Notes On Production Engineering

Lecture Notes in Production Engineering
For further volumes:

http://www.springer.com/series/10642
Berend Denkena and Ferdinand Hollmann (Eds.)
Process Machine Interactions

Predicition and Manipulation of Interactions
between Manufacturing Processes
and Machine Tool Structures
ABC
Editors
Prof. Dr.-Ing. Berend Denkena Dr.-Ing. Ferdinand Hollmann
Leibniz Universität Hannover German Research Foundation
Coordinator of Priority Program 1180 Program Director Engineering Science
Garbsen Bonn
Germany Germany
ISSN 2194-0525 e-ISSN 2194-0533

ISBN 978-3-642-32447-5 e-ISBN 978-3-642-32448-2
DOI 10.1007/978-3-642-32448-2
Springer Heidelberg New York Dordrecht London
Library of Congress Control Number: 2012947591
c Springer-Verlag Berlin Heidelberg 2013

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’s location, in its current version, and permission for use must always be obtained from Springer.
Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations
are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of pub-
lication, neither the authors nor the editors nor the publisher can accept any legal responsibility for any
errors or omissions that may be made. The publisher makes no warranty, express or implied, with respect
to the material contained herein.
Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)
Preface
The Priority Program 1180 “Prediction and Manipulation of Interactions between

Structure and Process” was funded by the German Research Foundation DFG from
2005 to 2012. It was initiated by researchers from the German Academic Society
for Production Engineering WGP and accompanied by a working group on process
machine interaction within the International Academy for Production Engineering
CIRP. The priority program dealt with the modeling and prediction of interactions
between machine structures and manufacturing processes in technical systems. The
objective was a sound reproduction of these interactions and a basic understanding
of the acting inter-relationships in order to be able to specifically influence and plan
manufacturing processes in the future. The understanding of process machine inter-
actions is a big issue in modern production technology. These interactions can be the
cause of erroneous processes, which directly lead to undesired quality problems. To
cope with rising quality demands and the ongoing need to increase production effi-
ciency in the future, 20 interdisciplinary research projects were funded for a 6 years
period by the DFG. In these projects, models and simulation tools were developed
for a variety of manufacturing technologies, such as cutting, grinding or forming.
Due to the intensive collaboration of researchers from different disciplines, such
as Production Engineering, Mechanical Engineering or Mathematics, it was possi-
ble to gain a deep understanding of process machine interactions. In addition, de-
tailed models and efficient simulation techniques to predict the interactions within a
reasonable calculation time were developed.
Process-machine-interactions have become a central research topic in production
engineering within the last years, not only in academic research but also in indus-
trial companies. Machine tool builders are expanding the use of simulation methods
to design machine tools, particularly considering process-machine-interactions. Ac-
cording to the resulting demand for access to research results and exchange, a series
of International Conferences on Process Machine Interactions has been success-
fully implemented with a steadily increasing number of participants. The Priority
Program 1180 contributed to this important research topic by providing elementary
experimental methods and mathematically verified computation models.
VI Preface
This book consists of the four parts “Basics”, “Grinding”, “Cutting” and “Form-
ing”. Part I “Basics” gives an overview of the applied and developed methods in
Measuring Technology, Modeling and Simulation and Mathematical Methods. The
following 3 parts “Grinding”, “Cutting” and “Forming” contain the main scientific
results and modeling approaches of all 20 research projects, covering a wide range
of topics, e. g. tool grinding, milling and deep drawing. Despite the large number
of different manufacturing methods investigated by the projects, some topics, such
as dynamic self-excitation of machine tools, known as “chatter vibration”, the static
deflection of machine tool parts due to the acting process forces or the influence of
thermal effects, were addressed by nearly all of the research projects.
Prof. Dr.-Ing. Berend Denkena Dr.-Ing. Ferdinand Hollmann

Leibniz Universität Hannover German Research Foundation
Coordinator of Priority Program 1180 Program Director Engineering Science
Contents
Part I: Basics
Measurement and Test Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
E. Abele, J.C. Aurich, B.-A. Behrens, D. Biermann, C. Brecher,
E. Brinksmeier, M. Czora, B. Denkena, U. Engel, K. Großmann, U. Heisel,
D. Heinisch, R. Hermes, B. Kirsch, F. Klocke, A. Krause, T. Kroiß,
R. Laurischkat, M. Löser, F. Mahr, H. Meier, M. Pischan, P. Rasper,
A.V. Scheidler, M. Storchak, E. Uhlmann, M. Weiß
Modeling and Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
C. Brecher, A. Bouabid, M. Deichmueller, B. Denkena, K. Großmann,
A. Hardtmann, D. Hömberg, R. Hermes, F. Klocke, M. Löser, O. Rott,
P. Steinmann, M. Weiß
Adaptive Finite Elements and Mathematical Optimization Methods . . . . 53
M. Andres, H. Blum, C. Brandt, C. Carstensen, P. Maaß, J. Niebsch,
A. Rademacher, R. Ramlau, A. Schröder, E.-P. Stephan, S. Wiedemann
Part II: Grinding

High-Performance Surface Grinding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
J.C. Aurich, A. Bouabid, P. Steinmann, B. Kirsch
Process Machine Interaction in Pendulum and Speed-Stroke
Grinding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
M. Weiß, F. Klocke, H. Wegner
Simulation of Process Machine Interaction in NC-Shape Grinding . . . . . 121
D. Biermann, H. Blum, A. Rademacher, A.V. Scheidler, K. Weinert
Modeling of Process Machine Interactions in Tool Grinding . . . . . . . . . . . 143
M. Deichmueller, B. Denkena, K.M. de Payrebrune, M. Kröger,
S. Wiedemann, A. Schröder, C. Carstensen
VIII Contents
Part III: Cutting

HPC - Stability Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
C. Brecher, R. Hermes, M. Esser
Development of a Stability Prediction Tool for the Identification
of Stable Milling Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
D. Hömberg, E. Uhlmann, O. Rott, P. Rasper
Synthesis of Stability Lobe Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
K. Großmann, M. Löser
Analysis of Industrial Robot Structure and Milling Process
Interaction for Path Manipulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
J. Bauer, M. Friedmann, T. Hemker, M. Pischan, C. Reinl, E. Abele,
O. von Stryk
Process Machine Interactions in Micro Milling . . . . . . . . . . . . . . . . . . . . . . 265
E. Uhlmann, F. Mahr, Y. Shi, U. von Wagner
Numerical Computation Methods for Modeling the Phenomenon
of Tool Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
B. Denkena, E.P. Stephan, M. Maischak, D. Heinisch, M. Andres
Dynamic and Thermal Interactions in Metal Cutting . . . . . . . . . . . . . . . . . 309
P. Eberhard, U. Heisel, M. Storchak, T. Gaugele
Surface Generation Process with Consideration of the Balancing State
in Diamond Machining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
C. Brandt, A. Krause, J. Niebsch, J. Vehmeyer, E. Brinksmeier, P. Maaß,
R. Ramlau
Modeling and Simulation-Based Optimization of a Turning Process . . . . 361
R. Britz, T. Maier, F. Schwarz, H. Ulbrich, M.F. Zaeh
Part IV: Forming

Advanced Forming Process Model - AFPM . . . . . . . . . . . . . . . . . . . . . . . . . 383
K. Großmann, A. Hardtmann, H. Wiemer, L. Penter, S. Kriechenbauer
Consideration of the Machine Influence on Multistage Sheet Metal
Forming Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
B.-A. Behrens, A. Bouguecha, R. Krimm, T. Matthias, M. Czora
Optimization of Tool and Process Design for the Cold Forging
of Net-Shape Parts by Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
T. Kroiß, U. Engel
Contents IX
Interaction Effects between Strip and Work Roll during Flat Rolling
Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
S. Puchhala, M. Franzke, G. Hirt
Increase of the Dimensional Accuracy of Sheet Metal Parts Utilizing a
Model-Based Path Planning for Robot-Based Incremental Forming . . . . 459
H. Meier, S. Reese, Y. Kiliclar, R. Laurischkat
Gear Rolling Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475
R. Neugebauer, U. Hellfritzsch, M. Lahl, M. Milbrandt, S. Schiller,
T. Druwe
Investigation of the Complex Interactions during Impulse Forming
of Tubular Parts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491
Fr.-W. Bach, M. Kleiner, A.E. Tekkaya
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513
Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515

List of Authors
Abele, E.
Technische Universität Darmstadt, Institute of Production Management,
Technology and Machine Tools
Andres, M.
Leibniz Universität Hannover, Institute for Applied Mathematics
Aurich, J. C.
University of Kaiserslautern, Institute for Manufacturing Technology
and Production Systems
Bach, F.-W.
Leibniz Universität Hannover, Institute of Material Science
Bauer, J.
Behrens, B.-A.
Leibniz Universität Hannover, Institute of Metal Forming and Metal Forming
Machine Tools
Biermann, D.
Technische Universität Dortmund, Faculty of Mechanical Engineering,
Institute of Machining Technology
Blum, H.
Technische Universität Dortmund, Faculty of Mathematics (LS X)
Bouabid, A.
Universität Erlangen-Nürnberg, Chair of Applied Mechanics
Bouguecha, A.
Machine Tools
Brandt, C.
University of Bremen, Center for Industrial Mathematics
XII List of Authors
Brecher, C.
RWTH Aachen University, Laboratory of Machine Tools and Production
Engineering, Chair of Machine Tools
Brinksmeier, E.
University of Bremen, Foundation Institute of Materials Science, Laboratory of
Precision Machining
Britz, R.
Technische Universität München, Institute of Applied Mechanics,
Department of Mechanical Engineering
Carstensen, C.
Humboldt University of Berlin, Department of Mathematics
Czora, M.
Machine Tools
De Payrebrune, K.M.
Technische Universität Bergakademie Freiberg, Institute for Machine Elements,
Design and Manufacturing
Deichmueller, M.
Leibniz Universität Hannover, Institute of Production Engineering and Machine
Tools
Denkena, B.
Tools
Druwe, T.
Fraunhofer Institute for Machine Tools and Forming Technology
Eberhard, P.
Universität Stuttgart, Institute of Engineering and Computational Mechanics
Engel, U.
Universität Erlangen-Nürnberg, Chair of Manufacturing Technology
Esser, M.
Engineering
Franzke, M.
RWTH Aachen University, Institute of Metal Forming
Friedmann, M.
Technische Universität Darmstadt, Department of Computer Science, Simulation,
Systems Optimization and Robotics Group
Gaugele, T.
Universität Stuttgart, Institute of Engineering and Computational Mechanics
List of Authors XIII
Großmann, A.
Technische Universität Dresden, Institute of Machine Tools and Control
Engineering
Hardtmann, A.
Engineering
Heinisch, D.
Tools
Heisel, U.
Universität Stuttgart, Institute for Machine Tools
Hellfritzsch, U.
Hemker, T.
Hermes, R.
Engineering, Chair of Machine Tools
Hirt, G.
Hömberg, D.
Weierstrass Institute, Berlin
Kiliclar, Y.
RWTH Aachen University, Institute of Applied Mechanics
Kirsch, B.
University of Kaiserslautern, Institute for Manufacturing Technology and
Production Systems
Kleiner, M.
Technische Universität Dortmund, Institute of Forming Technology and Lightweight
Construction
Klocke, F.
Engineering, Chair of Manufacturing Technology
XIV List of Authors
Krause, A.
University of Bremen, Laboratory for Precision Machining
Kriechenbauer, S.
Engineering
Krimm, R.
Machine Tools
Kröger, M.
Technische Universität Bergakademie Freiberg, Institute for Machine Elements,
Design and Manufacturing
Kroiß, T.
Universität Erlangen-Nürnberg, Chair of Manufacturing Technology
Lahl, M.
Laurischkat, R.
Ruhr-University Bochum, Chair of Production Systems
Löser, M.
Engineering
Maaß, P.
Mahr, F.
Technische Universität Berlin, Institute of Machine Tools and Factory Management
Maier, T.
Technische Universität München, Institute for Machine Tools and Industrial
Management, Department of Mechanical Engineering
Maischak, M.
Matthias, T.
Machine Tools
Meier, H.
Ruhr-University Bochum, Chair of Production Systems
Milbrandt, M.
Neugebauer, R.
List of Authors XV
Niebsch, J.
Johann Radon Institute for Computational and Applied Mathematics,
Austrian Academy of Sciences
Penter, L.
Engineering
Pischan, M.
Puchhala, S.
Rademacher, A.
Technische Universität Dortmund, Faculty of Mathematics (LS X)
Ramlau, R.
Johannes Kepler University Linz, Industrial Mathematics Institute
Rasper, P.
Reese, S.
RWTH Aachen University, Institute of Applied Mechanics
Reinl, C.
Rott, O.
Weierstrass Institute, Berlin
Scheidler, A. V.
Schiller, S.
Schröder, A.
Schwarz, F.
Shi, Y.
Technische Universität Berlin, Departments of Mechanics, Chair of Mechatronics
and Machine Dynamics
XVI List of Authors
Steinmann, P.
Universität Erlangen-Nürnberg, Chair of Applied Mechanics
Stephan, E. P.
Storchak, M.
Universität Stuttgart, Institute for Machine Tools
Tekkaya, A. E.
Technische Universität Dortmund, Institute of Forming Technology and Lightweight
Construction
Uhlmann, E.
Ulbrich, H.
Technische Universität München, Institute of Applied Mechanics,
Department of Mechanical Engineering
Vehmeyer, J.
von Stryk, O.
von Wagner, U.
Technische Universität Berlin, Departments of Mechanics, Chair of Mechatronics
and Machine Dynamics
Wegner, H.
RWTH Aachen University, Laboratory for Machine Tools and Production
Engineering
Weinert, K.
Weiß, M.
RWTH Aachen University, Laboratory for Machine Tools and Production
Engineering
Wiedemann, S.
Wiemer, H.
Engineering
Zaeh, M.
Part I
Basics
Chapter 1
Measurement and Test Techniques
E. Abele, J. C. Aurich, B.-A. Behrens, D. Biermann, C. Brecher,

E. Brinksmeier, M. Czora, B. Denkena, U. Engel, K. Großmann,
U. Heisel, D. Heinisch, R. Hermes, B. Kirsch, F. Klocke, A. Krause,
T. Kroiß, R. Laurischkat, M. Löser, F. Mahr, H. Meier, M. Pischan,
P. Rasper, A. V. Scheidler, M. Storchak, E. Uhlmann, and M. Weiß
Abstract. Nowadays, different measurement and test techniques are used to inves-
tigate the interaction between processes and machine tool structures. Machine and
workpiece properties are determined after analyzing the individual factors of
process metrology, which have an effect on the process. This chapter explains the
measurement methods for the structural analysis of the machine tool as well as for
manufacturing processes and for the workpiece analysis. In addition, an overview
of different measurement and test techniques based on selected examples related
to the priority program 1180 is given.
1.1 Introduction
This chapter analyzes the metrological possibilities in order to determine the inte-
raction between process and machine structure. The metrological analysis is sub-
divided into three main sections: Section 1.2 refers to the structural analysis of the
machine tool, section 1.3 to the process analysis and section 1.4 to the analysis of
the workpiece. In section 1.2 different measurement and test techniques of the
static and the dynamic machine tool behavior, the kinematics and the temperature
of machine tools are described. Section 1.3 describes different measurements and
test techniques for force, acoustic emission and temperature, whereas section 1.4
describes refuse to surface and geometry assessment. Here, a small imported out-
line can be obtained and its influence must be analyzed to understand the interac-
tion between process and machine tool structure for specific examples.
1.2 Structural Analysis of Machine Tools

Working precision, performance, environmental behavior and reliability of ma-
chine tools affect the quality of manufactured products and the efficiency of the
processes significantly. Technological progress in the machine tool industry and
the competition pressure to increase productivity ask for higher performance,
higher spindle speeds, higher feed rates and longer material removal rates. Hence,
new and optimized machine tool structures have to be developed. In addition, due
B. Denkena and F. Hollmann (Eds.): Process Machine Interactions, LNPE, pp. 3–27.
springerlink.com © Springer-Verlag Berlin Heidelberg 2013
4 E. Abele et al.
to significantly increased process loads the demands concerning the working accu-
racy of the machine tool have increased as well. Thus, apart from performance
values like spindle power and speed or feed rate the structural and mechanical
properties of the machine tool have to be known in order to assess the accuracy
and productivity [1].
To analyze and improve the machine tool behavior it is essential to describe the
machine tool characteristics by defined parameters. Despite the good progress in
calculating machine tool parameters, the experimental determination of the struc-
tural properties is still essential for the parameterization and evaluation of machine
tool models. The accuracy of machine tools is primarily affected by deviations at
the tool-workpiece interface. Depending on the transmission behavior of the
machine tool thermal, static or dynamic loads result in kinematic and geometric
deviations from the desired working motions. Therefore, the thermal and dynamic
behavior limits the theoretical performance of the machine tool. Due to dynamic
instabilities and displacements caused by thermal dilatation the full potential of
machines cannot be exploited, which results in a negative impact on the productiv-
ity. To ensure the efficiency of machine tools the interactions between process and
structure have to be determined. In the following sections, different measurement
methods for the analysis of the static, dynamic and thermal structure, their beha-
vior as well as the kinematics are described.
1.2.1 Static Machine Tool Behavior

Static process forces between the workpiece and the tool lead to a static stress on
all components and joints of the machine structure included in the flow of force.
Thus, the stiffness properties of the machine are the sum of the individual stiff-
nesses of the concerned elements. The influence of the machine properties on the
workpiece is usually of particular interest; hence, investigations of the machine
tools stiffness focus on the interface between tool and workpiece. Therefore, the
relative displacement between the tool and the workpiece as a result of static
process loads has to be investigated. While the process load is usually applied in
the three Cartesian coordinate directions, the measured displacements are divided
into tilts and deflections. These are each described by a translational and a tilting
stiffness matrix, in which the main stiffnesses are located at the principal diagonal
of the matrix with the cross stiffnesses next to them. Figure 1.1 shows an example
of the analysis of the main stiffness in the z-direction and the tilting about the x-
axis by applying a load in z-direction. Machine structures usually show a progres-
sive development of stiffness. After overcoming the clearances in the bearings, the
guidance and the screw connections as well as the internal friction in the gears and
seals, the stiffness rises with increasing load. Changing contact conditions and in-
ternal friction in the joints and contact points cause a hysteresis between loading
and load relieving. In this section, the determination of the static stiffness is ex-
plained by a press and an industrial robot.
1 Measurement and Test Techniques 5
Fig. 1.1 Characterization of static machine compliance (based on [1])
A method to determine the deflection of presses under a static load is described

in the standard DIN 55189 “Determination of the ratings of presses for sheet metal
working under static load“ for mechanical presses (part 1), as well as for hydraulic
presses (part 2) [2]. By means of this method the press is loaded by a hydraulic
system, which is applied torque-free by means of a compensation device in z-
direction (Fig. 1.2).
δz
Ram Δz
3 ΔFz
4 5 6
2
1
Fz
Non-
Linear
Bolster plate linear
z
y
Force
x
Pressure pad transducer
Distance 1 -
Compensation device 6 Displacement transducers
plates
(Cardan calotte)
Fig. 1.2 Measurement setup according to the DIN 55189 and resulting displacement
development
6 E. Abele et al.
The standard describes the determination of the deflection in forming direction

caused by a centric load and the tilting of the ram as well as its horizontal dis-
placement by an eccentric load. The investigated force-displacement characteristic
can be classified into two periods, an initial non-linear displacement and tilting pe-
riod, which occurs due to bearing clearances, and a period during which the press
deflects linear elastically. The relevant static press characteristics such as the hori-
zontal displacement, the tiltings about the x- and y-axis and the stiffnesses, are
also defined in DIN 55189 and can be determined by the recorded force-
displacement and force-tilting characteristics. The static press characteristics make
it possible to compare different types of presses with each other.
In the following paragraph, the determination of the static stiffness of an articu-
lated robot is shown. In Figure 1.3, the measurement setup as well as a typical
stiffness in the working space is presented.
Fig. 1.3 Experimental setup (left) and typical results (right) of an articulated robot
The setup consists of a force measurement rod to apply and detect tensile and
compressive forces and laser distance sensors to measure the displacement of the
robot due to the force. The difference in the tensile and compressive cycles
indicates hysteresis. The backlash amounts to approximately 0.2 mm at the mea-
surement point x = 1,900 mm, y = 0 mm and z = 600 mm in the base coordinate
system. Furthermore the measured curves show a slight s-shape due to a non-
linear structural behavior. Using the least-squares method a linear slope can be
fitted into the measured curve. The gradient of the measured curve then indicates
the stiffness.
1.2.2 Dynamic Machine Tool Behavior

The accuracy of a machine tool is determined by the deflection occurring at the
contact point between tool and workpiece at the specified target position. Apart
from the influences of the static loads, which are described in the previous chap-
ter, the dynamic behavior under varying loads is also a criterion for the perfor-
mance of a machine tool system. Unbalanced dynamic properties of this system
lead to oscillation phenomena, which can result in a poor surface quality of the
workpiece, increased machine and tool wear, tool breakage and damage of the
machine tool. The latter mentioned damages have to be considered particularly
with regard to the occurrence of regenerative chatter oscillations, which increase
as a result of the interactions between the dynamic machine tool behavior and
process behavior during the machining process [1], [4].
Therefore, the aim of investigating the dynamic machine behavior is to describe
possible weak points of the mechanical structure quantitatively using the tools of
frequency response measurement and experimental modal analysis [40]. A major
application of modal analysis in mechanical engineering is trouble-shooting. As an
example, chatter vibrations in machine tools are often caused by structural insta-
bilities, which can be identified using experimental modal analysis techniques.
Recently, the investigation of the dynamic machine tool behavior has become
more important for the configuration and alignment of simulation models. Nowa-
days, an important application is the correlation of finite element models with ex-
perimental data from modal analysis in order to improve the accuracy of structural
dynamic simulations. This is very useful for sensitivity analyses and the prediction
of the dynamic behavior due to structural modifications.
The investigation of the dynamic machine tool behavior is always based on the
measurement of the frequency-dependent rigidity of the structure [3], [4]. Among
signal processing and analog digital conversion (ADC), the required measurement
chain can be divided into three major systems. The first is the excitation of the
structure. This can be done in several ways. The most commonly used are an at-
tached shaker or a hammer blow. Electromagnetic or electrohydraulic shakers are
controlled by a signal generator providing the ability to induce various loads into
the structure. These loads can be, for example, sinusoidal, periodic, random or
transient. Especially sinusoidal loads, such as sine sweep or stepped sine, are pre-
ferably used to investigate non-linear system behavior by analyzing the structure
with varying load amplitudes. However, the use of a shaker requires a connection
to the structure, which remains attached throughout the test. This makes shaker
testing less flexible for in-the-field testing. In contrast, hammer testing provides
the advantage of inducing the excitation force in a contactless way. As the ham-
mer impact excites the structure over a wide frequency band, hammer testing is a
very fast and convenient way to determine the compliance behavior of a machine
tool. Since manipulation of the frequency content and amplitude of the compact is
limited, it is less appropriate for analyzing non-linear system behavior. [6]
The second subsystem of the measurement chain provides the detection of the
loads, which are induced into the machine tool structure. Therefore, piezoelectric
crystals or strain gauges are used, which are integrated into the force flux between
8 E. Abele et al.
the machine tool structure and the exciter. The third subsystem consists of the
measurement technique for detecting the vibration response of the machine tool
structure. In addition to systems for direct measurement of the displacement, e.g.
inductive transducers or optical measurement techniques piezoelectric accelero-
meters are used as well.
1.2.2.1 Frequency Response Function
The frequency response functions (FRF) represent the dynamic compliance beha-
vior of a machine tool structure in frequency domain. Also, process stability and
the occurrence of forced vibrations can be assessed on the basis of these mea-
surement data.
For the determination of the frequency response function Fast-Fourier-
Transformation-analyzers (FFT) are used. Therefore, the analog force and deflec-
tion/acceleration signals are sampled and digitalized. The sample rate determines
the frequency range and the number of samples defines the frequency resolution of
the analysis. To suppress high frequency disturbances analog force and deflection
signals must be filtered before the digitalization.
The sampled time signals can be weighted by a so-called window function to
avoid errors, which may occur when the signals are transformed into frequency
domain. The type of window function depends on the signal that has to be ana-
lyzed. Commonly used functions are transient window (impact force), exponential
window (response to an impact) and Hanning window. The transformation of the
weighted signals into frequency domain is carried out by a discrete Fourier Trans-
formation. The result of the transformation is a complex frequency spectrum,
which can be depicted either as real part and imaginary part, magnitude and phase
or as a Nyquist plot (Fig. 1.4). Separately from the depiction FRF describes the
frequency-dependent deflection answer of a mechanical system regarding the dy-
namic load acting on it.
Fig. 1.4 Principle of the measurement of frequency response functions

1.2.2.2 Experimental Modal Analysis
Experimental modal analysis is a method for determining the dynamic characteris-

tics of a structural system: its natural frequencies, mode shapes and damping fac-
tors. With these characteristics a mathematical model of the dynamic behavior of
the system, a so called “modal model”, can be formulated [5], [6]. The vibration of
a linear time-invariant system can be described by a linear combination of its
mode shapes, which are inherent to the dynamic system and determined by its
physical properties (mass, stiffness and damping) and their spatial distributions.
Coming from an analytical model, the system may be given in forms of partial dif-
ferential equations and their solution provides the natural frequencies and mode
shapes [7]. A more realistic physical model usually comprises mass, damping and
stiffness matrices, which characterize the system properties. By solving the eigen-
value problem the modal data can be obtained. Utilizing finite element analysis
almost every structure can be discretized into differential equations of motion and
hence permits theoretical modal analysis.
Experimental modal analysis is a technique used to determine the modal model of
a linear time-invariant system. By measuring the vibration response at one or more
locations and the excitation force at the same or a different location and calculating
their ratio several FRFs can be obtained. From these FRF-measurements a modal
model of the mechanical system can be derived. In order to obtain an accurate
Fig. 1.5 Geometry model with measurement locations of a horizontal milling machine
10 E. Abele et al.
modal model of the examined system the proper selection of excitation and re-
sponse locations is of particular importance. The mathematical models obtained
from modal analysis can also be used for the prediction of structure responses due
to exciting forces or for substructure coupling when a simple dynamic representa-
tion is more suitable than a complex finite element model.
Figure 1.5 shows a geometric model of a horizontal milling machine. Each
geometry point represents a measurement location in the machine tool. The struc-
ture was excited with an impulse hammer at several locations and the vibration
response was measured using tri-axial acceleration sensors. The obtained modal
model can be used, for example, to predict chatter vibrations or to identify struc-
tural instabilities.
1.2.3 Measurement of Kinematics

The accuracy of machine tools depends on a large variety of different influences.
Geometric deviations in dimension of machined and formed workpieces can, ac-
cording to [1], result in:
• Deviation of tool dimensions out of tolerances due to insufficient tool manufac-
turing
• Process-induced deviations such as tool wear or built-up edges
• Elastic deformation of the tool, the workpiece, clamping and support structures
• Deviations of the tool path regarding relative movement between tools and
workpieces including force-induced deviations of the machine tool structure
Depending on the process the listed influences need to be considered when model-
ing the process machine interaction. A lot of different methods are available to
measure the accuracy of the translational and rotational axes of machine tools. The
spectrum reaches from simple measuring setups with test gauges, measuring
straightness, parallelism, perpendicularity and concentricity up to complex and
highly accurate methods. Some of these methods are described below.
1.2.3.1 Circularity Test
The circularity test allows the determination of the accuracy of a circular path, in-
terpolated by a computer control unit. The deviations and vibrations can be traced
back to the control unit, the drives and the machine kinematics. The test can be
conducted using a double-ball-bar or a grid encoder. In the case of a grid encoder
a photoelectric sensor moves over a plate, which contains a very precise measur-
ing grid without contact. Using a double-ball-bar the circularity of the machine
tool is determined by a position-measuring system. The system is integrated in the
gauge. Performing the test with large radii gives information about the machine
geometry, whereas small radii are used for the evaluation of the feed drive dynam-
ics [1]. Afterwards, the measured path can be compared with the desired path.
Figure 1.6 presents the measuring setup with a grid encoder (left) and a double-
ball-bar (right).
Fig. 1.6 Measuring setup for the ball-bar test with a grid encoder (left) and a double-ball-
bar (right) [1]
1.2.3.2 Back-Step Test

During the back-step test several positions are approached from both sides. This is
repeated for each axis. The current position is detected by an external measuring
system at the tool-center-point (e. g. a grid encoder or a laser interferometer) and
is compared with the desired position. According to the VDI/DGQ 3441 standard
the parameters positional tolerance, positional deviation, reversal error, position-
ing scatter band and position uncertainty can be determined from the measurement
data (see Fig. 1.7).
Fig. 1.7 Procedure of the back step method (left); characteristic diagram and determined
parameters (right)
12 E. Abele et al.
1.2.3.3 Measurement of the Machine Axes using Laser Interferometry
An approach for the determination of the kinematics of a machine tool is the mea-
surement of the procedure movement of the machine axes. The measurement of
the procedure axis is briefly described on the basis of a face grinding machine
(Geibel & Hotz FS 635-Z CNC). The temporal response of the machine control
was examined for the input of a correcting variable. The velocity and the accelera-
tion of the machine table were measured. A laser interferometer system with a
scanning rate of fa = 20 Hz was used in order to avoid the influence of machine
vibrations on the velocity and acceleration measurements. The change of move-
ment between the interferometer and the retro reflector was measured with evalua-
tion software. [8]
For the investigations the strokes were measured by several sequential starting
points with constant, well-defined point distances at different workpiece veloci-
ties. The dependency of the workpiece velocity on the selected step size is shown
in Figure 1.8.
Fig. 1.8 Dependency of the workpiece velocity on the selected step size [8]
1.2.4 Thermal Machine Tool Behavior

Heat affects the static and dynamic properties of machine tools. The heat-related
deformation on the machine components varies according to the material proper-
ties, the machine´s geometry and the conditions of the heat transfer. Consequently,
the stiffness of the machine components is affected by the temperature. This has
an impact on the production process and leads to dimensional deviations of the
workpiece. The heat sources can be classified into internal and external sources
according to where the heat is generated.
The internal heat sources include thermal dissipation losses, which have their
origin in the limited electrical and mechanical efficiency of the machine compo-
nents. The external heat sources result from heat transfer mechanisms such as
conduction, convection or radiation caused by ambient heat flow. In addition, the

process-induced energy losses due to the friction between the tool and the work-
piece as well as the process heat have an impact on the temperature field of the
machine [9]. Temperature measurements on machine tools and machining centers
can be carried out according to the standards ISO 230-3 and ISO 10791-10
[10, 11]. The temperature distribution of the machine can be measured either at a
finite number of individual points using thermocouples (contact measurement) or
extensively via optical measurement systems (non-contact measurement via ther-
mography camera / pyrometer) [12].
Especially the infrared thermography is applicable for this kind of measure-
ment, for instance at press frames, because its surface has a homogenous radiance
constant. Therefore, the emission factor of the radiating object, which can be de-
termined by means of a reference measurement with an additional measuring sys-
tem, has to be known. For the measurements of a finite number at individual
points thermocouples or resistance thermometers can be used. These two types of
temperature sensors differ in their measurement accuracy, cost, size, capability of
measuring the surface temperature and vibration resistance (Tab. 1.1).
Table 1.1 Comparison of contact and non-contact measurement of temperature [12]
Contact measurement Non-contact measurement

lower costs no influence on measuring subject
more precise local and extensive temperature mea-
surement possible
easy to handle
Thermocouples are available for different applications and then classified into
different types. For example, thermocouples of the type T have an accuracy of
± 0.5 °C for a measurement range between approx. - 200 °C and 300 °C. Further-
more, this type is capable of measuring temperatures in fluids such as in the lubri-
cating oil system [13]. In Figure 1.9, some types of temperature sensors are
shown.
For measuring the temperature of the main eccentric shaft of a press electrically
insulated thermocouples with screw threads are often applied as close as possible
to the shaft. This can be done by fixing the sensor directly to the bearing of the ec-
centric shaft within the press frame or the connecting rod. For measurements
which do not allow a screwing fixation of the thermocouples the sensors can be
fixed with a thermal conductance paste and adhesive tape [9]. The signals of the
thermocouples can be recorded with a PC including a measuring board. The mea-
suring board should be equipped with an internal cold-junction compensation,
which is required for thermocouples. Within this compensation the reference tem-
perature is simulated by means of an integrated transistor. The difference in tem-
perature between the junction and the measurement point induces an electrical
voltage. For the measurement of the oil temperature in larger containers electrical-
ly-shielded resistance thermometers can be applied [13].
14 E. Abele et al.
Fig. 1.9 Different types of temperature sensors for measuring the temperature of machine
tool components
To consider thermal convection effects the temperature of the environment is

defined as the reference. The characteristic temperature profile of a machine con-
verges exponentially with time. The temperature at the components of the machine
rises with high gradients during the starting phase of the machine in usage and
converges gradually with further operating time towards a certain value. The tem-
perature gradients and the end temperature are much higher for machine tools used
for machining than for those used in metal forming. For the determination of a
steady thermal condition of the machine the temperature increase of all relevant
machine components has to be taken into account. Once a steady thermal condi-
tion is reached the production process of the workpiece is no longer influenced
significantly by temperature effects. Figure 1.10 shows the different thermal con-
ditions of a high speed stamping machine [9].
Fig. 1.10 Temperature profile of a high speed stamping machine [9]
1.3 Process Analysis

As depicted in section 1.2., precise knowledge about the machine tool structure is
essential in order to achieve the best productivity and accuracy. Since the machine
tool behavior interacts with the machining process, characteristics such as thermal,
statical or dynamical loads have to be taken into consideration. Therefore, an ex-
perimental determination of process factors has to be carried out. The measured
data such as process forces, temperatures, sound and vibration can be used for pa-
rameter identification and the evaluation of process models. In conjunction with
the identified parameters of the machine tool behavior the boundary conditions for
comprehensive process machine interaction (PMI) models can be defined.
In the following section, different measurement methods for the analysis of
process forces (see Sect. 1.3.1) are described. Measurement methods and applica-
tions of acoustic emission (see Sect. 1.3.2) and temperatures (see Sect. 1.3.3) are
given consecutively.
16 E. Abele et al.
1.3.1 Process Force Measurements

Process forces are commonly used values for the characterization of manufactur-
ing processes. Since there is a large variety of manufacturing processes where
force measurements are of interest, the boundary conditions also differ. According
to the different processes appropriate measuring devices have to be deployed con-
sidering the force magnitude and the process dynamics. For this purpose, the mea-
surement procedure as well as the post-processing of the measured data may vary.
In this section, two commonly used measurement methods are described, which
can be applied to cutting and forming processes. Subsequently, examples of force
measurements covering a force bandwidth from a few tenths of newtons to several
kilonewtons are given.
1.3.1.1 Force Transducer based on Strain Gauges
This kind of load cell consists of a steel body – the sensing device – which acts as
a spring. On this body, strain gauges are applied as measuring devices. Thus, the
forces are converted into elastic deformation. A calibration permits the correlation
between the elastic strains and the applied force [14].
The basic effect of a strain gauge is the change of resistance in an electrical
conductor due to the effect of mechanical stress, discovered by Wheatstone and
Thomson [15]. This change of resistance in a single wire is very small. For that
purpose, metal strain gauges with “wound wires”, which form a kind of grid, have
been developed. For an efficient production the grids are manufactured by etched
foil technology nowadays. Apart from metal strain gauges there are other types of
electrical resistive strain gauges, e. g. semi-conductor and vapor-deposited strain
gauges [16].
Strain gauge-based force transducers can be used for static and dynamic mea-
surements and are available in a variety of scales with nominal forces from about
10 N up to 5 MN. This range can be necessary, for example, for the measurement
of forming forces in cold forging. However, with larger nominal loads the height
of the transducers increases noticeably up to about 180 mm at 5 MN. Via a mea-
suring amplifier and an analog digital converter the output signal can be recorded
and processed electronically.
1.3.1.2 Force Measurement Based on Piezoelectric Elements
The piezoelectric effect is based on an interaction between electrical field strength,

electrical displacement and the mechanical factors displacement and stress. If the
piezoelectric element, e.g. quartz, is deformed mechanically, atoms in the crystal
lattice are displaced. This leads to an outward charge displacement. Piezoelectric
force transducers usually use the longitudinal effect in one direction. Long lasting
quasi-static measurements require special attention to avoid drift of the output sig-
nal. A detailed description of piezoelectricity and measuring with piezoelectric
sensors is given in [17].
Piezoelectric measurement devices, e.g. load washers and dynamometers, do
not need a sensing device showing a considerable elastic deflection, which the
measuring devices are applied on. In this case, the piezoelectric element itself is
exposed to the force and also emits the electric output signal. Piezoelectric measur-
ing devices possess a high stiffness and can be manufactured in a very compact de-
sign. Since they cover a large range of force magnitudes, which can be measured at
high sampling rates they are predestinated for the use in processes with interrupted
cutting conditions. The high resolution is also advantageous for measuring very
small cutting forces as they occur, e.g. in ultra precision cutting. Compared to force
transducers based on strain gauges, quartz load elements at a comparable measur-
ing range for forming processes have a more compact design and a higher stiffness.
However, piezoelectric force transducers are more expensive.
1.3.1.3 Force Measurement in Ultra-Precision Machining
In ultra-precision machining, all process parameters are - compared to conven-

tional machining - reduced or scaled down to some extent, e.g. cutting depth and
feed rate range within a few microns only. Thus, the process forces Fi are very
small as well (Fi ≤ 1.5 N). For this reason, dynamometers with the capability of
detecting very small forces in the specified range are required, e.g. the triaxial dy-
namometer described in Chapter 15 (piezoelectric, Kistler Type 9256A1, which
has a threshold of less than 0.002 N).
For most of the ultra-precision machining applications low rotational speeds
(150 rpm (≈ 2.5 Hz) < n < 5000 rpm (≈ 83.33 Hz)) are used. Therefore, process-
induced dynamic excitation with frequencies ν > 100 Hz can be neglected for ul-
tra-precision turning, given a continuous cut. Due to unbalances the machine tool
is excited by the rotation of the workpiece. For processes with discontinuous cut
(e. g. eccentric turning, circumferential milling (fly cutting)) an impact excitation
affects the tool and the machine tool whenever the tool engages the workpiece.
One major issue in measuring the forces for ultra-precision machining is the
post-processing (Fig. 1.11) of the data obtained. As the forces are very small, ex-
ternal sources (e. g. current flow of the cross table) can add noise to the measure-
ment data. To remove this noise low pass filters are used with cut-off frequencies
of the filters depending on whether the dynamic influences of the unbalances are
of interest or not.
2 1,5
Thrust force
N N
Thrust force
Force Fi
0 Cutting force 0,5

Cutting force
Feed force
-1 0
Feed force
-2 -0,5
0 50 100 150 s 250 0 50 100 150 s 250
Time t Time t
Fig. 1.11 Measured forces before (left) and after post-processing (offset and drift removed,
low-pass filter applied)
18 E. Abele et al.
1.3.1.4 Force Measurement in Micro and High Speed Cutting (HSC) Milling
For the force measurement of processes with interrupted cutting conditions at high
spindle speeds some particularities have to be taken into consideration. Due to the
periodical tooth engagement the machine tool structure is excited by high dynami-
cal loads. Depending on spindle speed and number of cutting teeth the tooth pass-
ing frequency may achieve values of up to several kHz. Since the cutting force
does not match an ideal sine wave the force signal also contains the harmonics of
the tooth passing frequency. Thus, if the actual cutting force has to be measured, a
very high sample frequency must be chosen. Another problem occurs when the
tooth passing frequency exceeds the measurement device’s eigenfrequency.
Therefore, it is also very challenging to obtain time signals of the actual process
forces acting on the cutter (e. g. to compare them with simulated force signals).
However, in a lot of cases it is much easier and less error-prone to evaluate low-
pass filtered or averaged signals (e. g. to identify cutting force coefficients).
Dynamometers based on piezoelectric elements are the most suitable solution
in terms of the highest possible eigenfrequencies and of achievable measuring res-
olution. The measuring device is usually located between the machine table and
the workpiece. Solutions for force sensors integrated into the tool holder are also
available.
1.3.1.5 Force Measurement in Cold Extrusion
Measuring the forces in cold extrusion processes requires the consideration of the
following specific process characteristics: The high flow stress in cold forging due
to forming at room temperature leads to high forces even if the die dimensions are
small. In the full forward extrusion process investigated in the project “Optimiza-
tion of Tool and Process Design for the Cold Forging of Net-Shape Parts by Simu-
lation” (chapter 19), forces of about 200 kN occur with a die diameter of only
12 mm above and 6 mm below the die shoulder. Thus, force transducers with a
high load capacity are necessary. Because of the high forces the presses and tool-
ing systems in use have to be stiff and compact. As a consequence, only limited
space is available for the force measurement. In contrast, the quasi-static characte-
ristic of extrusion processes with a smooth force increase does not imply high de-
mands on the dynamic behavior of the load cells.
Figure 1.12 shows the measurement setup in a tooling system for lateral cold
extrusion. The use of quartz load washers is required due to the limited height in
the press. The closed-die tooling system will be used on a single-acting stroke-
controlled press. The press has to apply the closing force, via disc springs, and the
forming force at the same time. To be able to measure both forces two quartz load
washers 9091A from Kistler Instrumente GmbH (nominal force 1200 kN) with a
height of only 28 mm are used. The washer (1) on the top is for measuring the en-
tire press force, the one (2) between the top tool plate and the disc springs for the
closing force. Subtracting these forces from each other leads to the punch force.
With the presented tooling system interactions between process, tool and machine
in closed-die forging will be investigated in the above-mentioned project (see
chapter 19).
Fig. 1.12 Tooling system for closed-die lateral cold extrusion with two quartz load washers
Kistler 9091A (1 and 2) for measuring press force and closing force
1.3.2 Acoustic Emission (AE) Measurements

Sensors for acoustic emissions to monitor process noise emitted during machining
and forming processes are vibratory systems. Their resonance points are deter-
mined by their construction. As a result, these sensors, which work proportionally
to acceleration, represent filter systems that have a dampening or amplifying effect
according to the frequency range. The AE sensor thus determines the evaluable
frequency spectrum by its frequency behavior to a great extent [18].
Acoustic emission Sensors are used in machining for different operations:
• first contact control,
• collision monitoring,
• balancing of grinding wheels,
• dressing monitoring in grinding,
• chatter detection,
• monitoring of grinding wheel wear and
• process monitoring of forming processes.
In the project “Process Machine Interaction in Speed-Stroke Grinding” the AE-
signal was used as a trigger signal (see Sect. 5.2). As the contact times between
the grinding wheel and the workpiece are in the range of milliseconds, the run-in
and run-out phase can take less than 2 ms. Hence, the sensitive AE-signal was
used to identify the exact moment of the first contact, which alleviates the analysis
of the force measurement.
20 E. Abele et al.
1.3.3 Process Temperature Measurements

Various methods and techniques are used to conduct temperature measurements.
One possible method is the use of thermo-chromic colors with special coating ma-
terials, which indicate changes in temperature by changing the color or tone [19].
Frequently used current methods are thermo-electric and radiation measurement
methods [20]. The so-called tool-workpiece-thermal elements and thermocouples
are used in thermo-electric measurements [21]. Pyrometry involves a non-contact
measurement of the absolute temperature; i. e. the self-radiation of the body is
measured without contacting the object itself [22]. In contrast, thermography in-
volves a measurement of the temperature distribution, i. e. relative differences in
temperature are measured but not, as in pyrometry, the absolute values. The radia-
tion measurement methods in general have a significantly higher time resolution
in comparison to thermo-electric methods. However, pyrometry is faster than
thermography due to its simple setup. Moreover, these are non-contact measure-
ment methods, which guarantee a considerably higher flexibility of the measure-
ments. Non-contact measurement methods exhibit a specific measuring error due
to a surface layer that forms in free air on the surfaces to be measured. In recent
years, the measurement of cutting temperatures by means of pyrometry and ther-
mography has gained considerable importance. A short overview of the used me-
thods of temperature measurement for cutting and grinding is given.
1.3.3.1 Investigation of the Temperatures in Cutting
Thermocouples have a relatively low time resolution and it is difficult to place

them directly in the cutting zone. Single-wire thermal elements represent an ex-
ception since they can easily be placed in the secondary cutting zone [23]. The
main problem for both of these methods as well as for the tool-workpiece-thermal
element measurement method is the calibration of the measuring chain. Hence,
these measurement methods are not commonly used. Only an average cutting
temperature is measured in the contact zone when the tool-workpiece-thermal
element method is used [21].
1.3.3.2 Investigation of the Temperatures in Grinding
In grinding, the major part of energy input is dissipated as heat. The heat flows in-
to the chips, the coolant, the environment, the workpiece and the grinding wheel.
The degree of thermal damage of the workpiece surface depends on the machining
parameters. Diffusion, chemical reaction and soft annealing appear in the contact
area as a result of thermal stresses. These effects depend exponentially on the heat
development [24]. The productivity is limited by the heat development at increas-
ing process parameters [25]. The quantification of these thermal effects in
grinding can be applied to specify the workpiece properties which are subsequent-
ly required in the application [26]. Based on this knowledge, simulations can be
implemented to determine adequate parameters for the NC-machine in advance.
One method for determining the transformation of the microstructure by heat input
is to examine micro-sections [27].
An overview of different methods for temperature measurements is presented

in [28]. Temperature measurements in grinding are primarily performed using
thermocouples and thermographic cameras. The disadvantage of these temperature
measurement methods is that the temperature can only be determined at the work-
piece. The best results using thermocouples are achieved using epoxy resin for
fixation [29]. Temperature measurements on the workpiece surface using a ther-
mographic camera are influenced by the coolant supply, which affects the work-
piece surface. Therefore, the workpiece has to be isolated from the coolant supply
[30, 31]. An option is to transport the infrared radiation from the contact area by
means of a glass fiber [32]. A further development is to measure directly inside
the grinding wheel using micro sensors. Here, the temperatures are measured
in-situ in the contact area between the grinding wheel and the workpiece. The
measurement signal is transferred by complex equipment in the grinding wheel;
therefore the temperature measurement is very complex [33]
1.3.3.3 Experimental Investigation of the Temperature in the Cutting Zones
To determine the temperature in the cutting zone, experimental investigations are

conducted by means of semi-artificial thermocouples [34-36]. This method can be
used to investigate the temperature distribution in the workpiece and in the chip.
One leg of the thermocouple is made of constantan wire, the other from the ma-
terial to be machined. The basic scheme of this measurement method is shown in
Figure 1.13 a. According to the scheme the wires are welded to the workpiece
with a condenser welding machine. Each constantan leg is placed on a preset
height hi and length relative to the border of the workpiece. If the exact start of the
measurement, which is determined by a trigger signal, and the cutting speed are
known, the distances li can be calculated. Thus, the exact position of the individual
constantan legs relative to the point of the wedge can be calculated and the exact
position of the point is to be measured accordingly. A specimen with the welded
constantan wires is shown in Figure 1.13 b.
Fig. 1.13 a) Scheme of the setup for temperature measurement and b) workpiece with
welded thermocouples
22 E. Abele et al.
In this example, C45 was used as test material and the standard carbide plates
P20 of the company Walter AG were used as inserts. A value of 5° was selected
for the rake angle and a value of 8° for the clearance angle. Characteristic temper-
atures in the primary cutting zone and in the chip are shown in Figure 1.14 [36].
Regarding the starting point of the cutting process, the temperature signal for the
sensor position can be identified. Thus, the temperature in the cutting zones and in
the basic material can be determined. The signal shape and the amplitude during
cutting correspond to the position of the temperature sensor or the constantan leg
respectively in the different layers of the material.
Fig. 1.14 Characteristic signal profile in the primary cutting zone and in the chip
In practice, this method can only be applied for comparably large cutting depths to
achieve reasonable results and so as to be able to determine the position of the constan-
tan leg precisely. This method was used in [37] for the experimental investigation of
the temperatures in the primary and tertiary cutting zone and in the base material.
1.4 Workpiece Analysis

Parts or their surfaces are machined to enable specific functions or tasks. Depend-
ing on these functions the geometry and/or the surface properties are specified. To
validate whether the machined parts meet these specifications their properties have
to be determined after machining. Below, examples are given, which describe the
assessment of surface topography and part geometry. Additionally, some post-
processing procedures are presented.
1.4.1 Surface Assessment

The roughness of technical surfaces is commonly detected via tactile measure-
ments. Therefore, the probe tip, usually a pyramid-shaped diamond, is moved over
the surface of the workpiece at a constant feed rate. The topographical data is de-
scribed by the vertical position shift of the probe tip with reference to the measur-
ing position in the so-called primary profile. To gain the waviness and the rough-
ness profile the measured data is processed with filters which are standardized
within the ISO Standard 11562 [38]. The waviness profile is smoothed and yields
long-wave information without roughness data. On the other hand, the roughness
profile consists only of short-wave information without waviness data. The com-
monly used characteristic values of the surface roughness are derived from the
roughness profile. These are the arithmetic mean surface roughness Ra, the ten-
point height Rz and the maximum roughness Rmax. Of course, there are more
roughness values available for various tasks having different specifications, which
will not be mentioned and explained here.
The arithmetic mean surface roughness Ra is an integral value over the whole
measuring length, which is also used in industrial application and in research.
When calculating Ra, peak values, which can be disadvantageous for some appli-
cations, are averaged. In these cases, the ten-point height Rz is more suitable. To
determine this value the whole measuring length is divided into five sections. In
each section, the maximum roughness, i. e. the difference of the maximum and
minimum height of the profile, is determined. The average value of these five val-
ues is called Rz. The maximum roughness Rmax is the difference of the maximum
and minimum height of the profile over the entire measuring length. It is a very
important value for technical applications such as sealing surfaces where peak
values as well as outliers can lead to difficulties.
Further methods for surface assessment are optical measurement techniques
like interferometry or white light interferometry. For the measurement a beam of
light is divided via a beamsplitter to illuminate two surfaces - workpiece and pla-
nar reference. The reflected beams from workpiece and reference are recombined
and directed to interfere at a detector. Depending on the phase shift of the measur-
ing and reference beam typical fringe patterns can be observed, e. g. parallel
fringes for tilted planar surfaces.
For white light interferometry (Fig. 1.15) the workpiece is scanned in vertical di-
rection so that the light reflected from the surface points interfering with the refer-
ence beam at certain heights will be recorded as surface data. The interferometric
measurement principle is only applicable for optical surfaces; it provides the same
roughness values as mentioned above but considers an area instead of a line profile.
Fig. 1.15 White Light Interferometer (left) and measured surface topography of a diamond
turned sample (right)
24 E. Abele et al.
The main advantages of optical measurements are the short measuring time
(only a few seconds) and, most importantly for optical surfaces, the absence of
any mechanical contact during the measurement. Therefore, the surface cannot be
damaged. Another advantage is the possibility of obtaining spatial structure in-
formation for the surface topography instead of only 2-dimensional information
(profilometry). One disadvantage is the low capability of measuring surfaces with
high slopes which can occur e.g. for spherical, aspherical, structured and free-
formed surfaces. If the slopes exceed a certain angle (typically > 15° to 25°) de-
pending on the numerical aperture of the applied objective, the reflected light
will not enter the microscope objective. For these areas, no measurement data is
recorded.
1.4.2 Geometry Assessment

Similar to the described surface assessment geometric measurements can be car-
ried out either by using mechanical (tactile) techniques or optical techniques.
Therefore, the same advantages and disadvantages apply. Additionally, a fast
evaluation of complex workpieces with a high decomposition can be realized for
optical geometry measurements.
White light fringe projection belongs to the 3-D imaging techniques, which can
create a spatial exposure without the operation of additional axes. The application
range of white light fringe projection lasts from quality conformance tests, reverse
engineering and medical applications to archaeology.
A scanner reaching with white light fringe projection consists of a CCD camera
and a projector. For the registration of a surface the projector generates a periodi-
cal equidistant interference fringe pattern. This projection is distorted by the ob-
ject´s geometry. During the projection, the generated striped pattern is captured by
the CCD-camera. The result of the distortion is a level difference in the structured
image, which is a rate for the altitude h. The dimension h can be calculated by the
scanner software using the principle of triangulation. The distance between projec-
tor and camera is the length L (see Fig. 1.16), and the operating distance A can be
found between the sensor and the surface area. The width of the measuring field is
designated by M, while the angle θ is related to the position of the camera and the
projector. The angle of projection α is between the projector and the surface area.
The observation angle of the camera to the current measuring point is designated
with β. The formula for the calculation of h is given by [39]:
Fig. 1.16 Measuring configuration of a white light fringe projection [38]
The workpiece is scanned from several different positions during the exposures.
After the scanning process is finished the individual exposures are coarsely-
merged to a 3-dimensional general view. This happens by means of characteristic
features on the surface. Afterwards, a precision alignment algorithm is used. The
calculated 3-D model can be exported. The scanner as well as the scanned part can
be seen in Figure 1.17.
Fig. 1.17 Scanner system (a) and a scanned workpiece (b)
The determination of the resulting deviation is evaluated by comparing the

milled part with the reference model. For this purpose, the deviations of all dot
pitches of the actual to the reference model are calculated and plotted in a false
color image [36].
1.5 Conclusion
This chapter presents a general overview of the state of the art for different mea-
surements and test techniques, which are used for the analysis of the interaction
between process and machine tool structure. It has been shown that, depending
upon a special process and machine tool, different measurement and test
26 E. Abele et al.
techniques can be used, depending on the special parameters to be determined.

Additionally, this chapter provides information as to which measuring and testing
procedures are used for the investigation of the process machine interactions.
References
[1] Weck, M.: Werkzeugmaschinen 5: Messtechnische Untersuchung und Beurteilung,
dynamische Stabilität. Springer, Berlin (2006)
[2] DIN55189, Ermittlung von Kennwerten für Pressen der Blechverarbeitung bei sta-
tischer Belastung. Beuth Verlag (1988)
[3] Schindler, J.: Experimentelle Strukturanalyse im Werkzeugmaschinenbau. Antriebs-
technik 34(5), 30 (1995)
[4] Weck, M.: Dynamisches Verhalten spanender Werkzeugmaschinen. Einflussgrößen,
Beurteilungsverfahren, Meßtechnik. Aachen (1971)
[5] He, J., Fu, Z.-F.: Modal Analysis. Butterworth-Heinemann, Oxford (2001)
[6] Ewins, D.J.: Modal Testing: Theory, Practice and Application, 2nd edn. Research
Studies Press Ltd., Hertford-shire (2000)
[7] Heylen, W., Lammens, S., Sas, P.: Modal Analysis Theory and Testing. Katholieke
Universiteit Leuven (2007)
[8] Jansen, T.: Entwicklung einer Simulation für den NC-Formschleifprozess mit To-
russchleifscheiben. Dissertation, Technische Universität Dortmund, Essen, Band 43
(2007)
[9] Derenthal, M.-J.: Bewertung und Optimierung der thermischen Eigenschaften schnel-
llaufender Umformanlagen. Dissertation, Leibniz Universität Hannover (2009)
[10] DIN/ISO 10791-10. Test conditions for machining centres – Part 10: Evaluation of
thermal distortion, Genf. (2007)
[11] ISO 230-3, Test code for machine tools- Part 3: Determination of thermal effects,
Genf. (2007)
[12] Behrens, B.-A., Reuß, C., Vieregge, T.: Thermografie zur Gesenküberwachung.
Schmiede-Journal, Ausgabe, 33–36 (September 2009)
[13] Bernhard, F.: Technische Temperaturmessung. Springer, Berlin (2003) ISBN:
3540626727, 1. Auflage
[14] Rohrbach, C.: Handbuch für elektrisches Messen mechanischer Größen, p. 498. VDI-
Verlag, Düsseldorf (1967)
[15] Thomson, W.: On the Electro-dynamic Qualities of Metals. Philosophical Transac-
tions of the Royal Society of London (1856)
[16] Hoffmann, K.: An Introduction to Measurements using Strain Gauges. Hottinger
Baldwin Messtechnik GmbH, Darmstadt (1989)
[17] Tichy, J., Gautschi, G.: Piezoelektrische Messtechnik, Physikalische Grundlagen,
Kraft-, Druck- und Beschleunigungsaufnehmer, Verstärker. Springer, Heidelberg
(1980)
[18] Klocke, F.: Process Design. Manufacturing Processes 2 – Grinding, Honing, Lapping,
pp. 251–287. Springer, Berlin (2009)
[19] Rosetto, S., Koch, U.: On Investigation of Temperature Distribution on Tool Flank
Surface. CIRP Annals 19(3), 551–557 (1971)
[20] Müller, B.: Thermische Analyse des Zerspanens metallischer Werkstoffe bei hohen
Schnittgeschwindigkeiten. Dissertation. RWTH Aachen (2004)
[21] Vieregge, G.: Zerspanung der Eisenwerkstoffe. Verlag Stahleisen M.B.H., Düsseldorf
(1970)
[22] De Witt, D.P., Nutter, G.D.: Theory and Practice of Radiation Thermometry. Wiley,
New York (1988)
[23] Kitagawa, T., Kubo, A., Maekawa, K.: Temperature and Wear of Cutting Tools in
High-speed Machining of Inconel 718 and Ti-6Al-6V-2Sn. Wear 202, 142–148
(1997)
[24] Kim, N.K., Gou, C., Malkin, S.: Heat Flux Distribution and Energy partition in
Creep-feed grinding. CIRP Annals 46(1), 227–232 (1997)
[25] Lavine, A.S., Malkin, S., Jin, T.C.: Thermal aspects of grinding with CBN wheels.
CIRP Annals 38(1), 557–560 (1989)
[26] Fischbacher, M.: Schleifen - Möglichkeiten zur Beherrschung der Prozesswärme.
IDR 42(I), 44–47 (2007)
[27] Büttner, A.: Das Schleifen sprödharter Werkstoffe mit Diamant-Topfscheiben unter
besonderer Berücksichtigung des Tiefschleifens. Dissertation. Technische Universität
Hannover, Hannover (1968)
[28] Davies, M.A., Ueda, T., M’Saoubi, R., Mullany, B., Cooke, A.L.: On The Measure-
ment of Temperature in Material Removal Processes. CIRP Annals 56(2), 581–604
(2007)
[29] Shen, B., Xiao, G., Guo, C., Malkin, S., Shih, A.J.: Thermocouple Fixation Method
for Grinding Temperature Measurement. Journal of Manufacturing Science and Engi-
neering 139, 0510141–0510148 (2008)
[30] Hoffmeister, H.-W., Maiz, K.: Wärmemanagement beim Flachprofilschleifen. In:
Hoffmeister, H.-W., Denkena, B. (eds.) Jahrbuch Schleifen, Honen, Läppen und Po-
lieren, vol. 61, pp. 17–26. Vulkan-Verlag, Essen (2004)
[31] Hoffmeister, H.-W., Machanova, I., Maiz, K.: Simulation of Grinding Processes with
FEA. In: CIRP International Workshop on Modelling of Machining Operations, pp.
329–333 (2005)
[32] Hwang, J., Kompella, S., Chandrasekar, S., Farris, T.N.: Measurement of Tempera-
ture Field in Surface Grinding Using Infra-Red (IR) Imaging Systems. Journal of Tri-
bology 125, 377–383 (2003)
[33] Brinksmeier, E., Heinzel, C., Meyer, L.: Development and Application of a Wheel
Based Process Monitoring System in Grinding. CIRP Annals 54(1), 301–304 (2005)
[34] Körtvelyessy, L.V.: Thermoelement Praxis. Vulkan Verlag, Essen (1981)
[35] Frohmüller, R., Knoche, H.-J., Lierath, F.: Aufbau und Erprobung von Temperatur-
messeinrichtungen durch das IFQ im Rahmen des Schwerpunktprogramms Spanen
Metallischer Werkstoffe mit hoher Geschwindigkeit. In: Spanen Metallischer
Werkstoffe Mit Hohen Geschwindigkeiten Kolloquium des Schwerpunktprogramms
der DFG, pp. 108–115 (1999)
[36] Zacher, M.: Integration eines optischen 3D-Sensors in ein Koordinatenmessgerät für
die Digitalisierung komplexer Oberflächen. RWTH Aachen, Dissertation (2004)
[37] Heisel, U., Storchak, M., Stehle, T., Korotkih, M.: Temperaturbestimmung in den
Zerspanzonen. wt Werkstattstechnik Online 100(5), 365–370 (2010)
[38] EN ISO 11562, Geometrical Product Specifications (GPS) – Surface texture: Profile
method – Metrological characteristics of phase correct filters (1998)
[39] Abele, E., Bauer, J.: Kamerabasierte Bahnkorrektur für das Fräsen mit Industrierobo-
tern. wt Werkstatttechnik Online, 9 (2010)
[40] Brecher, C., Denkena, B., Grossmann, K., Steinmann, P., Bouabid, A., Heinisch, D.,
Hermes, R., Löser, M.: Identification of Weak Spots in the Metrological Investigation
of Dynamic Machine Behaviour. Production Engineering – Research and Develop-
ment 9(6), 679–689 (2011)
Chapter 2
Modeling and Simulation
C. Brecher, A. Bouabid, M. Deichmueller, B. Denkena, K. Großmann,

A. Hardtmann, D. Hömberg, R. Hermes, F. Klocke, M. Löser,
O. Rott, P. Steinmann, and M. Weiß
Abstract. One focus of the Priority Program 1180 is the prediction of process
machine interactions. The investigated manufacturing processes as well as the ma-
chine tool behavior and the physical phenomena vary within the projects of this
program. So depending on the issues that were investigated, the modeling ap-
proach that is best suitable for the specific problem has to be applied. To predict
the interactions these models have to be coupled and simulated. Besides the mod-
eling approaches different simulation techniques have also been applied. This
chapter gives an overview of the applied models of the structural machine beha-
vior and the manufacturing processes, the coupling of these models as well as the
simulation techniques that were used.
2.1 Introduction
The prediction of process machine interactions requires models of the subsystems

“machine” and “process”. Which modeling approach is appropriate for a specific
problem depends on the type of process, the character of the investigated interac-
tion and the physical phenomena relevant for this interaction. However, other
factors may also have an impact, for example, the computation time or the effort
for modeling as well as the effort of a parameter identification based on measured
data.
A comprehensive overview of current issues and approaches in the field of pre-
dicting process machine interactions is given by Brecher et al. [1]. This chapter
focuses on issues that are relevant within the projects of the Priority Program
1180. It gives basic information about the modeling techniques and the coupling
methods. Most of the projects within the priority program scrutinized the interac-
tions between the process forces and the static and dynamic displacements at the
contact zone of tool and workpiece.
To describe the dynamic behavior of the machine multi-body systems, finite
element models and analogous models are applied. The first two approaches con-
tain structural information of the machine. This makes it much easier to apply
changes of structural parameters. The analogous models are abstract models where
information about the structure is lost but they can be parameterized much easier
30 C. Brecher et al.
using measured frequency response functions. The different approaches to model

the machine behavior differ, for example, in terms of degrees of freedom or the
ability of coping with nonlinearities.
The applied approaches of process models differ for cutting, grinding and form-
ing processes. In cutting, simple empirical models are applied in most cases. In
grinding, empirical models as well as finite element models are applied. In form-
ing simulations, only finite element approaches were used. To couple and simulate
the process and machine models coupled simulations as well as model integration
within the same software were applied.
2.2 Models of the Machine Behavior
2.2.1 Multi-Body Models
A multi-body simulation (MBS) system generally consists of different stiff bodies,

which can conduct defined movements. They are connected by different kinds of
joints like revolute, prismatic, ball or cardan joints. Movements and reactions of
the bodies can be simulated using algebraic-kinematic relations and external
forces in the time domain. The main advantage of MBS is the possibility to simu-
late large scale movements and rotations. In contrary to a MBS a linear FE-
simulation e. g. is able to represent the dynamic behaviour of a machine tool only
in one defined position. Thus, non-linearities cannot be simulated [2].
In a flexible MBS model, the advantages of both MBS and FE-simulation are
combined. As the FEM is able to simulate deformations of parts and the MBS can
realize movements of the modelled parts the dynamic machine behaviour can be
described for every position of the machine tool slides.
For the implementation of a flexible MBS the structural parts of a machine tool
have to be converted into flexible bodies. A flexible body allows the description of
its flexible properties in defined points of force transmission. The count of possi-
ble part deformations is reduced by means of modal superposition. Hence, linear
part deformations can be described by the combination of linear eigenmodes.
Fig. 2.1 shows the difference of a flexible MBS model and a moveable flexible
MBS using the example of a machine table. As the flexible MBS uses fixed ele-
ments to constrain the machine table to the machine bed the moveable flexible
MBS realizes a movement of the machine table. The occurring forces are
transferred to the machine bed by dividing the load onto different nodes of the
guideway. Hence, a variable load transmission can be realized and the machine
behavior can be depicted in more detail.
2 Modeling and Simulation 31
Fig. 2.1 Principle of flexible and moveable flexible multi-body simulation [3]
2.2.2 Finite Element Models

In contradiction to the multi-body systems treated in the last section, structural
analyses are primarily concerned with deformable structures, which consist of an
infinite number of single material points. Also, structural analyses are more inter-
ested in the distribution of physical quantities within the structure, e. g. the stress
distribution.
In machine tools, the finite element method can be applied in order to analyze
both the static and the dynamic machine behavior. In particular, time-dependent
and coupled processes are of high interest. Machine tools consist, in many cases,
of spinning cylindrical parts, which permit some simplifications of the solution
procedure through their form and periodic motion. Then, computationally more ef-
ficient solution techniques can be used, e. g. the Arbitrary Lagrangian Eulerian
(ALE) approach. In the following, the basics of a transient finite element solution
procedure are presented, exemplarily for the case of a spinning wheel, which can
represent a grinding wheel, for example. Detailed information can be found in [4].
Bold symbols denote vectors or tensors while scalars are denoted in normal font
style.
In the ALE approach, three states (or configurations) of the spinning wheel can
be distinguished, Fig. 2.2. B0 denotes the non-deformed state. In this configura-
tion, each of the material points constituting the wheel can be localized by the co-
ordinates X0. By means of a time-dependent rotation matrix R(t) one passes to the
rotated (or reference) configuration Br, where material points are described by the
coordinates
X = R (t ) ⋅ X 0 (2.1)
X0 describes the geometry of the spinning wheel at the idle state and does not de-
pend on the time t. X is time-dependent since the position of a material point
changes due to the rotation, regardless of whether the angular velocity ω is con-
stant or not.
Finally, the deformation of the spinning wheel is given at the deformed (or to-
tal) configuration Bt by the function ϕ0 or ϕ, depending on the choice of the con-
figuration one takes as reference, i. e. B0 or Br respectively. Thus, the coordinates
x of a material point in the total configuration can be given by
x = ϕ0 ( X 0 , t ) = ϕ ( X(t ) , t ) = ϕ ( R(t ) ⋅ X 0 , t ) (2.2)
The total deformation ϕ depends on the time t in two ways: indirectly via X(t) and,
in addition, in an explicit way via t. The dependency on X(t) includes the fact that
the less important part of the kinematics, i. e. the rigid body rotation, is anticipated
in the model formulation. The direct dependency on time t, however, expresses the
most important part of the kinematics, i. e. the transient deformation due to the
process forces. For the position vector x, a total and a partial time derivative x
and ∂ t x can be defined, depending on whether the rigid body rotation is included
or not.
x = ϕ 0 ( X 0 , t ) = ϕ ( X(t ) , t ) = ϕ ( R (t ) ⋅ X 0 , t )
Initial ϕ0 Deformed
Configuration B0 X0
Configuration Bt
x
X
R ⋅ X0 ϕ
Reference
Configuration Br
Fig. 2.2 Configurations corresponding to the ALE approach. In addition to the initial non-
deformed state and the total deformed state there is an intermediate rotated state one refers
to when solving the problem
The stress tensor can be defined in different ways. Relating a force increment
dF to a surface element da, both in the total configuration, results in the Cauchy or
true stress tensor σ. Taking, however, the surface element in the initial or rotated
configuration dA0 or dA leads to the Piola-stress tensor P0 or P respectively.
In the following, the basic relations of the finite element model of the spinning
wheel are given. All equations are written with respect to the rotated configuration
Br. The balance of momentum writes
Div P + b r = ρ r x (2.3)
where ρr is the material density. The external body forces br, caused by gravity for
example, are not important for the present presentation. The total acceleration x
can be determined by totally deriving the position vector x twice. After some deri-
vations and transformations one obtains the following relations:
Total velocity:  +∂ ϕ = F⋅X

x = F ⋅ X  +v (2.4)
t
Total acceleration: [
x = G : X ]
 + [ F ⋅ Ω + 2 L ]⋅ X
 ⊗X  +a (2.5)
Herein, the following entities have been used:
Deformation gradient: F = ∂ Xϕ (2.6)
Velocity gradient: L = ∂ 2Xtϕ = ∂ X v (2.7)
Gradient of the deformation gradient: G = ∂XF (2.8)
Local velocity: v = ∂ tϕ (2.9)
Local acceleration: a = ∂ tt2ϕ = ∂ t v (2.10)
Spin tensor: [Ω ij ] = ω0 − ω

0 
(2.11)

Guiding velocity:  = Ω⋅X

X (2.12)
Stresses are related to the strains by means of a material law, e. g. according to the
Neo-Hookean model. In this model, the isotropic strain energy function W is given
by
μ
W= [ C : 1 − 2 ] − μ lnJ + λ ln 2 J (2.13)
2 2
The Piola-stress results from the relation
P = F ⋅S (2.14)
with the Piola-Kirchhoff stress
S=2
∂W λ
=
∂C 2
[ ]
J 2 − 1 C −1 + μ 1 − C −1 [ ] (2.15)
In Eq. (2.13)-(2.15), λ and μ designate the Lamé parameters. In addition, the fol-
lowing entities have been used:
Right Cauchy-Green strain tensor: C = Ft ⋅ F (2.16)

Jacobian of the deformation gradient F: J = Det F (2.17)
Finally, the material stiffness results from
∂S ∂ 2W
C=2 =4 (2.18)
∂C ∂C ⊗ ∂C
Now, when the spinning wheel problem is discretized first with respect to space
and then time and when substituting Eq. (2.3) for Eq. (2.4)-(2.12), one can show,
that the forces at an arbitrary node I can be written at time step tn+1 as:
 [ ][ F ]
I  h  h dV
n +1 = − ρ r ∂ X N ⋅ X
FIspn ⋅X
h
Spinning forces: n +1 r (2.19)
Br
FIintn +1 =  [F ]
⋅ S hn +1 ⋅ ∂ X N I dVr
h
Internal forces: n +1 (2.20)
Br
Inertia forces: FIinen + 1 = ρ

Br
r N I a hn +1 dVr (2.21)
⋅ h
n +1 =
Coriolis-type forces: FIcor 
Br
2 ρr NI Lhn+1 ⋅ X dVr (2.22)
where the notation (⋅) h means that after spatial discretization the corresponding
entity (⋅) has, of course, to be written in discretized form, i. e. as a function of the
value at node I, weighted by the element shape function N I. The local velocity and
acceleration at time step tn+1 can be written according to the Newmark-scheme as
γ  γ  γ 
v n +1 = ∂ t d n +1 ≈ [d n +1 − d n ] + 1 −  v n + Δt − Δt  a n (2.23)
β Δt  β  2β 
 1 
a n +1 = ∂ tt2 d n +1 ≈
1
[d n+1 − d n ] − 1
vn −  − 1 a n (2.24)
β Δt 2 β Δt  2β 
Herein, Δt and dn+1 denote the time increment to be appropriately chosen and the
time-dependent nodal position vector to be solved for respectively. Also, it has
been assumed that the state of the wheel is known at time step tn. The parameters β
and γ appearing in Eq. (2.23)-(2.24) are the Newmark parameters to be chosen
such that
(β , γ ) =  1 , 1  (2.25)
4 2
Relations (2.19)-(2.22) in sum give the force residual vector
−R n +1 = Fnspn
+1 + Fn +1 + Fn +1 + Fn +1
int ine cor
(2.26)
to be iteratively solved, e. g. by means of the Newton-Raphson procedure when

appropriate boundary conditions have been defined. The linearization of the force
residual vector R leads to the stiffness matrix K, which relates the force at a node I
with the displacement at a node J caused by that force. In analogy to the force, the
stiffness can be given in form of the following single contributions:
 [ h
IJ n + 1 = − ρ r ∂ X N ⋅ X
K spn ][ ∂ ]
N J ⋅ X h I dV r
I
Spinning part: X (2.27)
Br
n +1 =  [∂ ]
⋅ S hn +1 ⋅ ∂ x N J I dVr
geo I
Geometric part: K IJ xN (2.28)
Br
Material part: 
h I
[
IJ n +1 = Fn +1 ⋅ ∂ x N ⋅ C n +1 ⋅ ∂ x N ⋅ Fn +1 dVr
K mat h J h
][ ] t
(2.29)
Br
 I 1 
Inertia part: IJ n +1 =
K ine ρ
Br
r N
 β Δt 2
N J  I dVr

