master surface. The tube-tube interface elements differ from the axisymmetric slide line elements in two ways. In the rst place there is assumed to be a nite clearance between the tubes, which has the effect that the separation distance h is nite even when contact occurs. In the second place for the three-dimensional element ITT31 there is a second possible slip direction in the plane of the cross-section of the tubes. The derivations for the ITT elements follow much along the same line as the derivations for the ISL elements. The contact equation can be written in the form
0 n (h + h ) = x 1 0 x :
o
(5.1.215)
Here x is a point on the outer tube that potentially contacts point x1 on the inner tube, as shown in Figure 5.1.23.
5.1.24
X1
Figure 5.1.23 Tube-tube contact. In this equation ho is the (positive) radial clearance between the tubes. In a similar way as was done for the ISL elements, the contact equation can be written in the form
n(h + ho ) = 0Ni(g) xi :
(5.1.216)
The sign reversal as compared to Equation 5.1.22 is related to the internal nature of the contact as compared to the external nature for the regular slide line elements. The linearized form of Equation 5.1.216 is
n(h + ho ) + nh =
0ts 0 N x :
i i
1
As before, we assume that during contact h = 0. In the contact direction this yields
h =
0N n x ;
i i i i
(5.1.217)
With use of
n 1 t = 0, it readily follows
0h t 1 n 0 N t 1 x :
o
(5.1.218)
t 1 n = 0n 1 t = n s 0
which transforms Equation 5.1.218 into
s =
1 1 + n ho (
dN i ds
n 1 xi ; n 1 xi ) ;
(5.1.219)
Interface Modeling
0N t 1 x
i
+ ho
dN i ds
(5.1.220)
where n is the segment curvature as dened by Equation 5.1.210. For the three-dimensional tubedef tube interface element, one can dene the transverse slip direction s = t 2 n, which yields
5.1.25
s2
= 0N
s 1 xi (= ho s 1 n):
(5.1.221)
The initial stress stiffness terms are again obtained by taking the second variations in ho , s, and
s2 . From Equation 5.1.217 follows
dh = n sds
(5.1.222)
N 0 dds n 1 dx :
i i
(5.1.223)
0s (1 + h
o i i
n)
dN i ds
n dxi
N 0x 1 n dds (1 + h
+ho xi +
o n )ds
(5.1.224)
s2 ds2 ho
1 n ds
dN i dN j ds
n dxj
This expression is symmetric. In the t-direction the virtual work term has the form
Ft (s + ho t n) = Ft s;
def
0N t 1 x :
i i i i
(5.1.225)
N 1 0 t 1 x dds
d2 x dg 2
ds
(5.1.226)
Note that
dt = d = =
dx ds
dx dg
dN i ds dN i ds
(I
0 t t) 1
2 dx d2 x dx i + 2
dg
0 dg
nn
1 dg
d2 x
3 dx dg 2
dg
dg
dg
1 (I 0 t t)ds
(5.1.227)
(n n + s s ) d x i
0 (
+ s s)ds;
5.1.26
s def 0s 1 =
d2
In this expression the terms involving s vanish for element type ITT21. Equation 5.1.220 and Equation 5.1.227 in Equation 5.1.226 yields
Ni s 1 dx i ds dN i 0 xi 1 t ds ds N 0 n sdh + xi 1 n ddsi dh N 0 n hds + h ddsi n 1 dxi In the s-direction one simply looks at the rst variation in s2 :
x dg 2
x 2 : dg
(5.1.228) Substitution of
s = s 1 0 s ds +
(5.1.229)
Ni ds 0 N i ds : ds2 = xi 1 0s ds
d
(5.1.230)
Note that
d =
n n 1 ds + t t 1 ds = 0n s 1 dn 0 t s 1 dt:
s2 = 0 xi 1 s
(5.1.231)
At the junction of two segments along a slide line, a discontinuity in slope may occur. This discontinuity can cause convergence problems since during iteration a contact point might move back and forth between two segments. Hence, it is useful to smooth the transition between segments. Consider rst the transition between two linear segments (Figure 5.1.24). The junction of the two segments is connected by a Hermitian polynomial between the points xa and xb , which are located on the segments:
Interface Modeling
xa = x2 + (1 0 )x3 ; xb = x4 + (1 0 )x3 :
5.1.27
X2
X1 X4 Xa X3
Figure 5.1.24 Transition between linear segments.
X Xb
The positions of xa and xb are, hence, determined by the smoothing factor , which is specied directly by the user in the range 0 0:5. We choose g as the parameter coordinate on the smoothed segment, with 01 g 1. At the extreme values for g the coordinates are xa and xb , and for the coordinate derivatives we choose
dx a d
x3 0 xa = (x3 0 x2 );
dx a d
With the Hermitian interpolation functions we can calculate the position as a function of g:
x = (g3 0 3g + 2)xa +
1 + ( 4 4 1
1 4
g3 0 g2 0 g + 1)
1 4
dx a d
0g3 + 3g + 2)xb
+ 1 4 (
g3 + g2 0 g 0 1)
dx b d
x=
g2 0 2g + 1)x2 +
(4
(5.1.232)
Note that the smoothing has in fact been done with a second-order polynomial. In the formulation of the interface contact and friction equations, the treatment of a smoothing segment is identical to the treatment of a regular segment. A transition between quadratic segments is shown in Figure 5.1.25. It is readily established that in this case
= =
0 d x a = ( 0 4 2 )x 2 + ( 0 4 + 8 2 )x 3 + ( 3 0 4 2 )x 4 d
dx b d
=(
0 + 4 2 )x 6 + ( 4 0 8 2 )x 5 + ( 0 3 + 4 2 )x 4 :
5.1.28
X2 X3 Xa X Xb X4
Figure 5.1.25 Transition between quadratic segments.
X6 X1 X5
As before, is specied directly by the user. Substitution of these equations in the Hermite shape functions yields
x=
10 4
[02g3 + ( 4 0 1)g2 + (02 + 2)g 0 1]x2 3 2 + [ 4g + (08 + 4)g + ( 4 0 8)g + 4]x3 2 2 + [(8 0 6)g + (06 + 4)]x4 3 2 + [04g + (08 + 4)g + (04 + 8)g + 4]x5 1 3 2 + [ 2g + (+4 0 1)g + ( 2 0 2)g 0 1]x6 :
(5.1.233)
If x3 = 1 (x2 + x4 ) and x5 = 1 (x4 + x6 ), the equations reduce to the same equations as were used 2 2 for smoothing between linear segments.
Self contact
Self contact is available in the context of a surface folding and touching itself. Appropriate contact elements are generated internally for each node on the surface. A node is allowed to contact all of the surface segments, with the exception of the segments that are adjacent to the node. Since a node is simultaneously a master and a slave (producing symmetric master-slave relationships), overconstraints would occur, for example, if the meshes on both sides of the interface matched exactly. If only a pair of nodes matches in two dimensions, no problem occurs due to the smoothing carried out on the master side of the interface. To avoid solver problems due to overconstraining, three-dimensional self-contact is activated only if penalty-type contact is used. The mathematical treatment of contact elements is the same as described above, with a few modications in two dimensions. When two adjacent segments of the surface fold forming a sharp crack, the contact algorithm becomes pure master-slave instead of symmetric to prevent redundant
Interface Modeling
5.1.29
contact constraints. It has been arbitrarily chosen that of these two segments the shortest is the slave and the longest is the master.
References
Contact formulation for ABAQUS/Standard contact pairs, Section 21.2.2 of the ABAQUS Analysis Users Manual Tube-to-tube contact elements, Section 23.3.1 of the ABAQUS Analysis Users Manual Slide line contact elements, Section 23.4.1 of the ABAQUS Analysis Users Manual
5.1.210
5.1.3
ABAQUS/Standard provides two formulations for modeling the interaction between a deformable body and an arbitrarily shaped rigid body that may move during the history being modeled. The rst is a smallsliding formulation in which the contacting surfaces can only undergo relatively small sliding relative to each other, but arbitrary rotation of the surfaces is permitted. This formulation is discussed in Small-sliding interaction between bodies, Section 5.1.1. The second is a nite-sliding formulation where separation and sliding of nite amplitude and arbitrary rotation of the surfaces may arise. This formulation is discussed in this section. The nite-sliding rigid contact capability is implemented by means of a family of contact elements that ABAQUS automatically generates based on the data associated with the user-specied contact pairs. At each integration point these elements construct a measure of overclosure (penetration of the point on the surface of the deforming body into the rigid surface) and measures of relative shear sliding. These kinematic measures are then used, together with appropriate Lagrange multiplier techniques, to introduce surface interaction theories (contact and friction). A library of interaction theories is provided in ABAQUS these may be thought of as a library of surface constitutive models. In this section we discuss only the kinematics of the interacting surfaces. The surface constitutive models are described in Chapter 4, Mechanical Constitutive Theories. Let A be a point on the deforming mesh, with current coordinates xA . Let C be the rigid body reference nodethe node that denes the position of the rigid bodywith current coordinates xC . Let A0 be the closest point on the surface of the rigid body to A at which the normal to the surface of the rigid body, n, passes through A. Dene r as the vector from C to A0. The geometry described by these quantities is shown in Figure 5.1.31.
A r n A C
Interface Modeling
5.1.31
to
A along 0n:
Then if h < 0c there is no contact between the surfaces at A, and no further surface interaction calculations need be done at this point. Here c is the clearance below which contact occurs. For a hard surface c = 0, but ABAQUS/Standard also allows a softened surface to be introduced in which c may be nonzero (although c is usually very small compared to other dimensions). If h 0c the surfaces are in contact. To enforce the contact constraint we will need the rst variation of h, h, and its second variation, dh. These quantities are now derived. Let S , = 1; 2 be locally orthogonal, distance measuring surface coordinates on the surface at 0 A . The S measure distance along the tangents t to the surface at A0 : these tangents are constructed according to the standard ABAQUS convention for such tangents to a surface in space. As the point A and the rigid body move, the projected point A0 will move along. The movement consists of two parts: movement due to motion of the rigid body and motion relative to the body
nh = 0xA + xC + r:
xA
xC + C 2 r + t
;
surface
where is the slip of point and due to slip along the surface
The normal
nh + h(C 2 n +
For hard contact h = 0 exactly, and for soft contact we will assume kinematic equation, thus, becomes
To obtain the second = 0. In addition, it will be assumed that dh = h = 0, which is accurate for relatively hard contact. It then directly follows that
= t 1 (xA 0 xC ) 0 (r 2 t) 1 C : variation of h, it will again be assumed that h ndh = dr;
5.1.32
ndh = d(C 2 r)j + C 2 drjC + dtj + dtjC + t d ; where we have used dxA = d xC = dC = 0. The rst term corresponds to a second-order variation on the vector r for rigid body rotations around point C and is given by (see Rotation variables, Section 1.3.1): d(C 2 r)j = C 1 dC r 0 1 C dC 1 r 0 1 r 1 C dC : 2 2
The second term in the expression for the second variation is obtained with the previously used expression for the slip along the surface: C 2 drjC = C 2 t d : The third term follows from the expression for the rigid body rotation:
dtj
dC
2 t
:
d
; @ 2r @S @S
Finally, the fourth term is obtained by differentiation along the surface coordinates:
dtjC
where
@ t d @S @ t @S
@ t @S
is the surface curvature matrix. Substitution of the last four expressions in the expression for the second variation yields
1 2
C dC
1 r 0 1 r 1 C dC 2
2 td + dC 2 t + d + td :
As in the rst variation, one can split the second variation into a normal and tangential components. For the normal component one nds
d
5.1.33
n 1 t = 0; hence,
n 1 = n 1
Similarly
@ t @S
@n 0t 1 @S
t 1
@n : @S
t 1 = t 1
@ t @S
@t 0t 1 @S :
If the local surface coordinate system is created by projection of a tangential Cartesian X Y system onto the surface, it is readily established that the last terms vanish. Hence, we will assume that the last term in the second variation is zero. The nal result is obtained by substitution of the expressions for the rst-order variation of the slip in the expressions for the second variation. After some reordering and with n 1 this furnishes
dh = ( xA 0 xC ) 1 t t
+(
1 (d x A 0 d x C )
1 1
xA 0 xC ) 1 t (t 2 r) + t(t 2 n) 1 dC
0
+ C
+(
C
d
The rst two terms of the expression for dh will only need to be included if slip occurs, whereas the expression for d
only needs to be taken into account if frictional forces are transmitted. _ For dynamic applications we need the velocity and acceleration terms h and to calculate impact h forces and impulses correctly. These terms are
_ _ _ h = 0 n 1 (x A 0 x C _ 0 C 2 r)
amp;
Reference
Contact formulation for ABAQUS/Standard contact pairs, Section 21.2.2 of the ABAQUS Analysis Users Manual
5.1.34
5.2.1
The contact modeling capabilities in ABAQUS allow access to a library of surface constitutive models. Part of this library in ABAQUS/Standard is the denition of the contact pressure between two surfaces at a point, p, as a function of the overclosure, h, of the surfaces (the interpenetration of the surfaces). Two models for p = p(h) are provided as described below.
Hard contact
In this case
p h
The contact constraint is enforced with a Lagrange multiplier representing the contact pressure in a mixed formulation. The virtual work contribution is
5 = pdh + dph:
This model provides an exponential ph relationship, as shown in Figure 5.2.11. The user denes an initial contact distance, c, and a typical pressure value, po , which is the pressure value at zero overclosure (h = 0). Then, we dene
p p
=0 = (exp(1) 0 1)
dp dh dp dh po
h c
+1
exp
h c
+1 01
and
=0
p = (exp(1) 0 1)
o
1 h
c
1 + 2 exp c + 1 0 c c
h
Interface Modeling
To avoid numerical difculties at high penetration (h > 6c), a linearized pressure-overclosure relation with continuous slope is used.
5.2.11
p0 -c h
dp 1 k0 = dh h = -0.9999c -c -0.9999c
The pressure-overclosure (p-h) relationship can be entered directly in tabular form as shown in Figure 5.2.12.
Softened contact dened with a linear pressure-overclosure relationship
The linear pressure-overclosure relationship is similar to the tabular relationship except that the linear form requires only a single value to be input to dene the slope and the curve always passes through the origin.
Softened contact implementation
A mixed formulation is used because the exponential stiffness associated with softened contact tends to slow down convergence or, due to the development of excessive contact stresses, may cause divergence. For the mixed formulation the virtual work contribution is
5.2.12
Pressure p
(pn,hn)
(p3,h3)
Overclosure h
5 = p(h 0 h) + p h;
(5.2.11)
where p is the contact pressure, h is the actual overclosure, and h(p) is the overclosure associated with the contact pressure, p. A local Newton loop is used to calculate h for the current value of p. The linearized form of this contribution is
d 5 = p dh + dp h 0 dh p dp; dp
(5.2.12)
where dh=dp = (dp=dh)01 is evaluated for the overclosure h. Since there is no term involving hdh, there is a zero on the diagonal of the Jacobian. A zero on the diagonal is not desirable because it may lead to equation solver problems if a rigid body mode is constrained only by contact elements. Hence, a small reference stiffness k0 is introduced by splitting the contact pressure as follows:
p
= q + k 0 (h + c )
and
p
= q + k0 h;
where q is a Lagrange multiplier and k0 is a small reference stiffness (see Figure 5.2.11):
k0 dp = dh =00 9999 :
h : c
Substituting for the pressure p in Equation 5.2.11 and Equation 5.2.12, we obtain
Interface Modeling
5 = q(h 0 h) + h
k0 h
( 0 h ) + q + k 0 (h + c )
(5.2.13)
and
d
5 = q(dh 0 dh) + h
k0 dh
( 0 dh) + dq + k0 dh
(5.2.14)
5.2.13
Further,
dh
= dh dp = dh (dq + k0 dh): dp dp
:
(2 0 dh k0)k0 1 0 dh k0 dp dp 1 0 dh k0 0 dh dp dp
#
dh dq
=0
+ k 0 (h + c ) + k 0 (h 0 h ) h0h
:
In the mixed formulation the difference between the actual and the calculated overclosure h 0 h will go to zero as part of the iterative solution process. The difference must be sufciently small to obtain an accurate solution. The admissible error in h 0 h is set to 0:005c for p p0 . For 0 p p0 the admissible error is interpolated linearly between 0:005c and 0:1c, where 0:1c represents the tolerance level at p = 0:0; alternatively, the tolerances can be specied by the user as part of the solution controls.
Viscous damping option
In addition to the surface constitutive models described above, where the contact pressure is a function of the surface overclosure, ABAQUS/Standard allows for the denition of a viscous pressure that _ is proportional to the relative velocity, h, at which the surfaces approach or separate from each other. This option is intended for the regularization of snap-through problems involving contact where convergence difculties arise due to the sudden violation of contact constraints. The damping pressure, f , is given by
f
_ _ = f (h; h) = (h)h ;
where is the damping coefcient. This coefcient is specied as a function of the overclosure, h, as follows: (0 for h < 0c0 (h) = 0(h + c0 )=(1 0 )c0 for 0c0 h < 0c0 0 for 0c0 h 0 where 0 is the value of the damping coefcient at zero overclosure and is the fraction of the overclosure interval [0c0; 0] over which the damping coefcient is equal to 0. The virtual work contribution associated with the damping pressure is
5 = f h :
5.2.14
The contribution to the stiffness matrix for the Newton solution is given by the linearized form of the virtual work contribution:
d5 = hdf = h @f @h
(0
_ _ @h = h d h + @h dh
_ @f @ h _ @ h @h dh;
dh
where
d dh
for h < 0c0 0 =(1 0 )c0 for 0c0 h < 0c0 0 for 0c0 h 0
In static analysis the velocity is dened as the displacement increment divided by the time h= t, and the stiffness contribution reduces to increment. Therefore, h
_ =1 1 _
d5 = h
d 1h
+ dh : dh 1t 1t _ h+
1t
In the case of dynamics @ h=@h is dened by the dynamic time integration operator, and the stiffness contribution can be written as
d5 = h
d
dh
dh ;
where
and are the Hilber-Hughes-Taylor time integration operator parameters. The viscous damping option cannot be used in a Riks analysis since velocity is not dened.
Reference
Contact pressure-overclosure relationships, Section 22.1.2 of the ABAQUS Analysis Users Manual
Interface Modeling
5.2.15
5.2.2
ABAQUS/Standard provides a surface-based capability for modeling fully saturated porous media. The surface-based capability can be used for small or nite sliding. Only ow normal to the surface is considered; tangential ow cannot be modeled.
The pore uid constraints on the contact interface
Let p1 and p2 be the pore pressures at the two sides of the interface. It is at all times required that the pore pressures on opposite sides be equal:
p2
0 p1 = 0 :
Similarly, let q1 and q2 be the volume ow rate densities normal to the interface at the two sides, _ and let h = v2 0 v1 = n 1 (v2 0 v1 ) be the relative velocity of the two sides in the direction of the interface normal. It is assumed that the interface is lled with uid at all times. Hence, continuity requires that
q1
_ + q2 = 0h ;
0 q1 = q ^
_ ^ = 0 1 (h + q ); 2 1 _ ^ q 2 = 0 (h 0 q ): 2
q1
The contribution of the interfacial virtual work equation and its linearized form are rst obtained in the general form including nite sliding. The equations are then specialized to the various formulations implemented in ABAQUS. Since we want to achieve force and volume ow rate equilibrium at each side of the interface, as well as obtain continuity in the pore pressures, we add the following integral to the virtual work equation: Z
5 pp =
i
where is an arbitrary Lagrange multiplier and A is the interface area. Eliminating q1 and q2 and using a suitable choice for ,
Interface Modeling
we obtain
5 pp = 0
i
Z
where
Since ABAQUS uses displacements (not velocities) and uxes integrated over be multiplied by t to obtain
where
1h is theZincremental change in h in the direction of the interface normal. Linearization yields ipp = 0 d 5 dh p + 1hd p + dph + pdh + 1t(dq p + qd p + dp q + pd q) dA: ^^ ^ ^ ^^ ^ ^
A
5ipp = 0
Z
dA;
Closed contact
1h = 0 and d = 0; therefore, the virtual work simplies to ipp = 0 5 ph + 1t(^ p + p q) dA: q ^ ^ ^
Z
d5ipp = 0
Z
dph
dA:
For steady-state analysis the transient terms can be omitted, and the terms involving uid ow are written in rate form. In this case we can assume that the interface displacements vanish, which leads to the simplied virtual work contribution
5ipp = 0 ss
Z
dA
d5ipp = 0 ss
Small sliding
Z
dph
When the small-sliding contact formulation is used, the terms d p, pdh, d p, and pd q in the linearized virtual work equation will vanish.