(2.30)
 γ ⋅ h
IJ n +1 =
Coriolis-type part: K cor 
Br
ρr N I

2
β Δt
∂ x N I ⋅ X  I dVr

(2.31)
Herein, I denotes the identity matrix. By summing all stiffness parts, the total
stiffness results in
K n +1 = K nspn
+ 1 + K n + 1 + K n + 1 + K n + 1 + K n +1
geo mat ine cor
(2.32)
which again results together with the force residual R in the Newton-Raphson step
K n + 1 ⋅ Δd n + 1 = R n + 1 (2.33)
where Δdn+1 denotes the iterative increment of the nodal displacement vector.
Once the nodal position vector dn+1 is known all other entities, e. g. the stress or
the strain tensor, can be determined.
2.2.3 Analogous Models

The multi-body and the FE models are based on the structure of the mechanical
system. In some cases, it sufficient to describe the dynamic behaviour by abstract
models, for example when only the dynamic behaviour at the tool center point
(TCP) is of interest for the desired simulation results.
The dynamic behaviour of the structure is represented by a system of masses
and spring/dampers, which act like the machine at observed points. In a simple
case, the frequency response function at one point can be written as the sum of the
frequency responses of n single degree of freedom mass spring damper systems
G ( jω ) = −m ω
1
(2.34)
n n
2
+ jωd n + c n
with the modal mass m, the damping d and the stiffness c. The parameters of these
models can be identified from frequency response functions measured at the real
machine structure. For this, commercial software for a modal analysis or self-
written routines can be used. In many cases, the fitting between the dynamic be-
havior of the real structure and the model is better than the one that can be
achieved by structure based models, especially in the case of multi-degree of free-
dom behavior of the machine, see chapter 1. Since the abstract models contain
only a few degrees of freedom the computation time for simulations is much
shorter than when structure-based models – like FE or MBS models - are applied.
A disadvantage of the analogous models is the loss of information about the
structure of the machine. These abstract models are only valid for a given configu-
ration or state of the machine. Therefore, a change of local parameters or a change
of the position of the axes can only be considered by a new set of modal parame-
ters. A structure-based model can help to identify local parameters of the real
structure that cannot be measured directly. An analogous model is not appropriate
for such a task.
2.3 Process Models
2.3.1 Cutting Force Models

In addition to the term describing the dynamic machine tool behavior the cutting
process term is needed to resemble the process-machine interaction in a closed
loop. The cutting process responds to a change in the cutting geometry with an al-
teration in process forces. The calculation of process forces regarding the dynamic
variation of the depth of cut was the priority of the research in the years 1960-
1980 [5]. Cutting forces are not only influenced by a relative motion between the
workpiece and cutting tool (inner chip thickness modulation) but also by the ripple
left on the workpiece surface (outer chip thickness modulation) at the previous
revolution (turning) or by the last cutting edge (milling). Not only the modeling of
the milling process demands for a considerable effort but also the determination of
the force coefficients is known as an extensive procedure. In addition to modeling
the determination of force coefficients, which had been defined by the simulation
model, was extensive. Radharamanan summarizes the work on this field of re-
search in detail [6].
A common method to estimate the cutting forces acting in turning and milling
processes is the analytical description of the interrelationships at the cutting edge
[7, 8]. There are three analytical formulations in literature, which differ in their
mathematical depiction and the determination of the characteristic force
coefficients, Fig. 2.3. The method of mechanistic description is based on the
modeling of so-called shear planes. In this regard, a forming process is adopted in
the ablated material. The assumed shear stress influences the amount of the
calculated process forces decisively. In this context, for example, the shear plane
model by Merchant [9] or the "Slip-line field" model by Fang [10] have to be
mentioned.
Mechanistic Model Models based on Exponential Functions Linear Models
e.g. by Fang et al., e.g. by Kienzle e.g. by Stèphán, Feng e.g. by Altintas
Ernst & Merchant
 b⋅h  −
Fi = b ⋅ K i1.1 ⋅ h1 mi Fi = b ⋅ K i ⋅ h i + b ⋅ K ie Fi = K ic ⋅ b ⋅ h + b ⋅ K ie
x
Fs = τ s ⋅  
 sin φc 
Ki1.1=f(vc, WZ / WST) Ki=f(vc, WZ / WST) Kic=f(vc, WZ / WST)
1-mi=f(vc, WZ / WST) xi=f(vc, WZ / WST) Kie=f(vc, WZ / WST)
αc
F F F
Cutting Tool
φc
Workpiece
h h h
Model Parameterization Model Parameterization Model Parameterization Model Parameterization

through Experimental Analysis through Cutting Tests through Cutting Tests through Cutting Tests
Fig. 2.3 Analytical formulation of cutting forces
Especially for the simulation of process stability linear models and models
based on exponential functions have been established to estimate the process
forces. Using such simulation models emerging friction between the workpiece
and the cutting edge can be captured by appropriate additional terms [7, 11].
These linear models assume a linear relationship between process forces and chip
thickness. There are different approaches for this purpose, e. g. by Tlusty [12],
Altintas [13] and Weck [14].
Models based on exponential functions derive process forces out of a non-linear
relationship to the chip thickness. Appropriate models for the description of the
processes have been developed, for example by Stepan [15] and Feng [16].
Investigations have shown that the feed motion-dependence of the dynamic force
variation may explain differences in process stability, which were also partially
observed in practice. The parameterization of the shown force models is
carried out on the basis of cutting tests and therefore valid only for one defined
cutting edge-workpiece material combination. In some cases, the parameterization
can be adapted to different experimental conditions with the help of adjusting
factors [17].
Especially for the depiction of chip formation, burr formation and the chip tem-
perature the cutting simulation using the finite element method is an important
tool [18]. In such a simulation tool, the chip formation is discretized for a small
surrounding between the workpiece and the cutting tool by finite elements in a
sufficiently small mesh size. These simulations are not linear since large dis-
placements and deformations, temperature and strain rate-dependent plasticity of
the workpiece at the cutting point as well as contact between tool and workpiece
must be taken into account. The results of the force calculation using FEM do not
provide sufficiently accurate results for the passive force in particular.
For mapping the cutting forces in order to simulate the process stability of a
machining process in most cases the afore-mentioned analytical models are used.
However, the problem is modeling the dynamic cutting forces using data from sta-
tionary chipping processes. Fig. 2.4 shows the geometric engagement of a cutting
edge for dynamic cutting conditions in one plane. The rapid change of the oscillat-
ing engagement conditions leads to digressive results in cutting force calculation.
Regardless of the approach of modeling process forces (mechanistic, empirical)
a multiplicity of research works have demonstrated, that dynamic effects have to
be taken into account additionally for stability simulation in the case of improving
process models. These models consider the inner chip thickness modulation
(wave-cutting) and the outer chip thickness modulation (wave-removing) as an
independent input parameter as well as a phase shift between the dynamic cutting
force and chip thickness modulation.
 The entire current average

depth h is composed of static h = xstat + xd(t) - xd(t-Tt) vc
and dynamic portions.
 Location and size of the
primary shear zone vary. xstat xd(t-Tt)
 A direct connection between
xd(t)
the chip thickness h and the
process force F is in contrast
to the mechanical conditions . F Fp
Fc
vc : Cutting Speed xstat : Static Portion of Tool Deflection

F: Total Process Force xd(t) : Dynamic Portion of Tool Deflection
Fc : Cutting Force (in Direction vc) xd(t-Tt) : Dynamic Portion of Tool Deflection
Fp : Passive Force (perpendicular to vc) Previous Revolution
Fig. 2.4 Plain dynamic cutting
In particular, the variation of the position of the shear plane is quite often used
for the determination of the dynamic cutting forces from existing static models.
Kim and Lee [19], for example, give an analytical description of the shear angle
regarding the inner and outer chip modulation. This dynamic force model
considers the transfer function of the tool and its derivation in the direction of
cutting speed and direction of the chip thickness variation. Although the
mentioned models can be used for stability simulation they have not been applied
for this purpose so far.
Van Brussel [20] and Werntze [21] propose simpler, empirical models. These
take into account the influence of the inner and outer chip thickness modulation in
terms of a proportional relation and a phase shift. The phase shift allows the
modeling of a time lag between changes in chip thickness and the corresponding
changes in force.
2.3.2 Abrasive Machining

In contrast to conventional machining where workpiece material is removed by a
small number of defined cutting edges a large number of abrasive particles or
grains acts in abrasive machining processes. In loose abrasive processes like po-
lishing or lapping, the grains are not connected to each other and move indepen-
dently from each other. In bonded abrasive processes like grinding or honing, the
grains are connected by a bond, e. g. resin, vitrified or metal bond. Numerous
models to describe grinding processes exist today. They can be divided into fun-
damental approaches, kinematic models, finite element method (FEM), molecular
dynamics, artificial neural networks and rule based models. An overview is given
in [22, 23]. Process machine interaction models for grinding are summarized in
[1, 24]. Grinding models describe the influence of various parameters like depth of
cut or cutting speed on forces, temperatures or surface roughness, for example.
Since a universal model for all grinding processes has not yet been developed [25]
it depends on the grinding operation and the contact conditions, which model is
suitable. In general, the grinding models can be divided into microscopic and ma-
croscopic approaches which are described in the following.
Microscopic Approaches Macroscopic Approaches
Modelling FEM, Kinematic- Kinematic-geometrical

methods geometrical simulations simulations (Dexel, Voxel)
Application Detailed simulation of Simulation of material
area chip formation, grinding removal, geometric
forces, heat and surface description of contact area,
quality and application of grinding
force models
Restrictions High computation time, Lower resolution in

high effort to model simulation result than with
grinding wheel geometry microscopic approaches,
surface quality cannot be
simulated.
Fig. 2.5 Modeling approaches for grinding processes

2.3.2.1 Microscopic Approaches
The aim of microscopic process models is to calculate e. g. local stresses or tem-

perature, or to gain knowledge on chip formation mechanisms. The microscopic
shape of a grinding wheel is determined either by topography measurement [26] or
by generative modeling using a mathematical description of the grain morphology
[27]. Mostly, statistical methods are applied for modeling the grain distribution
[28]. Kinematic-geometrical models, as exemplarily described in [27], assume an
ideal chip formation without plowing or similar effects. Process forces are calcu-
lated on the basis of undeformed chips of single grains. In other approaches, the
finite element method is used to model the engagement of grinding wheel and
workpiece [29]. Due to the high computation time, especially of 3D-FEM, only
small parts of a grinding wheel are modeled, normally on a small set of grains or
single grit scratching tests [30].
2.3.2.2 Macroscopic Approaches
In macroscopic approaches, the grinding process is modeled by calculating the en-

gagement of grinding wheel and workpiece geometrically from a macroscopic
view, i. e. no grains are modeled. In macroscopic approaches, parameters like
equivalent chip thickness heq or geometrical contact length lg are calculated and
empirical or FEM grinding force models applied [31]. In the kinematic-
geometrical simulation, the workpiece is discretized using dexels, voxels, or
boundary representation models, for example. A more detailed description of
process models for surface grinding, NC-shape grinding, pendulum and speed
stroke grinding and tool grinding can be found in the section “Grinding” of this
book.
2.3.3 Metal Forming

During an optimal process design the determination of the stresses, forces and
energy is an important topic. While the processes of the sheet-metal forming are
often limited by workpiece-lateral demands such as tensions and instabilities the
process limits of the bulk forming can usually be detected in the tools. Hence,
considering temporally and locally different loads the tool and workpiece must be
dimensioned in such a way that the tool is predicted from plastic deformations and
breakage and the tool wear adjusted to the desired life time. For the optimal ma-
chine selection the knowledge of value and localization of the necessary forming
force as well as the value of the deformation energy are important. At present,
there is a set of methods for the pre-determination of the stresses and forces. Ta-
ble 2.1 shows a selection of widely-used methods [32].
Table 2.1 Methods for the determination of stresses, forces and energy: dxi - infinitesimal
small dimension, kf – yield stress, V – forming volume, η – deformation efficiency, W –
deformation energy, Fm – temporal mean value of deformation force F, σc, τc – contact
stresses, σm – temporal and local mean value of normal stress [32]
Method Number input value output value

Increase of simplifying
Increase of effort and
of dxi
Method of deformation energy 0 kf, V, η σm, Fm, W
assumptions
Strip method 1 kf, V, τc or μ σ, F, (W)

accuracy
Slip-line method 2 kf, τc σ,

Method of upper and lower 3 kf, τc or μ F
bounds
Finite-Element-Method 3 kf, σc, τc or μ σ, F
The methods are tabulated according to the increasing value of an infinitesimal

small dimension dxi, which is the basis for the selected section models. Such as the
number of simplifying assumptions is reduced, the effort and accuracy of the cal-
culation as well as the number of the included small dimensions are increased. The
first two methods are regarded to the elementary theory, the other ones to the
higher theory of plasticity.
The majority of elementary methods is valid under the following conditions
[33]:
• The workpiece volume remains constant during the forming process.
• The material behaves homogeneously and isotropically, i. e. locally different
material properties are generally not considered.
• The elastic deformations are negligibly small compared with the plastic defor-
mations.
• The yield stress is given as function of material, (equivalent) strain, strain rate
and temperature.
• The material behavior follows the Tresca’s yield criterion (Maximum Shear
Stress Theory) and the v. Mises’ yield criterion (Distortion Energy Theory) re-
spectively.
• The contact shear stress is given with a friction formulation, e. g. using the fric-
tion value µ or the friction shear factor m.
• Forces of inertia and weight are neglected.
The most frequently applied theoretical methods are
• Method of deformation energy,
• Strip method,
• Slip-line method,
• Method of the upper and lower bounds,
• Method of weighted residuals and
• Finite-Element-Method.
The method of deformation energy is the most elementary method to calculate the
deformation forces. This method is based on the law of energy conservation with
the applied outer energy
Wa = F ⋅ s (2.35)
equaling the inner strain energy

1
Wi = k fm V ⋅ φ . (2.36)
η
In this formula, kfm is the mean value of yield stress and ϕ the equivalent plastic
strain. The inner energy is equivalent to the total deformation energy
Wtot = Wid + WFR + WSh + WB , (2.37)
summarizing the expressions for ideal, frictional, shearing and bending energy.
The ratio of ideal deformation energy and total deformation energy is referred to
as the deformation efficiency
Wid
η= . (2.38)
Wtot
With these preliminary considerations, the determination of deforming forces can

be derived for several forming processes. In order to exemplify this procedure the
derivation of the forming forces is shown for solid forward extrusion and deep
drawing in Fig. 2.6 and Fig. 2.7.
1
FExt Wa = FExt ⋅ s u , Wi = k fm V ⋅ φ
η
su A0
with V = A 0 s u and φ = ln
h0 d 0 (A0) A1
1 A
Wa = Wi : FExt ⋅ s u = k fm A 0 s u ⋅ ln 0
η A1
1 A0
FExt = k fm A 0 ⋅ ln (2.39)
η A1
d 1 (A1 )
Fig. 2.6 Solid forward extrusion

1
Wa = FD , max ⋅ s u , Wi = k fm V ⋅ φ
η
d0 dp d1
with V = π d 1s0 ⋅ su and φ = ln
d0
d1
FBH FBH
FD Wa = Wi :
s0 1 d
su FD ,max ⋅ s u = η k fm πd1 ⋅ s 0 ⋅ s u ⋅ ln d 0
1
rM
1 d1
FD ,max = k fm π d 1s 0 ⋅ ln (2.40)
η d0
Fig. 2.7 Deep drawing of a cylindrical cup
In both examples, the friction, shear and bending behavior are summarized in
the deformation efficiency factor. More detailed approaches considering the fric-
tion between die and workpiece result in the following formulas, calculating the
maximal forming forces in solid forward extrusion [34]
2  μ  A0 
FExt = k fm A 0 ⋅  tan α + 1 +  ln  + π ⋅ d0 ⋅ kf 0 ⋅ μ ⋅ h0 (2.41)
3  sin 2α  A1 
and for deep drawing processes [35]
 μπ  d 2 ⋅ μFBH  s 
FD,max = πd m s 0 e 2 1.15k fmI ⋅ ln p +  + k fmII 0  . (2.42)
  d m πd m ⋅ s 0  2rM 
  
This includes the blankholder force FBH, the mean wall diameter d m = d1 + s 0 and
the outside diameter of the flange when the drawing force has achieved a maxi-
mum value d p = 0.7d 02 + 0.3d 2m .
The presented elementary methods provide a solution for a set of problems
without substantial effort. This is achieved with simplifying assumptions deviating
from the real behavior. If the simplifying assumptions do not meet the require-
ments of accuracy, the Finite-Element-Method (FEM) could be used, however,
with a higher effort and costs. Thus, the following factors can be considered in an
improved way:
• different material characteristics,
• complicated geometry of the parts as well as
• inhomogeneous and unsteady mechanical and thermal behavior.

• The FEM is used in metal forming for the simulation of forming processes to
determine instantaneous and intermediate states as well as the design of highly-
stressed active elements of metal-forming tools (e. g. forging or extrusion dies)
[36].
2.4 Coupling of Models
2.4.1 Analytical Considerations

2.4.1.1 General Setting
In general, independent representations for structure and process are the starting
point for the development of models describing the process-structure interaction.
Depending on the properties of interest engineers can choose among a large varie-
ty of structure models. Mathematically, structure models are understood as equa-
tions of motion corresponding to
• a complex multi-body system,
• an abstract system with multiple degrees of freedom describing the dynamics of
a selected point of a large structure,
• a system of partial differential equations discretized with finite elements in
space,
• a coupled system of ordinary and discretized partial differential equations.
The above-mentioned models can be summarized by the following system of dif-
ferential (algebraic) equations
( )
M (t , q, p M )q = f t , q, q , p f , (2.43)
with pM, pf denoting parameter vectors, which are usually determined fitting struc-
ture simulation results to the corresponding experimental data. The vector q
represents the general set of coordinates. If (2.43) has been directly derived from
modal analysis data, the mass matrix is usually constant and the right hand side is
a linear function of q and q .
Process models relate a geometrical parameter vector g and the relative velocity
v between tool and workpiece to the cutting force vector F acting on the tool. In
milling, the geometrical parameter vector consists of cutting width b and uncut
chip thickness h. A large class of process models can be expressed in terms of the
force they exhibit
~
F = F ( g , v, p g ) , (2.44)
where pg represents an empirical parameter vector depending on tool and work-

piece material and on the tool geometry, see section 2.3.1.
In stationary cutting or grinding processes with sufficiently small cutting
forces, the geometrical parameters and the relative velocity do not deviate notice-
ably from the desired values regulated by the machine control system. However, if
the cutting forces become large, increasing machine structure oscillations induce
larger variations of the geometry parameter vector and the relative velocity. This,
in turn, leads to undesired cutting force variations, which possibly cause a further
increase of the machine structure oscillations. In order to describe these process
structure interaction phenomena both aspects have to be incorporated into the
model equations. The variation of the geometrical parameter can be formulated as
a non-linear functional involving the history of the state variable arising from the
structure model (2.43), i. e.
t
g = [t , q(.)] = g~ (s, q(s ))ds .
 (2.45)
t0
The relative velocity between tool and workpiece is usually also a non-linear func-
tion depending on the state variable and the corresponding time derivative, i. e.
v=~ v (t , q, q ) (2.46)
Moreover, an additional term appearing on the right hand side of (2.43) has to be
introduced to model the effect of the varying process forces arising from (2.44)
and (2.45), (2.46). Since the forces usually act on the tool or, in terms of the struc-
ture model, on an element representing the tool an additional model equation has
to be developed to get the appropriate contribution to the right hand side of (2.43).
In a general setting, such an expression is given by
f F = f F (t , q, F ) .
~
(2.47)
Thus, the modified structure model involving the additional term due to the pres-
ence of process forces reads
( ) ~
M (t , q, p M )q = f t , q, q , p f + f F (t , q, F ) (2.48)
As outlined above, phenomena related to process structure interactions can be

modeled by strongly coupled systems, possibly involving the history of the state
variable. The equations (2.48), (2.47), (2.46) and (2.45) are an abstract example
for the model equation corresponding to an interacting system. In each applica-
tion, the adoption of the general system may lead to different equations, which
have to be solved with tailored numerical algorithms, as shown in the following
sections. However, in order to illustrate the different aspects of the general model
we focused on a simple milling system at first.
2.4.1.2 Example: A Simple Milling System
A simple system possessing all the important features to model stability problems
in milling is illustrated in Fig. 2.8.
Ff
fz
ϕ h
dx Fc
m
kx
y
ky dy
x
Fig. 2.8 Scheme of a simple milling system
For the present example, the equation of motion representing the machine struc-
ture reduces to the system
 m 0  x  dx 0  x   k x 0  x   Fx 
   = −   −    +   , (2.49)
 
 0 m  y  0 d y  y   0 k y  y   Fy 
The unknown model parameters are pM = m and pf = (dx,dy,kx,ky)T. Note that (2.49)
is a very simple form of (2.48). The cutting force model corresponding to (2) is a
linear function relating the cutting cross section h ap and the cutting forces acting
on the tip of the cutting edge, i. e.
 Ff   K f 
  =  a p h . (2.50)
 Fc   K c 
The parameter vector is pg = (Kf,Kc)T. The expression corresponding to (2.45), i. e.
the variation of the geometrical parameter, consists of a stationary part hstat and a
dynamic part hdyn. The dynamic part hdyn represents the modulation of the uncut
chip thickness due to the structure oscillations
h = hstat + hdyn = f z sin ϕ + ( x(t ) − x(t − τ )) sin ϕ + ( y(t ) − y(t − τ )) cosϕ (2.51)
Note that instead of h nonlinear models usually involve the positive part of the un-
cut chip thickness, i. e. h+ = max(h,0). Since the cutting force components in
(2.50) are given in the rotating reference frame of the cutter an additional trans-
formation has to be introduced to get the corresponding force components into the
global reference frame, i. e.
 Fx   sin ϕ cos ϕ  F f 
  = −γ (ϕ )  , (2.52)
F
 y  − cos ϕ sin ϕ  Fc 
with γ: [0, 2 π ]  {0,1} denoting a function that switches from one to zero, if the
corresponding tooth is not cutting. Again, equation (2.52) can be interpreted as a
simple version of (2.47). Since the expressions for the cutting forces, the uncut
chip thickness, and the transformation of the cutting forces are given explicitly
(2.52), (2.51), (2.50) and (2.49) can be summarized by the following first order
delay differential equation (DDE)
u (t ) = [A − C (t )]u (t ) + C (t )u (t − τ ) + bstat (t ) , (2.53)
where C(t) denotes a non-smooth τ-periodic matrix. The vector bstat(t) represents
the external forces related to the stationary uncut chip thickness hstat. The incorpo-
ration of process structure interaction effects leads to an additional state-dependent
term hdyn in the uncut chip thickness (2.51). Without this term the matrix C(t) va-
nishes and (2.53) reduces to an inhomogeneous linear ordinary differential equa-
tion (ODE) with constant coefficients. Simulating the solution of (2.53) with the
Matlab dde23 solver and comparing it to the corresponding solution v(t) of the
system with C(t) = 0 for two different values of ap reveals the additional benefit of
the interaction model.
0.9 60
0.8
50
0.7
0.6 40
|u−v|/|v|
|u−v|/|v|
0.5
30
0.4
0.3 20
0.2
10
0.1
0 0
0 0.02 0.04 0.06 0.08 0.1 0 0.02 0.04 0.06 0.08 0.1
time / [sec] time / [sec]
Fig. 2.9 Stable (a) and unstable (b) solutions of system (2.53)
In the stable case, the solution of (2.53) converges to the solution of the corres-
ponding ODE. Consequently, the relative difference of both solutions converges to
zero. Such an evolution of the relative difference is shown in Fig. 2.9(a). In the
unstable case, the solution of (2.53) does not converge to the solution of the cor-
responding ODE, which remains bounded for all times. Consequently the relative
difference of both solutions diverges, as illustrated by Fig. 2.9(b). In the experi-
ments, the evolution of the DDE-solution is called chatter. While the DDE model
properly reproduces the onset of chatter the ODE model has a bounded solution of
all positive values of ap and is thus not capable of reproducing chatter phenomena.
2.4.2 Simulation
In the previous sections, the different approaches of modeling of the machine and
the process behavior have been presented. The coupling of the models differs
for the different processes. This section gives a short overview of the simulation
techniques, which have been applied within the different projects of the priority
program.
Within the Priority Program 1180, the analysis of the process machine interac-
tions for cutting operations focuses on the coupling of process forces and the
deflection of the tool. As shown in the example of a simple milling model, the
coupling for cutting processes is given by a change of the intersection between
tool and workpiece:
• change of cutting geometry caused by the deflection of the tool/workpiece
• modulation of cutting forces caused by the changing cutting geometry
For the computation of the tool workpiece intersection the following approaches
have been applied:
• intersection between a discrete workpiece (e. g. Dexel model) and the envelope
of the tool (e. g. CSG model)
• intersection between a discrete workpiece and a micro-model of single grains as
well as a macro-model of the grinding wheel for abrasive processes
• analytical computation of the chip thickness, as shown in the example in the
preceding section
In sheet metal forming, the position and deformation of the tool represent the
boundary conditions for the FE simulations of the forming of the blank. The re-
sulting forces at the nodes of the blank model act on the tool causing deflections
and deformations. For the computation of static or dynamic interaction respective-
ly two principle methods have been applied.
• Offline coupling, i. e. iterative computation of process forces and deflections of
the ram
• Integration of the model of the machine in the FE model of the forming process
For the simulation of commercial software tools as well as in-house developments
were used. The coupling was carried out in the same solver or by coupled simulations.
• Exchange of forces and deflections between two simulators with a fixed time
step
• Exchange of forces and deflections between two simulators with a variable time
step
• Exchange of forces and deflections in the same tool (model integration, time
domain simulations)
• Mode-dependent exchange of forces and deflections in the same tool.
• Integration of a force model in the equations of motion. Taking the repetitive
tooth engagement into account leads to a system of delayed differential equations.
The different simulation techniques are illustrated in Fig. 2.10

modeling techniques
machine behavior process behavior

static/dynamic forces
compliance temperatures
machining results
stability
workpiece: geometry,
surface roughness
model coupling
offline coupling
FE-models code export math tools
FE-models time domain
+ model reduction frequency domain
multi-body models code export

multi-body models alternative
+ model reduction co-simulation digital block simulation
time domain
analogous models
in-house development in-house development
Fig. 2.10 Simulation techniques applied in the Priority Program 1180
2.5 Conclusion
An overview of the applied modeling approaches has been provided. It has been
demonstrated which approach is suitable under specific conditions regarding the
machine and process behavior. An overview of the coupling methods and the si-
mulation techniques has also been given. The choice as to which approach has to
be applied depends on the intension of the simulation of the process machine inte-
raction, i. e. the interaction phenomenon that is to be investigated. This chapter
has given only basic information on the principle methods. Detailed information
will be provided in the following chapters, showing the application of these ap-
proaches to the project specific problems.
References
[1] Brecher, C., Esser, M., Witt, S.: Interaction of manufacturing process and machine
tool. CIRP Annals – Manufacturing Technology 58, 588–607 (2009)
[2] Klocke, F., Brecher, C., Sitte, B., Weiß, M.: Analyse der dynamischen Wechselwir-
kungen bei Pendel- und Schnellhubschleifprozessen. In: Jahrbuch Schleifen, Honen,
Läppen und Polieren, vol. 64, Ausgabe, Essen (2010)
[3] Hoffmann, F.: Optimierung der dynamischen Bahngenauigkeit von Werkzeugma-
schinen mit der Mehrkörpersimulation. Dissertation, WZL of RWTH Aachen Univer-
sity (2008)
[4] Oden, J.T., Lin, T.L.: On the general rolling contact problem for finite deformations
of a viscoelastic cylinder. Computer Methods in Applied Mechanics and Engineer-
ing 57, 297–367 (1986)
[5] Tlusty, J.: Analysis of the State of Research in Cutting Dynamics. Annals of the
CIRP 27(2), 583–589 (1978)
[6] Radharamanan, R.: The Measurement of the Dynamic Cutting Coefficients and the
Analysis of Chatter Behaviour in Turning. Katholieke Universiteit Leuven, Disserta-
tion (1977)
[7] Altintas, Y., Weck, M.: Chatter Stability of Metal Cutting and Grinding. In: Annals of
the CIRP, vol. 53(2), pp. 619–642 (2004)
[8] Stephenson, D.A., Agapiou, J.S.: Metal Cutting. Theory and Practice. Marcel Dekker
Inc., New York (1996)
[9] Merchant, M.E.: Mechanics of the metal cutting process. Journal of Applied Phys-
ics 16, 318–325 (1945)
[10] Fang, N., Jaeahir, I.S.: Analytical Predictions and Experimental Validation of Cutting
Forces Ratio, Chip Thickness, and Chip Back-Flow Angle in Restricted Contact Ma-
chining Using the Universal Slip-Line Model. International Journal of Machine Tools
and Manufacture 42, 681–694 (2000)
[11] Clausen, M.: Zerspankraftprognose und -simulation für Dreh- und Fräsprozesse.
Hannover. Diss (2005)
[12] Tlusty, J., Ismail, F.: Basic Non-Linearity in Machining Chatter. Annals of the
CIRP 30(1), 299–304 (1981)
[13] Altintas, Y.: Modeling Approaches and Software for Predicting the Performance of
Milling Operations at MAL-UBC. Machining Science and Technology 4(4), 445–478
(2000)
[14] Weck, M., Brecher, C.: Werkzeugmaschinen. Automatisierung von Maschinen und
Anlagen; 6. Auflage. Springer (2006)
[15] Stépán, G., Insperger, T.: Stability of the Milling Process. Periodica Polytechnica Ser.
Mech. Eng. 44 (2000)
[16] Feng, H.-Y., Azeem, A., Wang, L.: Simplified and Efficient Calibration of a Mecha-
nistic Cutting Force Model for Ball-End Milling. International Journal of Machine
Tools and Manufacture 44, 214–268 (2004)
[17] Witt, S.: Integrierte Simulation von Maschine, Werkstück und spanendem Ferti-
gungsprozess. Dissertation. Aachen. RWTH Aachen, Werkzeugmaschinenlabor
RWTH Aachen (2007)
[18] Marsolek, J., Fleischer, J., Schmidt, C., Schermann, T.: Simulation von Zerspa-
nungsprozessen mit Abaqus. Tagungsband ABAQUS Benutzerkonferenz, Nürnberg
(2005)
[19] Kim, J.S., Lee, B.H.: An Analytical Model of Dynamic Cutting Forces in Chatter Vi-
bration. International Journal of Machine Tools and Manufacture 31(3), 371–381
(1991)
[20] van Brussel, H.: Dynamische Analyse van het Verspaningsproces. Katholieke Un-
iversiteit Leuven, Dissertation (1971)
[21] Werntze, G.: Dynamische Schnittkraftkoeffizienten. Bestimmung mit Hilfe des Digi-
talrechners und Berücksichtigung im mathematischen Modell zur Stabilitätsanalyse.
RWTH Aachen, Dissertation (1973)
[22] Brinksmeier, E., Aurich, J.C., Govekar, E., Heinzel, C., Hoffmeister, H.W., Peters, J.,
Rentsch, R., Stephenson, D.J., Uhlmann, E., Weinert, K., Wittmann, M.: Advances in
Modeling and Simulation of Grinding Processes. Annals of the CIRP 55(2), 667–696
(2006)
[23] Tönshoff, H.K., Peters, J., Inasaki, T., Paul, T.: Modelling and Simulation of Grind-
ing Processes. Annals of the CIRP 41(2), 677–688 (1992)
[24] Aurich, J.C., Biermann, D., Blum, H., Brecher, C., Carstensen, C., Denkena, B.,
Klocke, F., Kröger, M., Steinmann, P., Weinert, K.: Modelling and simulation of
process: machine interaction in grinding. Production Engineering 3(1), 111–120
[25] Mackerle, J.: Finite Element Analysis and Simulation of Machining: an Addendum. A
Bibliography (1996-2002). International Journal of Machine Tools & Manufac-
ture 43, 103–114 (2003)
[26] Hou, Z.B., Komanduri, R.: On the Mechanics of the Grinding Process-Part 1. Sto-
chastic Nature of the Grinding Process. International Journal of Machine Tools &
Manufacture 43, 1579–1593 (2003)
[27] Zitt, U.-R.: Modellierung und Simulation von Hochleistungsschleifprozessen. Uni-
versity of Kaiserslautern, Dissertation (1999)
[28] Chang, H.-C., Junz Wang, J.-J.: A stochastic grinding force model considering ran-
dom grit distribution. International Journal of Machine Tools and Manufacture 48,
1335–1344 (2008)
[29] Doman, D.A., Warkentin, A., Bauer, R.: Finite element modeling approaches in
grinding. International Journal of Machine Tools and Manufacture 49(2), 109–116
[30] Werner, K., Klocke, F., Brinksmeier, E.: Modelling and Simulation of Grinding
Processes. In: 1st European Conference on Grinding. Aachen, pp. S.8-1 – S.8-27
(2003)
[31] Denkena, B., Deichmueller, M., Kröger, M., Popp, K.M., Carstensen, C., Schroeder,
A., Wiedemann, S.: Geometrical Analysis of the Complex Contact Area for Modeling
the local Distribution of Process Forces in Tool Grinding. In: Proceedings of the 1st
International Conference on Process Machine Interaction, pp. 289–298 (2008)
[32] Voelkner, W.: Spannungs- Kraft- und Arbeitermittlung beim Umformen. Ferti-
gungstechnik und Betrieb 25(12), 739–743 (1975)
[33] Voelkner, W.: Ein Beitrag zur Streifen- und Gleitlinienmethode. Wissenschaftliche
Zeitschrift der TU Dresden 25(3), 613–618 (1976)
[34] Lange, K. (ed.): Handbook of Metal Forming. Society of Manufacturing Engineers,
Dearborn, MI (1994)
[35] Petzold, W., König, J., Eberlein, L. (eds.): Umform- und Zerteiltechnik. Lehrbrief
Tiefziehen und Drücken, Dresden (1981)
[36] Neugebauer, R. (ed.): Umform- und Zerteiltechnik. Verlag Wissenschaftliche Skrip-
ten, Chemnitz (2005)
Chapter 3
Adaptive Finite Elements and Mathematical
Optimization Methods
M. Andres, H. Blum, C. Brandt, C. Carstensen, P. Maaß, J. Niebsch,

A. Rademacher, R. Ramlau, A. Schröder, E.-P. Stephan, and S. Wiedemann
Abstract. This chapter focuses on the topics of mathematical research in the pri-
ority program and on the application of the results to engineering problems. First, a
posteriori estimates and adaptive methods based on them are presented for contact
problems. In detail, frictional contact problems using a linear elastic material law,
which are discretized by a dual-dual method, are considered. Using a different ap-
proach, adaptive methods are derived for elasto-plasticity with contact and friction.
Then results for contact problems involving inertial effects are introduced. Second,
parameter identification and inverse problems are considered. After an introduction
into the general problem of setting and solution techniques, the balancing of a rotat-
ing system is discussed as a prototypical industrial application.
3.1 Introduction
Modeling, simulation and optimization are typical steps in computer-aided man-
ufacturing. The mathematical topics addressed in this priority program are fo-
cussed on developing efficient simulation and optimization techniques as well as
on parameter-identification tasks based on inverse problems.
The basic ideas are introduced by using a well-known example. We consider
a heat conduction problem in a perfectly isolated bar: Let us assume that the bar
consists of an inhomogeneous material of conductivity c(x) and specific heat ζ (x),
the bar is heated at its left endpoint and the resulting temperature at the endpoint on
the right is measured. The real temperature distribution inside the bar is denoted by
u , which is a function of time t and a spatial variable x ∈ [0, 1]. The heat distribution
in the bar can be modeled by the one dimensional heat equation, i.e. u satisifes
ζ (x)u̇(t, x) − div c(x) ∇u(t, x) = 0 for t ≥ 0, x ∈ [0, 1],

dx u(t, 0) = q0 (t, 0) for t ≥ 0,
d
(3.1)
u(0, x) = u0 (x) for x ∈ [0, 1].
springerlink.com c Springer-Verlag Berlin Heidelberg 2013
54 M. Andres et al.
The heat inflow is described by the function q0 and the initial temperature by u0 .
First, we address the task of efficiently approximating the solution u for known
c, ζ , q0 and discuss the various error terms involved. In order to approximate u we
compute a discrete solution ukh using a space-time finite element method (FEM).
However, we are not able to determine the exact solution of the FEM on a computer
but get an approximate solution ũkh . For example, we only determine approximate
solutions of the linear systems.
Thus, three types of errors have to be considered: The modeling error u − u,
the discretization error u − ukh , and the numerical error ukh − ũkh . The modeling
error cannot be estimated using mathematical techniques, one has to validate the re-
sults by experiments instead. The numerical error is negligible, if suitable numerical
techniques are used. In the run-up to the computation, a priori error estimates of the
discretization error show the convergence of the discrete solution ukh to u as well as
the convergence rate. However, a posteriori error estimates provide an estimate of
the discretization error for a computed discrete solution. They also yield informa-
tion about the distribution of the error. Based on this information, adaptive FEMs
modify the scheme such that the discretization error is reduced optimally. Hence, the
efficiency of the simulation increases. In the Sections 3.2, 3.3 and 3.4, some results
concerning adaptive FEMs are presented for contact problems . The application of
these methods in the simulation of engineering problems is discussed in Chapters 6,
7 and 14.
Next, we consider the case, where the parameter c, which is allowed to vary spa-
tially, is not known at all or only a rough estimate is available. The mathematical
technique of parameter identification (inverse problems) provides a systematic ap-
proach to determine such model parameters. The most widely used approach takes
an a priori guess c1 , simulates the resulting data u1 (t, 1) and determines an update
by minimizing the deviation between the simulation and the measured data g(t)
min u1 − g2 (3.2)

c
by a suitable, stable updating strategy.

The same approach can be used for optimizing manufacturing processes. Here,
the model and the simulation depend on structure and process parameters, which
can be manipulated in order to obtain optimal manufacturing results. We are also
able to determine their influence on unsatisfactory manufacturing results with the
help of parameter identification. Such tasks give rise to inverse problems requiring
specific regularization techniques, see Section 3.5. Examples of the application of
these techniques to engineering problems are described in Chapters 6, 9, and 15.
3.2 Adaptive Dual-Dual Methods for Frictional Contact

Problems in Linear Elasticity
In order to present an adaptive method for elastic contact problems we consider the
following contact problem in 2D elasticity with Coulomb friction. Assume a linear
3 Adaptive Finite Elements and Mathematical Optimization Methods 55
elastic body occupying an open-bounded domain Ω ⊂ 2 . The smooth boundary

Γ := ∂ Ω is divided into three disjoint parts, the Dirichlet boundary ΓD , where a
homogeneous Dirichlet condition holds, the Neumann boundary ΓN , where a pre-
scribed traction is acting on the body and the contact boundary ΓC where the body
is supposed to come into contact with a rigid foundation. Then, the displacement
vector field u in Ω satisfies the following partial differential equation (PDE):
− div σ (u) = f in Ω ,
u=0 on ΓD ,
σ (u) · n = t0 on ΓN ,
(3.3)
un ≤ g; σn ≤ 0; (un − g)σn = 0 on ΓC ,
if |σ t | < μ f |σn |, then ut = 0 on ΓC ,
if |σ t | = μ f |σn |, then ∃ s ≥ 0 : ut = −s σ t on ΓC ,
where the stress tensor σ (u) is connected to the strain tensor ε (u) via Hooke’s law
for linear elasticity. The volume body force f ∈ L2 (Ω ), the prescribed traction t0 ∈
1 1
H− 2 (ΓN ), the positive gap function g ∈ H 2 (ΓC ) and the friction coefficient μ f are
assumed to be given. To handle the Coulomb friction numerically it is approximated
by a sequence of contact problems using Tresca friction, as described by Nečas et
al. in [25]. For this reason, the friction law in (3.3) is replaced by the following
|σ t | ≤ F ; σ t · ut + F |ut | = 0 on ΓC , (3.4)
with a given friction function F ∈ L2 (ΓC ).

The aim is to derive a formulation, which can be handled numerically and consid-
ers the stress tensor σ as primary unknown. Approximating the stresses with finite
elements saves error-prone post-processing and therefore leads to a better prediction
of the real contact situation. In the following, we will present a dual-dual formula-
tion for problem (3.3). A detailed derivation of this formulation and a corresponding
a posteriori error estimator is carried out in Andres [2].
Starting from the complementary energy minimization problem, see Sokolnikoff
[41], we apply Fenchel’s duality theory as described in Ito and Kunisch [26]. This
results in a saddle-point formulation of dual type involving Lagrange multipliers for
the governing equation, the symmetry of the stress tensor as well as the boundary
conditions on the Neumann and the contact boundaries. For the saddle-point prob-
lem, we can derive an equivalent variational inequality problem using results from
Ekeland and Témam [20]. Both formulations include a non-differentiable functional
arising from the frictional boundary condition. Therefore, an additional dual La-
grange multiplier denoting the friction force is introduced. This procedure yields a
dual-dual formulation of a two-fold saddle-point structure involving the stress ten-
sor σ as primary variable as well as the displacements u in Ω and ϕ , λt and λn on
the Neumann and contact boundaries respectively. Furthermore, the rotation tensor
η and the friction force ν are considered as additional unkowns.
56 M. Andres et al.
We use the PEERS elements, see Arnold et al. [3], to approximate (σ , u, η )

and continuous, piece-wise linear functions to approximate the displacements on
the boundaries as well as piecewise constant functions to approximate the friction
force. The corresponding discrete variational inequality problem reads:
Find ((σh , νĥ ), (uh , ηh , ϕ h̃ , λt,h̃ , λn,h̃ )) such that the following variational inequal-
ities are satisfied for all test functions ((τ , κ ), (v, ξ , ψ , μt , μn ))
ã(σh , τ ) + B((uh , ηh , ϕ h̃ , λt,h̃ ), τ ) + dC,n (λn,h̃ , τ )= g(τ )

B((v, ξ , ψ , μt ), σh ) − q(μt , νĥ ) = f (v, ψ )
(3.5)
dC,n (μn − λn,h̃, σh ) ≤ 0 ∀ μn ≥ 0
q(λth̃ , κ − νĥ ) ≤ 0 ∀ |κ | ≤ 1.
The bilinear forms and linear forms are defined by

ã(σ , τ ) := σ: −1 τ dx, dC,n (μn , τ ) := μn , τn
ΓC ,
Ω

B((v, ξ , ψ , μt ), τ ) := u · div τ + η : τ dx + ψ , τ · n
ΓN + μt , τt
ΓC , (3.6)
Ω
q(μt , κ ) := μt , κ
ΓC , g(τ ) := g, τn
ΓC , f (v, ψ ) := (v, f)0 + ψ , t0
ΓN .
0

(−4, 1) ΓN t0 = 2 4
−800 (1− x2 + x16 )
ΓD Ω ΓD
ΓC g (4, −1)
Fig. 3.1 Geometry, boundary parts and rigid foundation (red).
The discrete system (3.5) can be solved by a nested Uzawa-type algorithm, see
Glowinski et al. [22] or Ekeland and Témam [20] for an explanation of Uzawa
algorithms. The choice of different mesh-sizes h, h̃ and ĥ within the discrete for-
mulation is necessary due to the discrete inf-sup conditions for the bilinear form
B((v, ξ , ψ , μt ), τ ) + dC,n (μn , τ ), on the one hand, and for the bilinear form q(μt , κ ),
on the other hand. They ensure the convergence of the nested Uzawa algorithm.
Within the algorithm, the new iterates for the normal displacement and the friction
force are projected onto the corresponding convex sets in each iteration step.
As a numerical example, we consider the domain Ω := [−4m, 4m] × [−1m, 1m]
with boundary Γ divided into the Dirichlet part ΓD := {−4m} × [−1m, 1m]∪{4m} ×
−3
x 10
2 100
0 0
normal stress [N/m2]

−100
displacement [m]
−2
−200
−4
−300
−6
−400
−8
−500
−10 −600
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3
contact boundary ΓC contact boundary Γ
C
(a) Normal displacement λnh̃ on ΓC . (b) Normal stress σnh on ΓC .

−3
x 10 600
4
3 400
tangential stress [N/m ]
2
2
200
displacement [m]
0 0
−1 −200
−2
−400
−3
−4 −600
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3
contact boundary Γ contact boundary Γ
C C
(c) Tangential displacement λth̃ on ΓC . (d) Tangential stress σth on ΓC .
Fig. 3.2 Approximated solutions on the contact boundary with mesh-size h = 0.0078m cor-
responding to 4 million degrees of freedom.
[−1m, 1m], the Neumann part ΓN := [−4m, 4m] × {1m} and the contact part ΓC :=
[−4m, 4m] × {−1m}.
We choose Young’s modulus E := 200 000 N/m2 and shear modulus G :=
80 000 N/m2 . The friction coefficient is μ f = 0.5 and the volume body force is
set to zero. On the contact boundary, we assume the body Ω to come into contact
with a rigid foundation, which has a positive distance g = 0.01m to Ω . In Figure 3.1
we show the domain and the distribution of the boundary.
In Figure 3.2, we display normal displacement and normal stress and tangential
displacement and tangential stress respectively. Due to the large system (3.5), the
number of degrees of freedom grows fast using the h-version. Since our solution
algorithm uses a nested loop to solve the inequalities in (3.5) the solution time is
very sensitive to the number of unknowns. Therefore, we are interested in adaptive
methods based on error estimators, which refine the finite element mesh in those
regions of the domain, where the estimated error is large.
58 M. Andres et al.
0.5
−0.5
−1
−4 −3 −2 −1 0 1 2 3 4
(a) After 4 adaptive iteration steps
0.5
−0.5
−1
−4 −3 −2 −1 0 1 2 3 4
(b) After 12 adaptive iteration steps
0.5
−0.5
−1
−4 −3 −2 −1 0 1 2 3 4
(c) After 24 adaptive iteration steps
Fig. 3.3 Meshes created during the adaptive refinement process using the maximal local
error.
Using a Helmholtz decomposition of the error of the stress tensor similar to

Carstensen and Dolzmann [14], we can derive a residual error estimator involv-
ing additional indicators for all Lagrange multipliers. We use two different adaptive
methods based on the residual error estimator. The difference lies in the decision
process whether a finite element is refined or not. The first one determines the el-
ement with the largest error estimate ηmax and refines all elements T whose error
estimate ηT is larger than a constant θ ∈ (0, 1] times the maximal error ηmax , see
Figure 3.3(a)-(c). The second one is the fixed-fraction method, see Bangerth and
Rannacher [4]. The estimates for all elements are first sorted by magnitude. Then,
the constant θ denotes the percentage of refined elements, see Figure 3.4(a)-(c).
The methods lead to comparable adaptive refinements. There are three areas of
the domain Ω , where the elements are refined frequently. The first one is near the
Neumann boundary at the top of the domain. The indicator concerning the residuum
of the Neumann boundary condition σh · n − t0 has a big influence there and in
0.5
−0.5
−1
−4 −3 −2 −1 0 1 2 3 4
(a) After 1 adaptive iteration step
1
0.5
−0.5
−1
−4 −3 −2 −1 0 1 2 3 4
(b) After 5 adaptive iteration steps
1
0.5
−0.5
−1
−4 −3 −2 −1 0 1 2 3 4
Fig. 3.4 Meshes created during the adaptive refinement process using the fixed fraction
method.
addition the boundary condition is changing in the corners. Moreover, the residuum
of the friction force (σh )t − νĥ F leads to a refinement of the elements near the
part of the contact boundary where the domain is in contact. Finally, the indicator
referring to the antisymmetric part of the stress tensor (σh − σhT ) leads to a refine-
ment of those elements near the two corners at the bottom of the domain. As in the
first case, the change of boundary conditions in the corners enforces this refinement
additionally.
The application of adaptive methods for the dual-dual formulation of a frictional
contact problem leads to a faster convergence of the algorithm. First, the two-
fold saddle-point structure of the formulation involves several Lagrange multipli-
ers some of them requiring a different mesh-size. For this reason even the coarsest
discretization exhibits a large number of degrees of freedom. Therefore, a uniform
mesh-refinement strategy can be applied ony few times, which may result in an
unsatisfactory accuracy of the solution in the contact zone. Second, the nested
Uzawa algorithm is very sensitive to the number of unknowns. Hence, applying
adaptive methods decreases the convergence time of the algorithm.
60 M. Andres et al.
3.3 Adaptive Mixed Schemes for Elasto-Plasticity with Contact

and Friction
In this section, we will discuss static contact problems in elasto-plasticity and the
approximation of their solutions by adaptive finite elements . We will focus on one
quasi-static time step of elasto-plasticity with linear kinematic hardening and homo-
geneous initial conditions. In elasto-plastic material models, the strain ε (u) is usu-
ally decomposed into the trace-free plastic part P ∈ Q := {q ∈ L2 (Ω )2×2
sym |tr(q) = 0}
and the elastic part e := ε (u) − P. The stress is decomposed asσ = σ b + σ p with
the backstress σ b = Hp and the plastic stress σ p , see [16]. Here, the tensor H
describes the hardening behavior of the elasto-plastic material. Similar to Sec-
tion 3.2, the stress σ is coupled to the elastic part of the strain via Hooke’s law
σ (u, P) = C(ε (u)−P), where C is the tensor of elasticity. To model elasto-plasticity
with contact and friction we extend the system (3.3) by the inclusion
σ (u, P) − HP ∈ ∂ j(P) (3.7)
cf. [13, 23, 24]. Here, j is the dissipation functional with j(Q) := σy (Q : Q)1/2 and
σy > 0 is the yield stress in uni-axial tension. The equivalent variational inequality
is to find W ∈ K := {Z = (v, P) ∈ W := V × Q | vn ≤ g on ΓC } so that
a(W, Z − W) + Ψ (Z) − Ψ (W) ≥ (Z − W) (3.8)
for all Z ∈ K, where we set V := {v ∈ (H 1 (Ω ))2 | v = 0 on ΓD } and
a((u, P), (v, Q)) := (σ (u, Q), ε (v) − Q)0 + (HP, Q)0,

Ψ (v, Q) := j(Q) dx + μ f |vt | ds,
Ω ΓC
(v, Q) := (f, v)0 + (t0 , v)0,ΓN .
The formulation (3.8) is again equivalent to a minimization problem, which enables

us to resolve the functional Ψ by Lagrange multipliers . The stationary conditions of
the resulting saddle-point problem leads to the following mixed formulation : Find
(W, λ p , λn , λt ) ∈ W × Λ p × Λn × Λt such that
a(W, Z) = (Z) − (λ p , χ (Z))0 − λn , γn (Z)

− (λt , γt (Z))0,ΓC ,
(3.9)
(μ p − λ p , χ (W))0 + μn − λn , γn (W) − g
+ (μt − λt , γt (W))0,ΓC ≤ 0
for all (Z, μ p , μn , μt ) ∈ W × Λ p × Λn × Λt , where χ (v, Q) := σy Q, βn (v, Q) := vn ,

βt (v, Q) := vt and
Λ p := {μ ∈ Q | μ : μ ≤ 1},
Λn := {μ ∈ H −1/2(ΓC ) | ∀v ∈ H+ (ΓC ) : μ , v
≥ 0},
1/2
Λt := {μ ∈ L2 (ΓC ) | |μ | ≤ 1},
1/2
with H+ (ΓC ) := {v ∈ H 1/2 (ΓC ) | v ≥ 0} and μ , v
denotes the usual dual pairing
of μ and v. We obtain a mixed discretization scheme by exchanging the space W
with the finite dimensional space Wh := Vh × Qh where Vh is the space of continu-
ous, piece-wise bilinear finite elements and Qh is the space of piecewise constant
finite elements on a quadrilateral mesh Th . Furthermore, we replace Λ p , Λn , and
Λt by sets Λ p,h , Λn,H , and Λt,H , which are constructed on the basis of piecewise
constant finite elements. The solution of the discrete mixed scheme is denoted by
(Wh , λ p,h , λn,H , λt,H ). The indeces h and H indicate mesh-sizes of the finite element
discretizations. To construct the set Λ p,h we apply the mesh, which is already used
to define Wh since the resulting inclusion Λ p,h ⊂ Qh guarantees the unique existence
of the discrete Lagrange multiplier λ p,h of the plastic variable. To ensure the unique
existence of the Lagrange multipliers λn,H and λt,H we have to make sure that a
discrete inf-sup condition is fulfilled. This can be done, for instance, by choosing
a coarser mesh-size H ≥ h defining their finite element discretizations, for details
see [38]. To include multiple yield surfaces we have to add some further plastic-
strain variables and dissipation functionals, which can again be resolved by extra
Lagrange multipliers into order to include them in the proposed mixed scheme. The
structure of the mixed formulation (3.9) remains the same. We refer to [16, 17, 40]
for more details.
In many standard discretization approaches for elasto-plasticity, the non-linearity
j is resolved by regularization, which allows for the application of Newton’s
method, cf. [17, 16]. A mixed discretization scheme, as discussed in this section,
results in a finite dimensional saddle-point problem, which can be solved, for in-
stance, by methods of quadratic programming or by simple Uzawa-type algorithms,
see [40, 23].
The main goal of our investigations is to derive a posteriori error estimates for
the proposed mixed scheme. For this purpose, we combine error estimates for mixed
schemes developed for problems in elasto-plasticity, on the one hand, and for contact
problems with friction, on the other hand. It is shown in [40] that the error of the
purely elasto-plastic problem without contact can be estimated by
W − Wh + λ p − λ p,h 0 η p := η + dev(σ (Wh ) − HPh) − σy λ p,h 0
where 1/2
η := ∑ h2K RK + ∑ h E RE
T ∈T E∈E 0 ∪E
is the usual residual type estimator with the residuals

[σn (Wh )]0,E , E ∈ Eh◦
RT := f + div σ (Wh )0,K , RE :=
g − σn(Wh )0,E , E ∈ EhN .
To include contact with friction we have to add some additional terms accounting for
the geometrical and complimentary conditions as well as the frictional conditions,
cf. [36]. We then obtain
62 M. Andres et al.
W − Wh + λ p − λ p,h 0 + λn − λn,H −1/2,ΓC + λt − λt,H 0,ΓC

η p + (g − γn(Wh ))− 1/2,ΓC + |(λn,H , (g − γn (Wh ))− )0,ΓC |
+ |Ψ (Wh , 0) − (λt,H , γt (Wh )|
where ( f )− := − max{0, − f (x)} denotes the negative part of a function f . The term
dev(σ (Wh )−HPh )− σy λ p,h 0 results from the use of quadrilaterals. It vanishes, if
dev(Wh ) ⊂ Qh holds. This is the case, if piece-wise linear finite elements on triangles
are applied. In this case, the estimates for purely elasto-plastic problems without
contact are similar to those for standard schemes, cf. [9, 15, 33].
In a numerical experiment, we apply the a posteriori error estimates within an
adaptive scheme, where we use the fixed-fraction strategy, as described in Section
3.2. We set Ω := [0m, 1m]2 and the material parameters E := 2, 500N/m2, ν := 0.25,
ϑ := 500N/m2, σy := 10N/m2 and μ f := 0.5. Furthermore, we define H := ϑ I
(a) After 3 adaptive iteration steps (b) After 5 adaptive iteration steps
Fig. 3.5 Meshes created during the adaptive refinement process.

(a) Plastic strain p (b) Plastic stress λ p
Fig. 3.6 (a) the norm of the plastic strain, (b) norm of the Lagrange multiplier
where I, is the identity tensor. The outer forces are set to zero and the obstacle
function is given as φ (x) := (x − 1/2)2 + 0.98. Figure 3.5 shows meshes created by
successive adaptive refinements on an initial mesh with 256 elements. We observe
refinements towards the contact area on the upper part of the boundary and the lower
corners, where Dirichlet conditions change to Neumann conditions. In Figure 3.6,
the resulting plastic strain and the Lagrange multiplier for it are shown.
3.4 Adaptive Finite Element Methods for Dynamic Contact

Problems
In Sections 3.2 and 3.3, adaptive finite element methods for static contact problems
are presented. Here, we focus on dynamic contact problems, where inertial effects
are considered. When neglecting frictional effects, the strong problem formulation
is an extension of the static one given in (3.3) and reads
ρ ü − divσ (u) = f in I × Ω ,
u(0) = u0 ; u̇(0) = v0 in Ω ,
u = 0 on ΓD , (3.10)
σ (u) · n = 0 on ΓN ,
un ≤ g; σn ≤ 0; (un − g)σn = 0 on ΓC ,
with the initial displacement u0 and the initial velocity v0 . The weak problem for-
mulation is
(ρ ü, v − u) + (σ (u), ε (v − u)) ≥ (f, v − u) (3.11)
where the variational inequality has to hold for all v in the test space V and all t ∈ I.
Two discretization approaches are commonly used: The first one combines a fi-
nite difference method in time with a finite element method in space, see [18, 27].
64 M. Andres et al.
An alternative is given by space-time finite element methods [6, 29]. The temporal
discretization has to be carried out carefully because it is difficult to ensure energy
conservation and numerical stability.
For both approaches, we derive a posteriori error estimates and start our descrip-
tion with the finite difference/ finite element approach. There, Rothes method is
used, i. e. the temporal direction is discretized first. We obtain a sequence of spa-
tial continuous but temporally discrete problems after the time discretization. The
structure of these problems is similar to static contact problems. Consequently, the
same techniques are used for the spatial discretization and for a posteriori error es-
timation. We modify the estimators given in [37] and apply them to the dynamic
problems. They lead to an estimate, which can be evaluated easily and fast. How-
ever the error is only estimated in the energy norm and no temporal effects are taken
into account. The detailed derivation is found in [5, 7, 8]. Currently, we are extend-
ing this approach by goal-oriented error estimators, which are originally designed
for static contact problems (see [39]).
The second approach relies on a space-time Galerkin discretization of the dy-
namic contact problems. We estimate the error in a user-defined error-functional J,
for instance the von Mises equivalent stress in a certain region. By utilizing a so-
called dual problem, which represents the influence of a space-time point on the
error J(u) − J(uh ), the a posteriori error estimate is derived. It consists of the primal
and the dual residual weighted by the approximation error of the dual and the primal
problem, respectively, as well as a term representing the error in the contact stresses
weighted by the dual solution. We obtain information about the spatial and the tem-
poral distribution of the error. However, a discrete solution of the dual problem has
to be calculated, i. e. a linear wave is additionally solved. More details are presented
in [29].
The a posteriori error estimates provide information about the spatial and, in the
dual-weighted residual (DWR) approach, temporal distribution of the discretization
error. To make use of this characteristic we have to localize the estimates, i. e. a
value is assigned to each single mesh cell. Two different techniques can be applied:
Integration by parts, see, e. g., [1], and filtering [10]. By considering hyperbolic
problems, some technical additional steps are carried out to obtain appropriate re-
finement indicators, see [29]. For instance, a special treatment of the hanging de-
grees of freedom is necessary.
Once the spatial and temporal error distribution is known, we utilize adaptive
refinement strategies, which select the mesh cells for refinement. A well-known
strategy is the fixed fraction strategy (see Section 3.2). Apart from this one, we
employ the more complex optimal mesh strategy [34], which optimizes the error
over the effort.
Employing a dynamic mesh approach, i. e. the meshes change between the time
steps, special attention has to be paid to the additional errors due to the mesh
changes. The data transfer has to be carried out in the appropriate way prescribed
by the space-time bilinear form. Typically, we have to calculate L2 -projections. Fur-
thermore, the meshes of two time steps have to be connected closely. To ensure this,
we employ a mesh regularization strategy [29].
Fig. 3.7 Schematic view of the bar and the rigid foundation
Finally, we present a numerical example from [29]. A 2D Signorini problem is

considered. The basic domain Ω = [0 m, 0.05 m] × [0 m, 0.2 m], a bar, is depicted in
Figure 3.7, the time interval is I = [0 s, 1 s]. We set the body force f = (0.5, 0) N/m2
and assume homogeneous initial conditions. Furthermore, a linear elastic material
law with modulus of elasticity E = 10 N/m2 and Poisson ratio ν = 0.33 is used. We
choose the density ρ = 10 kg/m2 . The distance between the elastic body and the
rigid foundation is g = 0.005 m. We are interested in the temporal mean value of the
von Mises equivalent stress σv (u) in a sub-domain B := [0 m, 0.05 m]2 of Ω , i.e.
T
J(u) = σv (u(x)) dx dt, σv := σ11 (u)2 + σ22(u)2 + 3σ12(u)2 . (3.12)

0 B
0.0055
eta
0.005
0.0045
0.004
k
0.0035
0.003
0.0025
0.002
0.0015
0.001
0 0.2 0.4 t 0.6 0.8 1
Fig. 3.8 Temporal mesh for η
In the next paragraphs, we will compare the results of the special contact estimator
η and the DWR estimator for the unconstrained problem ηe as well as different
adaptive methods. We use a fixed spatial mesh with 4, 096 cells and perform five
adaptive refinement iterations to compare the temporal meshes. The temporal mesh
based on η is depicted in Figure 3.8. It uses larger time steps for small t and smaller
time steps for large t.
For the discussion of the spatial meshes, we choose a uniform decomposition of
the time interval I with M = 400 time steps. The spatial mesh created in the constant
mesh approach based on η is presented in Figure 3.9. The left end of the contact
zone and the left corners of Ω are well-resolved. In the left corners, we observe
stress singularities due to the geometry and the change of the boundary conditions.
66 M. Andres et al.
Fig. 3.9 Adaptive mesh resulting from the constant mesh approach
(a) m = 1 (b) m = 100
(c) m = 200 (d) m = 300
(e) m = 400
Fig. 3.10 Meshes created during the adaptive refinement process in the dynamic mesh
approach
7000
eta
6000
5000
Nm
4000
3000
2000
1000
0
0 0.1 0.2 0.3 0.4 0.5
t 0.6 0.7 0.8 0.9 1
Fig. 3.11 Number of mesh elements in the single time steps
In Figure 3.10, the adaptive meshes created in the dynamic mesh approach are
depicted. For m < 200, we mainly observe additional refinements in the left corners,
where the stress singularities occur. In the 200th time-step, the left boundary of the
contact zone and B are well-resolved. The additional refinements are more and more
concentrated in B for m > 200. In Figure 3.11, we can see that the number of mesh
cells increases for t > 0.4. It attains its maximum for t ≈ 0.7 and then decreases.
In Figure 3.12, the convergence behavior for different refinement techniques is
compared. More detailed information about the dynamic mesh approach based on
η is presented in Table 3.1, where
0.1
CM eta
CM eta_e
DM eta
DM eta_e
Uniform
J(w)−J(wkh )
0.01
J(w)
0.001
1e-04
10000 100000 1e+06 1e+07 1e+08 1e+09
M
∑ Nm
m=1
Fig. 3.12 Convergence in the output functional J with respect to the total number of mesh
cells for different refinement techniques
|J(u) − J(uh )| J(u) − J(uh ) J(u) − J(uh )

Erel := , Ieff = , e
Ieff :== .
|J(u)| η ηe
The results in Table 3.1 substantiate the accuracy of the a posteriori error estimate.
From Figure 3.12, we deduce that the adaptive methods based on ηe as well as the
dynamic mesh approach based on η are very efficient. They lead to comparable re-
sults. However, the constant mesh approach based on η is less effective. The results
nearly correspond to the results based on uniform refinement.
Table 3.1 Quantitative results of the error estimator
M
M ∑ NM Erel Ieff e
Ieff
n=1
50 12800 3.725 · 10−2 2.692 2.708
96 42912 1.700 · 10−2 3.119 3.132
182 176552 6.339 · 10−3 2.294 2.300
352 725908 3.351 · 10−3 2.532 2.536
678 4177824 5.664 · 10−4 1.049 1.050
3.5 Parameter Identification and Inverse Problems

3.5.1 Basic Considerations
We will now turn to the second objective of this chapter, namely mathematical meth-
ods for the parametrization and optimization of simulation models. The starting
point for every simulation is an abstract model of the technical process under inves-
tigation. Such models typically depend on certain system and process parameters,
e.g. material constants in force law, stiffness and damping parameters for machine
68 M. Andres et al.
structures, feed rates and depth of cut. In real-life situations some of these parame-
ters are not known precisely or are acceptable only in a small range of manufacturing
conditions.
This general framework of parameter identification can be used for different tasks
such as
• model calibration (parametrization),
• determining optimal structure and process parameters for achieving a desired
result (optimal control),
• determining the causes, if a desired result is not achieved with a given parameter
setting (process monitoring and correction).
In addition, the mathematical formulation of the solution process typically adds
some artificial parameters, such as step sizes or regularization parameters for sta-
bilizing the numerical solvers. Hence, one needs to calibrate the system by deter-
mining optimal values for those parameters. This becomes a challenging task, if
those parameters cannot be measured directly. Hence, they have to be estimated
from indirect measurements, e. g. the determination of damping parameters from
vibration measurements of a mechanical structure. This is the typical task of in-
verse problems: determining an unknown parameter from indirect measurements.
The forward mapping, which relates the parameters to the measurements, is typ-
ically called parameter-to-state mapping, which needs to be inverted to solve the
inverse problems.
There is a large variety of mathematical methods for the determination of such
parameters, if they are scalar. However, the development of mathematically justi-
fied numerical schemes for the identification of distributed parameters (functions,
such as spatially varying material constants or time-dependent process parameters)
is the subject of ongoing research. The technical difficulty in determining these pa-
rameters increases even more, if the parameter-to-state mapping is ill-posed, which
means that different and diverging parameter settings give - within the available
measurement precision - rise to the same data. In mathematical terms, the inverse
problem needs a regularization (stabilization).
In order to describe one of the most successful regularization schemes (the so-
called Tikhonov-regularization) let us introduce a formal notation of the inverse
problem. We denote the searched parameters by p, where we assume that p = p(t, x)
is a function
depending on time and/or location. We assume that p has finite energy,
i. e. p2 (s) ds < ∞ .
Let us assume, that we cannot determine p, e. g. the unbalances of a rotating
system, directly. Hence, we have to rely on indirect measurements such as vibra-
tion measurements, g, at some sensor locations. The relation between p and g is
determined by a model or operator A, which describes the manufacturing process
as well as the measurement device. This model allows us to simulate artificial mea-
surements for different parameter settings p1 . Again, in general, g = g(t, x) is a
distributed parameter, i. e. a function of time and/or location. We formally write
A(p) = g. However, measurement devices typically have limited precision and we
assume that only δ

imprecise and noisy measurement data g are available, which
δ
fulfill g − g ≤ δ . Here, δ denotes the measurement precision.
In its abstract form, we face the task of determining a set of parameters p such
that
A(p) ∼ gδ . (3.13)
Let us emphasize that for ill-posed problems it does not make sense to request
A(p) = gδ , which would lead to highly oscillating solutions with infinite energy.
Instead, we stabilize the inversion by determining p as the minimizing solution of
2

min A(p) − gδ + αΦ (p) , (3.14)
where Φ (p) is called the regularization, term which controls the energy of p and at
the same time models some a priori information on the optimal parameter setting.
The classical setting
Φ (p) = p(s)2 ds (3.15)
leads to parameters of minimal energy. This type of least squares approach has been
investigated since the early 1960’s. However, only recently, so-called sparsity reg-
ularization schemes have been introduced: there, one is not interested in restrict-
ing the energy of the parameter p but one aims at determining parameters, which
are“well-localized”. For an illustration of this concept, let us consider a cutting pro-
cess. One might want to determine the optimal, time-varying feed rate of the cutting
tool in order to obtain a pre-defined surface g. In addition, the model A allows us to
simulate the resulting surface g for given feed rate f . Based on experience, a feed
rate f0 is suggested, which, however, leads to a sub-optimal result. The searched for
parameter, in this case, is the change of the feed rate, e. g. fopt = f + p and we want
to determine p such that A( f + p) ∼ g. Enforcing the side condition, one prefers for
the modification of the feed rate f over a minimal time span, leads to a modified
Tikhonov-functional. Solutions of this type are obtained with penalty terms of the
form
Φ (p) = |p(s)| ds . (3.16)
This minor change in the formulation of the minimization procedure has a tremen-
dous impact on the structure of the minimizer as well as on the complexity of the nu-
merical schemes needed for approximating the minimizers: the classical L2 -penalty
term (3.15) allows to use Hilbert space techniques leading to least square approx-
imations, which have been well-developed for decades. In contrast, sparsity pro-
moting L1 -penalty terms (3.16) require optimization techniques, which were first
analyzed in 2004 by [19].
3.5.2 A Prototypical Inverse Problem in Ultra Precision

Machining
The problem of computing optimal balancing weights for rotating systems is of par-
ticular interest in the high precision cutting of surfaces in optical quality. In this
70 M. Andres et al.
preliminary section, we have developed a mathematical procedure for computing

balancing weights for the rotating spindle of an ultraprecise cutting machine, see
Figure 3.13. This leads to a linear inverse problem. In Section 15.4 we have in-
cluded a model of the machine-process interaction. This is based on modeling the
cutting forces, which are determined from the parameters of the cutting process as
well as from the rotational unbalances of an imperfect spindle. On the other hand,
the cutting forces excite forces on the spindle, which in turn can be interpreted as
a non-stationary unbalance of the spindle. The determination of optimal process
parameters as well as the determination of optimal balancing procedures leads to
non-linear inverse problems. The remainder of this section is a summary of our re-
sults presented in [12] and [28].
workpiece holder main spindle

spindle casing
tool holder
granite plate
z-axis
granite board
x-axis
Fig. 3.13 Ultraprecision turning lathe [11]
3.5.2.1 Mathematical Model for Determining Balancing Weights

In order to apply the regularization techniques described in the previous section, we
need to determine the operator A and the data g related to the problem of computing
balancing weights.
The starting point for our approach is the basic vibration equation of the structure
mechanics
Mü(t) + Du̇(t) + Su(t) = p(t), (3.17)
in order to determine the vibrations u of the structure. To derive the system matrices
we use the Finite Element discretization of the machine structure. Figure 3.14 shows
the division of the machine into the main parts: spindle rotor, spindle casing, rotor
part of the motor and motor casing. This discretization method will be explained in
detail in Section 15.4.
The matrix M denotes the mass matrix, S the stiffness matrix and D the damp-
ing matrix of the system. The vibrations u of each node of the discretization are
collected in the vector u and the corresponding loads in vector p. Each node has 6
degrees of freedom: the displacements u, v and w in each space direction, the tor-
sion angle βz and the cross section slopes βx and βy . Since no information about the
damping properties is available the damping is neglected in a first attempt, i. e. we
consider the equation
Mü(t) + Su(t) = p(t), (3.18)
y
motor coupling spindle
casing
x
Ctorsion casing
Engine
En
rotor z
Caxial
Clateral C
spindle
C
engine
Cengine Ccasing
granite base
Fig. 3.14 Modeled parts of the turning lathe [12]
3.5.2.2 Solution of the Vibration Equation in Presence of Unbalances

If we only consider unbalances in our cutting machine and ignore unbalances in-
duced by the process forces, the right hand side p = punb (t) of Equation (3.18) has
harmonic entries depending on the angular velocity ω . In practice, the revolution
speed n in rpm is given, therefore we have ω = 260π n. An unbalance is modeled as
a mass Δ m displaced from the shaft by a vector r = rei·φ . Here, φ is the angle to a
given zero mark. If the displaced mass rotates with angular velocity ω , it induces a
harmonic centrifugal force (or load)
punb (t) = ω 2 p0 eiω t , with p0 := Δ mrei·φ .
We assume that the induced vibration is also harmonic with the same frequency ω ,
i. e., u(t) = u0 eiω t , with a complex and time independent amplitude u0 . By inserting
this into (3.18) we get a (time-independent) expression for the vibration amplitude
of the vibration u0 ,
u0 = (−ω 2 M + S)−1 p0 . (3.19)
Therefore, u is determined via the solution u0 of (3.19).

72 M. Andres et al.
3.5.2.3 Balancing
We will now discuss the inverse problem of balancing a rotating system. As men-
tioned in the last section, the mounting of a workpiece onto the spindle often results
in an unbalanced system. This unbalance can be projected onto the two balancing
planes. Reconstructing this projected unbalance distribution allows us to determine
the necessary balancing weights, which have to be mounted in the balancer rings.
The reconstruction is carried out with the help of the above given model. To this
end, the workpiece has to be added to the model, i. e., we have to include an element
in the rotating part of the spindle. Balancing then means computing the balancing
weights as well as their positions from measurements of the vibration.
In order to determine the balancing weights from vibration measurements we
have formulated the problem as an operator equation
A(p) = u. (3.20)
Here, p denotes the unbalance distribution we are looking for and u the vibrations
induced by this unbalance distribution measured at positions, where sensors are
mounted. The measurements for u are obtained from an idle run of the machine,
hence we have no forces from the cutting process and punb and u are harmonic and
related via their absolute values or amplitudes p0 and u0 by (3.19):
u0 = Ap0 , A = (−ω 2 M + S)−1 .
As explained in Subsection 3.5.1, the operator A is usually not continuously invert-

ible, which means that for given noisy data uδ with a data error of δ ≥ u0 − uδ 2
the function pδ0 = A−1 (uδ0 ) might be an arbitrary bad approximation of the true
unbalance distribution p. The computation of pδ0 can be stabilized by minimizing
(3.14), i. e. by solving,
2
δ δ
p0,α = min Ap0 − u0 + α Φ (p0 ) (3.21)
p0 2
instead. The penalty term Φ (p) acts as a stabilizer and prevents large values of
Φ (p). Typical choices of Φ are, e. g.
1/p
Φ (p0 ) = p0 p := ∑ |pi0 | p , 0 < p ≤ 2. (3.22)
i
Problems, which are not continuously invertible, are called ill-posed and require
regularization [21]. The functional (3.21) is well-known as Tikhonov-functional and
forms a regularization method [31, 32]. The parameter α is called the regularization
parameter. It has to be chosen according to the amount of noise in the data and
assures the convergence of pδ0,α to p0 as δ → 0. A popular rule for the choice of α
is Morozov’s discrepancy principle, [35, 30].
In order to find the underlying unbalance distribution projected onto the bal-
ancers, we have minimized the functional (3.21) with Φ (p) = p22 . For a numeri-
cal simulation, we place a specified unbalance at the node of the workpiece and use
our model to compute the related vibrations of the platform. We then consider the
inverse task of determining the unbalance (mass, location, phase) from the vibration
data.
In order to simulate a real-life situation it was assumed that the data can only
be obtained at two sensor positions. To simulate measurement errors we disturbed
the data at the sensor positions with noise (20 − 30%). The sensors are fixed in
y-direction on the spindle casing above the two balancers. We then reconstructed
the unbalance projected onto the balancing planes by minimizing (3.21), i. e., we
reconstructed an unbalance distribution, which produces approximately the same
vibrations at the sensors as the original unbalance created by the workpiece but is
located at the two balancer positions. As an example for an unbalance created by
a workpiece, we set 50 gmm at an angle of 30◦ to the zero mark at the workpiece
node in the model. The corresponding amplitudes of the vibrations for a frequency
range of [5, · · · , 50] Hz ’measured’ at the first sensor are plotted in Figure 3.15 (line
without markers). The vibrations were disturbed with a noise level of 26 %, i. e.
(u − uδ )/u = 0.26. In the balancing planes, we reconstructed unbalances of 26
gmm in the first and 26.5 gmm in the second plane, both with a position of 30◦ .
The balancing masses at the balancers can be placed on a radius r = 41.5 mm, thus
the according masses are determined as 0.63 g in the first plane and 0.64 g in the
second, both having to be placed under an angle of 210◦, see Figure 3.16. As shown
in Figure 3.15 (line with star markers), the setting of those balancing masses reduces
the vibration amplitude significantly.
Fig. 3.15 Vibration amplitude for a workpiece unbalance of 50 gmm with 30◦ angle before
and after balancing [12]
74 M. Andres et al.
Fig. 3.16 Balancing masses and angles for a workpiece unbalance of 50 gmm with 30◦ ,
balancing radius 41.5 mm. [12]
3.6 Conclusions and Outlook

The potential of adaptive methods in reducing the problem complexity has been
proven over the last years. We considered the application of adaptive methods for
the dual-dual formulation of a frictional contact problem, which leads to a faster
convergence of the algorithm. First, the two-fold saddle-point structure of the for-
mulation involves several Lagrange multipliers, some of them requiring a different
meshsize. For this reason, even the coarsest discretization exhibits a large number of
degrees of freedom. Therefore, a uniform mesh refinement strategy can only be ap-
plied few times which may result in an unsatisfactory accuracy of the solution in the
contact zone. Second, the nested Uzawa-algorithm is very sensitive to the number
of unknowns. Hence, applying adaptive methods decreases the convergence time of
the algorithm.
Furthermore, we stated an a posteriori estimate for the discretization of the mixed
formulation of elasto-plasticity with linear kinematic hardening and contact. The es-
timate holds for lower order finite element discretizations based on triangles, quadri-
laterals and hexahedrons. Moreover, we have illustrated the application of adaptive
algorithms to frictional contact problems in elasto-plasticity.
Adaptive methods lead to effective discretizations applied to dynamic contact
problems; and the underlying a posteriori error estimates based on the DWR method
estimate the discretization error accurately. However, the implementation of such
methods is very complex and should be improved, e. g. by using parallelized algo-
rithms in the evaluation of the a posteriori estimates. A further reduction of comput-
ing time and memory requirement will enhance the applicability of these techniques
to real-world engineering problems. Furthermore, the a posteriori error estimates
and adaptive methods are extended to frictional and even more complex thermal
coupled contact problems.
The second part of this chapter was devoted to parameter identification and in-
verse problems. The balancing of a rotating system is a prototypical example for
the use of inverse problem theory in industrial applications. The expensive multiple
step process of finding the best balancing weights and their positions can be reduced
to one computation using only one set of vibration measurements. We are paying
for this reduction of costs with the necessity of creating a mathematical model of
the rotating machinery, which relates unbalance distributions in the system to the
resulting vibrations. On the other hand, once such a model is created it can be used
repeatedly, e. g., in ultra precision machining, for every new machining process with
an unbalance-situation originating from the mounting of a new workpiece.
References
1. Ainsworth, M., Oden, J.T.: A Posteriori Error Estimation in Finite Element Analysis.
John Wiley & Sons, New York (2000)
2. Andres, M.: Dual-dual formulations formulations for frictional contact problems in me-
chanics. Ph.D. thesis, Institute of Applied Mathematics, Leibniz Universität Hannover
(2011)
3. Arnold, D.N., Brezzi, F., Douglas Jr., J.: PEERS: a new mixed finite element for plane
elasticity. Japan J. Appl. Math. 1(2), 347–367 (1984),
http://dx.doi.org/10.1007/BF03167064, doi:10.1007/BF03167064
4. Bangerth, W., Rannacher, R.: Adaptive Finite Element Methods for Differential Equa-
tions. Birkhäuser, Basel (2003)
5. Biermann, D., Blum, H., Jansen, T., Rademacher, A., Scheidler, A.V., Schröder, A.,
Weinert, K.: Space adaptive finite element methods for dynamic Signorini problems in
the simulation of the NC-shape grinding process. In: Proceedings of the 1st CIRP Inter-
national Conference on Process Machine Interactions (PMI), pp. 309–316 (2008)
6. Blum, H., Jansen, T., Rademacher, A., Weinert, K.: Finite elements in space and time for
dynamic contact problems. Int. J. Numer. Meth. Engng. 76, 1632–1644 (2008)
7. Blum, H., Rademacher, A., Schröder, A.: Space adaptive finite element methods for dy-
namic obstacle problems. ETNA, Electron. Trans. Numer. Anal. 32, 162–172 (2008)
8. Blum, H., Rademacher, A., Schröder, A.: Space adaptive finite element methods for dy-
namic signorini problems. Comput. Mech. 44(4), 481–491 (2009)
9. Bostan, V., Han, W., Reddy, B.D.: A posteriori error estimation and adaptive solution
of elliptic variational inequalities of the second kind. Appl. Numer. Math. 52(1), 13–38
(2005), doi:10.1016/j.apnum.2004.06.012
10. Braack, M., Ern, A.: A posteriori control of modeling errors and discretisation errors.
Multiscale Model. Simul. 1(2), 221–238 (2003)
11. Brandt, C., Krause, A., Brinksmeier, E., Maaß, P.: Force modelling in diamond machin-
ing with regard to the surface generation process. In: Proceedings of the 9th International
Conference and Exhibition on Laser Metrology, Machine Tool, CMM and Robotic Per-
formance, LAMDAMAP 2009, London, June 30-July 2, pp. 377–386 (2009)
12. Brandt, C., Niebsch, J., Ramlau, R., Maass, P.: Modeling the influence of unbalances for
ultra-precision cutting processes. Zeitschrift für Angewandte Mathematik und Mechanik
(2011) (in press)
13. Carstensen, C.: Numerical analysis of the primal problem of elastoplasticity with hard-
ening. Numer. Math. 82(4), 577–597 (1999), doi:10.1007/s002110050431
76 M. Andres et al.
14. Carstensen, C., Dolzmann, G.: A posteriori error estimates for mixed FEM in elasticity.
Numer. Math. 81(2), 187–209 (1998),
http://dx.doi.org/10.1007/s002110050389,
doi:10.1007/s002110050389
15. Carstensen, C., Klose, R., Orlando, A.: Reliable and efficient equilibrated a posteriori
error finite element error control in elastoplasticity and elastoviscoplasticity with hard-
ening. Comput. Methods Appl. Mech. Engrg. 195(19-22), 2574–2598 (2006)
16. Carstensen, C., Valdman, J., Brokate, M.: A quasi-static boundary value problem in
multi-surface elastoplasticity: part 1 - analysis. Math. Methods Appl. Sci. 27, 1697–1710
(2004)
17. Carstensen, C., Valdman, J., Brokate, M.: A quasi-static boundary value problem in
multi-surface elastoplasticity: part 2 - numerical solution. Math. Methods Appl. Sci. 28,
881–901 (2005)
18. Czekanski, A., El-Abbasi, N., Meguid, S.A., Refaat, M.H.: On the elastodynamic solu-
tion of frictional contact problems using variational inequalities. Int. J. Numer. Meth.
Engng. 50, 611–627 (2001)
19. Daubechies, I., Defrise, M., De Mol, C.: An iterative thresholding algorithm for linear
inverse problems with a sparsity constraint. Communications in Pure and Applied Math-
ematics 57(11), 1413–1457 (2004)
20. Ekeland, I., Témam, R.: Convex analysis and variational problems. Classics in Ap-
plied Mathematics, vol. 28, english edn. Society for Industrial and Applied Mathematics
(SIAM), Philadelphia (1999); Translated from the French
21. Engl, H.W., Hanke, M., Neubauer, A.: Regularization of Inverse Problems. Mathematics
and its Applications, vol. 375. Kluwer Academic Publishers Group, Dordrecht (2000)
22. Glowinski, R., Lions, J.-L., Trémolières, R.: Numerical analysis of variational inequali-
ties. Studies in Mathematics and its Applications, vol. 8. North-Holland Publishing Co.,
Amsterdam (1981); Translated from the French
23. Han, W.: Finite element analysis of a holonomic elastic-plastic problem. Numer.
Math. 60(4), 493–508 (1992), doi:10.1007/BF01385733
24. Han, W., Reddy, D.: Plasticity. Springer (1999)
25. Hlaváček, I., Haslinger, J., Nečas, J., Lovı́šek, J.: Solution of variational inequalities in
mechanics. Applied Mathematical Sciences, vol. 66. Springer, New York (1988); Trans-
lated from the Slovak by J. Jarnı́k
26. Ito, K., Kunisch, K.: Lagrange multiplier approach to variational problems and appli-
cations. Advances in Design and Control, vol. 15. Society for Industrial and Applied
Mathematics (SIAM), Philadelphia (2008)
27. Laursen, T.A.: Computational Contact and Impact Mechanics. Springer, Heidelberg
(2002)
28. Niebsch, J., Ramlau, R., Brandt, C.: On the interaction of unbalances and surface quality
in ultra-precision cutting machinery. In: SIRM 2011, Darmstadt, Germany (2011)
29. Rademacher, A.: Adaptive finite element methods for nonlinear hyperbolic problems of
second order. Ph.D. thesis, Technische Universität Dortmund, Published in Verlag Dr.
Hut, München (2009)
30. Ramlau, R.: Morozov’s discrepancy principle for tikhonov regularization of nonlinear
operators. Numer. Funct. Anal. and Optimization 23, 147–172 (2002)
31. Ramlau, R.: A steepest descent algorithm for the global minimization of the tikhonov-
functional. Inverse Problems 18(2), 381–405 (2002),
http://stacks.iop.org/0266-5611/19/i=2/a=312
32. Ramlau, R.: Tigra - an iterative algorithm for regularizing nonlinear ill-posed problems.
Inverse Problems 19(2), 433–467 (2003),
http://stacks.iop.org/0266-5611/19/i=2/a=312
33. Rannacher, R., Suttmeier, F.-T.: A posteriori error estimation and mesh adaptation for
finite element models in elasto-plasticity. Comput. Methods Appl. Mech. Eng. 176(1-4),
333–361 (1999)
34. Richter, T.: Parral multigrid method for adaptive finite elements with application to 3d
flow problems. Ph.D. thesis, Ruprecht-Karls-Universität Heidelberg (2005)
35. Scherzer, O.: The use of Morozov’s discrepancy priciple for Tikhonov regularization for
solving nonlinear ill-posed problems. Computing 51, 45–60 (1993)
36. Schröder, A.: A posteriori Error Estimation in Mixed Finite Element Methods for Sig-
norini’s Problem. In: Numerical Mathematics and Advanced Applications. Proceedings
of ENUMATH 2009, pp. 801–808 (2010)
37. Schröder, A.: Fehlerkontrollierte adaptive h- und hp-Finite-Elemente-Methoden für Kon-
taktprobleme mit Anwendungen in der Fertigungstechnik. Ph.D. thesis, Universität Dort-
mund (2005)
38. Schröder, A.: Mixed Finite Element Methods of Higher-Order for Model Contact Prob-
lems (2009); Humboldt Universitt zu Berlin, Institute of Mathematics, Preprint 09-16
39. Schröder, A., Rademacher, A.: Goal-oriented error control in adaptive mixed FEM for
Signorinis problem. Computer Methods in Applied Mechanics and Engineering 200(1-
4), 345–355 (2011)
40. Schröder, A., Wiedemann, S.: Erro restimates in elastoplasticity using a mixed method.
Applied Numerical Mathematics 61, 1031–1045 (2011)
41. Sokolnikoff, I.S.: Mathematical theory of elasticity, 2nd edn. McGraw-Hill Book Com-
pany, Inc., New York (1956)
Part II
Grinding
Chapter 4
High-Performance Surface Grinding
J.C. Aurich, A. Bouabid, P. Steinmann, and B. Kirsch
Abstract. This chapter presents experimental as well as modelling and simulation

approaches to investigate a high-performance surface grinding process. The com-
plex material removal mechanisms generate transient cutting forces that cover a
wide range of excitation frequencies. The generated cutting forces impact the
grinding machine and lead to deformations, which depend on the machine’s me-
chanical properties. In general, these deformations have an influence on the cut-
ting forces. Deformations lead to a change of the depth of cut and, therefore, the
cutting forces change. Thus, there are three aspects of great importance: the
process, the machine and the process machine interaction. Advances in investigat-
ing the process are covered first. Afterwards, a new approach to model the ma-
chine, its deformation behavior and the way the machine interacts with the process
will be described.
4.1 Introduction
In contrast to machining processes with defined cutting edges, grinding is affected
by a large number of cutting edges. These cutting edges are single grains, whose
number, geometry and relative position to the workpiece are not known. Although
the depth of cut in high-performance surface grinding is in the range of millime-
ters, the depth of cut of the single grains is in the sub-micrometer range. For this
reason, process machine interactions (PMI) in the range of micrometers have great
influence on the single grains and hence on the complete process [1; 2; 3].
To approximate these PMI, a process model was coupled with a machine model.
The process model, a kinematic-geometrical simulation of surface grinding
(KSIM), is able to deliver undeformed chip parameters of each single grain in the
contact zone and a high-frequency signal of the grinding force for each point of the
time discretization. Based on the simulated grinding force, the machine model, a fi-
nite element model for spinning motions (SFEM), simulates PMI of the modeled
machine structure in terms of the resulting displacements of the grinding wheel.
Experimental and numerical studies will be presented. The experimental inves-
tigations have been carried out to gain an enhanced understanding of the influence
of PMI on the process and the process results.
The numerical studies include the determination of undeformed chip parame-
ters and a detailed description of the machine model. After presenting different
coupling strategies of KSIM and SFEM, the result of a combined simulation will
be described.
82 J.C. Aurich et al.
4.2 Experimental Investigation of Process Machine Interactions

in Grinding
Experimental setup
The surface-up grinding experiments were executed on a high performance grinding
machine with high stiffness. Two series of tests were carried out. In the first series, a
resin bonded CBN (cubic boron nitride) grinding wheel (B181) with a diameter of
Ø 400 mm and a width of 12 mm was used. The second series of tests was aiming at
higher loads of the machine system to achieve higher displacements, corresponding
to stronger PMI. To achieve these higher loads, i. e. higher grinding forces, specific
material removal rates higher than those of the first series were necessary. Therefore,
another grinding wheel had to be used. This grinding wheel was an electro-plated
CBN grinding wheel (B251) with a diameter of Ø 400 mm and a width of 24 mm.
The workpiece material used for both series of tests was a heat-treated steel AISI
4140H (42CrMo4V, DIN EN 10132-3, 55 ± 2 HRC) with dimensions of
125 mm x 50 mm x 10 mm.
Common measurement techniques were applied using a dynamometer for
measuring normal and tangential forces and non-contact hall effect probe to meas-
ure the effective spindle power. In addition, the displacements of the grinding
wheel relative to the headstock were measured (to examine the process machine
interactions). The application of optical measurement systems was not possible
due to the high amount of cooling lubricant, which was used during the experi-
ments. Also, tactile sensors could not be used because of the high rotational
speeds. For this reason, the displacements were measured using eddy-current sen-
sors. They provide high-resolution signals and can be used in wet environments.
In the first series of tests using the resin bonded grinding wheel, the measuring
surface used for the sensors was a high-precision shoulder on the grinding wheel.
In the second series of tests using the electroplated grinding wheel, the flange of
the grinding wheel was used as measuring surface.
First series of tests – resin-bonded grinding wheel

The parameters used in the first series of tests are listed in Tab. 4.1. The focus of
the experiments was the examination of PMI based on the measured displace-
ments of the grinding wheel. Therefore, the influence of grinding wheel speed,
feed rate and depth of cut was evaluated.
The average values of the normal grinding forces during stationary grinding
ranged from Fn = 300 – 1,100 N, the tangential grinding forces ranged from
Ft = 50 - 300 N. The displacements in normal and tangential direction reached 7
and 1 µm respectively. However, the displacements in tangential direction could
not be reliably measured because of the small absolute values.
4 High-Performance Surface Grinding 83
Table 4.1 Grinding parameters for first series of tests
grinding wheel diameter 400 mm; resin-bonded; B181 V240 (DIN ISO 6106)
grinding wheel speed: 60 - 100 m/s
depth of cut: 0.25 - 1 mm
grinding parameters
feed rate: 600 – 4,800 mm/min
maximum specific material removal rate: 20 mm³/mm s
lubricant cooling 300 l/min at 12 bar; wheel cleaning 100 l/min at 25 bar
The relative dynamic stiffness during the stationary grinding process can be
calculated based on the measured normal grinding forces and displacements . The
calculated relative dynamic stiffness varies from 120 - 200 N/mm (Fig. 4.1).
Variations of the wheel speed and the depth of cut do not influence the relative
dynamic stiffness. The main influence on the relative dynamic stiffness in normal
direction was exerted by the feed rate. Different feed rates result in different ma-
chine loading times. Feed rates of 1,200 mm/min or less reveal a relative dynamic
stiffness of 120 - 160 N/µm. Feed rates of 2,400 mm/min or higher (grinding pass
< 3 s) reveal an almost constant relative dynamic stiffness of approximately
200 N/µm. It seems that the inertia of the traveling column of the grinding ma-
chine of more than 10 tons influences the measured displacements due to the short
time periods of loading.
Fig. 4.1 Relative dynamic stiffness from the first series of tests
Second series of tests – electro-plated grinding wheel

As mentioned above, the electro-plated bonding of the grinding wheel used for the
second series of tests allows much higher loads than the resin bonding. The maxi-
mum specific material removal rate investigated was 87.5 mm³/mms (Tab. 4.2).
Table 4.2 Grinding parameters for second series of tests
grinding wheel diameter 400 mm; electro-plated; B251; 0.2 carat/cm²

grinding wheel speed: 63 - 100 m/s
depth of cut: 0.25 - 2 mm
grinding parameters
feed rate: 1,000 – 6,000 mm/min
maximum specific material removal rate: 87.5 mm³/mms
lubricant cooling 420 l/min at 12 bar; wheel cleaning 110 l/min at 25 bar
Compared to the investigations with the resin bonded grinding wheel the spe-
cific material removal rate was significantly increased. However, the measured
normal and tangential grinding forces were smaller than the forces reached with
the resin-bonding wheel (Fn = 140 - 780 N and Ft = 30 - 260 N). The resulting
maximum measured displacements in normal direction were smaller (4 µm), too.
Similar to the first series of tests, the relative dynamic stiffness during the sta-
tionary grinding process was calculated based on the measured normal grinding
forces and displacements. The relative dynamic stiffness ranged between
110 N/µm and 210 N/µm (see Fig. 4.2). The same correlations for the relative dy-
namic stiffness as for the resin-bonded grinding wheel were observed. Variations
of wheel speed and depth of cut showed minor influence whereas the feed rate had
a strong influence.
Fig. 4.2 Relative dynamic stiffness from the second series of tests
In this second series of tests, the quality of the ground workpieces was eva-
luated. The surfaces were classified into three categories:
• no grinding marks
• grinding marks
• grinding marks and burn
According to these categories, the whole parameter range can be mapped into ac-
ceptable process results (no grinding marks) and inacceptable process results
(grinding marks or grinding marks and burn). The classification of the ground sur-
face was carried out visually since measurement with tactile sensors was not suit-
able. Fig. 4.3 shows the results of the mapped parameters.
Fig. 4.3 Quality of the ground surface
Similar to the relative dynamic stiffness, the feed rate is the dominating factor
for the occurrence of grinding marks and burn. This can be seen when comparing
different feed rates at constant material removal rates (e. g. 50 mm³/mms at
63 m/s): At a feed rate of 2,000 mm/min there are no grinding marks or burn.
However, at 3,000 mm/min grinding marks occurred. At 4,000 mm/min (and
6000 mm/min) grinding marks as well as burn were detected. The occurrence of
grinding marks can be attributed to the higher feeding path per wheel revolution at
higher feed rates and the resulting higher chip thicknesses. The observed burn was
caused by less cooling lubricant volumes per time unit when applying higher feed
rates. Less grinding marks and burn occurred at higher wheel speeds. This results
from the better cooling lubricant supply at higher wheel speeds and to the smaller
feeding path per wheel revolution.
Despite the correlation of the influence of feed rate on the relative dynamic
stiffness and on the workpiece quality no direct correlation between these two pa-
rameters could be found. Higher relative dynamic stiffness does not necessarily
result in higher or lower workpiece quality. When taking, for example, the feed
rate 5,000 mm/min at grinding wheel speed 100 m/s the relative dynamic stiffness
is approximately constant for all applied depths of cut. In contrast, the workpiece
quality ranges from no grinding marks to grinding marks and burn.
4.3 Modeling and Simulation
4.3.1 Process Model

In the following, the process model used is presented. Therefore, the concept of
the model is described and some simulation results are shown.
The process model used is based on a kinematic-geometrical simulation

(KSIM) of surface and cylindrical grinding developed by Warnecke and Zitt [4, 5]
and enhanced in further works [6, 7, 8]. The concept of the KSIM is based on the
observed micro and macroscopic cause-and-effect chain in grinding. KSIM was
developed and programmed in Fortran language as an offline tool for analyzing,
dimensioning and optimizing grinding processes. In contrast to empirical process
models, which correlate process inputs and process outputs mathematically, the
KSIM simulates the grinding process as a penetration between the enveloping
profile of the grinding wheel and the workpiece. Therefore, the micro and macro-
geometry of the grinding wheel and the workpiece as well as the process kinemat-
ics have to be modeled.
To gain realistic simulation results the grinding wheel topography has to be
modeled precisely. For this reason (in addition to the reproduction of the macro
geometry), there is a special focus on the micro geometry of the grinding wheel.
To model the topography in KSIM a “synthetic” generation of the grinding wheel
was chosen instead of digitalizing the topography. This has the advantage that the
modeling of the topography is not limited to the quality of the measuring system
and to the available grinding wheels. Instead, the topography is generated based
on microscopically-identified characteristic values.
The modeling of the micro-geometry in KSIM accounts for the stochastically-
affected nature of grinding wheels. The geometrically-undefined shape as well as
the unspecified number and distribution of the single abrasive grains are consi-
dered in KSIM. In modern high-performance, grinding tools are used, which con-
sist of a metal body and diamond or cubic boron nitride (CBN) abrasive grains for
the most part. Those tools use metallic, ceramic or resin bonding. Most modeling
concepts of diamond or CBN grains use abstract descriptions via simple geome-
trical bodies [1]. With this method, the morphology of real grains is only roughly
approximated. The modeling of the grains in KSIM is based on complex basic
geometries (ellipsoid, tetrahedron, cuboid and octahedron) according to the possi-
ble crystal morphologies of CBN and diamond. These basic geometry models are
modified under statistic variations to gain more realistic grain models. Different
grain sizes can be regarded by the statistic distribution of the long and short grain
axis lAI and lAK. This detailed geometrical mapping of the grinding wheel topology
in KSIM also enabled wear and dressing to be taken into account.
The KSIM simulates the grinding process by overlapping the modeled grinding
wheel and a geometrically-modeled workpiece. The overlapping between grinding
wheel and workpiece is given by the process kinematic, which can be calculated
from the motion of the machine axis by a coordinate transformation. This coordi-
nate transformation enables even complex process kinematics to be simulated
based on the process type and machine setup.
The output parameters of the simulation are the experimentally not measurable
undeformed chip thickness (hcu), chip length (lcu), chip width (bcu) and chip cross
section (Acu) of every single grain. In the simulation, ideal cutting is assumed,
which means that ploughing of material does not occur and each grain removes
the whole material volume encountered. The superposition of every single grain
contact enables the examination of the resulting macro and micro-geometry of the
workpiece. Based on the simulated chip parameters a specific grinding force can
be calculated depending on the process type and the workpiece material. The chip
cross section and the specific grinding force are used to calculate the forces of
each single grain based on a simplified approach of Kienzle. The Kienzle ap-
proach can be used to calculate cutting forces, e. g. for turning or milling
processes. The simplified approach is based on a specific cutting force kc,sim (an
experimentally-determined parameter), the undeformed chip thickness and the
width of the undeformed chip. The total process force can then be calculated by
cumulating the forces from each kinematic grain (i. e. each grain that removes ma-
terial) (Fig. 4.4).
Fig. 4.4 Flow chart of the kinematic simulation
For the simulations to be carried out the modeled grinding wheel with the di-
mension of Ø 400 x 3 mm² was used. By using a smaller grinding wheel width
than the wheel used in the experiments the calculation time of the kinematic simu-
lation could be reduced (without reducing the simulation quality [9]).
Based on the applied manufacturing process (electroplating) a grain protrusion
height of 35 % of the nominal grain size of 251 µm was used to model the grind-
ing wheel. This means that about 163 µm of the grain height were embedded in
the bonding, resulting in an average grain protrusion height of 88 µm. Because of
the statistic nature of the modeling process the grain size distribution leads to sig-
nificant deviations of the average grain protrusion height. This results in grains
with high protrusion heights that remove the whole material, resulting in chip
thicknesses of 10 to 15 µm. Those grains would wear very fast under real grinding
conditions. This rapid wear would lead to a stationary grinding process characte-
rized by a high number of kinematic grains distributing the material removal
process over many grains with less chip thickness. Since in KSIM only one wheel
revolution is simulated with no wear a numerical dressing process was imple-
mented in KSIM to model grinding wheels in a stationary wear state [8]. Within
this numerical dressing process, the grain protrusion height of every single grain is
ideally cut down to a specified limit (dressing height, see Fig. 4.5) to fit real grind-
ing conditions, for the simulations in this study 90 µm.
Fig. 4.5 Numerical dressing process
The simulated process parameters were chosen corresponding to those of the

experimental investigations, which have been carried out. These parameters are
listed in Tab. 4.3.
Table 4.3 Model specifications and process parameter for kinematic simulation
modeled grinding diameter 400 mm; width 3 mm; B251; 1,200 grains/cm²
wheel
process parameter grinding wheel speed: 63 - 100 m/s
for kinematic simu- depth of cut: 0.5 - 1 mm
lation
feed rate: 1,000 – 5,000 mm/min
maximum specific material removal rate: 83.33 mm³/mms
In each kinematic simulation, two grinding wheel revolutions are simulated.

The first grinding wheel revolution is simulated with an ideally even workpiece.
After this revolution, the contact zone is prepared, i. e. a ground surface is gener-
ated. The second grinding wheel revolution is then operated in this ground surface.
This is comparable to a real grinding process where the grains always penetrate a
ground surface in the contact zone. The following simulation results have been ob-
tained from the second revolution (Fig. 4.6):
• mean value of the average chip thickness of all kinematic grains (hcu,med)
and
• number of momentarily active grains within the contact zone (N’mom).
Generally, an increase of either the feed rate or the depth of cut results in higher
specific material removal rates. This means that a higher material volume has to
be ground in the same time, resulting in a higher chip thickness and a higher num-
ber of momentarily active grains. The results show a progressive increase for the
chip thickness and a degressive increase of the number of active grains with
increasing feed rate. The increase of active grains is limited due to the constant
contact length at a constant depth of cut, which limits the maximum number of
available grains for the material removal process. For this reason, the chip thick-
ness shows a progressive increase because it has to compensate the degressive in-
crease of active grains with increasing material removal rates.
Fig. 4.6 Simulation results
A comparison of the average chip thickness and the number of active grains for
different grinding wheel speeds reveals higher values for lower grinding wheel
speeds. At higher grinding wheel speeds, less material has to be removed within
each grinding wheel revolution. Thus, the chip thickness and the number of active
grains decrease.
4.3.2 Machine Model
The modeling of the machine has been carried out by means of the finite element
method on the base of the so-called Arbitrary Lagrangian Eulerian (ALE) ap-
proach. The system to be modelled consists of a grinding wheel, which rotates
with a constant angular velocity (Fig. 4.7). The notations used are briefly ex-
plained in the following.
Grinding Wheel ω = const.

R*
u* Ro Ri
Workpiece
δ
δ0
u
F0(t)
Fig. 4.7 Model of the grinding wheel. For the machine model there are two important radi-
uses: the inner radius Ri which the wheel is fixed at, and the expanded radius R* that serves
as reference to determine the wheel’s deformation
The grinding wheel has an inner radius Ri, which is assumed to remain constant
throughout the simulation since the grinding wheel is fixed to the spindle. Due to
the centrifugal forces caused by the angular velocity ω the outer radius Ro expands
to the radius R* by the radial expansion u*.
Starting-up processes enabling the grinding wheel to accelerate from the idle
state up to the velocity ω do not have to be considered. It will be assumed that at
the time t = 0 the grinding wheel already rotates with the velocity ω and thus has
the outer radius R*. The depth of cut, which the grinding wheel is initially plung-
ing into the workpiece with, is denoted by δ0. The cutting force resulting from δ0
is denoted by F0. Due to this force, which is assumed to only radially act at one
point of the wheel’s circumference, the grinding wheel deforms in radial direction
by the amount u, defined relatively to the expanded radius R*. In tangential direc-
tion, the force application point is retained by defining an additional bearing. The
deformation u is taken to be positive, if the outer radius R* increases. The depth of
cut, which results when the machine’s deformation is taken into account, is simply
given by δ = δ0 + u. According to this nomenclature one can distinguish initial
values (δ0 and F0(δ0)), which are set up at the beginning of the process, and effec-
tive values (δ and F(δ)), which result during the cutting process, for both the depth
of cut and the cutting force.
The machine model can now be used either independently or in conjunction
with a process model to run both uncoupled and coupled simulations respectively.
Simulation results are presented in a later section.
4.3.3 Coupling Strategies

The basic concept of the coupling strategies is an exchange of the simulation re-
sults between the machine (SFEM) and process (KSIM) model. That means that
the output data of one model is used as input data for the other model and vice
versa. The coupling algorithm is based on general modeling strategies for coupled
problems, i. e. Neumann and Dirichlet boundary conditions are applied. In the fol-
lowing, four coupling strategies are presented. Strategies 2 and 4 are closely re-
lated to each other.
Coupling Strategy 1 - data exchange and iteration until convergence after
multiple time steps (each wheel revolution)
The process model (KSIM), as presented in chapter 4.3.1, provides simulation re-
sults after every grinding wheel revolution. For this reason, data exchange of
process and machine model is carried out after each grinding wheel revolution.
The resolution of the simulated grinding force signal depends on the grinding
wheel speed and the wheel diameter. Also, one grinding wheel revolution consists
of a number of time steps (here 1,000), comparable to the number of data points
for one revolution in real measurements. When modeling, for example, the grind-
ing wheel presented in 4.3.1 (Ø 400 mm, grinding wheel speed 100 m/s) the simu-
lated grinding force signal for one wheel revolution has a resolution of 80 kHz.
The coupled simulation starts with the generation of a force signal by the
process model. For this force signal an ideal grinding machine is assumed, i. e. no
machine deformations are considered. This force signal is used as input data for
the machine model. The machine model generates displacements of the grinding
wheel for one wheel revolution based on the input force signal. After resetting the
workpiece topography (to simulate the same grinding wheel revolution as before)
the process model is restarted based on the generated displacements by the process
model. In doing so, the penetration of the grinding wheel and the workpiece is
modified, which results in a different grinding force signal. This altered force sig-
nal again serves as input for the machine model and so forth. This iteration is re-
peated until “convergence” occurs, i. e. the simulated force and / or displacement
signals (average values) of two consecutive iteration steps show minor deviations
(Fig. 4.8).
Fig. 4.8 Coupling strategy 1 - data exchange after each wheel revolution
Coupling Strategy 2 - data exchange and iteration until convergence for each
time step
The principle of coupling strategy 2 is similar to that of coupling strategy 1. The
difference is that the data exchange between the process and machine model
(grinding force and displacement) is conducted after each time step within one
grinding wheel revolution instead of the complete revolution (Fig. 4.9).
Fig. 4.9 Coupling strategy 2 - data exchange and convergence for each time step
In other words, coupling strategy 1 is applied for each single time step until con-
vergence is reached. When convergence of force and / or displacement is achieved
for the current time step the iteration cycle will be started again for the next time
step. When using 1,000 time steps for one wheel revolution coupling strategy 2 con-
sists of 1,000 iteration cycles, each with numerous iteration steps. This results in
very large simulation times so that a simplification of each iteration cycle represents
an interesting alternative solution, as will be discussed in coupling strategy 4.
The main advantage of coupling strategy 2 is the possibility to simulate only a
part of one grinding wheel revolution. That means computing time can be saved, if
less than one wheel revolution is needed.
Coupling Strategy 3 - asynchronous data exchange for each time step without
iteration
In coupling strategy 3, no iteration cycles are used. As in coupling strategy 2, the
force signal at time step tn is used as an input for the machine model. The calcu-
lated displacement then serves as input for the process model, however, at time
step tn+1 (Fig. 4.10).
Fig. 4.10 Coupling strategy 3 - asynchronous data exchange for each time step
Coupling Strategy 4 - data exchange and iteration until convergence for each
time step based on an empirical cutting force model
As already mentioned, this coupling strategy corresponds to coupling strategy 2
and represents coupling in a strong form. That means that a converged coupled so-
lution is achieved at every time step. The coupling iterations are, however, carried
out based on the assumption that the process force depends linearly on the cutting
depth, i. e.
F0 (δ 0 ) = c ⋅ δ 0 (4.1)
where c denotes the “process stiffness”. This assumption has been undertaken
based on the linear relation between the chip cross section and the specific grind-
ing force, to be found in the Kienzle approach, for example, and furthermore
enables a considerable acceleration of the coupled simulation.
Eq. (4.1) is only valid for an ideally stiff grinding wheel, which the cutting
depth is constant for throughout the whole process. For a real deformable grinding
wheel, however, the grinding force is not a function of the cutting depth set up but
of the actual (or effective) one. It will be assumed that the initial depth of cut δ0
leads to a wheel deformation u, which simply has to be added to δ0. By multiply-
ing the sum [δ0 + u] by the process stiffness c an adjusted force value is then
obtained and can be subsequently used to calculate the deformation again. This
coupling procedure can be written as:
(δ0  F0  u0)  (δ1  F1  u1)  … (δconv.  Fconv.  uconv.)

1. coupling convergence iteration
uncoupled solution
iteration (coupled solution)
In summary, the models of the process, the machine and the process machine
interactions presented above constitute a complete simulation procedure, which
can now be used to run coupled simulations.
4.3.4 Coupling Results

The simulation results are summarized in Figs. 4.11 to 4.13. Of high interest for
the following discussion are differences between the coupled and the non-coupled
simulation.
A typical force signal generated by KSIM for one wheel revolution is given in
Fig. 4.11a. A fast Fourier transform of the force signal shows a dominating contri-
bution of frequencies of up to 5 kHz (Fig. 4.11b). Beyond this frequency, the
grinding grits excite the machine at almost all frequencies.
The KSIM and SFEM models were coupled using coupling strategy 1. In
SFEM, only static terms were activated just in order to check the convergence be-
havior of the coupled process machine system. Fig. 4.11c shows how the mean
values over one grinding wheel revolution for the force and the displacement con-
verge after about 6 iterations to 910 N and 6.2 µm respectively. These values are
smaller than the values gained after the first iteration (1,050 N and 7.2 µm), i. e.
smaller than the values corresponding to the uncoupled solution.
Coupling strategy 1 is well suited for the case of quasi-static loading. In simu-
lating a high-performance grinding process it has been observed that the high
loading and deformation rates lead to an abrupt change of the state of single grits
from contact to non-contact. This effect led to excessive fluctuations in the simu-
lated force and deformation values causing the coupled simulation to diverge (de-
spite the reasonable average values). For this reason, coupling strategy 1 is not
suitable for coupled simulations involving vigorous dynamics.
In order to investigate dynamic properties of the coupled process machine sys-
tem a coupling strategy, which guarantees a strong coupling has to be used, i. e.
either strategy 2 or 4. The first one comprises more modeling details, the last is
more economic in terms of simulation times. In addition, the fast Fourier trans-
form of the force signal simulated in KSIM suggests carrying out the coupled
simulation within a wide range of loading frequencies. In the following, the force
signal provided by KSIM is substituted by a set of harmonic depths of cut with
different frequencies ranging from 0 to 20 kHz. This set results in a set of harmon-
ic loads, which are now able to excite the grinding wheel at different frequencies.
The process machine interaction is then simulated using coupling strategy 4. The
corresponding simulation model is based on the parameters listed in Tab. 4.4. All
of them represent parameters commonly encountered in real grinding processes.
For the process stiffness c a high value is used in order to emphasize the effect of
coupling.
Table 4.4 Parameters for the coupled simulation, grouped into machine, process and coupl-
ing parameters
Machine model:
Inner radius Ri : 63.5 mm
Outer radius Ro : 200 mm
Elastic modulus E : 210 GPa
Poisson ratio ν : 0.3
Material density ρ : 7.85 g/ccm
Boundary conditions : – No displacement at the inner boundary.
– No tangential displacement at the point the
load is acting on.
Initial (setup) load F0(t) : Results readily form the product [c ⋅ δ0(t)].
Process model:
Angular velocity ω : 200 rad/s, i. e. roughly about 2,000 rpm.
Depth of cut δ0(t) : Harmonic function with a fixed mean value of 5
µm, a fixed amplitude of 1 µm and variable fre-
quencies ranging from 0 to 20 kHz.
Coupling model:
Process stiffness c : 200 N/µm
The time-dependent force and deformation signals are exploited in the follow-
ing by means of their frequency and amplitude. Since the amplitude of a time de-
pendent signal is generally not constant with respect to time the amplitude mean
value over the whole signal is used.
The frequency spectrum of the grinding wheel can be obtained by varying the
excitation frequencies and simulating the resulting mean deformation amplitude
for each frequency. Also, the amplitude spectrum can be normalized by the de-
formation amplitude, which results when the grinding wheel is loaded by a quasi-
static force. A normalized representation enables different simulation models to be
compared to each other.
Fig. 4.12a shows how coupling shifts the eigenfrequencies gained from an un-
coupled simulation to higher values. Note that the frequency spectrum obtained from
a coupled simulation can no longer be considered as a property of the grinding
wheel but of the whole system consisting of grinding wheel, process and interaction.
To answer the question as to why the eigenfrequencies increase when coupling
is included one considers the applied force and the resulting deformation in time
domain, Fig. 4.12b,c. All forces shown are normalized by dividing them by the
same value, namely by the amplitude of the initial load F0(t), so that the initial and
the effective force can be readily compared to each other with regard to both am-
plitude and frequency. However, all deformations are normalized by dividing
them by their maximum value within the considered time interval. This has the
advantage that, in normalized form, forces and deformations can be represented in
the same diagram. A comparison of the frequencies is also possible. However, the
deformation amplitudes resulting from different simulations cannot be readily
compared to each other since deformations are not normalized by the same entity.
Also, only some special loading frequencies will be presented here:
• the loading frequency of f0 = 12.8 kHz, which corresponds to an eigen-
frequency of the uncoupled process machine system and
• the loading frequency of f0 = 13.3 kHz, which corresponds to an eigen-
frequency of the coupled process machine system.
f0 = 12.8 kHz
Although this loading frequency corresponds to an eigenfrequency of the grinding
wheel as given in Fig. 4.12a there is no significant increase of the deformation
amplitude throughout the coupled simulation. Fig. 4.12d shows that through cou-
pling the effective force signal acquires smaller amplitudes than the initial one be-
cause a deformed grinding wheel generates a smaller depth of cut than an unde-
formed one. In the first excitation cycle, the maximum amplitude of the force
effectively acting at the grinding wheel is only about 18 % of the initial force. An-
other more important aspect is the fact that through coupling the force signal ac-
quires a higher frequency. This is the reason why the shifting effect observed in
Fig. 4.12a occurs in direction of higher and not smaller frequencies. Fig. 4.12b
shows that the force, which is effectively applied, and the resulting deformation
oscillate with different frequencies. Under this circumstance, the deformation am-
plitude remained almost constant over the whole loading history. So, for reso-
nance, effective force and deformation must first synchronize. This happens at the
frequency of 13.3 kHz.
a
1.6
1.4
1.2
Normal Force in kN 1.0
0.8
0.6
0.4
0.2
0
0 2.5 5.0 7.5 10.0 12.5

Time in ms
b
0.025
Amplitude in kN
0.02
0.015
0.01
0.005
0
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 38
Frequency in kHz
c Mean normal force Mean grinding wheel deformation
1.2 0
1.0 -2
Deformation in µm
0.8 -4
Force in kN
0.6 -6
0.4 -8
0.2 -10
0 -12
1 2 3 4 5 6
Iteration
Fig. 4.11 Results of the coupled simulation based on coupling strategy 1. (a) Force signal
simulated by KSIM. (b) FFT-transform of the force signal. (c) Convergence behavior of the
coupled simulation.
a Non-coupled Coupled
log | A / A(static,non-coupled) |
Mean Deformation Amplitude
100
10
Loading Frequency in kHz

0,1
0 2 4 6 8 10 12 14 16 18 20 22
b Radial Deformation c
12.8 kHz 13.3 kHz
Effective Force
Force /Displacement
1.0 4
Normalized
0.5 2
0 0
-0.5 -2
-1.0 -4
10 Time in ms 10.25 10 Time in ms 10.25
d e
Loading Frequency: 12.8 kHz Non-coupled
Initial Force Effective Force Coupled
1.0
Normalized Force
0.5
Force
-0.5
-1.0
0 Time in ms Time
01
Fig. 4.12 Results of the coupled simulation based on coupling strategy 4. (a) Frequency
spectrum. (b)-(c) Effective force and displacement in time domain for special loading fre-
quencies. (d) Influence of coupling on the loading frequency. (e) Influence of coupling on
the variation of the effective force over time.
f0 = 13.3 kHz
As shown in Fig. 4.12c, effective force and resulting deformation oscillate for this
loading frequency (after some loading cycles) with the same frequency. Only
through their synchronisation can the coupled process machine system get into
resonance.
Another important difference between the coupled and uncoupled system is that
force and deformation are inversely influenced by each other. In the case of cou-
pling, the deformation patterns resulting from a harmonic load, a beat for example,
is mirrored into the force pattern and can thus also be retrieved there (Fig. 4.12e).
This is not the case for an uncoupled simulation.
In Fig. 4.13, the eigenmodes of the grinding wheel are presented. They are just
briefly described in the following. At the first eigenfrequency, the grinding wheel
is oscillating in vertical direction and thus behaves approximately like a rigid
body. The second eigenform represents an alternating dilation and contraction in
horizontal and vertical directions. In the third eigenmode, the grinding wheel takes
a triangular form. In the forth one, the outer wheel boundary expands and con-
tracts uniformly. Finally, polygonal wheel forms result at higher eigenfrequencies.
@ 5.2 kHz @ 8.4 kHz @ 11.2 kHz
@ 12.8 kHz @ 15.2 kHz
Fig. 4.13 Eigenmodes of the spinning grinding wheel (ω = 200 rad/s). The contour plots
represent the field of the radial deformation. Positive values mean an increase in the ex-
tended radius of the outer wheel R* due to deformation.
4.4 Conclusion
Extensive experimental research was conducted to examine PMI during grinding.
In doing so, displacements of the grinding wheel were measured and the resulting
relative dynamic stiffness was mapped for a large parameter field. High stiffness
was identified for high feed rates for different grinding wheels. The examination
of the workpiece quality revealed grinding marks and burn when applying high
feed rates. The experiments conducted enhance the understanding of the cause-
and-effect chain in grinding and the influence of PMI on the process and the proc-
ess result.
Also, a complete simulation procedure enabling the modelling of the process,
the machine and their interaction was discussed in detail. There are differences be-
tween results gained from uncoupled and coupled simulations, which can be
summarized as follows:
In the uncoupled process machine system:
• Loading frequencies, which correspond to eigenfrequencies of the ma-
chine, lead to a critical (non-stable) process.
• The grinding force results in a machine deformation that, however, has no
inverse influence.
• The knowledge of the frequency spectrum of the machine provides in-
formation about critical loading frequencies, which are not allowed, if the
process has to be kept stable.
In contrast, simulations of the coupled process machine system revealed that:
• Coupling is more complex than just scaling the amplitude and the fre-
quency of the initial load.
• A loading frequency, which corresponds to an eigenfrequency of the ma-
chine, does not necessarily lead to a critical process.
• Vice versa, a loading frequency, which lies outside the eigenfrequencies,
can be critical.
The last three aspects suggest that the knowledge of the natural frequencies of the
machine does not generally suffice to judge, whether a harmonic load is critical or
not. This again underlines the importance of simulation techniques in general and
of coupled simulations in particular.
Concerning simulation techniques one can say that there are quite efficient
simulation techniques to simulate the deformation behavior of the grinding wheel
due to transient process forces. This is primarily enabled by the use of the ALE-
approach instead of the conventional Lagrangian one so that the kinematically less
relevant rigid body motion is a priori omitted. This also enables the application of
the force at a fixed point without contact-searching steps.
There is still need to bring simulation and experimental techniques closer to-
gether. Providing confident information about the real workpiece surfaces is still
not reliably possible. Also, real-time simulations of high-performance grinding
processes are currently not available.
The physical phenomena taking place during PMI are quite complex and not
always accessible for measuring setups so that efficient simulation techniques are
of great importance to investigate the grinding process. Although the simulation
techniques presented in this chapter do not completely fit experimental results
they can still be considered as an important step in understanding the phenomena
involved in the PMI.
References
[1] Brinksmeier, E., Aurich, J.C., Govekar, E., Heinzel, C., Hofmeister, H.-W., Klocke, F.,
Peters, J., Rentsch, R., Stephenson, D.J., Uhlmann, E., Weinert, K., Wittmann, M.:
Advances in Modelling and Simulation of Grinding Processes. CIRP Annals - Manu-
facturing Technology 55(2), 667–696 (2006)
Klocke, F., Kröger, M., Steinmann, P., Weinert, K.: Modelling and Simulation of
Process - Machine Interaction in Grinding. Production Engineering 3(1), 111–120
(2008)
[3] Herzenstiel, P., Bouabid, A., Steinmann, P., Aurich, J.C: Experimental Investigation
and Computational Simulation of Process-Machine Interactions during High-
Performance Surface Grinding. In: Proceedings of 1st International Conference on
Process-Machine Interaction, PZH Hannover, pp. 267–278 (2008)
[4] Warnecke, G., Zitt, U.: Kinematic Simulation for Analyzing and Predicting High-
Performance Grinding Processes. CIRP Annals – Manufacturing Technology 47(1),
265–270 (1998)
[5] Zitt, U.: Modellierung und Simulation von Hochleistungsschleifprozessen. Disserta-
tion, TU Kaiserslautern (1999)
[6] Braun, O.: Konzept zur Gestaltung und Anwendung definiert gesetzter Hochleis-
tungsschleifscheiben. Dissertation, TU Kaiserslautern (2008)
[7] Aurich, J.C., Herzenstiel, P., Sudermann, H., Magg, T.: High-performance dry grind-
ing using a grinding wheel with a defined grain pattern. CIRP Annals – Manufacturing
Technology 57(1), 357–362 (2008)
[8] Herzenstiel, P., Aurich, J.C.: Numerical and Experimental Investigations of a Grinding
Wheel with a Defined Grain Pattern. In: CIRP Conference on Modeling of Machining
Operations, vol. 12, pp. 567–574 (2009)
[9] Aurich, J.C., Herzenstiel, P., Kirsch, B., Steffes, M.: Experimental and Numerical Stu-
dies of a Surface Grinding Process. In: Proceedings of the 2nd International Confe-
rence on Process Machine Interaction (PMI) Session G1 (2010)
Chapter 5
Process Machine Interaction in Pendulum
and Speed-Stroke Grinding
M. Weiß, F. Klocke, and H. Wegner
Abstract. The complex interaction of process forces and machine structure affects
the quality of ground workpieces, especially in highly-productive machining
processes, if machines are operated at their limits. In speed-stroke grinding, the
highly-dynamic process forces are caused by high workpiece velocities and high
acceleration of the machine table. These forces are influenced by the process pa-
rameters, the material properties, the coolant application and the grinding tool
specification. The paper describes the approach to simulate the process machine
interaction in speed-stroke grinding by a coupled model. The machine is modeled
by a multi-body simulation, which can depict the static and dynamic behavior of
the machine for every working position. This machine model is coupled with an
analytical-empirical force model, which predicts the process forces regarding the
process parameters, the coolant application and the workpiece material. The ma-
chine control system is implemented in the model as well. The ability to model a
speed stroke grinding process, including the machine, the control system and the
process itself can be used to predict and improve the workpiece quality regarding
the measurement accuracy minimizing time and cost intensive experiments.
5.1 Introduction
In many cases, grinding is the only economical process to machine parts. Materials,
which are especially difficult to machine, are ground to achieve high productivity
and meet the requirements in terms of surface qualities. This paper focuses on
speed-stroke grinding. Speed-stroke grinding is defined as a surface grinding
process with high table speeds. The table speed of most of the conventional grind-
ing machines is limited to 30 m/min. Pendulum grinding with table speeds higher
than 50 m/min is defined as speed-stroke grinding [1]. In speed-stroke grinding the
chip formation mechanisms change to bigger chips causing lower specific grinding
energies at high table speeds. The resulting lower heat transfer to the workpiece can
be used to improve the workpiece quality or the productivity of grinding processes.
Speed-stroke grinding can be applied for machining highly heat-resistant materials
and ceramics as well as hardened steel [2, 3, 4].
High table speeds result in short contact times between the grinding wheel and
the workpiece. Short run-in and run-out phases lead to high gradients of process
forces. The values of these forces are influenced by the complex interaction
102 M. Weiß, F. Klocke, and H. Wegner
between the process parameters, coolant application and the mechanics of chip
formation specific to the behavior of different workpiece materials. The material
removal in grinding is realized by an amount of single grinding grits with an unde-
fined shape. Several grinding grits are in contact with the workpiece surface at the
same time, realizing material removal. Hence, it is a complex challenge to predict
the forces occurring in the grinding process, their influence on the excitement of
the machine structure and, thus, the workpiece quality.
In the following, an approach is shown, where the machine is modeled by a
flexible multi-body simulation and the occurring forces by an analytical-empirical
force model. These two models are coupled to comprehend the process-machine
system. As an input for the simulation model and to validate the simulation results
experimental investigations have to be conducted. The results of the analysis of
the process and the machine behavior were used to set up the simulation model.
Afterwards, experimental investigations of certain influences on the machine be-
havior and process forces resulted in a more detailed and more exact model of the
whole process and lead to a potential of optimization for industrial pendulum and
speed-stroke grinding processes.
5.2 Experimental Investigation

To realize a solid data base as an input for modeling the machine properties and
the occurring process forces and vibrations have to be analyzed. The machine
behavior is examined by a modal analysis of the used speed-stroke grinding ma-
chine. The influence of different materials, process parameters and coolant appli-
cation on the process forces were examined by multiple grinding experiments with
different setups, as described below.
5.2.1 Experimental Setup

For the experimental investigations a BLOHM PROFIMAT 408 HTS speed stroke
grinding machine was used. In comparison to conventional surface grinding ma-
chines this machine is equipped with linear drives for the machine table move-
ment. Thus, the workpiece velocity can reach up to vw = 200 m/min. The table is
driven with a maximum acceleration of aw = 50 m/s2. Because of the high table
speeds and acceleration, especially at the reversal of the machine table, an impulse
decoupling with spring-dampening elements and an eddy current brake is mounted
beneath the machine table to reduce the impact on the structure and the process.
The machine and the 3D CAD model of the machine interior are shown in
Fig. 5.1. For the experiments, a precision height adjustment for coolant application
was designed and mounted to the headstock. This allows a repeatable adjustment
of different nozzles regarding the nozzle angle, axial position and height of the
coolant application. Hence, it is possible to conduct experiments with defined coo-
lant application parameters. For the experiments, the dynamic forces during pen-
dulum and speed-stroke grinding processes were measured by a multi-component
force dynamometer.
5 Process Machine Interaction in Pendulum and Speed-Stroke Grinding 103
Height adjustment
Height scale
Angle and position
y
adjustment
Needle nozzle
z Grinding wheel
Workpiece
Multi component
force dynamometer
x Machine table
Fig. 5.1 CAD drawing of machine interior with precision nozzle height adjustment
The static and dynamic behavior of the machine was examined by an experi-
mental modal analysis of the machine tool. A hydraulic actor unit was mounted to
the machine. This actor unit applied dynamic loads with different frequencies to
the grinding wheel. A piezo-electric sensor unit measured the response of the ma-
chine structure to the dynamic loads. This procedure was repeated with different
sensor and actor positions in x, y and z direction so that the cross relations be-
tween the excitations could be measured. More information about modal analysis
can be found in Section 1.2 of this book. The data was used to parameterize the
machine model. Thus, a comparison between the modeled and the real machine
behavior can be carried out.
5.2.2 Experimental Results

In the following, the results of the experiments are shown. At first, the measured
machine behavior is described. The results of the grinding experiments and the
evaluation of occurring grinding forces are shown below.
Modal analysis of machine behavior

As a basis for the research of the interaction of the machine and the grinding
process the static and dynamic machine behavior needs to be determined. The fre-
quency response function of the used machine is shown exemplarily in Fig. 5.2.
The static and dynamic load is induced in y-direction as well as the measured def-
lection. Variations of the tool centre point in y-direction are causing deviations in
dimension of the ground workpiece as they are influencing the depth of cut instan-
taneously. The system was preloaded with a force of 1,000 N. Measurements
show that the static elasticity of the machine is 0.03 µm/N. The first three eigen-
frequencies are at 38, 58 and 78 Hz with maximum amplitude higher than the stat-
ic flexibility. A modal analysis was conducted to identify the eigenmodes of the
system. The results and the comparison with the modeled machine behavior are
shown in the chapter Modeling and Simulation.
10-1 3. eigenfrequency 78 Hz
FRF w m:2abs:+X/w m:Piezo:+X
Synthesized FRF w m:2abs:+X/w m:Piezo:+X
Elasticity Gyy [µm/N]
measured frequency response function

modeled frequency response function
10-2
2. eigenfrequency 58 Hz
1. eigenfrequency 38 Hz
10-3
0 100 200 300 400 500 600 700 800 900

Frequency [Hz]
180°
Phase φ
0°
Synthesized FRF w m:2abs:+X/w m:Piezo:+X
-180°
0 100 200 300 400 500 600 700 800 900
Frequency [Hz]
Fig. 5.2 Frequency response function (Gyy) of test machine, preload 1,000 N
Process forces
In the experiments, the dynamic forces during pendulum and speed stroke grind-
ing processes are measured by a multi-component force dynamometer and record-
ed on a PC. The measurement of specific normal and tangential forces and the
acoustic emission (AE) signal during two passes of the grinding wheel is shown in
Fig. 5.3. It was recorded in the middle of a speed-stroke process at a ground vo-
lume of V’w = 500 mm3/mm. The AE-signal is important for the analysis of the
data. Every time the grinding wheel gets in contact with the workpiece a peak in
the AE-signal can be detected. Thus, the exact time frame from the first to the last
contact of the grinding wheel with the workpiece can be specified.
To explain different effects the measurement is separated into nine phases. At
the beginning of Phase 1 the workpiece is on the left side of the machine and acce-
lerates towards the grinding wheel. Because of the moment of inertia the specific
tangential forces show a characteristic peak during the acceleration. At the begin-
ning of Phase 2 the workpiece is engaged by the grinding wheel. With a work-
piece velocity of vw = 80 m/min and a depth of cut ae = 15 µm the run-in time is
very short. It takes less than 2 ms from the first contact of the grinding wheel until
the maximum depth of cut is reached. However, the normal force needs a certain
time to reach its stationary level at the end of phase 2.
% (scaled) tangential force norm al force /

F‘n / N/mm
60
Specific
50
40
30
20
10
1 3 5 7 9
0
2 4 6 8
40
F‘t / N/mm
Specific
20
0
-20
-40
46,7 46,8 46,9 47 47,1 47,2 47,3 47,4
Time / s
AE-Signal
100
80
60
40
20
0
1 3 5 7
vs vs vs vs
vw vw vw vw
Run in
1 Table acceleration 2 Coolant phase + up grinding
Run out
3 Up grinding 4 Coolant phase
5 Table deceleration – Change table direction – Table acceleration
Run in
6 Coolant phase 7 Down grinding
Run out
8 Coolant phase + down grinding 9 Table deceleration
Grinding wheel: 10B 181 V (ceramic bond) Q‘w = 20 mm3/mms

Coolant: Emulsion (4%) vs = 80 m/s
Needle nozzle, 8° nozzle angle, 110 l/min vw = 80 m/min
Workpiece material: 100Cr6 (700 HV) ae = 15 µm
Fig. 5.3 Force and acoustic emission signals during speed-stroke grinding
Responsible for this effect is the coolant. With the shown machine setup the
coolant nozzle is on the right side of the grinding wheel. If the workpiece is
approaching from the left side, the main coolant flow is not able to get into the
contact zone. The coolant jet hits the workpiece from the right side. Until the
workpiece moves further the coolant jet builds up a hydro-dynamic pressure in
and in front of the contact zone, as described in [6, 7]. Depending on flow rate,
coolant speed, nozzle angle, nozzle position and process parameters the coolant
adds a significant value to the normal force [8]. Besides that, the hydrodynamic
coolant pressure influences the dampening behavior of the system [9].
During phase 3 the process is nearly stationary. At the end of this phase, the
normal forces drop rapidly because of the disengagement of the grinding wheel.
After this, the coolant forces drop with a time delay in phase 4. As the gap
between the workpiece and the grinding wheel gets wider while the workpiece
moves apart the measured normal force decreases. In phase 5, the workpiece table
decelerates, changes its direction and accelerates in the opposite direction, which
can be seen in the tangential force progression. Phase 5 shows that the level of
acoustic emission is lower than in phase 1 and 9 even though the grinding wheel is
not in contact with the workpiece. The reason for this phenomenon is that the coo-
lant jet hits the workpiece from the side during phase 1 and 9. When the grinding
wheel is at the right side of the grinding wheel there is no coolant splashing
against the workpiece.
Phase 6 is characterized by an increase of normal forces while the workpiece is
not in contact with the grinding wheel yet. This is related to the dynamic coolant
forces, which build up because of the narrowing gap between the workpiece
and the grinding wheel. The run-in takes place at the beginning of phase 7. Before
the workpiece runs out of contact the forces drop as a result of the deflected
coolant jet.
The described effects show the complexity of forces occurring during the grinding
process. To model the forces the main influences have to be taken into consideration.
In order to quantify the effects of process parameters workpiece material and coolant
application numerous systematic experiments have been conducted.
The measured force signal contains a lot of information about the process and
the influences mentioned above. To separate the effect of the chip formation from
the coolant induced forces the data has to be analyzed as shown in Fig. 5.4. On the
upper left, measured specific normal forces plotted against the grinding wheel cir-
cumferential speed vs up and down grinding are shown. The forces drop with
higher rotation speed of the grinding wheel from above 30 N/mm to 20 N/mm.
The coolant forces were measured after every fifth stroke of the process by
conducting one stroke without a depth of cut ae. The results are shown exemplarily
in Fig. 5.4 on the upper right. The coolant induced forces increase up to a certain
point – here 80 m/s – before they begin to decrease at higher circumferential
speeds. For conventional processes a similar trend was described by Heinzel [10].
Higher grinding wheel circumferential speeds lead to a higher flow rate through
the contact zone. Hence, less coolant ponds in front of the contact zone and the
pressure in this area drops. As a result, the specific coolant normal forces de-
crease.
If the specific coolant normal forces F’nKSS are subtracted from the measured
specific grinding normal forces F’n, the specific cutting normal forces F’c,n can be
identified. This is necessary to evaluate the separate effects and to simplify
the modeling of the occurring forces in grinding as every effect can be examined
separately.
Specific grinding normal
Specific coolant normal

40 40
force F‘n KSS / N/mm

force F‘n / N/mm
30 30
20 20
10 10
0 0
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120140
Grinding wheel circumferential Grinding wheel circumferential
speed vs / m/s speed vs / m/s
40
Specific cutting normal
Up-grinding
force F‘c, n / N/mm
30 Down-grinding
20
10
0
0 20 40 60 80 100 120 140
Grinding wheel circumferential
speed vs / m/s
Material: Grinding parameters: Coolant application:

100 Cr 6 (AISI 52100) vw = 120 m/min Oil
Grinding wheel: ae = 10 μm Needle nozzle, 11°
CBN 10B 181 V Q‘w = 20 mm3/mms Flow rate: 62 l/min
Fig. 5.4 Evaluation of cutting and coolant-induced forces
Besides the hydrodynamic pressure, which builds up in and in front of the con-
tact zone, additional coolant forces can be induced due to the ground geometry.
An example is shown in Fig. 5.5. Narrow gaps between parts of the grinding
wheel, which are not in contact with the workpiece, and the workpiece lead to ad-
ditional hydro-dynamic forces. The smaller the gap, the higher these forces. If
such workpiece geometries are ground, the additional coolant forces need to be
considered.
In addition to the forces, the coolant has an influence on the dampening beha-
vior of the whole process machine system. To evaluate these effects the following
experimental setup was used. A workpiece made up of the two different materials
is ground. On one side, the workpiece consists of hardened 100Cr6 and on the
other side of Ureol, an easy to machine and homogenous modeling material. This
workpiece was ground with high table speed. The machining of the modeling
workpiece material creates very small forces compared to the bearing steel.
Grinding wheel
Additional coolant
forces
Total depth of
cut
FC
Workpiece
Contact area
Gap between grinding
wheel and workpiece
Fig. 5.5 Additional coolant forces by means of workpiece geometry
The experimental setup is shown in Fig. 5.6. As a result, the forces drop rapidly
at the junction of the two materials. If no coolant is used, a deviation of the profile
height can be measured on the surface of the soft material. When using coolant the
deviation is much smaller due to dampening and pre-load effects of the fluid in
and in front of the contact zone. Although several experiments with this setup
were accomplished quantitative statements about the dampening behavior could
not been derived yet as the collected data scatters in a big range. As a conse-
quence, the experimental setup and measurement need to be improved.
Without coolant
Profile / µm
Height/ µm
Easy to machine 25
of
modeling material
0
0
Profile
(Ureol)
-25
0 25 50 75 100
Distance / mm
With coolant
Profile / µm
25
Height of
00
100Cr6 -25
0 25 50 75 100
Distance / mm
Grinding wheel: 10B 181 V Q‘w = 20 mm3/mms

Coolant: Emulsion (4%) vs = 90 m/s
Needle nozzle, 8° nozzle angle, 110 l/min vw = 120 m/min
Workpiece material: 100Cr6 (700 HV) ae = 10 µm
Fig. 5.6 Influences of coolant dampening and pre-load effects

Another important influence on the occurring forces is the workpiece material.

Experiments have been conducted with the aim of evaluating the process forces
grinding different materials with speed-stroke grinding parameters. As an exam-
ple, the specific normal forces depending on the hardness of the material and the
table speed are shown in Fig. 5.7. Smaller chip thicknesses at low table speeds
lead to more friction and forming processes in chip formation compared to higher
table speeds. Thus, the specific grinding forces are lower at high table speeds [11].
The effect of the material hardness on the specific normal forces decreases with
higher table speeds. For certain groups of workpiece materials, this kind of model-
ing is sufficient. However, there are different effects on the grinding forces regard-
ing different workpiece materials. As shown in the figure, the specific normal
forces are higher while grinding the ductile and tough nickel-based alloy Inco-
nel 718 at different table speeds compared to 100Cr6. Hence, a force model needs
to be adapted to different workpiece material properties. Composition and struc-
ture of the workpiece material are the main influences on the grindability of work-
piece materials. Furthermore, the workpiece temperature has an influence on the
machinability. The necessary force to form a chip decreases at higher temperatures
but as the energy flow to the workpiece is comparatively low in speed-stroke
grinding this effect is neglected in the following model.
50
Inconel 718
Specific cutting normal
40
force F‘c,n / N/mm
30
100Cr6
20
10 vw = 20 m/min vw = 80 m/min
vw = 50 m/min vw = 120 m/min
0
0 200 400 600 800
Hardness / HV
100 Cr 6 (AISI 52100) vs = 120 m/s Emulsion 4 %
and Inconel 718 ae = 60, 24, 25, 10 µm Needle nozzle (11°) +
Grinding wheel: Q‘w = 20 mm3/mms clearing nozzle
CBN 10B 181 VP Flow rate: 110 l/min
Fig. 5.7 Specific cutting normal forces grinding different materials
The material removal in grinding results from various contacts between the
grinding wheel and the workpiece. In addition, the grinding grits have irregular
shapes and they move on a certain path around the centre of the grinding wheel.
Because of this, some of the grits just plow through the workpiece surface, some
grits form chips and some are not in contact with the workpiece surface. A large
amount of the energy is transferred into elastic and plastic deformation of the
workpiece material. Hence, the grinding wheel properties, depending on influ-
ences like grit size, bond type or dressing conditions, influence the occurring
forces significantly. Different grinding wheels were tested and need to be imple-
mented in the modeling as well, as shown in the following chapter.