Reference
Pore uid contact properties, Section 22.4.1 of the ABAQUS Analysis Users Manual
5.2.22
COULOMB FRICTION
5.2.3
COULOMB FRICTION
An extended version of the classical isotropic Coulomb friction model is provided in ABAQUS for use with all contact analysis cababilities. The extensions include an additional limit on the allowable shear stress, anisotropy, and the denition of a secant friction coefcient. The standard Coulomb friction model assumes that no relative motion occurs if the equivalent frictional stress 2 2 eq = 1 + 2
is less than the critical stress, crit , which is proportional to the contact pressure, p, in the form
crit
p;
where is the friction coefcient that can be dened as a function of the contact pressure, p; the slip rate, eq ; the average surface temperature at the contact point; and the average eld variables at the contact _ point. Rate-dependent friction cannot be used in a static Riks analysis since velocity is not dened. In ABAQUS it is possible to put a limit on the critical stress:
crit
min(p; max );
where max is user-specied. If the equivalent stress is at the critical stress (eq = crit ), slip can occur. If the friction is isotropic, the direction of the slip and the frictional stress coincide, which is expressed in the form i i _ = ; eq eq _ where i is the slip rate in direction i and eq is the magnitude of the slip velocity, _ _
eq _
q2
1 _
+ 2 : _2
As will be shown later, the same laws can be used for anisotropic friction after some simple transformations. The above behavior can be modeled in ABAQUS/Standard in two different ways. By default, the condition of no relative motion is approximated by stiff elastic behavior. The stiffness is chosen such that the relative motion from the position of zero shear stress is bounded by a value crit . (In the ABAQUS Analysis Users Manual crit is referred to as the allowable maximum elastic slip.) The critical slip value, crit , can be specied by the user. If it is not specied by the user, crit is, by default, set to 0.5% of the average length of all contact elements in the model. It is worth noting that this approximate implementation method can also be considered an implementation of a nonlocal friction model; that is, a friction model for which the Coulomb condition is not applied pointwise but weighted over a small area with a so-called mollifying function (Zhong, 1989). See Oden and Pires (1983) for further discussion of nonlocal friction models. Optionally, the relative motion in the absence of slip can be made exactly zero with the use of a Lagrange multiplier formulation. Although this procedure appears attractive because of the exact sticking constraint, it has two disadvantages:
Interface Modeling
5.2.31
COULOMB FRICTION
1. The additional Lagrange multipliers increase the cost of the analysis. 2. The presence of rigid constraints tends to slow or sometimes prevent convergence of the Newton solution scheme used in ABAQUS/Standard. This is likely to occur in areas where contact conditions change. A special case of friction in ABAQUS/Standard is so-called rough friction, where it is assumed that there is no bound on the shear stress; that is, no relative motion can occur as long as the surfaces are in contact. Rough friction is implemented with the Lagrange multiplier method. In ABAQUS/Explicit the relative motion in the absence of slip is always equal to zero if the kinematic contact algorithm is used with hard tangential surface behavior; at the end of each increment the positions of the nodes on the contact surfaces are adjusted so that the relative motion is zero. With the penalty contact algorithm in ABAQUS/Explicit the relative motion in the absence of slip is equal to the friction force divided by the penalty stiffness.
Elastic stick formulation
e In the elastic stick formulation in ABAQUS/Standard, the elastic tangential slip
i l is dened as the reversible relative tangential motion from the point of zero shear stress. The elastic slip is related to the interface shear stress with the relation
i
= ks
iel ;
=
crit :
crit
where ks is the (current) stiffness in stick, which follows from the relation
ks
Since crit may be dependent on contact pressure, slip rate, average surface temperature at the contact point, and eld variables, ks may change during the analysis. The behavior remains elastic as long as the equivalent stress does not exceed the critical stress; hence,
el
i t
Consistent linearization of this expression yields if crit max ; (5.2.31) @ p p if crit p: @p crit The contributions from the contact pressure are nonsymmetric for the second case. Since the slip rate is zero in the elastic stick formulation, derivatives with respect to the slip velocity are not needed. The above expressions hold if the equivalent shear stress remains less than the critical stress. If the equivalent stress exceeds the critical stress, slip must be taken into consideration so that the crit is satised. Let the starting situation be characterized by the elastic slip el . condition eq i The critical stress at the end of the increment follows from the contact pressure, p, and the slip rate, sl eq .
di = k s d i d i = k s d i + i
5.2.32
COULOMB FRICTION
be
e Let the (as yet unknown) elastic slip at the end of the increment be i l and the slip increment sl i . Consistency requires that
1
i =
iel 0
el + 1
isl ; i
and the shear stress at the end of the increment follows from the elasticity relation,
i
The slip increment is related to the stress at the end of the increment with the backward difference approach: i sl sl i eq : crit
1 =
el With these equations and the critical stress equality eq crit ; it is possible to solve for i , el sl and i . Elimination of i and i from the above equations yields,
1 1
i = i
i
1 isl ,
crit
crit 0 el i
sl + i 1 eq crit
or
def
=
el + 1
i; i
pr
i
i = + 1 sl crit: crit eq
where
pr def eq
pr pr = ( 1 )2 + ( 2 )2 :
Substitution in the expression for i and introduction of the normalized slip direction ni furnishes the nal result
pr = ipr = eq
Interface Modeling
i
= nicrit:
(5.2.32)
5.2.33
COULOMB FRICTION
where t is the time increment in a static analysis. In the case of dynamics t is scaled by the Hilber-Hughes-Taylor time integration operator parameters,
and . For the iterative solution scheme this equation must be linearized. Some straightforward algebra yields
pr i d = d
eq crit 0 ninj
crit d
jpr + ni dcrit pr pr eq
i
@
@p
crit
dp + @
_crit d_ eq
sl @ sl
eq
With the expression for the equivalent slip the nal result is
+p
@
@p
if
p dp + ninj 1t @@ d
j
_
eq
crit crit
= max; = p: 1
(5.2.33)
if
In this case the unsymmetric terms may have a strong effect on the speed of convergence of the Newton scheme. Hence, use of the unsymmetric equations solver is strongly recommended for the analysis of problems in which sliding friction occurs. In the case of dynamics t is scaled by the Hilber-Hughes-Taylor time integration operator parameters.
Frictional stress due to forced sliding in static analysis
If the user species different velocities for two bodies in contact as a predened eld, it is assumed that the slip velocity follows from the prescribed motion and is independent of the displacement values. The frictional stresses then follow from
i = eq
where and
i _ ; eq _
eq eq _
= p;
i are the tangential slip velocities derived from the user-dened elds (i = 1; 2). _
Friction contributions for complex eigenvalue extraction analysis
= (_ 1)2 + (_ 2)2:
At the contact nodes at which the velocity differential is imposed, the linearized shear stress can be expressed in the following form:
dp + ninj p @@ d_ j : eq _
COULOMB FRICTION
In the equation above the rst term is generated by the friction forces acting in the direction perpendicular to the direction of slip. The third term exists if the friction coefcient depends on velocity. In the complex eigenvalue extraction analysis both terms contribute to the damping matrix. The second term contributes to the stiffness matrix.
Viscous stick formulation for steady-state transport
For steady-state transport a viscous stick formulation is used. In this case the viscous tangential v slip rate, i , is related to the interface shear stress by
i
= s
iv ; _
=
crit : _
crit
s
The allowable viscous slip is dened as a fraction of the relative tangential velocity
crit
= 2Ff !R;
where Ff is a user-dened slip tolerance, ! is the angular spinning velocity, and R is the radius of a point on the contact surface in the undeformed conguration.
Exact stick formulation
In ABAQUS/Standard Lagrange multipliers are used to enforce exact sticking conditions. A constraint term enforced with Lagrange multipliers qi is added to the virtual work statement: (5.2.34) S where qi is a Lagrange multiplier. By taking the rate of change of Equation 5.2.34, the rate of virtual work is obtained: (5.2.35) S where the last term results from a nonlinear relation between nodal displacements and relative motion i . This term is nonzero only for contact with nite sliding (see Finite-sliding interaction between deformable bodies, Section 5.1.2, and Finite-sliding interaction between a deformable and a rigid body, Section 5.1.3). To obtain a complete formulation, a relationship between i , qi , and i must be dened. For sticking conditions a suitable relation is
3
5 = (i
i + 1
iqi)dS;
Z
Interface Modeling
i
= q i + k0 1
i ;
5.2.35
COULOMB FRICTION
with k0 a reference stiffness selected internally in ABAQUS. The reference stiffness can be considered to model a spring that is in parallel to the sticking constraint. Since the constraint prevents relative motion, the reference stiffness has no physical signicance and is added only to eliminate zero terms on the diagonal of the stiffness matrix that could cause equation solver problems. The variational and rate forms of the equation for i are readily obtained as
i
= q + k0
i
and
d = d q + k 0 d
:
i i i
in Equation 5.2.34 and Equation 5.2.35 vanish. With the backward difference method, the frictional stress is obtained as
d53 = (k0
id
i +
i dqi + qid
i + id
i )dS: (5.2.36) S If the element is slipping, 1
i can take an arbitrary value. Consequently the terms involving qi
i
1i = crit 1
eq ; if
dp p + ninj 1t @@ d
j
_
eq
(5.2.37)
di = 1 ( 0 ninj )d
j + ni d 5 3 =
Z h crit
+ p @ @p
crit crit
= max; = p:
(5.2.38)
if
(ij 0 ninj ) id j + id i dS ZS h 1 eq crit @ 3 d5 = 1 (ij 0 ninj ) id j + dp + p @p ni idp eq S i p @ + 1t @ eq ninj id j + id i dS _ 1 (
if if
crit
= max ;
(5.2.39)
crit
= p:
Observe that for two-dimensional problems the term involving ij 0 ni nj vanishes. In the case of dynamics t is scaled by the Hilber-Hughes-Taylor time integration operator parameters. To determine whether an element sticks or slips, the following tests are used. If the element is currently sticking, it is checked whether the magnitude of the frictional stress exceeds the critical value. Hence, if eq > crit ; the state is changed to slipping.
5.2.36
COULOMB FRICTION
If the element is slipping, the direction of slip is compared with the frictional stress i t applied at the start of the iteration. If at the end of the iteration
()
1
(t) < 0;
i i
In anisotropic friction the friction coefcients in the two (principal) directions are different; hence, the critical shear stresses are different:
crit 1
= 1 p
1
and
crit 2
= 2p:
The two critical shear stresses are at the extreme points of the friction ellipse, given by the equation
2 2
1
2
= p2 : =
(5.2.310)
= 1 ( 2 + 2 ): 2 1 2 =
q
1 2
eq
and the average critical stress
+ 2; 2
crit
= min(p; max); =
where max is the user-specied limiting shear stress, which is applied to the scaled shear stresses. No slip will occur if eq < crit , and slip can occur if eq crit . If slip occurs, it is assumed that the direction of slip is governed by an associated slip law:
_ = @ eq = @ eq d 1 = 1 @ @ d
1 1 1
eq
and
_ = @ eq = @ eq d 2 = 2 : @ @ d
2 2 2
Interface Modeling
eq
_ def 1 1 = _
and
_ def 2 2; = _
(5.2.311)
5.2.37
COULOMB FRICTION
_ eq
_1 _2 =
2 +
2:
i eq
As a result, the direction of the scaled slip and the scaled shear stress coincide:
_ = _ i : eq
Since these equations have exactly the same structure as the equations for isotropic friction, the same solution algorithms can be used to get the incremental and linearized equations in terms of the scaled stress and slip. The actual stresses are readily obtained from the scaled stresses with Equation 5.2.310, and the linearized equations are scaled with the same factors. In the anisotropic elastic stick formulation, it is assumed that the stiffness in stick in terms of the scaled stress and slip is constant: crit ks = ; crit hence, the scaled stress is related to the scaled elastic slip by the relation
i
= ks
el : i
and
In terms of the actual stress and strain in the 1- and 2-directions, this implies
1
with
el = k1 1
2
el = k2 2 ;
k1
2
1
ks ;
k2
2
2
ks :
Hence, the anisotropy in terms of the stiffness in stick is more pronounced than the anisotropy in terms of critical stress.
Reference
5.2.38
5.2.4
ABAQUS/Standard allows heat ow across an interface via conduction or radiation. Generally, both modes of heat transfer are present to some degree. Regardless of the mode, heat transfer across the interface is assumed to occur only in the normal direction.
Conduction
= k ( A 0 B );
where q is the heat ux per unit area crossing the interface from point A on one surface to point B on the other, A and B are the temperatures of the points on the surfaces, and k is the gap conductance. The derivatives of q are 1 @k @q = k + ( A 0 B ) A 2 and
@q = B
0k + 1 @k (A 0 B ); 2
1 ( A + B ): 2
where
=
Radiation
The heat ow per unit area between corresponding points is assumed to be given by
q = F ( A
2
0 Z ) 4 0 ( B 0 Z ) 4 ;
where Z is the value of absolute zero temperature on the temperature scale being used; q is the heat ux per unit surface area crossing the gap at this point, from surface A to surface B ; A and B are the temperatures of the two surfaces; and F is the gap radiation constant derived from the emissivities of the two surfaces. The derivatives of q are
@q A
Interface Modeling
= 4 F ( A 0 Z ) 3
and
@q B
= 0 4 F ( B
0 Z )3 :
5.2.41
Jacobian matrix
5 = qAA 0 qAB ;
where A is the area. The contribution to the Jacobian matrix for the Newton solution is
d 5 = dqAA 0 dqAB ;
where
@q @q dq = dA + dB :
For tied thermal contact the temperature at point A is constrained to have the same temperature as point B . The Lagrange multiplier method is used to impose the constraint by augmenting the thermal equilibrium statement as follows:
where is the Lagrange multiplier. The contribution to the Jacobian matrix for the Newton solution is
Reference
Thermal contact properties, Section 22.2.1 of the ABAQUS Analysis Users Manual
5.2.42
5.2.5
For coupled temperature-displacement analysis in ABAQUS/Standard the user can introduce a factor, , which denes the fraction of frictional work converted to heat. The fraction of generated heat into the rst and second surface, f1 and f2 , respectively, can also be dened. This heat generation capability is to be used only in a coupled temperature-displacement analysis. The heat fraction, , determines the fraction of the energy dissipated during frictional slip that enters the contacting bodies as heat. Heat is instantaneously conducted into each of the contacting bodies depending on the values of f1 and f2 . The contact interface is assumed to have no heat capacity and may have properties for the exchange of heat by conduction and radiation. Refer to Small-sliding interaction between bodies, Section 5.1.1, and Finite-sliding interaction between deformable bodies, Section 5.1.2, for explanations of the notation used for the shape functions and contact parameters involved in the small-sliding and slide line theory. Note that the shape functions for interpolation of the temperature eld may be different from the interpolation functions for the displacements; for example, if the underlying elements are of second order, the displacements are interpolated using quadratic functions, whereas the temperature eld is interpolated using linear shape functions. Hence, the temperature interpolator will be denoted with the symbol M and the displacement interpolation will be denoted with the symbol N . Only the heat transfer terms will be discussed in the following. The heat ux densitiesq1, going out the surface on side 1, and q2, going out the surface on side 2 are given by q1 = qk qr 0 f1qg
and
q2
= 0 q k 0 qr 0 f 2 q g ;
where qg is the heat ux density generated by the interface element due to frictional heat generation, qk is the heat ux due to conduction, and qr is the heat ux due to radiation. The heat ux density generated by the interface element due to frictional heat generation is given by
qg
where is the frictional stress, s is the incremental slip, and t is the incremental time. The frictional stress is dependent on the contact pressure, p; the friction coefcient, ; and the temperatures on either side of the interface. The heat ux due to conduction is assumed to be of the form
= s_ = 1s ; 1t
= (h; p; )(1 0 2 ) = (h; p; )1; where the heat transfer coefcient (h; p; ) is a function of the average temperature at the contact point, = 1 (1 + 2 ); the overclosure, h; and the contact pressure, p. 1 and 2 are the temperatures of side 1 2
qk
5.2.51
Interface Modeling
where F is the gap radiation constant (derived from the emissivities of the two surfaces) and Z is the absolute zero on the temperature scale used. Using the Galerkin method the weak form of the equations can be written as
Z
= F ( 1 0 Z ) 4 0 ( 2 0 Z ) 4 ;
3 Z
1 q1dS
1 qk
( + qr 0 f1qg )dS;
S
2 q2dS
2
where
^ = f11 + f22. The contribution to the Jacobian matrix for the Newton solution is
d
(5.2.51)
where N is the temperature at the N th node associated with the interface element. Note that the summation convention will be used for all superscripts. Therefore, the temperature variables can be written as follows:
( ) = M1N (s)N ;
2 s
( ) = M2N (s)N ;
^ ^ 1 (s ) = 1 M N (s ) N ; (s ) = M N (s ) N ; (s ) = M N (s ) N ; N N N N N ^ where M N (s) = f1 M1 (s) + f2 M2 (s) and M (s) = 1 (M1 (s) + M2 (s)). Substituting the above 2 5 = N
Z h
ds
+ 1M N
@qk
@qk
@1
@qr
d1
@qk
@q + @r d2
^N MN (qk + qr) d1ds 0 qg dM ds ds S @qr @qr M + 1M N @1 M1 + @2 M2M + @@qk 1M M + @qk M M dM 1 @
amp; +1M N
@qk @h
dh
@qk @p
dp @qk d M K @qg @s
amp; +1M N
^ amp; 0M N
@1 @
K @qr dM1
ds
K @qr dM2 K K + +
@2 ds @
1
@qg @
1
( @p dp + @
d1
+ @
@
2
d2
)+
ds
i
ds
@qk dM K + @
ds
K
ds
dS:
5.2.52
d 5 = N
Z h
@q @qk dh + k dp @h @p
ds
@
ds
@qr @1
= 4 F ( 1 0 Z ) 3 ; = (h; p; );
= 0 4 F ( 2 0 Z ) 3 ;
@qk @ 1 @qk @h
= @(h; p; ) 1; @h
@qg @ @ @p
j
= 1s ; 1t
j
@qk @p
For contact pairs and slide line elements, each integration point is associated with a unique slave node. N N If we associate M1 with the slave surface, then M1 will again have only a single nonzero entry equal N with respect to s vanish. In contrast, on the master surface there will to one and the derivatives of M1 N be multiple nonzero entries in M2 , which are a function of the position on the master surface at which contact occurs. For GAPUNIT and DGAP elements each contact (integration point) is directly associated with a node N N pair. Hence, M1 and M2 each have one nonzero entry that is equal to one, and all terms involving N and M2 with respect to s vanish. N derivatives of M1 The variations of overclosure, h, and slip, s, can be written as linear functions of the variations of displacement. These expressions, which determine the form of the matrix for contact elements, have
Interface Modeling
5.2.53
been derived in Small-sliding interaction between bodies, Section 5.1.1, and Finite-sliding interaction between deformable bodies, Section 5.1.2.
References
Fully coupled thermal-stress analysis, Section 6.5.4 of the ABAQUS Analysis Users Manual Thermal contact properties, Section 22.2.1 of the ABAQUS Analysis Users Manual
5.2.54
5.2.6
For coupled electrical-thermal analysis in ABAQUS/Standard the user can introduce a factor, , which denes the fraction of electrical dissipation in the contact zone converted to heat. The fraction of generated heat into the rst and second surface, f1 and f2, respectively, can also be dened. The heat fraction, , determines the fraction of the energy dissipated due to electrical current that enters the contacting bodies as heat. Heat is instantaneously conducted into each of the contacting bodies depending on the values of f1 and f2 . The contact interface is assumed to have no heat capacity and may have properties for the exchange of heat by conduction and radiation. The heat ux densities, q1, going out the surface on side 1, and q2 , going out the surface on side 2, are given by q1 = qk qr 0 f1qg
and
q2
= 0 q k 0 qr 0 f 2 q g ;
where qg is the heat ux density generated by the interface element due to electrical current, qk is the heat ux due to conduction, and qr is the heat ux due to radiation. The heat ux due to conduction is assumed to be of the form
qk
where the heat transfer coefcient h; is a function of the average temperature at the contact point, 1 1 2 , and overclosure, h. 1 and 2 are the temperatures of side 1 and side 2, respectively. 2 The heat ux due to radiation is assumed to be of the form
= ( + )
qr
= F 2( 1 0 Z ) 4 0 ( 2 0 Z ) 4 3;
where F is the gap radiation constant (derived from the emissivities of the two surfaces) and Z is the absolute zero on the temperature scale used. The electrical ux density, J , in the interface element is given in terms of the difference in the electric potential, ', across the interface:
where the gap electrical conductance g h; is a function of the overclosure, h, and the average temperature at the contact point, . '1 and '2 are the electric potentials of side 1 and side 2, respectively. In a steady-state analysis the heat ux density generated by the interface element due to electrical current is given by qg g h; '1 0 '2 2 g h; ' 2;
( )(
) =
( )(1 )
5.2.61
where is the fraction of dissipated energy converted to heat. In a transient analysis the average heat ux density is given by
qg
= 1t
g (h; )(1')2dt t 1 (h; ) 2(1' )2 + 1' 1' + (1' )23 ; t+1t t+1t t t 3 g
Z t+1t
where t is the time at the start of an increment and t is the time increment. Using the Galerkin method, the weak form of the equations can be written as
Z
1 q1dS
1 qk
( + qr 0 f1qg )dS;
S
Z
2 q2dS
2
where
^ = f11 + f22. The contribution to the Jacobian matrix for the Newton solution is ^ 5 = [1(dqk + dqr) 0 dqg]dS:
S
Z
(5.2.61)
( ) = M1N (s)N
and
2 s
( ) = M2N (s)N ; ( ) = M N (s ) N ;
Substituting the above
where N is the temperature at the N th node associated with the interface element. Note that the summation convention will be used for all superscripts. Therefore, the temperature variables can be written as follows:
1 (s ) = 1 M N (s ) N ; ^ ()= ( )+ ()
Z h
s
^( ) = M N (s)N ; ^
N
s
5 = N
1M N
amp
^ ; 0M N
@1
@qr
d1
1 + @qg d @ 1' @
d ' @qr @2
@qg
= 4 F ( 1 0 Z ) 3 ;
= 0 4 F ( 2 0 Z ) 3 ;
5.2.62
@qk
1 = (h; p; );
@qk @
@qg @
For small-sliding interface elements, each integration point is associated with a unique slave node. If N N we associate M1 with the slave surface, then M1 will have only a single nonzero entry equal to one. In N contrast, on the master surface, there will be multiple nonzero entries in M2 . Similarly, the contribution to the variational statement of electrical equilibrium is
1'J dS;
and the contribution to the Jacobian matrix for the Newton solution is
d
5=
1'
@J @
d
+ @@J' d1' 1
@J @
dS:
and
References
Coupled thermal-electrical analysis, Section 6.6.2 of the ABAQUS Analysis Users Manual Electrical contact properties, Section 22.3.1 of the ABAQUS Analysis Users Manual
Interface Modeling
5.2.63
ACOUSTIC-STRUCTURAL INTERACTION
5.2.7
ABAQUS provides two alternatives for modeling interaction between acoustic and structural media: surfacebased interaction or acoustic interface (ASIn) elements. Both are available in ABAQUS/Standard, but only the surface-based capability is available in ABAQUS/Explicit. If the special-purpose interface elements (ASIn) are used, interacting structural and acoustic nodes must be shared by the two meshes. The surfacebased capability can be used for structural and acoustic meshes that have different node numbering and whose surface meshes may not be spatially coincident. The ease of use and low computational cost of the surface-based procedure make it preferable to the element-based approach.