To model the complex interaction between the machine and the process with the
highly dynamic forces shown in the chapter above an approach with a coupled
model is used. The model is implemented in Matlab Simulink and calculates the
forces and deviations in discrete time steps. The first part of the model is the ma-
chine, which is represented by a moveable multi-body simulation (MBS). The oc-
curring grinding forces are predicted by an analytical-empirical process model.
These forces are calculated and transferred to the machine model. Due to the elas-
ticity of the machine these forces lead to a deviation of the tool centre point
and thus to a deviation of the ground workpiece shape. These deviations are
transferred to the force model, where the forces for the following time step are
calculated. The model operates in discrete time steps, which are arbitrary in the
simulation. Another important part of the model is the simulation of the machine
control system as the response time of the control system can lead to deviations in
speed-stroke grinding processes as well. The principle of the coupled simulation is
shown in Fig. 5.8. Coupled simulations permit the usage of different simulation
environments with data exchange by means of a suitable interface. Proper inter-
faces have to be designed [12]. Another advantage of coupled models is the possi-
bility to adapt each of the models separately so that changes and improvements
can be carried out more easily.
Simulation of the grinding process Simulation of the control system

F(t) F(t), M(t)
Process Position control
Machine tool
MBS
DBS DBS
s(t) s(t), v(t)
Quality Process stability +
Fig. 5.8 Coupled process – machine – control simulation

For the designed model the simulation of the process forces and the control sys-
tem is implemented into Matlab Simulink. In the following, the machine and the
process model are described in more detail.
Machine model
For the design of a MBS model of a machine the following steps have to be con-
ducted. The procedure of the development of a MBS machine model is shown in
Fig. 5.9.
1. CAD-model 3. Moveable flexible

MBS-model
Comparison with
experiments
2. FE-model
Abstraction and
meshing of machine
components Modelling of
moving parts
Fig. 5.9 Procedure for the development of a multi body machine model
Based on CAD data of the machine the relevant parts need to be abstracted.
Small bore holes and small radii are removed because they have no significant in-
fluence on the stiffness or the mass of the system. As a result, the FE-meshing of
the finite element model (FE-model), which is performed afterwards, is less com-
plex. Next, guiding systems, spindle bearings and screw connections are approx-
imated by simple spring elements. Afterwards, a comparison between the
measured machine behavior and the simulated machine is conducted to evaluate
the required stiffness of the joints. Furthermore, the relevant eigenmodes and ei-
genfrequencies can be depicted. Fig. 5.10 shows the 2nd eigenmode of the machine
system at 58 Hz exemplarily.
For the MBS-model the FE structure parts have to be converted into flexible
bodies. A flexible body allows the description of its flexible properties at defined
force transmission points. The calculation of the forces at these knots and the elas-
tic deformation of the parts is implemented in MSC.Adams and calculated by a
developed sub-routine for every position of the spindle and the machine table in
the machine. The special aspect about the developed MBS-model is the possibility
of realizing a movement of certain parts. This allows the simulation of movements
of the spindle and the machine table like in the real machine. More detailed infor-
mation about the machine model and the subroutines can be found in [13, 14].
2nd Eigenmode
Eigenmode of original machine Eigenmode simulation model
Eigenfrequency: 58 Hz Eigenfrequency: 57,4 Hz

Rotary motion of the spindle an torsion in Y-direction of the Y-Slide
Fig. 5.10 Comparison between measured and simulated eigenmode and eigenfrequency
Moveable MBS-models have the advantage of representing the machine beha-

vior for every position of the machine components. However, the design of a mo-
veable MBS-model of a machine is very time consuming and a lot of knowledge
about the machine structure and the stiffness of the joints is necessary.
Process model
The task of the process model is to calculate the occurring forces during the grind-
ing process, regarding the position of the grinding wheel, the grinding parameters
and other boundary conditions. Many grinding force models have been developed
over the last decades. Usually, a technological study of grinding processes results
in the development of a model, which is valid for a closely limited field with given
boundary conditions, as grinding is a very complex process of material removal
[15]. Many models to predict the grinding forces are presented in [15, 16, 17].
Most of these models consider the kinematic of the grinding process and an empir-
ical additive, which represents the influences of the grinding wheel and the work-
piece material.
For the prediction of the grinding forces during speed-stroke grinding an
adapted grinding force model was developed. The schematic representation of the
model is shown in Fig. 5.11. Depending on the workpiece position the current ma-
terial removal is calculated by using a chip longitudinal section Al, which
represents the intersection of the grinding wheel and the workpiece during one
time step of the simulation. This longitudinal section Al is multiplied by an expe-
rimental investigated force-ratio factor, which is dependent on the process
parameters, the material properties and the grinding wheel specification, to get the
cutting forces. Simultaneously, the coolant forces are calculated. These forces are
dependent on the workpiece position and the process parameters as well, which
was shown before. Additionally, coolant application parameters like nozzle angle,
nozzle type and coolant flow rate are considered [18]. The coolant forces and the
cutting forces are summarized to the total process forces.
Material properties
kc
vw Fc = Al ⋅ k c
Chip longitudinal vs
section Al Cutting forces
F [N]
vs Process parameters
vw and tool properties
Total process forces
Workpiece position
v KSS Nozzle
α

Q KSS
Coolant forces
Coolant parameters
Fig. 5.11 Schematic representation of the force model
In the following, the single modeling steps are described in more detail.
Fig. 5.12 shows the calculation of the chip longitudinal section during the run-in
phase. In position I, the grinding wheel first gets into contact with the workpiece.
During the run-in phase the contact length increases and thus the material removal
per time step also increases. This increasing material removal is proportional to
the area, which is bordered by the contact length between two time steps. Towards
time step III the maximum depth of cut is reached and the chip longitudinal sec-
tion does not change in case of an even workpiece surface. Grooves in the work-
pieces can also be represented by the model.
t0 t1 t2
x Surface with
deviations
Position: I II III
ΔAl2 Δt
ΔAl1 Ideal surface
Grinding wheel ae,0
ae,1
ΔAl3 ae,2
Al lg
vw Workpiece
A l: Chip longitudinal section ΔAli: Deviation of Chip longitudinal section

lg: Contact length
Fig. 5.12 Chip longitudinal section during run-in of the grinding wheel
With constant time steps Δt the change of the longitudinal chip section can be
described by geometrical values. For a constant workpiece velocity vw the longi-
tudinal chip section can be described as a function of the position x according to
equation 1 during the run-in phase.
A l,max 2 ⋅ A l,max (1)
A l,Run-In (x) = − 2
x2 + x ± ΔA li
lg lg
Al,max is the maximum chip longitudinal section, which is reached after the run-in
of the grinding wheel at the maximum depth of cut ae. ΔAli is the deviation of the
chip longitudinal section Ali due to an uneven surface or the geometry. At the run-
out, the chip longitudinal section is calculated with the following equation:
A l,max 2 ⋅ A l,max (2)
A l,Run-Out (x) = 2
x2 − x + A l,max ± ΔA li
lg lg
If the workpiece has no deviations at the surface, the chip longitudinal section
stays constant. Fig. 5.13 shows the cutting force characteristic of one stroke during
grinding. In speed-stroke grinding, the run-in and run-out phases are in the range
of milliseconds. If the table speed is 120 m/min at a depth of cut of 10 µm, these
phases take 0.935 ms, for example. If there are deviations in the surface, an addi-
tional ΔAli needs to be added for the calculation.
chip longitudinal
section / Al
Al,max Overrun
Al,Run-in Al,Run-out
0
Grinding wheel position x
Fig. 5.13 Characteristics of total chip longitudinal section during one stroke
By modeling the chip longitudinal section it is possible to calculate the intersec-

tion between the grinding wheel and the workpiece for surface grinding processes
[19]. This is equivalent to the actual material removal, taking the depth of cut and
the workpiece velocity vw into account. Uneven workpiece surfaces lead to a
changing material removal, which is also considered in the model. In unfavorable
cases, the change of the material removal rate in one stroke can lead to chatter ef-
fects. This could be provoked neither during the grinding experiments nor in the
simulation. This result corresponds to the modal analysis shown in chapter 2 as
there are no distinct weaknesses of the machine regarding chatter effects.
To predict the forces, which affect the machine structure quantitatively, the
modeling focuses on the process parameters, the workpiece material and the
coolant as these influences were emphasized as the significant influences in speed-
stroke grinding. First, the forces, which depend on the process parameters, are
calculated. The calculation of the forces is carried out by multiplying the chip lon-
gitudinal section with the determined force ratio factors kc (Equation 3). The chip
longitudinal section is dependent on the deph of cut ae, the workpiece velocity vw
and the chosen time step Δt, as explained above. The influences on the force ratio
factor kc are the grinding tool, the material properties, e. g. the hardness and the
sort of material, and the grinding wheel circumferential speed vs.
Fi = Al (ae , vw , Δt ) ⋅ kc (vs , Material, Tool) (3)
Specific force ratio factor
120 B181 LHV Define grinding wheel

k‘c, n / 103 N/mm3
1
100 specification and
80 10B 181 V circumferential speed vs
2
60 Determine specific
40 2
B251 V force ratio factor
20 1
0
0 20 40 60 80 100 120
Grinding wheel circumferential
speed vs / m/s
Adaption of kc
Set workpiece material
3
Specific force ratio
300 and material properties

Inconel 718
adaption / %
233 %
Determine adaption
100Cr6
200 4 4
factor
100
3 vw = 20 m/min vw = 80 m/min
vw = 50 m/min vw = 120 m/min
0
0 200 400 600 800
Hardness / HV
Fig. 5.14 Adaption of force ratio factor kc
As the different influences are taken into account the kc factor has to be adapted.
An example for the functionality of the model is shown in the following. Fig. 5.14
illustrates the calculation of the adapted force ratio factors. Different grinding
wheel specifications lead to specific force ratio factors kc at different circumferen-
tial speeds vs. The data was generated in several experiments for different grinding
wheels. In the example in Fig. 5.14, a grinding wheel with a grit size of 181 µm is
used.
The first diagram at the top of the Figure shows the determination of the specif-
ic force-ratio factor for a vitrified, bonded CBN wheel with a grit size of 181 µm
at a circumferential speed of 120 m/s. The specific force-ratio factor is
59,000 N/mm3, assuming defined standard conditions. These standard conditions
are hardened and tempered 100Cr6 as the workpiece material and a combined
needle nozzle with a clearing nozzle with a total coolant flow rate of 110 l/min.
After the “standard” force-ratio factor kc has been calculated it needs to be adapted
by the influences mentioned above. In the example in Fig. 5.14, the nickel base al-
loy Inconel 718 is chosen. As mentioned above, the material, which is difficult to
machine, leads to higher cutting forces. Thus, the force-ratio factor is adapted. In
this case, the forces are higher by a factor of 2.33. This leads to a total specific
force-ratio factor kc,tot of 137,470 N/mm3.
The coolant application affects the forces in two ways. On the one hand, the
coolant application has an influence on the chip formation, which leads to varying
forces. This is also considered by an adaption of the force ratio factor kc. On the
other hand the hydrodynamic pressure adds a value to the total forces. The simula-
tion program calculates these forces depending on the actual coolant conditions.
The coolant forces are calculated in the model by using the workpiece position as
the input factor as well. In combination with the process parameters and the coo-
lant application, parameters like coolant flow rate and nozzle angle, the occurring
forces can be predicted similar to the cutting forces. These are the most important
parameter adjustments for the coolant supply [20].
In a last step, the total process forces for the current time step are summarized.
Then the value is transferred to the machine model, which calculates the static and
dynamic displacement of the grinding wheel. With the displacement, the cycle
starts again with the calculation of the forces for the next time step. Hence, it is
possible to record the tool path and the resulting surface.
A comparison between a measured and simulated surface is shown in Fig. 5.15.
The measurement shows the characteristic super-elevation at the beginning. The
fast-run in of the grinding wheel results in a rapid raise of the process forces. On
this side of the workpiece, the system is pre-loaded by the hydrodynamic coolant
forces. Accordingly, the profile height on the left side is higher at this side com-
pared to the right. After a short period of vibration the workpiece is quite even.
The peak in the middle of the measurement of the real surface is used as a refer-
ence as the measurement length of the used surface measurement device is less
than 100 mm. On the right side of the workpiece the profile height drops. This is
related to the decreasing coolant force at the right side of the workpiece, as de-
scribed above. The height of this step varies depending on the process parameters
and coolant application.
30 Measurement
reference
Measured surface
Profile height [µm]
20 Simulated surface
10
-10
0 20 40 60 80 100 120 140 160 180

Length [mm]
100 Cr 6 (AISI 52100) vs = 120 m/s Emulsion 4 %
Grinding wheel: ae = 10 µm Needle nozzle, 11°
CBN 10B 181 V Q‘w = 20 mm3/mms Flow rate: 62 l/min
Fig. 5.15 Measured and simulated workpiece surface after speed-stroke grinding
5.4 Optimization
The approach to model the interaction between the speed-stroke grinding process
and the machine structure, as shown in this paper, can be used to improve produc-
tion processes. Furthermore, it builds the foundation for further research for the
understanding of the grinding process, especially speed-stroke grinding.
It was shown that the coupled simulation allows the prediction of the shape of
the speed-stroke ground workpiece, taking the process parameters, the coolant ap-
plication and the workpiece material into account. If the process result can be pre-
dicted without conducting time and cost-consuming experiments, a faster start of
production can be realized. With the knowledge of the maximum deviation of the
workpiece measurements the process time can be optimized. The roughing process
leads to a certain surface. The grinding minimum allowance for the finishing
strokes can be determined and, thus, the minimum cycle time can be depicted.
The developed coupled simulation can be adapted for other grinding processes.
On the one hand, the grinding force model needs to be adapted for different grind-
ing processes. In particular, the calculation of the current material removal rate
needs to be adapted to other kinematics. On the other hand, a new machine model
has to be built up for other machines. Depending on the process reduced machine
models with less build-up effort may be sufficient. The advantage of flexible mul-
ti-body simulations is the ability to simulate the machine behavior in every posi-
tion of the machine slides, the modeling effort, however, is very high, which is
shown above. In a lot of surface grinding processes the spindle moves within a li-
mited area in the machin only. As a consequence, the simulation of the machine
by means of a multi-mass spring system can be sufficient. This system can be es-
tablished in a substantially shorter time frame.
Another way to use the results of this work is to already evaluate the behavior
of machines in the design phase. Based on the CAD data the machine can be mod-
eled and the process behavior as well as the process result can be estimated with-
out building and testing the new machine. Structural weaknesses and illegitimate
process parameters can be determined, specific to the grinding processes.
5.5 Conclusion and Outlook

In speed-stroke grinding, the highly dynamic process forces cause deviations of the
machine structure. In this paper, the development of a coupled process-machine
model has been presented. An approach for modeling the process forces depending
on process parameters, material properties and coolant application parameters has
been described. The different effects, which influence the occurring process forces,
have been analyzed separately in various experiments. Coupled with the MBS-
model of the machine and the control system model the simulation is able to predict
the workpiece quality in terms of measurement and form deviations.
In addition, the presented results have revealed potential fields of research in
modeling the speed-stroke grinding process. In this project, only simple workpiece
geometries were analyzed. Theoretically, the prediction of more complex geome-
tries is possible. Another important issue is the influence of different coolant ap-
plications on the dampening behavior of the process machine system. From the
conducted experiments, no distinct statement regarding this influence can be
made.
A major goal of modeling production processes must be the evaluation of
process results without time and cost-consuming experiments. At the moment,
these experiments are still necessary to parameterize the developed models. The
approaches developed in the priority program 1180 can be used as a basis for fur-
ther research to understand the complex grinding process.
References
[1] Duscha, M., Klocke, F., D’Enremont, A., Linke, B., Wegner, H.: Investigation of
Temperatures and residual stresses in speed-stroke grinding via Fea Simulation and
practical tests. In: Proceedings in Manufacturing Systems, vol. 5(3), pp. 143–148
(2010)
[2] Zeppenfeld, C.: Schnellhubschleifen von gamma-Titanaluminiden. Gamma-
Titanaluminiden. RWTH Aachen University, Aachen (2005)
[3] Nachmani, Z.: Randzonenbeeinflussung beim Schnellhubschleifen. Dissertation,
RWTH Aachen University, Aachen (2008)
[4] Uhlmann, E., Sammler, C.: Schnellhubschleifen von Hochleistungswerkstoffen. Jahr-
buch Schleifen, Honen, Läppen und Polieren, vol. 64, pp. 155–168. Vulkan-Verlag,
Essen (2010)
[5] Weck, M.: Dynamisches Maschinenverhalten bei der Zerspanung mit undefinierter
Schneide. Werkzeugmaschinen 5 – Messtechnische Untersuchung und Beurteilung,
[6] Beck, T.: Kühlschmierstoffeinsatz beim Schleifen mit CBN. Dissertation, RWTH
Aachen University, Aachen (2002)
[7] Wittmann, M.: Bedarfsgerechte Kühlschmierung beim Schleifen. Dissertation, Uni-
versity of Bremen, Bremen (2007)
[8] Brinksmeier, E., Heinzel, C., Wittmann, M.: Friction, Cooling and Lubrication in
Grinding. Annals of the CIRP, Keynote Paper 48(2), 581–598 (1999)
[9] Marksoud, T.M.A., Mokbel, A.: Suppression of chatter in grinding using high-
viscosity coolants. Proceedings of the Institution of Mecanical Engineers, Part B
(Journal of Engineering Manufacture) 216, 113–123 (2002)
[10] Heinzel, C.: Methoden zur Untersuchung und Optimierung der Kühlschmierung beim
Schleifen. Dissertation, University of Bremen, Bremen (1999)
[11] Klocke, F.: Process Design. Manufacturing Processes 2 – Grinding, Honing, Lapping,
[12] Esser, M., Brecher, C., Witt, S.: Interaction of manufacturing process and machine
tool. Annals of the CIRP, Keynote Paper 58, 588–607 (2009)
[13] Hoffmann, F.: Optimierung der dynamsichen Bahngenauigkeit von Werkzeugma-
schinen mit der Mehrkörpersimulation. Dissertation, RWTH Aachern University, Aa-
chen (2008)
[14] Klocke, F., Brecher, C., Sitte, B., Weiß, M.: Analyse der dynamischen Wechselwir-
kungen bei Pendel- und Schnellhubschleifprozessen. Jahrbuch Schleifen, Honen,
Läppen und Polieren 64, 53–65 (2010)
[15] Tönshoff, H.K., Peters, J., Inasaki, I., Paul, T.: Modelling and Simulation of Grinding
Processes. Annals of the CIRP, Keynote Paper 41(2), 677–688 (1992)
[16] Brinksmeier, E., et al.: Advances in Modeling and Simulation of Grinding Processes.
Annals of the CIRP, Keynote Paper 55(2), 667–696 (2006)
process: machine interaction in grinding. In: Production Engineering – Research and
Development, vol. 3(1), pp. 111–120. Springer, Heidelberg (2009)
[18] Weiß, M., Klocke, F., Wegner, H.: Influence of coolant and workpiece material prop-
erties on dynamic grinding forces in speed stroke grinding. In: Proceedings of the 2nd
International Conference on Process Machine Interaction, Vancouver, G4 (2010)
[19] Klocke, F., Duscha, M., Hoffmann, F., Wegner, H., Zeppenfeld, C.: Machine –
Grinding Wheel –Workpiece Interaction in Speed-Stroke Grinding. In: Proceedings
of the 1st International Conference on Process Machine Interaction, Hannover, pp.
259–266 (2008)
[20] Wittmann, M., Heinzel, C., Brinksmeier, E.: Evaluating the Efficiency of Coolant
Supply Systems in Grinding. In: Production Engineering – Research and Develop-
ment, Braunschweig, vol. XI(2), pp. 39–42 (2004)
Chapter 6
Simulation of Process Machine Interaction
in NC-Shape Grinding
D. Biermann, H. Blum, A. Rademacher,

A.V. Scheidler, and K. Weinert
Abstract. The study focuses on the NC-shape grinding process when using toroid
grinding wheels and its simulation. First, the experimental investigation with re-
spect to the machine structure and its dynamic behavior, the process forces as well
as the temperature distribution in the workpiece and the grinding wheel are dis-
cussed. That forms the basis for the modeling and simulation of the NC-shape
grinding process. The simulation consists of a geometric-kinematical simulation
coupled with a finite element simulation. To validate the simulation, comparisons
between the quantities measured and the corresponding calculated values are car-
ried out. Subsequently to this validation the transferability of the simulation to
other grinding processes is studied. Furthermore, the simulation is utilized to op-
timize grinding processes, especially with respect to the NC data.
6.1 Introduction
Grinding processes are mainly used as surface finishing processes. The chip re-
moval is usually carried out by rotating grinding wheels with bonded abrasive
grains. The process is characterized by high cutting speeds and various contacts
between the workpiece and the single grains. Grinding is usually the last step in
the production chain of a workpiece and, accordingly, this phase plays a very im-
portant role. On account of this, defects caused by the grinding process would be
associated with high costs and must be avoided.
A toroid grinding wheel can be flexibly used for NC-shape grinding. Based on
the semi-circular profile of the grinding wheel and a line-by-line movement it is
possible to machine geometrically-complex surfaces. The different parameters re-
lated to NC-shape grinding with a toroid grinding wheel are demonstrated in
Fig. 6.1. The feed motion in the NC-shape grinding process is controlled by NC-
data, which provides excellent possibilities for grinding free-formed workpiece
surfaces. The use of CNC-technology and an adapted path control allows the
grinding of planes as well as profiles and complex shapes. The contact area be-
tween grinding wheel and workpiece varies during engagement along the tool
paths. In the case of complex engagement, the contact area is not measurable
without additional software tools. The shape grinding process is determined by the
shape of the tool, the kinematic conditions of the cut, the chip removal and the
surface produced [1].
122 D. Biermann et al.
Fig. 6.1 Different parameters of NC-shape grinding with a toroid grinding wheel [2]
The interaction between grinding wheel, workpiece and machine tool in an NC-
shape grinding process using a toroid grinding wheel is examined in the present
experiments. The research presented in this paper describes the process behavior
of the shape grinding process. The description of the interaction between grinding
wheel, workpiece and machine tool in a NC-shape grinding process requires vari-
ous simulation models and corresponding experiments.
In addition to obtaining a basic understanding of the process and its simulation,
the application of the simulation and its transferability to other processes are the
focus of the current research. Here, application means that the simulation results
are used to optimize the NC-shape grinding process. One approach we mainly
work on this purpose is the modification of the NC-data in such a way that grind-
ing errors are reduced or eliminated. First results related to path planning are pre-
sented in this chapter, where the depth of cut is increased to compensate for the
deflection of the grinding machine. It is considerably more difficult to reduce the
effect on the workpiece caused by oscillations, which result from the dynamic be-
havior of the machine, because restrictions relating to the machine tool control of
the grinding machine have to be taken into account. With respect to the practical
application of the simulation, the transferability of this simulation approach to dif-
ferent grinding machines and other grinding processes is of decisive importance.
The simulation is applied to other grinding wheels than the toroid one and the
workpiece material varied. Thereby, the parameters in the simulation model are
identified and it becomes obvious which parameters have to be adjusted. Further-
more, the dependency of the model's parameters on the modifications is described.
In the next section, we will discuss the experimental basis of the simulation and
describe the experiments relating to the dynamic behavior of the machine as well
as to measurements of force and temperature. The topic of Section 6.3 is the simu-
lation approach and a detailed description of single constituent parts of the simula-
tion. Furthermore, the determination of the model's parameters using numerical
methods is explained. Section 6.4 focuses on the validation of the simulation re-
sults. The application of the simulation to path planning and the transferability of
the simulation to other grinding processes are examined in Section 6.5. Finally, we
will draw some conclusions and provide an outlook on further developments.
6 Simulation of Process Machine Interaction in NC-Shape Grinding 123
6.2 Experimental Results

The accuracy and the dynamic behavior of the structure of the machine were in-
vestigated in order to model and simulate the shape-grinding process in its com-
plexity. In shape grinding, material removal occurs along the tool paths, which
vary in height. Thus, there is no engagement of the tool in the direction of feed
motion. Hereby, in the process, complex and no longer trivially measurable con-
tact surfaces between the tool and the workpiece can occur. In addition, the accu-
racy of the test machine used must be fundamentally taken into account in order to
be able to represent these contact surfaces in a way that corresponds realistically
to the process. Static, dynamic and thermal deformations as well as machine in-
duced errors can lead to variances, which cannot be accounted for solely by ex-
amining the tool, the contour of the workpiece and the programmed movement in
relation to each other. Furthermore, the machine must be suitable for the execution
of the programmed target motions. In this section, the measurement of the static
and dynamic machine tool behavior and the effects of this behavior on the surfac-
es of the workpiece, the generated temperatures and the forces are to be discussed.
6.2.1 Measurement of the Static and Dynamic Machine Tool

Behavior
The results of the experiments, which were conducted to measure the reactivity of
the traversing axes and the static as well as dynamic behavior of the machine
structure, will now be presented. First, the positioning accuracy of the machine
will be discussed and followed by the description of the dynamic properties of the
machine and their effects on the surface of the workpiece.
Examination of the Positioning Accuracy
The positioning accuracy was determined using a laser interferometer. In this way,
the absolute positioning error was determined. It appears that the positioning accu-
racy measured for the z and the y axes for a workpiece width of bmax = 20 mm and
a maximum profile height of hmax = 10 mm lies within the area of tolerance of
measurement precision. Slip occurs because of the belt drive of the x axis so that a
positioning can be realized only in one direction. Accordingly, the experiments
were conducted using down-grinding since small forces occur in this process [2].
The contours of the workpiece resulted from line-by-line traversing movements
of the toroid grinding wheel. A large number of grinding strokes with varying
amounts of material removal results from the process kinematics until the desired
workpiece contour is realized. The process is characterized by a high degree of
flexibility. The test machine is designed for a fast workpiece velocity of
vw, max = 35 m/min in its function as a surface and profile grinding machine. The
vertical traversing axis is however limited to traversing speeds of vf,y = 4 m/min.
The traversing speed must be constant. When interpolating the x and the y axes
the traversing velocity must conform to the slower speed. The interpolation of the
machine axes takes place linearly between the sampling points provided by the
NC program. Because of this, discontinuous transitions between the path segments
result, which lead to a significant faceting of the workpiece surface. Experiments
to determine the maximum traversing speed when the distance between the NC-
data points is Δ = 0.5 mm have shown that the machine reaches a maximum work-
piece velocity of vw = 2238 mm/min because the drive dynamics of the machine
using this NC-data points is limited. This has to be taken into account during the
shape grinding processes [2].
Dynamic Behavior of the Machine Structure

During the grinding process, forces are generated by the contact of the grinding
wheel with the workpiece, which directly affect workpiece quality or have an ef-
fect on the machine parts lying in the distribution of force. The grinding machine
system can thereby be induced to vibrate. Such a coupled system is determined by
mutual interaction between the machine structure and the process. The dynamic
behavior of the system is investigated in order to reach a better understanding of
the cause of the waviness and to be able to represent this in a simulation. For this
purpose, a modal analysis is conducted. The grinding machine is converted into a
simplified substitute model in order to determine the modal parameters. The subs-
titute structure is brought close into line with the grinding machine by using a total
of 106 nodes. The substitute system is understood to be dynamically linear, i. e.
we proceed on the assumption that a proportionality exists between excitation and
response. The linearity of the system is the pre-condition for ascertaining the fre-
quency response function [2].
Several modes arise from the modal analysis of the grinding machine, three of
which are largely identical with the frequencies found by measuring the flexibility
at the spindle. The related eigenfrequency forms are presented in Fig. 6.2. Basical-
ly, no significant relative displacements between the workpiece and the tool could
be seen at eigenmodes having a frequency of f < 50 Hz so that an influence on the
process and thus on the quality of the workpiece can be excluded here [2].
Fig. 6.2 Eigenmodes of the grinding machine investigated as determined by the modal
analysis along with a representation of the extreme position of deflection [2]
Two eigenmodes play an important role when dynamic forces in a vertical di-
rection affect the machine. Engaging the grinding wheel with the workpiece caus-
es a waviness with an eigenfrequency of f = 60.2 Hz. A strong vertical oscillation
of the spindle tower can be seen at this frequency, so the grinding spindle with its
tower oscillates along the y axis. The second eigenmode has an effect in the ver-
tical direction at an eigenfrequency of f = 143 Hz and represents its effect through
a vertical oscillation of the spindle. The spindle executes a horizontal oscillating
movement parallel to the machine table at an eigenfrequency of f = 84.6 Hz [2].
6.2.2 Force Measurement

The process forces are to be determined in order to develop a force model in the
simulation. These forces increase with increasing depth of cut ae and increasing
workpiece velocity vw. An example of the mean values of the radial forces for var-
ious workpiece velocities vw and depths of cut ae are shown in Fig. 6.3 [2].
Fig. 6.3 Forces at various depths of cut ae and workpiece velocities vw
Determining the Friction in the System

The friction coefficient is calculated by using Coulomb's Law to describe friction.
This law is simply an estimation because only a global force during the grinding
process can be measured. The friction coefficient calculated is nearly constant for
a grinding path in the quasi-stationary area. This coefficient, of course, behaves
differently depending on the engagement conditions during run-in and run-out.
This is presented for two conditions of engagement in Fig. 6.4. The coefficient
of friction also changes when the workpiece velocity vw and the depth of cut are
varied.
Fig. 6.4 Friction coefficients determined for different process parameters along the grinding
path
6.2.3 Temperature Measurement

Sect. 1.3.3 showed that temperatures have a considerable effect on the quality of
the workpiece. An important influence on temperature development is the coolant
supply to the contact area, which is optimized, as described in [3], for NC-shape
grinding using toroid grinding wheels. It has been shown that shoe nozzles are
best suited for this process because under constant machining parameters the least
workpiece damage in terms of grinding burn and cracks on the surface has been
found [3].
The process temperatures were measured using thermocouples and a thermo-
graphic camera in order to determine the influence of heat development during
NC-shape grinding. When measuring with thermocouples, it was seen in [4-7] that
the experimental results here, regarding the dependency of the temperature on the
parameters workpiece velocity vw, depth of cut ae, and the distance between the
thermocouple and the surface Δx, are comparable to the main results in case of the
toroid grinding wheel. The temperature increases with increasing workpiece ve-
locity and depth of cut and with the decreasing distance between the thermocouple
and the surface. An example is shown in Fig. 6.5 [8].
The temperatures measured using the thermographic camera during face grind-
ing are comparable to those found when using the thermocouple measurement
technique. For the same distance to the contact area and equivalent process para-
meters both methods yield similar temperature values. Thus, the thermographic
camera technique can be used when machining other surface contours with com-
plex and varying contact areas. It must be taken into account that only free-form,
two-dimensional surfaces can be machined because the camera can measure only
temperatures close to the workpiece surface. Since convex surfaces are produced
during the course of the experiments we conducted, one can get an idea of temper-
ature development when the contact area between grinding wheel and workpiece
Fig. 6.5 Dependency of temperature on the distance between the thermocouple and the con-
tact area [8]
is constantly changing. Temperature development at the contact area of these convex

surfaces is similar to the development of the process force along the grinding path.
The temperature increases with an enlargement of the contact area (Fig. 6.6) [8].
Fig. 6.6 Temperature development over workpiece length during shape grinding

For modeling the numerically controlled (NC) shape grinding process a holistic
approach is chosen. Consequently the reciprocal effects between structure and
process are considered. In line with the present state of research, simulation of the
process is conducted by linking two simulation tools. The geometric-kinematical

simulation describes the contact area Awgk between the grinding wheel and the
workpiece under idealized conditions. A finite element simulation takes the dy-
namics of the process into account.
The complete simulation consists of three parts: the geometric-kinematical si-
mulation, the finite element simulation and the removal predictor. The interaction
of these three parts is illustrated in Fig. 6.7 and described in this section. The
geometry, the material properties and the NC-data are the main inputs of the simu-
lation. The surface of the workpiece after the grinding process is the main output.
It is mainly determined by the actual depth of cut ae,act. The displacement of the
grinding wheel and the process forces are additional results.
The global simulation is controlled by the geometric-kinematical simulation.
Consequently, the global temporal discretization is implemented here. At the mo-
ment, equal time steps are used for the geometric-kinematical simulation and for
the finite element simulation. Later on, it may become necessary to use smaller
time steps for the finite element simulation in order to increase the accuracy.
Fig. 6.7 The simulation cycle in a time step
After the initialization of the different simulations, consisting mainly of the

reading of the geometric data and the usual finite element preparations, time step-
ping starts. The simulation cycle displayed in Fig. 6.7 is passed through in each
time step. It starts with the geometric-kinematical simulation. First, some prepara-
tion steps are necessary. The new position of the workpiece in relation to
the grinding wheel is determined by the NC-data. The resulting process force
Fres in the current time step is calculated from the geometric-kinematical data.
To calculate this force an empirical grinding force model is used, which is only
able to predict a resulting process force. No local or microscopic effects are taken
into account.
The data for the part of the workpiece surface which might be in contact with
the grinding wheel is passed on from the geometric-kinematical simulation to the
finite element simulation. This information is used to describe the contact con-
straints of the dynamic Signorini problem in the current time step. Then, the dis-
crete problem is solved and the normal contact stress at that time is determined.
The resulting accumulated contact force Fcon is calculated from the normal contact
stress and this value is returned to the removal predictor. The most important parts
of the grinding machine, the grinding wheel and the spindle, are taken into ac-
count in the finite element simulation. The stiffness of the other parts of the ma-
chine is modeled by elastic bearings.
The finite element simulation is able to calculate the accumulated contact force
Fcon. Thereby, the contact area is represented in detail. Fcon depends directly on the
actual depth of cut ae,act. So it is expected that a good approximation of ae,act has
been found, if Fcon and Fres agree. The predicted global process force Fres is stored
in the removal predictor.
Two steps are performed in the removal predictor. First, it tests whether the ac-
cumulated contact force Fcon and the predicted process force Fres match. If that is
the case, the cycle is left and the next time step is started. Otherwise, a corrected
value for the actual depth of cut ae,act is predicted and passed on to the geometric-
kinematical simulation. Here, the surface of the workpiece is modified according-
ly, and the cycle restarts. See [9] for more details.
6.3.1 Geometric-Kinematical Simulation

To simulate the grinding processes, various models presently exist, as described in
[5]. A geometric-kinematical simulation has been developed for NC-shape grind-
ing. The geometric-kinematical simulation aims to determine the contact situation
and is based on a dexel model, which represents the varying shape of the work-
piece, while the grinding wheel is represented as a solid torus. A dexel model has
been described in [10]. The material removal and the contact area of the grinding
wheel and the workpiece are evaluated by the instantaneous intersection of these
models. Therefore, the NC-data of the tool path generated by a CAD/CAM-
program can be used as input for the simulation [9, 11-13]. Fig. 6.8 shows model
elements of the geometric-kinematical simulation.
Fig. 6.8 Model elements of the geometric-kinematical simulation [2]
Force model
An empirical process force model was implemented in the geometric-kinematical
simulation based on a series of experiments. A specific force model for the NC-
shape grinding process based on the contact area of the geometric-kinematical
simulation was developed. This empirical force model is examined under ideal
conditions with deformation, vibration and temperature not being taken into
account [3].
The process forces increase with an accumulating workpiece velocity in an ex-
ponential way. This behavior is analogous to the behavior of the cutting forces
with defined cutting edges. Therefore, it is possible to determine a force model for
the grinding process, which is based on the Kienzle equation [14], usually em-
ployed for other cutting processes.
The process forces of the NC-shape grinding are calculated fairly exactly for
the geometric-kinematical simulation by the process force model. On average, the
results differ by 6 % from those of the measurements. Greater deviations only ap-
pear for large contact areas. By implementing the process force model into the
geometric-kinematical simulation the process forces, the contact area and the
workpiece shape can be calculated and used as the basis for the finite element si-
mulation [3].
6.3.2 Finite Element Simulation

The most relevant parts of the machine, the grinding wheel and the spindle, form
the domain of the finite element model. Since their deformation is small during
the grinding process a linear elastic material law is used. In addition, contact
constraints have to be included in the model because the displacements of the
grinding wheel are restricted by the workpiece. The dynamic behavior of the ma-
chine has to be taken into account, too. The friction between grinding wheel and
workpiece is described by Coulomb’s law. Therefore, an appropriate model is giv-

en by a dynamic Signorini problem.
Besides the dynamics, thermal effects are also considered. Thus, a thermo-
mechanical contact problem has to be solved, where the heat distribution and the
displacement are unknown and coupled. The heat distribution is described by the
heat equation. On the one hand, the displacement and the heat are interrelated by
the thermo-elastic material law. On the other hand, heat is generated by the fric-
tional contact between workpiece and grinding wheel. Mixed boundary conditions
are used to model the heat transfer to the coolant. The mathematical formulation
of this problem can be found in [15, 16].
The discretization of the dynamic Signorini problem can be carried out either
by finite difference schemes such as the Newmark method in time and finite ele-
ments in space [16] or by finite elements in space and time [17, 18]. Both methods
are implemented in the finite element library SOFAR [19], which is the platform
used for the calculations described here. The discretization of the heat equation is
described in [15, 16].
Precise knowledge of the workpiece surface is essential for the simulation. The
workpiece surface is discretized by the nailblock model of the geometric-
kinematical simulation. The finite element simulation has to evaluate this descrip-
tion of the surface at many different points. The nails lying in the possible contact
area are taken into account in the finite element simulation. Since the coordinates
of the nodes of the finite element mesh and of the nails are not identical, the eval-
uation has to be carried out via interpolation. Furthermore, a finite element mesh
of the workpiece has to be constructed in every time step in order to calculate the
heat, see [15].
Two approaches seem to be appropriate for calculating the normal and tangen-
tial contact stress. The first one is to calculate the normal and tangential contact
stress based on the displacement u in a post-processing step. The second possibili-
ty is to rewrite the variational inequality as a mixed problem. Here, the normal and
tangential contact stress is only the Lagrangian multiplier in the mixed problem.
Thus, it is calculated in combination with the displacements during the solution
process. The second formulation is preferred because a better accuracy is
achieved. The tangential contact stress is the value essential for calculating the
heat generated, which is proportional to the product of tangential stress and tan-
gential velocity.
In Fig. 6.9, the finite element model and a drawing of the spindle and the grind-
ing wheel are displayed. In the finite element model, the geometry has been
slightly simplified and is discretized by hexahedral elements. The material is li-
near elastic in the whole domain. However, the material parameters vary. The
areas where the different material parameters apply are indicated in the finite ele-
ment model found in Fig. 6.9. The first region consists of the spindle and the car-
rier of the grinding wheel. The second one contains the grinding layer. The third
region relates to the bearings. The material parameters of the bearings are used to
model the stiffness of the rest of the grinding machine. This approach is explained
in [20].
Fig. 6.9 Drawing and finite element model of the spindle and the grinding wheel
The algorithm for solving the discrete problems is complex. An extensive pres-
entation of this theme can be found in [16]. It includes an outer fix-point algo-
rithm, where a quadratic program with nonlinear constraints has to be solved in
each iteration. Therefore, a sequential quadratic programming (SQP) method is
applied. The rotation of the grinding wheel complicates the calculation. An arbi-
trary Lagrangian Eulerian (ALE) approach is used to take the rotational effects in-
to account. This is also described in [16].
To discretize the thermo-elastic problem of the workpiece two points have to be
considered. The first one is that the basic domain changes during the calculation
due to the material removal. Consequently, the initial data of a new time step must
be interpolated on the new mesh. The second point is directly related to this. A
new finite element mesh has to be constructed in each time step. This is carried
out by meshing the dexel model of the workpiece stored in the geometric-
kinematical simulation.
The finite element simulation is complex and therefore time consuming. A
posteriori error estimates have been derived and adaptive finite element methods
have been developed to achieve a given accuracy at minimal numerical costs. Two
different approaches are considered: In the first one, the error is estimated in the
global energy norm [21-23]. This approach is easy to implement and leads to good
meshes but the error is not measured in the relevant physical quantities. This dis-
advantage has been overcome by dual-weighted residual (DWR) based error esti-
mators, see [18, 24]. More details can be found in Section 3.4.
6.3.3 Coupling
The target depth of cut ae,tar is given by the NC-data. The machine, however, is not
able to realize this depth of cut. Instead, an actual depth of cut ae,act is measured,
which is smaller than the target depth of cut. The geometric-kinematical simula-
tion is only able to simulate the target depth of cut. It is combined with the finite
element simulation to obtain a better approximation of ae,act.
The process force Fres is predicted by the geometric-kinematical simulation and
the contact force Fcon is calculated by the finite element simulation. A good ap-
proximation of the actual depth of cut is found, if the contact force and the pre-
dicted process force are equal. The problem is to find such an approximation of
the actual depth of cut. The solution algorithm is described in this section.
The accumulated contact force Fcon is calculated from the actual normal contact
stress by integration. The contact force Fcon is a function of the actual depth of cut.
We minimize the deviation between Fres and Fcon measured in the 2-norm squared
as dependent on the actual depth of cut by using the golden cut algorithm. The op-
timal point is then the actual depth of cut sought. For an alternative approach and
more details, see [9].
6.3.4 Parametrization
Many of the model parameters, the frictional resistance coefficient and the heat
distribution factors, for instance, cannot directly be determined by experiments.
However, they must be accurately defined to ensure precise simulation results.
Numerical parameter identification techniques are used to calculate the model pa-
rameters and to overcome the difficulties related to direct measurement.
Based on an experimental design, several suitably chosen experiments are car-
ried out. Some quantities, which can be determined with small measurement error
and calculated by the simulation, are measured. The experiments are reproduced
numerically in the simulation, where some initial guesses, e. g. from literature, are
used for the parameters. The difference between the simulation results and the
measured results is calculated in an appropriate norm. Then, optimization algo-
rithms are applied to minimize the deviation, where the parameters are the optimi-
zation variables. The optimal solution then gives the parameters sought.
This technique has been used to determine the damping parameters and the
stiffness of the bearings. The experiments are described in [2] and the parameter
identification in [20]. For the determination of the parameters connected to the
model of the thermal effects, see [8, 15].
6.4 Simulation Results

During the past few years, a complex simulation has been created. It has been
tested and validated step by step, i. e. every part was independently considered.
Three major results are discussed here. The first one is the validation of the re-
moval model. To this end, an actually ground workpiece surface was compared
with the surface predicted by the simulation. The results of the frictional model are
substantiated by the comparison of the measured tangential forces during an actual
grinding process and the accumulated tangential stresses in the simulation. The
third aspect relates to the simulation of thermal effects. The temperatures calcu-
lated for some chosen points of the workpiece were checked against the tempera-
tures measured by thermocouples at the same points.
6.4.1 Validation of the Removal

The simulation presented here is tested by means of a simple grinding process.
The cutting speed was set at vw = 50 m/s and the workpiece velocity was
vf = 30 m/min. Thus, the grinding wheel rotated with n = 85 s-1. The target depth
of cut was set at ae = 40 µm. A non-water-based coolant was used. The process
kinematics was down-grinding.
The simulation was carried out as described above. The termination tolerance
was set at tol = 10 N. This seems to lead to a rough approximation. Experience
with the simulation, however, has shown that a smaller tolerance changes the val-
ues of the actual depth of cut by only about ae,act = 0.001 µm. Thus, the tolerance
is accurate enough in relation to the other model errors. Three to four passes of the
simulation cycle are needed to reach the given prescribed tolerance.
In Fig. 6.10, a cut through the workpiece is displayed. The measured result is
also provided for comparison. Temperature effects are neglected in this calcula-
tion. One can easily see that the results are good for the quasi-stationary process
phase. However, to get an accurate result during the run-in damping has to be con-
sidered in the simulation. For more details, see [2].
Fig. 6.10 The simulated and the actual workpiece surface
6.4.2 Validation of the Frictional Model

To validate the frictional model the accumulated tangential forces are calculated in
the simulation and compared to the tangential forces actually measured. In
Fig. 6.11, the measured forces and the simulated forces with a constant frictional
coefficient and a time-dependent one are shown. In the quasi-stationary area, both
approaches lead to good results. However, during the run-in and the run-out the
results using the time-dependent frictional coefficient are more accurate.
Fig. 6.11 Comparison of the tangential forces using the force model and the FEM based on
the frictional model
6.4.3 Validation of the Temperature Results

In this section, the temperatures measured in an experiment are compared with the
results of the corresponding simulation. A workpiece made of X210Cr12 is used.
The width is 19.7 mm, the length 80 mm and the height 22.3 mm. The depth of cut
is ae of 1 mm. The material removal is carried out line-by-line, where the width of
cut ap is 0.1 mm. The cutting speed vc is set at 50 m/s and the workpiece velocity
vw at 1,500 mm/min. The temperature of the coolant is 26°C and therefore
θs = 26 °C is used in the simulation.
In Fig. 6.12, the temperature measured by a thermocouple is compared to the
temperature calculated by the simulation at the same point. There, several passes
of the grinding wheel are examined. It is obvious that the time interval between
the different temperature peaks has been correctly calculated by the simulation.
Furthermore, the deviation of the calculated temperature from the measured one is
small. However, sometimes the simulated maximum temperature is larger than the
measured one and sometimes it is smaller. The overall development was predicted
correctly.
Only one pass is depicted in Fig. 6.13. There, we see that the qualitative trend
of the curve for the temperatures measured is reproduced by the one simulated.
The temperature decay was not predicted accurately, the temperature decay is too
slow in the simulation. However, the final, resulting temperature is correct.
Fig. 6.12 Comparison of measured and simulated temperature results
Fig. 6.13 Comparison of measurement and simulation results for one pass
6.5 Application of Simulation

In order to achieve an overall view of the simulation and the prediction of the NC
data it is necessary, on the one hand, to transfer the simulation to other grinding
processes and materials and, on the other hand, to integrate tool path planning into
the simulation. In the following sections, the present state of path planning and the
first experimental results using other grinding wheel contours will be discussed
6.5.1 Path Planning

The simulation described here was developed in order to produce, in future, an in-
tact workpiece surface and a workpiece with a high degree of dimensional accura-
cy. For this purpose, path planning is integrated into the simulation. In the first
step, the mean spindle displacement, when flute grinding, is examined. When pro-
ducing the flutes, the depth of cut, which does not change over the grinding path,
is adjusted as long as it takes for the mean actual depth of cut to reach the target
depth of cut along the grinding path. It can be seen that the simulation with path
planning over a grinding path having constant engagement conditions attains the
mean target cut of depth (Fig. 6.14). The positioning of the workpiece in the
machine has to be examined and the traverse velocities of the different axes taken
into account when path planning in order to integrate the path planning into
the simulation for the changing engagement conditions and to realize the path
planning for every time step.
Fig. 6.14 Flute grinding with and without path planning
6.5.2 Transferability to Other Processes

The simulation must be transferable to other grinding wheels and materials so that
a holistic simulation can be produced. First, the different grinding processes are
realized on the basis of different grinding wheel contours. In addition to a toroid
grinding wheel, a cylindrical and a profile grinding wheel are used. These grinding
wheels must be integrated into the simulation and analyzed to see whether the si-
mulation is suitable. In order to calibrate the simulation force measurement must
be undertaken once more to produce a force model and temperature measurements
have to be made for the different grinding wheels. The different engagement con-
ditions for the grinding wheels in the geometric-kinematical simulation are shown
in Fig. 6.15.
Fig. 6.15 Engagement conditions using different grinding wheels in the geometric-
kinematical simulation
The forces for different process parameters are measured. Based on the mean
value of the forces measured, the force coefficients at every workpiece speed for
all grinding wheel contours could be determined. The simulated forces are in good
agreement with the measured forces when flat surfaces are produced using cylin-
drical grinding wheels as well as when slanted surfaces are produced using a pro-
file grinding wheel. Comparing the forces, it becomes obvious that there is a great
variance in the forces in the area of the quickly-changing contact surfaces when
concave surfaces are produced. Here, only small depths of cut can be achieved us-
ing cylindrical and profile grinding wheels (Fig. 6.16). Therefore, it is impossible
to predict the forces in the simulation. The development of forces along the sur-
face contours can be simulated using toroid grinding wheels because, due to the
relatively large depth of cut, the contact surfaces are not so greatly changed as
they would be for the depths of cut used with cylindrical and profile grinding
wheels.
Fig. 6.16 Comparison of the measured and simulated forces along a grinding path using a
cylindrical grinding wheel for a concave workpiece
One can see that when the temperature in the process is measured the tempera-
tures behave as described in literature and in Sect. 6.2.3 when using both grinding
wheel contours. An example is shown in Fig. 6.17.
A further analysis of the grinding processes is now in progress. On the one
hand, the temperatures measured using the different grinding wheel contours are
being compared with the temperatures calculated in the finite element simulation.
On the other hand, other workpieces are being investigated.
Fig. 6.17 Dependency of temperature on the distance between the thermocouple and the
contact area using a profile grinding wheel

In this chapter, the experimental basis of the empirical models for NC-shape
grinding process are presented, the models themselves and also the numerical me-
thods for calculating approximate solutions have been described. The results of
three validation experiments presented here have shown that our method for simu-
lating the NC shape grinding process is an appropriate one. In particular, the iden-
tification of model parameters using numerical methods leads to the achievement
of acceptable results. Furthermore, first results concerning path planning show that
the simulation approach is applicable for optimizing the process. The demonstrat-
ed transferability of the simulation approach to other grinding processes reveals
the flexibility of this approach.
The application of the simulation presented here for optimizing grinding
processes will be extended. As an example, complex automatic modifications of
the NC data should be mentioned. One aim is to damp out oscillations in the ma-
chine structure but the restrictions of the grinding machine, which influence a rea-
lization of the NC data turn this into a difficult task. For this purpose, a lot of si-
mulation runs are needed, where the accuracy requirements vary strongly from a
rough approximation of the removal to a precise prediction of the temperature dis-
tribution. Consequently, a wide range of simulation options has to be provided
and the computing time minimized. It is planned to employ adaptive finite
element methods and modern high performance computing techniques to meet
these demands.
Considering the practical usage of this simulation outside academic research,
the key issue is the transferability of the approach to other grinding machines, oth-
er workpieces and other grinding processes. The results presented here concerning
this topic will be extended by conducting further experiments, which will mainly
concentrate on new workpiece materials.
References
[1] DIN EN 8589-11. Schleifen mit rotierendem Werkzeug - Einordnung, Unterteilung,
Begriffe. Beuth Verlag (2003)
[2] Jansen, T.: Entwicklung einer Simulation für den NC-Formschleifprozess mit To-
russchleifscheiben. Dissertation, Technische Universität Dortmund. Vulkan Verlag,
Essen (2007)
[3] Biermann, D., Blum, H., Jansen, T., Rademacher, A., Scheidler, A., Weinert, K.: Ex-
perimental analyses to develop models for NC-shape grinding with a toroid grinding
wheel. In: Denkena, B. (ed.) The 1st International Conference on Process Machine In-
teractions, Hannover, Germany, pp. 279–287, 3.9–4.9 (2008)
[4] Brecher, C., Esser, M., Witt, S.: Interaction of manufacturing process and machine
tool. CIRP Annals 58(2), 588–607 (2009)
[5] Brinksmeier, E., Aurich, J.C., Govekar, E., Heinzel, C., Hoffmeister, H.-W., Peters,
J., Rentsch, R., Stephenson, D.J., Uhlmann, E., Weinert, K., Wittmann, M.: Advances
in modelling and simulation of grinding processes. CIRP Annals 55(2), 667–696
(2006)
[6] Klocke, F.: Manufacturing Processes 2: Grinding, Honing, Lapping. Springer, Berlin
(2009)
[7] Tönshoff, H.K., Peters, J., Inasaki, I., Paul, T.: Modelling and simulation of grinding
processes. CIRP Annals 41(2), 677–688 (1992)
[8] Biermann, D., Blum, H., Rademacher, A., Scheidler, A.V.: Simulation of thermal ef-
fects in NC-shape grinding of free formed surfaces using toroid grinding wheels. Part
I: Experimental results. In: Proceedings of the CIRP 2nd International Conference
Process Machine Interactions, Vancouver, BC, Canada, pp. 10.6–11.6 (2010); digital
published
[9] Weinert, K., Blum, H., Jansen, T., Rademacher, A.: Simulation based optimization of
the NC-shape grinding prozess with toroid grinding wheels. Production Engineering –
Research and Development 1(3), 245–252 (2007)
[10] Stautner, M.: Simulation und Optimierung der mehrachsigen Fräsbearbeitung. Tech-
nische Universität Dortmund. Vulkan Verlag, Essen (2006)
process machine interaction in grinding. Production Engineering – Research and De-
velopment 3(1), 111–120 (2009)
[12] Biermann, D., Mohn, T.: A geometric-kinematical approach for the simulation of
complex grinding processes. In: CIRP International Conference on Intelligent Com-
putation in Manufacturing Engineering, Innovation and Cognitive Production Tech-
nology and Systems, Naples, Italy (2008)
[13] Weinert, K., Blum, H., Jansen, T., Mohn, T., Rademacher, A.: Angepasste Simula-
tionstechnik zur Analyse NC-gesteuerter Formschleifprozesse. ZWF, Zeitschrift für
wirtschaftlichen Fabrikbetrieb 101(7/8), 422–425 (2006)
[14] Kienzle, O.: Einfluss der Wärmebehandlung von Stählen auf die Hauptschnittkraft
beim Drehen. Stahl und Eisen 74, 530–551 (1954)
[15] Biermann, D., Blum, H., Rademacher, A., Scheidler, A.V.: Simulation of thermal ef-
fects in NC-shape grinding of free formed surfaces using toroid grinding wheels. Part
II: Modeling and FE-discretization. In: Proceedings of the CIRP 2nd International
Conference Process Machine Interactions, Vancouver, BC, Canada, pp. 10.6–11.6
(2010); digital published
[16] Blum, H., Kleemann, H., Rademacher, A., Schröder, A.: On solving frictional contact
problems part II: Dynamic case. Ergebnisberichte Angewandte Mathematik 377,
Technische Universität Dortmund (2008)
[17] Blum, H., Jansen, T., Rademacher, A., Weinert, K.: Finite elements in space and time
for dynamic contact problems. International Journal for Numerical Methods in Engi-
neering 76, 1632–1644 (2008)
[18] Rademacher, A.: Adaptive Finite Element Methods for Nonlinear Hyperbolic Prob-
lems of Second Order. Dissertation Technische Universität Dortmund. Verlag Dr.
Hut, München (2009)
[19] Blum, H., Kleemann, H., Rademacher, A., Schröder, A., Wiedemann, S.: SOFAR:
Scientific object oriented finite element library for application and research. Technic-
al report, Technische Universität Dortmund (2002),
http://www.mathematik.uni-dortmund.de/lsx/research/
software/sofar/index.html
[20] Biermann, D., Blum, H., Rademacher, A., Schäckelhoff, M., Scheidler, A.V., Wei-
nert, K.: Bestimmung der Materialparameter des Spindel-Schleifscheiben-Systems
mittels numerischer Parameteridentifikation. Ergebnisberichte Angewandte Mathe-
matik 376T, Technische Universität Dortmund (2008)
[21] Biermann, D., Blum, H., Jansen, T., Rademacher, A., Scheidler, A.V., Schröder, A.,
Weinert, K.: Space adaptive finite element methods for dynamic Signorini problems
in the simulation of the NC-shape grinding process. In: Denkena, B. (ed.) 1st Interna-
tional Conference on Process Machine Interactions, Hannover, Germany, pp. 309–
316, 3.9.–4.9 (2008)
[22] Blum, H., Rademacher, A., Schröder, A.: Space adaptive finite element methods for
dynamic Signorini problems. Electronic Transactions on Numerical Analysis 32,
162–172 (2008)
[23] Blum, H., Rademacher, A., Schröder, A.: Space adaptive finite element methods for
dynamic Signorini problems. Computational Mechanics 44(4), 481–491 (2009)
[24] Rademacher, A., Schröder, A.: Goal-oriented error control in adaptive mixed FEM
for Signorini’s problem. Computer Methods in Applied Mechanics and Engineer-
ing 200(1-4), 345–355 (2011)
Chapter 7
Modeling of Process Machine Interactions
in Tool Grinding
M. Deichmueller, B. Denkena, K. M. de Payrebrune,

M. Kröger, S. Wiedemann, A. Schröder, and C. Carstensen
Abstract. A systematic modeling approach to predict and manipulate the static

and dynamic process machine interactions in tool grinding is described. The mod-
eling approach is verified by experimental investigations gained by means of an
industrial tool grinding machine and separate test stands. It combines models of
the static and dynamic behavior of the grinding machine and its components with
a microscopic grinding process model. Material removal algorithms are applied to
cope with the changing shape and changing mechanical properties of the work-
piece during grinding. The interaction model has been applied in the process
planning phase to optimize tool paths and process parameters in order to reduce
resulting shape errors in ground tools.
7.1 Introduction
High performance cutting tools play an important role in modern manufacturing.

Within the production of an aircraft, for example, high-quality parts have to be
manufactured containing up to several thousand holes with low tolerances. Parts
produced for the car industry are also mostly high quality parts. In order to meet
the high quality requirements, even for high lot sizes, cutting tools made of
tungsten carbide are used, which is a very hard and strong material. In addition, a
high effort is spent on machine tests within the process ramp-up phase with the in-
tention of ensuring a fast and stable process. The worldwide market for cutting
tools has a volume of about 13 billion € [1], which also shows the significance of
the industry.
Due to the hardness of tungsten carbide cutting tools are manufactured by
grinding processes. As summarized in part I of this book, many research projects
have focused on understanding and controlling grinding processes [2]. In recent
years, the focus of research was widened to investigate the influence of the grind-
ing machine to the grinding process, i.e. the process machine interaction. In the
grinding of cylindrical cutting tools like drills or end-mills, the workpiece plays an
important role regarding the process machine interactions. The relatively low
stiffness of the workpiece – which also changes considerably during grinding –
leads to static deflections and vibrations and therefore to geometry errors on the
ground tool. In order to understand and predict the process machine interaction in
144 M. Deichmueller et al.
tool grinding it is essential to model both the generation of forces and the genera-
tion of deflections.
In the future, the process machine interaction should be considered in the plan-
ning phase of tool grinding processes to allow the fabrication of high quality tools
with low tolerances and without extensive prototype grinding. Within an interdis-
ciplinary research project of production engineers, mechanical engineers and ma-
thematicians, the authors developed a multi-scale simulation approach, which
represents the process-machine interactions in tool grinding in the microscopic
and the macroscopic scale.
Tungsten carbide tools are produced using deep grinding processes. Within one
production step, the flutes are ground, which can be up to several millimeters
deep, see Figure 7.1.
Fig. 7.1 Experimental setup for tool grinding
The occurring process forces lead to large deformations of the workpiece.

Therefore, it is modeled by an elastic beam model with changing moments of
inertia. In addition, the mechanical properties and dynamic transfer behavior of
workpiece clamping, grinding wheel-spindle-system and machine structure are
considered in the boundary conditions. Another significant issue is the exact mod-
eling of the occurring grinding forces. The stress distribution and, with this, the
grinding forces are dependent on the present process parameters, such as grinding
velocity, infeed and the shape of the contact area between grinding wheel and
workpiece. The challenge in predicting the grinding forces in tool grinding is that
process parameters, such as contact length or depth of cut, are not constant within
the contact area and change during grinding.
7.2 Modeling Approach

The modeling approach, which is presented in this paper, has been developed in
an interdisciplinary collaboration of the Institute of Production Engineering and
Machines Tools (IFW, Leibniz Universität Hannover), the Institute for Machine
Elements, Design and Manufacturing (IMKF, TU Bergakademie Freiberg) and the
7 Modeling of Process Machine Interactions in Tool Grinding 145
Department of Mathematics (IfM, Humboldt-Universität zu Berlin). It combines a

kinematic-based discrete material removal simulation with a detailed analysis of
the current contact area. With this, the current macroscopic shape of workpiece
and contact area can be computed considering local contact lengths and cutting
depths. Based on the contact analysis, a 3D finite element model for single and
multi-grain scratches is parameterized, which allows the calculation of the stress
distribution during the microscopic grain engagements. The microscopic grinding
force model is also the basis for the dynamic interaction model, which has been
developed to predict instable grinding conditions and consider the effect of ma-
chine and workpiece vibrations on the resulting grinding process. The modeling
approach is consequently validated by experimental data gained by means of an
industrial tool grinding machine and with tool grinding tests. The interrelation of
the macroscopically and microscopically-based models is illustrated in Figure 7.2.
Fig. 7.2 Modeling concept for the simulation of the process machine interactions in tool
grinding, interfaces and responsible research institutes
7.3 Modeling of the Machine Structure

Tool grinding machines are flexible machines with a high degree of freedom and
normally five machine axes. For experimental analysis of the process machine in-
teractions in tool grinding, a Walter Helitronic Power 5-axis tool grinding machine
is used. Via two axes, the spindle and grinding wheel can be positioned axially in
y and z-direction, whereas the position of the machine table can be defined
over one axial (x-direction) and two rotation axes (A and C), as illustrated in
Figure 7.3.
During the grinding process, forces occur, which deform the machine structure.
Since this has an effect on the quality of the produced workpiece, analyses were
carried out to characterize the deformation of the clamping of workpiece and
grinding wheel and the machine structure.
Fig. 7.3 Axis configuration of the used 5-axis CNC tool grinding machine (Walter Helitron-
ic Power)
7.3.1 Workpiece Clamping

The workpiece is fixed on the machine table via a jaw chuck. During the manufac-
turing process, the workpiece rotates around its center line and moves towards the
grinding wheel to realize a helical flute. Thereby, the out-of-roundness and the de-
formation of the clamping have an impact on the final geometry. Both effects can
be measured by tactile sensors during grinding tests. Therefore, an additional mea-
suring pin is added on the workpiece rod.
At the beginning of the grinding tests, the workpiece rotates 5 times before the
grinding wheel gets in contact. During grinding, the deformations in y and z-
direction of the workpiece are measured at the cylindrical pin at the tip of the
workpiece, as shown in Figure 7.4. Additionally, the corresponding process forces
in y-direction are displayed as well, see also [3].
In the first 30 seconds, the out-of-roundness of about 37 µm is noticeable in
this case. Comparing the amplitude of the sine waves with the largest deformation
of the workpiece of 315 µm, the deviation in form and position by the out-of-
roundness amounts to 12 % and cannot be neglected. In the workpiece model, the
out-of-roundness is thereby realized by an eccentric pre-positioning of the work-
piece, which 100 mm long with a diameter of 10 mm.
Fig. 7.4. Measurement of workpiece deformation, process forces and out-of-roundness of

the workpiece tip
7.3.2 Machine Table

In a separate test, the elasticity of the machine table is analyzed. To measure the
deformation of the clamping and the workpiece table a constant force F = 100 N is
loaded onto the workpiece and the deformation is measured at several positions.
The comparison of the measured data and the analytical solution for the static
bending of the workpiece rod with ideal stiff boundary conditions results in a li-
near difference, which can be assigned to the deformation of the machine
structure. From Figure 7.5 it can be seen that the calculated difference between
measured and calculated deflection curve of a workpiece (10 mm diameter, 100
mm length) amounts to Δy = 11.8 µm and the tilt to Δφ = 1.1 10-³ degrees at the
support.
Fig. 7.5 Measured deformation of workpiece and machine table for a constant load of
F = 100 N
Since the stiffness of the clamping support has an influence on the total dis-
placement of the workpiece, the elasticity of the machine table is implemented in-
to the workpiece model. Therefore, two spring elements are considered in the
model with a distance of 40 mm within the clamping, see also [4, 5].
7.3.3 Grinding Wheel Clamping

The elasticity of the spindle is measured in the same way as the deformation of the
machine table, see Figure 7.5. A constant force is loaded on the spindle and the
deformation is measured at different points. The stiffness of the spindle is detected
to cSpindle = 11.8 N/µm, which is in the same range of the stiffness of the clamping
and about 30 times higher than the stiffness of the workpiece with an elasticity of
about cWorkpiece = 0.4 N/µm measured at the tip of the workpiece. The deformation
of the spindle influences the contact conditions during the grinding process, there-
fore the elasticity has to be considered in the model, especially for modeling the
dynamic movement of the spindle during grinding.
7.3.4 Grinding Wheel

To manufacture super-hard cutting materials like high speed steels or tungsten
carbide, grinding wheels with diamond or CBN-grains are used. The abrasive
grains are placed in a bonding, where additives are implemented to improve the
heat conductivity or to enlarge the chip space for better chip removal and coolant
supply. Before grinding, the grinding wheel has to be trued and sharpened. With
the truing process, the macroscopic shape of the grinding wheel is defined. During
sharpening, the bond is set back until sharp grains appear on the surface of the
grinding wheel and sufficient chip space is created. This conditioning process is
necessary because the grinding wheel wears over time meaning that the abrasive
grains blunts or break out and the shape of the wheel changes.
Fig. 7.6 Measured surface and Fast Fourier analysis of a grinding wheel topography with
graining D91
The contact between grinding wheel and workpiece is characterized by thou-

sands of stochastically microscopic grain engagements. These grain contacts dy-
namically excite the system, which can cause chatter or lifting and has a negative
effect on the quality of the ground surface. The corrugation of the grinding wheel
surface and the excitation frequency are mainly dependent on the rotation speed
2πn. Therefore, the shape of a nominally cylindrical grinding wheel is measured
by a laser sensor and analyzed. The measured data are shown in Figure 7.6. It is
obvious that the grinding wheel surface is wavy with different frequencies, as
shown by the Fourier transformation. The fundamental oscillation is due to an out-
of-roundness of the wheel with the basic frequency n of the rotation speed whe-
reas the harmonics are based on additional waves. The high-frequency noise cor-
responds to the abrasive grains. Since the size of the grains, which overlap over
the bonding, is nearly normally distributed, as references and analysis have shown,
it can be described mathematically, [6]. To represent the high-frequency excitation
of the workpiece in the dynamic simulation the grinding wheel is modeled by a
combination of sine waves for the out-of-roundness and the periodic waves, and
additionally by a normal distribution for the grains. The amplitudes are taken from
measurements, see also [3].
7.3.5 Workpiece Model

The workpiece is considered to be one part of the whole structure which interacts
with the tool grinding process. The modeling approach for the workpiece consists
of two parts. The first part is responsible for modeling the current macroscopic
shape of the workpiece. The workpiece geometry and material removal are discre-
tized using a so-called dexel model. The second part describes the static and dy-
namic deflection of the workpiece deformation. Here, the Euler beam equation
and a finite element approach with beam elements, respectively, are applied. Both
parts of the workpiece model are described in detail below.
7.3.5.1 Dexel Model
In 1986, van Hook introduced “dexels” or “depth elements” as a discretization

method for workpieces [7]. Since then, dexels have been applied by several au-
thors for the simulation of manufacturing processes, e. g. [8, 9]. The workpiece is
discretized by a regular grid of dexels, which have a constant distance in two di-
rections and varying lengths in the third direction. The accuracy of the model is
high in dexel direction and relatively low perpendicular to it. The use of three in-
dependent dexel grids overcomes the problem of different model accuracy in the
coordinate directions. Figure 7.7 shows the discretization of the workpiece with
dexel grids in all three coordinate directions. A dexel Di,j can be identified by the
position i,j in the particular dexel grid. Undercuts can be simulated by adding a
dexel segment Di,j,d to the current dexel.
Fig. 7.7 Workpiece discretization with dexels
The cylindrical grinding wheel is approximated as a non-rotating polyhedron in

the easiest case. The movement of the grinding wheel is determined by a
NC-program, which can be interpreted by the simulation system. During one si-
mulation step, the grinding wheel moves from one position to the next in a step-
wise linear movement. The trace of the grinding wheel is called the sweep
volume. In each simulation step, the intersection of sweep volume and workpiece
is calculated. In case of an intersection, the specific dexels are shortened, divided
or removed [10].
7.3.5.2 Static Deformation
In each simulation step, the current workpiece’s cross sections are analyzed for
calculating the geometrical moments of inertia Iyy, Izz and Iyz at discrete positions xi
along the workpiece axis x. For example, the geometrical moment of inertia Izz can
be calculated with
bD ⋅ h3
IzzD = D + y2 ⋅ b ⋅ h ,
12 D D D (1)
I zz = ΣI zzD
as displayed in Figure 7.8.

Fig.7.8 Calculation of geometrical moments of inertia
The static oblique bending of the workpiece is calculated by integrating the li-
near beam equations
−I yz (x)M y (x) + I yy M z (x)

v ''(x) = ,
E(I yy I zz − I 2yz )
(2)
−I zz (x)M y (x) + I yz M z (x)
w ''(x) =
E(I yy Izz − I2yz )
in every simulation step, where v is the deformation in y-direction and w in

z-direction. The material (tungsten carbide) is assumed to be isotropic and homo-
geneous, with Young’s modulus E being 590,000 N/mm². In addition to the geo-
metrical moments of inertia I(x), the bending moments M are a function of
position x and time t. To determine M(x) a grinding force model is applied, which
is described in Section 7.4.2. In case the clamping is assumed to be rigid, the
boundary conditions are v(0) = w(0) = 0 and v’(0) = w’(0) = 0 or otherwise de-
pendent on the stiffness of the support. After calculating the bending lines v(x) and
w(x) the deformation at position xi is assigned to all dexels of dexel layer i.
7.3.5.3 Dynamic Deformation
The dynamic workpiece deflection is highly dependent on the momentary geome-

try, stiffness and forces loaded on the workpiece. Mathematical and experimental
investigations have shown the dependencies of the eigenfrequencies on the length
 of the ground flute and depth of cut ae.. In Figure 7.9, the first eigenfrequency of
a workpiece with longitudinal flutes is shown exemplarily. The variation of the
workpiece characteristics is visible. A finite beam element approach is used as
workpiece discretization, according to the dexel model, to represent the changing
properties, compare [11].
Fig.7.9 Simulated workpiece frequency characteristics depending on the geometry for lon-
gitudinal ground flutes
Each beam element has six degrees of freedom to represent bending and torsion
of the workpiece axis. The properties are assumed to be constant within one ele-
ment but can change over time. The equation of motion can be written as
(3)
Mq + Dq + Kq = f ,
with the mass matrix M, the damping matrix D, the stiffness matrix K, the vector
of the generalized coordinates q and the vector of forces and moments f. Because
the system of coupled equations has the size of 6n × 6n, with n the number of
elements, it is very large and takes long to solve. By means of a modal reduction,
the system can be reduced and the equations become linearly independent, which
simplifies the solving process. The modal transformation of the system in the
modal space is carried out by a transformation matrix Φ = qˆ 1 ,qˆ 2 ,...,qˆ n  with
qˆ i the eigenvectors of the system. Via the substitution of the node displacements
q = Φqmod (4)
and the multiplication with the transformation matrix from the left side the equa-
tion of motion can be written in modal space as
ΦT MΦq mod + ΦT DΦq mod + ΦT KΦq mod = ΦT f

(5)
q  mod + q mod +  q mod = fmod
By using special eigenvectors the modal mass matrix  is equal to the identity
matrix; and the modal stiffness matrix has a diagonal shape
 = diag{ω12 , ω22 ,..., ωn2 } with the eigenvalues as entries. Depending on the num-
ber of considered eigenvectors, the size of the system is reduced in modal space;
however the information about the neglected modes is also lost.
The reduction of Equation (5) to a system of first order with the substitution
z1 = q mod and z 2 = q mod is
d  z1   0 E   z1   0 
  = − −    +  −1 
 −  −    z 2   f mod  ,
1 1
dt z2  (6)
z =  z + f (t ).
The solution of the dynamic deformation of the excited system is a combination of

a homogeneous and a particular solution. To apply the method of variation of con-
stants to Equation (6) a general solution is given by
t

z (t ) = et z (t0 ) + e (t −τ ) f(τ ) dτ .
0
(7)
To calculate the momentarily dynamic deformation the solution is determined

step-wise. This means that the solution for time z(ti+Δt) is determined depending
on the last solved step z(ti), the step size Δt and the momentary excitation of f(ti).
It is assumed that the momentary excitation f(ti) is constant within the small step
size Δt. The matrix  depends on the geometry of the workpiece and has been
recalculated frequently.
After solving Equation (7) in modal space, the solution of the deformation has
to be retransformed into the physical space by Equation (4).
7.4 Tool Grinding Process Model
In addition to the detailed mathematical description of the relevant parts of the

machine structure in Section 7.2, a model describing the tool grinding process it-
self is needed to predict the process machine interaction. Tool grinding is charac-
terized by contact conditions between grinding wheel and workpiece, which are
more complex than in conventional surface grinding. During the grinding of the
helical flutes, the workpiece is fed into the grinding wheel with a helical move-
ment. The shape of the grinding wheel in combination with the relative movement
of workpiece and grinding wheel generates the shape of the flute. The contact area
has a complex shape with a variation of contact lengths and cutting depths. The
basis for the grinding force modeling is a detailed geometrical analysis of the con-
tact area, which is described in the following.
7.4.1 Model-Based Contact Area Analysis

To build a grinding process model, which can calculate the three dimensional dis-
tribution of the grinding stresses and the resulting load in tool grinding, a geome-
trical analysis of the contact area is necessary. The analysis is based on the dexel
model described in 7.3.5.1, where each dexel represents one fraction of the total
workpiece volume. In each simulation step, dexels intersecting with the grinding
wheel are shortened or removed. A big change in dexel length represents a high
material removal rate and vice versa. By dividing the removed volume VD of each
dexel i by the simulation time step size Δt, the distribution of the material removal
rate Qwa in the contact area can be determined.
Qwa(i) = VD(i) / Δt. (8)

Figure 7.10 a) exemplarily shows the occurring external material removal rates
Qwa for a flute grinding process.
As described in [12], the external material removal rate Qwa(i) is used to deter-
mine the equivalent chip thickness heq for each volume element, see Figure 7.10
b). For the calculation, a constant cutting speed vc is assumed, which is reasonable
due to the cylindrical shape of the grinding wheel. With the presented procedure,
it is possible to determine equivalent grinding parameters like heq and Qwa for
every dexel cut during one simulation step. It can be applied for changing contact
geometries and an arbitrary grinding wheel movement. The contact area analysis
is the basis for the application of grinding force models, which are described in the
following section.
Fig. 7.10 a) Calculated material removal rate Qwa(i) [mm³/s] and b) calculated equivalent
chip thickness heq [µm] for a tool grinding process (vft = 30 mm/min, d = 10 mm,
R = 62.0 mm)
7.4.2 Grinding Force Model

Two approaches for calculating the grinding forces are shown. Both use the
equivalent grinding parameters as input data. The microscopic finite element ap-
proach calculates the stress resulting from a single grain – workpiece contact. The
second approach determines the grinding force distribution empirically.
7.4.2.1 Microscopic Force Model

Within the tool grinding process, the interaction of grains and workpiece on a mi-
croscopic scale result in forces observable on a macroscopic scale. These forces
are measured under the machine table, see Figure 7.3, but experiments suggest
that the stresses vary throughout the contact zone. In order to include this varia-
tion, a process force model has to account the locally different contact conditions.
The contact zone is discretized into small sub-domains by the dexel model (see
7.2.5.1). The range of parameters for the engagement of single grains can be cal-
culated using the geometrical analysis explained in 7.3.1. and [12]. This analysis
also provides a method for the combination of the microscopic stresses stored in a
database (see Figure 7.2) to the stress distribution within the contact zone. Here,
the approach is to model the contact of grains and workpiece on a microscopic
scale for the different parameters of those sub-domains. This is carried out by
coupling variational inequalities describing the elasto-plastic behavior of the
workpiece and its contact with abrasive grains. Linear kinematic hardening is as-
sumed for the workpiece on microscopic scale denoted by Ω. This domain is fixed
on the free boundary ΓD; and the contact boundary is denoted by ΓC. For the case
of notation, only one time-step of the quasi-static problem with initial values set to
zero is considered here (for further details see [13]). The primal problem of elasto-
plasticity with linear kinematic hardening including contact is to find a displace-
ment field u and a plastic strain P such that
− div σ (u, P) = f in Ω,
σ (u, P) = ε (u) − P ,
u = 0 on Γ D ,
σ n (u, P ) = t0 on Γ N , (9)
u n ≤ g ; σ n ≤ 0; (u − g )σ n = 0 on Γ C ,
σ (u, P) − P ∈ ∂j (P).

Here, j (Q) := σ y (Q : Q)1/ 2 , is the non-differentiable part of the minimum plas-
Ω
tic work function with the yield stress σy > 0 in uni-axial tension, σ denotes the
stress,  is the elasticity tensor and  is the tensor describing the hardening. For
simplicity, the surface traction t0 and the outer volume force ƒ are assumed to be
zero from here on. M : N := MijNij denotes the usual scalar product of M and N.
The part of the boundary ΓN is assumed to be disjoint with ΓD and ΓC.
Since the unique analytic solution W := (u, P) for this partial differential equa-
tion and inclusion is generally unknown, a numerical approximation has to be
computed. For this purpose, the mathematical problem shown in (9) is reformu-
lated as a variational inequality; and then it is equivalently written as a saddle-
point problem introducing Lagrange multipliers. Moreover, the spaces, where the
solution is searched, are replaced by discrete ones. This results in the problem to
find Wh, λC,H, λp,h such that
a( Wh , Z h ) + ΓC
λC , H v h ·n + σ
Ω
y λh, p : Qh = 0
(10)
ΓC
( μC , H − λC , H )u h ·n +  σ y ( μ p ,h − λ p ,h ) : Q h ≤ 0
Ω
for all ( Z h , μC , H , μ p , h ) ∈ Wh × Λ C , H × Λ p , h . Here,
a( W, Z) :=  (ε (u ) − P ) : (ε ( v) − Q) + P : Q,
Ω
where n is the outer unit normal. It can be shown that the mathematical problem
shown in (10) has a unique solution (Wh, λC,H, λp,h). λC,H is the approximation of
the normal stress on the contact boundary, whereas λp,h is the approximation of
the plastic stress, cf. [14,15]. The finite dimensional space Wh is constructed on
hexahedral meshes. In order to ensure the uniqueness of λC,H, the discretization of
ΓC is carried out with a mesh-size H > h, see [16]. An error estimator presented in
Chapter 3 and [17] and an adaptive refinement algorithm allowing possible multi-
level hanging nodes, cf. [18,19], are used in order to improve the accuracy of the
discrete solution for a fixed number of degrees of freedom.
Another numerical approximation approach is the regularization of the varia-
tional problem obtained from Problem (9), which results in an energy minimization
problem over K := {v ∈ H 01 (Ω ) 3 v ( x )·n ( x ) ≤ g ( x ) on Γ C } of the smooth functional
1
 (Z) := a(Z, Z) + jδ (Z) (11)
2
where jδ ( Z) :=  σ y (Q : Q + δ 2 )1/2 , and δ the regularization parameter. In order to
Ω
use the Newton-Raphson method on the whole space V instead of K, the penalty
functional
1
2 ΓC
Ψ ( v ) := ( v·n − g ) 2+ (12)
is introduced. Here, ρ is the penalty parameter and (ƒ)+ := max(0,ƒ) denotes the
positive part of a real valued function ƒ. The functional Ψ is added to the energy
functional  ; and the finite element space Wh is used to compute a discrete ap-
proximation of Wρ,δ. The main drawback of this approach is its dependence on δ
and ρ, cf. [20, 21]. The normal component of the microscopic contact forces Fmicro
is computed from the Lagrange multiplier λc via
Fmicro =  λC . (13)
ΓC
If the regularized energy functional and the Newton-Raphson method are used, the
stress must be computed from u with Hooke’s law and the normal stress is inte-
grated over the contact boundary. Once parameterized, the computed values are
stored in the database and missing entries have to be interpolated or newly com-
puted. This parameterization of the microscopic force model is carried out via the
measurement of single grain scratches. Figure 7.11 shows the deformation com-
puted by the finite element approach with regularization after a single grain
scratch. The maximum infeed was set to 5 µm; and the grain moved on a circle
with a diameter of 150 mm. The red color indicates areas where the plowing effect
occurs, i.e. material piles up next to the scratch line. However, the resultant cutting
depth is lower than the nominal depth of cut due to the elastic deflection of tool
and workpiece.
Fig. 7.11 Resulting cutting depth of a simulated scratch test
Here, the results of main interest are the forces occurring during scratching.
Figure 7.12 shows the computed forces for different parameter configurations. The
change in the module of elasticity has the biggest influence on the forces.
This reduction of the module of elasticity could be interpreted as damage to the
material. This damage is likely to occur during the scratching itself (for an exam-
ple of brittle material, see [22]) or could be introduced by the pre-treatment of the
workpiece.
Fig. 7.12 Force during scratch test calculated with microscopic Finite Element simulation
A possible approach to explain macroscopic damage phenomena is based on

small cracks on the microscopic level [23]. Both discretizations of the elasto-
plastic material law introduced above can be extended to allow fissures. There-
fore, the faces between hexahedra in the finite element mesh can be doubled and
disconnected. In Figure 7.13, a small crack was introduced based on previously
computed stresses. This allows the simulation of crack growth along faces.
Fig. 7.13 Strain calculation by a Finite Element simulation with a priori crack
7.4.2.2 Empirical Force Model
To parameterize an empirical grinding force model, flat grinding experiments us-

ing a cylindrical grinding wheel and a rectangular workpiece have been carried out
on a Walter Helitronic Power tool grinding machine. The workpiece has been
tightly fixed to the machine table in order to minimize deformation for correct cor-
relation between grinding depth and force, compare Figure 7.14.
Fig. 7.14 Experimental setup to determine the influence of ae on grinding forces

A Kistler three-component dynamometer has been used to measure the grinding

forces for different depths of cut and feed speeds. As described in [24], the result-
ing forces for different depths of cut had to be related to geometrical parameters,
which can be evaluated both in the experiment, as described above, and in the 3D
dexel simulation. The factor heq/dlg has been found to suit this requirement well. It
relates the equivalent chip thickness heq to the local contact length dlg for each
segment of the contact area. This dimensionless value replaces the factors feed
speed, cutting speed and grinding wheel diameter, which are commonly used in
grinding force models [25]. Figure 7.15 shows the relationship between the me-
chanical stresses and heq/dlg.
Fig. 7.15 Relationship between mechanical stresses σn (solid line, ♦) and σt (dashed line,●)
and heq/dlg for a tool grinding process (vft = 15 - 50 mm/min, ae = 0.01 - 1mm, vc = 18 m/s)
The calculated stresses are fitted to a polynomial equation applying the least
squares method.
 h eq  h eq  2
σ n = 1.56 ⋅105 ⋅ + 9.92 ⋅109 ⋅    N

 d lg  d lg   mm²
   
(14)
 h eq  h eq 
2
σ t =  0.63 ⋅105 ⋅  
N
+ 3.43 ⋅109 ⋅ 
 d lg  d lg   mm²
   
This relationship is applicable for the parameter range shown in Figure 7.15. The
grinding force model has been implemented into the simulation system to visual-
ize the occurring stresses in tool grinding. Figure 7.16 shows the distribution of
the normal stresses in the contact area. It can be seen that the highest pressure acts
at the area with high equivalent chip thickness heq, cp. Figure 7.10 b).
Fig. 7.16 Calculated distribution of normal stress σn [N/mm²] for a tool grinding process
(vft = 30 mm/min, d = 10 mm, R = 62.0 mm)
7.4.3 Temperature in Contact Area

A main task in modeling of grinding processes is the determination of the generated
heat and its distribution into workpiece, grinding wheel, coolant, chips and envi-
ronment. Most approaches consider a 2D moving heat source model according to
Carslaw and Jaeger with a constant heat flux density throughout the contact area.
Applying this heat source model to a workpiece model using linear heat conduction
theory and assuming an adiabatic fixturing and an isotropic material model, it is
possible to calculate the transient temperature distribution inside the workpiece.
However, these approaches are limited to grinding processes with constant en-
gagement along the grinding wheel axis and relatively large workpieces where
losses to the environment can be neglected. To determine the heat flux density, one
has to know the generated heat and the area of contact. The heat flux into the work-
piece is dependent on the temperature acting in the contact area. Therefore, it is a
common task in research on grinding processes to determine the temperature in the
contact area, which is difficult, since the contact area is not easily accessible by
optical temperature measurement equipment like infrared cameras. Also, a high ef-
fort is needed to prepare the workpieces when using thermocouples being posi-
tioned in different positions inside the workpiece. With this indirect measurement
method the temperature in the contact area is calculated by extrapolating the tem-
peratures in different positions inside the workpiece with defined distance to the
contact area.
In tool grinding, where the workpieces are relatively small and the ground
flutes have a special concave shape it is not feasible to position the thermocouples
in a defined position inside the workpiece. The authors chose a measurement se-
tup, where temperature sensor and electronics are inside the grinding wheel, with
the advantage that the temperature in the contact area can be measured directly
with a measuring range of 150°C – 600°C, see Figure 7.17.
Fig. 7.17 a) Grinding Wheel made by FOS Messtechnik GmbH equipped with a tempera-
ture sensor b) electronics inside the grinding wheel with detector, batteries and wireless
equipment for sending the data to external receiver during grinding
Figure 7.18 shows measurement results using the sensor integrated grinding
wheel. A flute with a helical angle of 30° was ground into a tungsten carbide
workpiece with a diameter of 16 mm. It can be seen, that the temperature signal
(black line) rises to its maximum level of about 210°C when the grinding wheel is
entering the workpiece. The time delay of the temperature signal is due to the fact
that the measurement range of the temperature sensor starts at about 150°C. The
explanation for the higher signal level at the beginning of the measurement is that
the sensor is calibrated for the temperature of tungsten carbide. At the beginning,
there is no contact between grinding wheel and the tungsten carbide workpiece
and the signal is not valid. The process forces, which are measured in machine
tool coordinates (see Figure 7.3), and the measured temperature show similar cha-
racteristics: during the constant engagement conditions both, process forces and
temperatures stay constant. This characteristic implies that the process is thermo-
mechanically stable with constant heat flows due to the flood supply of coolant.
Fig. 7.18 Process forces and temperatures measured with the sensor integrated grinding
wheel (vf = 25 mm/min, vc = 18 m/s, workpiece diameter d = 16 mm)
With the help of the sensor integrated grinding wheel it is possible to analyze
the influence of the coolant supply on the grinding process. Figure 7.19 shows the
variation of the coolant jet velocity and the resulting temperature in the contact
and the ratio µ between tangential force Ft and normal force Fn. It can be seen,
that the lowest temperature and friction occurs when the coolant jet velocity is
equal to the cutting speed vc.
Fig. 7.19 Influence of velocity of coolant on temperature and grinding force ratio (process:
vft = 30 mm/min, vc = 18 m/s, workpiece diameter: 16 mm, helix angle: 30°, coolant: min-
eral oil)
7.5 Planning and Optimization of Tool Paths

With the presented models of machine structure, workpiece and tool grinding
process it is possible to calculate forces based on a material removal simulation
and a contact area analysis. For a prediction of the deflections occurring in tool
grinding it is necessary to couple both process model and structure model. Since a
deflection directly affects the contact conditions and therefore the acting grinding
forces, a coupling of process and structure model using an iterative algorithm is
required. The algorithm determines the deformation of workpiece and machine
structure during grinding at a static equilibrium of the grinding force and the
spring-back force of the workpiece until the residuum between the deformations
calculated in the last step and in the current step is smaller than a predefined value
ε. Then the material is removed, the actual geometry is defined and the following
calculation step is carried out.
Using the prescribed simulation method, a method has been developed to adapt
existing tool paths for reducing the effects of process machine interactions in tool
grinding. The procedure is described in Figure 7.20. Based on a simulation of the
grinding process using the original tool path, the workpiece deflection is calculated.
The workpiece deflection is then added to the tool path which means that the grind-
ing wheel cuts more material than before. The simulation and optimization is
repeated until the calculated workpiece cross sections are equal to the ideal cross
section which is determined by a simulation run with an ideally stiff workpiece.
Fig. 7.20 Method for optimizing tool paths in tool grinding
7.6 Application and Results
7.6.1 Simulation of Static Effects

The empirical grinding force model (compare Section 7.4.2.2) has been coupled
with the elastic workpiece model described in Section 7.3.5.2 to predict the grind-
ing forces in tool grinding considering the deflection of the workpiece. Figure 7.21
shows a comparison of simulated and measured grinding forces for the parameters
listed in the caption. The following effects are visible in the Figure: 1) The mean
value of the measured grinding forces matches the simulated forces very well
for the first flute, 2) The phases, where the grinding wheel enters and exits the
Fig. 7.21 Calculated (black lines) and measured grinding forces (oscillating lines) for the
grinding of two flutes of a tungsten carbide tool. (Grinding wheel: Q-Flute-2 D54 C75,
R = 62 mm, Workpiece: Tungsten Carbide, L = 100 mm, d = 10 mm, Process: Down
Grinding, vft = 30 mm/min, vc = 18m/s)
Fig. 7.22 Comparison of simulated cross-section using dexel-model and real cross-section
(photograph made with digital microscope); detail shows part of the contour with highest
deviation
workpiece, are predicted precisely by the simulation model, 3) The grinding forces
slightly rise during full-contact phase, both in simulation and measurement. This
effect is caused by the increasing stiffness of the workpiece when approaching the
workpiece clamping and the resulting higher actual depth of cut at the end of each
flute. 4) When grinding the second flute the simulated magnitude of Fy does not
exactly match the measured values.
Apart from the grinding forces, the resulting cross-section of the workpiece has
been simulated and compared with measurements. A Keyence optical microscope
was used to create a high resolution picture of the cross-section of the workpiece
tip, where the maximum shape error occurs. As can be seen in Figure 7.22, the dif-
ference between simulation and measurement is relatively low, which verifies the
high accuracy of the developed and implemented static process machine interac-
tion model. However, the cross-section resulting from grinding a flexible work-
piece is different from the ideal cross-section, which can be simulated using a ri-
gid workpiece, compare Figure 7.30.
7.6.2 Simulation of Dynamic Effects

The dynamic behavior of the workpiece and grinding wheel fixation as well as the
grinding wheel topography have a wide influence on the process forces and the
workpiece geometry. Due to the long computing time the dynamic effects are ana-
lyzed on geometrically simpler experiments, where a longitudinal flute is ground.
The workpiece has a diameter of 10 mm and a cantilevering length of 100 mm.
For the experiments and the simulation the adjusted depth of cut is 1 mm, the feed
speed is 100 mm/min and the cutting speed is set to 18 m/s. A flute of 30 mm
length is simulated with the introduced grinding process model; and the forces and
workpiece geometry are calculated.
7.6.2.1 Influence of the Grinding Wheel Topography
The wavy wheel surface causes fluctuations in the process forces due to varying
contact conditions. In Figure 7.23, calculated process forces with an ideal round
grinding wheel and with a grinding wheel including an out-of-roundness of
24 µm, a harmonic wave of 10th order with an amplitude of 1 µm and stochastical-
ly distributed grains of about 0.6 µm are shown. For the simulations, the dynamic
module from Section 7.3.5.3 is included. Additionally, the measured force signals
are also illustrated.
In the first few seconds, the contact length between grinding wheel and work-
piece increases as well as the mean value and the amplitude of the process forces
until the contact length gets quasi-constant. Comparing the mean values it is visi-
ble that the calculation with ideal round grinding wheel fits the measured data well
for the y-direction, whereas for the x-direction the forces are overestimated. The
calculation with the wavy grinding wheel shows a smoother change between run-
ning-in period and constant interval, similar to the measurements. For the y-
direction the forces are underestimated by about 10 N, whereas in x-direction the
simulation is in accordance with the measured data.
When comparing the amplitudes of calculated and measured forces, the high
fluctuation at the beginning of the experiments is not expressed by the simulation.
This is due to ideal initial conditions for the simulation. During the experiments,
Fig. 7.23 Calculated process forces Fx and Fy with an ideal round and a wavy grinding
wheel compared with measured data
influences of the coolant, for example, are also included in the measured force
signal, which causes larger amplitudes at the beginning of the process. For the
normal forces in y-direction, the increasing measured force amplitudes are also not
represented by the simulation but quasi-constant amplitudes are reached. This cha-
racteristic can also be related to the coolant, which has a higher influence in nor-
mal direction than in tangential direction. The accuracy could be increased by
means of additional analyses about the influence of coolant.
The consideration of the time histories of measured forces and the resulting
time-varying coefficient of friction show significant periodic fluctuations, which
depend on grinding wheel rotation, amplitude of the process parameters, feed, cut-
ting speed and feed rate. Worth mentioning is the measured phase shift between
the normal and tangential load, which is found in the measurements vary in severi-
ty and comparative measurements with other materials and fixations. As an exam-
ple, the relationship between the local friction coefficient and the acting normal
force FN is displayed for different depths of cut in Figure 7.24.
Fig. 7.24 Consideration of the timing behavior of the friction coefficient as a function of the
normal force and the depth of cut
7.6.2.2 Influence of the Dynamic Effects on Workpiece Geometry
When comparing the results of the workpiece geometry by including the dynamic
behavior of workpiece and grinding wheel, a difference in the geometry to the
ideal surface is visible, as shown in Figure 7.25.
This difference has its maximum at the tip of the workpiece, when the forces
cause the largest bending and decrease towards the clamping. When comparing
the surface with single measured data points, the calculated geometry error is un-
derestimated but shows qualitatively the same shape. The difference can be ex-
plained by the influence of the coolant, which is not represented in the simulation.
By considering an additionally measured coolant-force component of about 10 N,
as discussed in Section 7.6.2.1, the bending of the workpiece increases and the
geometry error gives a better approximation of the measured data points.
170 M. Deichmueller et aal.
Fig. 7.25 Calculated workpiiece geometry predicted by the dynamic module with and witth-
out coolant consideration
7.6.3 Simulation Stu

udies
The presented computatio on models have been implemented as a plugin into thhe
simulation system CutS, which
w has been developed at the IFW Hannover [8]. CuttS
is a universal framework for the simulation of machine movements and materiial
removal and for the determ mination of technological process characteristics of mann-
ufacturing processes. It iss possible to change the shapes and dimensions of workk-
piece and grinding wheel and to use different tool paths by loading NC-program ms
to the NC-control of CutS S. Since the developed calculation models are not bounnd
to one specific workpiecee geometry or process parameter combination, it is posssi-
ble to use CutS for simu ulation studies to gain more knowledge on tool grindinng
processes.
In order to determine suitable parameters such as dexel grid distance or simula-

tion time step, simulation studies on the influence of these parameters on the simu-
lation accuracy have been performed. Figure 7.26 shows the influence of the dexel
grid distance on the resulting cross-section area of the workpiece, measured along
the fluted part of the workpiece (mean value and min-max-bar). The higher the
number of dexels used for simulation, the lower is the grid distance and the better
is the simulation accuracy. It is noticeable that the mean value of the cross-section
area is nearly the same for all grid distances but there is a variation for larger grid
distances which can be up to 5 times the grid distance. The dexel grid distance
should be chosen smaller than 0.1 mm for achieving high quality simulation
results.
Fig. 7.26 Influence of dexel grid distance on simulation accuracy for calculating the work-
piece cross section of a ground two-flute twist drill (10 mm diameter, 30 ° helix angle)
Another important simulation parameter is the simulation time step, which is

the period of time represented by one simulation step. A lower simulation time
step leads to better results and longer computation times. Figure 7.27 displays the
influence of the time step on the simulation accuracy, described by the resulting
cross section area. For large time steps, the simulation accuracy is very low, which
is caused by the errors of the linear interpolation of the tool movement. For small
time steps the interpolation error is negligible, because the difference between two
tool positions is low and its movement can be linearized. Summarizing the results
from Figure 7.27, a simulation time step of 0.5 seconds or lower should be ap-
plied, the lower the better.
Fig. 7.27 Influence of simulation time step on simulation accuracy for calculating the
workpiece cross section of a ground two-flute twist drill (dexel grid distance: 0.1 mm)
Another application of the simulation system is the analysis of technological re-

lationships in tool grinding processes. Figure 7.28 exemplarily shows the influ-
ence of the feed velocity vft on the resulting process forces for grinding a two-flute
tungsten carbide twist drill with 10 mm diameter. In the analyzed parameter limits,
the process forces show a non-linear dependency on the feed velocity which re-
flects the relationship of the process force model as shown in Figure 7.15.
Fig. 7.28 Influence of feed velocity on calculated process forces accuracy for calculating
the workpiece cross section of a ground two-flute twist drill (simulation time step: 0.5 s,
dexel grid distance: 0.1 mm, for process parameters, see Figure 7.21)
7.6.4 Optimization
The main objective of the research on process machine interaction is to understand
the occurring effects during grinding and to predict process characteristics and
process outcome. With this knowledge, it is possible to optimize grinding
processes already in the process planning phase. In this case, the developed mod-
els have been used to adapt the path of the grinding wheel to reduce the shape er-
rors in the ground workpiece.
The optimization procedure starts with a simulation of the static workpiece def-
lection using the simulation model described in Section 7.6.1. During the simula-
tion, the workpiece deflection is computed at the point, where the grinding wheel
exits the contact area. At that point, the final flute profile is generated. Therefore,
it is a good position to evaluate resulting shape error. A characteristic deformation
profile for a two-flute tool is shown in Figure 7.29. It can be seen that the deflec-
tion constantly rises during the entrance phase of the grinding wheel and then de-
creases quadratically during full engagement due to the increasing stiffness of the
workpiece near the clamping. The deflection v in y-direction of the second flute is
significantly higher because of the reduced stiffness of the workpiece after grind-
ing the first flute.
Fig. 7.29 Calculated workpiece deflection (u,v,w) in (x,y,z)-direction at current position of

grinding wheel
The next step is the segmentation of the calculated workpiece deformation into
line elements. These elements are then translated into NC commands, which
change the movements of the grinding wheel. With the new NC-program, the si-
mulation of the deformation is repeated until the desired shape of the workpiece is
achieved. In this research, two optimization criteria are used to evaluate the cur-
rent shape of the workpiece. The first criterion is the difference of the core radius
of simulated and ideal cross-section. The second criterion is the dexel-wise com-
parison of the shape of a simulated ideal cross-section.
When the current NC-program leads to a workpiece shape fulfilling the quality
criteria, the optimization procedure is stopped. As an exemplary result, Figure 7.30
compares the cross-sections at the tip of a workpiece for two different simulation
runs with the ideal cross-section. The first cross section (left) has been simulated
using the original NC-program. A relatively large deviation of the resulting work-
piece shape from the ideal workpiece shape is visible with a maximum deviation of
160 µm, which corresponds to the deflection presented in Figure 7.29.
Fig. 7.30 Comparison of resulting workpiece cross-sections with uncompensated NC-

program (left) and compensated tool path (right)
The cross-section on the right of Figure 7.30 has been computed using an
adapted NC-program, which was the result of one optimization loop. It can be
seen that the error has been reduced significantly with a maximum deviation of 35
µm. The results indicate that the presented optimization procedure is able to im-
prove the workpiece quality in tool grinding, considering the existing process ma-
chine interactions.

An interdisciplinary modeling approach to predict the process machine interaction
in tool grinding has been presented. Models for the grinding process and the me-
chanical behavior of workpiece and machine have been developed and coupled
with each other. A simulation of the process kinematics has been combined with a
dexel-based material removal simulation to compute the varying workpiece shape

during the grinding of cutting tools. It has been possible to calculate the grinding
forces and the static and dynamic deflections occurring during tool grinding. For
the first time, it has been possible to calculate the stress distribution in the com-
plex contact area. For this purpose, an empirical grinding force model was devel-
oped, which connects the acting stress with the equivalent depth of cut heq divided
by the local contact length dlg. A finite element approach to calculate the stress
due to the microscopic grain engagement shows a high influence of the Young’s
modulus and the damage criteria. Based on the geometry-dependent mechanical
properties of the workpiece and the simulated contact forces, the quasi-static and
time-variant deflections can be predicted. The presented approaches have been va-
lidated by comparing measured and simulated workpiece geometry after grinding.
In addition, the developed simulation methods have been used to optimize the path
of the grinding wheel for a reduction of geometry errors of the ground tool. Since
the modeling approach is designed modularly it can be extended continuously in
the future, e. g. to include other grinding wheels or workpiece materials.
References
[1] European Cutting Tool Association: Zerspanungswerkzeuge: Marktvolumen welt-
weit, Fertigung Sonderausgabe – Werkzeuge 12, 7 (2010)
Klocke, F., Kröger, M., Steinmann, P., Weinert, K.: Modelling and Simulation of
Process - Machine Interaction in Grinding Production Engineering, vol. 3, pp. 111–
120 (2008)
[3] de Payrebrune, K., Kröger, M., Deichmueller, M., Denkena, B.: Investigation on the
Dynamics of Tool Grinding, Machine Dynamics Problems, vol. 33, pp. 92–104
(2009)
[4] Popp, K., Kröger, M., Deichmueller, M., Denkena, B.: Analysis of the Machine
Structure and Dynamic Response of a Tool Grinding Machine. In: 1st International
Conference on Process Machine Interactions, pp. 299–307 (2008)
[5] Popp, K., Kröger, M., Deichmueller, M., Denkena, B.: On Contact Modeling of
Workpiece and Grinding Wheel with Nonlinear Elements. In: 7th EUROMECH Solid
Mechanics Conference, pp. 281–282 (2009)
[6] Chang, H.-C., Wang, J.-J.J.: A stochastic Grinding Force Model considering Random
Grit Distribution. International Journal of Machine Tools and Manufacture 48, 1335–
1344 (2008)
[7] van Hook, T.: Real-Time Shaded NC Milling Display ACM SIGGRAPH Computer
Graphics, vol. 20, pp. 15–20 (1986)
[8] Denkena, B., Böß, V.: Technological NC Simulation for Grinding and Cutting
Processes Using CutS. In: 12th CIRP Conference on Modelling of Machining Opera-
tions, vol. II, pp. 563–566 (2009)
[9] Biermann, D., Mohn, T.: A Geometric-Kinematical Approach for the Simulation of
Complex Grinding Processes. In: 6th CIRP International Conference on ICME (2008)
[10] Denkena, B., Tracht, K., Yu, J.-H.: Advanced NC-Simulation based on the Dexelmo-
del and the HRMC-Algorithm Production, Engineering, vol. XIII(1), pp. 91–94
(2006)
[11] de Payrebrune, K., Kröger, M.: Modal Reduction of Dynamics of Tool Grinding. In:
2nd International Conference on Process Machine Interactions (2010)
[12] Denkena, B., Deichmueller, M., Kröger, M., Popp, K.M., Carstensen, C., Schroeder,
A., Wiedemann, S.: Geometrical Analysis of the Complex Contact Area for Modeling
the local Distribution of Process Forces in Tool Grinding. In: 1st International Confe-
rence on Process Machine Interactions, pp. 289–298 (2008)
[13] Han, W., Reddy, D.: Plasticity. Springer (1999)
[14] Brokate, M., Carstensen, C., Valdman, J.: A Quasi-Static Boundary Value Problem in
Multi-Surface Elastoplasticity. I: Analysis, Math. Methods Appl. Sci. 27(14), 1697–
1710 (2004)
[15] Schröder, A., Wiedemann, S.: Error Estimates in Elastoplasticity using a Mixed Me-
thod, Humboldt Universität zu Berlin, Institute of Mathematics, Preprint 11-01 (2011)
[16] Schröder, A.: Mixed Finite Element Methods of Higher-Order for Model Contact
Problems. Humboldt Universität zu Berlin, Institute of Mathematics, Preprint 09-16,
submitted to SINUM (2009)
[17] Schröder, A., Wiedemann, S.: Humboldt Universität zu Berlin, Institute of Mathemat-
ics, Preprint 11-02 (2011)
[18] Schröder, A.: Constraints coefficients in hp-FEM. Numerical Mathematics and Ad-
vanced Applications, 183–190 (2008)
[19] Schröder, A.: Constrained approximation in hp-FEM: Unsymmetric subdivisions and
multilevel hanging nodes. LNCSE, vol. 1(76), pp. 317–325 (2009)
[20] Han, W., Reddy, D.: On the Finite Element Method for Mixed Variational Inequali-
ties arising in Elastoplastiity. SINUM 32(6), 1778–1807 (1995)
[21] Brokate, M., Carstensen, C., Valdman, J.: A quasi-static Boundary Value Problem in
Multi-Surface Elastoplasticity. II: Numerical Solution, Math. Methods Appl.
Sci. 28(8), 881–901 (2005)
[22] Wang, C.Y., Clausen, R.: Marble Cutting with Single Point Cutting Tool and Di-
amond Segments. Machine Tools & Manufacture 42, 1045–1054 (2002)
[23] Lemaitre, J., Desmorat, R.: Engineering Damage Mechanics. Springer (2005)
[24] Deichmueller, M., Denkena, B., de Payrebrune, K.M., Kröger, M., Wiedemann, S.,
Schroeder, A., Carstensen, C.: Determination of Static and Dynamic Deflections in
Tool Grinding using a Dexel-Based Material Removal Simulation. In: 2nd CIRP In-
ternational Conference on Process Machine Interactions (2010)
[25] Tönshoff, H., Peters, J., Inasaki, I., Paul, T.: Modelling and Simulation of Grinding
Processes. CIRP Annals - Manufacturing Technology 41, 677–688 (1992)
Part III
Cutting
Chapter 8
HPC - Stability Simulation
C. Brecher, R. Hermes, and M. Esser
Abstract. The prediction of stable process parameters to maximize the productivi-

ty of milling machines has been an important field of research for a long time. In
the past, simulation tools allowing an assessment of the process stability have
been created. Nevertheless, the accuracy of predictions by simulation is not yet
high enough to make efficient use of stability simulation in production planning.
In this context, the article presents developments in the field of modeling the dy-
namic machine and process behavior for process machine interaction simulation in
the field of high performance cutting processes (HPC). On the process side, a
complex force model, which takes into account the effects of a phase shift be-
tween force generation and chip thickness variation, is introduced. Also, an analy-
sis of fast-rotating main spindle systems has been carried out, which considers
variations in the dynamic compliance behavior at the tool center point (TCP) due
to rotor dynamic effects and variation of the bearing rigidity. With the PrimeCut
software package, a calculation program is presented, which includes the state of
the art and new developments in the field of stability simulation.
8.1 Introduction
For a long time, the technologies of High Performance Cutting have contributed to
the fact that material removal rates in milling processes have greatly increased due
to new cutting materials and tool geometries [14]. The machine technology has al-
so been adapted to the requirements [1], [7]. Lightweight design coupled with a
high degree of rigidity permits high acceleration values and feed rates in the main
axes. Main spindle systems have also been subject to further development and
today offer rotating speeds and torques permitting highly-efficient machining
operations.
However, the productivity of machine tools is often limited, not by the installed
drive performance but by the regenerative vibrations, the so-called chatter [4],
[12]. In such cases, the machining process becomes instable due to the force com-
pliance interaction between the machine tool structure and the cutting process [3],
[6]. Under such extreme operating conditions, the result is occasionally an escalat-
ing shortening of the durability of individual machine components and breakage
of tools. In addition, instable processed workpieces are rejected due to surface
damage.
180 C. Brecher, R. Hermes, and M. Esser
In this field, the simulation of stability charts offers a cost and time-efficient
possibility of parameterizing machining processes with regard to their stability be-
havior or to use the process stability as an assessment criteria in the analysis and
design of machine systems [5], [11]. Today, there are numerous simulation possi-
bilities for predicting optimized process parameters, whose limited quality of re-
sults, however, prevents a sensible application to the planning of production. The
same applies to computer-aided evaluation of stable machining performance in the
design phase of machine systems [16].
The aim of the work accomplished in the priority program SPP 1180 “HPC sta-
bility simulation” and described in this article is the improvement of the accuracy
of simulated stability charts to determine process-optimized working parameters.
For this purpose, the sub-systems machine tool structure and cutting process
as well as their simulative depiction in the stability simulation regarding their in-
fluences on errors on the simulation results are determined first. Based on a sys-
tematic analysis, the potential for improvement is shown and implemented in the
modeling of the sub-systems. Besides the individual consideration of the machine
structure and machining process, their coupling and the method of stability simu-
lation concerning potential influences of errors are also analyzed. Here, the influ-
ences on simulation accuracy due to pre-determined simulation parameters are
analyzed as well as the program structure of the simulation models used in the
WZL regarding calculation time-efficiency and usability.
The knowledge obtained by the work carried out is summarized in the newly-
developed PrimeCut simulation software. Besides efficiently computing a stability
analysis of machining processes, the software allows for a calculation of surface
structures of processed workpieces.
8.2 Dynamic Compliance Behavior of Rotating Main Spindle

Systems
The dynamic compliance behavior of the machine structure is one of the two es-
sential input parameters of stability simulation. For modeling a machine system,
not only the compliance behavior of structural components must be acquired but
also all machine components in the force flow between the tool and the workpiece.
As a key component in the HPC processing, particular attention has to be paid to
the dynamic compliance behavior of the system comprising the main spindle and
tool. At these fast-rotating spindle-bearing systems, various influences cause a
compliance behavior, which changes with the speed of revolution. In general,
these rotation-speed-dependent effects can be divided into two classes. In the one
class, the heat generation in the spindle bearings as well as in the drive unit causes
changes in the mechanical properties of the rotating systems. Both changes of the
axial bearing pre-load as well as the radial bearing rigidity due to thermal expan-
sion can be identified as the main cause. The second class consists of rotor-
dynamic effects and the effect of the centrifugal force influence on fast-rotating
spindle systems. The influence of the centrifugal force has a similar effect as the
thermal expansion in widening the inner ring of the bearing and changing the
8 HPC - Stability Simulation 181
position of the rolling element. This effect causes a change in the radial and axial
bearing rigidity. The influences of thermal expansion and centrifugal force in
spindle bearings were investigated on individual test rigs [8] in the works and their
significance was shown in the work of [17].
Apart from the influence of similar spindle components, there is a change in the
compliance behavior of main spindle systems under revolution due to rotor-
dynamic effects at the spindle shaft itself (Fig. 8.1). Gyroscopic and centrifugal
forces influence the natural frequency and compliance amplitudes of the dynamic
behavior at the TCP. Dynamic process loads and the resulting bending of the spin-
dle shaft cause additional balancing influences. The spindle shaft with its slim
structure does not resemble the idealized Laval runner, yet follows the rules of ro-
tor-dynamics.
Gyroscopic effect Centrifugal effect

Impressed Centrifugal
Rotor
displacement
Mass force
Rotating
mass
ω ω
Axis of
gyroscope
Resultant
displacement Static and dynamic unbalance
Reference: Mode shape of spindle nose Dynamic unbalance Static unbalance
Axis of gyroscope
Standstill without Rotation with

gyroscopic effect gyroscopic effect Axis of balance point
Fig. 8.1 Rotor-dynamics of the spindle shaft and the tool [11]
For slim rotors, the resonance frequency is reduced due to the occurring gyros-
copic force. In a resonance case, the TCP conducts an elliptical movement in the
same direction of rotation as that of the tool. For external excitation of the rotating
spindle shaft, for instance due to dynamic process loads or imbalances because of
the bending of the shaft, a deflection in the direction opposite to the rotation is su-
per-imposed. Decisive for the simulative modeling of stability behavior is the ac-
tual dynamic compliance of the main spindle under the influence of rotation. For
this reason, the focus of the work being carried out is on the metrological determi-
nation of this behavior.
8.2.1 Test Rig for Measuring Rotating Spindle Systems

In the following, a test rig for measurements on rotating spindle systems is
presented, whose dynamic compliance behavior is metrological evaluated in prep-
aration of the measurements on real main spindle systems (Fig. 8.2). The mea-
surement at the whole system consisting of spindle and bearing now permits the
Belt drive gear Spindle shaft Spindle bearing arrangement

 Concept: double-O-bearing
 Single bearing: 4 * FAG

XC7014E.T.P4S.WKL
 Loose bearing function: ball lining
 Total preload: 1200 N
 Rigidity of preload: 573 N/mm
Bearing rigidity (simulation)
Rigidity [N/µm]
Loose bearing TCP 300 radial
Claw clutch with ball lining Fixed
bearing 200
 High speed spindle in open construction without axial
100
peripheral components
 Belt drive gear and claw clutch 0
0 10000 20000 30000
 nmax = 24.000 1/min Spindle speed [1/min]
Fig. 8.2 Test rig for the assessment of a rotating system [11]
possibility of investigating the theoretical background of rotor-dynamic effects

and their influence on the mode shapes of rotating main spindle systems.
The test rig follows the design of a machine spindle with external drive and
preloaded bearing concept. The dimensions of the spindle shaft as well as the ar-
rangement of the bearings correspond to the dimensions of modern motor spindles
with HSK-63 tool interface. Compared to the conventional spindle housing, the
bearings are positioned on two separated bearing blocks mounted on a rigid base
plate.
For measuring the compliance behavior at the TCP or the modal analysis of the
rotating spindle shaft, the auxiliary spindle test rig is equipped with the measuring
and excitation technology shown in Fig. 8.3. Using a self-designed impulse exciter
the system is excited in a frequency spectrum up to approximately 4,500 Hz. The
induction of an impulse-shaped force with this shock pendulum represents a pos-
sibility of inducing a homogenous force into the rotating system, which is easy to
carry out. The impulse force progression is measured by means of a piezo-electric
force element, which is integrated into the tip of the impulse exciter. Only contact-
less displacement sensors can be used for measuring the deflection response on ro-
tating spindles. Here, an eddy current sensor provides sufficient resolution quality.
However, the deflection responses measured at the excited system contain er-
rors derived from the non-circularity or out-of-balance of the spindle shaft. It is
not possible to filter out these measuring errors, which are dependent on the speed
of rotation, because an effective filtering would also impair frequencies relevant
for the measurements. However, the deflection signal can be determined from the
measurement signal by means of a software-side subtraction procedure in time
domain. For this purpose, the displayed measurement signals are divided into two
sections of equal length.
Belt guard Rigidity flange
Clutch
Spindle shaft
Loose bearing
support
Impulse exciter
Ball lining (shock pendulum)
Bearing Eddy current sensor

temperature control
Fixed bearing
support
Mounting plate
Machine base out of

rectangle profiles welded
and compound filled with
concrete
Fig. 8.3 Modal analysis of the test rig spindle under rotation [11]
The first section of the signal comprises the duration of the impulse excitation
and the decay of the deflection response. The second section serves as a reference
and contains only parts of the interference signal. The signal parts are aligned with
each other by means of a Best-Fit process and the interference signal is eliminated
by means of subtraction.
Fig. 8.4 shows the metrological acquired dynamic compliance behavior at TCP
for various speeds of revolution. Also shown here are the measurements at rest
and for 1,000 1/min in order to determine variation of the compliance behavior be-
tween still-standing spindle systems and systems with a very slow revolution
speed. Thermal effects as well as centrifugal force, out-of-balance and gyroscopic
effects exert no influence at low spindle speeds of 1,000 1/min. However, the
load-bearing lubrication film, which forms in the bearing under rotation, influ-
ences the damping properties of the bearing and thus the dynamic compliance be-
havior of the overall system. Thus, the measurement shows a slight change in the
resonance peaks to higher damping and less compliance.
In Fig. 8.5, the frequency ranges of the dominant mode shapes, here the shaft
bending of first, second and third order, are depicted. These mode shapes can
cause instabilities in the milling process but change their respective characteristics
in form and value of the amplitude superposition in dependence of the spindle
speed. All three mode shapes change in different ways at higher speeds of revolu-
tion. The bending vibration of the first order at 1,000 Hz reduces its amplitude to
approximately 40 % and splits into two mode shapes. A clean differentiation of
the two resulting peaks in the modal analysis is not possible. At 24,000 1/min, a
new mode shape arises at approximately 800 Hz, which can also be identified in
the modal analysis as a bending vibration of first order.
1
First order Second order Third order
bending vibration bending vibration bending vibration
Compliance [µm/N]
0.1
<
0.01
Spindle speed
0
0.001
1000 1/min
180
15.000 1/min
90 24.000 1/min
Phase [°]
-90
-180
0 1000 2000 frequency [Hz] 4000
Fig. 8.4 Dynamic behaviour of the test rig-spindle [11]
In the frequency range of the bending of the second order, at around 2200 Hz,
the amplitude rises and the phase response becomes flatter when the spindle speed
increases. The third dominant mode shape at 2,500 Hz shows a behavior resulting
from a significant influence on the gyroscopic forces. The resonance peak splits
into two peaks. A creation of a second resonance peak can also be clearly recog-
nized in the phase response. Fig. 8.5 shows the evaluation of the modal analysis
for the mode shapes of third order. For low spindle speeds, a definite oscillation of
the spindle tip in the direction of the excitation force can be seen in the results of
the modal analyses (second figure from the left). The mode shape changes at high-
er spindle speeds. A circular movement of the TCP occurs from the linear bending
vibration. This circular movement in the respective low-resonance frequency is in
n n
dx Fx
dy
Mode shapes of third order bending vibration
 Measurement in six n [min-1] f [Hz] n [min-1] f [Hz] n [min-1] f [Hz]
levels
0 3515 15.000 3458 15.000 3531
 Detection of two
directions 1.000 3497 24.000 3424 24.000 3552
Fig. 8.5 Verification of the splitting of resonance peaks [11]

the same direction as the rotation of the spindle and for the respective higher fre-
quency opposed to the spindle rotation (third and fourth figures from the left). The
behavior then corresponds to those of the gyroscopic effect expected for strong
angular deviation of the shaft.
8.2.2 Simulation of Real HPC Motor Spindles

The investigations at the auxiliary spindle test rig show that the compliance beha-
vior of the main spindle system including the tool must not be ignored for the
modeling of the dynamic machine behavior in the HPC stability simulation. For
this purpose, the auxiliary spindle test rig has been modeled with WZL’s own
software NewSpilad on a trial basis. Centrifugal and gyroscopic forces are also in-
cluded in the calculation. The speed-dependent rigidity of the spindle bearings is
determined by means of the WinLager calculation program and taken into account
in NewSpilad for the calculation of the overall spindle behavior.
Fig. 8.6 shows the comparison of metrological and simulative determined com-
pliance behavior of the spindle-bearing system. The comparison clearly shows that
the simulation does not depict the effects of the spindle rotation with sufficient ac-
curacy. The reason for this is the insufficient depiction of the bearing behavior
caused by linearizing errors at the working point and by ignoring the bearing tip-
ping in the bending vibrations of the shaft.
Due to the limited depiction quality the currently available simulation systems
have not been able to model the dynamic compliance behavior of main spindles in
the stability simulation with high accuracy. In order to improve the quality of the
results of stability simulation the metrological determined compliance behavior of
real main spindle systems considering the influence of the rotation speed is used.
1
Compliance [µm/N]
Simulation results: Standstill n = 0

spindle speed vs. compliance amplitude
0.1
]
/N
µm
c e[
li an 0.01
mp
Co Measurement
100 Simulation
0.001
1
Compliance [µm/N]
10-1 n = 24.000 1/min
10-2 0.1
24
10-3
18 in]
0 /m 0.01
00
1000 12
d [10 Measurement
2000 ee
3000
6 sp Simulation
dle 0.001
Frequency [Hz] 4000 0 in 0 1000 2000 3000 4000
Sp Frequency [Hz]
Fig. 8.6 Compliance simulation of the spindle test rig [11]

For measuring the compliance behavior on original main spindle systems the
same measurement setup is used as applied to the auxiliary spindle test rig. How-
ever, this method is not suitable for the measurement at the original milling tools
because the cutting edges and chip spaces of the mill permit neither contact excita-
tion nor an exact measurement of the deflection. For this purpose, the milling tools
are replaced by cylindrical carbide rods, so-called dummy tools, and mounted into
the spindle system to be investigated. For depicting the real compliance behavior,
the dummies have the same moment of inertia as the original tools (Fig. 8.7 left).
Comparable measurements between the original and dummy tool with the spindle
at rest provide very good approximations of compliance behavior.
Fexc Fexc 1
mm
mm
Dummy tool: D = 20 mm, Lges = 163 mm

Compliance [µm/N]
20
15
Motor spindle: nmax = 28.000 1/min, Pmax = 18 kW

0.1
A = 176 mm2 A = 148 mm2
Ix/y = 2485 mm4 Ix/y = 1189/2918 mm4
n=0
0.01 n = 10.000 1/min
n = 20.000 1/min
n = 28.000 1/min
0.001
0
Phase [°]
-90
Dummy- and original- tool
cemented carbide in tool holder
D = 20 mm, Lges = 163 mm -180
0 1000 2000 Frequency [Hz] 4000
Fig. 8.7 Compliance behavior of original and dummy tool [11]
The measurement results for a dummy tool are shown exemplarily in Fig. 8.7
for various spindle speeds. As for the auxiliary spindle test-rig, the measurement
results show a significant change of dynamic behavior over the speed range. Be-
cause of the good comparability of the measuring results at the original and dum-
my tools, the measurement results determined on the dummy will be used for the
stability calculations in the following. The simulation is taking into account the ef-
fects of spindle rotation and gyroscopic effects on the original main spindle sys-
tem. Fig. 8.8 shows the comparison between two simulated stability charts with
and without taking into account the influence of the spindle speed as well as the
cutting depth limit determined in machining tests. A shift of the stability maxi-
mum due to the consideration of spindle speed and an improvement of simulation
results in comparison with the stability limit determined in the machining tests can
be clearly identified. Clear improvement potential by taking into account rotation-
al speed-dependent compliance behavior is provided mainly in the region of high-
speed cutting of aluminum. Due to the rigid structure of the machine and the high
rotation speed of the mill (stable region beyond the highest stability minimum),
the dynamic behavior of the machine structure in the lower frequency range in
comparison with the dynamic behavior of the cutting tool and main spindle system
is irrelevant for the occurrence of dynamic chatter vibrations.
Apart from the depiction of the dynamic machine behavior, modeling of the
dynamic process behavior presents the second important aspect of stability simula-
tion. Linear, exponential and mechanical force models have been the state of the
art, [3], [20], for a long time but they ignore potential influences in the analytical
description of the process forces. Some authors [15], [18] point out that a real ma-
thematical relationship between chip thickness and dynamic cutting force is not
sufficient in all cases.
30 Comparison
Cutting tests
Cutting depth ap [mm]
of measurement
24 Stable cut and simulation
Chatter Power
li mit Cutting tool:
18 Chatter frequency
Mitsubishi
12 BXD 4000
D = 32 mm
2 cutting edges
6
Material:
0 Simulation with spindle aluminium 7010
Chatter frequency fR [Hz]
speed dependent compliance

1500 Simulation with Process:
compliance at n = 0 ae = 32 mm
1200
fz = 0.18 mm / tooth
900
600
300
0
15000 18000 21000 Spindle speed [1/min] 27000
Fig. 8.8 Stability chart in the machining of aluminium [11]
8.3 Modeling of the Dynamic Process Behavior
More detailed investigations show that the dynamic deflection of the cutting edge
inducing a highly-dynamic chip thickness variation must be included in the mod-
eling of the machining process. Examples are the consideration of the process
damping (effect of the contact between tool flank and workpiece surface) or the
deduction of an effective direction-corrected cutting speed [2] [9], [13]. The use of
a test-rig is necessary in order to be able to determine the influence of the cutting
dynamics on the cutting forces for a large range of process parameters. The in-
fluencing dimensions of the dynamic process force creation defined in the test rig
experiments are analyzed in further procedures to derive an analytical model with
improved reproduction accuracy of the cutting force generation at the cutting
edge.
8.3.1 Test Rig Concept for Dynamic Cutting Force Measurement

In addition to the metrological acquisition of the relevant process parameters, cut-
ting force and cutting depth modulation (dynamic chip thickness variation), the
test-rig has to fulfill the requirement to induce defined oscillation movements be-
tween tool and workpiece into the cutting process after having defined frequency
and amplitude. In addition, the process-relevant dimensions of cutting speed and
feed motion for the test process must be definitely set and permitted by the struc-
ture of the test-rig. A continuous turning process is suitable for implementing the
necessary requirements at the test-rig and the cutting force measurement as the
cutting parameters can be constantly set independently of each other. In contrast,
in the milling process the chip thickness varies over time and depends on the cut-
ting speed, feed motion and the number of cutting edges. The kinematic conditions
of the turning process with a still-standing tool are also more suitable to induce de-
fined oscillations into the process than into a milling process with a rotating tool.
Central component of the designed test-rig structure is the tool-holder shown in
Fig. 8.9 (right). This transfers the oscillation movement created by a piezo-
actuator in the cutting edge girder, which targeted deflections of the cutting edge
are introduced into the process by, orthogonally to the cutting speed. The use of a
stacked piezo-actuator for the system excitation is a necessary requirement for a
higher rigidity of the actuator system in the Z-direction, which is sensitive to the
variation of the chip thickness with also simultaneous high excitation frequencies.
Systems on an electric-hydraulic or electro-magnetic base provide either little fre-
quency bandwidth or an insufficient rigidity. Apart from the piezo-actuator used,
the guidance of the tool-holder resp. the oscillation movement of the cutting edge
represent the second relevant components for the capability of the test rig. Along
the Z-axis, the guidance must be designed very flexibly to realize the deflection of
the cutting edge, preferably free of power dissipation. At the same time, the cut-
ting forces in the remaining spatial directions should be dissipated by the greatest
possible rigid support in the framework. For this purpose, the tool holder is sup-
ported with a flexure joint each at its front and rear side in the framework in the
X-Y-level (Fig. 8.10). Flexure joints are selected in place of hydrostatic guidance
systems or rolling element guidances for the implementation of the test-rig con-
cept because they are almost free of friction for mean deflections as well as having
a high potential for geometrical and mass optimization of the moving components.
Due to the achievable mass reduction, the amplitude as well as the oscillation fre-
quency can be positively influenced without increasing the performance of the
piezo-actuator. In connection with the test-rig structure, the selected actuator sys-
tem achieves a frequency bandwidth of up to 2 kHz at a deflection of 40 μm. This
corresponds to the requirements determined in pre-tests for depicting instable cut-
ting experiments in the HPC processing, which are typically characterized by chat-
ter vibrations in the range between 100 Hz and 2,000 Hz.
Apart from the implementation of the guidance movement and force transmis-
sion, the tool holder serves as a carrier for the sensors for the metrological acquisi-
tion of the process. Three piezo sensors arranged on the same level between tool
holder and cutting edge girder are used for the acquisition of the cutting forces in
the three spatial directions. The inertia forces of the tool holder and the cutting
edge also acquired by the force sensors during the oscillation movement are ma-
thematically compensated by a down-streamed measuring program.
Flexure joint for

View A vf View A
Reference body tool guidance
(half section)
Work piece Overlaid
oscillation
Cutting edge movement:
A· sin(2πf· t)
nSP
y z
Spindle position measurement
with rotary encoder on the spindle x
Reference area for Eddy current sensor

eddy current sensor Piezoelectric sensor
Fig. 8.9 Guide and workpiece concept for dynamic cutting tests [11]
The relative compliance between workpiece and tool is acquired by means of

the eddy current sensor integrated in the cutting edge holder for calculating the
chip thickness modulation. For this purpose, the workpiece holder fixed to the
HSK63 tool holder is equipped with a reference surface serving as a basic dimen-
sion for the eddy current sensor (Fig. 8.9, center). Use is made of ring cylinders of
various test materials as test workpieces with the chip width of the cutting process
being pre-defined by the width of the ring. Fig. 8.9 (left) schematically shows the
engagement conditions as well as the positioning of the tool holder and the test
workpiece during the testing process. To perform a measurement the workpiece is
rotated by the main spindle of the milling machine according to the pre-defined
cutting speed. The feed motion by the machine system in the process is carried out
in the same direction as the dynamic cutter deflection along the Z-axis. For the du-
ration of the measurement, the ring surface is overcut again at each workpiece ro-
tation and machined in its whole width.
Apart from the relative compliance between tool and workpiece, the surface
contour of the workpiece, which was applied by the dynamic deflection of the cut-
ter in the previous overcutting, is also included in the calculation of the chip
thickness modulation. For this purpose, the upper chip thickness modulation (chip
thickness change from current tool deflection) and the lower chip thickness mod-
ulation (chip thickness change due to surface contour of the workpiece) are settled
with the aid of the workpiece angle of rotation. The rotary encoder integrated in
the machine system is read out for the acquisition of the rotation angle.
After having obtained all measuring data it must be revised for systematic er-
rors. The signal processing used includes the filtering of aliasing effects and the
compensation of amplitudes and phase errors by means of downstream measuring
hardware for signal conditioning. Apart from this, the effects of the mechanical
structure behavior and the inertia of the test-rig components have been taken into
account. For this purpose, a suitable user interface in the LabView software envi-
ronment has been created for the operation of the test-rig as well as for data acqui-
sition and data evaluation.
Support for
piezo actuator
Piezo actuator
Adapter
Flexure joints
y z
ard
ebo
Bas
Eddy current Force sensors Framework

sensor Cutting edge
Fig. 8.10 Design of the overall test-rig for dynamic cutting tests [11]
8.3.2 Modeling Approach of Dynamic Cutting Force

The modeling approach of dynamic cutting forces developed within the scope of
the project work provides the process response on an empirical base in the form of
cutting and passive forces on an excitation by means of a harmonic chip thickness
variation. For further depiction of the dynamic effects at the cutting edge, there is
a differentiation of the influences of the inner and outer chip thickness modula-
tion. Furthermore, it seems to make sense to select a complex depiction level for
the modeling. In this way, not only rigidity but also a damping with a phase shift
of 90° can be allocated to the process with respect to the upper or lower chip
thickness modulation. For instance, models according to these criteria are de-
scribed by van Brussel [19] and Werntze [21]. In a slightly altered form, the model
is defined by the following equations.
x i (t) = x̂ i ⋅ sin(ωt ) (8.1)
x o ( t ) = x̂ i ⋅ sin(ωt − ε) (8.2)
′ ,c (t ) = Fˆdyn
Fdyn ′ ,c ⋅ sin(ωt + ϕ Fc )
(8.3)
= xˆ i (Rci sin(ωt + ϕ ci ) + Rco sin(ωt + ϕ co − ε ) )
′ , p ⋅ sin(ωt + ϕ Fp )
′ , p (t ) = Fˆdyn
Fdyn
= xˆ i (R pi sin(ωt + ϕ pi ) + R po sin(ωt + ϕ po − ε ) )
(8.4)
The modeling approach is based on a complex dynamic force influence from the
chip outer side as well as from the chip inner side, whose wave profiles are shifted
against each other. Eq. 8.1, Eq. 8.2. xi and xo describe the compliance for the inner
and outer sides of chip thickness modulation. xo is shifted backwards to xi in terms
of time about the positive angle ε. All phase positions are referenced to the mod-
ulation of the inner chip thickness, thus to the current relative compliance between
tool and workpiece. p with reference to the chip width are calculated. The interre-
lation between chip thickness modulation and the resulting dynamic cutting force
F’dyn,c resp. passive force F’dyn are shown in Fig. 8.11. They apply for the equation
of the cutting force (Eq. 8.3) and also for the calculation of the passive force (Eq.
8.4) with different parameters. The resulting dynamic force is comprised of a part,
which the outer chip thickness modulation is responsible for, and which is scaled
in relation to this with Rjo and shifted about φjo in phase, as well as a correspon-
dingly scaled (Rji) and phase shift (φji) of the inner chip thickness modulation. The
two dynamic force components are added to the total passive or cutting force re-
spectively. With that, the displacement of the cutting edge exclusively along a sin-
gle direction is possible.
Phase shift between
Reference force
inner and outer chip
Cutting tool displacement
thickness modulation Outer chip

ε thickness modulation
Xo(t)
φo
Ro
F‘o(t)
F‘i(t) φi
Ri
Xi(t)
Inner chip
thickness modulation Time
Fig. 8.11 Dynamic force model in complex domain [11]
With that, the dynamic displacement of the cutting edge exclusively along the
modulation of both sides of chip thickness, the parameters for the inner and outer
modulation cannot be determined directly from individual force and deflection
measurements. Instead, the parameters are numerically calculated from the mea-
surement results of several cutting tests. For this purpose, cutting tests with the
same process parameters but various phase angles ε are performed. The variation
of the phase angles is set over a small change of rotation speed between the tests.
The round and square symbols in Fig. 8.12 (left) represent the individual experi-
ments with different phase angles ε under otherwise identical conditions.
Fig. 8.12 on the left side for the machining of steel material C45E shows the
progression of the cutting forces referenced to the cutting width and cutting depth.
The phase angle given is referenced to the modulation on the inner cutting side
(xi). Because of this reference the phase angles of the cutting force are usually
attributed to the region of the second and third quadrants, thus between -90°
and -270°. It must be noted that the absolute value of the cutting force reaches its
maximum approximately at a phase angle of ε = 180° and almost reaches a value
of zero when the upper and lower chip thickness modulation are in phase (ε = 0°).
However, the passive force hardly varies. Thus, it is mainly caused by the modula-
tion of the inner chip thickness.
Amplitude [N/mm²] Phase [°] Imaginary part [N/mm²]

3000 0 1500
-90 750
2000
-180 0
1000
-270 -750
Chip shape:
-360 -1500
0 90 180 270 360 -3000 -2000 -1000 0
Phase angle ε [°] Real part [N/mm²]
Cutting force: Amplitude phase Passive force: Amplitude phase Cutting force Passive force
vc = 200 m/min foscillation = 840 Hz Material: C45E

fU = 0.1 mm/U doscillation = 20 µmp-p Cutting tool: Sandvik R245 E-PL 4030
Fig. 8.12 Illustration and fitting of the dynamic cutting forces [11]
The extended lines represent the prediction of the introduced force model,
whose parameters have been calculated with the aid of the measuring values ac-
cording to a best-fit process. Thus, the equation system can only be uniquely
solved with four scalable results for four unknown parameters, it can be deter-
mined that two arbitrarily selected measurement results (respectively amplitude
and phase angle, that means four scalar values) are always sufficient to define the
four model parameters, respectively for cutting and passive force. On the right
side in Fig. 8.12, the same measuring values are presented in the complex plane.
The shown curves illustrate the rotation of the force vectors from the outer mod-
ulation around the peak of the force vector of the inner modulation. Here, the
small diameter of the passive force curve shows that the passive force is generally
determined by the inner chip thickness modulation and has no linear dependency
on the chip thickness. Conventional force models cannot represent this behavior.
The cutting force model introduced in the complex domain offers extended
possibilities depicting the real process force behavior in a simulative manner and,
as a result, to take into account numerous variables of process parameterization
and dynamic machine behavior. Within the test-rig investigations, the relevance of
individual influences on the progression of the process force has been investigated
and parameterizations for the calculation of the process forces have been deter-
mined. As an example, the potential of the dynamic cutting force model in the de-
piction of the influence of clearance flank wear on the process force is shown. Fig.
8.13 shows the resulting model parameters for the complex dynamic cutting force
model. Accordingly, at 120 Hz the tool wear has no influence. However, the influ-
ence rises drastically at higher oscillation frequencies. Cutting and passive force
coefficients of the inner chip thickness modulation increase strongly. The influ-
ence of the outer chip thickness modulation becomes less via the frequency.
Oscillation frequency
f = 120 Hz f = 600 Hz f = 1200 Hz
Amplitude coeff. [N/mm2]
4000
Test parameters
3200 RRoc vc = 200 m/min fU = 0.1 mm/U
RRic Material: C45E tool: R245 M-PM
2400
1600
800
0 R
Rop
45 R
Rip
0
ϕϕoc
Phase coefficients [°]
-45 ϕϕic
90
45
0
ϕop
-45 ϕip
-90
0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2
Wear mark width [mm] Wear mark width [mm] Wear mark width [mm]
Fig. 8.13 Influence of flank wear on the cutting of steel [11]
The phase shift of the passive force of the inner chip thickness modulation also
changes. For higher frequencies, the process can be damped and stabilized by
means of the effect of the time-shifted modulation of the inner and outer chip
thickness. A visualization of the effects of the process damping is given in Fig.
8.14. Here, simulation calculations for surface-overlapping between tool and
workpiece are carried out for a planar cutting process at a wavelength of 1.7 mm
and vibration amplitude of 0.05 µm. It is assumed that the collision between clear-
ance flank and workpiece surface causes additional forces, which lead to the ef-
fects shown in Fig. 8.13. This explains the positive phase shift of the passive force
to the inner chip thickness modulation. The force precedes the chip thickness as
there is already clearance flank contact with the inward movement of the cutter.
Thus, the cutter is already subject to the maximum reactional force before reach-
ing the deepest point.
Vibration R = 50µm
α = 11°
α = 11° α = 11°
Width of wear mark
= 0,05mm
2.5 Overlap area 2.5 2.5
Vc [10-3 mm²]
2.0 2.0 2.0
1.5 1.5 1.5

Vibration
1.0 1.0 1.0
0.5 0.5 0.5
0 0.4 0.8 1.2 1.6 0 0.4 0.8 1.2 1.6 0 0.4 0.8 1.2 1.6
Vc Cutting path [mm]
Cutting path [mm] Cutting path [mm]
Fig. 8.14 Flank face contact with a rounded and a worn tool [11]
Comprehensive cutting experiments with steel and aluminum on different ma-

chine systems were carried out within the project of the SPP 1180 for the verifica-
tion of the complex extended cutting force model. Fig. 8.15 shows an example of a
stability chart, which was identified on a vertical three-axes milling machine for
processing C45E steel. For modeling the machine dynamics, frequency response
functions measured with the machine at rest have been applied. The rotational beha-
vior of the spindle bearing system, as shown in the already presented auxiliary spin-
dle test-rig, has almost no influence on the dynamic compliance behavior in the low-
er range of rotation speeds and can therefore be ignored for the processing of steel.
The newly-developed model with the defined C45E parameters is compared
with the conventional linear process model within the scope of the process para-
meters recommended by the tool manufacturer. The comparison between the
measured and simulated stability chart in Fig. 8.15 shows that in the regions with
large deviations between the simulation results the complex extended process de-
piction leads to a clear improvement of the results.
10 Comparison
Cutting depth ap [mm]
Cutting tests of measurement

8 Stable cut and simulation
Chatter
6 Chatter frequency Cutting tool:
Sandvik
4 R390-32T16-11M
D = 32 mm
3 cutting edges
2
Material:
0 Simulation with steel C45E
Chatter frequency fR [Hz]
500 linear force model

Process:
400 ae = 32 mm
Simulation with advanced fz = 0.12 mm / tooth
300 complex force model
200
100
0
900 1100 1300 Spindle speed [1/min] 1700
Fig. 8.15 Stability chart in the machining of steel [11]

The deviations of the cutting experiments in the region of low revolutions can
thus be derived from different effects. At low cutting speeds, damping effects can
become more effective as these are taken into account in the parameter identifica-
tion. Different friction effects in chip formation, which may be responsible for
such deviations, are described by Dudzinski and Molinari [10]. In addition, espe-
cially for low cutting speeds, adhesion effects with particles of the workpiece ad-
hering to the tool also influence the dynamics of the cutting process and thus the
stability limit.
8.4 Method of Stability Simulation
Apart from the accuracy of the modeling and the input parameterization, the me-
thod of stability simulation itself has an influence on the quality of the simulation
results. The coupling of the modeling approaches described in the previous
chapters for process and machine behavior in time domain is described in the fol-
lowing. For the simulative investigations described here, the dynamic compliance
behavior of the machine structure is used as input parameter in form of a matrix of
direct and cross-frequency response functions. In the multi-dimensional stability
simulation, these frequency response functions (FRF) are taken into account by
means of the projection of process forces into the Cartesian spatial directions of
the machine system. However, these FRF cannot be processed by time-based si-
mulation algorithms and must therefore be treated as differential equations. The
overall behavior of the machine structure is thus depicted as an accumulation of
damped single-mass oscillators. All single-mass oscillators belonging to one direc-
tion are loaded with the same force in the simulation; the resulting deflections are
added together. The measured or simulated dynamic behavior of machine tools
can be described in this manner with a high degree of accuracy.
The dynamic cutting rigidity is coupled with the dynamic machine compliance
in the simulation model. Fig. 8.16 describes this coupling for a three-dimensional
milling process and several cutting edges. The model is illustrated in a simplified
manner concerning the consideration of the coordinate directions. Here, on the one
hand, a planar milling process with feed motion of the milling cutter in the x-y-
level is assumed. On the other hand, the dynamic cross compliances of the FRF
are ignored in the figure. The direction factors di,xd are set to zero while the respec-
tive cutting edge is not engaging into the workpiece. In this case, a force feedback
does not occur at this cutting edge. If a cutting edge is in engagement, the generat-
ed profile xdi is used for the cutting force calculation at the corresponding cut. The
profile xdi is delayed by the dead time Ti,i+1 between two cutting sequences and the
profile of the chip outer side is passed to the next cutter i+1. The value of the dead
times Ti,i+1 are the same for equally-divided cutters but can deviate from each oth-
er for a non-uniform pitch. Here, the cutting forces are determined over the linear
dynamic cutting force coefficient kcb by multiplying the dynamic part of the chip
thickness with kcb and the chip width b. The angle of the cutting force compared
to the chip thickness change (ßorth) is included in the direction factor dF,Fj, which
projects process forces from the coordinate system of the cutting edge to the ma-
chine coordinate system.
On the right side of Fig. 8.16, the sub-model of the second cutting edge is
shown in a sectional view. The lower part of the picture also displays the sub-
model of the same cutting edge but the model is extended such that the developed
complex force model can be shown. Here, xd1 and xd2 serve as input dimensions.
However, both are taken into account separately and multiplied with the force
coefficient Rpq for calculating the dynamic cutting and passive force and shifted
timewise by the phase angle φpq. The phase shift in the shown control loop is then
also taken as a dead time element. The dead time necessary for the depiction of the
phase shift is therefore dependent on the frequency of the respective active mode
shape.
Process model milling with n cutting edges Depiction of one cutting edge
Fx xges xd1
GFx,x Cutting edge 2
Fy yges Fx u2 xd2 xges
GFy,y dF,Fx b kcb dx,xd
Fy - yges
u1 xd1 dF,Fy dy,xd
dF,Fx b kcb - dx,xd
xd2
dF,Fy dy,xd
Essential enhancement for the complex force model
T1,2 xd1
Cutting edge 2, complex force model
u2 xd2 Fx F xd2 xges
dF,Fx b kcb dx,xd dFc,Fx c b - Fic ϕic Ric dx,xd
- Fy yges
dF,Fy dy,xd dFc,Fy Fip dy,xd
ϕip Rip
T2,3 Foc
With di,xd = 0 ϕoc Roc
dFp,Fx
if cutting edge i Fop
is not in contact dFp,Fy b - ϕop Rop
Fp
Tn-1,n
xdn xd2
un
dF,Fx b kcb - dx,xd
 Assumption of a plane milling process
dF,Fy dy,xd
 Neglecting the cross frequency response
Tn,1  Neglecting the nominal depth of cut
Fig. 8.16 Enhancement of the dynamic cutting force model [11]
The resulting dynamic force parts are summarized into dynamic cutting and
passive force. They are projected by means of cutting angle-dependent direction
factors into the Cartesian directions x and y of the machine tool and act again on
the structural compliance. The simulation structure was extended in order to set
the correct dead time to depict the phase angle φpq for a resonance frequency and
to take into account the identified influence of the frequency dependency on the
dynamic process force progression. On the right, Fig. 8.17 shows that simulation
loops must be created for the different mode shapes of the machine tool. Thus, an
interpolation of the model parameters is carried out for each mode shape and the
dead times for depicting the phase angle are calculated as well.
Dynamical compliance Individual consideration of each mode shape for

summarized in transfer function frequency dependent dynamical force coefficients
F x F
Consideration of n cutting edges x
GFx,x,Mode_1 1 x1
Cutting
F y edges F
F x F y 1 1- n y1
x ges x GFy,y,Mode_1
G Cutting
Fx,x
F y edges F x F
y ges y 2 x2
G 1- n GFx,x,Mode_2 Cutting
Fy,y edges
y F
2 1- n y2
GFy,y,Mode_2
Utilization of frequency dependent coefficients

x F
m xm
 One transfer function per mode shape GFx,x,Mode_m
y
Cutting
edges F
 Interpolation of the force coefficients GFy,y,Mode_m
m 1- n ym
regarding to thenatural frequencies

 Summation of the dynamic forces
Fig. 8.17 Enhancement of frequency–dependent model parameters [11]
8.5 Simulation of Processed Workpiece Surfaces

Beyond the prediction of process stability with a high efficiency, the PrimeCut
software environment developed by the WZL permits the calculation of machined
workpiece surfaces. Apart from the industrial relevance for improving the surface
topology of machined workpieces, the simulated surface structure presents the ba-
sis for the evaluation of the dynamic compliance behavior of machine tools after
processing has taken place. The identification of the oscillation frequency of the
machine structure is of particular interest. By means of suitable interpretation rou-
tines future conclusions are to be derived with regard to the dynamic weak points
of the machine structure, which otherwise are only possible using frequency re-
sponse analysis. A first step towards this is the simulation of processed workpiece
surfaces for known processes and the compliance behavior of the machine. For a
simulative depiction of the surface structure of an instable processed workpiece
the PrimeCut software tool uses the trajectory of the cutting edges of a milling
tool, which is made up of the milling rotation and the feed motion as well as of the
dynamic compliance of the TCP.
Fig. 8.18 illustrates a simulated surface structure based on the stability simula-
tion shown as an example in Fig. 8.15. A profile cut of the surface was metrologi-
cally, tactilely depicted and also determined by means of simulation. Fig 8.18
shows these profile cuts and also the results of a frequency analysis of the profiles.
The frequency is calculated from the feed motion of the mill referenced to the wa-
velength of the profile marks. Basically, the machining process of the surface pro-
file can be depicted. Apart from the dominant frequency peak at 4 Hz, a second
waviness at approximately 8 Hz is also correctly taken into account in the simula-
tion. However, the determined frequencies also show that the conclusion of a sur-
face profile to the basic oscillation frequency in milling due to the discontinuous
cutter engagement is not trivial.
15
Comparison
Surface profile [µm]
10 of measurement
and simulation
5 Cutting tool:
Sandvik
0 R390-32T16-11M
-5 D = 32 mm
3 cutting edges
-10
32 34 36 Groove length [mm] 40 Material:
steel C45E
Profile spectrum [µm]
6 Measurement
Simulation Process:
4 n = 1200 1/min
ae = 32 mm
Profile frequency referred ap = 5.4 mm
2 to feed rate fz = 0.12 mm / tooth
0 Chatter vibrations
0 5 10 Frequency [Hz] 20 at approx. 50 Hz
Fig. 8.18 Surface profile from instable machining of steel [11]
Based on systematic cutting experiments under instable process conditions and

the simulative verification, the influence of the oscillation frequency as well as the
oscillation amplitude on the formation of the surface structure in the remaining
time of the SPP 1180 are investigated and used to derive suitable conclusion pos-
sibilities. In this connection, Fig. 8.19 shows a test-rig structure, which can induce
dynamic weak points with defined a oscillation frequency and oscillation ampli-
tude in a milling process. With electro-hydraulic excitation, a clamped test-
workpiece can be dynamically excited orthogonally to the feed motion of the
milling tool.
The adjustment range of the vibration frequency and relative displacement is
sufficiently dimensioned with up to 300 Hz at an oscillation amplitude of 100 µm
in order to derive the behavior of the dynamic structural oscillations of machine
systems. The surface profiles generated on the test-workpieces (Fig. 8.19, right)
are evaluated based on the knowledge of the dynamic compliance behavior and
used as input for a simulative method reproducing the surface structure. The simu-
lation is carried out with the parameters rotation speed, number of cutting edges,
feed motion and mill diameter known from the process. The parameter determined
main frequency is varied in the model of the milling process. Simulation results
and measuring results are compared with each other with respect to the characte-
ristic dimensions, for instance the FFT of the surface. An exact description of such
a method is the subject of a current research work.
Inductive displacement sensor Dynamometer
Hydraulic actuator Work piece support
fchatter n
Spindle speed n = 700/min