The equations on the contact interface between the structure and the acoustic medium
In the surface-based method the tractions and volumetric acceleration uxes are computed between structural and acoustic media. In place of consistent distributed tractions or uxes on both media, one side (identied as the slave) receives point tractions/uxes based on interpolation with the shape functions from the other (master) side. Either the acoustic uid or the structural solid can be the slave or master, and no Lagrange multipliers are introduced in the formulation. The basis for deciding which to make slave or master is discussed in the ABAQUS Analysis Users Manual. The transient expressions for the coupled acoustic-structural problem are
Z
Vf
fs
frs
ft
Sfr [Sfrs
f 1
f 1
" " : dV +
Z
V
um
_ 1 um dV
Z ZV
St
um 1 um dV
um 1 t dS = 0
Sfs [Sfrs
um 1 n0 p dS 0
(structural medium), where n0 is the normal vector pointing into the uid. The uid-solid surface consists of the union of the directly coupled uid-solid region, Sfs , and a region coupled via a reactive acoustic surface or impedance boundary, Sfrs . Of primary interest here are those terms integrated over Sfs [ Sfrs , which couple the two variational equations. The uid impedance integral, over Sfr [ Sfrs , depends only on the acoustic pressure eld and its variations, so it is unaffected by the contact with the solid. The derivation for the steady-state case is formally identical to the transient case and will not be discussed here. For details of the differences in transient and steady-state acoustics in ABAQUS, see Coupled acoustic-structural medium analysis, Section 2.9.1. When ASIn elements are used (see Acoustic interface elements, Section 18.9.1 of the ABAQUS Analysis Users Manual), the formulation requires that the uid and solid elements be
Interface Modeling
5.2.71
ACOUSTIC-STRUCTURAL INTERACTION
geometrically and nodally conformal so that the shape functions for the structural displacements and the acoustic pressures are identical. The shape functions are integrated using standard methods to yield element matrices of dimension equal to the number of surface nodes on the element. The complete uid-solid coupling matrices are formed by the sum over the element faces; that is, a standard element assembly operation. The two nal coupling matrices have the sparsity pattern of the coupled uidsolid element faces. In surface-based coupling the interaction surface Sfs [ Sfrs is formed by the boundary between possibly nonconforming structural and acoustic meshes. Therefore, the uid-solid coupling matrix cannot be broken up into a sum over element faces as simply as in the ASIn case. To derive the coupling matrices in the surface-based procedure, we use a variation of the master-slave procedure used in small-sliding contact (see Small-sliding interaction between bodies, Section 5.1.1). At the start of an analysis, the projections xN of slave nodes onto the master surface are found, and the areas AN and normals n0 (xN ) associated with the slave nodes are computed. The projections are points p (xN) on the master surface; master nodes in the vicinity of this projection are identied. Variables at the slave nodes xN are then interpolated from variables at the identied master surface nodes near the projection p (xN ). Since the physical degrees of freedom for the uid and solid meshes are different, two cases must be treated. The two cases handle the discretization of the coupling terms differently.
Solid/structural master, uid slave
If the uid medium surface is designated as the slave, we constrain values at each uid node to be an average of the values at nearby master surface nodes (see Figure 5.2.71). The pointwise uid-solid coupling condition,
f @ x
1 @p
1 n0 + um 1 n0 = 0;
is enforced at the slave nodes, resulting in displacement degrees of freedom added to the uid slave surface. These slave displacements are constrained by the master displacements and thereby eliminated; the slave pressures are not constrained directly. Hence, the uid equation coupling term
S [S
fs
p n0
frs
@p 1 @ x dS
is equal to
0
Z
S [S
fs
p n0 um dS:
frs
This term is now approximated, at the slave node level, by the interpolated values of structural displacements at the nearby master nodes times the area of the slave node:
i where Ni (p (xN )) is the master surface interpolant evaluated at the projection of the slave node, um are the structural accelerations at the master nodes, and n0 (xN ) is the normal vector, pointing into
X 0 i p n0 1 um dS AN n (xN ) 1 Ni (p (xN )) um S [S i
fs frs
"
5.2.72
ACOUSTIC-STRUCTURAL INTERACTION
xN xN1
u1
Figure 5.2.71 Fluid slave. the uid, evaluated at the slave node. The summation extends over all master nodes i in the vicinity of the slave node projection p (xN ) (see Figure 5.2.71). The computation is repeated for each slave node xN on the surface Sfs [ Sfrs and assembled to form the entire coupling matrix. Similarly, the contribution to the pressure coupling term in the structural equation due to slave node xN is approximated by
S [S
fs
um 1 n0 p dS pN AN
frs
"
n0 (xN ) 1 Ni (p (xN ))
where pN is the acoustic pressure at slave node xN, and the sum is again over the master nodes i in the vicinity of the slave node projection. These expressions for the coupling terms result in matrices that are the transpose of each other. The normal vectors at the slave surface are used, so these vectors must be well-dened (see Surfaces: overview, Section 2.3.1 of the ABAQUS Analysis Users Manual).
Fluid master, solid/structural slave
If the solid medium is designated as the slave, the values on this surface are constrained to equal values interpolated from the master surface. The pointwise uid-solid coupling condition is again enforced at the slave nodes, resulting in acoustic pressure degrees of freedom added to the solid slave surface (see Figure 5.2.72). These slave pressures are constrained by the master surface acoustic pressures and eliminated; the slave displacements are not constrained directly. Hence, the contribution to the coupling term in the acoustic equation for a single slave node xN is approximated by
Z
Interface Modeling
S [S
fs
frs
2 3 N p n0 1 um dS AN n0 (xN ) 1 um
"
i H ( p (xN ))
5.2.73
ACOUSTIC-STRUCTURAL INTERACTION
P(xN) i pi xN p xN1 AN u2
i+1 pi+1
xN+1 u1
n(xN)
S [S
fs
um
n0 p dS AN
X
i
H (
i p (xN)) pi;
frs
where pi is the pressure at master node i and the sum is again over the master nodes in the vicinity of the slave node projection p (xN ).
Reference
Acoustic, shock, and coupled acoustic-structural analysis, Section 6.9.1 of the ABAQUS Analysis Users Manual
5.2.74
6.1.1
Many of the elements in ABAQUS allow centrifugal, Coriolis, and rotary acceleration forces to be included. This section denes these load types. It is assumed that the model (or that part of it to which these forces are applied) is described in a coordinate system that is rotating with an angular velocity, ! , and/or an angular (rotary) acceleration, ! d! =dt. Let A = [a1 ; a2 ; a3 ] be a right-handed set of unit, orthogonal vectors that form a basis for this ! system. Then, dA=dt = ! 2 A and d2A=dt2 = d! =dt 2 A + ! 2 dA=dt. If the angular velocity is cast as
! = ! n;
where
! is the magnitude of ! and n is the unit axis of rotation, the rotary acceleration is ! dn d! d! = n + ! = nROT A ; dt dt dt
!dn=dt represents the effect of the motion of the axis of rotation (precessional motion); ! = d! =dt 1 nROTA is the magnitude of the rotary acceleration; and nROTA is the axis of rotary acceleration. If dn=dt = 0, then = d!=dt and nROTA = n. In component form
where the term
!k = !nk
and
Let x0 be a point on the axis of rotation. The position of a material particle, x, can be written
d!k dt
d! k dnk n +! : dt dt
x = x0 + y i ai ;
where
yi , i = 1; 2; 3, are the coordinates of the point in the rotating basis system. dx dx0 dyi = + a + y i ! 2 ai dt dt dt i
and
! d2 x0 d2 y i dyi i i d! 2 + dt2 ai + 2 dt ! 2 ai + y ! 2 (! 2 ai ) + y dt 2 ai : dt We assume that the origin of the rotating system, x0 , is xed, so that d2 x dt2
=
dx 0 dt
With this simplication
d2 x 0 dt2
=0
dx dyi = a + y i ! 2 ai dt dt i
6.1.11
and
d2 x dt2
= ddty a 2
+2
dyi dt
2a
Z
+y !
2 (! 2 a ) + y
i
! d! dt
2a :
i
(6.1.11)
V0
0
d2 x dt2
1 x dV 0 ;
where 0 is the mass density of the body in its reference conguration, where its volume is V 0 and x is a virtual variational eld. For the part of the body described in the rotating system, the acceleration, ! d2x=dt2, is given by Equation 6.1.11, while x = yi ai only, since ! and d! =dt are prescribed and x0 is xed. Thus,
W A =
V0
0
d2 y i
yi + dt2
2
dyi dt
! 2 ai + yi ! 2 (! 2 ai ) + yi d! 2 ai 1 y j aj dt
dV 0:
Simplifying,
W A =
V0
0
d2 y i
2 aj )
+ y y (!
i j
! 1 ai ! 1 aj 0 (! 1 ! )ij ) + yi yj d! 2 ai 1 aj dt
dV 0:
V0
0
d2 y i dt2
yi dV 0;
is the usual consistent mass matrix term associated with acceleration of the material particles with respect to the rotating system. Writing the angular velocity of the rotating basis system as its components in that system, ! = !i ai , gives the second term as Z
where ijk is the alternator tensor. This term is the Coriolis force term and arises whenever there is velocity in the rotating system, which can happen in dynamic analysis or in quasi-static cases in which a constant velocity has been introduced (for example, by dening an initial velocity). The third term is, likewise, rewritten as
V0
2 ijk
dyi dt
yj !k dV 0;
0 yi yj (!i !j
0 !k !k ij )dV 0 :
6.1.12
Z
V0
0ijk yi yj
d!k dt
dV 0 :
This term is the rotary acceleration load term. In ABAQUS/Standard the centrifugal load, Coriolis, and rotary acceleration terms contribute to the load stiffness matrix. The centrifugal load term has a symmetric load stiffness matrix,
0
Z
Z
V0
0 !k !k ij )dV 0 ;
V0
@y l
dyi
dt
dy y ! dV ;
j k
and the rotary acceleration term has an antisymmetric load stiffness matrix,
0
Reference
V0
d!k dt
dV 0:
Concentrated and distributed loads, Section 19.4.2 of the ABAQUS Analysis Users Manual
6.1.13
BASELINE CORRECTION
6.1.2
ABAQUS/Standard offers baseline correction of acceleration records for time domain analysis. The correction technique is that proposed by Newmark (1973). An acceleration correction, a0 t , is added a a0, such that the mean to the raw data record, a t , to produce a corrected acceleration record, ac t square velocity over the time of the event is minimized. The acceleration correction is parabolic over any number of time intervals during the event:
()
()= +
()
a0 t
2 ( ) = C1 + C2 t 02T1 + C3 t 02T1 ; T T
T1 < t < T2 ;
where T1 ; T2 denote the limits of a time interval and Ck , k velocity minimization: Z T2 @ 2 dt vc t @Ck T1
[ ( )]
= 0;
where vc t is the corrected velocity record obtained by integrating the corrected acceleration record, ac t . It can be shown that the Ck are dened for each time interval by
()
= 4 1800:= 01890:= 2 where = 1T =T2; 1T = T2 0 T1 ; and A1 , A2, and A3 are dened as 1 Z 1T v( + T1) d; A1 = (1T )3 Z0 1 1T v( + T ) 2 d; A2 = 1 (1T )4 Z0 1 1T v( + T1) 3 d: A3 = (1T )5 0 Here v(t) denotes the uncorrected velocity record obtained by integration of the uncorrected acceleration record, a(t). These velocities are obtained by assuming the uncorrected and the corrected acceleration vary
:
C2 ; C3
8 9 < C1 =
0300:
0 1 9 38 900: 0630: < A1 + 0vc (T1 ) 0 v(T1 )1=(21T ) = 05760:= 4200:= 5 : A2 + 0vc (T1 ) 0 v(T1 )1=(31T ) ; ; 6300:=2 04725:=2 A3 + vc(T1) 0 v(T1 ) =(41T )
linearly over each time increment of the original acceleration history. This is not exact for the corrected acceleration record (because of the parabolic variation of the correction in time), but it is assumed that the acceleration history is discretized at sufciently small time increments so that this is an insignicant error.
Reference
6.1.21
6.2.1
For beam and truss structures immersed in uid (e.g., offshore piping and riser problems), ABAQUS/Standard provides a capability for introducing drag forces via Morisons equation, inertia loads, and buoyancy loads. Fluid drag is associated with velocities due to steady currents and any waves that may have been specied. Fluid inertia is associated with wave accelerations. Buoyancy has two components: the hydrostatic pressure measured to the mean uid level and the dynamic pressure caused by the presence of waves. Partial submergence is done automatically for all uid load types. Drag and inertia loads are considered in two forms: distributed loads along the length of the element (distributed drag loading is further divided into a component normal to the elements axis and a component along the tangent to the element) and point drag and inertia loads where the beam changes cross-section. Buoyancy loading is applied with a closed-end assumption; that is, it is assumed that the elements ends can support buoyancy loading normal to the elements cross-section. If the ends of the element are actually open endedthat is, the elements ends cannot support uid pressure loadspoint buoyancy forces are provided to remove the buoyancy forces at the ends of the element. This section documents the form of these loadings. It is assumed that the uid particle velocities and accelerations are known as functions of the current spatial location; they are dened by superimposing the steady current velocity and the wave velocity.
Continuously distributed drag and inertia loading
Distributed load types FDD, WDD, FDT, and FI on beam, truss, or rigid beam elements provide continuously distributed drag and inertia via Morisons equations. To specify this loading, the following denitions are used:
vf
af vp ap 1v t 1 vt 1 vn
is the uid particle velocity (dened by the steady current input and possibly additional contributions from wave denitions), is the uid particle acceleration when the waves are dened, is the velocity of a point on the element (nonzero during dynamic analysis steps only), is the acceleration of a point on the element (nonzero during dynamic analysis steps only),
CT CD CM CA
= (1v 1 t)t is the relative tangential (axial) velocity of the uid, = 1v 0 1vt is the relative transverse velocity of the uid,
is the tangential (axial) drag coefcient, is the transverse drag coefcient, is the transverse inertia coefcient, is the transverse added mass coefcient, is the uid density,
6.2.11
R h
is the structural velocity factor, and is the exponent for tangential drag.
2 FD = 1 CD D 1vn(1vn 1vn) 1 : 2
1
Then, the transverse drag force per unit length on the member is
FT = 1 CT D1vtj1vtjh01 ; 2
and the inertia force per unit length is
At points where the section changes size and, thus, exposes an end area to the uid (see Figure 6.2.11), additional drag and inertia forces arise.
Figure 6.2.11 Change in section of an immersed beam. For this case we need the additional denitions:
1A
t
Cn
is the change in area of the section, is an outward normal to the exposed area (see Figure 6.2.11), is the drag coefcient associated with the discontinuity,
6.2.12
is the tangential inertia coefcient associated with the discontinuity, is the tangential added mass coefcient associated with the discontinuity, and are uid and structural acceleration shape factors for the tangential inertia term.
Fds =
1 2 Cn
1A1vft 1vft
j
for for
This force is nonzero only when the relative velocity of the uid has a negative projection on the outward normal; i.e., when the uid is owing against the exposed surface. The inertia force is
ABAQUS assumes closed-end conditions when computing the distributed buoyancy loads (load type PB) for beams, pipes, rigid beams, and elbows. An open-end condition can be simulated by using load type TSB to remove the buoyancy load from the end of the element. For truss elements the effect of buoyancy is simply Archimedes principle; that is, a vertical force equal to the weight of the displaced uid is applied to the element. For buoyancy loading the beam is assumed to have a uniform circular pipe section with the effective diameter specied as part of the loading denition. Consider a pressure eld that varies with z , the vertical coordinate, where z k 1 x and k is the unit vector pointing in the vertical direction. Then
^ = p (z ) :
^ = g(z0 0 z);
where z0 is the vertical location of the free surface of the uid (either the free surface elevation of the uid inside the pipe or the elevation of the mean water level for external pressure), is the density of the uid, and g is the acceleration due to gravity. If waves are dened, the wave eld generates a dynamic pressure effect. In this case the pressure eld has a nonlinear variation with respect to location. Under the assumption of waves with small amplitude relative to the wave length and depth of the sea, we can assume that the dynamic pressure is slowly varying with respect to the wave direction; hence, derivatives of the pressure with respect to the coordinates in the plane parallel to the still water surface are neglected. However, the dynamic pressure eld has a nonlinear dependence on z . The pressure eld results in a nodal loading contribution that can be written as follows. Let p be the pressure loading contribution to the weak form of the equilibrium equations. This contribution can be written in terms of a nodal load vector, pN , as
p
5 = pN 1 xN ;
6.2.13
pN =
Zl
0
R p ^dx dNN ds ds
2
@p ^N k ds ; @z N
where R is the radius of the pipe (internal or external), p is the total pressure (including hydrostatic and dynamic pressure), NN are the shape functions associated with the nodes on the element, and l is the current length of the element.
Point buoyancy
Point buoyancy is needed to apply discrete buoyant forces on an exposed surface. Point buoyancy can be used to remove the closed-end loading condition on beam, pipe, or rigid beam elements by applying a negative load on the exposed area. Use the following denitions:
z w f zw zf
is the elevation of the point considered, is the mass density of the uid outside the pipe, is the mass density of the uid inside the pipe, is the free surface elevation of the uid outside the pipe, is the free surface elevation of the uid inside the pipe, is the outward normal to the exposed area, and is the gravitational acceleration. In ABAQUS it is assumed that
t g
g = 0 g k.
where
f1 =
and
0 1 n f2 = 0 1
n
Load stiffness
To ensure the quadratic convergence of the Newton method in ABAQUS, it is necessary to calculate the changes in the above forces with respect to changes in the kinematic solution for the structure. Thus, a load stiffness is calculated for all of these load types.