Depth of cut ap = 3 mm
fz Excitation freq. fr = 120 Hz
Work piece
Disc springs
Spindle speed n = 900/min
Depth of cut ap = 3 mm
Flexure joints Base plate Exciter freq. fr = 120 Hz
Fig. 8.19 Test-rig for the emulation of dynamic weak points
8.6 Conclusion
Within the framework of the SPP 1180, the possibilities for the stability simula-
tion of High-Performance-Cutting (HPC) processes have been improved by inves-
tigating the input data for the parameterization of the simulation models and the
methods of simulation itself. The depiction of the machine tool dynamics and of
the dynamic behavior of the cutting processes has been extended.
The dynamic behavior of the machine structure was emphasized to improve the
stability simulation. The main effect identified in the HPC processing of alumi-
num alloy is the change of the dynamic behavior of spindle-bearing systems dur-
ing high rotation speeds. A measurement setup and the corresponding analysis
software were developed to evaluate this effect. Besides, the rotation speed de-
pendent dynamic behavior could be investigated in detail by means of an auxiliary
spindle test-rig comprising an accessible spindle shaft for modal analysis. The
investigations have shown that substantial deviations from the behavior of the
spindle at rest occur at higher spindle speeds. Although current tools for structural
simulation can represent individual rotor-dynamic effects, they are not yet suitable
for the identification of the structural behavior at high spindle speed. The possibil-
ities of the metrological determination created are thus necessary in order to carry
out stability simulations for milling processes at high spindle speeds.
Furthermore, investigations of the dynamic process behavior were presented. A
test rig was developed and built up for direct metrological determination of the
dynamic cutting rigidity. This permits the exact and independent variation of
process and oscillation parameters for the orthogonal milling process. A large
number of test-rig experiments with different cutting edges and workpiece mate-
rials were the basis for a presentation of the dependencies in the determination of
dynamic milling forces. Especially cutting with rounded or blunt cutters has
shown that the usual linear and real depiction of the dynamic cutting rigidity is not
sufficient as input parameter for stability simulation. Much rather, a phase shift
between force and chip thickness modulation in the cutting process, i.e. a complex
linking of both dimensions, must be taken into account. A corresponding model
has been presented and parameterized for several combinations of cutting edges
and materials. The use of conventional models has shown to be sufficient for cut-
ting aluminum with sharp cutters and high cutting speeds. The depiction and
processing of steel materials, however, must be carried out with the more complex
models.
Beyond the simulation of the process stability, the prediction of the surface pro-
file of instable machined workpieces can also be useful for the identification of re-
cursive weak points. An appropriate method should be derived for this purpose in
the remaining project time of the SPP 1180.
References
[1] Abele, E., Munirathman, M., Tschannerl, M.: Hochgeschwindigkeitsbearbeitung
(HSC) - technologische Leistungsfähigkeit und noch zu überwindende Hindernisse.
In: Proceedings CIRP Int. Conference High Performance Cutting, Aachen (2004)
[2] Altintas, Y., Eynian, M., Onozuka, H.: Identification of Dynamic Cutting Force Coef-
ficients and Chatter Stability with Process Damping. Annals of the CIRP, Jg.57, H.1,
S.371–S.374 (2008)
[3] Altintas, Y., Weck, M.: Chatter Stability of Metal Cutting and Grinding. Annals of
the CIRP, Jg. 53, H.2, S.619–S.642 (2004)
[4] Andrae, P.: Hochleistungszerspanung von Aluminiumknetlegierungen. Diss. Univer-
sität Hannover (2002)
[5] Brecher, C., Esser, M., Paepenmüller, F.: Motor Spindles for HPC: Testing and Chat-
ter Simulation. In: Proceedings 2nd CIRP Int. Conference High Performance Cutting,
Vancouver, Canada (2006)
[6] Brecher, C., Esser, M., Witt, S.: Interaction of Manufacturing Process and Machine
Tool. Annals of the CIRP, H. 58 (2009)
[7] Brecher, C., Witt, S.: Static, Dynamic and Thermal Behaviour of Machine Tools with
Regard to HPC. In: Proceedings CIRP Int. Conference High Performance Cutting,
Aachen (2004)
[8] Butz, F.: Gestaltung der Loslagerung von Werkzeugmaschinenspindeln. Shaker, Aa-
chen (2007)
[9] Clancy, B.E., Shin, Y.C.: A Comprehensive Chatter Prediction Model for Face Turn-
ing Operation Including Tool Wear Effect. International Journal of Machine Tools
and Manufacture Jg.42, S.1035–S.1044 (2002)
[10] Dudzinski, D., Molinari, A.: Metal cutting and high speed machining. Kluwer, Bosten
(2002)
[11] Esser, M.:: Stabilitätssimulation für das HPC-Fräsen. Dissertation. Aachen. RWTH
Aachen, Werkzeugmaschinenlabor, WZL (2010)
[12] Großmann, K., Mühl, A., Löser, M.: Prognose von Stabilitätsgrenzen für das Fräsen.
ZWF Jg.101, H.7/8, S.416–S.421 (2006)
[13] Kals, H.J.J.: On the Calculation of Stability Charts on the Basis of the Damping and
the Stiffness of the Cutting Process. Annals of the CIRP, S.297–S.303 (1971)
[14] Munirathnam, M., Kreis, M.: Das Gesamtpaket muss stimmen - Trendbericht: Sys-
temdenken bei der HSC / HPC - Zerspanung steigert die Produktivität. In: WB
Werkstatt und Betrieb, H. Bd 139 Heft 6, pp. S.42–S.50 (2006)
[15] Peters, J., von Vanherck, P., van Brussel, H.: Die Messung der dynamischen
Schnittkraftkoeffizienten. In: Fertigung, H. Bd 3 Heft 2, pp. S.57–S.65 (1972)
[16] Schulz, A.: Systeme zur Schwingungsdämpfung von Werkzeugmaschinen. In: VDI-
Fortschrittberichte, Jg. 651 pp. S.40–S.56 (2005)
[17] Spachtholz, G.: Erweiterung des Leistungsbereiches von Spindellagern. Apprimus,
Aachen (2008)
[18] Tlusty, J.: Analysis of the State of Research in Cutting Dynamics. Annals of the
CIRP, Jg. 27, H. 2, S.583–S.589 (1978)
[19] van Brussel, H.: Dynamische Analyse van het Verspaningsproces. Katholieke Un-
iversiteit Leuven, Dissertation (1971)
[20] Weck, M., Brecher, C.: Werkzeugmaschinen. Automatisierung von Maschinen und
Anlagen, vol. 6. Springer, Auflage (2006)
[21] Werntze, G.: Dynamische Schnittkraftkoeffizienten. Bestimmung mit Hilfe des Digi-
talrechners und Berücksichtigung im mathematischen Modell zur Stabilitätsanalyse.
RWTH Aachen, Dissertation (1973)
Chapter 9
Development of a Stability Prediction Tool
for the Identification of Stable Milling Processes
D. Hömberg, E. Uhlmann, O. Rott, and P. Rasper
Abstract. This chapter deals with a new mathematical model to characterize the
interaction between machine and workpiece in a milling process. The model consists
of a multi-body system representing the milling machine and a linear thermo-elastic
workpiece model. An extensive experimental analysis supported the development
of the governing model equations. A numerical solution strategy is outlined and
complemented by simulations of stable and unstable milling processes including
workpiece effects. The last part covers the development of a new algorithm for the
stability analysis of large milling systems.
9.1 Introduction
The interaction of process and structure is the main reason for the unwanted chat-
ter phenomena in milling. Hence, the determination of stable cutting conditions for
given structures and the design of more efficient milling machines are important re-
search fields in production technology. Since the mathematical description of model
components like machine, workpiece and process is well understood by now, the
main challenge is to integrate them into a coupled model and to simulate the result-
ing system with tailored numerical algorithms to reproduce the stability limits and
the dynamical characteristics of the real system correctly.
Accordingly, the goal of the present project was the development of a coupled
model to study the dynamics of milling processes by means of time domain sim-
ulations. To cope with a necessarily long simulation time the machine structure is
treated as a multi-body system and the workpiece as a thermo-elastic body. Compre-
hensive experimental studies performed in close cooperation between engineers and
mathematicians provided the basis for the mathematical modeling, the identification
of model parameters and the validation of the numerical results.
The main achievements of the present work are the consistent mathematical mod-
eling of a complex milling system including the mathematical analysis of the derived
equations [10], the numerical implementation of a time-domain simulation system
204 D. Hömberg et al.
and its experimental validation, experimental studies, which have revealed a tem-
perature dependency of the stability limits.
The chapter is organized as follows: In Section 9.2, we will present the differ-
ent experimental studies carried out during the modeling procedure, as well as the
parameter identification and for the validation of the simulation results. Section 9.3
covers the derivation of the model equations, the development of numerical algo-
rithms and the presentation of simulation results. The last section is devoted to some
concluding remarks including the discussion of perspectives and further issues.
9.2 Experimental Analysis
9.2.1 General Setup

Machining tests were performed for experimental determination of process stabil-
ity behavior and cutting forces. Hence, a commonly used measurement setup for
milling experiments according to [12] was implemented to measure the specific cut-
ting forces and to analyze the process behavior.
The tests were performed on a five-axis machining center type MAP LPZ 500,
which features linear drives for the linear axes. Aluminum alloy AlZnMgCu1,5 (EN
AW-7075) blocks were machined for the general experiments using a one-edge end
mill cutter of High-Speed Steel (HSS) with a diameter of d = 8.0 mm and a side rake
angle of γ f = 23◦. The size of the workpieces was 150 × 100 × 50 mm. The milling
tool was fixed in a heat-shrinking tool holder with a HSK-A 63 mounting shank.
The workpiece was mounted on a 3-component-dynamometer type Kistler 9257A.
The dynamometer was connected to three charge amplifiers type Kistler 5011. For
the acquisition of the data, a measuring board from National Instruments with a
maximum sampling rate of 500 kHz and a resolution of 16 bit was applied. Figure
9.1 schematically displays the measurement setup.
heat shrinking 3 charge amplifiers PC with measuring

toolholder board
end mill cutter
work piece
dynamometer
machine table
Fx Fy Fz
Fig. 9.1 Experimental setup for machining tests

9 Development of a Stability Prediction Tool 205
9.2.2 Experiments for Parameter Identification

9.2.2.1 Measurement of Structural Dynamics
The identification of unknown model parameters is essential for the prediction of
process stability. Therefore, the measurement of the dynamic behavior of the ma-
chine tool structure is necessary. With a similar approach as Faassen et al. [12],
the dynamic characteristics of the machine structure and cutting tool can be de-
duced from measured mobility-frequency-response functions at the tool center point
(TCP). The tool was excited in x and y-direction of the machine tool coordinate sys-
tem using an impact hammer (Kistler 9722A500), while the machine tool was in an
idle state. The bandwidth was 5, 000 Hz. The response measurement in y-direction
was realized with a laser vibrometer (Polytec OFV 303). Therefore, errors due to the
additional mass of an accelerometer near the TCP were avoided as stated in [11].
9.2.3 Experimental Stability Analysis

9.2.3.1 Machining Tests for the Analysis of Process Stability and Cutting
Forces
The considered process in the machining tests was peripheral end milling. Flutes
with full immersion of the cutter (ae = 8 mm) were machined into the work-
piece. During the process, the cutting force components Fx , Fy and Fz according
to the machine coordinate system were measured at the workpiece using the 3-
component-dynamometer. The spindle speed was varied between n = 15, 800 rpm
and n = 19, 800 rpm. A feed rate per tooth of fz = 0.2 mm was set for each spin-
dle speed. The cutting depths were gradually increased with 0.5 mm increments
starting at 0.5 mm until the cutting process became unstable. Based on the data cap-
tured during the cutting tests the process stability behavior could be analyzed by the
generation of stability charts. The process data corresponding to stable machining
processes was used to determine cutting coefficients according to [23]. A sample
rate of 50 kHz was used for data acquisition.
9.2.3.2 Identification of Unstable Milling Processes

A method to distinguish stable and unstable process states is necessary for the anal-
ysis of the stability behavior of cutting processes. The regenerative effect is the main
reason for stable cutting processes to become unstable. Varying cutting forces lead
to self-excited vibrations and an additional relative shift between tool and work-
piece. These displacements create a characteristic waviness on the workpiece sur-
face. Thus, every following cut results in a chip-thickness modulation, which is
amplified by the machine tool vibrations [3], [2], [28], [33]. The characterizing fre-
quencies of these self-excited vibrations are called chatter frequencies. The domi-
nant chatter frequencies are generally located close to the eigenfrequencies of the
considered machine tool structure. Chatter leads to a worse product quality because
of additional self-excited vibrations. Therefore, it is necessary to define a stability

criterion in order to identify unstable milling processes and maintain a good product
quality.
The described effect makes it possible to identify process instabilities by con-
siderating force signals in frequency domain. Other possibilities are the analysis of
vibration signals near the TCP, the resulting surface roughness, the occurrence of
chatter marks and noise measurements during the machining process [12]. In this
research work, the recorded cutting force signals were transformed into frequency
domain using Fast-Fourier-Transformation (FFT) according to [3], [13] and [19] in
order to identify the dominant frequencies and related amplitudes. Figure 9.2 shows
the spectrum of the force signals for a stable and and an unstable milling process.
Since a one-edge end mill cutter was used the spindle speed of n = 17200 rpm
corresponds to excitation frequencies, which are multiples of f = 287 Hz.
Stable Process (ap = 1.5 mm)

Unstable Process (ap = 2.5 mm)
Excitation Frequencies
60
Relative force amplitude FRel
Tool:
N End Mill Cutter
z = 1; d = 8.0 mm
mm
Work piece:
Aluminium Alloy
30 EN AW-7075
Process Parameters:
15 ae = 8.0 mm
fz = 0.2 mm
n = 17200 rpm
0
0 750 1500 Hz 3000
Frequency f
Fig. 9.2 FFT of stable and unstable cutting processes
According to Sims [27], stable and unstable process states were determined by
the investigation of the ratio between the absolute value of the largest force am-
plitude, which can be allocated to the excitation frequency, and the largest force
amplitude of the occurring chatter frequency. If the ratio exceeds a value of 10 %,
the corresponding process is classified as unstable. In addition, the resulting surface
roughness, the occurrence of chatter marks and noise during the machining pro-
cess were also taken into account. A detailed description of the procedure is given
in [29].
9.2.3.3 Results of the Stability Analysis

The measured cutting force signals were analyzed using the illustrated method for
the identification of unstable cutting processes in order to generate stability plots
and determine stability limits. Figure 9.3 shows the resulting stability plot.
Calculated Stability Limit

Unstable Process
Stable Process
4 Tool:
End Mill Cutter
Cutting depth ap
mm z = 1; d = 8.0 mm
Work piece:
Aluminium Alloy
2 EN AW-7075; T = 25oC
Process Parameters:
1 ae = 8.0 mm; fz = 0.2 mm
0
15600 16700 17800 rpm 20000
Spindle speed n
Fig. 9.3 Stability plot for a feed-rate-per-tooth of f z = 0.2 mm
The variation of the stability limit is rather small in the analyzed range of spindle
speed. The maximum value is a p = 2 mm at spindle speeds around n = 17, 000 rpm
and n = 19, 400 rpm. For other spindle speeds, a p = 1.5 mm is the measured stability
limit.
The graph depicted in Figure 9.3 has been computed using the standard stabil-
ity prediction method proposed by Altintas et al. (see e.g. [2] or [1]). A mobility-
frequency-response function measured by means of hammer excitation and cutting
force coefficients determined from cutting force measurements under stable cutting
conditions have been provided as input data.
9.2.3.4 Possible Sources of Measurement Errors in the Process Analysis

The large gap between the predicted and the measured stability limits can probably
be explained with difficulties in the measurement of the frequency-response function
at the tool center point. Beyond that, the dynamometer which has been mounted be-
tween machine table and workpiece, might also affect the dynamical characteristics
of the structure. Both effects have been analyzed in [22]. The frequency-response
function of the work piece changes significantly, if the workpiece is not directly
mounted on the machine table but on the dynamometer. However, in both cases,
the average value of the measured workpiece frequency-response function is about
two orders of magnitude smaller than the corresponding value at the tool center
point [22]. Consequently, the effect of the dynamometer has been assumed to be
negligible.
However, recent results of machining tests have shown that the dynamic behavior
of the dynamometer-workpiece combination seems to have a much larger impact on
process behavior as expected. Although the amplitudes of the response functions
are very small, current experimental process analysis has shown that the shifting
of eigenfrequencies by use of a dynamometer seems to play a major role for the
stability behavior. Alternatively, the work piece accelerations during the machining
tests could be monitored to detect unstable processes. By defining a new stability
criterion this method would allow to remove the dynamometer and thus refine the
chatter detection.
9.2.4 Workpiece Effects in Milling

9.2.4.1 Experimental Setup and Measurement Procedure
The dynamic process behavior is mainly influenced by the spindle and cutting tool of
the machine tool system. Apart from that major influence, the geometry of the work-
piece also has an impact on the process [5], [8]. The effect of different workpiece
geometries on process stability was analyzed with an almost equal measurement
setup and experimental procedure as before (see Section 9.2.1 and 9.2.2).
In this case, a more flexible T-shaped workpiece compared to the blocks with a
size of 150 × 100 × 50 mm in the machining tests before was analyzed. The size
of the vertical bar was changed to obtain numerous and different geometries. The
width of the bar was varied between t1 = 10 mm and t2 = 20 mm and the height
between h1 = 50 mm and h2 = 150 mm to realize different dynamic behavior of
the workpieces in terms of flexibility. The T-shaped workpiece was mounted upside
down on the dynamometer so that the bar was in a vertical position. Flutes with
full immersion of the same end mill cutter were milled into the T-shaped workpiece
beginning at the top of the bar. Considering the stability plot above (Figure 9.3)
stable process states were analyzed using different workpiece geometries.
9.2.4.2 Analysis of the Influence of the Workpiece Geometry on Process

Stability
Machining test were carried out using different T-shaped workpieces. The figure
points out the huge impact of the workpiece geometry on process stability behavior.
With the use of the stability criterion described in Section 9.2.3.2, the considered
process state was classified as stable using the aluminum block as all relevant am-
plitudes in the frequency domain are related to the excitation with a frequency of
f = 267 Hz.
Stable Process (block)

Stable Process (bar: h =60 mm; t = 10 mm)
Unstable Process (bar: h =140 mm; t = 10 mm)
Excitation Frequencies
20
Relative force amplitude FRel
Tool:
End Mill Cutter
N
z = 1; d = 8.0 mm
mm Work piece:
Aluminium Alloy
10 EN AW-7075
Process Parameters:
5 ae = 8.0 mm; ap = 1.0 mm
fz = 0.2 mm
n = 16000 rpm
0
0 750 1500 Hz 3000
Frequency f
Fig. 9.4 Comparison of stability behavior considering different geometries.
The variation of the workpiece geometry leads to a varying dynamic process

behavior. While the machining process remained stable for a bar with a height of
h = 60 mm the amplitudes of the vibrations became higher in the frequency domain.
When a bar with a height of h = 140 mm was machined the cutting process became
unstable. Here, dominant chatter frequencies arise in the frequency domain apart
from the excitation frequency of f = 267 Hz.
The results show the big impact of the workpiece geometry and thus the effect of
the workpiece dynamics on process stability. It shows that there is need to consider
workpiece dynamics in stability prediction. Thus, the stiffness and damping of the
workpiece have to be analyzed and linked to the dynamic behavior of the machine
tool structure for the calculation of stability plots.
9.2.5 Temperature Effects

9.2.5.1 Experimental Setup and Measurement Procedure
To analyze the temperature influence on process stability and cutting forces the
measurement setup in Figure 9.1 was adjusted by using workpieces with differ-
ent pre-heating conditions. The pre-heating of the workpiece was realized with a
heating plate by Horst GmbH with three heating cartridges and a heater power of
P = 300 W. Thus, temperatures of to T = 200 ◦ C were achieved. The plate was
triggered by a temperature regulator device type HT MC1.
The workpiece was warmed-up to temperatures between TW P1 = 25 ◦ C and
TW P2 = 100 ◦ C, positioned upside down, by means of the described heating plate
before the machining started. In order to ensure homogeneous heating conditions
during the machining tests thermocouples type K were fixed on the sides of the
workpiece to control the local work piece temperature. In addition, the temperature
was monitored using a thermal imaging camera type Infratronics IR600. Due to this
approach it was possible to execute cutting tests with defined and homogeneous
heating conditions (Δ T = 1 K) of the workpiece.
9.2.5.2 Temperature Influence on Measurement Accuracy

The dynamometer is a piezo-electric sensor. For the analysis of thermal effects
on process stability and cutting forces a possible source for measurement er-
rors is the change in thermal conditions. In order to quantify these errors the
temperature-dependency of the three-component-dynamometer and the influence on
the measurement results were analyzed. The dynamometer was gradually heated to
temperatures from TW P1 = 25 ◦ C to TW P2 = 100 ◦ C. Then, force measurements were
performed using defined force signals. First, the dynamometer was loaded with de-
fined masses in a static way; and the arising differences between the measured force
signals and normal forces were determined. In a second step, an impact hammer was
used excite of the dynamometer at different temperatures. The measured amplitudes
and lengths of the force impulse signals by the hammer were recorded and subse-
quently compared with the measured signals of the dynamometer. The warming of
the hammer can be neglected due to the temporally short contact between platform
and impact hammer.
Only a small measurement inaccuracy of the dynamometer of about 1 % for the
static case and 0.5 % for the dynamic case could be identified as a result of these
approaches. Hence, a thermal decoupling between dynamometer and workpiece for
the later experiments was not required. However, the variance was considered later
in the analysis.
9.2.5.3 Analysis of the Influence of Temperature on Specific Cutting Forces

Linear cutting force models relate the cutting cross-section given by a p and h to
the cutting forces via an empirical constant; the so-called specific cutting force
coefficients. These coefficients depend on cutting speed, workpiece material and
cutter geometry (see e. g. [1], [31]). Since the workpiece material properties, i. e.
especially the yield stress, and the contact conditions change with temperature, it
is expected that the cutting force coefficients decrease for higher workpiece tem-
peratures. The effect of temperature on specific cutting forces has been analyzed
employing the measurement setup and experimental procedure presented in Section
9.2.1 and 9.2.2 in order to confirm this assumption. As suggested in [1], the pro-
cess parameters (spindle speed n = 16, 200 rpm, cutting depth a p = 1.5 mm and
feed-rate-per-tooth of fz = 0.2 mm) were been chosen such that the corresponding
process was stable (see Figure 9.3). The specific cutting force component Kc was
calculated as shown in [1] or [17]. Figure 9.5 shows the influence of different pre-
heating states of the workpiece between TW P1 = 25 ◦ C and TW P2 = 100 ◦ C on the
specific cutting force coefficient Kc .
Specific cutting force Kc 800 Tool:

End Mill Cutter
N
z = 1; d = 8.0 mm
mm2 Work piece:
Aluminium Alloy
600 EN AW-7075
Process Parameters:
500 ap = 1.5 mm
ae = 8.0 mm
fz = 0.2 mm
400
o
25 50 C 100
Workpiece Temperature TWP
Fig. 9.5 The dependency of the specific cutting force Kc on the preheating of the workpiece
A non-linear relationship between the workpiece temperature and the specific

cutting force coefficient Kc can be observed. The specific cutting force decreases
with increasing workpiece temperature. The decay is about 16 % in a temperature
range of TW P1 = 25 ◦ C to TW P2 = 50 ◦ C. The decay is less important but still visible
for higher workpiece temperatures. Based on these observations a new cutting force
model has been developed, which incorporates the effect of the workpiece tempera-
ture on the specific cutting force. Since the diagram illustrated in Figure 9.5 exhibits
an exponential decay, a similar relation was chosen for the modified specific cutting
force (9.9) presented in Section 9.3.4.
9.2.5.4 Results of the Stability Analysis for Different Workpiece

Temperatures
Experimental stability plots for different pre-heating conditions of the workpiece
were created with use of the described measurement procedure and the illustrated
method for the identification of unstable milling processes (see Figure 9.6).
Due to the pre-heating of the workpiece within the temperature range between
TW P1 = 25 ◦ C to TW P2 = 100 ◦ C, an increase of stability limits of at least Δ a p =
0.5 mm in the whole rpm-range with a maximum of Δ a p = 2.0 mm at n =
19, 400 rpm can be determined. Only at spindle speeds between n = 17, 800 rpm
and n = 18, 400 rpm is there no rise in the stability limits. In conclusion, higher
work piece temperatures have a stabilizing effect on the process behavior.
Because of the pre-heating of the workpiece, higher temperatures occur in the
cutting zone, which leads to a softening of the workpiece material so that the part of
the cutting force related to plastic deformation decreases. Partially, the temperatures
can reach such high levels that melting of the machined aluminum can be observed.
The viscous interface between chip and tool improves the chip flow, which results
in decreasing cutting forces [25], [4], [30], [14], which finally leads to increasing
stability limits.
Stability limit T = 100oC

5
Tool:
mm End Mill Cutter
Cutting depth ap
z = 1; d = 8.0 mm
3 Work piece:
Aluminium Alloy
2 EN AW-7075
T = 25…100 oC
1 Process Parameters:
ae = 8.0 mm; fz = 0.2 mm
0
15600 16700 17800 rpm 20000
Spindle speed n
Fig. 9.6 Stability limits in dependency of the temperature
9.3 Coupled System and Simulations

9.3.1 Modeling Concept
The goal of the presented research project was the development of a complex sim-
ulation system involving the dynamical characteristics of workpiece and machine
structure to study the dynamics of milling processes by means of time-domain sim-
ulations. Unstable processes can be identified by analyzing either the evolution of
the uncut chip thickness or the simulated cutting force spectrum. In both cases, the
precision of the predicted stability limit increases with the length of the simulated
time interval. However, due to the presence of high characteristic frequencies the
time-step size is strictly limited, which finally leads to unacceptable computation
times, especially for models with many degrees of freedom. In view of these prob-
lems, modeling the machine structure as a multi-body system and the workpiece as
a thermo-elastic work body seemed to be the best trade-off between accuracy and
efficiency.
From a macroscopic point of view, the largest part of the workpiece behaves like
a thermo-elastic body. Plastic deformations usually occur only in regions close to the
cutting edges and can therefore be incorporated by employing an empirical cutting
force model. Due to these simplifications, the material removal cannot be simulated
directly and has to be approximated by a heuristic approach. For milling processes,
where the difference between exit and entry angle is smaller than the pitch angle
of the cutter, such an approach can be constructed by applying a method of steps.
In each step the workpiece reference domain is considered to be constant and the
system is solved until the cutting edge leaves the workpiece. Before the next cutting
edge starts cutting, we construct a new workpiece reference domain based on the
previously computed solution and pursue the simulation. Consequently, the model-
ing and the simulation process consists of two parts. While the first part deals with
the phenomena occurring during one tooth period, the second part focuses on the
construction of a series of workpiece reference domains and thus on an implemen-
tation of the method of steps.
9.3.2 Multi-body System

The development procedure of a multi-body system representing a machine struc-
ture can be summarized as follows. The first step, an experimental modal analysis,
provides the eigenfrequencies and the corresponding mode shapes. Next, consid-
ering the measured mode shapes, the rigid bodies for the model can be defined.
Finally, after having implemented the mathematical model, the remaining free cou-
pling parameters have to be identified by comparing the experimentally and numer-
ically determined frequency-response functions for several points on the machine
structure using a least-squares approach. Based on the experimental data and addi-
tional FEM simulations a multi-body system has been developed, which matches
the measured frequency-response function. The final model, which is composed of
cutter, tool holder, two spindle segments, headstock, x-slider and a frame moving in
y-direction with respect to the fixed machine base, is illustrated in Figure 9.7.
The equation of motion describing the dynamics of the system can be summa-
rized by the following expression
M(t, q)q̈ = fI (t, q, q̇) + fE (t, q, q̇, η ) + gE (t, q, F) (9.1)
with gE representing the external forces and torques applied via joints on each body.
The parameter vector η contains the free coupling parameters, such as joint stiffness
or joint damping, while M denotes the state dependent mass matrix and fI represents
the inertia forces. The last term incorporates the cutting force vector F acting on the
cutter. The generalized coordinates q describe the relative motion between the rigid
bodies. Due to the presence of pre-defined parameters, such as feed and spindle
rotation speed the time appears explicitly in the expressions for M, fI , fE and gE .
Since the system is organized in a tree-like structure an iterative algorithm (c.f. [7])
can be applied to evaluate the equations of motion during a time-domain simulation.
9.3.3 Thermo-elastic Workpiece Model

The largest part of the workpiece behaves like a thermoelastic body. Plastic defor-
mations usually occur only in regions close to the cutting edges. In the framework
of a macroscopic description, the cutting forces F and the heat produced during
the chip formation can be modeled by employing an empirical approach. A volume
force sm
E occurring on the right hand side represents the cutting forces acting on
the workpiece. Similarly, a distributed heat source seE models the heat flux into the
workpiece. Note that both functions sm e
E and sE depend on the cutting forces and thus
headstock
x-slider
y-slider
upper
spindle
lower
spindle
toolholder
z y
x tool
Fig. 9.7 Multi-body system representing the milling machine
on the solution of the coupled system. As shown in [15], the equations of thermo-
elasticity read
ρ utt = div(σ ) + sm
E, (9.2)
2
σ = λ tr(ε )I + 2με − 3K αΘ I with K = (λ + μ ), (9.3)
3
1
ε= ∇u + (∇u)T , (9.4)
2
ρ cvΘt = κΔΘ − 3K α T0 div(ut ) + seE (9.5)
where λ and μ are the Lamé constants, ρ denotes the mass density, T0 the reference
temperature, cv the specific heat, κ the heat conductivity and α is the thermal expan-
sion coefficient. The function u represents the deformation field, I identity matrix
and tr(.) denotes the trace operator.
The heat equation has been formulated in terms of the deviation Θ from the
reference temperature T0 . Thus, the actual temperature T is given by T = Θ + T0 .
The workpiece is fixed on a large and rigid machine table, a configuration, which can
be approximated by imposing Dirichlet boundary conditions on the corresponding
part of the workpiece boundary, i.e.
u(t, x) = 0 for x ∈ ΓT , (9.6)

Θ (t, x) = 0 for x ∈ ΓT . (9.7)
9.3.4 Coupling and Material Removal Models

9.3.4.1 Coupling of Process and Structure Models during One Tooth Period
The cutting forces acting on cutter and workpiece can be described by empirical
models involving the uncut chip thickness, i. e.
F̂ = a p K̂(Tce ) h, (9.8)
where a p denotes the depth of cut. The vector K̂ denotes an empirical parameter, the
so called specific cutting force, which is usually assumed to be constant or depend-
ing on the cutting speed. However, as shown in Section 9.2.5, the cutting forces
decrease due to higher workpiece temperatures. This effect has been incorporated
into the cutting force model by multiplying the constant specific cutting force vector
std
K̂ = [K f , Kc , K p ]T by an empirical function involving the work piece temperature
Tce at the cutting edge, i. e., we define
(cTce )−b ,
std
K̂(Tce ) = K̂ (9.9)
with further fit parameters b and c. The uncut chip thickness h generally depends
on the current position of the cutting edge, the workpiece deformation and the
shape of the workpiece surface created by the preceding tooth. In the mathemati-
cal models delay terms are often used to incorporate the effect of the surface shape
(see [10]). However, to allow for a numerically stable and realistic coupling to the
workpiece, an alternative approach is pursued here. The workpiece surface shall
be constructed employing a real material removal model similar to the approaches
presented in [9], [32].
To derive a formula for the uncut chip thickness, we assume at first that the work-
piece reference configuration ΩR is given and show how the uncut chip thickness
can be derived from the current cutter position given by the solution of (9.1) and
the displacement field corresponding to (9.2). Note that for presentation purposes
we consider only the special case that the cutter axis is parallel to the z-axis of the
workpiece reference frame. In the general case, additional transformations depend-
ing on the solution of (9.1) have to be applied to compute the vector components in
the reference frame of interest.
As illustrated in Figure 9.8, a point on the cutting edge can be characterized
by the vector rce = r̃ce (t, q, z) = rca + rae . Note that the dashed lines in Figure 9.8
represent the ideal tooth path without machine and workpiece deformations. The
current workpiece domain Ω (t) is given by the reference configuration ΩR and the
displacement field corresponding to (9.2), i.e. x = X + u(t, X) with x ∈ Ω (t) and
X ∈ ΩR . As shown in [10], [23] and indicated by the bar labeled ’h’ in Figure 9.8,
the uncut chip thickness is the distance between a point ’ce’ on the cutting edge and
the workpiece surface measured in the direction of rae . Mathematically, this can be
formulated as follows

0 / Ω (t),
if rce ∈
h(t, q, z, Ω (t)) = (9.10)
max
∗
h∗ otherwise,
h ∈H
with the set H being defined as

rae
H = x ∈ R+ | rce − x ∈ Ω (t) . (9.11)
rae
Combining the expression for the uncut chip thickness with the empirical cutting
force model (9.8) divided by a p gives a force per unit length, which possibly assume
different values on each z-level, i.e.
R̂ = K̂ h(t, q, z, Ω (t)). (9.12)
Since the components of the vector R̂ (see Figure 9.8) are given with respect to
the reference frame corresponding to the cutting edge, (9.12) has to be transformed
by means of an orthogonal transformation O(ϕ (z)) in the cutter reference frame to
compute the cutting force per unit length acting on the cutter, i. e.
R̃ = O(ϕ (z))R̂ = O(ϕ (z))K̂ h(t, q, z, Ω (t)), (9.13)
with ϕ (z) denoting the angle between rce and the x-axis of the cutter reference
frame. Integrating (9.13) along the cutter axis finally provides the resultant force
acting on the cutter.
y y
x z x
z Ω(t) Ω(t)
^
rca ^
Rf Rc
A(t,z)
rce
h yc h
cutting rae Δφ yc
φ(z) Δφ
edge rae
cutter xc cutter xc
axis axis
fz fz
Fig. 9.8 Uncut chip thickness (left) and definition of the volume force (right)
Similar to the resulting force on the cutter, the volume forces sm

E acting on the
right hand side of (9.2) also have to be calculated from the relative cutting force
given in (9.12). Applying a second orthogonal transformation Q(t, q), which de-
pends on the solution of (9.1) to (9.13) provides the components of the relative
cutting force in the workpiece reference frame, i. e.
R = Q(t, q)R̃ = Q(t, q)O(ϕ (z))K̂ h(t, q, z, Ω (t)). (9.14)
Since the presented model is a macroscopic approach, which cannot reproduce the
chip formation, the relative cutting forces have to be distributed over an area A(t, z)
located between cutting edge and workpiece surface. As illustrated in Figure 9.8,
for a given point on the cutting edge rae and a given level z the area A(t, z) is defined
as the set of all points in Ω (t) enclosed by the disc-ring segment with label ’A(t, z)’.
The segment is defined by the arcs through rae with radius D/2 and length Δ ϕ D
and through (1 − h/rae )rae with radius D/2 − h and length 2Δ ϕ (D/2 − h). Thus,
the volume force reads

0 if (x, y) ∈
/ A(t, z),
sE (t, x, y, z) =
m
R (9.15)
A(t,z) otherwise.
The heat conducted into the workpiece can be computed from the specific cutting
forces and the given cutting conditions (see [26]). If the shear angle appearing in the
corresponding expressions is estimated by an analytical formula as, for example,
proposed in [20], the total heat flux into the workpiece denoted by H(h, K̂) can be
written in explicit form. The corresponding term appearing on the right hand side
of (9.5) reads
0 if (x, y) ∈
/ A(t, z),
sE (t, x, y, z) = H(h,K̂)
e
(9.16)
A(t,z) otherwise.
9.3.4.2 Material Removal Model

As mentioned before, the workpiece surface can be constructed by employing a
material removal model. The main idea is to construct a volume based on the cut-
ting edge path and workpiece deformations, which can be subtracted from a given
workpiece domain by means of Boolean operations (see e.g. [32]).
To this end, recall that during one tooth period each point on the cutting edge
follows a certain path depending on the motion of the cutter. In a subinterval, some
points of the cutting edge penetrate the deformed workpiece surface and the cutter
is cuts. Thus, a workpiece deformation can be associated to each point rce of the
cutting edge in the workpiece domain Ω (t), i. e.
rce = Xce + u(t, Xce ), Xce ∈ ΩR . (9.17)
With solution X∗ce (t, rce ) of the above equation and the cutting edge points we define
a new point y, which corresponds either to the cutting edge or to the new shape of
the reference domain, i. e.

rce / Ω (t),
if rce ∈
y(t) = (9.18)
X∗ce (t, rce ) otherwise.
Monitoring these points y(t) ⊂ R3 during a time interval, which encloses

the actual cutting period, gives a set of points describing an open surface in the
three-dimensional space. From such a surface, we construct a set Ωc ⊂ R3 , which
represents the points travelled by the cutting edge and incorporating the workpiece
deformations. The new workpiece reference domain ΩRnew can be found by subtract-
ing the domain Ωc from the given workpiece reference domain ΩR , i.e.
ΩRnew = ΩR \ Ωc . (9.19)
The presented strategy leads to a series of workpiece domains, each incorporat-

ing the motion of the cutting edge and the corresponding workpiece deformations
during the preceding tooth path. Together with the expression for the uncut chip
thickness, the model leads to a non-linear system of coupled ordinary and partial
differential equations. The history of workpiece and cutter motion is stored in the
workpiece surface.
9.3.5 Numerical Algorithm, Implementation and Parameter

Identification
The main solution algorithm is composed of two parts. While the first part deals
with the solution of the coupled system on a constant reference domain, the second
part is focused towards the construction of a series of reference domains and the
corresponding initial conditions.
As shown in the previous sections, the process structure interaction leads to a
strong coupling of workpiece and machine model. In addition to the finite element
discretization of the workpiece equations, a tailored time-integration algorithm has
been developed for the coupled system guaranteeing a fully-implicit coupling of the
complete system. This strategy required a large programming effort. Since standard
finite element libraries do not allow the incorporation of multi-body systems, and
contrarily, the standard multi-body simulation packages do not provide any tools
to integrate coupled PDE-systems into the simulation environment, the numerical
implementation of both parts had to be developed in the framework of an in-house
library. The result is a powerful milling simulation system. Its main features are an
optimal resolution of the coupling effects by an efficient implicit time-integration
scheme, and as a consequence of the material removal model, the spatial resolution
of the machined workpiece surface. An example for the generated workpiece sur-
face employing a Dexel model similar to [34] is illustrated in Figure 9.9 for a stable
process and in Figure 9.10 for an unstable process, respectively. In addition, the pre-
sented approach allows for the identification of machine and workpiece parameters
by means of standard Gauss-Newton methods.
The identification process was carried out in close cooperation between experi-
menters and mathematicians. Based on the standard experiments, i. e. measurement
of mobility-frequency response functions, cutting forces and temperatures, the ex-
perimental and the numerical methods have been successively adjusted in order to
finally provide a realistic milling model.
0.25
uncut chip thickness / [mm]
0.15 0.25

0.2 0.2
0.1
0.15 0.15
0.1 0.1
0.05
0.05 0.05
0 0 0
1.6 1.7 1.8 1.9 2 0.031 0.0315 0.032 0.0325 0.033 0.065 0.066 0.067
time / [sec] x 10 −3 time / [sec] time / [sec]
Fig. 9.9 Example for the material removal simulation, i. e. the evolution of the workpiece
shape, and the corresponding uncut chip thickness in case of a stable cut
0.15 0.25 0.5

0.2 0.4
0.1
0.15 0.3
0.1 0.2
0.05
0.05 0.1
0 0 0
1.5 1.6 1.7 1.8 1.9 0.031 0.0315 0.032 0.0325 0.033 0.065 0.066 0.067
−3
time / [sec] x 10 time / [sec] time / [sec]
Fig. 9.10 Example for the material removal simulation, i. e. the evolution of the workpiece
shape, and the corresponding uncut chip thickness in case of an unstable cut
9.3.6 Simulations
The coupled model and the developed simulation algorithm provide a tool to inves-
tigate the characteristics of milling processes involving the dynamics of machine
and workpiece. Although the temperature effects presented in Section 9.2.5 can
be reproduced with the presented model, an illustration of the related phenomena
is beyond the scope of this work. Therefore, we focued on effects, which can be
simulated with an elastic workpiece model. Due to this restriction the workpiece
equations can be reduced to system (9.2)-(9.4) with a constant temperature, i.e.
Θ = 0. The damping effects in the workpiece were incorporated into the model
on the space discrete level, introducing a so-called Rayleigh damping term. Again,
the unknown damping parameters were determined by comparing simulated and
measured frequency-response functions. Recall that the vector of unknown param-
eters in (9.1) has been fitted so that the multi-body system reproduces the measured
frequency-response functions at the tool centre point (TCP). Moreover, the cutting
force model employed in [22] to compute the stability limit illustrated in Figure
9.3 is similar to (9.8). Consequently, a system composed of the machine model and
a rigid workpiece almost has the same stability limits as the system analyzed to
compute the red line in Figure 9.3.
As observed in the experiments, a supple workpiece structure or the wrong choice
of cutting depth and spindle speed may destabilize the milling process. In order to
display these effects numerically we considered three examples.
In the first simulation, the workpiece structure is rather stiff and the process
parameters have been chosen so that no chatter occurs (a p = 0.5 mm and n =
16, 400 rpm, see Figure 9.3).
50 0.3
rel. force spectrum / [N / mm]
40 0.25
0.2
30
0.15
20
0.1
10 0.05
0 0
0 1000 2000 3000 4000 5000 0 0.05 0.1 0.15
frequency / [Hz] time / [sec]
Fig. 9.11 Relative cutting force spectrum and evolution of the uncut chip thickness corre-
sponding to a stable cut
Figure 9.11 illustrates the spectrum of the cutting force in x-direction divided
by a p and the evolution of the uncut chip thickness. Both diagrams indicate that the
process is stable. No chatter peaks arise in the force spectrum and the uncut chip
thickness converges to the stationary evolution.
For the second example, the spindle speed has been set to n = 17, 800 rpm. In this
case, the process parameters correspond to unstable cutting conditions. Additional
chatter peaks occur in the relative force spectrum (spectrum of the cutting force in
x-direction divided by a p ) depicted in Figure 9.12. According to the considerations
0.4
rel. force spectrum / [N / mm] 50 Simulation

Experiment
40 tooth passing freq. 0.3
30
0.2
20
0.1
10
0 0
0 1000 2000 3000 4000 0 0.05 0.1 0.15
Fig. 9.12 Simulated and measured relative cutting force spectrum and evolution of the uncut
chip thickness corresponding to an unstable cut
60 0.4
rel. force spectrum / [N / mm]
50
0.3
40
30 0.2
20
0.1
10
0 0
0 1000 2000 3000 4000 0 0.05 0.1 0.15
Fig. 9.13 Relative cutting force spectrum and evolution of the uncut chip thickness corre-
sponding to an unstable cut due to a supple workpiece
outlined in Section 9.2.4.2 and 9.2.3.2 the additional chatter peaks clearly indicate
that the corresponding process is unstable.
The evolution of the uncut chip thickness illustrated in Figure 9.12 confirms this
result. In contrast to the evolution shown in Figure 9.11, the uncut chip thickness
does not converge to the stationary evolution but increases noticeably after a short
decay at the beginning and remains on a high level until the end of the simulation.
In the third example, the process parameters have been set to the same values as
in the first simulation (a p = 0.5 mm and n = 16400 rpm). The stiff workpiece uti-
lized in the first example has been replaced by a more supple structure. Especially
on top, the supple workpiece exhibits a high compliance, which can destabilize the
previously stable milling process.
Analyzing the evolution of the uncut chip thickness reveals that the process does
not converge to a stationary regime. Large workpiece oscillations lead to an increas-
ing uncut chip thickness and thus to increasing cutting forces. The additional chatter
peaks appearing in the relative force spectrum clearly indicate that the correspond-
ing process is unstable. In contrast to the second example, the highest chatter peak
is not located at 2, 834 Hz at but at 408 Hz. Thus, the weak spot in the structure
leading to chatter vibrations is different in the two examples.
9.4 Conclusions
The goal of this chapter was the development of a complex milling model to in-
vestigate effects of machine and workpiece structure on the stability of milling pro-
cesses. The experimental analysis in Section 9.2 reveals that the stability limits can
be increased by pre-heating the workpiece. Moreover, it illustrates how milling pro-
cesses can become unstable due to a lack of workpiece stiffness. The experiments
conducted for the identification of machine parameters display that the frequency-
response functions at the TCP strongly depend on the excitation method and vary
with the angular spindle position. The simulations in Section 9.3 clearly demonstrate
that the developed model is capable of reproducing the instability effects observed
in the experiments. For the first time a new stability analysis method allows the
determination of stability limits of large DDE-systems with periodic coefficients.
The results are promising and open up various directions for future research.
A challenging task would be to investigate the stability of milling processes with
respect to variations in the machine design.
Finally, from the application point of view, an efficient numerical tool for the
systematic derivation of stability diagrams is most desirable. The developed stability
analysis tool is a first step in this direction. The improvement of the numerics and the
exploitation of model reduction techniques to increase the efficiency of the method
will be subject to further research.
References
1. Altintas, Y.: Manufacturing Automation–Metal Cutting Mechanics, Machine Tool Vibra-
tions, and CNC Design. Cambridge University Press (2000)
2. Altintas, Y., Budak, E.: Analytical prediction of stability lobes in milling. Ann. of the
CIRP 44(1), 357–362 (1995)
3. Altintas, Y., Weck, M.: Chatter stability of metal cutting and grinding. Ann. of the
CIRP 53(2), 619–642 (2004)
4. Amin, A.K.M.N., Abdelgadir, M.: The Effect of Preheating of Work Material on Chatter
During End Milling of Medium Carbon Steel Performed on a Vertical Machining Center
(VMC). J. of Manuf. Sci. and Eng. 125, 674–680 (2003)
5. Budak, E.: Mechanics and dynamics of milling thin walled structures. Ph.D. Thesis,
University of British Columbia, Vancouver, Canada (1994)
6. Bueler, E.: Error Bounds for Approximate Eigenvalues of Periodic-Coefficient Linear
Delay-Differential Equations. SIAM J. Numer. Anal. 45(6), 2510–2536 (2007)
7. Bremer, H., Pfeiffer, F.: Elastische Mehrkörpersysteme. B. G. Teubner, Stuttgart (1992)
8. Campa, F.J., López de Lacalle, L.N., Lamikiz, A., Sánchez, J.A.: Selection of cutting
conditions for a stable milling of flexible parts with bull-nose end mills. J. of Mater.
Proc. Technol. 191, 279–282 (2007)
9. Campomanes, M.L., Altintas, J.: An improved Time Domain Simulation for Dy-
namic Milling at Small Radial Immersions. Transactions of the ASME (2003),
doi:10.1115/1.1580852
10. Chelminski, K., Hömberg, D., Rott, O.: On a thermomechanical milling model. Nonlin-
ear Analysis Real World Applications 12, 615–632 (2010)
11. Ewins, D.J.: Modal Testing: Theory and Practice. Res. Stud. Press LTD. (1986)
12. Faassen, R.P.H., van de Wouw, N., Oosterling, J.A.J., Nijmeijer, H.: Prediction of regen-
erative chatter by modelling and analysis of high-speed milling. Int. J. of Mach. Tools &
Manuf. 43, 1437–1446 (2003)
13. Faassen, R.P.H.: Chatter Prediction and Control for High-Speed Milling - Modelling and
Experiments. Dissertation, TU Eindhoven, Netherlands (2007)
14. Gatto, A., Iuliano, L.: Chip Formation Analysis in High Speed Machining of a Nickel
Base Superalloy with Silicon Carbide Whisker-Reinforced Alumina. Int. J. of Mach.
Tools & Manuf. 34, 1147–1161 (1994)
15. Haupt, P.: Continuum Mechanics and Theory of Materials. Springer, Berlin (2000)
16. Hughes, T.J.R.: The finite element method. Dover Publications, New York (2000)
17. Insperger, T., Stépán, G.: Stability of the milling process. Period. Polytech. – Mech.
Eng. 44(1), 47–57 (2000)
18. Insperger, T., Stépán, G.: Updated semi-discretization method for periodic delay-
differential equations with discrete delay. Int. J. Numer. Methods Eng. 61(1), 117–141
(2004)
19. Insperger, T., Mann, B.P., Surmann, T., Stépán, G.: On the chatter frequencies of milling
processes with runout. Int. J. of Mach. Tools & Manuf. 48, 1081–1089 (2008)
20. Lee, E.H., Schafer, B.W.: The Theory of plasticity applied to a problem of machining.
Journal of Applied Mechanics 18, 405–413 (1951)
21. Mann, B., Insperger, T., Bayly, P.V., Stépán, G., Schmitz, T.L., Peters, D.A.: Effects of
Radial Immersion and Cutting Direction on Chatter Stability in End-Milling. In: N.N.
(ed.) Proceedings of the International Mechanical Engineering Conference and Exposi-
tion (IMECE), New Orleans, LA (2002)
22. Rasper, P., Rott, O., Hömberg, D., Uhlmann, E.: Analysis of uncertainties in the stability
prediction for milling processes. In: Altintas, Y. (ed.) Proceedings, CIRP 2nd Interna-
tional Conference Process Machine Interactions (2010)
23. Rott, O., Rasper, P., Hömberg, D., Uhlmann, E.: A milling model with thermal effects in-
cluding the dynamics of machine and work piece. In: Denkena, B. (ed.) Proceedings, 1st
International Conference on Process Machine Interactions, PZH Produktionstechnisches
Zentrum GmbH, Hannover-Garbsen (2008)
24. Rott, O., Jarlebring, E.: An iterative method for the multipliers of periodic delay-
differential equations and the analysis of a PDE milling model. In: Vyhlidal, T. (ed.)
Proceedings of the 9th IFAC Workshop on Time Delay Systems, Prague (2010)
25. Schulz, H.: Hochgeschwindigkeitsfräsen metallischer und nichtmetallischer Werkstoffe.
Carl Hanser Verlag, München (1989)
26. Shaw, M.C.: Metal Cutting Principles. Oxford University Press, Oxford (2005)
27. Sims, N.D.: The self-excitation damping ratio: A chatter criterion for time-domain
milling simulations. J. of Manuf. Sci. and Eng., Trans. of the ASME 127(3), 433–445
(2005)
28. Uhlmann, E., Mense, C.: Analysis of the Milling Machine Operation Behaviour for Pro-
cess Stability and Machine Tool Optimisation. In: N.N. (ed.) Proceedings of 10th CIRP
Int. Workshop on Model. of Mach. Oper., Reggio Calabria, Italy (2007)
29. Uhlmann, E., Rasper, P.: Temperaturabhängiges Stabilitätsverhalten–Untersuchung des
Temperatureinflusses auf das Stabilitätsverhalten beim Umfangsstirnfräsen. In: wt Werk-
stattstechnik online, Düsseldorf, Germany, vol. 7(8), pp. 464–469 (2009)
30. Venuvinod, P.K., Lau, W.S., Narasimha Reddy, P., Rubenstein, C.: On the Formation of a
Fluid Film at the Chip Tool Interface in Rotary Machining. Ann. of the CIRP 32, 59–64
(1982)
31. Weck, M., Teipel, K.: Dynamisches Verhalten spanender Werkzeugmaschinen. Springer,
Berlin (1977)
32. Weinert, K., Surmann, T.: Geometric Simulation of the Milling Process for Free Formed
Surfaces. In: Weinert, K. (ed.) Simulation Aided Offline Process Design and Optimiza-
tion in Manufacturing Sculptured Surfaces, Witten (2003)
33. Zatarain, M., Munoa, J., Peigné, G., Insperger, T.: Analysis of the Influence of Mill Helix
Angle on Chatter Stability. Ann. of the CIRP 55(1) (2006)
34. Zeng, W., Peng, X., Leu, M.C., Zhang, W.: A Novel Contour Generation Algorithm
for Surface Reconstruction From Dexel Data. Journal of Computing and Information
Science in Engineering 7, 203–210 (2007)
Chapter 10
Synthesis of Stability Lobe Diagrams
K. Großmann and M. Löser
Abstract. Chatter vibrations during machining lead to poor workpiece surfaces and
increased tool wear. In the worst case, the tools and even the main spindle can be
damaged. Nowadays, the surface regeneration is considered to be the main effect
causing chatter instabilities. Regenerative chatter is initiated by repetitive tooth en-
gagement where the currently engaged tooth cuts the surface produced by the pre-
ceding tooth. In a stability lobe diagram (SLD), the stable and unstable areas are
separated by the graph of a critical cutting parameter plotted against the spindle
speed. Stability lobe diagrams can be used to optimize machining processes in
terms of maximizing material removal rate under stable cutting conditions. These
SLDs are computed by time domain simulations. However, this consumes a lot of
computational time. Thus, several time efficient algorithms in discrete time as well
as frequency domain have been developed in the last decades. This chapter scruti-
nizes under what conditions different algorithms in frequency domain can be
applied. The processes are separated regarding cutting conditions and dynamic be-
havior so that the most time efficient algorithm can be chosen for each class.
10.1 Introduction
To predict the stability boundaries of cutting processes the interactions within the
closed-loop of the coupled sub-systems machine and process is examined,
Fig. 10.1. Regenerative chatter is caused by the repetitive engagement of a tooth
into the surface cut by the preceding tooth. In Fig. 10.1, this is represented by the
time delay T.
Machine
x 
y 
Fx,y_stat Gxx ( jω) Gxy (j ω)  
G ( jω) G (j ω)
 yx yy 
Fx 
F 
 y aFx  ΔFP + -
a 
 Fy 
ap kt, kr T [a ay ]
Δh(t,t-T)
x
Δh(t)
Process
Fig. 10.1 Closed loop of the process-machine interactions [1]
226 K. Großmann and M. Löser
The figure shows an example where the process machine interactions in x-y-
plane are investigated and the z direction is neglected. The dynamic behavior of
the machine is represented by the transfer function matrix G(jω), which contains
the direct transfer functions and the cross transfer functions in x and y direction
respectively. However, the process force is a function of chip thickness h. Thus,
the deflections in x-y-plane have to be transformed into the direction of chip
thickness h. This is conducted by means of so-called directional coefficients (ax,
ay). With the dynamic chip thickness Δh the dynamic process forces ΔFp can be
computed by applying an empirical cutting force model. Here, this is represented
by the tangential coefficient kt and the radial coefficient kr. The process forces
now have to be transformed into x-y-plane again to close the loop of the process
machine interaction.
One method to compute stability boundaries of this closed loop is to model and
simulate the interactions in time domain. Modeling non-linearities in the time do-
main is much easier than in the frequency domain; at the same time, however, the
computation time to calculate SLDs increases significantly. One reason for this is
that a certain amount of “real time” has to be simulated to ensure the detection of
unstable cutting using time signals.
It is more time efficient to analyse the process machine interactions in fre-
quency domain. Analytical methods in frequency domain are the oldest algorithms
to predict the stability boundaries and are based on the research works by Tlusty
and Polacek as well as by Tobias and Fishwick [2, 3]. Algorithms in frequency
domain apply stability criteria like the Nyquist criterion on the open loop transfer
function of the process-machine interactions, [1, 4]. In milling operations, the di-
rectional coefficients shown in Fig. 10.1 are time-variant and vary periodically
with the tooth-passing frequency. To apply the methods in frequency domain av-
erage directional coefficients are used. Since the average directional coefficients
equal zeroth order Fourier coefficients this method is called zeroth order approxi-
mation (ZOA-method).
Operations with highly intermittent cutting conditions, such as low immersion
milling, show very strong time variance, which is assumed to affect the process
stability. In these cases, the assumption of constant average directional coeffi-
cients may lead to incorrect predictions of stability boundaries. Various authors
have presented computational time-efficient methods in discrete time domain to
determine stability lobe diagrams for highly intermittent cutting operations. Bayly
et al. have presented a time-finite element analysis [5], Insperger and Stepan have
developed the semi-discretization method [6].
To take the time variant behavior of directional coefficients in frequency do-
main into account Budak and Altintas expanded their single frequency solution to
the multi-frequency solution [7]. This was later on applied to low immersion mil-
ling by Merdol and Altintas [8].
Most of the studies that apply these advanced time efficient methods focus on
the investigation of single influences (for example: single degree of freedom be-
havior, helix angle, low immersion milling). In this chapter, the complex influence
of different effects is discussed for methods working in frequency domain. ZOA
method and two slightly different methods based on the multi-frequency solution
10 Synthesis of Stability Lobe Diagrams 227
are investigated. The applicability of these methods is scrutinized for a pattern of

different cutting conditions and different characteristics of the dynamic machine
behavior. The presented works were conducted within a project of the priority
program SPP 1180. One goal of the project is to define classification numbers,
which allow the selection of an appropriate algorithm before the computation of a
stability lobe diagram.
10.2 Computation of SLDs in Frequency Domain
10.2.1 Directed Frequency Response Functions

The open loop transfer function of the process machine interactions can be written
as:
G 0 ( jω ) = a p ⋅ (1 − e − jωT ) ⋅ G g (iω ) (10.1)
1
with T delay T =
n⋅Z
Z number of teeth
n spindle speed
ωT tooth passing frequency ω T = 2π ⋅ n ⋅ Z
Here, Gg(iw) is the directed frequency response function. The Nyquist criterion
can be applied to this transfer function:
 < 1 stable

Re(G 0 ( jω )) = = 1 stability boundary
 (10.2)
 > 1 unstable
Im(G 0 ( jω )) = 0
Simply spoken, the Nyquist criterion checks if an input signal leaves the open loop
without phase shift (imaginary part equals zero) and amplified (real part > 1).
Since the output signal is equivalent to the input signal of the subsequent pass
through of the loop the signal will be more and more amplified and the closed
loop becomes unstable.
In peripheral milling this criterion cannot be applied under any condition since
the coordinates in the x-y-plane are coupled, Fig. 10.2. The simplest approach for
the solution of this problem is by handling the interaction not in the machine coor-
dinates x and y but by investigating the transfer function at the direction of the
chip thickness h. Therefore, the deflections are transformed into the direction of
chip thickness and the chip thickness-dependent process forces are transformed
back into the direction of the machine coordinates. The transformation is carried
out by so-called directional coefficients [1].
Regarding regenerative chatter the static process forces and therefore the static
chip thickness can be neglected, [4]. The static chip thickness is the chip thickness
that would occur without relative displacements between workpiece and tool. The
fz
transformation
Fr
tool
ϕz Fx Δh= sinϕ·Δx + cosϕ·Δy
Ft
Ft= Kt·ap·Δh
Δx Fr= Kr·ap·Δh
Fr Δy
Δh Fx= -cosϕ·Ft - sinϕ·Fr
hstat Fy= sinϕ·Ft - cosϕ·Fr
Fy
Ft workpiece
ϕ
ϕ ϕe
Fig. 10.2 Cutting conditions in peripheral milling
dynamic chip thickness Δh is the change of h caused by the present deflection and
the deflection at the time of the preceding tooth engagement.
Δx    x (t )  x ( t − T) 
Δh = [sin ϕ cos ϕ ] ⋅   = [sin ϕ cos ϕ ] ⋅   −  
  y(t )  y(t − T ) 
(10.3)
 Δy 
Taking into account that the deflections are the reaction of forces acting on the
flexible machine structure this can be written in frequency domain.
G G xy  Fx 
Δh = (1 − e − jωT ) ⋅ [sin ϕ cos ϕ ] ⋅  xx ⋅ 
G yx G yy   Fy 
(10.4)
[ G
Δh = (1 − e − jωT ) ⋅ a x a y ⋅  xx ] G xy  Fx 
⋅ 
 yx G yy   Fy 
G
The dynamic process forces are a function of dynamic chip thickness Δh:
Fx  − k ⋅ cos ϕ − k r ⋅ sin ϕ  a 
F  = a p ⋅ t  ⋅ Δh = a p ⋅  Fx  ⋅ Δh (10.5)
 y  k t ⋅ sin ϕ − k r ⋅ cos ϕ  a Fy 
Equalizing the vector of process forces in (10.4) with (10.5) and taking the time-
delayed re-engagement of the subsequent tooth into account leads to the transfer
function of the open loop where Gg denotes the directed frequency response
function:
G 0 = a p ⋅ (1 − e − jωτ ) ⋅ a x [ ] G
a y ⋅  xx
G xy  a Fx 
⋅ 
G yx G yy  a Fy 
(
G 0 = a p ⋅ D ⋅ a x a Fx G xx )
+ a y a Fx G yx + a x a Fy G xy + a y a Fy G yy
(10.6)
G 0 = a p ⋅ D ⋅ (a xx G xx + a xy G yx + a yx G xy + a yy G yy ) = a p ⋅ D ⋅ G g
The directional coefficients depend on the immersion angle ϕ and are therefore
time-dependent. So, to apply (10.6) in frequency domain average directional
coefficients have to be used. Applying the Nyquist criterion and rearranging (10.6)
gives the frequency-dependent critical depth of cut apcrit, [1]
−1
a pcrit (ω ) = (10.7)
2 ⋅ Re(G g ( jω ))
Since only positive cutting depths are relevant for cutting operations only frequen-
cies with negative real parts of the directed frequency are considered. Further-
more, the relation between chatter frequency and spindle speed can be derived
from the Nyquist criterion:
60 ⋅ f
n=
 1
z m − arctan
{
Re G g ( jω ) 

} (10.8)

 π Im G g ({jω ) 
 }
With (10.7) and (10.8) the relation between cutting depth and spindle speed and,
consequently, the stability lobe diagram (SLD) can be computed. For this purpose,
the dynamic behaviour of the machine has to be assumed as constant within the
observed range of spindle speed.
(10.7) shows that only frequencies are relevant, whose real parts of the directed
frequency response function are negative. For a multi-degree of freedom system
(MDoF system) it is possible to divide the directed transfer function Gg into sec-
tions, which can be assigned to a specified mode respectively. For each of these
sections a mode-dependent stability boundary can be computed. The total stability
lobe diagram can now be determined by choosing the minimum of the mode-
dependent boundaries at every spindle speed. However, it has to mentioned that
this is only valid, if the assumption of averaged directional coefficients can be
made and if the directional coefficients are used to compute the directed transfer
function Gg as shown in (10.6). The directed transfer function Gg takes the cou-
pling of x and y direction into account. Insperger and Stepan have shown that
computing stability lobe diagrams for x and y direction separately by neglecting
the geometrical coupling and superposing these SLDs will lead to an incorrect
prediction of stability boundaries, [9].
10.2.2 Time Variant Behavior

In milling operations, the directional coefficients are time-variant and vary period-
ically with the tooth passing frequency. Using average directional coefficients al-
lows a time-efficient computation of the process stability in the frequency domain,
as shown in the section before.
As mentioned in the introduction, several authors have introduced computa-
tional time-efficient algorithms in discrete time domain to take these effects into
account. However, since this chapter deals with the computation of stability in
frequency domain the following section focuses on an algorithm based on the mul-
ti-frequency solution presented by Altintas and Merdol [8].
Altintas and Merdol [8] utilize the fact that the directional coefficients are peri-
odic with tooth-passing frequency ωT. Because of this they can be expanded into a
Fourier series.
∞ T

1
a xx ( t ) =
r = −∞
a xx ,r ⋅ e jrωT t , a xx ,r =
T 
a xx ( t ) ⋅ e − jrωT t dt
0
(10.9)
Taking into account that the relation between time and immersion angle ϕ is given
by the tooth-passing frequency and that the directional coefficients are zero, if the
teeth are not engaged, the Fourier coefficients can be written as [8]:
ϕ ex
Z
a
− jrZϕ
a xx ,r = xx (ϕ ) ⋅ e dϕ (10.10)
2π
ϕ st
Where ϕst denotes the angle where the teeth engage the workpiece and ϕex denotes
the angle where the teeth exit the workpiece. Since the Fourier coefficients of ze-
roth order are equivalent to the average directional coefficients the methods using
average directional coefficients are called zeroth order approximation (ZOA).
Fig. 10.3 depicts an example of absolute values of Fourier coefficients.
0.5 Fourier coefficients of

axx
0.4 directional coefficients
ayy
a [108·N/m²]
0.3
up milling
0.2 dtool= 12 mm
0.1 λ= 0°
ae= 2 mm
0
number of teeth Z= 4
r= 0 r= 1 r= 2 r= 3 r= 4
Fig. 10.3 Example for Absolute Values of Fourier Coefficients
By applying the Fourier series the process forces as a function of the change of
the relative deflection between tool and work piece can be written as follows:
 hr 
F( t ) = a p ⋅ 
  A r e jrωT t  ⋅ Δx( t )
 (10.11)
 r =− h r 
This relation in time domain can be transformed into frequency domain by using
the Laplace transformation. The Laplace transforms are given by:
(
L{Δx (t )} = L{x(t ) − x(t − T )} = 1 − e jωT ⋅ x( jω ) = D ⋅ x( jω ) )
Le{ jrωT t
}
⋅ x(t ) = x( j(ω − rω T ))
(10.12)
Applying these Laplace transformed is slightly different to the approach intro-

duced in [8] but leads to the same results.
 hr 
F( jω) = a p ⋅ D ⋅ 
  A r ⋅ x( j(ω − rωT ))

 r =−h r 
 r h 
F( jω) = a p ⋅ D ⋅ 
  A r ⋅ G( j(ω − rωT )) ⋅ F( j(ω − rωT ))

 r =−h r 
F( j(ω − pωT )) = a p ⋅ D ⋅ (10.13)
 hr 
⋅
  A r ⋅ G( j(ω − (p + r )ωT )) ⋅ F( j(ω − (p + r )ωT ))

 r =−h r 
 hr 
Fp = a p ⋅ D ⋅ 
  A r ⋅ G p+r ⋅ Fp+r 

 r =− h r 
For a Fourier order hr=1 follows:
F−1 
 
 F0  = a p ⋅ D ⋅
 F1 
 
(10.14)
 A −1 ⋅ G −2 ⋅ F− 2  A 0 ⋅ G −1 ⋅ F−1   A1 ⋅ G 0 ⋅ F0  
     
⋅   A −1 ⋅ G −1 ⋅ F−1  +  A 0 ⋅ G 0 ⋅ F0  +  A1 ⋅ G 1 ⋅ F1  
  A ⋅ G ⋅ F   A ⋅ G ⋅ F  A ⋅ G ⋅ F  
  −1 0 0   0 1 1   1 2 2 
The right side of (10.14) shows forces F with indices of r = ±(h r + 1) . Since the
absolute values of Ar converge to zero for increasing order r the terms with indices
of order greater than hr are set to be zero. It follows:
F−1   A 0 A1 0  G −1 0 0  F−1 
  A    
 0
F = a ⋅ D ⋅ A A 1⋅ 0 G0 0  ⋅  F0 
p  −1 0 
 F1  G 1   F1 
   0 A −1 A 0   0 0 (10.15)
F out = a p ⋅ D ⋅ A ⋅ G ⋅ F in
h h h h
G0 = ap ⋅ D ⋅ Ah ⋅Gh
This leads to the transfer function of the open loop. Therein, the coordinates are
coupled. To apply the Nyquist criterion the investigated system needs only one
coordinate or – in the case of a multi variable system – the coordinates have to be
decoupled. The coordinates can be decoupled by a modal transformation. The
modal coordinates correspond to the eigenvalues of G0. The Nyquist criterion can
now be applied to every one of the ζ·(2·r+1) eigenvalues. ζ is the number of the
structural coordinates (ζ= 2 for a stability analysis in the x-y-plane), [8].
Since the depth of cut is a scalar value the transfer function can be scaled
with ap.
G0 =
1
ap
(
⋅ G 0 = eig D ⋅ A h ⋅ G h ) (10.16)
Fig. 10.4 shows an example of the real and imaginary parts of decoupled transfer
functions. In this example, the dynamic behaviour has relevant eigenfrequencies of
up to 5 kHz. Since the matrix of transfer functions contains the transfer functions at
frequencies shifted by multiples of the tooth passing frequency at a specified fre-
quency, the superposition of transfer functions is periodic with tooth passing fre-
quency. So the stability analysis need not be carried out for the whole bandwidth of
5 kHz but only for a frequency band of the tooth-passing frequency ft.
ft
100 100
Re(G0) [m-1]
50 0
Im(G0) [m-1]
0 -100
Im(G0) [m-1]
100
-50 0
-100 -100
-100 0 100 0 1000 2000 3000 4000 5000
Re(G0) [m-1] f [Hz]
Fig. 10.4 Decoupled and Scaled Transfer Functions of the Open Loop
These sets of transfer functions have to be computed for every discrete spindle
the stability boundary has to be determined at. The stability boundaries are given
by the reciprocals of the real parts of the intersections between transfer function
and real axis. In [10] Altintas et al. assert that “the most conservative and positive
depth of cut must be considered as a final solution.” So, the critical depth of cut is
given by:
1
a p _ crit =
(
Re G 0 (Im = 0) max ) (10.17)
Fig. 10.5 demonstrates the computation of the stability boundary for the example
mentioned before. The relevant parts of the transfer functions are drawn as bold
lines and the critical depth of cut is ap_crit= 11.2 mm.
However, the assumption that the critical depth of cut is given by the most con-
servative real part, is not correct in every case. However, if all intersections with
the real axis are analyzed, the algorithm can also be used, if multiple stability
boundaries occur at a specified spindle speed. All algorithms in frequency domain
100 10
50 5
Re(G0) [m-1]
Re(G0) [m-1]
0 0 1/apcrit= 89.1 m-1
apcrit= 11.2 mm
-50 -5
-100 -10
-100 0 100 80 90 100
Im(G0) [m-1] Im(G0) [m-1]
Fig. 10.5 Determination of Critical Depth of Cut
apply the simplified Nyquist criterion shown in (10.2). The complete criterion
analyzes not only the intersection of the transfer locus with the real axis. It says
the process is stable, if the Nyquist point is encircled counter-clockwise at least
once. An intersection with the real axis where the imaginary part changes from
positive to negative values will not lead to an unstable process.
Fig. 10.6 shows a simple example to demonstrate this. It depicts a section of a
SLD for a one-dimensional process (only x-direction has been taken into account).
The dynamic system has an SDoF behavior (m= 0.572 kg; d= 140 Ns/m;
c= 2.2·107 N/m). The cutting force coefficients are kt= 3.6·109 N/m² and
kr= 2.25·109 N/m². The figure compares the stability boundaries determined by
time domain simulations with the boundaries determined by the multi-frequency
solution. The figure also shows the transfer locus for a spindle speed of nspin-
dle= 10,600 rpm. Point I marks a counter-clockwise intersection and the process
gets unstable at a depth of cut of apcrit= 7.3 mm. The clockwise intersection at
point II compensates the instability and the process becomes stable again up to the
depth of cut marked by point III.
10.2.3 Cutting-Depth-Dependent Behavior

Not only the radial immersion has an influence on the “smoothness” of the cutting
forces. An increasing helix angle will cause forces that change less rapidly over
time. Assuming the parameters of the process force model are unaffected by the
helix angle, the average forces – i. e. the zeroth order Fourier coefficients - are al-
so independent from the helix angle. However, the higher order Fourier coeffi-
cients are influenced by the helix angle. This can be illustrated by the process
forces.
Re(G0) [m-1] Re(G0) [m-1]

-400 -200 0 200 0 50 100 150
500 increasing depth of cut
20
10
Im(G0) [m-1]
Im(G0) [m-1]
III II I
0 0
-10
-20
-500
nspindle = 10600 rpm
20
III apcrit= 16.4 mm

15 process gets unstable
II apcrit= 12.4 mm
apcrit [mm]
process gets stable

10
I apcrit= 7.3 mm
process gets unstable
5
frequency domain
time domain
0
1 1.05 1.1
nspindle [rpm] 4
x 10
Fig. 10.6 Multiple Stability Boundaries at one Spindle Speed
Fig. 10.7 shows simulated process forces Fy for different helix angles at axial
cutting depths ap= 10 mm and ap= 20 mm but in both cases for the same radial
depth of cut ae= 3mm. The figure also depicts the spectra of the process forces.
For the straight fluted mill there is a sharp change of the forces when the cutting
edge leaves the workpiece. The force at zero frequency (which equals the average
force) is independent of the helix angle. The forces at the tooth-passing frequency
and their harmonics decrease with increasing helix angle. The decreasing effect is
more significant for the axial depth of cut of ap= 20 mm.
The magnitudes of the process forces at tooth-passing frequency and their har-
monics correspond to the Fourier coefficients. Thus, for helicoidal mills the values
of the Fourier coefficients depend on the axial depth of cut ap. In this case, the
value of the Fourier coefficients is given by [11]:
− j⋅r⋅2⋅π ⋅a / p
1− e p
A *r = A r ⋅ (10.18)
j⋅ r ⋅ 2 ⋅π ⋅ a p / p
100
λ= 0
250 Fy(f) [N] ap= 10 mm
Fy(t) [N] 80 λ= 10°
200
60 λ= 20°
150
average force
λ= 30°
100 40
50 20
0 0
0 5 10 0 500 1000 1500 2000
t [ms] f [Hz]
200
500
Fy(t) [N]
Fy(f) [N] ap= 20 mm
400 150
300
average force
100
200
50
100
0
0
0 5 10 0 500 1000 1500 2000
t [ms] f [Hz]
Fig. 10.7 Process Forces and Spectra of Process Forces
Ar represents the Fourier coefficients for the straight-fluted mill and p is the
pitch of a tool with Z teeth.
d ⋅π
p = Tool (10.19)
Z ⋅ tan(λ )
With increasing axial depth of cut the values of the Fourier coefficients decrease.
The Fourier coefficients differ for the different helix angles. The zeroth order
coefficients are independent of the helix angle but the values of the higher order
coefficient decrease with an increasing helix angle λ.
Similar to (10.15), the open loop transfer function can be written as follows:
G 0 = a p ⋅ D ⋅ A *h ⋅ G h (10.20)
Decoupling the coordinates in (10.20) and applying the Nyquist criterion could
still be carried out by computing the eigenvalues, as described in the chapter be-
fore. This can be achieved by computing the eigenvalues for the characteristic eq-
uation
[
det I + ΛA h G h = 0 ] (10.21)
(
with Λ = a p 1 − e jωT , [8]. )
However, if the decoupled system has to be computed for every ap, it is less
time consuming to apply a different approach. For a multi-variable system the sta-
bility analysis can be carried out for the determinant of the matrix of open loop
transfer functions minus identity matrix:
G 0 = det G 0 − I (10.22)
The behavior of the whole system is therefore summarized in a transfer function

for a single coordinate. The open loop is stable, if the transfer function does not
encircle the point {0, 0·i}. This way, the determination of stability boundaries is
numerically easier, especially for a Fourier series expansion of higher order.
Since the elements of matrix of Fourier coefficients A *h depend on the axial
depth of cut the transfer function cannot be scaled with ap as described by (10.16).
Thus, applying the Nyquist criterion only provides information, if a specified
combination of spindle speed and axial depth of cut results in a stable process or
not. A change of axial depth of cut ap leads to different transfer locuses and the
stability boundary at a given spindle speed has to be determined iteratively. In
the following sections, the above-described algorithm is called Multifreq_det
algorithm.
10.2.4 Summary
In the previous three sections, different algorithms for the computation of stability
lobe diagrams have been shown. The first and simplest one is the well-known ze-
roths order approximation (ZOA). The last two approaches are expanded versions
of the multi-frequency solution presented by Merdol and Altintas. One of them
uses the computation of eigenvalues to decouple the system of multiple “frequen-
cy coordinates” (Multifreq_eig) and the other one computes the determinants of
the transfer matrices (Multifreq_det). Which algorithm is best suitable for the de-
termination of a SLD depends on the effects that have to be taken into account.
However, the more complex the algorithm gets, the more computational time is
needed. But even the algorithm for a depth-of-cut-dependent behavior is less time-
consuming than time domain simulations. In the following section, these algo-
rithms are applied to cutting operations with different process and machine
behavior.
10.3 Application
10.3.1 Models of Process and Machine