Reference
6.2.14
6.2.2
This is a linearized wave theory based on irrotational ow of an inviscid incompressible uid. The linearization is achieved by assuming the wave height a is small compared to the wavelength and the still water depth. It is also assumed that the uid is of uniform depth (that is, the bottom is at). Since we have irrotational ow, there exists a ow potential, , obeying
52 = 0
and giving the uid particle velocities as
(6.2.21)
v=
Now assume there is a potential energy per unit mass, G, (in this case associated with the gravity eld). Then, equilibrium is given by
@ : @x
(6.2.22)
@v @t
@v 1v @x
=
@p 0 @G 0 @ x ; @x
(6.2.23)
where is the uid density and p is the pressure. Substituting Equation 6.2.22 in Equation 6.2.23, we obtain 2 @ @v @G @p
@ x@t
@x
1v
0 @ x 0 @ x :
@ 1 + v1v @t 2
=
0G 0 (p 0 p0 ) + F (t);
(6.2.24)
since the uid is assumed to be incompressible; thus, is constant. Here F is an arbitrary function of t and p0 is the pressure in the air just above the free surface. For convenience we choose F (t) = 0: The term 1=2v 1 v can be neglected compared to the other terms (this can be shown, from the resulting solution, to be consistent with the order of approximation that the wave height is small compared to the wavelength). By choosing the z -coordinate to point vertically upward, the gravity potential is conveniently chosen as
G = g (z 0 z s );
@ + g (z 0 zs ) + p 0 p0 = 0: @t p = p0 + g(zs 0 z ) + pdyn :
6.2.21
(6.2.25)
From this equation the total pressure at a point below the instantaneous uid surface is
Hence, the total pressure is the air pressure plus the hydrostatic pressure plus the dynamic pressure, pdyn , where pdyn is given by @ : pdyn = 0 @t Let be the elevation of the uid surface at time t above the mean (undisturbed) uid surface level, zs . Since the position of the free surface is a part of the solution, there are two boundary conditions that must be applied on z = . The rst is a dynamic equilibrium condition at the interface between the uid and the air. Since the interface is assumed to have no mass, the forces normal to the interface in the uid and the air must be equal. If the surface tension of the interface is neglected, the pressure in the water and the air must be equal at the interface. Assuming that the pressure due to the motion of the air is negligible (which can be shown to be reasonable), the air pressure can be approximated by its undisturbed value (Whitham, 1974). The dynamic boundary condition then implies p = p0, where p is the pressure in the water at the free surface and p0 is the pressure in the undisturbed air. Since is assumed to be small with respect to the depth of uid, the boundary condition can be made linear by applying it at z = zs instead of at z = zs + . With these assumptions Equation 6.2.25 provides the boundary term
@ =0 : g @t z=zs
1
(6.2.26)
The second boundary condition on the free surface comes from the kinematics of the free surface. Let the free surface be given by
@ @t
0 @z = 0: @t
If we assume that the wave height is small compared to the wavelength (max << ); then we can approximate @z by the velocity vz , so that @t
@ 0 vz = @ 0 @ = 0 at z = zs : @t @t @z Eliminating between Equation 6.2.26 and Equation 6.2.27 gives @ @2 +g @t2 @z The boundary condition on the bottom, z = zb , is vz
= =0
(6.2.27)
at z
= zs :
(6.2.28)
@ @z
=0
at z
= zb :
(6.2.29)
6.2.22
The problem is now dened by Equation 6.2.21, Equation 6.2.28, Equation 6.2.29, and the requirement that the solution be a plane wave periodic in the horizontal plane, such that
(x1 ; x2 ; z; t) = (d 1 x 0 c t; z );
where d is the direction of wave propagation and c is the wave speed. We solve the problem by assuming that = P (z )8(x1 ; x2; t). Since functions, Equation 6.2.21 provides the two equations
and
are independent
where A1 and A2 are constants and is the phase angle of the wave in degrees ( provides an arbitrary choice of origin in time and is chosen so that the vertical displacement of a uid particle is a minimum when s = 0, t = 0, and = 0). There is no motion at the bottom of the uid in the vertical direction, so by Equation 6.2.29 we nd
2 3 360 ;
(6.2.210)
A1 = 0
Substituting this into Equation 6.2.210 gives
2 3
A2 tanh kzb :
2
where C is a constant. The dispersion relation can be obtained by substituting Equation 6.2.211 into Equation 6.2.28 and setting z = zs , giving
2 3 g c2 = tanh k(zs 0 zb ) : (6.2.212) k The wave frequency ! is related to the wave period by ! = 2= . The constant k is called the wave number and is related to the wavelength by k = 2=, so that ! = ck. The free surface elevation above the undisturbed uid surface, , is given by Equation 6.2.26:
2 3; 360
(6.2.211)
=C
(6.2.213)
6.2.23
Writing the wave amplitude (half the wave height) as a, this denes
a = 0C
so that Equation 6.2.211 can be rewritten
2 3 kc cosh k (zs 0 zb ) g 3
this one wave component. The term 1 (v 1 v) in Equation 6.2.24 was neglected because the wave amplitude 2 a is small compared to the wavelength . This implies, from Equation 6.2.23, that the uid particle acceleration is approximated as a = @ v=@t; that is, the convective part of the acceleration, @ v=@ x 1 v, is neglected. Since the theory is linear, any set of waves can be superposed by linear superposition of its components: X
2 3 ga cosh2 k(z 0 zb ) 3 2 sin k (s 0 c t) + =0 kc cosh k(zs 0 zb ) 360 2 3 (6.2.214) 0s 0 c t ga cosh2 (2=)(z 0 zb ) 3 1 =0 sin 2 + : 2 cosh (2=)(zs 0 zb ) 360 This solution provides uid particle velocities v = @=@ x and accelerations throughout the uid for
=
components
N ;
where N is the potential Equation 6.2.210 of the summarized as follows. For z zs : Velocity potential:
2
N th
N
360
ui =
N
360
and vertically,
2 X sN g aN N sinh2 (2=N )(z 0 zb ) 3 cos 2 uz = 0 0 t 2 N cosh (2=N )(zs 0 zb ) N N components
N
360
6.2.24
horizontally,
N
360
and vertically,
N
360
sN a cosh2 (2=N )(z 0 zb ) 3 sin 2 ai = 02g dNi N 0 t N cosh (2=N )(zs 0 zb ) N N components
X
N
360
and vertically,
N
360
z=0
components
aN cos 2
sN 0 t N N
N
360
2 N
2
N
tanh
2 2
N
z s 0 zb ) :
Dynamic pressure:
=N )(z 0 zb ) sN N 2 3 cos 2 pdyn = 0g aN 0 t + 360 ; N cosh (2=N )(zs 0 zb ) N components 2 3 X @pdyn sN N aN sinh2 (2=N )(z 0 zb ) 3 cos 2 = 02g 0 t + 360 ; @z cosh (2=N )(zs 0 zb ) N N components N
X
cosh (2
and
@ 2 pdyn @z 2
N
360
6.2.25
Airy wave theory is a linearized theory; however, the wave amplitude can be large compared with the size of a structure. We, therefore, must make an assumption about the wave kinematics below a crest and above the mean water level. The assumption used here follows modied Airy wave theory as described in Hansen (1988) and Faltinsen (1990). The free surface boundary condition has been made linear in Equation 6.2.26. Above the mean or undisturbed surface level zs the velocity, acceleration, and dynamic pressure are extrapolated from their values at the mean surface level. Hence, for zs < z
Accordingly,
Reference
6.2.26
6.2.3
Assume that an innite series of plane, uniform waves travels through the uid in the positive S -direction. The z -coordinate is chosen to be positive in the vertical direction, so the gravity potential is G = g(z 0 z0), where z0 is an arbitrary datum. Assume that the uid is inviscid and incompressible. The uid particle velocities are derivable from a ow potential
52 = 0
v=0
Equilibrium is
(6.2.31)
@ : @x @p 0 @G 0 @ x ; @x
@v @t
@v 1v @x
=
where is the uid density and p is the pressure. Writing @ v=@t in terms of the ow potential then gives
Integrating with respect to equation
@2
@ x@t
@v 1v @x
@G @x
@p : @x
x (note that is constant since the uid is incompressible) gives the Bernoulli
@ @t
0 1 v 1 v = G + p 0 p0 + gF (t); 2 0 1 v 1 v = g (z 0 z s ) + p 0 p 0 ; 2
where F (t) is an arbitrary function (which for convenience is set to zero) and p0 is the atmospheric pressure. Substituting in the gravity potential, this is
@ @t
where zs is the undisturbed surface level. From this equation the total pressure at a point below the instantaneous uid surface is p = p0 + g(zs 0 z ) + pdyn : Hence, the total pressure is the air pressure plus the hydrostatic pressure plus the dynamic pressure, pdyn , where pdyn is given by @ 1 0 2v 1 v : pdyn = @t
6.2.31
Let is
@ 1 0 v1v @t 2
z = +zs
= g;
assuming the pressure at the surface is negligible. Assuming the waves are uniform, of wavelength and period , and that they travel in the positive S -direction means that the solution as a function of S and t must appear in terms of a phase angle
= 2
S t 0
360
2
S 0 ct +
360
where c = is the wave celerity. This means that, for any function in the solution,
@ @t
Thus, at the free surface boundary
= 0c
@ : @s
= cvs ;
@ @ = 0c @t @s
or
(6.2.32)
A further boundary condition at the free surface is that the uid particle velocity relative to the wave celerity must be tangential to the slope of the wave:
@ vz j : = vs 0 c z=+zs @s
At the seabed (z = zb), there is no uid motion in the vertical direction:
(6.2.33)
0vz = @ jz=zb = 0: @z
The problem now consists of nding a potential function, , that satises Equation 6.2.31the boundary condition at the seabedas well as the boundary conditions at the surfaceEquation 6.2.32 and Equation 6.2.33. Stokes proposed a power series solution to this problem, and Skjelbreia and Hendrickson (1960) have obtained that solution to fth-order. The potential function is assumed to be
6.2.32
where = 2=, the Aij are constants that depend on the ratio of water depth to wavelength (zs 0 zb )=, and is a parameter. The wave prole, (), is assumed to be
2 = = (A11 + 3A13 + 5 A15) cosh[ (z 0 zb )] sin c c 0 (2 A22 + 4A24 ) cosh[2 (z 0 zb )] sin 2 + (3 A33 + 5A35 ) cosh[3 (z 0 zb )] sin 3 0 4A44 cosh[4 (z 0 zb )] sin 4 + 5A55 cosh[5 (z 0 zb )] sin 5;
(6.2.34)
= 0 cos + (2 B22 + 4 B24) cos 2 0 (3 B33 + 5 B35) cos 3 + 4 B44 cos 4 0 5 B55 cos 5;
where the Bij are constants for a given water depth and wavelength. Finally, it is assumed that
(6.2.35)
c2 = c2 (1 + 2C1 + 4 C2) 0
and that
(6.2.36)
F = 2C3 + 4 C4:
Skjelbreia and Hendrickson obtain the 18 constants Aij ; Bij , Ci, and c0 from matching terms in equal powers of and cos in the free surface boundary conditions, Equation 6.2.32 and Equation 6.2.33. They give the constants as functions of s = sinh[ (zs 0 zb )]; c = cosh[ (zs 0 zb )]; as
c2 = g s=c; 0 A11 = 1=s; A13 = 0c2 A15 A22 A24 A33 A35
5c2 + 1 ; 8s 5 1184c1 0 1440c8 0 1992c6 + 2641c4 0 249c2 + 18 =0 ; 1536s11 3 = 4; 8s 192c8 0 424c6 0 312c4 + 480c2 0 17 = ; 768s10 13 0 4c2 = ; 64s7 512c12 + 4224c10 0 6800c8 0 12808c6 + 167704c4 0 3154c2 + 107 ; = 4096s13(6c2 0 1)
6.2.33
A55 = 0 B22 = c
2c 2 + 1 ; 4s 3 272c8 0 504c6 0 192c4 + 322c2 + 21 ; B24 = c 384s9 8c 6 + 1 ; B33 = 3 64s6 88128c14 0 208224c12 + 70848c10 + 54000c8 0 21816c6 + 6264c4 0 54c2 0 81 ; B35 = 12288s12(6c2 0 1)
B44 = c B55 =
C1 =
192000c16 0 262720c14 + 83680c12 + 20160c10 0 7280c8 12288s10(6c2 0 1)(8c4 0 11c2 + 3) 7160c6 0 1800c4 0 1050c2 + 225 ; + 12288s10(6c2 0 1)(8c4 0 11c2 + 3)
Skjelbreia and Hendrickson (1960) have a factor +2592 multiplying c8 in the equation for C2. This was corrected to 2592 by Nishimura et al. (1970). They then obtain equations for (= 2=) and . The wave height is
H=
(6.2.37)
c2 =
2
(6.2.38)
Given the wave period, wave height, and water depth, Equation 6.2.37 and Equation 6.2.38 must be solved simultaneously for the wavelength, , and the parameter . This is done with a Newton method, using the Airy (linear) wave solution as an initial guess.
6.2.34
Fluid particle velocities and accelerations for Stokes 5th order wave
360
D1 = A11 + 3 A13 + 5A15 ; D2 = 0(2 A22 + 4 A24); D3 = 3 A33 + 5A35 ; D4 = 04 A44; D5 = 5 A55: vs = 0 @ @s
and and
vz = 0
@ ; @z
@v @v @v a z = z + z vs + z vz : @t @s @z Since the solution appears in terms of the phase angle , it follows that @v @v @v = 0c =0 : @t @s @s as = @vs @t
+
This allows the acceleration components to be written as
@vs @v vs + s vz @s @z
a s = ( v s 0 c)
@vs @vs + vz @s @z
and
a z = ( v s 0 c)
@vz @s
+ vz
@vz : @z
@ 1 0 v1v : pdyn = @t 2 Substitution of the expression for yields: vs = 0ce1 ; vz = 0ce2 ; 3 2 c 2 as = 0(1 + e1 )e3 + e2 e4 ; 3 2 c 2 (1 + e1 )e4 + e2 e3 ; az = c 1 pdyn = 0 e + v1v ; 1 2 @pdyn 2 c =0 (e 4 + e1 e 4 + e 2 e 3 ) ; @z @ 2 pdyn 4 2 c (e + e e + e e + e e + e e ) ; =0 z2 6 1 6 4 4 2 5 3 3
6.2.35
where
e1
n=1
5 X 5 X 5 X 5 X
n=1
X nD
5 2
n cosh
n S t (z 0 zb ) cos 2n 0
360
e2 = e3 = e4 = e5 =
and
nDn sinh
n S t (z 0 zb ) sin 2n 0
2
360
; ; ;
n=1
n S t (z 0 zb ) sin 2n 0
360
n=1
360
360
n=1
; :
e6 =
5 X
n=1
n Dn cosh
3
360
=
where
360
The Stokes wave eld is a spatial description of the wave eld. All wave eld quantities are calculated up to the instantaneous uid level. The wave eld denes velocity, acceleration, and dynamic pressure at spatial locations for all values of time. Hence, the velocity, acceleration, and dynamic pressure are determined by using the current (for geometrically nonlinear analysis) or reference (for geometrically linear analysis) location of the structure at the current time in the appropriate equations. The time used in the wave eld equations is the total time for the analysis, which accumulates over all steps in the analysis (static, dynamic, etc.).
Reference
6.2.36
6.3.1
ABAQUS provides a capability for introducing generalized forces on acoustic and solid media associated with the arrival of dilatational waves. This capability applies to acoustic scattering problems and problems involving blast loads in air or water. Thus, the capability is available in transient dynamic procedures in ABAQUS/Standard and ABAQUS/Explicit. Consider a dynamic problem involving uid and coupled solid domains, excited by a propagating wave in the uid arriving from outside these domains. When the mechanics of a uid can be described as linear, wave elds in the uid can be superimposed. Therefore, the observed total pressure in the uid can be decomposed into two components: the incident wave itself, which is known, and the wave eld excited in the uid due to reections at the uid boundaries and interactions with the solid. To compute the latter, scattered solution, it is sufcient to apply loads at the boundaries of the uid and solid domains corresponding to the effects of the incident wave eld. The uid mechanical behavior is nonlinear when the uid is capable of undergoing cavitation. In that case superposition of the incident wave and the response due to the boundaries, the solid, and the cavitating uid regions to the incident wave loading is not valid. A total wave formulation (see Coupled acoustic-structural medium analysis, Section 2.9.1) is used in ABAQUS/Explicit to handle the incident wave loads on an acoustic medium capable of undergoing cavitation. In the total wave formulation the incident wave loading is applied as traction on the boundary of the modeled acoustic domain as the wave impinges on this domain from an external source. The default scattered wave formulation applicable in the absence of cavitation is presented below.
Scattering formulation
The equations for coupled uid-solid interaction in ABAQUS are developed in Coupled acousticstructural medium analysis, Section 2.9.1. Here, we proceed from the coupled system:
(M f
M NM uM
fr
(6.3.11) These equations couple the total pressure in the uid to the displacements in the solid. They are valid for any combination of uid and solid domains in a particular model: the matrix Sfs is dened on all the interacting uid and solid surfaces. The uid traction
NM _ C(m) uM
IN
=+ =
h QN iT pQ + P N 0 Sfs
PM T M Sfs
3
P Pf
01 @p 1 n0 = u + u 1 n0 f _f T = f @ x f
(6.3.12)
is a quantity (with dimensions of acceleration) that describes the mechanism by which the solid motion drives the uid; the uid drives the solid by the pressure on the solid surface. In Equation 6.3.12, n0 is the inward normal on the boundary.
6.3.11
To proceed, we decompose the total pressure into the known incident wave component and the unknown scattered component: p pI + pS : Then
(M f
PQ P P P P Mfr Q )pQ + (CfP Q + Cfr Q )pQ + (KfP Q + Kfr Q + K Q )pQ S _S S
=+
(M f
PQ
+ Mfr
PQ
)p I + ( C f
PQ
+ Cfr
PQ
)p I + ( K f _
PQ
P P Kfr Q + K Q )pQ I
2 PM 3
P M Sfs M TS
Pf P 0
(6.3.13)
Sfs
M TI ;
and
M NM uM
NM C(m) uM _
IN
QN Sfs
iT 2
p S Q + pI Q
+P
(6.3.14)
We see that the displacements in the solid are driven by the sum of the incident pressure, which forms an applied boundary traction, and the scattered pressure, the unknown eld in the uid. The incident eld is independent of the scattered eld by convention. Therefore, it can be shown that it is a solution to the equation
(M f
PQ
+ Kfr
PQ
P K Q )p Q I
P M Sfs M TI ;
+Mfr =
PQ
P M Sfs M TS
)p S + ( C f
PQ
Pf P :
+ Kfr
PQ
P K Q )p Q S
(6.3.15)
This equation, using the scattered pressure as the unknown eld variable, is solved together with Equation 6.3.14 using explicit or implicit integration or using the steady-state formulation (Coupled acoustic-structural medium analysis, Section 2.9.1). The scattered uid traction, TS , depends on the incident pressure through the decomposition above and the solid motion at the boundary. In the absence of an impedance condition at this boundary, this results in the relation
TS
= =
=
01 @pS 1 n0 = u +
u 1 n0 0 T _ I @x
f
u+ u+
u 1 n 0 0 uI + u I _ _ f f 1 @pI 1 n0 : u 1 n0 + _ f f @ x
1 n0
(6.3.16)
If an impedance condition is applied at the uid-solid boundary, application of the total pressure decomposition and Equation 2.9.113 results in the relation
6.3.12
TS
= u+
0 0
Virtual mass formulation
f f
c1
1
c1
1 1 1 pS 0 + pS pS 0 _ f k1 c1 k1 1 1 1 pI 0 + pI : pI 0 _ f k1 c1 k1
_ u f
1
n0 +
@pI f @ x
1
n0
(6.3.17)
The virtual mass approximation is a simplication of the general scattered-eld form of the coupled system Equation 6.3.11. Physically, it corresponds to considering the uid wave speed to be very large compared to the characteristic structural wave speeds of interest; that is, uid disturbances propagate and distribute in the eld at an extremely high rate compared to the disturbances in the structure. This can be modeled by imposing an incompressibility condition on the uid. Formally, applying such a condition results in the suppression of the terms related to time derivatives, volumetric drag, and impedance in the uid equation for the coupled system:
KfP Q pQ M
NM M
=+ =
P Sfs M uM
u
NM M + C(m ) u + _
Sfs
QN
iT
f
h
1 2 PM 3
Sfs
n0 1
QN Sfs
iT
@pM I @x
N
(6.3.18)
pI
+P
The uid will also be assumed to have homogeneous mass density, f . The uid pressures, pQ , can be eliminated from the structural equation by inverting the uid stiffness, KfP Q to yield
M NM uM
+ C (m ) u _
NM M
IN
0 0
h h
QN Sfs QN Sfs
iT iT
(K f
PQ
01 2 S P M 3 u M
fs
pI
0
iT
QN Sfs
iT
(K f
PQ
0 1 2S P M 3 fs
f
n0 1
@pM I @x
PN:
(6.3.19)
Dening
NM MVMA
QN Sfs
(K f
PQ
01 2 S P M 3 ;
fs
(6.3.110)
M NM
NM MVMA uM
+ C (m ) u _
NM M
+I
QN Sfs
iT
pI
NM I 0 1 MVMA n0 1 @px @ f
P N : (6.3.111)
This equation makes the term virtual mass clear: the effect of an incompressible uid surrounding a NM structure is to add inertia to the structural motion, MVMA. For some shapesin particular, cylinders, disks, and spheresthe virtual mass is a priori known from classical solutions.
6.3.13
QN The matrix operator Sfs is the discrete form of the integral over the wetted or loaded surface (see Coupled acoustic-structural medium analysis, Section 2.9.1). If the structure is long and thin, it is often modeled with beam elements. Then it is appropriate to replace the surface integral that
generates the
QN Sfs
iT
pI
Sfs
H N pI n0 dS;
(6.3.112)
pI
0
Z
Z
Vwetted
HN
@pI dS; @x
(6.3.113)
where Vwetted is the volume enclosed by the wetted surface. If H N is chosen to be the same interpolation function used for the beam and if the loading is similarly interpolated, this term becomes
[Sfs ]
T
pI
= =
Vsolid
H N H M dS @pM I : @x
@pM I @x
0
1
s
(6.3.114)
NM
Therefore, the virtual mass equations for a beam immersed in an incompressible uid are
0 1
NM
+ MVMA
NM
u
NM M + C(m ) u + _
f
MVMA +
NM
s
NM
@pM I @x
+P
(6.3.115)
This equation states that an immersed beam is loaded by a term due to the pressure gradient in the uid due to the incident wave and that it experiences an additional amount of inertia due to the presence of the uid.
Scattering load amplitude
The incident eld in the uid is assumed known; moreover, it is assumed to be associated with a separable solution to the scalar wave equation of the form
p I (x ; ) p t ( )p x (x );
where is the retarded time, which is a function of the true time, t, and the position, x. In applications of interest here, the incident wave is produced by an excitation outside the computational domain. However, its temporal part, pt(t), is specied through an amplitude time history at a single point inside the computational domain, referred to as the standoff point, x0 . The spatial term, px (x), is restricted to forms modeling plane and spherical wavefronts only, emanating from a specied source point, xs . An incident wave produces a time-varying pressure at a spatial point of interest xj of the form
6.3.14
pI (xj ; t)
p t (t)p x (x j ): kx s 0 x 0 k kx s 0 xj k
(6.3.116)
xj is expressed by either
p x (x j ) =
p x (x j ) = 1
for plane waves. The amplitude observed at a point xj is related to that at the standoff point through this spatial variation, delayed by the time required for the wave to travel from the standoff point to the point xj :
pI (xj ; t) = pt
Rj
p t ( j )p x (x j );
R0
c0
0 R0 p
x( j )
(6.3.117)
k xs 0 x0 k; k xs 0 xj k
s 0
and
Rj
for plane waves. j is referred to as retarded time because it includes a shift corresponding to the time required for the wave to move from the standoff point to xj . Of course, the delay may be positive or negative, depending on the relative distances between the point of interest, xj , and the source and between the standoff point and the source. The incident wave boundary tractions in the solid equation Equation 6.3.14 result from direct substitution of Equation 6.3.117. The uid boundary tractions Equation 6.3.14 and the tractions in Equation 6.3.115 require the evaluation of the gradient of the incident wave pressure:
@pI (xj ; t) @x Rj Rj pt p x c0 @ x Rj @ x 1 @ 1 @ R 0 + p t px R0 : _ + px pt c0 @ x R0 @ x
=
0 px pt _
1 @
The gradient simplies for spherical waves and slowly moving source points to
@pI (xj ; t) @x
=
01
c0
pt _
0R
1
j
pt
R
R2 j
(x j
0 xs )
6.3.15
and to
@pI (
xj ; t) @x
=
01 _
c0 pt
1
R0
(x 0
0 xs )
(x j
0 xs ) 0 xs )
( x0
for plane waves. The time derivatives of the incident pressure eld can be written
pI ( _
xj ; t) = pt px _
and
pI (
xj ; t) = pt px ;
again neglecting the small terms associated with source point motion.
Effect of structural motion
Generally, the effects of the motion of the source point, xs (t), are included in the dened amplitude pt (t) of the incident wave eld (specied at the standoff point x0 ). That is, it is assumed that the stated amplitude reects the time history at a specic xed standoff point, including the effects of the source point motion, if any. For example, in underwater shock loading the incident wave eld is produced by a pulsating gas bubble, which migrates toward the free surface of the water. Therefore, the corresponding incident wave amplitude model for this effect includes a moving source point. The structural model affected by the incident wave may be in motion, as well. For example, a ship can move with respect to an explosive charge source during the loading. Because the wave motion is usually extremely fast compared to the ship motion and because incident wave pulse durations are typically short, some approximations apply. First, it can be assumed that the motion of the standoff point, x0 , can be described by its position at time zero and by the average velocity during the incident wave loading, v0 . In addition, the magnitude of this velocity can be assumed small compared to the speed of wave propagation, so the wave equation itself is unaffected by the relative motion. Furthermore, it can be assumed that the specied amplitude history, pt (t), corresponds to observations made at the position of the standoff point at time zero, x0 (0). Under these assumptions the effect of structural motion during incident wave loading can be modeled in a reference frame xed with the standoff point, so the source point appears to move with v0 :
xs (t) 7! xs (0) 0 v0 t:
6.3.16
This motion also affects the perceived arrival time of the wave at the standoff point. Dening the time of arrival at x0(0) as t and the time of arrival at x0 (t) as tI , we can express the difference between these times in terms of v0 :
c0jtI 0 tj = kxs (t) 0 v0 t 0 x0 (0)k 0 kxs (t) 0 x0 (0)k: This time shift is maximum when the movement v0 is aligned with xs (t) 0 x0 (0). Therefore, tkv0 k (xs (t) 0 x0 (0)) c 0 j tI 0 tj k x s (t) 0 kxs (t) 0 x0 (0)k 0 x0 (0)k 0 kxs (t) 0 x0 (0)k jtjkv0k:
Since we have assumed that the standoff motion is slow compared to the wave speed, we neglect this small time shift.
Underwater explosion bubble load amplitude
An underwater explosion can lead to the formation of a highly compressed bubble that propels the surrounding water radially outward, generating an outward-propagating shockwave. As the bubble expands, the pressure inside decreases until it is considerably below the ambient pressure. After reaching a maximum radius with a minimum pressure, the bubble begins to contract. The contraction proceeds until the bubble collapses to a minimum radius. Because of a large pressure generated inside the bubble during this stage, the bubble begins to expand again, generating a second outwardpropagating wave. Once the bubble expands to a second maximum radius, it contracts again. The same expansion-contraction sequence can repeat many times. At the same time during this pulsation process, the bubble migrates upward under the force of buoyancy. As long as the bubble does not reach the free water surface, the expansion-contraction sequence continues, each time with reduced amplitude.
Geers-Hunter model
Geers and Hunter (2002) proposed a phenomenological model that treats an underwater explosion as a single bubble event comprised of a shockwave phase and an oscillation phase, with the rst phase providing initial conditions to the second. Based on this model, the volume acceleration during the shockwave phase is given by
P [0:8251 exp(01:338t=Tc ) + 0:1749 exp(00:1805t=Tc)]; (6.3.118) f c 1=3 1=3 1=3 in which Pc = K (mc =ac )1+A and Tc = kmc (ac =mc )B , where mc and ac are the mass and radius of the explosive charge, respectively, and K , k, A, and B are constants for charge material; f
V (t) =
4ac
is the mass density of the uid. _ Integration with V (0) = 0 and further integration with V (0) = (4=3)a3 then yields c
_ V (t) = 4ac
V (t) = a3 + c
4 3
f
6.3.17
a=
3 4
1=3
and
a= _
1 4
_ V =a2 :
These expressions are evaluated at tI = 7Tc to determine the initial conditions for the subsequent bubble response calculations during the oscillation phase. This choice was validated since, for a single set of charge constants, the initial condition values for values of tI between 3Tc and 7Tc produce essentially the same response during the oscillation phase, as demonstrated by Geers and Hunter (2002). The following are the equations of motion for the doubly asymptotic approximation model to describe the evolution of the bubble radius, a, and migration, u, during the oscillation phase:
l0 l _ 0 c1 l0 0 a2 0 1 u2 0 2 u a1 ; _ _ _ a 3 3 f l1 _ u = 02 _ 0 c1 l1 0 2au ; __ a f 1 g 2 l1 1 1 1 g l0 _ + _ + (1 + )u _ 0Z ; + a 2 + u2 0 _ cg l0 = (6.3.119) (1 + ) 2 2 f 3 a 3 a f g g 1 3 1 g g l1 2 _ 1+ + + CD u ; + 2 au 0 1 0 __ _ l1 = ga 0 cg 2 (1 + ) f f f a a 8 cg c 1 cg g l1 g1 _ + 2+ 0 c g 3 CD u2 ; + au + __ _ 10 ga 0 cg 2 g 1 = (1 + ) cf cf f a a f 8 a=0 _
where
g cg ; f cf
and
f
(P g
0 pI + f gu) + 3
1
"
l1 a
2
g f
g 1 a
=
2 #
;
Pg
=
Kc
Vc V
; g
=
c
Vc V
; cg
= cc
Vc V
1 ( 01) 2
s
;
and
cc
Kc ; c
and c is the charge mass density, cf is the sound speed in uid, V = (4=3)a3 is the current volume of the bubble, Kc is the adiabatic charge constant, Vc is the volume of charge, is the ratio of specic heats for gas, g is the acceleration due to gravity, and pI = patm + f gdI (where patm as the atmospheric pressure and dI as the depth of the charges center). CD is an empirical ow drag parameter, which impedes the bubbles migration. _ _ Seven initial conditions are needed. The rst two are a(tI ) = aI , a(tI ) = aI , the second two _ are u(tI ) = 0, u(tI ) = 0, the fth one is l1 (tI ) = 0, and the remaining two are determined as
02
1
gI f
aI _ cf
+ I
aI ZI cf
6.3.18
and
g1(tI ) =
0I01
1
0 gI
f
2
g cf
a2 I
with
ZI = f (PgI
01
0 pI ) 0 1 3
1
0 gI
f
f gI
ga 2
I
cgI
Then Equation 6.3.119 can be solved by using any suitable method for nonlinear ordinary differential equations; in ABAQUS, a fourth-order Runge-Kutta integrator with variable time steps is used. The incident pressure induced during the bubble response can be expressed as
pI (xj ; t)
p t (t)p x (x j );
a A
c
(6.3.120)
where
p t (t) = f
4 V (t);
Rj
(6.3.121)
with V given by Equation 6.3.118 for the shockwave phase (tI < 7Tc ), and
p t (t) =
f
4
7Tc );
p x (x j ) =
1
Rj
Here Rj kxs 0 xj k. Since the constants A and B are both substantially smaller than one, pt(t) for the shockwave phase above is only weakly dependent on Rj . Thus, for simplicity, Rj can be treated as a constant when the gradient of the pressure is evaluated. In this way all formulations in the previous two sections can be still used without losing any signicant accuracy.
Waveless model
The model of Geers and Hunter (2002) can be simplied to ignore the energy losses due to waves in the uid and the gases. This waveless form of the equations more closely reproduces the results of earlier research on this phenomenon. In the waveless model the shockwave phase of the loading is unchanged from the previous case. The equations of motion for the evolution of the bubble radius and migration during the oscillation g phase are simplied, however, using the assumptions c1 7! 0; f 7! 0;and cg 7! 0:Under these f assumptions, the waveless equations of motion for the bubble dynamics are
6.3.19
l0 ; a l1 u = 02 _ ; a
a=0 _
_ l0 _ l1 _ g 1
a2 + u 2 _ _
1 3
+
2 3
u _
l1 a
0 Z;
(6.3.122)
= au __
= 2au; __
0 ga + 3 CD u2; _ 8 0 pI + f gu) + 1 3
Vc V
where
f
=
(P g
l1 a
2
;
Pg
Kc
g
= 0;
cg
= 0:
The initial conditions are determined as for the previously discussed case. Since the equation for g1 is now decoupled from the others, it does not need to be solved.
Reections outside the computational domain
ABAQUS allows the user to specify planes outside the computational domain that reect the incident wave. This reected wave is superposed, with a suitable time delay or phase shift, onto the wave arriving at the standoff point via the direct path, forming the total incident wave eld for that source. This functionality is implemented for spherical or planar waves and reection planes at any orientation, which may have arbitrary impedance properties. Only a single reection from each plane is considered. However, multiple reections inside the nite element computational domain are considered automatically.
Spherical waves
Consider the schematic arrangement of source point, standoff point, and reection plane A as shown in Figure 6.3.11. The reection plane can be specied by a normal vector, nA , and some point on the plane, xA . The distance from the source to the plane is dA k (xA 0 xS ) 1 nAk, and the plane is assumed to have a specic characteristic impedance, ZA . This information, together with the behavior of the source, is sufcient to characterize the reected eld in terms of an image source, located at the point xSA xS + 2dAnA . The source will be derived in steady state and then extended to the transient case. The actual source is given by
PS
pt e0i
~
2 kx
S 0xk c0
kx S 0 x0 k ; kx S 0 x k
6.3.110
"standoff point" incident wave bearing medium has sound speed c0 , mass density f
Figure 6.3.11 Schematic of incident wave loading. and the image source is given by
2 3 0i
kxSA00xk kxS 0 x0 k ; c PSA pA e ~ kxSA 0 xk where the normalization length kxS 0 x0 k is chosen for convenience.
Generally, a nonzero imaginary ~ part of the planes impedance ZA will imply that pA will also have a nonzero imaginary part. The ~ unknown pA is found by enforcing the impedance condition at the plane:
PSA + PS
and the conservation of linear momentum:
= i
ZA (~ f 1 nA) ; u
where is the complex density of the medium, including volumetric drag effects. Combining these ~ two equations and simplifying, we have
0
2 uf = @@x [PSA + PS ] ; ~~
2
In ABAQUS the impedance property is specied using admittance constants 1=c1 and 1=k1,
1 1 1 0 ZA c0 + i
~dA ~ 5: pA = pt 4 ~ ~ 1 1 01 0 ZA c0 + i
~dA ~
3
1 = 1 + i
; Z c k
A
6.3.111
so it is convenient to use
pA ~
Q+ pt ~ Q0
3 2 1 i
1 1 c1 + k1 0 c0 ~ + i
~dA 5: = p t 4 1 i
1 ~ 1 0 c1 0 k1 0 c0 + i
~dA ~
This describes the source strength for a spherical image source in steady state, but in the transient ~ case the complex density needs to be eliminated using = f + i
: The expression for the image
source can be expanded to
0 c
0 d1 0 i
f 0 i
k 0 i
c1 +
2 f pA ~ c k
c1
0 d1 + i
f + i
k 0 i
c1 0
2 f pt : ~ c k
A
In ABAQUS the effects of uid volumetric drag, the spreading loss due to the distance to the reection plane, and the complex part of the admittance 1=k1 are ignored, so the amplitude of the reected wave is related to the amplitude of the incident wave by
" 1 1 # c1 0 c0 f ~ pA pt ~ ~ 1 0 0 1 Q pt :
c1 c0 f
Therefore, the reected spherical load is similar to the direct load, with magnitude reduced by the reection impedance effect and by the greater distance travelled. In the time domain inversion of the steady-state result, retaining a phase shift corresponding to the arrival time at the standoff point, yields
kx S 0 x 0 k ; kxSA 0 xk
0
j
nsa
Soft and hard limits
xSA 0 xj kxSA 0 xj k :
0 and
pA
= 0 pt :
6.3.112
A hard reecting plane refers to a zero uid particle motion condition, implying 1=ZA
p
0. Then
A = pt :
Planar waves
The schematic for reected plane waves is similar to that for spherical waves, except for the geometry of the reection. In contrast to the spherical case, when planar waves reect from a boundary (see Figure 6.3.12), the direction of the reected wave is common for all points on the wavefront. The location of the perceived source of the reected wave is different, however. To calculate the phase shift for the reected wave incident at an arbitrary point xj correctly, the locations of these perceived sources need to be calculated.
Xs
Direct
do , n
Xs d sa n sa X0 Xj
Direct
Reflecte
?
X
d 'n v
X sa
Figure 6.3.12 Schematic of incident wave loading using planar waves. The reected plane wave load can be calculated on the basis of the reected wave direction and the time delay associated with the additional path length of the reected wave. Referring to Figure 6.3.12, we see that the reected wave direction is given by
kx S 0 x S k + kxj 0 xv k: ^ ^
The time delay of interest for point xj , j , is related to the difference of these distances via the acoustic wave speed:
c0 j = ( x j
0 x 0 ) 1 n 0 j + kx S 0 x0 k 0 k xj 0 x v k: ^ ^
The rst term on the right is the same as the delay due to the direct path; the rest is the increment due to the additional path length of the reection. Some geometric manipulations allow this expression to be written in terms of the locations of xj , xS , x0 , and the location of the spherical-wave virtual source, xsa : d 1 d 2 sa 0 c0 j = j(xj 0 x0 ) 1 n0j + 2rv kdsakr0 ; where
r0 =
kx S 0 x 0 k;
and
rv =
Reected planar waves are subject to reduction in amplitude due to impedance conditions at the reection plane in the same manner as spherical waves, but there is no spatial attenuation. Unlike spherical waves, planar waves may yield no reections for some orientations of the specied reection plane: if the direction of the wave is parallel to the reection plane, or strikes at an obtuse angle, no reections are generated.
Reference
6.3.114
6.4.1
ABAQUS/Standard allows for the simulation of uid penetrating into the surface between two contacting bodies and application of the uid pressure normal to the surfaces. The surface-based contact approach is used to model the interactions between the bodies, where one surface denition provides the master surface and the other surface denition provides the slave surface. Both surfaces can be deformable, or one can be rigid. A single slave nodebased penetration criterion is used. Fluid will penetrate into the surface between the contacting bodies from one or multiple locations, which are exposed to the uid, until a point is reached where the contact pressure is greater than the critical value specied by the user, cutting off further penetration of the uid. The critical contact pressure is introduced to account for the asperities on the contacting surfaces. The higher this value, the easier the uid penetrates. The default value of the critical contact pressure is zero, in which case uid penetration occurs only if the contact pressure is zero and contact is lost. When a node has a contact status of OPEN, its nearest neighboring nodes are considered to be subjected to the uid pressure as well. The nodes initially exposed to the uid, which are specied by the user, will always be subjected to the uid pressure irrespective of the contact status at these nodes. The pressure penetration load will be applied normal to the element surface based on the pressure penetration criterion described above at the beginning of an increment and will remain constant over that increment even if the uid penetrates further during that increment. A nodal integration scheme is used to integrate the distributed pressure penetration load over an element; the variation of the distributed load over an element will be determined by the load magnitudes at the elements nodes, which are coincident with the base points. Consider the contact interaction of three nodes101, 102, and 103on the slave surface made up of faces of two rst-order elements, 1 and 2, with a master surface made up of faces of two elements, 4 and 5, which are described by nodes 201, 202, and 203 as shown in Figure 6.4.11. If the uid with a pressure magnitude of f has penetrated up to node 102 on the slave surface, the variation of the distributed load over element 1 is given by
P1
f N1
f N2
f;
f N1 ;
where N1 and N2 are the shape functions on the face of a rst-order element. For an element-based master surface we must also consider how the uid pressure is applied to the master surface, which depends on the location of the anchor point on the master surface. The anchor point is chosen as the point on the master surface closest to the last slave node subjected to the uid pressure (pressure penetration tip). All the nodes between the anchor point and the node initially exposed to the uid on the master surface, as specied by the user, are considered to be subjected to the uid pressure. The pressure penetration capability does not require that the contacting surfaces have matching meshes; however, the best accuracy is obtained when meshes are initially matching. For initially nonmatching meshes the equivalent uid pressure applied on the master surface can be evaluated based on equilibrium
6.4.11
Figure 6.4.11 Pressure penetration with nonmatching meshes. considerations. For the problem illustrated in Figure 6.4.11, the anchor point corresponding to slave node 102 is D. For the element associated with the anchor point (element 5) on the master surface, part of the load is transferred from element 1 and part of the load is transferred from element 2 on the slave surface. The load distribution on element 5 is illustrated in Figure 6.4.12, where L5 is the element length. Because the uid pressure will be simulated as a distributed load on the contacting surfaces, the variation of the load over an element needs to be described. For a rst-order element if Q202 and Q203 are the equivalent forces at nodes 202 and 203 due to the distributed load shown in Figure 6.4.12, the variation of the distributed load over element 5 on the master surface is given by
P5
=4
Q202
02
L5
Q203
N1
+4
Q203
02
L5
Q202
N2 :
P1
P2
202
D (102) L5
(103)
203
master surface
Figure 6.4.12 Stress distribution over an element on the master surface with nonmatching meshes.
Reference
Pressure penetration loading, Section 22.1.6 of the ABAQUS Analysis Users Manual
6.4.12
PRESSURE STIFFNESS
6.5.1
In geometrically nonlinear analysis pressure loads are applied on the deformed structure. Hence, the equivalent nodal loads are dependent on the nodal displacements. This dependency leads to additional contributions to the Jacobian in the solution procedure used in ABAQUS/Standard. The external virtual work is Z
W
E
u 1 p n dA;
where A is the surface on which the pressure is applied; n is the normal to this surface, pointing into the material; u is the virtual displacement eld; and p is the pressure magnitude.
Pressure load stiffness on a surface in three-dimensional space
The expression
2 @ x dg dh; @h
where x are the current coordinates of a point on the surface, and the surface parametric coordinates (g, h) are chosen to give the correct sign to n through the cross product. The external virtual work is then given by Z
W E
p u
g;h
1 @ x 2 @ x dg dh; @g @h
0dW = 0
E
Z
g;h
p u
@du
@g
2 @ x + @ x 2 @du @h @g @h
dg dh;
Now
n dA = k 2
@x @g
t dg;
where k = (0; 0; 1) is a unit vector out of the plane of the model, t is the thickness of the twodimensional solid (which is assumed to be constant), and the surface parametric coordinate g is chosen to give the correct sign to n through the cross product. The external work is then given by
W
E
=
g
p u
1 k 2 @ x t dg; @g
6.5.11
PRESSURE STIFFNESS
0dW E
Reference
=0
p u
1 k 2 @du t dg: @g
Concentrated and distributed loads, Section 19.4.2 of the ABAQUS Analysis Users Manual
6.5.12
6.5.2
ABAQUS provides for loads per unit length in the beam cross-sectional directions as distributed load options for the beam elements (load types P1, P2). Since these are follower forces, they have a load stiffness; and this stiffness can sometimes be important especially in the case of buckling prediction by eigenvalue extraction. The symmetric form of this load stiffness is included in ABAQUS/Standard (see Hibbitt, 1979, and Mang, 1980). This form is developed below. The external virtual work on the beam is
where the pressure load, , is given by the externally prescribed pressure magnitude, p, as = p ; where = 1 or 2 denes the particular cross-sectional direction of the load. Therefore, = (01) 2(d =dS ), where = 2 when = 1, and = 1 when = 2 so that
p u dS;
1
p n n x
e = (01)
d 1 dS; 2 dS
x u
where S is the material coordinate along the beam. Now assuming that the load magnitude, p, is externally prescribed so that it does not change with position, the rate of change of e with change in position, d , is Z
Now and so
Thus,
dS
This load stiffness is not symmetric, except in the case of a beam in a plane with xed ends (or no ends, such as a ring), in which case the rst term is exactly zero and the second gives the symmetric form
(01)
p 1
ddu
d 2 + dS 2 d dS
u u dS: u
In ABAQUS, even for the general beams in three dimensions, the load stiffness is introduced as the symmetric part of d e above.
Reference
Concentrated and distributed loads, Section 19.4.2 of the ABAQUS Analysis Users Manual
6.5.21
PRESSURE ON ELBOWS
6.5.3
Elbow elements are often used to model pipelines in which the curvature of the pipe can change signicantly while the pipe is subjected to uniform or hydrostatic pressure. Therefore, pressure loadings that include large geometry changes are developed for these elements, as described in this section. The virtual work contribution of pressure on the lateral surface of the elbow is
WL
p=
p x 1 n dA;
where
p x
n
A
is is is is
pressure magnitude, variational displacement at a point on the midsurface of the lateral wall of the elbow, normal to the lateral wall midsurface, and area of space occupied by the lateral surface in the current conguration.