The following investigations use the example of an up-milling operation with cy-
lindrical end mills. Table 10.1 shows the parameters of this reference process as
well as the coefficients of the process force model. To describe the process forces
a linear force model was used [12]. The parameters were conducted by means of
cutting tests for aluminium AA7075 workpieces.
Table 10.1 Parameters of the Reference Process
Process Up-milling
Number of teeth Z= 4
Tool
Tool diameter dTool= 12 mm
Material AlZn5,5MgCu (AA7075)
Workpiece Tangential force coefficient kt= 830 N/mm²
Radial force coefficient kr= 225 N/mm²
The dynamic behavior is represented by modal models. The parameters of these

models have been identified from measured frequency response functions of dif-
ferent spindle tool systems. In all cases, cross compliances have been neglected.
The frequency response functions are therefore computed as follows:
nx
−m
1
G xx =
x ,m ω + jωd x ,m + c x ,m
2
m =1
ny
(10.23)
−m
1
G yy =
y, m ω + jωd y ,m + c y,m
2
m =1
G xy = G yx = 0
with number of eigenmodes n, the modal parameters mass m, damping d and

stiffness c.
10.3.2 Rotational Symmetric SDoF System

The single degree of freedom system is meant here as the single degree of freedom
behavior in the x and y-directions respectively. Such a behavior with just one do-
minant eigenfrequency is typical for tools with a large length-to-diameter ratio.
Fig. 10.8 shows the measured frequency response function of a carbide dummy
tool with a ratio of l/d= 8. It shows a dominant eigenfrequency at 728 Hz. The re-
sponse functions are nearly the same for x and y- direction. The modal parameters
identified from the measured data and used for the computation of stability lobe
diagrams are shown in Table 10.2.
3 Gxx
compliance [µm/N]
Gyy
0
400 600 800 1000 1200
f [Hz]
Fig. 10.8 Reference for a SDoF Behavior
Table 10.2 Modal Parameters of the SDoF System
Gxx Gyy
Mass m [kg] 0.922 0.925
Damping d [kg/s] 69 69
7
Stiffness c [N/m] 1.93·10 1.95·107
For the example of SDoF, several authors showed systems that may occur at
small radial immersion additional stability lobes. These additional stability lobes
cannot be predicted by the ZOA-method. Zatarain et al. [11] investigated the in-
fluence of the helix angle on chatter stability by using the multi-frequency solution
by Merdol and Altintas as well as the semi-discretization method by Insperger and
Stepan. This work showed that for increasing helix angles the additional lobes
transform into closed instability islands. The same behavior can be shown for this
example here.
30
Multifreq_eig
25 time domain
20
apcrit [mm]
15
10
5
0.5 1 1.5 2 2.5

nspindle [rpm] 4
x 10
Fig 10.9 SLDs of the SDoF System, λ= 0°, ae= 0.5 mm

Fig. 10.9 shows a comparison of SLDs computed by time domain simulation

and by the Multifreq_eig algorithm. The time domain and the frequency domain
solution are in good agreement and show the additional stability lobes.
30
Multifreq_eig
2.5
25 λ= 0°
20 Multifreq_det
apcrit [mm]
λ= 30°
15 ZOA
10
5
0.5 1 1.5 2 2.5

nspindle [rpm] 4
x 10
Fig. 10.10 SLDs of the SDoF System, ae= 0.5 mm
With increasing helix angle, the additional stability lobes transform into unsta-
ble islands within the stable area. This can be shown with the multi-frequency so-
lution as well as by time domain simulations. Since the directional coefficients
depend on depth of cut the Multifreq_det algorithm has to be used in the case of a
helix angle λ= 30°. For a helix angle of λ= 0 the Multifreq_eig algorithm is used.
For a helix angle of λ= 30° the ZOA algorithm shows nearly the same results as
the multi-frequency solution, except the unstable islands, Fig. 10.10.
30
Multifreq_eig
25 λ= 0°
20 Multifreq_det
apcrit [mm]
λ= 30°
15 ZOA
10
5
0.5 1 1.5 2 2.5

nspindle [rpm] 4
x 10
Fig. 10.11 SLDs of the SDoF System ae= 3 mm
For larger radial immersions the time variant behavior of the directional coeffi-
cients gains less impact on the stability boundaries. At a radial depth of cut of
ae= 3 mm, no impact of the helix angle can be shown and the ZOA algorithm
results in practically identical stability boundaries to the multi-frequency solution,
Fig. 10.11. In this case, the fast ZOA method can be applied without loosing accu-
racy of the predicted stability boundaries.
10.3.3 Rotational Symmetric MDoF System

As shown in the example, in the previous chapter the dynamic behavior of spindle
tool systems with long slender tools is dominated by the eigenmodes of the free pro-
jecting part of the tool. Tools with a lesser length-to-diameter ratio eigenmodes of
the spindle and the spindle stock gain more impact on the dynamic behavior at the
tool centre point (TCP). This results in the behavior of a multi-degree of freedom
system (MDoF System) with several modes that may cause chatter instability.
0.6
compliance [µm/N]
0.4
0.2
0
0 1000 2000 3000 4000 5000
f [Hz]
Fig. 10.12 Dynamic Behavior of the Rotational Symmetric MDoF System
Fig. 10.12 depicts a frequency response function in y-direction measured at a

tool with a length-to-diameter ratio of l/d=2.5. Assuming a rotational symmetric
system, the frequency response function in x-direction is set to be equal to the re-
sponse function in y-direction.
30
Multifreq_eig
25 λ= 0°
Multifreq_det
apcrit [mm]
20 λ= 30°
ZOA
15
10
5
0.5 1 1.5 2 2.5
nspindle [rpm] 4
x 10
Fig. 10.13 SLDs of the Symmetric MDoF System, ae= 3 mm

The investigated SDoF system has no impact on the stability boundaries at a

radial depth of cut of ae= 3 mm. This is different for the symmetric MDoF system.
For the straight-fluted mill an impact on the stability can be seen around the spin-
dle speed of about 20,000 rpm. However, for a helix angle of λ= 30° the multi-
frequency solution is in good agreement with the stability boundaries obtained by
the ZOA method, Fig 10.13. An increase of the radial depth of cut will minimize
the impact of the time-variant directional coefficients even for the straight-fluted
mill, Fig. 10.14.
20
Multifreq_eig
λ= 0°
15
Multifreq_det
apcrit [mm]
λ= 30°
10 ZOA
0
0.5 1 1.5 2 2.5
nspindle [rpm] 4
x 10
Fig. 10.14 SLDs of the Symmetric MDoF System, ae= 5 mm
10.3.4 Non-rotational Symmetric MDoF System

Usually spindle and milling tool are rotational symmetric systems. In most cases,
however, the spindle stock is non-rotationally symmetric. Especially for tools with
a small length-to-diameter ratio this may have an impact on the dynamic behavior
at the tool centre point. In addition to Fig. 10.12, Fig. 10.15 shows the measured
frequency response functions in x and y-direction. The response functions differ
for the different directions.
0.6 Gxx
compliance [µm/N]
Gyy
0.4
0.2
0
0 1000 2000 3000 4000 5000
f [Hz]
Fig. 10.15 Dynamic Behavior of the Non-Symmetric MDoF System
Fig. 10.16 and Fig. 10.17 show a comparison of the SLDs computed by multi-
frequency solution and ZOA method for radial immersions ae= 3 mm and
ae= 5 mm respectively. In both cases, the ZOA method provides acceptable results
for a helix angle of λ= 30°. However, compared with the results for the symmetric
system the zero helix angle also has an impact at higher radial immersions.
30
Multifreq_eig
25 λ= 0°
Multifreq_det
apcrit [mm]
20 λ= 30°
ZOA
15
10
5
0.5 1 1.5 2 2.5
nspindle [rpm] 4
x 10
Fig. 10.16 SLDs of the Non-Symmetric MDoF System, ae= 3 mm
20
Multifreq_eig
λ= 0°
15
Multifreq_det
apcrit [mm]
λ= 30°
10 ZOA
0
0.5 1 1.5 2 2.5
nspindle [rpm] 4
x 10
Fig. 10.17 SLDs for the Non-Symmetric MDoF System, ae= 5 mm
10.4 Conclusions
The behavior of the mechanical system in combination with the process behavior
has an impact on the stability. This means that the process and machine behavior
determines which effects have to be taken into account for an accurate prediction
of stability boundaries.
In a case study - presented in this chapter - algorithms for the prediction of sta-
bility boundaries in frequency domain have been applied to different combinations
of process and machine characteristics. A comparison with results in time domain
has demonstrated that frequency domain algorithms are applicable, even if the sys-
tem shows a time-variant behavior. It has been demonstrated that in some cases a
simplification can be made so that the time efficient zeroth order approximation is
applicable. Table 10.3 shows a pattern of the computational methods for the inves-
tigated combinations of process and machine behavior.
Table 10.3 Pattern of Applicable Frequency Domain Algorithms for the Investigated Ref-
erence Process
Symmetric Symmetric Non-Symmetric

SDoF MDoF MDoF
Straight-fluted
a e / d ≈ 0.04 Multifreq eig Multifreq eig Multifreq eig
Helicoidal
Multifreq det Multifreq det Multifreq det
a e / d ≈ 0.04
Straight-fluted
a e / d ≈ 0.25 ZOA Multifreq eig Multifreq eig
Helicoidal
a e / d ≈ 0.25 ZOA ZOA ZOA
Straight-fluted
a e / d ≈ 0.4 ZOA ZOA Multifreq eig
Helicoidal
a e / d ≈ 0.4 ZOA ZOA ZOA
This pattern is valid for the investigated reference process. Ongoing works deal
with the definition of classification numbers to expand this pattern for a general
process and machine behavior. Some investigated issues that have to be quantified
are, for example, the influence of tool diameter and cutting force coefficient as
well as the quantification of the influence of the dynamic behavior, i. e. number of
modes and dynamic stiffness.
References
[1] Weck, M., Brecher, C.: Werkzeugmaschinen Fertigungssysteme Bd.5 Messtechnische
Untersuchung und Beurteilung, dynamische Stabilität. Springer, Heidelberg
(2006)
[2] Tlusty, J., Polacek, A.: Beispiele der Behandlung der selbsterregten Schwingungen
der Werkzeugmaschinen, 3. Forschungs- und Konstruktionskolloquium Werkzeug-
maschinen und Betriebswissenschaft (3. FoKoMa). Vogel-Verlag, Coburg (1957)
[3] Tobias, S.A., Fishwick, W.: A Theory of Regenerative Chatter. The Engineer, Lon-
don (1958)
[4] Altintas, Y., Budak, E.: Analytical Prediction of Stability Lobes in Milling. Annals of
the CIRP 44(1), 357–362 (1995)
[5] Bayly, P.V., Halley, J.E., Mann, B.P., Davies, M.A.: Stability of Interrupted Cutting
by Temporal Finite Element Analysis. Journal of Manufacturing Science and Engi-
neering 125, 220–225 (2003)
[6] Insperger, T., Stepan, G.: Semi-discretization method for delayed systems. Interna-
tional Journal for Numerical Methods in Engineering 55, 503–518 (2002)
[7] Budak, E., Altintas, Y.: Analytical Prediction of Chatter Stability in Milling. Part I:
Modelling, Part II: Applications, Transactions of ASME, Journal of Dynamic Sys-
tems, Measurement and Control 120, 22–36 (1998)
[8] Merdol, S.D., Altintas, Y.: Multi Frequency Solution of Chatter Stability for Low
Immersion Milling. ASME Journal of Manufacturing Science and Engineering 126,
459–466 (2004)
[9] Insperger, T., Stepan, G.: Stability Transition between 1 and 2 Degree-of-Freedom
Models of Milling. Periodica Polytechnica Mechanical Engineering 48(1), 27–39
(2004)
[10] Altintas, Y., Stepan, G., Merdol, D., Dombovari, Z.: Chatter stability of milling in
frequency and discrete time domain. CIRP Journal of Manufacturing Science and
Technology 1, 35–44 (2008)
[11] Zatarain, M., Munoa, J., Peigne, G., Insperger, T.: Analysis of the Influence of Mill
Helix Angle on Chatter Stability. Annals of the CIRP 55(1), 365–368 (2006)
[12] Altintas, Y.: Modeling Approaches and Software for Predicting the Performance of
Milling Operations at MAL-UBC. Machining Science and Technology 4(4), 445–478
Chapter 11
Analysis of Industrial Robot Structure
and Milling Process Interaction for Path
Manipulation
J. Bauer, M. Friedmann, T. Hemker, M. Pischan, C. Reinl,

E. Abele, and O. von Stryk
Abstract. Industrial robots are used in a great variety of applications for hand-
ling, welding, assembling and milling operations. Especially for machining opera-
tions, industrial robots represent a cost-saving and flexible alternative compared to
standard machine tools. Reduced pose and path accuracy, especially under process
force load due to the high mechanical compliance, restrict the use of industrial ro-
bots for machining applications with high accuracy requirements. In this chapter, a
method is presented to predict and compensate path deviation of robots resulting
from process forces. A process force simulation based on a material removal cal-
culation is presented. Furthermore, a rigid multi-body dynamic system’s model of
the robot is extended by joint elasticities and tilting effects, which are modeled by
spring-damper-models at actuated and additional virtual axes. By coupling the re-
moval simulation with the robot model the interaction of the milling process with
the robot structure can be analyzed by evaluating the path deviation and surface
structure. With the knowledge of interaction along the milling path a general
model-based path correction strategy is introduced to significantly improve accu-
racy in milling operations.
11.1 Introduction
Major fields of machining applications for industrial robots are automated pre-
machining, deburring and fettling of cast parts or trimming of carbon-fiber-
reinforced laminate. Due to its kinematic structure with 6 axes the robot can cover
a large working space and is able to reach difficult workpiece positions so that it
can be applied to perform complex machining operations. Therefore, compared to
standard machine tools, industrial robots offer an economic and flexible machin-
ing alternative. However, industrial robots do not provide a high absolute and re-
peat accuracy. Current industrial robot systems reach a repeat accuracy of 0.06
mm. Under process load, e. g. in milling operation, an additional deflection of the
tool center point (TCP) occurs. Measured deflections of 0.25 mm under process
loads of 100 N in earlier tests [1] have confirmed the expected compliance. There-
fore, when using an industrial robot for milling applications, inaccuracies of the
246 J. Bauer et al.
serial robot kinematic, the low structural stiffness and the effective process forces
lead to path deviations. The results are unwanted trajectory deviations, which lead
to errors in dimension and a reduced surface quality of the workpiece. These devi-
ations mainly consist of a static offset overlaid with a low frequency oscillation of
the tool [2].
In order to increase the milling accuracy with present process forces of the mil-
ling operation a model-based path manipulation module for robots is developed.
This module consists of a robot and a milling force model to predict path errors in
advance of a real milling operation. Additionally, the simulation module allows
the investigation and analysis of the interaction of milling force and motion of the
robot. Finally, the proposed strategy for modeling and simulation serves as the ba-
sis for an efficient offline correction of path deviations.
Overall Objective
The machining operation induces process forces, which act on the machine struc-
ture. The high compliance of the structure causes a deviation of the programmed
path at the TCP. This path displacement causes a new cutting condition of the cut-
ter including a variation of the process force (cf. Fig. 11.1). Thus, milling robots
are characterized by a close interaction of the cutting process with the mechanical
robot structure. An early consideration of the interactions of the milling robot in
the production process can reduce manufacturing failures as well as efforts to ma-
nually re-teach the robot’s path.
Fig. 11.1. Model based tool path compensation
The general aim is the development of a strategy to improve the accuracy in ro-
bot based milling operations by prediction and offline compensation of static and
low frequency path deviations. Furthermore, the optimization of cutting parame-
ters and a prediction of milling properties (accuracy in dimension, surface struc-
ture) is in the focus.
In order to simulate the milling force a standard cutting force model presented
in [3] is implemented and adapted for industrial robots. The simulation of the
11 Analysis of Industrial Robot Structure and Milling Process Interaction 247
robot’s motion dynamics is based on the Newton-Euler-formulation. The model is

based on a fine granular description of the kinematic structure and dynamical
properties. It allows the introduction of arbitrary rotational axes to model elastic
deformation. Additional properties are added to consider tilting of the axes and
backlash of the gears.
For both sub-models, different experimental investigations are conducted to de-
termine the model parameters describing the physical behavior of the robot and
the machining process. Both models are initially tested independently and then
coupled to simulate the machine process interaction. Two methods of model-based
offline path compensation - based on ideal milling forces and deviation mirroring -
are presented. Finally, robot milling experiments are carried out to proof the com-
pensation concepts.
11.2 Robot and Milling Force Modeling

In order to study the interaction of machine structure, i. e. the industrial robot and
the removal process, a fine granular robot model and a milling force model are
developed.
11.2.1 Extended Robot Kinematic and Dynamics Modeling

In this section, the modeling and simulation of the kinematic and dynamics of the
robot's motion are discussed. Special interest is taken in the modeling of elastici-
ties of the robot, which have a high impact on the robot's motion, if the robot op-
erates under present process forces. The presented approach is based on a modular
modeling methodology, which allows the integration of arbitrary axes of motion
without the need to re-implement the equations of the robot dynamics.
Modeling the Kinematical Structure
As elastic motions of the robot may not only appear around the axes of the robots
drives but also around additional axes, an extended kinematical model has been
derived. In this section, the kinematic model is discussed using homogeneous
transformations in a 4x4-matrix representation.
For better readability, the abbreviations
1 0 0 x 1 0 0 0
0 1 0 y  0 cos(α ) − sin(α ) 0
Trans ( x, y , z ) :=  , Rot ( x; α ) := 
1 z  0 
,
 0 0  0 sin(α ) cos (α )
0 0 0 1  0 0 0 1 
 cos(α ) 0 sin(α ) 0   cos(α ) − sin(α ) 0 0
 0 1 0 0  sin(α ) cos(α ) 0 0
Rot ( y; α ) :=   , Rot ( z; α ) := 
 − sin(α) 0 cos (α ) 0   0 0 1 0 
 0 0 0 1   0 0 0 1 
248 J. Bauer et al.
are used to represent translations and rotations. In standard Denavit-Hartenberg-

notation (e. g. [4]), a rigid link with a revolute joint is described by
link dh ,i = Rot ( z;θ i )·Trans (0, 0, di )·Trans (0, 0, ai )· Rot ( x;α i )
with θi being the current position of the joint and the constants di, ai and αi de-
scribing the relative position of the next joint. These joints, which are driven to
move the robot, are referred to as actuated joints. It should be noted that this con-
vention does not contain information on the precise placement of the joint. As
Trans (0, 0, pi )·Rot ( z;θi )·Trans (0, 0, d i − pi ) = Rot ( z; qi )·Trans (0, 0, d i )
is valid, joints can be anywhere on the z-axis of the previous link's coordinate
frame. To allow for a precise placement of the joint the model is extended by the
joint displacement pi, leading to
linkwrep,i = Trans(0,0, pi )·Rot ( z;θi )·Trans(0,0, di − pi )·

·Trans(0,0, ai )·Rot ( x;αi ).
This model is extended by two variable rotations around axes, orthogonal to the
joint's axis, leading to the final extended kinematics model
linkext ,i = Trans(0, 0, pi )·Rot ( z;θi )·Rot ( x;θ x ,i )·Rot ( y;θ y ,i )·Trans(0, 0, di − pi )·

·Trans(0, 0, ai )·Rot ( x; α i )
with θx,i and θy,i being the rotation angles caused by elasticities around the addi-
tional axes. These additional revolute axes are referred to as virtual joints (cf. Fig.
11.2). It should be noted that these virtual joints are optional. One or both virtual
joints can be added to each link, if additional elasticities need to be modeled for
this link.
Additional Dynamics Properties
Simulating the motion dynamics of the robot requires additional parameters for
each link of the robot. Namely, these are the mass mi , the inertia tensor Ii and the
center of mass comi for each link. In the implementation used for this work, the
center of mass is described with respect to the coordinate frame defined by linkext,i.
The inertia tensor is resolved at the center of mass in a coordinate frame using the
same orientation as the frame defined by the link.
Fig. 11.2 Modeling of the robot with tilting at virtual axes
Modeling the Drives and Elasticities
Simulation of the robot's motion requires information

τi
of the torques acting in the joints (either real or vir-
tual) of the robot. As no information of the robot's
drivetrain nor the control strategy of the motors are
known, the assumption is made that each motor is at -s i
its desired position qi and moves with the desired rate si ( q i − θi )
q i . The motors are coupled to the joints by means of
a spring-damper-system with the spring stiffness Ki,
the damping Di and si the backlash of the gear lead-
ing to the equation
Fig. 11.3 Modeling of
 (( qi − si ) − θ i ) , if ( qi − θ i ) ≥ si
 backlash
τ i = Di ·(θi − qi ) + K i · (( qi + si ) − θ i ) , if ( qi − θ i ) ≤ − si
 0 , else
for the torques in the actuated joints (cf. Fig. 11.3). Note that this convention also
allows for elasticities and damping around the actuated axes, which may occur be-
cause of the gears used in the drivetrain. Torques acting on virtual joints are calcu-
lated in the same way with the exception that the desired position and rate always
are zero and that there is no backlash,
τ x| y ,i = K x| y ,i ·θ x| y ,i + D x| y ,i ·θx| y ,i .
250 J. Bauer et al.
Forward Dynamics Simulation
The dynamics of a robot's motion is described by
M ( θ )·
θ + C (θ , θ ) + G ( θ ) = τ + S ( Fxyz , tool , θ )
with θ = (θ1 , θ x ,1 , θ y ,1 ,..., θ n , θ x , n , θ yn )T being the positions of all joints (actuated or

virtual), τ = (τ 1 , τ x ,1 ,τ y ,1 ,..., τ n , τ x , n ,τ yn )T likewise being the torques in all joints
and S being the torques caused by milling projected into the respective joints.
Different ways exist to calculate the mass-matrix M, vector of coriolis forces C
and gravitational forces G in this model. The modular approach chosen describing
the robot's structure is well-suited for algorithms based on the Newton-Euler-
formulation of robot dynamics. Two widely used algorithms of this group are the
Composite-Rigid-Body-Algorithm (CRBA, see [5]) and the Articulated-Body-
Algorithm (ABA, [6]). Both algorithms yield the same solution but differ in run-
time with the CRBA performing better than the ABA for systems with few joints.
Implementation
To allow the efficient simulation of the robot's motion dynamics an object-oriented

framework has been developed using C++. Within this framework, the robot's
structure is modeled as a chain of modeling entities consisting of the robot's base,
variable and fixed rotations and rigid bodies (consisting of a fixed translation in
combination with the body's mass, center of mass and inertia tensor). Additional
modeling entities not used for this research include variable translations (to de-
scribe prismatic joints) and forks (to build tree-shaped structures beyond the kine-
matic chain). To allow for arbitrary structures these modeling entities can be
combined in any order so that one is not limited to the structure described above.
Similar approaches have been used successfully for the simulation of industrial ro-
bots [7], biomechanical systems [8] and autonomous mobile robots [9].
The framework provides methods to solve the robot's kinematics and dynamics
equations, yielding solutions for the direct kinematics, the inverse dynamics and
the forward dynamics. Currently, the forward dynamics is based on the CRBA and
ABA to allow the selection of the better performing algorithm, depending on the
complexity of the robot.
Due to this modular description of the robot the structure of the simulated robot
can be exchanged easily without the need to derive new equations of motion.
Thus, it is possible to select (and re-select) the virtual joints required for a specific
milling simulation, depending on the concrete robot. For the purpose of parameter
estimation and trajectory optimization the framework also allows the calculation
of derivatives of the simulated robot's motion with respect to any modeling para-
meters.
This feature is based on the ADOL-C library [10] for automated derivation of
C++-functions. Depending on the current use of the developed framework one can
either use the special types provided by ADOL-C (if derivatives are required) or
the standard floating-point types of the machine (no derivatives available, but
faster execution). By this feature, it is possible to use the same implementation of

the model either for parameter estimation or for the simulation of the milling
process as well as for the application of numerical optimal control methods [11],
without changes in the source code.
11.2.2 Process Force Calculation

The calculation of milling forces is based on a material removal simulation that al-
so calculates complex cutter workpiece engagement conditions and therefore the
chip geometry. Based on the chip geometry the milling forces are calculated using
a standard model presented in [3]. While other methods for cutter workpiece en-
gagement simulations are introduced in [12, 13, 14, 15] the dexel representation
used in this work enables an efficient computation of the engagement condition.
Especially in applications of milling with industrial robots the accuracy of the
dexel-based method is considered to be appropriate. In order to increase the calcu-
lation accuracy, [16] recommends the usage of a multi-dexel model for the repre-
sentation of the workpiece (cf. Fig. 11.4).
+ + =
Fig. 11.4 Dexel representation of a workpiece in three directions in space
Thereby, the description of a single dexel is made by line equations,

p = dt + s
Here, p is a point on the line, d is the line direction, t is the scaling factor and s the
starting point in space. The workpiece is modeled using a multi-dexel representa-
tion compromising dexel directions in x, y and z. The discretization of the work-
piece is defined by the dexel distances dx, dy and dz. In order to receive a
sufficient accuracy of the force calculation the relation kd between the dexel
discretization and the tool radius R is considered,
dx dy dz
kd = = = = 0.05 .
R R R
252 J. Bauer et al.
During a simulated milling operation the tool moves through the workpiece in dis-
crete time steps ti. The incremental size of the time step is dt.
dttooth 2tooth
dt = .
nint
In every time step ti=1,…, nint the material removal is calculated, where dttooth 2tooth
is the tooth-passing time calculated by
60
dttooth 2tooth =
N z ·n
with the number of revolutions n and the number of teeth N z .
11.2.2.1 Calculation of the Chip Geometry
The material removal simulation calculates a point cloud representing the outer
volume of a single chip, which the discrete chip geometry is extracted from. The
chip geometry is described by the angular discretized chip thickness h(φ,z) with
the entry and exit angles φin, φout. As the chip thickness varies along the cutting
edge the chip gets subdivided into discs of the height dz and in dφ in angular di-
rection (cf. Fig. 11.5).
z radial
h(φ,z)
y
y
ap dz
dz
x dφ
x
Fig. 11.5 Discretized chip geometry
The chip thickness is calculated in three steps. In the first step, the set of points
is subdivided into disks of the height dz. As a next step, inside each disc the max-
imum angular points are defined as restricting points by φin and φout (cf. Fig.
11.6a). Due to the complex contact conditions and the complex chip geometries
the entry and exit angles vary over the depth of cut ap. The cross-sectional geome-
try of each disk is calculated based on the cylindrical form of the cutter at the dis-
crete time of ti and ti-1. The area between the input angle φin and exit angle φout is
discretized into multiple sub-areas with the angular distance dφ. The value of the
discrete chip thickness h(φ,z) with respect to φ is calculated by a line intersection
with the two cylinders in the final step (cf. Fig. 11.6b).
Fig. 11. 6 Calculation of the chip thickness h(φ,z)
Adding all disc levels together, a chart of the chip thickness h(φ,z) is extracted,
which forms the base of the force calculation. This is the most frequently per-
formed arithmetic operation during the removal simulation.
11.2.2.2 Cutting Force Model
For the prediction of the cutting forces a standard cutting force model based on
Altintas [3] is implemented, where the time delay responsible for the chatter is
neglected. According to the chip discretization the cutter is sectioned into discs of
height dz (cf. Fig. 11.7).
Fig. 11.7 Cutting force calculation
In each disc e, Frta,j represents the force per tooth j in radial, tangential and axi-
al direction,
Frta , j,e = K c ·dz·h j (ϕ , z ) + K e ·dz .
254 J. Bauer et al.
Depending on the angular position of the tooth j of a disc, the corresponding chip
thickness hj(φ,z) is inserted. The cutting force coefficients Kc=[Krc, Ktc, Kac] and
Ke=[Kre, Kte, Kae] need to be identified in advance.
A transformation of Frta,j,e with T(φ) and the subsequent summation over all
teeth Nz and discs Ne results in the process force Fxyz,tool, given in a non-rotating
tool coordinate frame,
Ne N z
Fxyz,tool =  T (ϕ )·F
e =1 j =1
j rta, j,e .
The force Fxyz,tool acts at the TCP of the robot and thus represents the interface to
the robot model.
11.3 Analysis of the Mechanical Robot Structure

In order to identify the mechanical model parameters of the robot several experi-
ments are performed including
• modal analysis,
• stiffness measurement within the working volume of the robot,
• stiffness measurement of the structural robot components.
In the following sub-sections, the investigations are presented and the consequen-
tial model adaptations are implemented.
11.3.1 Modal Analysis

The modal analysis is used to determine eigenfrequencies, eigenforms and modal
damping of the robot structure. The robot structure is excited by a hammer im-
pulse at a defined position. The response is recorded with a tri-axial acceleration
sensor, which the transfer functions are derived from for all 109 measuring points.
Figure 11.8 shows the robot structure and the measuring grid.
Fig. 11.8 Modal analysis of the robot

Based on the transfer functions, the eigenforms, eigenfrequencies and modal

damping is extracted. The robot KUKA KR 210 [17] has six dominant eigenfre-
quencies below 100 Hz. In Table 11.1, these frequencies are summarized and ex-
plained briefly.
Table 11.1 Eigenfrequencies and eigenforms of the robot
Eigenfrequency Modal Damping

Eigenform Description
[Hz] [%]
1 8.4 6.37 Oscillation about axis 1
2 11.1 1.02 Tilting of axis 1 around y
3 16.9 0.75 Tilting of axis 1 around y
4 20.6 0.16 Oscillation about axis 3
Oscillation about axis 2 and 3 plus
5 24.1 0.11
deformation of fork (axis 5)
Tilting of axis 1, 2 and 3 plus
6 57.9 0.64
torsion of the whole structure
11.3.2 Static Stiffness within Working Space

Measuring of the static stiffness in the working space is performed in x, y, and z-
direction at nine positions within the relevant working area of the robot. Static
forces Fx, Fy, and Fz are applied by a force measurement rod. Laser distance sen-
sors are mounted to an external measurement frame to capture the robot’s dis-
placements Δx, Δy and Δz (cf. Fig. 11.9). The static stiffness kxx, kyy, kzz is
calculated by
Fx = k xx Δx , Fy = k yy Δy , Fz = k zz Δz .
Fig. 11.9 Test-rig of the stiffness measurement

256 J. Bauer et al.
While evaluating the stiffness at a single measuring point by 3 cycles of tensile

and compressive force of F ∈[−1500 , 1500] N , the robot’s pose remains fixed.
Figure 11.10 illustrates the robot’s compliance - calculated by hxx=(kxx)-1,
hyy=(kyy)-1 and hzz=(kzz)-1 - at each point of the measuring grid. A global trend in
all directions towards a higher compliance is observed at those measuring points,
which have a larger distance to the robot base.
4 4 4
2 2 2
0 0 0
1 1 1
2 2 2 1
1 1
3 2 3 2 3 2
3 3 3
Fig. 11.10 Direct compliance within the working area (z = 900 mm)
In a similar way, the rotational and tilting stiffness are experimentally meas-
ured. While the force load is applied in defined directions at the spindle the dis-
tance sensors measure the tilting and rotation at each axis sequentially. In Table
11.2, the measured stiffnesses are summarized.
Table 11.2 Rotational robot stiffness at axes 1 – 6
Mech. Robot Component Rotational Stiffness [Nm/rad]

Axis 1 8,937e-6
Axis 2 6,343e-6
Axis 3 2,357e-5
Axis 4 7,804e-5
Axis 5 1,047e-4
Axis 6 2,052e-4
11.3.3 Parameter Identification

For a full parameterization of the robot model, mass, stiffness, and damping ma-
trices M, K and D need to be determined. The model adjustment is split into the
static and the dynamic adaption. Within the static model adjustment the measured
component stiffnesses (Sect. 11.3.2) are deployed in the robot model. The virtual
robot is positioned at points p1 to p9 corresponding to the experiment and a static
load is applied in the simulation. Hence, the deviations are simulated and the stiff-
ness of the full mechanical structure is determined. By adjusting the model’s
rotational and tilting stifffness values in order to compromise the measured stifff-
ness, the model gets em mpirical updated. The simulated and the measured stifff-
nesses within the working g area are compared in Figure 11.11.
2000 2000 2000

Messung Messung Messung
Simulation Simulation Simulation
n
1500 1500 1500
1000 1000 1000
500 500 500
0 0 0
1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9
Fig. 11.11 Comparison of sim

mulated and measured working area stiffness
For the model dynamiccs adaption, the mass and damping matrix have to be de-
termined. The mass and the t location of the centers of mass for each link are exx-
tracted from the technicall documentation. The inertia tensors of all rigid links arre
derived from CAD-representations. By calculating the frequency response funcc-
tions (FRF) of the virtuall robot compliance Hxx, Hyy, and Hzz and comparing them m
to the corresponding robo ot’s FRF, the dynamical correlations of the robot moddel
are evaluated (cf. Fig. 11.12). In order to improve correlation, the mass and stifff-
ness model parameters arre slightly adjusted. Finally, the peak height of selecteed
eigenfrequencies is adjustted by modifying the damping parameters to confirm thhe
measured FRFs.
Fig. 11.12 Frequency respon

nse function of the compliance H
After updating the mod del parameters, the simulated robot compliance results iin
a reasonable correlation to
t the measured compliance frequency response functioon
in the frequency band of f = 0 to 50 Hz. Especially the frequency and amplitude oof
the resonance peaks correelate well.
11.3.4 Simulation and

a Validation
The validation of the proccess force model is carried out separately from the robot
model. Figure 11.13 shows results of machining a test workpiece of aluminum m
3.1325 with a portal roboot. The used cutter is a two-fluted poly-crystalline cuttter
with d = 10 mm. The milliing parameters are n= 8000 rpm, vf = 50 mm/s, ap = 1 mmm.
258 J. Bauer et al.
100
Fx mea
0 Fx sim
-100
0 5 10 15 20 25 30 35 40 45
Zeit [s]
100
Fy mea
0 Fy sim
-100
0 5 10 15 20 25 30 35 40 45
Zeit [s]
0
Fz mea
-200 Fz sim
-400
0 5 10 15 20 25 30 35 40 45
Zeit [s]
Fig. 11.13 Validation of the process force model
The cutting forces are recorded by a Kistler force dynamometer during the ma-
chining operation. In order to calculate the mean cutting forces, the force signal is
filtered by a low-pass filter with a cut-off frequency of fcorner = 100 Hz. As the
main eigenfrequencies of the robot structure are lower than 60 Hz (cf. Table 11.1)
the consideration of frequency bands above 100 Hz are neglected for coupled ro-
bot milling simulations. The comparison of measured and simulated cutting forces
Fx, Fy, Fz (Fig. 11.13) shows a sufficient congruence of an overall quadratic mean
below 20 % of each force direction.
11.4 Model-Based Compensation

The two compensation strategies are characterized by a model-based approach us-
ing ideal milling forces (Sect. 11.4.1) and by geometrically mirroring the deviation
path (Sect. 11.4.2). The aim is the offline correction of static and low-frequency
deviations before the real milling process is started. Furthermore, this allows an
improvement of machining parameters as well as estimations concerning dimen-
sional accuracy or surface quality.
11.4.1 Compensation by Ideal Milling Forces

The proposed model-based compensation strategy consists of three steps: (1) A
simulation run assuming an idealized robot, (2) the computation of reference joint
positions and (3) a transformation into compensating track points.
Assuming a known ideal TCP trajectory, this idealized path is initially simu-
lated without considering that elasticities in the robot’s dynamics, i. e. real joint
positions, match the ideal set values ( qreal = qset and qreal = qset ) for virtual and ac-
tuated joints. Thus, a path h through a sequence of points on the workpiece is simuu-
lated using the robots kin nematics. This generates corresponding trajectories of thhe
process forces at the TCP P. As the computation is based on a discretization of thhe
workpiece and an infiniteesimal stepsize Δt is deployed in an accurate simulation,
force signals contain significant noise components. Therefore, low-pass filtering is
applied to get smooth trajectories (cf. Fig. 11.14) for further use.
Using the recursive Neewton-Euler-method [4], the inverse dynamics is solveed
to compute idealized to orques τideal(tk) for a selection of ideal joint positionns
{qideal (tk ) | k = 1, ... , kmax } . Then, with τ ideal and with the assumptioon
qcomp = qideal ,
K·(qcoomp − qideal ) + D·( qcomp − qideal ) = τ ideal
is solved for each tk as a linear equation system. This results in compensationnal

joint positions qcomp, which h includes virtual joints as well as the actuated ones.
As the robot input req quires for certain positions pcomp(tk), the calculated com
m-
pensational joint positions qcomp(tk) are transformed into base coordinates by solvv-
ing the direct-kinematics for each tk. Thus, the deviation according to the stiffness
of actuated and virtual joiints is considered in the resulting sequence of compensaat-
ing set points {qcomp (tk ) | k = 1, ... , kmax } .
Fig. 11.14 Deviation, correcttion and resulting path in simulation
The proposed model-b based strategy is computed very efficiently as only onne
simulation run with an id dealized, kinematic robot model is performed. The evenn-
tual solution of one lineaar equation system per interpolation point can be carrieed
out in real-time.
11.4.2 Compensatio
on by Deviation Mirroring
The second approach of path
p compensation is based on a geometrical considera-
tion of deviations caused by milling forces acting on the cutter (cf. Fig. 11.15(1.))).
The compensation strateg gy is applied in radial and axial direction of the cutteer.
260 J. Bauer et al.
Figure 11.15(2.) illustrates the idea of calculating a new NC-code, based on mir-
roring path points on the planned NC-path. Thus, when the milling operation is
conducted with the modified NC-code, a coincidence of planned and the resulting
TCP position is expected (cf. Fig. 11.15(3.)).
Fig. 11.15 Concept of path compensation by mirroring the cutter position.
To calculate a compensated path the machining operation is initially simulated

using the original NC-code. Thus, corresponding deviations of the TCP become
available and the offset values for selected points on the deviated path to corres-
ponding NC-path points are calculated. By mirroring these offsets at the planned,
ideal NC-path the compensational path points are computed. In order to increase
the overall path accuracy additional path points may be inserted into the original
NC-path for compensation.
11.4.3 Experimental Investigation

During the experiment a milling path describing a 90-degree corner is machined.
Within this test, static deviation and dipping in the corner are investigated. Based
on the ideal milling path the simulation initially returns deviations of 0.35 mm in y
and 0.06 mm in x-direction. The compensational NC-path is generated by mirror-
ing certain points on the ideal NC-path. The simulation with this compensational
NC-path shows a significant reduction of the path deviation under present process
force (cf. Fig. 11.16).
To validate these results from simulation a low feed rate of 1.5 mm/s and a mil-
ling depth of ap = 0.5 mm is initially chosen for the experiments as a full slotting
operation. This setting allows the neglecting of process forces and the manufactur-
ing of a workpiece without recognizable deviations. Subsequently, a milling op-
eration with a feed rate of vf = 50 mm/s and ap = 1.5 mm is performed. Due to the
process forces the TCP deflects and the resulting deviation at the workpiece is
measured by a microscope relative to the previous reference operation. Deviations
between 0.03 mm and 0.44 mm are measured on 7 positions along the path (cf.
Fig. 11.17). In the corner, a large difference of 1.69 mm between the planned and
the resulting path is measured due to path blending effects. The root mean square
error of all seven measured points equals erms = 0.72mm.
Fig. 11.16 Simulation of path

h deviation and robot program modification
The experimental millling operation is repeated with the modified robot proo-
gram including the comp pensating path points. The deviation is reduced primarilly
within the strait path segment. The root mean square error of erms = 0.72 mm is
now reduced to erms = 0.15 mm. In the corner, only a moderate error reduction waas
achieved because of the roobot’s internally-controlled path blending.
Fig. 11.17 Test workpiece without/with

w compensation
262 J. Bauer et al.
In z-direction, the profile of the machined part with and without compensation
is measured on a coordinate measuring machine. The compensated path is slightly
closer to the supposed NC-path on the straight path segments. In the corner, the
dipping of the cutting tool is reduced by ∆z = 0.029 mm (cf. Fig. 11.18).
Fig. 11.18 Z-deviation of the machined workpiece
The main reason for only achieving a moderate compensation in z-direction is

related to the relatively low forces acting on the cutter in z-direction. While mean
forces Fx,1 = -130 N, Fy,1 = 58 N in the first and Fx,2 = 108 N, Fy,2 = 59 N in the
second straight path segment were observed, the z-force is only Fz,1 = 24 N and
Fz,2 = 39 N. The simulated mean force in both segments is Fz = 35 N, which the
compensation is calculated on. Thus, the path correction off-set based on the cal-
culated deviation is small compared to x and y- direction. Furthermore, the internal
numerical control of the available KUKA robot additionally causes path errors,
which cannot be described by the proposed robot model. This effect is even more
dominant when only a small deviation appears due to the low force level.
11.5 Conclusion
Industrial robots are a cost-saving and flexible alternative for machining applica-
tions in a low-accuracy area. A reduced position and path accuracy, especially un-
der process force load due to the high mechanical compliance, restricts the use of
industrial robots for further machining applications.
The proposed strategy is built on a coupled simulation module consisting of a
process force model and a robot model. This validated simulation enables the pre-
diction of a force-induced path deviation in robot milling operations. Experiments
have shown that a milling error occurs as a summation of robot control-contouring
error and the process force deviation of up to 0.44 mm. Depending on the control
system and the level of cutting force, the presented approach compensates the er-
ror caused by process force deviation to a large extend. The influence of the con-
trol-contouring error on the overall path deviation is more dominant in the case of
only small process forces. Since this control-contouring error could not be covered
in the simulation by the so-far implemented robot model, the path correction
approach results in only moderate improvements for this case of small process
forces. For further improvements, these effects of the robot’s internal path plan-
ning and control algorithms should be identified and implemented into model.
However, in the case of dominant process forces the path compensation con-
cept is able to reduce the milling error significantly.
Acknowledgments. The authors thank Max Stelzer, who carried out very helpful proofs
of concept in the first phase of the project.
References
[1] Abele, E., Bauer, J., Rothenbücher, S., Stelzer, M., Stryk, O.: Prediction of the Tool
Displacement by Coupled Models of the Compliant Industrial Robot and the Milling
Process. In: Conference on Process Machine Interaction, Hannover (2008)
[2] Weigold, M.: Kompensation der Werkzeugabdrängung bei der spanenden Bearbei-
tung mit Industrierobotern. PhD Thesis TU Darmstadt (2008) ISBN 978-3-8322-
7178-7
[3] Altintas, Y.: Manufacturing Automation: Metal Cutting Mechanics, Machine Tool
Vibrations and CNC Design. Cambridge University Press (2000)
[4] Craig, J.J.: Robotics. Addison-Wesley (1989)
[5] Walker, M.W., Orin, D.E.: Efficient Dynamics Computer Simulation of Robotic Me-
chanisms. Journal of Dynamic Systems, Measurement, and Control 104, 205–211
(1982)
[6] Featherstone, R.: Rigid Body Dynamics Algorithms. Springer (2008)
[7] Höpler, R.: A unifying object-oriented methodology to consolidate multibody dynam-
ics computations in robot control, PhD Thesis TU Darmstadt (2004)
[8] Stelzer, M.: Forward Dynamics Simulation and Optimization of Walking Robots and
Humans, PhD Thesis TU Darmstadt (2007)
[9] Friedmann, M.: Simulation of Autonomous Robot Teams with Adaptable Levels of
Abstraction, PhD Thesis TU Darmstadt (2009)
[10] Walther, A., Griewank, A.: ADOL-C: A Package for the Automatic Differentiation of
Algorithms Written in C/C++. Version 2.1.12, Documentation,
https://projects.coin-or.org/ADOL-C
[11] von Stryk, O.: User’s Guide for DIRCOL (Version 2.1): a direct collocation method
for the numerical solution of optimal control problems, TU Darmstadt (2000),
http://www.sim.informatik.tu-darmstadt.de/sw/dircol
[12] Rehling, S.: Technologische Erweiterung der Simulation von NC-
Fertigungsprozessen, PhD Thesis Universität Hannover (2009)
[13] Surmann, T.: Geometrisch-physikalische Simulation der Prozessdynamik für das
fünfachsige Fräsen von Freiformflächen, PhD Thesis TU Dortmund (2005)
[14] Stautner, M.: Simulation und Optimierung der mehrachsigen Fräsbearbeitung, PhD
Thesis TU Dortmund, München (2002) ISBN 3-8027-8732-3
[15] Selle, J.: Technologiebasierte Fehlerkorrektur für das NC-Schlichtfräsen, PhD Thesis
Universität Hannover (2005)
[16] Weinert, K., Müller, H., Kreis, W., Surmann, T., Ayasse, J., Schüppstuhl, T.,
Kneupner, K.: Diskrete Werkstückmodellierung zur Simulation von Zerspanprozes-
sen. In: Zeitschrift für wirtschaftlichen Fabrikbetrieb: ZWF, München, pp. 385–389
(2002) ISSN 0932-0482
[17] KUKA Roboter GmbH. Data sheet KR 210-2. Gersthofen - Germany (July 2009),
http://www.kuka-robotics.com
Chapter 12
Process Machine Interactions in Micro Milling
E. Uhlmann, F. Mahr, Y. Shi, and U. von Wagner
Abstract. In this chapter, both analytical and experimental studies on the process
stability of micro milling are presented. The investigations are carried out in order
to improve a comprehensive model, which describes interactions of the dynamic
cutting forces and the dynamic machine tool behavior including the end mill. Do-
minant chatter frequencies at different operating points are determined by analyz-
ing the process forces, acoustic signals as well as optical measurement signals.
Results are documented and discussed by means of stability lobe diagrams. The
findings are confirmed by analyses of the milled surfaces. Finally, some sugges-
tions for improving the parameter identification are given.
12.1 Introduction
Significant Process Machine Interactions (PMI) can be observed in micro milling,
e. g. by comparing the surface roughness of the manufactured workpiece to the vi-
bration-response signals measured on the machine tool housing [18]. Alternating
loads caused by interrupted cutting conditions interact with the flexible end mill
and the machine tool structure. Deviations of the Tool Center Point (TCP) are the
consequences.
In this context, a precise modeling of the interacting elements such as the machine
tool structure, the end mill as well as the cutting conditions is necessary for
process simulation. Stability predictions containing this information help to im-
prove the machining of complex micro-structures.
Based on the experimental modal analysis of the machine tool structure and the
operational vibration analysis of the micro end mill a comprehensive spatial mul-
tiple degrees of freedom (MDOF) model is described in this chapter. It is com-
posed of oscillator chains and a flexible tool, which is modeled as a rotating
Euler-Bernoulli beam using finite elements (FE).
The equations of motion are derived with Hamilton’s principle including gy-
roscopic effects as a result of the rotational motion of the end mill. Besides, a
self-excited part due to the internal material damping of the tool is included. The
dynamic cutting forces, which occur during the process, are coupled to the MDOF
model. Therefore, a geometrical milling force model is developed, which contains
significant characteristics of micro cutting such as ploughing effects below the
critical chip thickness.
266 E. Uhlmann et al.
Experiments are carried out in order to verify the process stability behavior.
The simulated and experimentally-determined stability lobe diagrams are then
compared and analyzed in detail.
12.2 Modeling and Stability Prediction

Due to PMIs in micro milling it is not sufficient to model the end mill as a lumped
mass for chatter prediction. Experimental modal analysis and operational vibration
analysis are carried out in order to obtain the characteristics of the machine tool
structure and the end mill.
Figure 12.1 shows a series of modal analyses conducted on a
WISSNER Gamma 303 HP 3-axis micro milling machine tool. In all experiments,
an electro-dynamic shaker is used to generate noise signal as excitation, which is
measured with a piezo-electric force transducer. The response signal is measured
with a tri-axial accelerometer.
Modal Analysis 1 Modal Analysis 2
reference points
spindle shaker
-6
x 10
3
mode shapes
|
N/m xx
|H | (25°)
xx
2
Magnitude
|Hyy|
1.5
0.5
0
500 1000 1500 2000 2500 3000 3500 4000 Hz 5000
Frequency
Fig. 12.1 Experimental modal analysis of the machine tool structure at different reference
points
In modal analysis 1, the reference point is assigned to a location on the sliding

carriage of the machine tool. The dominant mode shapes of the machine tool
structure can be identified. However, the coherence at points on the tool shank is
not sufficient. Therefore, the shaker and the accelerometer are mounted at the top
of the tool shank to identify the frequency response functions (FRFs) at the tool
12 Process Machine Interactions in Micro Milling 267
clamping. The measured FRFs in modal analysis 2 are shifted at different posi-
tions, where |Hxx| (25°) means a 25-degree rotation of the Spindle-Tool Holder-
Tool system (STT) from the initial measure position. All these phenomena indi-
cate that the spindle configuration significantly affects the modal results of the
system at non-rotating state. Considering the measured mode shapes, the first
three natural frequencies of the FRFs can be traced back to the sliding carriage of
the machine tool. The fourth mode is supposed to be caused by the clamping con-
dition at the STT system. Considering the significant redundant mass effect of the
accelerometer, the shaker and other unknown influences, the resonance frequency
position and the damping factor of this mode are not reliable and not directly uti-
lizable.
Due to the small dimensions of the micro end mill it is challenging to carry out
a classical experimental modal analysis directly. Figure 12.2 shows an operational
vibration analysis to study the resonance frequency of the tool.
Fig. 12.2 Operational vibration analysis of the micro end mill with a shaker as excited base
An end mill with a nominal diameter of 0.5 mm and a cantilever length of

20 mm is clamped. The assembly is mounted on a large shaker, which produces
noise signals as excitation. The excitation is measured with a tri-axial accelerome-
ter. Several points, including one on the clamp, are measured with a laser
vibrometer along the tool shank and the tip. The FRF (defined as response veloci-
ty/excitation acceleration) at tip point 8 is illustrated in Figure 12.2 (right). The
first deflection mode shape corresponds to the dominant resonance at about
8.7 kHz. It should be noted that point 2 at the clamped end also exhibits a small
transverse deflection, which means that the theoretical clamping point is located
between points 1 and 2, and the effective cantilever length is more than 20 mm.
This is caused by the imperfect clamping conditions of the simple assembly.
However, it can be concluded that the first bending resonance frequency of the
micro end mill with a cantilever length of 20 mm should be higher than 8.7 kHz
[6, 8].
Compared to the models built with FRFs in the frequency domain the system
models in the time domain are generally more complex. However, they also have
advantages, e. g. there are more methods available for the modeling and an
analytical stability analysis. The complex dynamics of the system such as the rota-
tion of the spindle-tool-system can be comprised. In this context, a spatial system
model composed of oscillator chains and finite elements (FE) is utilized in order
to incorporate both the machine tool modes and the flexible tool modes. Mean-
while, it also meets the requirements for process simulation and stability
prediction.
In order to identify the system’s parameters by an experimental modal analysis
the model is firstly treated without rotational effects, i. e. regardless of the spindle
speed and the according rotary inertia of the end mill. Subsequently, the rotational
effects are taken into consideration within the formulation of the equations of mo-
tion for further application.
12.2.1 System Model without Rotational Effects

Without rotational effects, the equations of motion are decoupled in two orthogon-
al directions. Figure 12.3. shows the configuration of a simplified model, which
describes the dynamic behavior at the tool clamping (see Fig.12.1. reference point
2). The tool is modeled by an Euler-Bernoulli beam with finite elements.
x1 x2 x3 x4 a) b)
y4
jY q6
mx1 mx2 mx3 Y
Z y3 q2 q7
uY
kx1 , dx1 , ... ,kx4 ,dx4 my3 y2 q3 X
s
q5
my2 Y uZ
y1 L q8
X jZ
z my1 q1
y q4
ky1 ,dy1, ... ,ky4 ,dy4 Z
x
Fig. 12.3 a) Spatial system model; b) Finite element of the micro end mill
The global reference frame (xyz) is inertially fixed on the milling plane. Rotat-
ing this frame by 90° about the y-axis defines the non-rotating reference frame
(XYZ) to simplify the description of the element kinematics while considering the
rotational effects.
When using the time-dependent generalized coordinates
qe=(q1, q2, q3, q4, q5, q6, q7, q8)T
at the two nodes of the element, the translational displacements of the element
ue=(uZ, uY)
can be expressed as
 uZ  =  φ1 −φ2 0 0 φ3 −φ4 0 0  q ,
u   0 0 φ φ 0 0 φ φ  e (12.1)
 Y  1 2 3 4 
where the shape functions φ1 to φ4 can be found in [14].

For the Euler-Bernoulli beam the shear deformation is neglected. Thus, the ro-
tations (φY, φZ) and translations (uZ, uY) are related by
∂uZ ∂u
ϕY = − , ϕZ = Y . (12.2)
∂s ∂s
The system equations of motion can be derived using Hamilton’s principle
t1 t1
δ  (T − U )dt +  δ Wdt = 0 , (12.3)

t0 t0
where T and U are the kinetic and potential energy of the system respectively, δW
is the virtual work of non-conservative forces and δ represents the variational
operator.
Because of the different geometries of the tool shank, the taper and the cutting
edges at the tool tip, it is not accurate enough to model the micro end mills by a
uniform beam. This can be demonstrated by the first three normalized bending
mode shapes at different boundary conditions as shown in Figure 12.4. Note that
the shank, tip lengths and diameters as well as the taper length considerably affect
the natural frequencies and mode shapes of the tools [6]. In this case, a micro end
mill with a cantilever length of 20 mm, a shank diameter of 3 mm, a nominal di-
ameter of 0.5 mm and the tip length of 1 mm is used as a demonstration. It can be
seen that the first bending mode is dominated by the shank. For clamped-free and
sliding-free conditions they are similar to a uniform beam. The second and third
bending modes are dominated by the taper, especially the tip deflections.
clamped-free: sliding-free: free-free:
1st bending mode 1st bending mode 1st bending mode

2nd bending mode 2nd bending mode 2nd bending mode
3rd bending mode 3rd bending mode 3rd bending mode
Fig. 12.4 Normalized mode shapes of the tool for different boundary conditions
To combine the modes of the tool and the machine, the parameters (mass, stiff-
ness and damping) of the oscillator chains should be identified. Sub-structure
coupling analysis is a useful method for this purpose. The detailed derivation of
this method can be found in [12, 16]. Generally, the entire machine tool system is
divided into two sub-structures: The complex machine tool structure including the
upper part of the tool shank (FRFs can be measured by experimental modal
analysis) and the rest of the tool including the difficult to measure tool tip (FRFs
usually obtained using beam theories or FE software). The identification of the
FRFs at the coupling is challenging due to the rotational degrees of freedom
(DOF). There are several methods to experimentally measure rotational FRFs or
analytically derive them from measured translational FRFs [5]. To simplify the
modeling the following assumptions are made.
For the analyzed micro milling machine tool, the clamping conditions be-
tween the end mill, the tool holder, and the spindle are perfect, i. e. sufficient
clamping length and quasi-isotropic contacts. The rotational DOF can be neg-
lected and the beam can be coupled to the oscillators with a sliding-free boun-
dary condition. Since the measure point is approximately located at the contact
position between the end mill and the tool holder, the effect of rotational DOF is
supposed to be excluded. Moreover, a rigid coupling is adopted here, i. e. the
rest of the end mill is coupled at the last oscillator without spring or damper
elements (see Fig. 12.3).
In order to identify the parameters of the oscillators the system model is sepa-
rately treated in the milling plane. In each direction, the parameters can be identi-
fied with the help of the measured and computed receptances using curve-fitting
methods. A demonstration of two typical measurements in x and y-directions is
shown in Figure 12.5. It should be mentioned that the identified oscillator parame-
ters need to be furthermore optimized for chatter prediction, since the resonance
frequencies and the damping factors are unverified especially at the fourth peak,
which can be traced back to the clamping conditions at the tool shank.
x 10-6 x 10-6
2 2
measured |H | measured |H |
xx yy
N/m identified |Hxx| N/m identified |H |
yy
Magnitude
1 1
0.5 0.5
0 0
0 1000 2000 3000 Hz 5000 0 1000 2000 3000 Hz 5000
Frequency Frequency
Fig. 12.5 Identified and measured FRFs
The mode shapes of the coupled system can be computed with the identified
parameters of the machine tool modes. Figure 12.6 shows the fourth and fifth
normalized mode shapes in y-direction.
Oscillator chain
th
in y-direction 4 mode shape 5th mode shape
undeformed shape undeformed shape

deformed shape deformed shape
Fig. 12.6 Normalized mode shapes of the coupled system in y-direction
As seen, for this kind of system modeling the fourth mode is dominated by the
oscillator chains including the tool as a rigid body. In contrast, the fifth mode in-
cludes a bending mode of the tool. Torsion, tension, compression effects and cor-
responding modes are not included here.
12.2.2 System Model with Rotational Effects

According to the oscillator chains configuration and the coordinate system selec-
tion, the rotational speed has no effect on the kinematics of the lumped mass.
Thus, their equations are trivial and can be found in [17]. Here, attention has to be
paid to the formulation of the finite element equations.
To describe the kinematics of the beam elements, a rotating frame (x1y1z1) is at-
tached on the neutral fiber of the beam. The rotation matrix (direction cosine ma-
trix) between the fixed frame (XYZ) and the rotating frame (x1y1z1) is defined by
introducing two intermediate frames as well as three successive rotations:
1. φ2 about Y defining the frame (x2y2z2),
2. φ3 about z2 defining the frame (x3y3z3),
3. φ1 about x3 defining the frame (x1y1z1).
It is assumed that a non-holonomic constrain
ωe ⋅ e X = Ω (12.4)
holds, where ωe is the angular velocity vector of the beam elements with respect to
the fixed frame, eX is the unit vector of coordinate X in the fixed frame, Ω is the
constant spindle’s angular velocity. It should be pointed out that after linearization
about small deformations, the rotations (φ2, φ3) can be substituted approximately
by the rotations (φY, φZ) shown in Figure12.3 b).
The total element’s kinetic energy is composed of translational and rotational
parts and can be expressed as
L L
1 1
Te =
20 µ e vTe v e ds +  ωTe Θe ωe ds ,
20
(12.5)
where μe is the mass per unit length of the finite element, the absolute veloci-
ty ve is the derivative of ue=(uZ, uY) with respect to time in the non-rotating frame
and Θe is the mass moment of inertia of the element. While neglecting the shear
and torsion deformations as well as the axial loads, the potential energy of the
element can be expressed as
L
1
2 0
Ue = EI e u ′′e T u′′e ds , (12.6)
where E is the Young’s modulus, Ie is the second moment of area of the element
and
ue˝=∂2ue/∂s2. (12.7)
The internal viscous damping of the tool is assumed to be proportional to the ele-
ment bending stiffness EIe with an introduced damping factor di [19, 21]. By var-
iational calculation the linearized element’s equations of motion are
e + ( di K e − ΩG e ) q e + ( K e + di ΩK se ) q e = fe ,
Meq (12.8)
where the element matrices are similar as stated in [14, 21]. It is noteworthy that a
self-excited term arises due to the rotation and internal viscous damping of the
tool, which differs from regenerative effects in cutting processes.
The assembled system’s equations of motion consisting of the finite elements
and oscillator’s equations of motion are accomplished with the compatibility con-
straints at the corresponding element nodes. In total, the comprehensive model has
170 degrees of freedom (DOF). It should be noted that the system’s equations ex-
pressed in the fixed frame have a simple matrix form due to the non-rotating oscil-
lator chains. When they are transformed into the rotating frame, time-dependent
terms (cosΩt and sinΩt) appear in the oscillator-related components [17].
12.2.4 Process Model for Micro Milling

A complex geometrical model is implemented in order to calculate the dynamic
cutting forces. Regarding the cutting process, one significant difference between
macro and micro cutting is the influence of the cutting edge radius rβ. While in
macro milling rβ is significantly smaller than the chip thickness h, in micro milling
rβ is of the same order as h. Above the critical chip thickness hcr shear deformation
is dominant. Below this value elastic deformation becomes dominant and the ef-
fect of ploughing occurs [11].
Geometrical correlations can be used to estimate the critical chip thickness,
hcr = rβ ⋅ (1 − cos Θ m ) (12.9)
where Θm is the critical angle, which can be determined from the negative effec-
tive rake angle γeff (γeff = π/2 – Θm).
Empirical investigations from several researchers report a strong dependency of

the workpiece material to the ratio hcr/rβ as follows:
• micro cutting of copper values of 0.1 to 0.2 can be found in [13],

• aluminum and steel values of 0.25 and 0.38 are reported in [20].
• L’Vov [10] derived a neutral point of Θm = 45° independent from friction. In
this case, with (12.8) hcr/rβ is 0.293.
To predict the surface roughness the correlation
2
d d 2  fz 
Rkin = − −
4  2 
(12.10)
2
can be extended by the critical chip thickness as [3]
f z2 hcr  d ⋅ hcr 
Rth = + 1 + . (12.11)
4d 2  2 ⋅ f z2 
Figure 12.7 shows the critical chip thickness and the theoretical surface roughness
based on the above-mentioned theories.
2 2
theoretical roughness Rth
critical chip thickness hcr
0.4 0.4
µm µm 0.293
0.293
0.2
0.2
1 1
hcr/rβ = 0.1
hcr/rβ = 0.1
0,5 0,5
0 0
0 2 4 6 µm 10 0 2 4 6 µm 10
cutting edge radius rβ cutting edge radius rβ
Fig. 12.7 Influence of the cutting edge radius r on the critical chip thickness hcr (left) and
on the theoretical surface roughness Rth (right)
Since the geometrical estimation of the surface roughness already indicates sig-
nificant influence of the cutting edge radius, the cutting forces and the process sta-
bility are supposed to be affected as well [8]. Figure 12.8 shows a schematic view
of the geometrical conditions at the cutting edge considering this point.
workpiece
+ Δy t effective
Δx t rake angle
γeff
cutting
78,°
Ch
edge
end mill
h(ϕ) wedge angle β
ip
y t-τ
+Δ
cutting edge
τ
Δx t- M clearance
angle α
rβ
h(ϕ)
hcr
s
y α workpiece
x
Fig. 12.8. Schematic view of the cutting force model
The chip thickness (12.12) can be described by a function of the rota-

tion angle φ, which is divided into segments φ(z) in order to consider the helix an-
gle λ of the end mill. Furthermore, information about the trochoidal cutting edge
trajectory, self-induced and separately induced vibrations resulting in deflections
are also included.
h (φ ) = f z ⋅ sin(φ ) + f z ⋅ n ⋅ N z ⋅ φcurrent + Δxt −τ + Δyt −τ + Δx + Δy (12.12)

      
 
hstatic hdynamic hprevious cut δ seperately induced
By multiplying (12.12) with a screen function g(φ) (1 or 0), as described in [4], the
interrupted cutting conditions as they occur in milling can be set according to the
cutting parameters. The volume below the critical chip thickness Vpl is also de-
rived geometrically to take the ploughing effect into consideration.
   