The product of the surface normal and the differential area can be rewritten in terms of material coordinates S 0 along the pipe and around the pipe section:
n dA =
where r0 is the initial pipe radius, so that
r0 @
1 @x
@x 2 @S 0 r0 d dS 0;
p WL =
1 @x p 0 x r @
@x 2 @S 0 r0 d dS 0;
where A0 is the area of space occupied by the lateral surface in the reference conguration. For hydrostatic pressure, the pressure magnitude is a function of position:
p = (z 0
0 x 1 k ) (z 0 p z 1 ) ; 0
where
p1 z0 z1
is is is is
the reference pressure magnitude, the zero pressure height, the reference height, and (0; 0; 1), a unit vector in the vertical direction.
x = x + (r0 + ur )r + ut t + y 3 a3;
6.5.31
PRESSURE ON ELBOWS
where
x (S 0 )
is the position of a point on the pipe axis, is the radial displacement, is the tangential displacement, is a1 cos + a2 sin , and is 0a1 sin + a2 cos ; with a1 (S 0 ) and a2(S 0 ) being the cross-sectional basis vectors.
ur (; S 0 ) ut(; S 0 )
r t
2 [(r0 + ur )r + utt + y3 a3 ];
t
3
and the derivatives of the position with respect to the parametrization are
@x @
= (
@ur @
0 ut)r + (r0 + ur + @u )t + @y a3 @ @
and
+ [
dS dx
dS 0
Assuming that (1) terms in twist in the pipe, (2) terms in y3 and its derivatives can be ignored due to negligible warping in the pipe, (3) a3 (dS=dS 0 ) = dx=dS 0 , (4) the stretch dS=dS 0 is unity, and (5) ur and ut are small compared to r0, we p arrive at the following expression for the integrand of WL :
x r0 p @x 1 @ x 2 @S 0 @
@ 1 a3 + @S 0 1 a3 + (r 1 a3 + ut @St0 1 a3] a3: @S 0 (@ r=@S 0 ) 1 t and (@ t=@S 0 ) 1 r can be ignored due to negligible
0
@S @y3
@r + ur )
@S
@S
p x
r0
+ + +
! p ! p ur p ut r0 r0
(1 + r0
@S 0 @r
@r
1 a3)(r0 + ur + @u ) @
@ur
): (1 + r0 0 1 a3 )(ut 0 r0 @S @
For closed-end loading the virtual work contribution of pressure on the end-caps of the elbow is
WE =
E0
y
1 @ y 2 @ y r d dr; @r @
6.5.32
PRESSURE ON ELBOWS
where r and are the material coordinates in a two-dimensional cylindrical coordinate system of points on the end-caps of the elbow element, y represents position on the end-caps, and E 0 is the area of space occupied by the end-caps of the elbow element. We assume the following deformation for the end-caps:
y(r; ; S 0 ) = x(S 0 ) +
r r0
where S 0 is the parameter that identies the end-cap being considered. The assumed deformation arises naturally on considering a deformation of the end-caps in which radial rays of the reference end-cap conguration remain straight lines under deformation. It can be shown easily that the assumed deformation of the end-caps is differentiable as long as the deformations of the circumferential curves of the end-caps are differentiable. For the end-cap boundary shapes that arise in applications (primarily ovalized modes), the assumed deformation will be locally invertible so that integration of functions over the deformed surface is not likely to be a problem. Ignoring the terms due to warping in the expression for position on the lateral surface, the rst variation of position on an end-cap is
r y = x + 0 [ (ur r ! ! ! ! 0 ut! 1 a3 )r + (ut + (r0 + ur )! 1 a3 )t + (ut ! 1 r 0 (r0 + ur )! 1 t)a3];
=
@ur @
r0
[ (r0 + ur )r + utt]
t
and
@y @
r r
0[(
0 ut )r + (r0 + ur + @u )t ]: @
@ut @
The integrand of the expression for the virtual work of pressure on the end-caps can now be expressed as
p r y
1 @y 2 @y @r @
0u
r t @u
@
g[
1
r0
2 x
1n +
r r0
! 3 !
1 fu t r 0 ( r 0 + u r ) t g a 3 1 n ] ;
where n is a3 if the center of the end-cap is node 1 of the element and 0a3 if the center is node 2 or 3 of the element. The load stiffness for the pressure loading, which by denition is the rst variation of the virtual p p work of the pressure load, is given by d(WL + WE ). The following expressions are required for its calculation:
p @x d( 0 x r @
@x 2 @S 0 )
dp r0
@x 1 @ x 2 @S 0 @ @x p x 1 @x 2 @S 0 ! @ @ ! g [( @d! 2 r + d! 2 @Sr0 ) 1 a3 + + f @r @S 0 (1 + r0 @S 0 1 a3 ) x
@S 0
@r
! 1 d! 2 a3 ]
6.5.33
PRESSURE ON ELBOWS
@r @dut @ut @
p r0
(1 + r0
@S 0
a3 ) x 1 [ (dur +
@dur @
! )d! t ] @ p @ur @r ! + 0 (1 + r0 0 a3 )! [ dur (ut ) + (r0 + ur )(dut r @S @ @ut @dut dut(r0 + ur + ) ut(dur + ) a3 @ @ @ur @ut ! ) u t (r 0 + u r + ) d! + (r0 + ur )(ut @ @ p @dut @r ) + 0 (1 + r0 0 a3 )ur ( dur + r @S @ p @dur @r + 0 (1 + r0 0 a3 )ut ( dut ) r @S @
+ (dut 0
)t + (ut 0
@ @ur
)r + (r0 + ur +
! )d! 2 r
1 f 0
0 @du @
g 2 a3 ]
1 1
and
d( p r y
1 @y 2 @y ) @r @
dp r
y
1 @y 2 @y @r @
1
r0
+ p[ 2r0dur + dur [ 2 x
@ut @ r r0
+ ( r 0 + ur )
@dut @
0 dut @u 0 ut @du @ @
r
1n +
3 (a 3
+ p[ r 0 + 2 r 0 u r + ( r 0 + ur ) [
r
1
r0
! 2 x d!
2 n ]:
1
In the above dp is nonzero only in the case of hydrostatic pressure, when it is given in the rst case by
dp =
0 (z 0 p z 1 ) k 1 d x 0 0 (z 0 p z 1 ) k 1 d y : 0
1
Reference
Pipes and pipebends with deforming cross-sections: elbow elements, Section 15.5.1 of the ABAQUS Analysis Users Manual
6.5.34
SLIDING CONSTRAINT
6.6.1
SLIDING CONSTRAINT
The sliding constraint has a variety of uses. For example, this MPC is used in conjunction with other MPC types to constrain a shell element mesh to a solid element mesh. The MPC maintains consistency with standard shell theory by forcing initially straight lines through the thickness to remain straight despite rotation and displacement. When applied to solid element nodes on the shell-solid interface, this MPC enforces a kinematic approximation of compatibility with the shell model. The theory of this constraint is as follows: Let P 1 , P n be the points dening the line; and let P m be the node that must lie on this line. The direction of the line is given by
n = (xn 0 x1)=l;
n n
where
l
= (x
0 x 1 ) 1 (x 0 x 1 )
t
31
2
Let i; j; k be base vectors in the x-, y-, z -directions in the global coordinate system. Then, dene a unit vector normal to the line as
1
=
n2i jn 2 ij
unless
= i( n2 =
n3
= 0),
2k jn 2 kj
n
=n
2 t1
and n now form a set of orthonormal base vectors with t1 and t2 normal to the line joining P 1 and P . The constraint can be imposed by the condition that the line joining the node m to node 1 be perpendicular to t1 and t2; that is,
t ; t
2,
(x
0 x1 ) 1 t 1 = 0
and
(x
m
0 x1 ) 1 t 2 = 0
We now choose a local coordinate numbering system such that i is the global direction on which t1 has its largest projection: jt1j > jt1j and jt1j jt1 j; where j 6= i; k 6= i: i j i k Likewise, we choose global direction
j
such that
tj
and
j
j 2j j 2 j
> tk ;
where
k:
6.6.11
SLIDING CONSTRAINT
Using this denition of i; j; k the constraint conditions can be written explicitly in terms of coordinate components of node m as
x i ti
m 1 + xm t1 + xm t1 = x 1 1 t 1 j j k k
+
m 2 xj tj
and
m 2 x i ti
+ xk
tk
=x
1t : i; j; k
m m xi ; x j
avoids dividing
0 0
ti t j ]
2 1
= x (t
1 2
tj
0 t2 1 ) 0
tj
m 1 2 x k (t k t j
0 0
tk t j )
2 1
and
m 1 2 x j [ ti t j
tj t i ]
1 2
= x (t
2 1
ti
0 t1 1 ) 0
tj
m 1 2 x k (t k t i
t k t i ):
2 1
The above equations will enforce the desired constraint. We also need the derivatives of these constraints. These are
( dx
m
dx
) t + (x
0 x1 ) 1
dt
=0
and
(d x
m
dx
) t + (x
0 x1 ) 1
d n )n
dt
= 0;
where
dt
=(
0 t1 1
and
dt
=(
0 t2 1
m
d n )n :
0 0
dx
) t
1 1 0 (x 0 x1 ) 1 nt1 1
m
dn
=0
and
(d x
dn
dx
) t
n
1 2 0 (x 0 x1 ) 1 nt2 1
to give
(i
dn
= 0:
1
l
0 nn) 1 (
1
=l )
dx
0 0 0
dx
);
and, therefore,
t
1 1
dn
= (t
1( 1(
dx
dx
and
t
dn
= (t
=l )
dx
dx
):
1 t1 0
dx
1 t 1 0 (x 0 x 1 ) 1 n (t 1 ) 1 (
m
=l
dx
dx
)=0
6.6.12
SLIDING CONSTRAINT
and Let
P
= 1=l (
i; j; k
x m 0 x1 ) 1 n .
m 1 dxi ti
+ dxj
tj
+ dxk
tk
0 0
Pd
xn 1 t1 0 (1 0 P )dx1 1 t1 = 0
+ dxj
tj
m 2 dxk tk
Pd
xn 1 t2 0 (1 0 P )dx1 1 t2 = 0:
we obtain
m 1 2 dxi (ti tj
Pd
0 2 1) + ( 1 2 0 2 1 )+ x 1 (t1 2 0 t2 1 ) 0 (1 0 ) x1 1 (t1 2 0 t2 1 ) = 0
ti t j
m dxk tk tj
tj
tk t j
tj
P d
tj
tj
and
0 1 2 ) + ( 2 1 0 1 2 )+ 0 x 1 (t2 1 0 t1 2) 0 (1 0 ) x1 1 (t2 1 0 t1 2) = 0 In the two-dimensional case n lies in the or plane and t2 = (0 0 01). This implies that the
m 2 1 dxj (tj ti
Pd t j ti
m dxk tk ti
ti
tk t i
ti
P d
ti
ti ;
x y
r z
x 1 1 t 1 0 xm t1 ; k k xn 1 t1 0 (1 0 P )dx1 1 t1 = 0:
m 1 dxi ti
+ dxk
tk
Pd
Reference
General multi-point constraints, Section 20.2.2 of the ABAQUS Analysis Users Manual
6.6.13
6.6.2
The shell to solid constraints SS LINEAR, SS BILINEAR, and SSF BILINEAR are used to connect shell elements to a solid element mesh. These MPCs are used in conjunction with the sliding constraint SLIDER (see Sliding constraint, Section 6.6.1). The SLIDER MPC maintains consistency with standard shell theory by forcing initially straight lines through the thickness to remain straight despite rotation and displacement. Thus, the shell to solid MPC must enforce the remaining constraints: The displacement of the shell node at the interface must be equal to the displacement of the corresponding point on a line of nodes through the thickness of the solid; The rotation of the shell node at the interface must be compatible with the rotation of the corresponding line of nodes through the thickness. The three MPC types impose essentially the same constraints but use different weighting factors to reect the nature of the interpolations in the solid elements. SS LINEAR is used with rst-order elements, SS BILINEAR is used at the edges of second-order elements, and SSF BILINEAR is used for the middle of second-order elements. The MPCs can be used in two-dimensional as well as three-dimensional models. The degrees of freedom will automatically adapt to the dimensionality of the problem. The shell to solid MPCs can be used with any number of points through the thickness of the solid. The weighting functions for the nodes in the solid will be chosen based on this number. The displacement constraint for the shell node is obtained by setting the displacement of the shell node, us ; equal to the weighted average of the displacements of the nodes in the solid:
us =
Xw u ;
n i=1 i i c
(6.6.21)
where the MPC selects the appropriate weighting factors, wi; based on the MPC type and the location of the nodes. If the SLIDER MPC is used to keep the nodes in the solid on a straight line, the choice of weighting factors does not inuence the solution. For the formulation of the rotation constraint, we assume that the nodes on the solid remain on one in the undeformed line. Hence, this line of nodes can be represented by the normalized direction conguration and in the deformed conguration. Let the rotation of the shell node be given by the nite rotation vector, . Then , , and are related by the equation
Nn
^ ^ where C = exp[] with the skew symmetric matrix form of the rotation vector . See Rotation variables, Section 1.3.1, for a more detailed discussion of nite rotations. This constraint equation does not completely dene the rotation of the shell node: any solution to Equation 6.6.22 can be augmented by a rotation f = f around the line of nodes in the solid, where f can be chosen arbitrarily. Hence, Equation 6.6.22 only constrains the nite rotation vector to within two components. The linearized form of the rotation constraint is
C N = n;
1
(6.6.22)
6.6.21
2 n = dn;
(6.6.23)
where is the linearized rotation. See Rotation variables, Section 1.3.1, for a more detailed discussion of the linearized rotation. We now dene two local directions, s and t, so that s, t, and n form a right-handed, orthonormal, local coordinate system. We then project Equation 6.6.23 onto s and t, which yields ( 2 n) 1 s = dn 1 s; ( 2 n) 1 t = dn 1 t:
With some standard vector algebra these equations can be transformed into the form
t
1 = dn 1 s;
n
s
1 = 0dn 1 t:
(6.6.24)
The change in the normal can be expressed in terms of the displacement difference between the two extreme nodes 1 and n in the continuum:
1 = h 1u 1 (I 0 nn); where 1u = un 0 u1 is the displacement difference and h = n 1 (xn 0 x1 ) is the distance between the 1 1 = h 1u 1 s;
s
1 1 = 0 h 1u 1 t: ( )
(6.6.25)
These constraint equations are formulated in terms of local components of . To obtain the constraint , we choose a basis ei; ej ; ek that is a cyclic permutation of the in terms of global components of global basis e1 ; e2 ; e3 , such that nk ni and nk nj . The constraints can then be written in component form as follows:
i ti i si
1 + t + t = h 1u 1 s; 1 + s + s = 0 h 1u 1 t:
j j k k j j k k
(no summation)
k k
(6.6.26)
The linearized constraints shown above are used directly in geometrically linear analysis, linear perturbations, and ABAQUS/Explicit. In geometrically nonlinear analysis in ABAQUS/Standard, the linearized constraints are used to solve for the general nonlinear constraint Equation 6.6.22 with a Newton method. When the rotation is large, the permutation i; j; k may change to maintain the conditions that nk ni and nk nj .
Reference
General multi-point constraints, Section 20.2.2 of the ABAQUS Analysis Users Manual
6.6.22
REVOLUTE JOINT
6.6.3
REVOLUTE JOINT
A revolute joint is a joint between two nodes in which the rotations of the nodes differ by a relative rotation about an axis that is xed in and, therefore, rotates with the joint. A simple example is a hinge. A revolute joint is implemented in ABAQUS/Standard as a multi-point constraint, dening the total rotation of the constrained (slave) node (the rst node of the MPC), (S ), as the total rotation of the master node (the second node of the MPC), (M ), followed by the relative rotation J , about the axis of the joint :
The joint axis, , also rotates with the rotation of the master node:
a = C ( M ) a 0 :
1
(6.6.31)
Thus, the joint imposes three constraints (each component of the angular velocity of the slave node is constrained) but introduces an additional degree of freedom in the form of the relative rotation J . This means the joint provides a total of two constraints to the model if J is not prescribed and three constraints if it is. The virtual work contribution of the three nodes of the joint is
MM M + M J J = 0; where MS is the total moment at node S , MM is the total moment at node M , and M J is the moment
1
MS
S +
M S + MM )
M + (
MS a + M J )J = 0:
1
If there are no further constraints associated with the nodes of the joint, M and J are independent variations, so the constrained virtual work equation implies that
M S = 0M M
and that
MJ =
0 M S 1 a = M M 1 a:
6.6.31
REVOLUTE JOINT
Because the revolute is implemented in this manner, the relative rotation in the joint J appears as a degree of freedom in the model (degree of freedom 6 at the third node of the MPC). Thus, a moment, M J , can be applied in the joint by specifying its value as a concentrated load; J can be given a prescribed variation in time by specifying a boundary condition; or stiffness and/or damping can be associated with relative rotation of the joint by attaching a spring and/or dashpot to ground to this degree of freedom (a spring or dashpot to ground is used because the variable is a relative rotation).
Reference
General multi-point constraints, Section 20.2.2 of the ABAQUS Analysis Users Manual
6.6.32
UNIVERSAL JOINT
6.6.4
UNIVERSAL JOINT
A universal joint is a joint between two nodes containing orthogonal hinges that provide two axes of relative rotation in the joint. A universal joint is implemented in ABAQUS/Standard as a multi-point constraint, dening the total S rotation of the constrained (slave) node (the rst node of the MPC), ( ), as the total rotation of the master node (the second node of the MPC), ( M ), followed by two relative rotations: 1 about the rst axis of the joint 1 , then 2 about the second axis of the joint 2 (which is orthogonal to 1 ):
The rst joint axis, 1 , rotates with the rotation of the master node:
a1 = C(M ) a10:
1 1 1
The second joint axis has this rotation plus the rotation about the rst joint axis:
a2 = C(1a1) C(M ) a20:
M + a _
_ 1 1 + 2 2 ;
M + 1a1 + 2a2 :
(6.6.41)
Thus, the joint imposes three constraints (each component of the angular velocity of the slave node is constrained) but introduces two additional degrees of freedom in the form of the relative rotations 1 and 2. This means the joint provides a total of one constraint to the model if 1 and 2 are not prescribed or up to three constraints if they are. The virtual work contribution of the joint is where S is the total moment at node S , M is the total moment at node M , and M1 and M2 are the moments in the joint hinges. Applying the constraints (Equation 6.6.41), this is
MS
+
S MM
M + M1 1 + M2 2 = 0;
M + (MS a1 + M1 )1 + (MS a2 + M2 )2 = 0: M If there are no further constraints associated with the nodes of the joint, , 1 and 2 are independent
(
M S + MM )
M S = 0M M ;
6.6.41
UNIVERSAL JOINT
M1 = 0
and
M2
MS 1 a = MM 1 a = 0M S 1 a = M M 1 a :
1 1 2 2
Because the universal joint is implemented in this manner, the relative rotations in the joint, 1 and 2 , appear as degrees of freedom in the model (degree of freedom 6 at the third and fourth nodes of the MPC). Moments M1 and M2 can, therefore, be applied in the joint by specifying their values as concentrated loads; 1 and 2 can be given prescribed variations in time by specifying boundary conditions; or stiffness and/or damping can be associated with relative rotations of the joint by attaching springs and/or dashpots to ground to these degrees of freedom (springs or dashpots to ground are used because the variables are relative rotations).
Reference
General multi-point constraints, Section 20.2.2 of the ABAQUS Analysis Users Manual
6.6.42
6.6.5
The V LOCAL MPC in ABAQUS/Standard constrains the velocity components at the rst MPC node to be equal to the velocity components at the third node along local, rotating, directions. These local directions rotate according to the rotation at the second node. In the initial conguration the rst local direction is from the second to the third node of the MPC. The global z -axis is used if these nodes coincide. The velocity of the rst node u1 , is as follows:
u1 = _
3 X e2u_ 3:
i
=1
1 u1 =
where where
i t
3 X e2 + 1 1 u3 ;
i
=1
i t
1 2
e2 + 1 1 def C + 1 1 e2 ; = 2 2
t t t
1
i t
C + 1 1 = C( 1 12) 2 2
t t
is the increment of rotation dened by half of the magnitude of the increment of rotation at the second node of the constraint, 2 , and 2t , i ; ; , are the local directions at the beginning of the increment. i
=1 2 3
Reference
General multi-point constraints, Section 20.2.2 of the ABAQUS Analysis Users Manual
6.6.51
KINEMATIC COUPLING
6.6.6
KINEMATIC COUPLING
A kinematic coupling constrains a group of slave nodes to the translation and rotation of a master node in a customized manner; combinations of slave node degrees of freedom are selected to participate in the constraint. Since each slave node has a separate relationship with the master node, the kinematic coupling constraint can be considered as the combination of general master-slave constraints. These general constraints are described below.
Rotations
Each possible combination of selected constraints on a slave node rotation results in rotation relationships that are unique but analogous to existing MPC types: zero constrained rotational degrees of freedom results in a pin constraint; one, results in a universal constraint; two, results in a revolute constraint; and three, results in a beam constraint. Each of these combinations is treated according to the appropriate theory employed in ABAQUS/Standard. To implement these constraints, an additional node is created internally for each slave node. For example, for revolute and universal constraints this additional node is used in a similar manner to the nodes required in the specication of the analog MPC types.
Translations
The additional internal node, described above, also reects the motion of the slave node, relative to that of a fully constrained slave node in the following manner. Let m be the position of the master node and s be the position of the slave node in the reference conguration. The reference conguration position of the slave node with respect to the master node is then
N =X 0X
s
^ x =x
s
n;
m . The selectively
(6.6.61)
n = C ( ) 1 N;
m m
where C ( ) is the rotation matrix associated with the master node rotation, constrained slave node position can be described as
x e
= ^ + yi
s
e;
i
where yi are the translation degrees of freedom at the additional node and i are the current conguration base vectors, which rotate from the reference global Cartesian base vectors i according to
C (m ) 1 ei :
6.6.61
KINEMATIC COUPLING
This base vector rotation is made regardless of the choice of rotation constraint at the slave node. Constraint or release of slave node translation degree of freedom i can now be described as the constraint (yi = 0) or release of translation degrees of freedom on the additional node. With these constraints on yi , Equation 6.6.61 can be used to dene the constraint equations.
Linearized form
+ n + yi ei + yi ei
where r = xs 0 xm .
Second-order form
= x
+
2 r + y e ;
i i
(6.6.62)
0( 1 d
m
)r + 2 (
1 r)d
+ 2 (d
1 r)
+ yi d
2e
+ dyi
2e:
i
(6.6.63)
Reference
Kinematic coupling constraints, Section 20.2.3 of the ABAQUS Analysis Users Manual
6.6.62
Chapter 7 References
References 7.1
REFERENCES
7.1.1
REFERENCES
Agah-Tehrani, A., E. H. Lee, R. L. Mallett, and E. T. Onat, The Theory of Elastic-Plastic Deformation at Finite Strain with Induced Anisotropy Modeled as Combined Isotropic-Kinematic Hardening, Metal Forming Report, Rensselaer Polytechnic Institute, Troy, New York, 1986. Al-Ani, A. M., and J. W. Hancock, J-Dominance of Short Cracks in Tension and Bending, Journal of the Mechanics and Physics of Solids, vol. 39, pp. 2343, 1991. Allik, H., The Application of Finite and Innite Elements to Problems in Structural Acoustics, Computational Mechanics 91: Proceedings of the International Conference on Computational Engineering Science, ICES Publications, Atlanta, Georgia, 1991. Anagnastopoulos, S. A., Response Spectrum Techniques for Three Component Earthquake Design, Earthquake Engineering and Structural Dynamics, vol. 9, pp. 459476, 1981. Aoki, S., K. Kishimoto, and M. Sakjata, Crack Tip Stress and Strain Singularity in Thermally Loaded Elastic-Plastic Material, Transactions of the ASME, Journal of Applied Mechanics, vol. 48, no. 2, pp. 428429, 1981. Aravas, N., On the Numerical Integration of a Class of Pressure-Dependent Plasticity Models, International Journal for Numerical Methods in Engineering, vol. 24, pp. 13951416, 1987. Arruda, E. M., and M. C. Boyce, A Three-Dimensional Constitutive Model for the Large Stretch Behavior of Rubber Elastic Materials, Journal of the Mechanics and Physics of Solids, vol. 41, no. 2, pp. 389412, 1993. Ashwell, D. G., and R. H. Gallagher, Editors, Finite Elements for Thin Shells and Curved Members, John Wiley and Sons, London, 1976. Astley, R. J., G. J. Macaulay, and J. P. Coyette, Mapped Wave Envelope Elements for Acoustical Radiation and Scattering, Journal of Sound and Vibration, vol. 170, pp. 97118, 1994. Atomic Energy Commission Regulatory Guide 1.60, Design Response Spectra for Seismic Design of Nuclear Power Plants. Atomic Energy Commission Regulatory Guide 1.92, Combining Modal Responses. Barlow, J., Optimal Stress Locations in Finite Element Models, International Journal for Numerical Methods in Engineering, vol. 10, pp. 243251, 1976. Barnett, D. M., and R. J. Asaro, The Fracture Mechanics of Slit-Like Cracks in Anisotropic Elastic Media, Journal of the Mechanics and Physics of Solids, vol. 20, pp. 353366, 1972. Bathe, K. J., and C. A. Almeida, A Simple and Effective Pipe Elbow ElementLinear Analysis, Transactions of the ASME, Journal of Applied Mechanics, vol. 47, no. 1, 1980. Bathe, K. J., and E. N. Dvorkin, A Continuum Mechanics-Based Four-Node Shell Element for General Non-Linear Analysis, International Journal of Computer Aided Engineering Software, vol. 1, 1984.
References
7.1.11
REFERENCES
Bathe, K. J., and E. L. Wilson, Large Eigenvalue Problems in Dynamic Analysis, Proceedings of the ASCE, EM6, 98, pp. 14711485, 1972. Batoz, J. L., K. J. Bathe, and L. W. Ho, A Study of Three-Node Triangular Plate Bending Elements, International Journal for Numerical Methods in Engineering, vol. 15, pp. 17711821, 1980. Bayliss, A., M. Gunzberger, and E. Turkel, Boundary Conditions for the Numerical Solution of Elliptic Equations in Exterior Regions, SIAM Journal of Applied Mathematics., vol. 42, no. 2, pp. 430451, 1982. Bear, J., Dynamics of Fluids in Porous Media, American Elsevier Publishing Company, Dover, New York, 1972. Belytschko, T., Survey of Numerical Methods and Computer Programs for Dynamic Structural Analysis, Nuclear Engineering and Design, vol. 37, pp. 2334, 1976. Belytschko, T., and L. P. Bindeman, Assumed Strain Stabilization of the Eight Node Hexahedral Element, Computer Methods in Applied Mechanics and Engineering, vol. 105, pp. 225260, 1993. Belytschko, T., J. I. Lin, and C. S. Tsay, Explicit Algorithms for the Nonlinear Dynamics of Shells, Computer Methods in Applied Mechanics and Engineering, vol. 43, pp. 251276, 1984. Belytschko, T., B. L. Wong, and H. Y. Chiang, Advances in One-Point Quadrature Shell Elements, Computer Methods in Applied Mechanics and Engineering, vol. 96, pp. 93107, 1992. Bergan, P. B., G. Horrigmoe, B. Krakeland, and T. H. Soreide, Solution Techniques for Non-Linear Finite Element Problems, International Journal for Numerical Methods in Engineering, vol. 12, pp. 16771696, 1978. Bergstrm, J. S., and M. C. Boyce, Constitutive Modeling of the Large Strain Time-Dependent Behavior of Elastomers, Journal of the Mechanics and Physics of Solids, vol. 46, pp. 931954, 1998. Betegn, C., and J. W. Hancock, Two-Parameter Characterization of Elastic-Plastic Crack-Tip Fields, Journal of Applied Mechanics, vol. 58, pp. 104110, 1991. Bettess, P., C. Emson, and T. C. Chaim, A New Mapped Innite Element for Exterior Wave Problems, Numerical Methods in Coupled Systems, Edited by R. W. Lewis et al, John WIley & Sons, 1984. Bilby, B. A., G. E. Goldthorpe, and I. C. Howard, A Finite Element Investigation of the Effect of Specimen Geometry on the Fields of Stress and Strain at the Tip of Stationary Cracks, Size Effects in Fracture, Institution of Mechanical Engineers, London, pp. 3746, 1986. Bilkhu, S., Private communication, 1987. Budiansky, B., and J. L. Sanders, On the Best First-Order Linear Shell Theory, Progress in Applied Mechanics, The Prager Anniversary Volume, Macmillan, London, pp. 129140, 1963. Burnett, D. S., A Three-Dimensional Acoustic Innite Element Based on a Prolate Spheroidal Multipole Expansion, Journal of the Acoustical Society of America, vol. 96(5), pp. 27982816, 1994. Calladine, C. R., A Microstructural View of the Mechanical Properties of Saturated Clay, Geotechnique, vol. 21, no. 4, pp. 391415.
7.1.12
REFERENCES
Chen, W. F., and D. J. Han, Plasticity for Structural Engineers, Springer-Verlag, New York, 1988. Chu, C. C., and A. Needleman, Void Nucleation Effects in Biaxially Stretched Sheets, Journal of Engineering Materials and Technology, vol. 102, pp. 249256, 1980. Clough, R. W., and J. Penzien, Dynamics of Structures, McGraw-Hill, New York, 1975. Cofn, L. F., Jr., The Flow and Fracture of a Brittle Material, Journal of Applied Mechanics, vol. 72, pp. 233248, 1950. Cohen, M., and P. C. Jennings, Silent Boundary Methods for Transient Analysis (in Computational Methods for Transient Analysis), Ed. T. Belytschko and T. R. J. Hughes, Elsevier, 1983.
References
Cormeau, I., Numerical Stability in Quasi-Static Elasto-Visco Plasticity, International Journal for Numerical Methods in Engineering, vol. 9, pp. 109127, 1975. Cotterell, B., and J. R. Rice, Slightly Curved or Kinked Cracks, International Journal of Fracture, vol. 16, pp. 155169, 1980. Cowper, G. R., Gaussian Quadrature for Triangles, International Journal for Numerical Methods in Engineering, vol. 7, pp. 405408, 1973. Crank, J., The Mathematics of Diffusion, Clarendon Press, Oxford, 1956. Criseld, M. A., A Fast Incremental/Iteration Solution Procedure that Handles Snap-Through, Computers and Structures, vol. 13, pp. 5562, 1981. Criseld, M. A., Snap-Through and Snap-Back Response in Concrete Structures and the Dangers of Under-Integration, International Journal for Numerical Methods in Engineering, vol. 22, pp. 751767, 1986. DeGroot, S. R., and P. Mazur, Non Equilibrium Thermodynamics, North Holland Publishing Company, North Holland, Amsterdam, 1962. Desai, C. S., Finite Element Methods for Flow in Porous Media in Finite Elements in Fluids, vol. 1, Wiley, London, pp. 157181, 1975. Deshpande, V. S., and N. A. Fleck, Isotropic Constitutive Model for Metallic Foams, Journal of the Mechanics and Physics of Solids, vol. 48, pp. 12531276, 2000. Dodge, W. G., and S. E. Moore, Stress Indices and Flexibility Factors for Moment Loadings on Elbows and Curved Pipes, Welding Research Council Bulletin, no. 179, December 1972. Drucker, D. C., and W. Prager, Soil Mechanics and Plastic Analysis or Limit Design, Quarterly of Applied Mathematics, vol. 10, pp. 157165, 1952. Du, Z. -Z., and J. W. Hancock, The Effect of Non-Singular Stresses on Crack-Tip Constraint, Journal of the Mechanics and Physics of Solids, vol. 39, pp. 555567, 1991. Dupuis, G., Stabilit Elastique des Structures Unidimensionelles, ZAMP, vol. 20, pp. 94106, 1969. Eleiche, A. S. M., A Literature Survey of the Combined Effects of Strain Rate and Elevated Temperature on the Mechanical Properties of Metals, Air Force Materials Laboratories, Report AFMLTR72125, 1972.
7.1.13
REFERENCES
Engelmann, B. E., and R. G. Whirley, A New Explicit Shell Element Formulation for Impact Analysis (in Computational Aspects of Contact, Impact and Penetration), Ed. R. F. Kulak and L. E. Schwer, Elmepress International, 1990. Engquist, B., and A. Majda, Absorbing Boundary Conditions for the Numerical Simulation of Waves, Mathematics of Computation, vol. 31, pp. 629651, 1977. Ericsson, T., and A. Ruhe, The Spectral Transformation Lanczos Method for the Numerical Solution of Large Sparse Generalized Symmetric Eigenvalue Problems, Mathematics of Computation, vol. 35, pp. 12511268, 1980. Erdogan, F., and G. C. Sih, On the Crack Extension in Plates under Plane Loading and Transverse Shear, Journal of Basic Engineering, vol. 85, pp. 519527, 1963. Faltinsen, O. M., Sea Loads on Ships and Offshore Structures, Cambridge University Press, 1990. Farin, G., Curves and Surfaces for Computer Aided Geometric Design, Academic Press, San Diego, Second Edition edition, 1990. Farin, G., Smooth Interpolation to Scattered 3D Data in Surfaces in Computer Aided Geometric Design, Barnhill, R. E. and Boehm, W., eds., North-Holland, pp. 4363, 1983. Flanagan, D. P., and T. Belytschko, A Uniform Strain Hexahedron and Quadrilateral with Orthogonal Hourglass Control, International Journal for Numerical Methods in Engineering, vol. 17, pp. 679706, 1981. Flugge, W., Tensor Analysis and Continuum Mechanics, Springer-Verlag, New York, 1972. Gao, H., M. Abbudi, and D. M. Barnett, Interfacial Crack-Tip Fields in Anisotropic Elastic Solids, Journal of the Mechanics and Physics of Solids, vol. 40, pp. 393416, 1992. Geers, T. L., and K. S. Hunter, An Integrated Wave-Effects Model for an Underwater Explosion Bubble, Journal of the Acoustical Society of America, vol. 111(4), pp. 15841601, 2002. Gibson, L. J., and M. F. Ashby, The Mechanics of Three-Dimensional Cellular Materials, Proceedings of the Royal Society, London, A 382, pp. 4359, 1982. Gibson, L. J., M. F. Ashby, G. S. Schajer, and C. I. Robertson, The Mechanics of Two-Dimensional Cellular Materials, Proceedings of the Royal Society, London, A 382, pp. 2542, 1982. Gordon, J. L., OUTCUR: An Automated Evaluation of Two-Dimensional Finite Element Stresses According to ASME Section III Stress Requirements, Paper No. 76WA/PVP-16, ASME Winter Annual Meeting, December 1976. Green, A. E., and P. M. Naghdi, A General Theory of an Elastic-Plastic Continuum, Archives of Rational Mechanics and Analysis, vol. 17. Grimes, R. G., J. G. Lewis, and H. D. Simon, A Shifted Block Lanczos Algorithm for Solving Sparse Symmetric Generalized Eigenproblems, SIAM Journal on Matrix Analysis and Applications, vol. 15, pp. 228272, 1994. Grote, M., and J. Keller, On Nonreecting Boundary Conditions, Journal of Computational Physics, vol. 122, pp. 231243, 1995.
7.1.14
REFERENCES
Gudehus, G., Editor, Finite Elements in Geomechanics, Wiley and Sons, 1977. Gupta, A. K., Response Spectrum Method in Seismic Analysis and Design of Structures, Blackwell Scientic Publications, 1990. Gurson, A. L., Continuum Theory of Ductile Rupture by Void Nucleation and Growth: Part IYield Criteria and Flow Rules for Porous Ductile Materials, Journal of Engineering Materials and Technology, vol. 99, pp. 215, 1977. Hansen, H. T., Nonlinear Static and Dynamic Analysis of Slender Structures Subjected to Hydrodynamic Loading, University of Trondheim, Norway, 1988.
References
Hayashi, K., and S. Nemat-Nasser, Energy-Release Rate and Crack Kinking under Combined Loading, Journal of Applied Mechanics, vol. 48, pp. 520524, 1981. He, M. -Y., and J. W. Hutchinson, Kinking of a Crack out of an Interface: Tabulated Solution Coefcients, Harvard University, Cambridge, Massachusetts, Division of Applied Mechanics, 1989. Hegemier, G. A., Evaluation of Models for MX Siting, Volume IIReinforced Concrete Models, Systems, Science and Software, Report SSSR804155, La Jolla, California, 1979. Hegemier, G. A., and K. J. Cheverton, Evaluation of Reinforced Concrete Models for Nuclear Power Plant Application, EPRI, Palo Alto, California, 1980. Hibbitt, H. D., Some Follower Forces and Load Stiffness, International Journal for Numerical Methods in Engineering, vol. 14, pp. 937941, 1979. Hibbitt, H. D., Special Structural Elements of Piping Analysis, Pressure Vessels and Piping; Analysis and Computers, ASME, New York, 1974. Hibbitt, H. D., and B. I. Karlsson, Analysis of Pipe Whip, EPRI, Report NP1208, 1979. Hilber, H. M., T. J. R. Hughes, and R. L. Taylor, Collocation, Dissipation and Overshoot for Time Integration Schemes in Structural Dynamics, Earthquake Engineering and Structural Dynamics, vol. 6, pp. 99117, 1978. Hilderbrand, F. B., Introduction to Numerical Analysis, McGraw Hill, 1974. Hilleborg, A., M. Modeer, and P. E. Petersson, Analysis of Crack Formation and Crack Growth in Concrete by Means of Fracture Mechanics and Finite Elements, Cement and Concrete Research, vol. 6, pp. 773782, 1976. Hjelm, H. E., Elastoplasticity of Grey Cast Iron, FE-Algorithms and Biaxial Experiments, Ph.D. Thesis, Chalmers University of Technology, Division of Solid Mechanics, Sweden, 1992. Hjelm, H. E., Yield Surface for Grey Cast Iron under Biaxial Stress, Journal of Engineering Materials and Technology, vol. 116, pp. 148154, 1994. Hoenig, A., Near-Tip Behavior of a Crack in a Plane Anisotropic Elastic Body, Engineering Fracture Mechanics, vol. 16, pp. 393403, 1982. Holman, J. P., Heat Transfer, McGraw-Hill, 1990.
7.1.15
REFERENCES
Hughes, T. J. R., and A. Brooks, A Theoretical Framework for Petrov-Galerkin Methods with Discontinuous Weighting Functions in Finite Elements in Fluids, vol. 4, edited by R. H. Gallagher, D. H. Norrie, J. T. Oden, and O. C. Zienkiewicz, John Wiley & Sons, 1982. Hughes, T. J. R., and W. K. Liu, Nonlinear Finite Element Analysis of Shells, Part I: Three-Dimensional Shells, California Institute of Technology, Pasadena, California, 1980. Hughes, T. J. R., R. L. Taylor, and W. Kanoknukulchai, A Simple and Efcient Element for Plate Bending, International Journal for Numerical Methods in Engineering, vol. 11, no. 10, pp. 15291543, 1977. Hughes, T. J. R., and T. E. Tezduyar, Finite Elements based upon Mindlin Plate Theory with Particular Reference to the Four-Node Bilinear Isoparametric Element, Journal of Applied Mechanics, pp. 587596, 1981. Hughes, T. J. R., and J. Winget, Finite Rotation Effects in Numerical Integration of Rate Constitutive Equations Arising in Large Deformation Analysis, International Journal for Numerical Methods in Engineering, vol. 15, pp. 18621867, 1980. Hurty, W. C., and M. F. Rubinstein, Dynamics of Structures, Prentice-Hall, Englewood Cliffs, New Jersey, 1964. Hussain, M. A., S. L. Pu, and J. Underwood, Strain-Energy-Release Rate for a Crack under Combined Mode I and Mode II, ASTM-STP-560, pp. 228, 1974. Hutchinson, J. W., Singular Behaviour at the End of a Tensile Crack in a Hardening Material, Journal of the Mechanics and Physics of Solids, vol. 16, pp. 1331, 1968. Hutchinson, J. W., and Z. Suo, Mixed Mode Cracking in Layered Materials, Advances in Applied Mechanics, vol. 29, pp. 63191, 1992. Irons, B., and S. Ahmad, Techniques of Finite Elements, Ellis Horwood Limited, Halsted Press, John Wiley and Sons, Chichester, England, 1980. Ikeda, T., Fundamentals of Piezoelectricity, Oxford University Press, New York, 1990. Johnson, D., Surface to Surface Radiation in the Program TAU, Taking Account of Multiple Reection, United Kingdom Atomic Energy Authority Report ND-R-1444(R), 1987. Kaliske, M., and H. Rothert, On the Finite Element Implementation of Rubber-like Materials at Finite Strains, Engineering Computations, vol. 14, no. 2, pp. 216232, 1997. Kawabata, S., Y. Yamashita, H. Ooyama, and S. Yoshida, Mechanism of Carbon-Black Reinforcement of Rubber Vulcanizate, Rubber Chemistry and Technology, vol. 68, no. 2, pp. 311329, 1995. Der Kiureghian, A., A Response Spectrum Method For Random Vibration Analysis of MDF Systems, Earthquake Engineering and Structural Dynamics, vol. 9, pp. 419435, 1981. Kennedy, J. M., T. Belytschko, and J. I. Lin, Recent Developments in Explicit Finite Element Techniques and Their Applications to Reactor Structures, Nuclear Engineering and Design, vol. 96, pp. 124, 1986.
7.1.16
REFERENCES
Kfouri, A. P., Some Evaluations of the Elastic T-Term Using Eshelbys Method, International Journal of Fracture, vol. 30, pp. 301315, 1986. Kilian, H.-G., Equation of State of Real Networks, Polymer, vol. 22, pp. 209216, 1981. Kilian, H.-G., H. F. Enderle, and K. Unseld, The Use of the van der Waals Model to Elucidate Universal Aspects of Structure-Property Relationships in Simply Extended Dry and Swollen Rubbers, Colloid and Polymer Science, vol. 264, pp. 866876, 1986. Kleiber, M., H. Antnez, T. D. Hien, and P. Kowalczyk, Parameter Sensitivity in Nonlinear Mechanics, John Wiley & Sons, 1997.
References
Kroenke, W. C., Classication of Finite Element Stresses According to ASME Section III Stress Categories, ASME 94th Winter Annual Meeting, November 1973. Kroenke, W. C., G. W. Addicott, and B. M. Hinton, Interpretation of Finite Element Stresses According to ASME Section III, Second National Congress on Pressure Vessels and Piping, June 1975. Kumar, V., M. D. German, and C. F. Shih, An Engineering Approach for Elastic-Plastic Fracture Analysis, Electric Power Research Institute Report NP-1931, Project 1237-1, Electric Power Research Institute, Palo Alto, CA, 1981. Kwata, K., S. Asai, and H. Takeda, A Solution for the IAEA International Piping Benchmark Problem, Century Research Corporation, Tokyo, 1978. Lamb, H., Hydrodynamics, Dover Publications, New York, 1945. Larsson, S. G., and A. J. Carlsson, Inuence of Non-Singular Stress Terms and Specimen Geometry on Small-Scale Yielding at Crack Tips in Elastic-Plastic Material, Journal of the Mechanics and Physics of Solids, vol. 21, pp. 263278, 1973. Lee, J., and G. L. Fenves, Plastic-Damage Model for Cyclic Loading of Concrete Structures, Journal of Engineering Mechanics, vol. 124, no. 8, pp. 892900, 1998. Lemaitre, J., and J.-L. Chaboche, Mechanics of Solid Materials, Cambridge University Press, 1990. Leppington, F., Inverse Scattering in Modern Methods in Analytical Acoustics, edited by D. G. Crighton, et al., Springer Verlag, pp. 551564, 1992. Lianis, G., Small Deformations Superposed on an Initial Large Deformation in Viscoelastic Bodies, Proceedings of the 4th International Congress on Rheology, part 2, pp. 109119, Interscience, New York, 1965. Lin, C. S., and A. C. Scordelis, Nonlinear Analysis of RC Shells of General Form, ASCE, Journal of Structural Engineering, vol. 101, pp. 523538, 1975. Lindholm, J. S., and R. L. Besseny, A Survey of Rate Dependent Strength Properties of Metals, Air Force Materials Laboratory, Report AFMLTR69119, 1969. Lubliner, J., J. Oliver, S. Oller, and E. Oate, A Plastic-Damage Model for Concrete, International Journal of Solids and Structures, vol. 25, no. 3, pp. 229326, 1989.
7.1.17
REFERENCES
Lysmer, J., and R. L. Kuhlemeyer, Finite Dynamic Model for Innite Media, Journal of the Engineering Mechanics Division of the ASCE, pp. 859877, August 1969. MacNeal, R. H., A Simple Quadrilateral Shell Element, Computers and Structures, vol. 8, pp. 175183, 1978. MacNeal, R. H., Derivation of Element Stiffness Matrices by Assumed Strain Distributions, Nuclear Engineering and Design, vol. 70, pp. 312, 1982. Maiti, S. K., L. J. Gibson, and M. F. Ashby, Deformation and Energy Absorption Diagrams for Cellular Solids, Acta Metallurgica, vol. 32, no. 11, pp. 19631975, 1984. Maiti, S. K., and R. A. Smith, Comparison of the Criteria for Mixed Mode Brittle Fracture Based on the Preinstability Stress-Strain Field, Part I: Slit and Elliptical Cracks under Uniaxial Tensile Loading, International Journal of Fracture, vol. 23, pp. 281295, 1983. Mang, H. A., Symmetricability of Pressure Stiffness Matrices for Shells with Loaded Free Edges, International Journal for Numerical Methods in Engineering, vol. 15, pp. 981990, 1980. Matthies, H., and G. Strang, The Solution of Nonlinear Finite Element Equations, International Journal for Numerical Methods in Engineering, vol. 14, pp. 16131626, 1979. Mentrey, Ph., and K. J. William, Triaxial Failure Criterion for Concrete and its Generalization, ACI Structural Journal, 92:31118, May/June, 1995. Mills-Curran, W. C., Calculation of Eigenvector Derivatives for Structures with Repeated Eigenvalues, AIAA Journal, vol. 26, no. 7, pp. 867871, 1988. Morman, K. N., Analytical Prediction of Elastomeric Mount Dynamic Characteristics, Report L2143-2, Ford Engineering Computer Systems, Car Engineering, Dearborn, Michigan, 1979. Morman, K. N., and J. C. Nagtegaal, Finite Element Analysis of Sinusoidal Small-Amplitude Vibrations in Deformed Viscoelastic Solids. Part I: Theoretical Development, International Journal of Numerical Methods in Engineering, vol. 19, pp. 10791103, 1983. Morse, P., and K. Ingard, Theoretical Acoustics, McGraw-Hill, 1968. Mullins, L., Effect of Stretching on the Properties of Rubber, Journal of Rubber Research, vol. 16, pp. 275289, 1947. Murff, J. D., Private communication, 1982. Murthy, D. V., and R. T. Haftka, Derivatives of Eigenvalues and Eigenvectors of a General Complex Matrix, International Journal of Numerical Methods in Engineering, vol. 26, pp. 293311, 1988. Nagtegaal, J. C., D. M. Parks, and J. R. Rice, On Numerically Accurate Finite Element Solutions in the Fully Plastic Range, Computer Methods in Applied Mechanics and Engineering, vol. 4, pp. 153177, 1977. Nakamura, T., and D. M. Parks, Antisymmetrical 3-D Stress Field near the Crack Front of a Thin Elastic Plate, International Journal of Solids and Structures, vol. 25, pp. 14111426, 1989.
7.1.18
REFERENCES
Nakamura, T., and D. M. Parks, Determination of Elastic T-Stress along Three-dimensional Crack Fronts Using an Interaction Integral, International Journal of Solids and Structures, vol. 28, pp. 15971611, 1992. Nayak, G. C., and O. C. Zienkiewicz, A Convenient Form of Invariants and its Application to Plasticity, Proceedings of the ASCE, Engineering Mechanics Division, vol. 98, no. St4, pp. 949954, 1972. Needleman, A., and V. Tvergaard, Necking of Biaxially Stretched Elastic-Plastic Circular Plates, Journal of the Mechanics and Physics of Solids, vol. 25, pp. 159183, 1977. Newman, M., and A. Pipano, Fast Modal Extraction in NASTRAN via the FEER Computer Program, NASA TM X-2893, pp. 485506, 1973. Newmark, N. M., J. A. Blume, and K. K. Kapur, Seismic Design Spectra for Nuclear Power Plants, Power Division Journal, ASCE, 99, P02, Proceedings Paper 10142, pp. 287303, 1973. Nguyen, H. V., and D. F. Durso, Absorption of Water by Fiber Webs: an Illustration of Diffusion Transport, Tappi Journal, vol. 66, no. 12, 1983. Nishimura, H., M. Isobe, and K. Horikawa, Higher Order Solutions of Stokes and Cnoidal Waves, Journal of Faculty Engineering, University of Tokyo (B), vol. 34, no. 2, 1977. Nuclear Standard NE F 9-5T, Guidelines and Procedures for Design of Class 1 Elevated Temperature Nuclear System Components, United States Department of Energy, Nuclear Energy Programs, March 1981. Oden, J. T., Mechanics of Elastic Structures, McGraw-Hill, New York, 1967. Oden, J. T., and N. Kikuchi, Finite Element Methods for Constrained Problems in Elasticity, TICOM, University of Texas at Austin, Austin, Texas, 1980. Oden, J. T., and N. Kikuchi, Finite Element Methods for Constrained Problems in Elasticity, Fifth Invitational Symposium on the Unication of Finite ElementsFinite Differences and Calculus of Variations, Edited by H. Kardestuncer, University of Connecticut, Storrs, CT. Oden, J. T., and E. B. Pires, Nonlocal and Nonlinear Friction Laws and Variational Principles for Contact Problems in Elasticity, Journal of Applied Mechanics, vol. 50, pp. 6773, 1983. Ogden, R. W., and D. G. Roxburgh, A Pseudo-Elastic Model for the Mullins Effect in Filled Rubber, Proceedings of the Royal Society of London, Series A, vol. 455, pp. 28612877, 1999. Ohtsubo, H., and O. Watanabe, Flexibility and Stress Factors for Pipe BendsAn Analysis by the Finite Ring Method, ASME paper 76PVP40, 1976. Palaniswamy, K., and W. G. Knauss, On the Problem of Crack Extension in Brittle Solids under General Loading, Mechanics Today, Edited by S. Nemat-Nasser, Vol. 4, Pergamon Press, 1978. Papoulis, A., Probability, Random Variables, and Stochastic Processes, McGraw-Hill, New York, 1965. Parks, D. M., The Virtual Crack Extension Method for Nonlinear Material Behavior, Computer Methods in Applied Mechanics and Engineering, vol. 12, pp. 353364, 1977.
References
7.1.19
REFERENCES
Parks, D. M., Advances in Characterization of Elastic-Plastic Crack-Tip Fields, Topics in Fracture and Fatigue, Edited by A. S. Argon, Springer Verlag, 1992. Parks, D. M., and C. S. White, Elastic-Plastic Line Spring Finite Elements for Surface Cracked Plates and Shells, Journal of Pressure Vessel Technology, vol. 104, pp. 287292, 1982. Parlett, B. N., The Symmetric Eigenvalue Problem, Prentice-Hall, Englewood Cliffs, New Jersey, 1980. Parlett, B. N., and B. Nour-Omid, Toward a Black Box Lanczos Program, Computer Physics Communications, vol. 53, pp. 169179, 1989. Parry, R. H. G., Editor, Stress-Strain Behaviour of Soils, G. T. Foulis and Co., Henley, England, 1972. Penzien, J., and M. Watabe, Characteristic of 3-Dimensional Earthquake Ground Motions, Earthquake Engineering and Structural Dynamics, vol. 3, pp. 365373, 1975. Powell, G., and J. Simons, Improved Iterative Strategy for Nonlinear Structures, International Journal for Numerical Methods in Engineering, vol. 17, pp. 14551467, 1981. Puso, M. A., A Highly Efcient Enhanced Assumed Strain Physically Stabilized Hexahedral Element, International Journal for Numerical Methods in Engineering, vol. 49, pp. 10291064, 2000. Ramaswami, S., Towards Optimal Solution Techniques for Large Eigenproblems in Structural Mechanics, Ph.D. Thesis, MIT, 1979. Ramm, E., Strategies for Tracing the Nonlinear Response Near Limit Points, Nonlinear Finite Element Analysis in Structural Mechanics, Edited by E. Wunderlich, E. Stein, and K. J. Bathe, Springer-Verlag, Berlin, 1981. Rice, J. R., The Line Spring Model for Surface Flaws, The Surface Crack: Physical Problems and Computational Solutions, J. L. Sedlow, Editor, ASME, 1972. Rice, J. R., Continuum Mechanics and Thermodynamics of Plasticity in Relation to Microscale Deformation Mechanisms, Constitutive Equations in Plasticity, Argon, A. S., Editor, MIT Press, Cambridge, Massachusetts, 1975. Rice, J. R., and G. F. Rosengren, Plane Strain Deformation Near a Crack Tip in a Power Law Hardening Material, Journal of the Mechanics and Physics of Solids, vol. 16, pp. 112, 1968. Rodal, J. J. A., and E. A. Witmer, Finite-Strain Large-Deection Elastic-Viscoplastic Finite-Element Transient Response Analysis of Structures, NASA contractor, Report 159874, NASA Lewis, 1979. Rots, J. G., and J. Blaauwendraad, Crack Models for Concrete: Discrete or Smeared? Fixed, Multi-directional or Rotating?, HERON, vol. 34, no. 1, pp. 159, 1989. Sandler, I. S., A New Computational Procedure for Wave Propagation Problems and a New Procedure for Non-Reection Boundaries, Computer Methods in Applied Mechanics and Engineering, vol. 164, pp. 223233, 1998. Scheller, J. D., and R. Mallet, Numerical Evaluation of an Inelastic Piping Elbow Element, ASME Paper 79PVP41, 1979.
7.1.110
REFERENCES
Schoeld, A., and C. P. Wroth, Critical State Soil Mechanics, McGraw-Hill, New York, 1968. Schreyer, H. L., R. F. Kulak, and J. M. Kramer, Accurate Numerical Solutions for Elastic-Plastic Models, Transactions of the ASME, Journal of Pressure Vessel Technology, vol. 101, no. 3, 1979. Shaw, J., and S. Jayasuriya, Modal Sensitivities for Repeated Eigenvalues and Eigenvalue Derivatives, AIAA Journal, vol. 30, no. 3, pp. 850852, 1991. Shih, C. F., and R. J. Asaro, Elastic-Plastic Analysis of Cracks on Bimaterial Interfaces: Part ISmall Scale Yielding, Journal of Applied Mechanics, pp. 299316, 1988. Shih, C. F., B. Moran, and T. Nakamura, Energy Release Rate along a Three-Dimensional Crack Front in a Thermally Stressed Body, International Journal of Fracture, vol. 30, pp. 79102, 1986. Siegel, R., and J. R. Howell, Thermal Radiation Heat Transfer, McGraw Hill, New York, 1980. Sih, G. C., Some Basic Problems in Fracture Mechanics and New Concepts, Engineering Fracture Mechanics, vol. 5, pp. 365377, 1973. Sih, G. C., P. C. Paris, and G. R. Irwin, On Cracks in Rectilinearly Anisotropic Bodies, International Journal of Fracture Mechanics, vol. 1, pp. 189202, 1965. Simo, J. C., On a Fully Three-Dimensional Finite-Strain Viscoelastic Damage Model: Formulation and Computational Aspects, Computer Methods in Applied Mechanics and Engineering, vol. 60, pp. 153173, 1987. Simo, J. C., (Symmetric) Hessian for Geometrically Nonlinear Models in Solid Mechanics: Intrinsic Denition and Geometric Interpretation, Computer Methods in Applied Mechanics and Engineering, vol. 96, no. 2, pp. 189200, 1992. Simo, J. C., and F. Armero, Geometrically Nonlinear Enhanced Strain Mixed Methods and the Method of Incompatible Modes, International Journal for Numerical Methods in Engineering, vol. 33, pp. 14131449, 1992. Simo, J. C., D. D. Fox, and M. S. Rifai, On a Stress Resultant Geometrically Exact Shell Model. Part III: Computational Aspects of the Nonlinear Theory, Computer Methods in Applied Mechanics and Engineering, vol. 79, pp. 2170, 1989. Simo, J. C., and M. S. Rifai, A Class of Assumed Strain Methods and the Method of Incompatible Modes, International Journal for Numerical Methods in Engineering, vol. 29, pp. 15951638, 1990. Simon, H. D., The Lanczos Algorithm with Partial Reorthogonalization, Mathematics of Computation, vol. 42, pp. 115142, 1984. Skjelbreia, L., and J. Hendrickson, Fifth Order Gravity Wave Theory, Proceedings of the 87th Coastal England Conference, Chapter Ten, den Haag, 1960. Smeby, W., and A. Der Kiureghian, Modal Combination Rules for Multicomponent Earthquake Excitation, Earthquake Engineering and Structural Dynamics, vol. 12, pp. 112, 1984. Sobel, L. H., In-Plane Bending of Elbows, Computers and Structures, vol. 7, pp. 701715, 1977.
References
7.1.111
REFERENCES
Sofronis, P., and R. M. McMeeking, Numerical Analysis of Hydrogen Transport Near a Blunting Crack Tip, Journal of the Mechanics and Physics of Solids, vol. 37, no. 3, pp. 317350, 1989. Spence, J., An Upper Bound Analysis for the Deformation of Smooth Pipe Bends in Creep, Second IUTAM Symposium on Creep in Structures, Goteberg, 1970, Springer Verlag, 1972. Spring, K. W., Euler Parameters and the Use of Quaternion Algebra in the Manipulation of Finite Rotations: a Review, Mechanisms and Machine Theory, vol. 21, pp. 365373, 1986. Storkers, B., On Material Representation and Constitutive Branching in Finite Compressible Elasticity, Journal of the Mechanics and Physics of Solids, vol. 34, no. 2, pp. 125145, 1986. Strang, G., Linear Algebra and Its Applications, Academic Press, New York, 1976. Strang, G., and G. Fix, An Analysis of the Finite Element Method, Prentice-Hall, New York, 1973. Stroh, A. N., Dislocation and Cracks in Anisotropic Elasticity, Philosophical Magazine, vol. 7, pp. 625646, 1958. Stroud, A. H., Approximate Calculation of Multiple Integrals, Prentice-Hall, 1971. Suo, Z., Singularities, Interfaces and Cracks in Dissimilar Anisotropic Media, Proceedings of the Royal Society, London, A, vol. 427, pp. 331358, 1990. Sutter, T. R., C. J. Camarda, J. L. Walsh, and H. M. Adelman, Comparison of Several Methods for Calculating Vibration Mode Shape Derivatives, AIAA Journal, vol. 26, no. 12, pp. 15061511, 1988. Symonds, P. S., Survey of Methods of Analysis for Plastic Deformation of Structures under Dynamic Loading, Brown University, Division of Engineering Report, June 1967. Symonds, P. S., T. C. T. Ting, and D. N. Robinson, Survey of Progress in Plastic Wave Propagation in Solid Bodies, Brown University, Division of Engineering Report, June 1967. Tada, H., C. Paris, and G. R. Irwin, The Stress Analysis of Cracks Handbook, Del Research Corporation, Hellertown, Pennsylvania, 1973. Takeda, H., S. Asai, and K. Kwata, A New Finite Element for Structural Analysis of Piping Systems, Proceedings of the Fifth SMIRT Conference, Berlin, 1979. Tanaka, T., and D. J. Fillmore, Kinetics of Swelling of Gels, Journal of Chemical Physics, vol. 70, pp. 12141218, 1979. Tariq, S. M., Evaluation of Flow Characteristics of Perforations Including Nonlinear Effects With the Finite Element Method, SPE Production Engineering, pp. 104112, May 1987. Taylor, R. L., P. J. Beresford, and E. L. Wilson, A Nonconforming Element for Stress Analysis, International Journal for Numerical Methods in Engineering, vol. 10, pp. 12111219, 1976. Theocaris, P. S., A Higher Order Approximation for the T-Criterion of Fracture in Biaxial Fields, Engineering Fracture Mechanics, vol. 19, pp. 975991, 1984. Thompson, C. J., Classical Equilibrium Statistical Mechanics, Oxford University Press, New York, 1988.
7.1.112
REFERENCES
Timoshenko, S. P., Strength of Materials; Part III, Advanced Theory and Problems, D. Van Nostrand Co., Princeton, New Jersey, Third Edition edition, 1956. Treloar, L. R. G., The Physics of Rubber Elasticity, Clarendon Press, Oxford, Third Edition edition, 1975. Tvergaard, V., Inuence of Voids on Shear Band Instabilities under Plane Strain Condition, International Journal of Fracture Mechanics, vol. 17, pp. 389407, 1981. Twizell, E. H., and R. W. Ogden, Non-linear Optimization of the Material Constants in Ogdens Stress-Deformation Function for Incompressible Isotropic Elastomers, Journal of the Australian Mathematical Society, Series B 24, pp. 424434, 1986.
References
Wang, B. P., Improved Approximate Methods for Computing Eigenvector Derivatives in Structural Dynamics, AIAA Journal, vol. 29, no. 6, pp. 10181020, 1991. Wang, Y. -Y., A Two-Parameter Characterization of Elastic-Plastic Crack Tip Fields and Application to Cleavage Fracture, Ph.D. Thesis, Department of Mechanical Engineering, MIT, Cambridge, MA, 1991. Weber, G., and L. Anand, Finite Deformation Constitutive Equations and a Time Integration Procedure for Isotropic, Hyperelastic-Viscoplastic Solids, Computer Methods in Applied Mechanics and Engineering, vol. 79, pp. 173202, 1990. Whitham, G. B., Linear and Nonlinear Waves, John Wiley & Sons, 1974. Wilkinson, J. H., The Algebraic Eigenvalue Problem, Oxford University Press, Oxford, 1965. Williams, M. L., On the Stress Distribution at the Base of a Stationary Crack, Journal of Applied Mechanics, vol. 24, pp. 109114, 1957. Wilson, E. L., A. Der Kiureghian, and E. P. Bayo, A Replacement for the SRSS Method in Seismic Analysis, Earthquake Engineering and Structural Dynamics, vol. 9, pp. 187192, 1981. Wilson, E. L., R. L. Taylor, W. P. Doherty, and J. Ghaboussi, Incompatible Displacement Models in Numerical and Computer Models in Structural Mechanics, Eds. S. F. Fenves, N. Perrone, A. R. Robinson, and W. C. Schnobrich, Academic Press, New York, 1973. Wu, T. H., Soil Mechanics, Allyn and Bacon, Boston, 1976. Xu, G., F. Bower, and M. Ortiz, An Analysis of Non-Planar Crack Growth under Mixed Mode Loading, International Journal of Solids and Structures, vol. 31, pp. 21672193, 1994. Yeoh, O. H., Some Forms of the Strain Energy Function for Rubber, Rubber Chemistry and Technology, vol. 66, pp. 754771, 1993. Yu, C. C., and J. C. Heinrich, Petrov-Galerkin Methods for the Time-Dependent Convective Transport Equation, International Journal for Numerical Methods in Engineering, vol. 23, pp. 883902, 1986. Yu, C. C., and J. C. Heinrich, Petrov-Galerkin Method for Multidimensional, Time-Dependent, Convective-Diffusive Equations, International Journal for Numerical Methods in Engineering, vol. 24, pp. 22012215, 1987.
7.1.113
REFERENCES
Zhong, Z. H., Contact Problems with Friction, Proceedings of Numiform 89, pp. 599606, Balkema, Rotterdam, 1989. Zienkiewicz, O. C., The Finite Element Method, McGraw-Hill, London, Third Edition edition, 1977. Zienkiewicz, O. C., C. Emson, and P. Bettess, A Novel Boundary Innite Element, International Journal for Numerical Methods in Engineering, vol. 19, pp. 393404, 1983. Zienkiewicz, O. C., and G. N. Pande, Time Dependent Multilaminate Model of RocksA Numerical Study of Deformation and Failure of Rock Masses, International Journal for Numerical and Analytical Methods in Geomechanics, vol. 1, pp. 219247, 1977.
7.1.114