( )
2
 rβ − rβ − hcr
2
   d
( ) ⋅(r )
2
V pl =  rβ2 ⋅ arcsin  2  − rβ − rβ − hcr
2
β − hcr ⋅ 2 (12.13)
  2 ⋅ rβ  
   
The cutting forces in radial and tangential direction can now be calculated by
z =a p
 (
K ⋅ a ⋅ hφ ) + ( K ⋅ V )
xf
Fr = r p p ,r pl (12.14)
z =0  
cutting ploughing
z =a p
 (
K ⋅ a ⋅ hφ ) + ( K ⋅ V ) ⋅ (1 + μ )
xf
Ft = t p p ,t pl f (12.15)
z =0     
cutting ploughing friction
Ki and μf are specific constants which have to be determined through experiments

or taken from appropriate literature. xf describes the non-linear behavior of the
specific force constants according to different ranges of the feed per tooth fz [9].
In order to investigate the cutting edge effect experiments are carried out with
different materials. The choice orients to commonly used materials in micro mil-
ling especially for the die and mold fabrication (e. g. powder metallurgic steel,
AmpColoy). Figure 12.9 shows the specific force constants.
5 5
10 10
specific force Kr
specific force Kt
2 2
N/mm N/mm
3 3
10 10
2 2
10 10
0 50 100 µm 200 0 50 100 µm 200
depth of cut ap depth of cut ap
5
10
PM X 190 CrVMo 20 (52HRC)
specific force Ka
AmpColoy944
2
N/mm Cu
PMMA
3 CuZn39Pb1
10
ae = d (full immersion cutting)
d = 1 mm (TiAlN coating)
2 n = 29.000 rpm
10
0 50 100 µm 200
depth of cut ap
Fig. 12.9 Specific forces in radial, tangential and axial direction for different materials
To analyze the complex PMIs in micro milling, brass (CuZn39Pb1) is selected

to be the most appropriate material. The specific force constants only possess mi-
nor non-linearities in a large field of operating points and lower tool wear can be
estimated compared to hardened steel. Nevertheless, the cutting forces are high
enough to evoke chatter.
At this point, the cutting forces can be calculated in radial and tangential direc-
tion. By applying the standard rotation matrix
Θφ =  − sin(φ ) − cos(φ )  (12.16)

 − cos(φ ) sin(φ ) 
the cutting forces can be written in x and y- direction and compared to the meas-
ured process forces.
Fxy = Θφ [ Fr Ft ] T (12.17)
Figure 12.10 shows calculated cutting forces under variation of the helix angle
(left) and the cutting edge radius (right).
Variation of the helix angle λ Variation of the cutting edge radius rβ

d = 1 mm λ = 0° d = 1 mm rβ = 1 µm
Kr = 2900 N/mm 2 λ = 10° Kt = 2900 N/mm2 rβ = 5 µm
Kt = 2900 N/mm2 Kr = 2900 N/mm
2
λ = 20° rβ = 10 µm
fz = 10 µm fz = 10 µm
ap = 70 µm λ = 30° ap = 70 µm λ = 30°
rβ = 1 µm λ = 40° µf = 0,1
4 4
force in x-direction Fx
N N
0 0
-2 -2
-4 -4
4 4
force in y-direction Fy
N N
0 0
-2 -2
-4 -4
0 90 180 deg 360 0 90 180 deg 360
rotation angle φ rotation angle φ
Fig. 12.10 Influence of the helix angle (left) and the cutting edge radius (right) on the
cutting forces
The calculation shows a significant influence of these parameters. The sudden

rise of the cutting forces at small helix angles is noticeable. Also, increased cutting
forces due to a higher cutting edge radius can be seen, which is typical for wearing
end mills. This means that neglecting these effects might cause strong divergences
when simulating the cutting forces and applying them to the machine tool struc-
ture. Especially the dynamic excitation of the end mill is affected [15].
The above-described force model consists of the major influences, which occur
in micro milling. However, because of its non-linear and complex character it is
difficult to be transferred into the frequency domain. It is therefore rather predes-
tinated for simulations in the time domain. Thus, for the following analyses in the
time domain some simplifications are made:
• The feed per tooth fz is constantly set to 10 µm, which reduces the influences of
the ploughing effect,
• all experiments are carried out in full-immersion cutting to reduce high dynam-
ic excitation frequencies at the entrance and exit of the cutting edge.
The used cutting force model corresponds to the implementations described in [4].
Solely the specific force constants Ki are adapted according to the varying depth of
cut ap and the adjusted spindle speed n (see Fig. 12.9).
12.2.5 Stability Prediction

Stability charts can be calculated to predict stable cutting processes and to im-
prove the efficiency of the production. In macro-scale milling, two different types
of instabilities can appear. They are referred to as (secondary) Hopf instability and
period doubling (also flip instability). Period doubling usually occurs at low radial
immersion and when using cutting tools with a small number of cutting edges. It is
therefore under steady investigation for high speed cutting (HSC) processes. Sev-
eral analytical methods for stability prediction have been successfully developed
both in the frequency domain [1, 4] and in the discrete time domain [2, 7]. The
methods vary regarding the required computing time, which strongly affects their
efficiency when high-order DOF systems are analyzed. In this context, the FE
model is only applied to the case of full immersion cutting. Thus, the zeroth-order-
approximation method [1] is used.
For stability analyses, the linearized form of the cutting force at a stationary
operation point, i. e. the constant feed rate, is coupled to the system equations of
motion. The equations can be transformed into the frequency domain by approx-
imating the force directional coefficients with only the zero order of their Fourier
series. Although the spindle speed is explicitly involved in the governing delay-
differential-equation (DDE) by the gyroscopic and internal damping terms, the
change of the structural dynamics (i. e. FRF at tool tip point) due to the spindle
speed can be neglected for the considered spindle speed range. Reasons for that
are the non-rotating modeling of the oscillator chains and relatively small rotary
inertia of the end mill.
12.3.1 Experimental Set-Up

All experiments are carried out on a WISSNER Gamma 303 HP 3-axis milling
machine tool using a Precise SC3062 high-frequency spindle with a maximum
spindle speed of 60,000 rpm. Due to effects such as tool wear and premature tool
breakage the micro milling process can still be seen as hardly reproducible. There-
fore, only one type of end mill is used. The two-fluted end mills with a nominal
diameter of d = 1 mm are TiAlN-coated and possess a helix angle of λ = 30°. All
experiments are conducted with brass (CuZn39Pb1) as workpiece material. To en-

able statistical analyses three tests are carried out at each operating point.
Different measuring devices are used to gather signals describing the process
characteristics.
• the process forces are measured with a Kistler Dynamometer 9256C2,
• the acoustic characteristics are recorded with a conventional microphone
(maximum sample rate: 16 kHz),
• the tool deflection and its velocity are measured with a Polytec Vibrometer
OFV-353.
All signals are logged with a National Instruments data acquisition system
NI9162–USB and signal-treated (e. g. FFT) with MATLAB.
12.3.2 Results
The verification of the calculated stability charts requires a reliable criterion for
unstable process conditions. Therefore, it is common practice to divide measured
signals into their spectral components by applying a Fourier transformation. In
case of unstable process conditions chatter frequencies can then be detected. Fig-
ure 12.11 exemplarily shows the spectra of measured forces and microphone sig-
nals at a spindle speed of n = 28,000 rpm and increasing values of the depth of cut
ap from 60 µm to 105 µm.
0.2 0.2
Magnitude
0.1 0.1
110 110
µm µm
p
p
a
a
ut
ut
0 80 0 80
fc
fc
0 0
o
2 4 6 kHz 60 2 4 6 kHz 60
pth
pth
10 10
Frequency
de
Frequency
de
Fig. 12.11 Spectra of the force signals (left) and microphone signals (right)
Both signals clearly reveal typical chatter frequencies above an ap-value of

90 µm. It is remarkable that almost all chatter cases come along with multi-chatter
frequencies. However, high magnitudes above 6 kHz can only be found at unsta-
ble operating points. Their appearance is therefore used as a chatter criterion. Re-
garding the microphone signals it can be stated that the frequencies with greater
magnitude are related to the dominant natural frequency of the STT system. Other
frequencies arise from a combination between the basis chatter frequency and the
tool engagement frequency. This phenomenon is usually observed in macro-scale
milling with low radial immersion but not in full immersion cutting processes,
where only the frequency around the most flexible mode is addressed for chatter
[1, 4, 7]. Even though the spectrum of the microphone signals shows a lot more
noise, presumably due to the minimal quantity lubrication (MQL), chatter fre-
quencies could reliably be found. More frequency contents can be seen in the
spectrum of the force signals. The most significant is the cutting edge engagement
frequency. Others are the spindle frequency and its multiples. All decisions as to
whether an operation point is stable or unstable could be made from measurements
from both devices yielding the same result.
Considering the large number of measured data a way of clearly representing
the chatter frequencies is needed. For this purpose, the Campbell diagram is pre-
destinated. Here, the chatter frequencies at different spindle speeds are visualized.
Spindle speed dependencies of the chatter frequencies and possible causes for
non-linearities or other phenomena can be analyzed in detail. Figure 12.12 shows
Campbell diagrams for unstable operating points between 26,000 rpm and
33,000 rpm.
chatter frequencies scanned from force signals

chatter frequencies scanned from microphone signals
multiples of the spindle frequency n x fsp
calculated spindle speed dependend chatter frequencies
Fig. 12.12. Campbell diagrams from the force (left) and microphone spectra (right)
The parallel dashed lines signify the multiples of the spindle frequency. Espe-
cially in the Campbell diagram of the force signal it can be seen that some fre-
quencies are located on those lines. All frequencies between the dashed lines are
possible chatter frequencies. According to the theory of possible Hopf chatter fre-
quencies [7], the chatter frequencies correspond to basic eigenfrequencies between
7 kHz and 7.2 kHz, which is presumably caused by the STT system. Therefore,
the chatter frequencies are compared to calculated spindle-speed-dependent chat-
ter frequencies. A good accordance can be stated for both diagrams. However, the
microphone spectra are considered to be the superior method for chatter detection.
Figure 12.13 shows the analytically computed stability charts, calculated by using
the previously detected chatter frequencies, as well as the experimentally deter-
mined stability charts.
natural frequency damping ratio

Mode 4 7085.56 Hz & 7246.88 Hz 0.0127 in x and 0.0051 in y direction
Mode 2 799.09 Hz & 799.10 Hz 0.0289 in x and y direction
Mode 3 1736.64 Hz & 1736.78 Hz 0.0189 in x and y direction
Mode 1 337.38 Hz; 0.0165 in x and y direction
experimentally determined stability limit
250
µm
unstable cutting
depth of cut ap
150
100
stable cutting
50
0
26.000 27.000 28.000 29.000 30.000 31.000 rpm 33.000
Spindle Speed
Fig. 12.13 Experimentally determined stability chart and simulated stability lobes with dif-
ferent chatter frequencies
It can be seen that the fourth mode of the machine tool model is dominant. The
measured basic chatter frequency corresponds to this mode. The asymmetric mod-
eling in x and y-directions also takes the mode interaction into account. Influences
of other modes cannot directly be determined from the chatter frequencies due to
their small significance compared to the fourth mode. However, their effects can
be observed from the milled surfaces as shown in Figure 12.14.
200 µm 200 µm 200 µm
Fig. 12.14 Milled surfaces with stable cutting conditions at 28,000 rpm (left), unstable cut-
ting conditions at 31,000 rpm (middle) and at 29,000 rpm (right)
The milled surface with stable conditions shows steady traces of the cutting
edge trajectories. In contrast to that, unstable cutting conditions lead to different
surfaces. Effects of overlaid vibrations can be seen additionally to the kinematic of
the cutting edge trajectories. At 29,000 rpm, high-frequent deviations of the TCP
are reasons for that. This corresponds to the fourth mode above of 7 kHz. At
31,000 rpm, a waviness caused by lower frequent vibrations can be seen. It is sup-
posed to be caused by the first three modes.
The chatter frequencies as seen in Figure 12.11 and Figure 12.12 vary in their
magnitude as well as in their frequency value. One reason for that are changing
conditions of the cutting process and the machine tool structure. Analyses of the
tool holder and the clamping conditions are carried out in order to point out the
most significant influences. Figure 12.15 shows the run-out as a function of
the spindle speed measured at the tool shank. Also, the chatter frequencies at
different clamping conditions are analyzed.
10
µm
0.02
6
deflection δ
velocity
4 m/s
loose
2
0
0 0 2 4
10.000 20.000 30.000 40.000 rpm 60.000 6 kHz stiff
Frequency 10
Spindle Speed
Fig. 12.15 Run-out as a function of spindle speed (left) and chatter frequency of the end
mill at different clamping conditions (right)
It can be seen that the tool run-out is strongly influenced by the spindle speed.
Therefore, the consideration of this effect within the comprehensive model is ne-
cessary. High magnitudes of frequencies, also at odd multiples of the spindle, and
tool engagement frequency in force can be explained by that.
A significant influence on the dominant chatter frequency depending on the
clamping conditions of the tool can also be stated. The difference of the chatter
frequency can amount up to 2 kHz, which strongly affects the calculation of the
stability chart.
12.4 Optimization
As depicted in the preceding section, chatter frequencies strongly depend on the
variation of tool clamping and process conditions. This complicates the analytical
determination of stability charts. A reliable method to identify chatter frequencies
without extensively repeating the milling operations at different depths of cut ap
would therefore be helpful. For this purpose, an experimental set-up is illustrated
in Figure 12.16.
Piezo-ceramic actuators are clamped between the workpiece and the clamping
vice on the machine tool. The workpiece is excited with different signals during
the milling process in order to evoke chatter.
clamping vice
micro end mill -3
x 10
4
velocity
m/s
80
µm
0 60
p
ta
0 50
cu
2 4 6 kHz 40
of
Frequency 10
laser beam
h
piezo-actuators
pt
from Vibrometer
de
experimental setup without excitation
1 1
-1
(m/s )/V 0.5
80 80
µm µm
0 60 0 60
p
p
a
a
50 50
ut
ut
0 2 4 0 2 4
c
c
6 kHz 40 6 kHz 40
of
of
Frequency 10 Frequency 10
h
h
pt
pt
de
de
excitation with chirp signal
Fig. 12.16 Workpiece excitation by piezo-actuators
During the milling operation, the velocity of the micro end mill is measured us-
ing a laser vibrometer. Chatter frequencies can be determined from the spectra of
the velocity of the end mill as soon as the process reaches the critical depth of cut
ap (in this case ap = 60 µm). When a chirp signal is applied to the piezo-actuators
the chatter frequencies can already be detected from the frequency response func-
tion (velocity/voltage) at a depth of cut of ap = 50 µm. The similar phenomenon is
also observed using noise signals as excitation, with more frequency contents,
however.
Thus, this method allows the location of chatter frequencies at a very early
stage of instability. Furthermore, no measurement equipment has to be applied,
which possibly changes the dynamic behavior. The intended milling parameters
can be used and the tool wear can be reduced compared to conventional methods.
12.5 Conclusion
This chapter has presented both analytical and experimental studies on the process
stability of micro-milling . Experimental modal analyses of the machine tool struc-
ture as well as of the end mill were carried out. Subsequently, a comprehensive
model including the machine tool dynamics as well as the process forces, was de-
veloped in order to calculate stability charts. The model includes rotational and
gyroscopic effects of the spindle-tool holder-tool-system (STT). The model for the
dynamic process forces is based on a geometrical description of the cutting
process. Influences of the helix angle as well as the cutting edge radius were taken
into account. Experimental determinations of the specific cutting forces were then
carried out to complete a preliminary parameter identification.
Using the model, calculated and experimentally determined stability charts
were compared. Measured force, acoustic and optical signals were analyzed and
employed as chatter criterion. The spindle speed dependence of the chatter fre-
quencies were visualized in Campbell diagrams in order to verify the basic chatter
frequencies. It is noticeable that the first bending resonance frequency of the ana-
lyzed end mills under stiff clamping conditions was detected at 8.7 kHz whereas
the measured chatter frequencies during the milling process were between 7 kHz
and 8.5 kHz. It can therefore be concluded that the chatter frequencies are not only
due to the dynamic behavior of the end mill but arise from complex interactions of
the structure eigenfrequencies. Measurements of the milled surfaces confirm this
presumption. Since the chatter frequencies significantly influence the stability of
the system, experiments on the tool run-out and the clamping conditions of the
tool holder were carried out. Divergences of the chatter frequencies of up to 2 kHz
were detected. As a result of the sensitive dynamic behavior an experimental setup
for chatter detection was introduced. The workpiece was excited with different
signals by piezo-electric actuators. Thereby, a reliable in-situ detection of the chat-
ter frequencies without modifying the dynamic system (e. g. by the use of addi-
tional measurement equipment) could be realized.
References
[1] Altintas, Y., Budak, E.: Analytical prediction of stability lobes in milling. CIRP An-
nals 44, 357–362 (1995)
[2] Bayly, P.V., Halley, J.E., Mann, B.P., Davies, M.A.: Stability of interrupted cutting
by temporal finite element analysis. ASME J. Manuf. Sci. Eng. 125, 220–225 (2003)
[3] Brammertz, P.H.: Die Entstehung der Oberflächenrauheit beim Feindrehen- Bericht
aus dem Laboratorium für Werkzeugmaschinen und Betriebslehre der TH Aachen,
Industrie-Anzeiger, Essen, Nr.2, pp. 25–32 (1961)
[4] Budak, E., Altintas, Y.: Analytical Prediction of Chatter Stability Conditions for Mul-
tidegree of Freedom Systems in Milling. Part I: General Formulation, Part II: Appli-
cation of the General formulation to Common Milling Systems. Transactions of the
ASME J. Eng. Ind. 120, 22–36 (1998)
[5] Duarte, M.L.M., Ewins, D.J.: Rotational degrees of freedom for structural coupling
analysis via finite-difference technique with residual compensation. Mech. Syst. Sig-
nal Process 14(2), 205–227 (2000)
[6] Filiz, S., Ozdoganlar, O.B.: A three-dimensional model for the dynamics of micro-
endmills including bending, torsional and axial vibrations. Precis. Eng. 35, 24–37
(2011)
[7] Insperger, T., Stépán, G.: Updated semi-discretization method for periodic delay-
differential equations with discrete delay. Int. J. Numer. Methods Eng. 61, 117–141
(2004)
[8] Jun, M.B.G., DeVor, R.E., Kapoor, S.G.: Investigation of the Dynamics of Microend
Milling—Part II: Model Validation and Interpretation. ASME J. Manuf. Sci.
Eng. 128, 901–912 (2006)
[9] Kienzle, O., Victor, H.: Spezifische Schnittkräfte bei der Metallbearbeitung.
Werkstatttechnik und Maschinenbau 47(5), 224–225 (1957)
[10] L’Vov, N.P.: Determining the minimum possible Chip-Thickness. Machines & Tool-
ing 40(4), 45–46 (1968)
[11] Malekian, M., Park, S.S., Jun, M.B.G.: Investigation of Critical Chip Thickness and
Micro Ploughing Forces. In: Proceedings of CIRP 2nd International Conference on
Process Machine Interactions, Vancouver, Canada (2010)
[12] Mascardelli, B.A., Park, S.S., Freiheit, T.: Substructure coupling of microend mills to
aid in the suppression of chatter. ASME J. Manuf. Sci. Eng. 130, 011010 (2008)
[13] Moriwaki, T., Shamoto, E.: Ultraprecision Ductile Cutting of Glass by Applying Ul-
trasonic Vibration. Annals of the CIRP 41(1), 141–144 (1992)
[14] Nelson, H.D., McVaugh, J.M.: The dynamics of rotor-bearing systems using finite
elements. ASME J. Eng. Ind. 98(2), 593–600 (1976)
[15] Schauer, K.: Entwicklung von Hartmetallwerkzeugen für die Mikrozerspanung mit
definierter Schneide, Dissertation, TU Berlin (2006)
[16] Schmitz, T.L., Donaldson, R.R.: Predicting high-speed machining dynamics by sub-
structure analysis. CIRP Ann. - Manuf. Technol. 49(1), 303–308 (2000)
[17] Shi, Y., Mahr, F., von Wagner, U., Uhlmann, E.: A spatial multiple degree of free-
dom machine tool model for micro milling simulation. In: Proc. 2nd CIRP-PMI,
MM06, Vancouver (2010)
[18] Uhlmann, E., Mahr, F., Shi, Y., von Wagner, U., Eßmann, J.: Interactions between
mechanical vibrations and surface roughness during the micro milling process. In:
Proceedings of CIRP 1st International Conference on Process Machine Interactions,
Hannover, Germany, pp. 978–973 (2008)
[19] Wauer, J.: Kontinuumsschwingungen. Vieweg+Teubner Verlag, Wiesbaden (2008)
[20] Yuan, Z., Zhou, M., Dong, S.: Effect of diamond tool sharpness on minimum cutting
thickness and cutting surface integrity in ultraprecission machining. Journal of Mate-
rials Processing Technology 62, S.327–S.330 (1996)
[21] Zorzi, E.S., Nelson, H.D.: Finite element simulation of rotor-bearing systems with in-
ternal damping. ASME J. Eng. Power 99(1), 71–76 (1977)
Chapter 13
Numerical Computation Methods for Modeling
the Phenomenon of Tool Extraction
B. Denkena, E.P. Stephan, M. Maischak, D. Heinisch, and M. Andres
Abstract. Tool extraction is a phenomenon, where the end mill slips out of the
shrink-fit chuck in axial direction during the cutting process. This leads to severe
damage of the workpiece, the tool and in some cases even the machine spindle. So
far, this is an unexplained problem with no repeatability. In this article, experi-
mental investigations such as scanning electron microscopy (SEM) and residual
stress measurements on the clamping surface of shrink-fit chucks affected by tool
extraction are presented. Furthermore, results from experiments on a special test-
rig and a mathematical approach, which aims at the prediction of failures due to
Process Machine Interaction, are described. Within the mathematical approach, a
finite element model of the tool and the tool holder is linked with a cutting force
simulation. The dynamic behavior of the spindle is determined by frequency re-
sponse function measurements. From these measurements, a modal model is de-
duced and coupled with the finite element model of the tool holder. The presented
mathematical model is used to compute the resulting stresses in the interface of
those components due to process forces.
13.1 Introduction
Thermal shrink-fit tool holders are widely used in high speed (HSC) and high per-
formance cutting (HPC), where high torque transmission and excellent concentric-
ity is required [1]. The clamping mechanism of shrink-fit chucks is based on an
interference fit between the chuck and the tool shank. By heating up the chuck,
typically by a high-frequency inductor, the inner diameter extends and the tool can
be fed into it. At cooling down, the chuck reverts to its original size and thus
clamps the tool. In comparison to other tool holders the clamping forces of shrink-
fit chucks are very high and concentricity typically is below 3 µm at 3 times the
tool diameter. Hence, especially for HSC milling operations, where spindle speeds
exceed 10,000 rpm, shrink-fit tool holders are commonly used due to the simplici-
ty of the tool holder’s design, which contributes to its balance. There are no
clamping mechanisms, which have to be counterbalanced (Fig. 13.1) [2].
Analytical dimensioning of shrink-fits can be found in DIN 7190 [3], which is
adequate for calculating the main parameters such as the joint pressure and the re-
quired interference depending on the intended loads. Rondé [4] performed
286 B. Denkena et al.
extensive experimental investigations on thermal shrink fit holders and developed

analytical approaches for the calculation of transferable forces and the deforma-
tion behavior of collet chucks. Furthermore, he pointed out that shrink-fit chucks
are excellent for transferring the highest torque compared to the other investigated
collet chucks and are therefore well-suited for HSC and HPC.
Fig. 13.1 Shrink-fit tool holder Ø 16 mm [Schunk]
Tool extraction is a phenomenon, where the end mill slips out of the shrink-fit
chuck during the cutting process, which leads to severe damage of the workpiece,
the tool and in some cases even the spindle [5, 6]. There is also the risk of injury
to the worker by splintering parts. Tool extraction is a problem mainly reported
from high speed or high performance cutting [7].
Fig. 13.2 Extracted milling cutter
Figure 13.2 shows a milling cutter, which slipped out of the shrink-fit chuck
and broke off during high speed milling of aluminum. From reported cases it is
known that an extraction of several millimeters sometimes occurs after only a few
minutes of operating time. In other cases, shrink-fit malfunction appeared after
months of operation. Until now, no specific repeated pattern has been recognized.
13 Numerical Computation Methods for Modeling the Phenomenon 287
The appearance of tool extraction can also not be connected to obvious chatter vi-
brations. Furthermore, there are no obvious faults related to the shrinking process.
Hence, the cause of this phenomenon is presently unknown but it is legitimately
assumed that it is an effect from the interaction of the cutting process and the
tool/tool holder/spindle dynamics. Thus, for investigating the phenomenon of tool
extraction the dynamic behavior of the spindle as well as the milling process have
to be considered.
Solution Approach
The objective is the comprehension and subsequently the prediction of critical

machining conditions (see Fig. 13.3). Therefore, after analyzing damage cases, the
phenomenon is analyzed by creating a simulation model of the tool and the shrink-
fit holder. This model enables the transient calculation of the stresses in the con-
tact zone between the tool shank and the holder’s clamping bore. The interaction
with the cutting process and the spindle dynamics is performed by coupling the
tool/tool holder model with a cutting force calculation on one side and a dynamic
model of the spindle on the other side.
Fig. 13.3 Solution approach
Supplementary to the simulations, experimental investigations are performed.

Shrink-fit holders and tools from damage cases are analyzed. Experiments for the
verification of the simulation models are carried out, and test series on a test rig
are conducted to reproduce the phenomenon of tool extraction. The comprehen-
sion as to which circumstances make the interface between the tool and the tool
holder fail enables the prediction of critical machining conditions.
The next section presents selected damage cases and shows results from ana-
lyses of the accordant shrink-fit holders. Furthermore, the test-rig for reproducing
the tool extraction experimentally is described as well as results from those expe-
riments. Subsequently, the simulation model of the tool and the tool holder is in-
troduced, followed by simulation results and a short summary.
13.2 Experimental Investigations

As the phenomenon of tool extraction is not yet known in detail, the analysis of
damage cases may help to understand the interaction between the tool shank and
the tool holder’s clamping surface. In this section, two damage cases are de-
scribed, where tool extraction occurred. Furthermore, the clamping surface of sev-
eral shrink-fit tool holders subjected to tool extraction has been analyzed and the
results are discussed here. Finally, a test-rig for reproducing tool extraction under
pre-determined conditions is shown and results are presented.
13.2.1 Damage Cases

The following damage cases occurred at high speed milling of aluminum work-
pieces for aerospace parts. In aerospace industry, integral components made from
aluminum are widely used. To avoid joints those parts are manufactured from sol-
id blocks. The advantage is the optimal usage of the material properties at a mini-
mum weight. As integral components typically have a material removal rate of
more than 95 %, high cutting speeds and feed rates are required for economic cut-
ting processes. In those high speed cutting processes tool extraction is a known
problem and leads to severe damage, not only to the tool and the workpiece but in
some cases also to the spindle and, thus, causes a significant economic loss.
Figure 13.4 shows a workpiece of Al7075, where the tool, a Ø25 mm serrated end
mill, slipped out of the holder by 5 mm at a revolution speed of 16,000 rpm and a
feed rate of 21 m/min. The depth of cut was continuously increased by a 5°-angled
tool path at 15 mm width of cut. At the corner, where the direction of feed changes
by 90 degrees and an increase of the immersion angle occurs, the cutter started to
move out of the shrink-fit holder in axial direction. After approximately 30 mm
the cutting forces exceeded the material limits and the tool broke off. The marks
on the workpiece right before the point of process interruption indicated severe vi-
brations of the tool. At that point, the end mill had slipped out of the shrink-fit
holder by approximately 5 mm.
In contradiction to the previous case, the tool damage described in the follow-
ing occurred at high performance cutting of titanium during a test series. As
shown in Figure 13.5, the workpiece was machined by slot milling with an overlap
of 17 mm with a Ø25 mm three-edged end mill. The depth of cut was 50 mm at a
spindle speed of 636 rpm and a feed rate of 0.15 mm per tooth.
Fig. 13.4 Aluminum workpiece subjected to tool extraction
Fig. 13.5 Tool extraction at titanium workpiece and broken tool with tool holder
As titanium is a high-tensile material, machining is carried out at comparatively

slow cutting speeds and the cutting forces are much higher compared to the ma-
chining of aluminum. In this case, a reinforced shrink-fit holder with a wall thick-
ness of 17 mm was used. The interference of the tool shank and the holder is
24 µm and thus lies within the tolerance compared to other shrink-fit holders.
Measurements with a tri-axial dynamometer mounted between the workpiece re-
vealed that process forces were increasing from line to line due to increasing wear
of the cutting edges.
13.2.2 Analysis of Shrink-Fit Chucks Subjected to Damage

In order to investigate the causes of tool extraction, shrink-fit holders from several
damage cases were analyzed by geometry measurements, scanning electron mi-
croscopy (SEM), residual stress measurements, hardness testing as well as surface
and roughness measurements. For comparison, new and normally used tool hold-
ers were also analyzed in the same manner. Measurements of the interference be-
tween tool shank and the holder’s clamping bore showed that the tool holders in
the damage cases do not have significant lower interference fits than other holders.
The subsequent analyses concentrated on the clamping surface of the shrink-fit

holders. Therefore, the holders were serrated into smaller parts as shown in Figure
13.6 in order to gain access to the clamping surface. With these parts surface mea-
surements at different positions were performed using SEM. Moreover, measure-
ments of the residual stresses in the clamping surface were performed at the five
axially-ordered spots designated in Figure 13.6.
Fig. 13.6 Shrink-fit holder subjected to tool extraction
With an X-ray diffractometer of the type GE XRD 3000 P the stresses were
measured in axial (x) and circumferential (y) direction. For the presented data,
damaged shrink-fit chucks with different tool diameters were analyzed. The re-
sults were compared to new shrink-fit chucks and to those already subjected to
several clamping cycles.
Figure 13.7 shows the mean values of the residual stresses divided into three
groups. New shrink-fit chucks, those subjected to several clamping cycles for ref-
erence and holders from damage cases. Subdivided by circumferential and axial
direction, mean values for each of the three groups were calculated and related to
the mean stresses of the new shrink holders.
The mean stress amplitudes proved that – compared to the new and normal-
lyused – the residual stresses in holders affected by tool extraction are significant-
ly higher than in new or normally used chucks. Especially in the circumferential
direction, the residual stresses increased at a higher percentage than in axial direc-
tion. This indicates that the material properties in the clamping surface are not
only affected in axial direction when tool extraction occurs but rather in circumfe-
rential direction.
Fig. 13.7 Measurements of residual stress in clamping surfaces
Thus, with the help of SEM measurements the clamping surface is further in-
vestigated. In Figure 13.8 a) a section of the clamping surface of a shrink-fit chuck
being used for several clamping cycles without tool extraction is shown. On the
surface, the grinding grooves from the manufacturing process can clearly be iden-
tified in circumferential direction. Moreover, the surface is flattened due to the
previously applied clamping pressure.
The bright marks on the clamping surface of the shrink-fit chuck in Figure 13.8
b), which was subjected to tool extraction, corroborate the theory of the tool shank
revolving and flexing in the clamping bore and piece-wise slipping in the axial di-
rection.
Fig. 13.8 SEM measurement of new and damaged clamping surface

Thus, neglecting thermal effects so far, it seems that not only axial slipping be-
tween tool shank and shrink-fit chuck but rather circumferential slipping is a main
phenomenon when tool extraction occurs. Hence, it is assumed that, due to dy-
namic interaction between the tool and the workpiece, tool extraction is a combi-
nation of axial and circumferential slipping. Once static friction is lost, axial force
induced by the right-handed helix angle of the milling cutter leads to a relative
axial movement.
13.2.3 Experimental Investigations on Test Rig

In order to experimentally determine the dynamic conditions, which lead to tool
extraction, a special test-rig has been designed and built up (Fig. 13.9). With this
test-rig, different tool holders can be exposed to dynamic loads at revolution
speeds of up to 20,000 rpm. The tool holder is mounted on a conventional milling
spindle and a carbide cylinder is clamped in the holder as a dummy tool. The car-
bide cylinder itself is connected to a rotor shaft designed for transferring the loads
to the tool/tool holder interface. The loads are applied via electromagnetic actua-
tors in radial and axial direction. With a second milling spindle, connected by an
axially-elastic clutch, torque loads can also be applied. The displacement of the ro-
tor shaft is measured at the radial lamination core in radial and axial direction with
five eddy current sensors. The rotational displacement between tool and tool hold-
er is measured via rotary-pulse generators in the spindles. The test-rig is controlled
via a rapid prototyping system, which allows fully-automated operation and the
deterministic application of harmonic and non-harmonic load spectra.
Fig. 13.9 Spindle test-rig at IFW

More than 100,000 tests were performed with diameter 12 and 16 mm tool-
tool-holder combinations at usual interference fits. However, test results showed
that dynamic loads applicable by the test-rig are too low to enforce shrink-fit fail-
ures in terms of relative axial or rotational displacement of tool shaft and tool-
holder. This might not only be explained by the forces but also by differences in
the structural dynamic behavior of the tool/tool holder/spindle assembly in the
test-rig compared with that in a cutting process. Thus, widened shrink-fit chucks
were used for further tests. At these chucks, the clamping bore was previously wi-
dened by honing. With this process, the cylindricity and surface quality of the
clamping bore are only minimally affected as honing is also the finishing process
in the production of shrink-fit chucks.
In the following, results from tests with widened shrink-fit chucks with a no-
minal diameter of 16 mm are presented. In these tests, tool extraction occurred
under several dynamic loads. The parameters defining the dynamic load are ampli-
tudes of radial and axial force, excitation frequency and torque. The loads in radial
(Fr(t)) and axial (Fa(t)) direction are then defined as a harmonic function:
Fr ( t ) = Ax sin ( 2π f x t ) + F0, x + Ay sin ( 2π f y t + ϕ xy ) + F0, y
(13.1)
Fa ( t ) = Az sin ( 2π f z t + ϕ xz ) + F0, z
The torsional load is defined as constant (M). The interference-fit as a parameter is

induced by using shrink-fit chucks with different widening. Furthermore, proper-
ties of the clamping bore have to be considered, such as cylindricity, surface
roughness and the coefficient of friction between tool shank and clamping surface.
During the tests, the radial, axial and angular displacement of the rotor shaft is
measured. Radial displacement is measured on the left and right side of the radial
laminaton core. As the measurement location on the right is closer to the tool/tool
holder assembly its amplitude y2 is used within the analysis of the test results. In
Figure 13.10, the ranges of the varied parameters and the discrete parameter val-
ues are shown. As the system of rotor shaft, tool, tool holder and spindle shows
frequency-dependent dynamic behavior when excited by harmonic loads, the exci-
tation frequency is an important parameter. Within only small frequency ranges
large variations in the resulting amplitude may occur. Thus, frequency is the pa-
rameter with the most variation. In order to reduce the number of experiments on-
ly few parameters were varied, as shown in Figure 13.10. Other parameters
remained constant or zero, i. e. the tests were performed at a constant revolution
speed of 1,000 rpm, only radial force in y-direction, constant axial force and
torque.
Analyzing the amplitude y2 (near the tool) reveals the dynamic behavior of the
system under forced vibration in radial y-direction. In Figure 13.11 (top), the y2 is
plotted against the excitation frequency fy for Ay = 1,250 N and F0,z = 250 N and
M ≥ 30 Nm. The diagram clearly shows some of the resonances of the system. A
modal analysis of the rotor shaft shows the first bending eigenfrequency at 130 Hz
with a very high amplitude so that no tests are possible around this frequency. The
second and third eigenfrequency can be found at 335 and 417 Hz, both with a sim-
ilar mode shape, where bending occurs at tool and tool holder. The histogram in
Fig. 13.10 Ranges and values for varied parameters
Figure 13.11 (bottom) shows the distribution of failure occurrence related to exci-
tation frequency. The histogram includes all experiments, where an axial dis-
placement and/or a rotational displacement between tool shank and tool holder
occurred (both referred to as “failure”). Obviously, more than 90 % of the failures
occurred around an excitation frequency of 400 Hz, the third resonance in
y-direction of the rotor shaft/tool/tool holder/spindle system, even though there are
higher amplitudes at other frequencies. Thus, it can be concluded that frequencies
exciting mode shapes with high bending at tool and tool holder promote tool ex-
traction. Furthermore, test analyses revealed that the rotational load also plays an
important role as tool extraction has only happened under significant torque so far.
Of course, other parameters, especially the interference fit, are also responsible for
the combination of load parameters leading to tool extraction.
In Figure 13.12, an example of tool extraction occurring during a test is shown.
The excitation frequency fy is 398 Hz with an amplitude Ay of 1,250 N and a tor-
que load M of 35 Nm. The bottom right plot shows the rotational displacement be-
tween tool shank and tool holder. At the time the torque load is applied, the tool
starts revolving in the shrink fit chuck. At approximately the same time, the tool
moves axially out of the holder, both movements slightly decrease with time. It is
characteristic within this test that an axial displacement can occur even without an
axial load. In most of the tests, the rotational displacement is much higher than the
axial displacement.
Fig. 13.11 Amplitude y2 against frequency (top) and occurrence of failures (bottom)
Fig. 13.12 Example of single analysis of tool extraction

In Figure 13.13, the clamping surfaces of a shrink-fit chuck subjected to expe-

rimental tests but without tool extraction (a) is compared to surfaces from shrink-
fit chucks, where tool extraction occurred (b - d). In Figure 13.13a, only the
grooves from honing can be seen whereas b, c and d all show significant stripes in
circumferential direction, apparently caused by the revolving of the tool shank. In
axial direction, no significant marks can be found. This is due to the fact that the
ratio of circumferential displacement and axial displacement related to the clamp-
ing surface in all experiments with tool extraction lies between 3.5 and 20,000.
The ratio tends to rise with increasing interference fit. Comparing these results
with the analysis of the damage cases presented in section 13.2.2, it is assumed
that in the case of tool extraction a sudden increase of torque first leads to a re-
volving of the tool followed by an axial movement primarily caused by axial
process force. In order to understand what happens in the interface of the tool and
the tool holder, finite element simulations are performed with a mathematical
model of the tool and the tool holder is described in the following sections.
Fig. 13.13 Clamping surface of shrink-fit chuck without (a) and with tool extraction (b - d)
13.3 Mathematical Modeling

For a numerical simulation allowing the prediction of tool extraction during a cut-
ting process a finite element model of the tool holder and the shrunk tool is essen-
tial. The governing equation describing the cutting process is the visco-elastic
wave equation. Equipped with appropriate boundary and initial conditions, the
displacement field u(x,t) occurring in the system of tool and tool holder during the
process satisfies the following boundary and initial value problem (BIVP).
 ( x, t ) − β div σ ( u ( x, t ) ) − div σ ( u ( x, t ) ) = f ( x, t )
ρ ( x)u in Ω × ( 0, T )
u ( x, t ) = U ( x, t ) on Γ D × ( 0, T )
(13.2)
σ ( u ( x, t ) )·n = T ( x, t ) on Γ N × ( 0, T )
u ( x, 0 ) = u ( x, 0 ) = 0.
In the above problem, the domain Ω represents the structure of tool and tool hold-
er and its boundary ∂Ω is decomposed into the two disjoint sets ΓN and ΓD. The
given data is the density function ρ(x), which is assumed to be independent of
time. Moreover, volume forces f(x,t) as well as boundary tractions T(x,t) on the
Neumann boundary part ΓN and inhomogeneous displacements U(x,t) on the Di-
richlet boundary part ΓD are assumed to be given during the whole process. More
precisely, the boundary tractions result from the cutting process and act at the tip
of the tool and the inhomogeneous displacements result from the dynamic beha-
vior of the spindle interacting with the tool holder. Both the cutting forces and the
vibrations of the spindle were modeled separately to save a time-consuming finite
element model consisting of all of the four mentioned structures (spindle, tool,
tool holder and workpiece), see also the upper right image in Figure 13.3. In the
following sub-sections, the finite element model of the tool and the tool holder,
the model of spindle dynamics, the cutting force model and the coupling of the
three models are described in detail.
13.3.1 Finite Element Model of Tool and Tool Holder

In the above BIVP, the associated material law is Hooke’s law for linearly-elastic
material. The tool is made of carbide and since the considered workpiece is alumi-
num no plastic deformations are assumed to occur in the tool or in the tool holder.
A transient three-dimensional finite element model is used to approximate the
BIVP for the structure of tool and tool holder. The time discretization is performed
with the Discontinuous Galerkin Method (DGM), see [8], using piece-wise qua-
dratic basis functions. Frictional contact conditions on the contact interface be-
tween tool and tool holder within the model have been neglected since the two
structures are assumed to stick along the whole contact interface after the shrink-
ing process up until critical tangential stresses are exceeded, i. e. tool extraction
occurs. Hence, in the above problem, the domain Ω represents the compound of
tool and tool holder owning inhomogeneous material parameters, see Figure
13.14.
Fig. 13.14 Cross-section of the FE-model of tool and tool holder
The shrinking process is approximated by a linear heat strain law. The corres-
ponding contact pressure is modeled by applying a negative change of temperature
to the tool holder. The required temperature change is computed with the follow-
ing formula
Δr
ΔT = (13.3)
αT ri
where ∆r denotes the interference fit, αΤ represents the coefficient of thermal ex-
pansion of the tool holder and ri considers the inner radius of the tool holder. This
approach leads to a volume force denoted by f(x,t) in the BIVP, which is indepen-
dent of time. In the weak formulation, the volume force reads
α T ΔTψ (t )  (3λ ( x) + 2μ ( x))div(ϕ ( x))dx (13.4)

Ω
where ψ(x) and ϕ(x) denote temporal and spatial basis functions respectively and
λ(x) and μ(x) are the Lamé coefficients depending on the materials.
To simulate the process numerically a transient three-dimensional finite element
method is implemented considering the close-to-process structure of tool and tool
holder. Therefore, the elastic wave equation is re-written into a system of first or-
der equations in time by introducing the velocity v as an additional unknown vari-
able such that the DGM in time can be applied
u − v = 0
(13.5)
ρ ( x ) v ( x, t ) − β div σ ( v ( x, t ) ) − div σ ( u ( x, t ) ) = f ( x, t ) .
By applying the DGM in time and standard FEM in space to the above problem,
the corresponding discrete weak formulation results in a linear system of equa-
tions for the coefficients of the finite element basis functions of displacement and
of velocity. To circumvent large displacements due to the rotation of the system
the displacements and the velocities are described in rotational coordinates. By
further applying the approach of Li and Wiberg [9] the system can be decoupled
into a system concerning the velocity and three linear equations for the displace-
ment coefficients. Finally, the stiffness matrix resulting from the divergence term
in the elastic wave equation is used to approximate the material damping with
damping factor β within the process. Let ⊗ be the tensor product, then the two li-
near systems are:
 − k − 
 M ρ v + F1 − 2 Au 
 v1   
 
Cρ ⊗ M ρ + CC ⊗ M C + C A ⊗ A · v 2  =  k − 
 F2 − Au 
(13.6)
v   2
 3 
 k − 
 F 3 + Au 
 3 
 k 
 Au − + Av 3 
15
 A 0 0   u1   
    − k k k 
 0 A 0 · u =
 2   Au + Av 1 + Av 2 − Av 3
(13.7)
 0 0 Au   2 2 3
  3  
 − k Av + k Av + k Av 
 1 2 3 
 4 4 10 
Within the systems the matrix Mρ denotes the mass matrix, MC the mass matrix
resulting from the coriolis force and A = K – ω2MZ the sum of the stiffness matrix
K and the mass matrix MZ resulting from the centrifugal force multiplied by the
square of the revolution velocity ω. The three by three matrices Cρ, CC, and CA
(see 13.8) are the corresponding coefficient matrices resulting from the analytical
time integration of the DGM and the decoupling of the system, where ϕι are the
hat basis functions and the first bubble function in time.
 12 1 −1
  −32 −1 1
  8 β24+ 3 k 4β +k −10 β − 3 k

 4β +5k 
2 3 3 3 24 60
   −1  8 β +3k −10 β − 7 k
Cρ =  −21  , CC = kω  3  , C A = k  24
−2
 (13.8)
1 1 1
2 3 3 3 24 60
 13 −1
0   13 1 −4   −10 β − 7 k −10 β − 3 k 30 β +13 k 
3 3 15  60 60 225 
The time step is denoted by k and the vectors u– and v– represent the coefficient
vectors of the displacement and the velocity at the end of the previous time inter-
val respectively. Finally, the vectors F1, F2, and F3 correspond to the right hand
side vectors resulting from the volume force and the boundary conditions. The two
linear systems have to be solved in each time step.
A prediction of an extraction of the tool can be made by observing the stresses
occurring at the interface of tool and tool holder during the process simulation.
Using the DIN 7190 norm [3], which concerns the necessary contact pressure of
shrunk tools, an extraction may be possible, if the ratio of contact pressure and
tangential stresses falls below a determined bound. Of course, this may occur only
locally so that experimental tests have to be made to validate the simulations.
13.3.2 Model of Spindle Dynamics

Since the spindle is not close-to-process a direct implementation of the whole
structure into the transient finite element model described in Section 13.3.1 seems
unnecessary. However, the dynamic behavior of the spindle may lead to devia-
tions and vibrations of the whole system, which cannot be neglected.
For this reason, a sub-structure modeling technique using experimental modal

analysis (EMA) is applied. General techniques for sub-structure modeling are de-
scribed in [10–12]. Receptance coupling techniques for the tool/tool hold-
er/spindle assembly are especially subject to numerous works related to stability
prediction for cutting processes [13–20].
The model of the spindle dynamics obtained from EMA is characterized in
modal degrees of freedom. In order to integrate such a model in a finite element
simulation it is transformed to a model containing physical degrees of freedom re-
ferred to a virtual point P at the joint between the tool holder and the spindle, see
Figure 13.15a.
Fig. 13.15 Measurement of spindle dynamics with HSK flange from separated tool holder
To derive the model in modal degrees of freedom 24 frequency response func-

tions (FRFs) were measured on an HSK-A-63 flange of a separated shrink-fit
chuck, as depicted in Figure 13.15b. The system was excited at two reference
points in radial direction by using an impact hammer with integrated force sensor.
The resulting excitation of the structure was measured by four tri-axial accelera-
tion sensors mounted on the HSK-A-63 flange. The obtained acceleration FRFs
were integrated twice in order to obtain the receptances. A modal model with 34
degrees of freedom within the frequency band of 200 to 2,500 Hz was created us-
ing a poly-reference parameter estimation method.
By using the modal mass, damping and stiffness matrices MΦ, DΦ, KΦ and the
incomplete mode shape matrix Φ  obtained from the EMA, the system can be
transformed into the following ordinary differential equation (ODE) with both
physical and modal degrees of freedom representing the dynamic behavior of the
spindle at the HSK flange:
x(t ) + DT x (t ) + KT x(t ) = F(t )
MT 
(13.9)
x(0) = x (0) = 0
Within this system, MT, DT, and KT denote the transformed mass, damping and
stiffness matrix respectively. F(t) is the force vector resulting from excitation of
Fig. 13.16 Comparison of measured and modeled FRFs
the interface between tool holder and spindle. The first six components of x(t) de-
note the three translational and rotational degrees of freedom at the center point P.
Figure 13.16 compares the drive point measurements with the estimated main de-
grees of freedom in radial x and y-direction from the transformed modal model.
As can be clearly seen, the FRFs estimated from the transformed modal model re-
produce the measurements well.
13.3.3 Cutting Force Model

The boundary tractions T(x,t) in the BIVP resulting from the cutting process are
computed by a cutting force model [21–23]. They correspond to the cutting forces
of the process and are indispensable as they cause an excitation of the structure of
tool and tool holder. The cutting force model is based on the assumption that the
forces are proportional to the infinitesimally small cross-sectional surface Aij and
the associated arc length bij . The part of the cutting edge indenting the workpiece
from the tip of the tool to the cutting depth ap is divided into n chips, where i de-
notes the chip number and j the number of the cutting edge, see Figure 13.17. To
compute the surface A ij the cutting edge trajectory is used to determine the chip
thickness hij . According to Altintas [24], the proportionality of the tangential,
radial and axial force components Ftra to Aij and bij is linear and demands the
experimental determination of shear force coefficients Ktra,c as multipliers to the

surface and cutting coefficients Ktra,e as multipliers to the arc length.
Once the coefficients are determined for the process the forces acting at each
cutting edge j are computed to
{Ftra } j = K tra , c Aij + K tra ,e b ij .
i
(13.10)
As the revolution velocity is assumed to be constant the tool rotation angle φ is a
linear function of time t.
Fig. 13.17 Model for determining the cutting edge trajectories and the cutting forces [23]
If the forces are transformed into Cartesian coordinates by the transformation

xyz
matrix Ttra , i. e.
{F } {Ftra } j
i
= Ttra
xyz i
xyz
(13.11)
j
the corresponding process force at rotation angle φ can be computed as the sum
over all chip numbers and all nz cutting edges to
nz
Fxyz (ϕ ) =  {Fxyz } .
n
i
(13.12)
j
j =1 i =1
However, this procedure assumes an ideal stiff tool, which is not the case in reali-
ty. For this reason, the model is extended considering self-excitations of the tool,
which may be caused by the process forces. In this way, the chip thickness hij is
computed by considering the cutting edge trajectory at the rotation angle φ as well
as a given deviation of the cutting edge at the same position and time. This will re-
sult in changes of the undeformed chip cross-section Aij and therefore in different
process forces.
13.3.4 Coupling of the Models

To predict a possible tool extraction during a cutting process the excitation of the
structure of tool and tool holder resulting from the process forces has to be taken
into account. Furthermore, the structure is affected by the vibrations of the spin-
dle, which also result from the process forces. Hence, it is essential to couple the
cutting force model on one side and the model of spindle dynamics on the other
side with the transient finite element model of tool and tool holder, see Figure
13.18.
Fig. 13.18 Coupling of the models
The coupling of the process forces is accomplished via a Dirichlet-to-Neumann

mapping (DEN). The cutting edges in the cutting force model are discretized as n
force vectors acting on the corresponding point of the cutting edge. Therefore, the
finite element mesh at the tool tip is discretized such that the element edges of the
mesh follow the cutting edges in the helix angle α, see also the left image of Fig-
ure 13.18. In this way, the n force vectors can be interpolated along the edges to a
line force corresponding to the boundary traction T(x,t) in every time step. By ap-
plying the DGM the cutting forces contribute to the three vectors F1, F2, and F3 on
the right hand side of the system (13.6).
The implementation of the interaction of the structure of tool and tool holder
with the cutting force process is explained in the following. The coupling process
starts with an ideally stiff tool in the cutting force model, i. e. the deflection of the
tool is U0xyz (t0 ) = 0. The starting deflection of the tool for all subsequent time steps
is determined by the deflection of the previous time step. The computed cutting
0
forces Fxyz (ti ) depending on the process parameters result in a deflection U1xyz (ti ) of
the tool for every time step ti. By using this deflection within the cutting force
model in the same time step as described at the end of Section 13.3.3, the new
1
forces Fxyz (ti ) are calculated with the effective tool deflection. This procedure is
j −1
j
repeated until the residual of two consecutive forces Fxyz (ti ) − Fxyz (ti ) falls below
a given bound. When the iteration loop has converged, the stresses on the interface
of tool and tool holder can be computed. The coupling procedure at time step ti is
visualized in Figure 13.19.
Fig. 13.19 Coupling of the cutting force model and the FE-simulation of tool and tool
holder [6]
On the other side, the vibrations of the spindle excited by the cutting forces in-
teract with the tool holder at the corresponding interface, see Figure 13.18. When
the dynamic behavior of the spindle is known in terms of the ODE (13.9), it is
coupled via a DEN mapping with the FE model of tool and tool holder by the in-
homogeneous Dirichlet data U(x,t) in the BIVP (13.2). Here, the Dirichlet boun-
dary ΓD is defined as the interface between tool holder and spindle. Starting with
U(x,0) = 0 the displacement U(x,ti) is approximated by the solution x(ti) of the
ODE (13.9). In detail, for every x on ΓD and every time step ti, the corresponding
displacement is the sum of the translation in the first three components of x(ti) and
the rotation of x around the center point P with rotation angles in the fourth to
sixth component of x(ti)
The force vector F(t) is necessary to solve the ODE (13.9). The deflection of
the tool, which is caused by the cutting forces, results in an oscillation of the Di-
richlet boundary ΓD. Hence, the forces and moments in the first six components of
the force vector F(t) on the right hand side of (13.9) are computed by the solution
vectors u1, u2,, and u3 of the linear system (13.7). By computing the stress tensor
σ(u) due to Hooke’s law, the forces and moments are computed to
{F(ti )}13 :=  σ (u ( x, ti )) ⋅ n dx
ΓD
(13.12)
{F (ti )}46 :=  ( x − P ) × (σ (u ( x, ti )) ⋅ n)dx
ΓD
where n denotes the normal outward to the Dirichlet boundary ΓD. Since the ei-
genfrequencies expressed by the model of spindle dynamics are not affected by an
excitation the last 28 components of F(ti) are set to zero. To ensure that the correct
data is exchanged within the coupling an iterative loop, as shown in Figure 13.19,
is applied for every time step. This leads to a series of force vectors Fj(ti). Again
the inner iteration stops, if the difference of two consecutive force vectors
F j (ti ) − F j −1 (ti ) falls below a predefined bound. The cycle for a fixed time step
ti is visualized in Figure 13.20.
Fig. 13.20 Coupling of the spindle dynamics and the FE-model of tool and tool holder [6]
13.4 Application of Simulation Methods

To approximate a cutting process with the help of the simulation methods de-
scribed in Section 13.3 the coupling of the different models has to be validated.
For the coupling of the cutting force model with the finite element model of tool
and tool holder the cutting forces of a cutting process are measured. These meas-
ured forces were compared with the forces resulting from the cutting force model
when the coupling of the two models mentioned above is applied to the same
process.
Fig. 13.21 Transient simulation of process forces considering dynamic tool deflection
The effect of the coupling, i. e. the consideration of the tool deflection within
the cutting force model is visualized in Figure 13.21. The figure shows the forces
in radial (Fr), axial (Fa) and tangential direction (Ft), which act on the cutting
edges of the tool during one revolution of the cutting process. The continuous
lines represent the forces from the cutting force model considering tool
deflections, whereas the dashed lines are computed by the cutting force model
considering an ideally stiff tool. It turns out that the consideration of the tool dy-
namics has significant influence on the amplitude of the cutting forces, which
should not be neglected. Furthermore, effects such as regenerative chatter can only
be considered, if the models are coupled.
With the help of the coupled models, simulations with various load cases are
performed to investigate the stresses in the contact zone of the tool and the shrink-
fit chuck. As an example, Figure 13.22 shows the change in tangential stress in
axial direction related to the initial stress state of the shrunk tool during a transient
process simulation for a single time step.
Fig. 13.22 Change in tangential stress in tool/tool holder interface due to cutting process
The prediction of relative movement of the tool and the tool holder is made
possible by observing the stresses in the contact zone. If the tangential stress ex-
ceeds a certain limit defined by the normal stress and a friction coefficient tool ex-
traction is assumed, the simulation is stopped.
Summary
The article shows experimental and mathematical approaches for the investigation
of the phenomenon of tool extraction. The analysis of damage cases exhibits that
tool extraction is caused by the dynamic interaction of the tool/tool holder/spindle
structure with the cutting process. SEM and residual stress measurements of the
clamping surface of damaged shrink-fit holders corroborate the thesis that the tool
shank revolves in the clamping bore while simultaneously slipping out in axial di-
rection. With the help of a test-rig the phenomenon of tool extraction is repro-
duced under various load cases and main parameters are analyzed.
Furthermore, simulations with a finite element model of the tool and the tool
holder are performed concentrating on the stresses in the interface between the
tool shank and the tool holder. In order to consider the dynamic behavior of
the spindle a modal model is deduced from frequency-response-function-

measurements and coupled with the FE-model. The cutting process is simulated
by geometrical intersection of the tool trajectory with the workpiece. The process
forces are determined according to a cutting force model presented by Altintas
[24]. The coupling of these models takes into account the tool deflection induced
by the process forces and vice versa. Within transient simulations, the change of
the stresses in the interface region resulting from the process machine interaction
are computed aiming at the prediction of tool extraction.
References
[1] Schulz, H., Rondé, U.: Werkzeuge schrumpfspannen. Werkstatt und Betrieb 127(11),
873–875 (1994)
[2] Eastman, M.: Shrink-Fit Toolholding, Cutting Tool Engineering 49(3) (1997)
[3] Deutsches Institut für Normung e.V., Pressverbände - Berechnungsgrundlagen und
Gestaltungsregeln. Beuth Verlag, Berlin (February 2001) 17.040.10; 21.120.10/
[4] Rondé, U.: Untersuchung von Systemen zum Spannen von Zylinderschaftwerkzeugen
unter besonderer Berücksichtigung ihrer Eignung für die Hochgeschwindigkeitsbear-
beitung, Dissertation, Darmstadt (1994)
[5] Denkena, B., Stephan, E.P., Maischak, M., Heinisch, D., Andres, M.: Numerical
computation methods for process-oriented structures in metal chipping. In: Denkena,
B. (ed.) 1st International Conference on Process Machine Interactions, vol. 1, pp.
247–256. PZH Produktionstechn, Zentrum (2008)
[6] Denkena, B., Stephan, E.P., Maischak, M., Heinisch, D., Andres, M., Krüger, M.:
Investigations on Dynamic Tool, Structure and Process Interaction. In: Altintas, Y.
(ed.) 2nd International Conference on Process Machine Interactions, Vancouver
(2010)
[7] Fladerer, F.: Haltemomente sind beim Schrumpfen ein Maß für die Produktivität. Ma-
schinenmarkt 31(32), 28–29 (2007)
[8] Johnson, C.: Numerical solution of partial differential equations by the finite element
method. Dover Publ., Mineola (2009)
[9] Li, X.D., Wiberg, N.E.: Structural dynamic analysis by a time-discontinuous Galerkin
finite element method. Internat. J. Numer. Methods Engrg. 39(12), 2131–2152 (1996)
[10] Duarte, M.L.M.: Experimentally derived structural models for use in further dynamic
analysis, Dissertation, University of London (1996)
[11] Ewins, D.J.: Modal testing, Theory, practice and application. Research Studies Press,
Baldock (2000)
[12] Allen, M.S., Mayes, R.L., Bergman, E.J.: Experimental modal substructuring to
couple and uncouple substructures with flexible fixtures and multi-point connections.
Journal of Sound and Vibration 329(23), 4891–4906 (2010)
[13] Altintas, Y., Weck, M.: Chatter Stability of Metal Cutting and Grinding. CIRP An-
nals - Manufacturing Technology 53(2), 619–642 (2004)
[14] Park, S.S., Altintas, Y., Movahhedy, M.: Receptance coupling for end mills. Interna-
tional Journal of Machine Tools and Manufacture 43(9), 889–896 (2003)
[15] Filiz, S., Cheng, C.H., Powell, K.B., Schmitz, T.L., Ozdoganlar, O.B.: An improved
tool-holder model for RCSA tool-point frequency response prediction. Precision En-
gineering 33(1), 26–36 (2009)
[16] Schmitz, T.L.: Torsional and axial frequency response prediction by RCSA. Precision
Engineering 34(2), 345–356 (2010)
[17] Özsahin, O., Ertürk, A., Özgüven, H.N., Budak, E.: A closed-form approach for iden-
tification of dynamical contact parameters in spindle-holder-tool assemblies. Interna-
tional Journal of Machine Tools and Manufacture 49(1), 25–35 (2009)
[18] Ertürk, A., Özgüven, H.N., Budak, E.: Analytical modeling of spindle-tool dynamics
on machine tools using Timoshenko beam model and receptance coupling for the
prediction of tool point FRF. International Journal of Machine Tools and Manufac-
ture 46(15), 1901–1912 (2006)
[19] Ahmadian, H., Nourmohammadi, M.: Tool point dynamics prediction by a three-
component model utilizing distributed joint interfaces. International Journal of Ma-
chine Tools and Manufacture 50(11), 998–1005 (2010)
[20] Kolar, P., Sulitka, M., Janota, M.: Simulation of dynamic properties of a spindle and
tool system coupled with a machine tool frame. The International Journal of Ad-
vanced Manufacturing Technology, 1–10 (2010)
[21] Clausen, M.: Zerspankraftprognose und –simulation für Dreh- und Fräsprozesse,
Hannover (2005)
[22] Denkena, B., Tracht, K., Schmidt, C.: A flexible force model for predicting cutting
forces in end milling. Production Engineering – Research and Development 13(2),
15–20 (2006)
[23] Denkena, B., Schmidt, C.: Experimental Investigation and Simulation of Machining
Thin-Walled Workpieces. Production Engineering – Research and Development 1(4),
343–350 (2007)
[24] Altintas, Y.: Manufacturing automation. Metal cutting mechanics, machine tool vi-
brations, and CNC design. Cambridge Univ. Press, Cambridge (2000)
Chapter 14
Dynamic and Thermal Interactions
in Metal Cutting
P. Eberhard, U. Heisel, M. Storchak, and T. Gaugele
Abstract. This contribution presents a physical cutting process model based on

the Discrete Element Method (DEM), which allows the simulation of dynamic and
thermal interactions in metal cutting. Core component of the approach is the DEM
model of a solid with elastic-plastic deformation modes, which is verified in stan-
dardized tensile and Charpy impact tests as well as other non-standardized tests.
The model is enhanced such that the thermo-dynamics of a solid due to heat con-
duction can be included, which is also verified in different tests. The applicability
to model-cutting processes is shown in the simulation of orthogonal cutting
processes. The results of the simulation are compared to experimentally obtained
results for both forces as well as temperatures. For verification purposes, an FEM
model is made, which predicts both forces on the tool as well as temperatures.
14.1 Introduction
For decades, there has been a steady increase of the quality of machine tools. For
further improvement, a better understanding of the interactions between the ma-
chine tool and the physics of the manufacturing process is necessary. A mapping
of the process to an appropriate physical model is often challenging from a com-
putational mechanics point of view as there are often large deformations as well as
a steady evolution of boundaries and new surfaces. For a mesh-based approach,
this may prove to be awkward due to constant remeshing and a mapping of quanti-
ties between successive meshes. Alternatively, a meshless method like the DEM
naturally includes the separation of material and large deformations. A conven-
tional DEM model consisting of rigid spherical particles for granular material is
enhanced by cohesive visco-elastic-plastic interactions. This generates a ‘granular
solid’, which is subjected to virtual material tests to verify it and to determine pa-
rameter values. The mechanical state of the particles is extended to include ther-
mal properties and heat conduction, too. Thereby, an established workpiece model
is machined in orthogonal cutting processes. The obtained results are compared to
experimentally-obtained results and FEM references. The experiments are made
on a specifically designed test bed, which allows the measurement of forces on the
tool during cutting. To include different cases of cutting, the stiffness of the test
310 P. Eberhard et al.
bed as well as the angle of inclination of the tool can be modified so oblique cut-
ting processes can also be considered.
14.2 Experimental Investigations

For verification purposes experimental investigations are made. The following
section describes the setup and the results of the experiments.
14.2.1 Test Setup

To investigate the interaction processes in cutting a test bed is set up, whose CAD
model and machine prototype are shown in Figure 14.1. The bed of the test bed is
manufactured from massive polymer concrete slabs und carries the tool holder,
which is mounted in gantry design. In addition, the gantry is connected to the bed
via solid tension rods. The pre-stressing of the tension rods is adjustable. For this
reason, flexibilities can be systematically introduced in order to verify the coupl-
ing model. Moreover, a table is mounted onto the bed, which carries the tool hold-
er via a measurement platform. The table drive is realized by a linear direct drive,
which enables a continuous modification of the cutting speed from 0 to 200 m/min
and hence covers the conventional cutting speed range. The drive system consists
of a linear engine, a servo control unit including power electronics and a PC as su-
perordinate control. An important objective in the design of the test bed is to
largely disable the dynamic behavior of the structure when measuring the process
forces.
Fig. 14.1 CAD-model and general view of the test bed
To examine the stiffness of the test bed different measurements are performed
in feed direction (x-axis) and in the direction of depth setting (z-axis). A stiffness
of approximately 490 N/µm is determined in z-direction for the table and tool
14 Dynamic and Thermal Interactions in Metal Cutting 311
holder, which represents an excellent value. The stiffness of the gantry in x-

direction is 230 N/µm, which is also sufficient for the investigations to be con-
ducted.
In order to investigate to what extent the stiffness of the structure is influential,
flexibilities are systematically introduced into the structure by means of specifical-
ly developed stiffness elements. Thus, the stiffness between tool holder and gantry
as well as between workpiece and table can be reproducibly modified. This is car-
ried out by means of a “stiffness element”, which consists of two plates with disc
springs placed in-between, see Figure 14.2a. The stiffness of the elements is sub-
sequently modified in the range from 1,000 to 10,000 N/mm, whereby the effect of
the static stiffness, apart from the possible minor stiffness of the tool, can be rela-
tively well distinguished from the effect of the static stiffness of the entire setup.
The modification of the stiffness of the element is guaranteed by the number of
disc springs. Figure 14.2b exemplifies the examination of two different static
stiffnesses of the element. The diagram in Figure 14.2b suggests that the stiffness
element guarantees a maximum displacement of 0.5 mm. If further load is applied,
the element shuts and its two plates operate as one solid common body.
To investigate the effect of the static stiffness on the resultant forces and to de-
termine the resultant forces in orthogonal and free three-dimensional cutting, a
measuring chain was assembled and set up. The signals were recorded and
processed as well as controlled and evaluated by means of a program developed
under the LabVIEW GUI 7.1, which consists of three separately operating Lab-
VIEW programs. One program serves to record and visualize the data, another to
calibrate and another to statistically evaluate the resultant forces.
Free three-dimensional cutting is guaranteed by the twisting of the tool head
around the angle of inclination λS. The general view of the test setup for three-
dimensional cutting is shown in Fig. 14.3. In the investigations of the machining
of steel C45, the angles of inclination of 15° and 45° are used and the cutting force
Fx, the shear force Fy and the thrust force Fz are recorded.
a) b)
Fig. 14.2 Stiffness element and its static stiffness
Fig. 14.3 Performing a free three-dimensional cutting process on the test bed
The heat flows are established by experimental determination of the tempera-

ture change in the parts and assemblies. To investigate the effect of the ambient
temperature on the heat flows inside the machine structure, the test bed is put into
a climate chamber together with the entire measuring equipment required for the
test. In the course of the investigations, the ambient temperature is varied in a
range from +14°C to +36°C, see [4].
The experimental investigations to determine the temperature in the primary,
secondary and tertiary cutting zones are performed by means of semi-artificial
thermocouples, which are based on the Seebeck effect [5], [6] and [7]. Two me-
thods are applied. With the first method the temperature distribution in the work-
piece and in the chip can be examined [8]. One thigh of the thermocouple is made
of constantan wire, the other thigh of the material to be machined. Thus, a thermo-
couple of the type “J” is generated. With the second method, the temperature dis-
tribution at the border between wedge and chip can be examined [8]. In this case,
one thigh of the thermocouple is formed by constantan foil, which is clamped in-
between two carbide plates. The workpiece is used as the second thigh. The con-
stantan foil and all the tools are isolated from the upper and lower inserts. Depend-
ing on the relative position of the upper and lower inserts to each other, and de-
pending on the type of tool sharpening, the position of the foil can be modified
relative to the wedge point. Hence, the cutting temperatures can be measured at
different points, mainly in the secondary cutting zone. As is the case in the first
method, a thermocouple of type “J” is also generated. The signals are recorded
and processed as well as controlled and evaluated by means of a program devel-
oped under the LabVIEW GUI 7.1.
14.2.2 Temperature and Force Measurements

The following section summarizes the measurement of forces and temperatures.
14.2.2.1 Forces in Orthogonal and Free Three-Dimensional Cutting
The investigations of the resultant forces in both orthogonal and free three-
dimensional cutting represent a data basis for the verification of cutting models.
Figure 14.4 shows the course of the cutting and thrust force in the orthogonal ma-
chining of steel C45 with different cutting speeds from 50 to 150 m/min at differ-
ent cutting depths. As expected, the dependencies show a significant rise of the
forces with increasing cutting depths. The increase of the cutting speed, however,
has only little effect on the resultant forces.
In three-dimensional cutting, the components of the resultant force, i. e. the cut-
ting force Fx, the shear force Fy and the thrust force, increase, as expected, with in-
creasing cutting depth, see Figure 14.5. The size of the cutting force, in contrast,
hardly changes with an increasing angle of inclination λS. The course of the thrust
force flattens with a greater angle of inclination and the same maximum values.
The shear force Fy increases with an increasing angle of inclination. This reaction
is intensified with increasing cutting depths ap. The increase of the shear force
with an increasing angle of inclination can be explained with the extension of the
cutting length or width.
3500 1600
N 100 m/min
N
Cutting force
Thrust force
150 m/min
2000
800
1000 400 100 m/min

150 m/min
0 0
0 0,1 0,2 0,3 0,4 mm 0,6 0 0,1 0,2 0,3 0,4 mm 0,6
Depth of cut Depth of cut
Fig. 14.4 Resultant forces in orthogonal machining of steel C45
14.2.2.1 Investigation of the Temperatures in the Cutting Zones
Moreover, investigations on thermal phenomena in the cutting zones are con-

ducted in order to verify the cutting model. The temperatures in the primary, sec-
ondary and tertiary cutting zones and in the base material are used as references.
Characteristic signals in the primary cutting zone and in the chip are shown in
Figure 14.6.
2500
Cutting force
1500
1000
500 S=15°
S=45°
0
0 0,1 0,2 mm 0,4
Depth of cut
a)
800
S=15°
N S=45°
Shear force
400
200
0
0 0,1 0,2 mm 0,4
Depth of cut
b)
800
N
Thrust force
400
200 S=15°
S=45°
0
0 0,1 0,2 mm 0,4
Depth of cut
c)
Fig. 14.5 Dependency of the resultant forces on the angle of inclination λS
The shape of the signal and the amplitude during cutting correspond to the posi-
tion of the temperature sensor and to the constantan thigh in the different layers of
the material respectively. In practice, the method with the welded constantan wire
[8] can only be applied in comparably large cutting depths so as to realize a rea-
sonable resolution and to be able to precisely determine the position of the con-
stantan thigh. Therefore, this method was applied in experimental investigations to
determine the temperature in the primary and tertiary cutting zone and in the base
material. The method on the basis of the constantan foil clamped in-between two
inserts does not exhibit such a restriction and was hence used as the basic method
for recording the cutting temperatures in the secondary cutting zone. The change
in temperature in the secondary cutting zone or in the chip, depending on the
Fig. 14.6 Characteristic course of the signal in the primary cutting zone and in the chip
cutting speed and the distance L of the measuring point and on the position of the
constantan foil relative to the cutting edge of the die plate, is shown in Figure 14.7
for a free three-dimensional cutting process. The figure shows that the change in
temperature at a cutting speed of 50 m/min is extreme. The maximum is reached at
a distance of 1 mm between the measuring point and the cutting edge. This corres-
ponds to the known temperature distribution in the secondary cutting zone. At cut-
ting speeds of 100 m/min and 200 m/min, a temperature maximum is only reached
at greater distances to the cutting edge. The results of the FEM cutting model are
summarized in Table 14.1 for a distance of 0.25 mm to the cutting edge. The max-
imum difference between the simulated and experimentally recorded results
amounts to 13.8 %, which can be regarded as an acceptable error in thermal
investigations.
800 800
°C
Temperature
°C
Temperature
400 400
L=0,25
200 50 m/min 200 L=0,50
100 m/min L=0,70
L=1,00
0 200 m/min 0 L=1,25
0 0,4 0,8 mm 1,4 0 50 100 150 m/min 250

Distance from cutting edge Cutting speed
Fig. 14.7 Temperature change in the secondary cutting zone
Table 14.1 Simulated and experimentally recorded temperatures in the secondary cutting
zone
v, m/s Tsim, °C Texp, °C

50 370 319
100 475 438
14.3 Material and Cutting Models

Two cutting models are set up. One is based on the FEM, the other is built on a
mesh-free DEM approach.
14.3.1 FEM
The FEM model was developed with the commercial product LS-DYNA for the
orthogonal and three-dimensional cutting process [9], [10]. In setting up the model
it is assumed that the conditions along the cutting edge are the same. A boundary
condition for the workpiece is that the lower workpiece edge is to be fixed in both
directions. Regarding the tool it is assumed that the edges averted from the work-
piece are fixed in the direction of depth setting. The movement of the tool is pre-
determined by a rheonomic constraint in feed direction.
The algorithm for the plastic deformation comprises the following calculation
methods:
• Initialization (initial conditions, boundary conditions, maximum inte-
gration time step size and step size for remeshing)
• Organizing the time cycle
• Calculating the clamping condition
• Evaluating the fraction criterion (Johnson-Cook criterion, geometrical
criterion and criterion of maximum tensile stress. If one criterion is
fulfilled for one element, the element is removed.)
• Controlling the step size to renew the FE mesh
The calculation is organized by means of the developed program manager

OCFEM. The program manager has its own pre-processor to prepare the entry da-
ta, a mesh generator, the control of the LS-DYNA solver and the LS-DYNA post-
processor. Furthermore, another post-processor is available to extract the diagrams
of the cutting force components and of the specific chip forming work. A sche-
matic diagram of the operating mode of the OCFEM with external modules and
files is shown in Figure 14.8. The data transfer between the individual models
takes place by means of ASCI files. The developed FEM model was verified by a
comparison of the experimental and calculated values. The cutting and thrust
forces and the chip compression coefficient Кa were used as references. Steel C45,
which was cut with a cutting speed of 100 m/min at a depth of cut of 0.1 mm, was
used as test material. A value of 8° was selected for the cutting angle and a value
of 5° for the angle of clearance. The results of the verification of the FEM cutting
model are presented in Table 14.2. The f is a chip load per revolution. The results
suggest that the model provides adequate results for the cutting force but inade-
quate values for the thrust force and chip compression coefficient, i. e. values that
are clearly too low.
Fig. 14.8 Schematic diagram of the program setup with OCFEM
Table 14.2 Data for FEM verification
f, Кa Fx, N Fz, N
mm/rev Exp. FEM Err. % Exp. FEM Err. % Exp. FEM Err. %
2.72 -30 561±45 26 383±37 80
0.15 2.6 1.9 -27 519±36 416 20 334±31 76 77.24
2.4 -13 558±50 4.35 369±60 47.38
2.4 -21 940±55 11.5 473±57 77.6
0.30 2.2 1.9 -14 875±92 832 4.9 430±55 106 75.3
2.5 -16 534±39 4.8 377±38 51.2
14.3.2 Discrete Element Method

Some fundamentals of the Discrete Element Method (DEM) are summarized in
the following. This method is then used for cutting simulations.
14.3.2.1 Dynamics of Particle Systems
The DEM models the motion of usually free and rigid particles, which interact
with each other, and the system boundaries in case of physical contact. A DEM
algorithm can be sub-divided into three major steps, which are neighborhood
search, force computation and integration of the equations of motion. The neigh-
borhood search delivers the information as to which particles are in physical con-
tact and can therefore interact. Based on this information, the forces acting on the
particles are evaluated. Usually, a force-displacement approach is employed,
which calculates penalty forces based on negligible overlaps. The last step is to in-
tegrate the equations of motion.
Neighborhood search algorithms are not explained here. For information about
those and more detailed information about the method in general refer to [11].
The translational dynamics of a free particle i with position vector ri is readily
described using Newton’s equation, see [12],
miri = f i (1)
with the vector of external forces fi. If rotational dynamics also has to be consi-
dered, Euler’s equation of rigid body motion can be used, which reads
Ii ⋅ ω
 i + ωi × I i ⋅ ωi = t i (2)
where Ii is the particle’s inertia tensor, ωi is its rotational velocity and ti the vec-
tor of external torques. The differential equations (1) and (2) can be solved with
several numerical integration schemes. The model setup with dynamically moving
particles and evolving boundaries in conjunction with impacts indicates an inte-
grator, which is not sensitive to jumps in the right hand side of the equation. Fur-
thermore, it should be possible to implement it in an implicit manner for stability
reasons as in a general case the frequencies of the system are not known in ad-
vance for all times. For these reasons, an implicit Newmark scheme is chosen. See
[11] for the details.
14.3.2.2 Particle Force Laws
The right hand side of equations (1) and (2) contains force laws, which are eva-
luated based on force-displacement relationships. To generate a repulsive poten-
tial, which is used, if two separate bodies collide, either linear springs are used or,
limited to spherical particles, a Hertz contact force, see [13]. To define a realistic
material model of engineering materials, however, cohesive forces are needed.
Depending on particle shape, the physics of the problem and other factors, differ-
ent approaches can be employed. In this approach, spheres are used as the only
particle shape as the contact geometry is then determined fast and computational-
ly-efficient, thus allowing for a finer discretization. The chosen approach is to
connect adjacent spheres with massless, visco-elastic-plastic rods in their centro-
ids, which generates what will be called a ‘granular solid’.
The visco-elastic-plastic rods are represented by a force law, which has been
developed based on a piece-wise linear hardening model in [14]. Subjecting a duc-
tile specimen to uni-axial loading generates a stress-strain relationship, as depicted

in Figure 14.9. Loading beyond the yield point σ0.2 causes plastic flow of the ma-
terial. Analogously, the strain ε0.2 corresponding to σ0.2 can be defined. Subsequent
loading in the opposite direction causes the material to flow plastically further, if σ
< σy2 with |σ0.2| > |σy2|. Based on this observation, the model as illustrated in Fig-
ure 14.10 is developed. The stress strain relationship is governed by
σ = Erodε 0.2 + k (ε − ε 0.2 ) . (3)
A loading below σ0.2 results in a linear elastic state of stress with Young’s mod-
ulus E. Further loading beyond the elastic limit results in plastic flow with harden-
ing determined by parameter k. In that case, the initial stress-strain-curve is shifted
in the stress-strain-plane by the parameters σ0 and ε0, which are obtained by equat-
ing the lines g(ε) and h(ε), see [15] for a detailed explanation.
σ σ g()
σy
σ0.2 σ0.2
h()
σ0
0 y
σy2
Fig. 14.9 Stress-strain hysteresis Fig. 14.10 Elastic-plastic force law
The state of stress σij in a rod connecting two particles i and j results in a force act-
ing on the particles it connects. The elastic plastic force fi,ep acting on particle i is
calculated by
f i, ep =  σ ij Aij (4)
j
where Aij is the cross sectional area of the rod. It is calculated as arithmetic mean
of the radius of both particles involved. As each real material sustains loading only
to a certain limit, the rod connection is removed, if
σ ≥ σ max (5)
where σmax is the maximum bearable stress of the rod connection. This can also be
expressed as maximum bearable strain εmax.
14.3.2.3 Mechanical Properties of the Solid
The cohesive force interaction is applied to all particles, which mechanically inte-
ract in the initial configuration. The procedure to generate an initial configuration
is roughly sub-divided in a deterministic approach and two random approaches. In

the deterministic approach, the particles are arranged as a mono-disperse distribu-
tion in a regular, face-centered cubic lattice (fcc), which is a purely geometric
problem. The random approaches require a full simulation each time until a static
equilibrium of the system is reached. The first random approach is to drop a dis-
tribution of particles into a container under the influence of gravity and wait until
they come to a rest. The second approach uses a containment holding an aggregate
of particles, which is sufficiently smaller in volume than the containment. In the
course of the simulation, the particles are inflated and move randomly within the
box. Increasing particle radii cause increasing pressure on the box walls, which is
used as a trigger to stop the simulation as soon as a pre-defined pressure threshold
is reached. The random approaches can be used in a meaningful manner with
poly-disperse distributions.
The generated solid is subjected to several tests to assess its mechanical proper-
ties. These are a test for isotropy, a tensile test and a Charpy impact test. For an
isotropy test, specimens are subjected to slow tensile and torsional strain to ex-
amine how many particles are needed and how they should be arranged. Tensile
tests are made according to the appropriate standard, which allows for comparison
with experimental results and results from literature. The objective of a tensile test
with regard to a granular solid is to determine parameter values for the interaction,
which governs elastic-plastic behavior. For a more detailed explanation of tensile
and isotropy tests and some results, see [16]. To examine how the model behaves
under dynamic loading as opposed to static loading, as in the case of the tensile
test, a Charpy Impact Test is conducted as specified in the corresponding standard
[17]. In a Charpy impact test, a notched specimen with a square cross-section. as
shown in Figure 14.11, is hit by a hammer on the opposite side of the notch, as in-
dicated in Figure 14.12. The hammer is fixed to a pendulum and released from a
pre-defined height, thus hitting the specimen with a known kinetic energy. As the
specimen breaks during the impact and absorbs energy, the pendulum with
the hammer reaches a smaller amplitude after impact than it had before. The
amount of absorbed energy can then be readily calculated from the difference of
amplitudes.
hammer
55mm specimen
10mm
2mm
counter bearing
Fig. 14.11 Standardized test specimen Fig. 14.12 Test setup

A Charpy impact test is simulated with a regular specimen, i. e. the particles are
arranged in an fcc lattice and an inflated specimen with parameters as given in Ta-
ble 14.3.
Table 14.3 Parameters of Charpy impact test specimen
regular specimen inflated specimen

particle radius r [m] 4*10-4 4*10-4
number of particles [1] 20000 20000
Erod [N/m2] 5.31*1010 11.2*1010
Ε0.2 [1] 0.285% 0.295%
ε [1/s] 1.0 1.0
3
ρ [kg/m ] 2700 2700
The amount of absorbed energy for both types of specimen is shown in Fig-
ures 14.13 and 14.14. Increasing the maximum strain εmax before rupture for a rod
increases the amount of energy it can absorb. This can be seen in Figure 14.13,
where εmax is chosen to be 0.07, 0.08 and 0.09. The best agreement with the expe-
rimental value of 6.99 J is found for εmax = 0.08. In a similar manner, the absorbed
work is determined for a random specimen as seen in Figure 14.14. Here, the best
agreement is found for εmax = 0.09. An alternative setup of the test is given in [18].
0.07 0.08
8 8
0.08 0.09
0.09 0.10
6 experiment 6 experiment
energy [J]
energy [J]
4 4
2 2
0 0
0 0.01 0.02 0.03 0 0.01 0.02 0.03
displacement [m] displacement [m]
Fig. 14.13 Absorbed energy for regular Fig. 14.14 Absorbed energy for inflated
specimen specimen
14.3.2.4 Thermal Interaction
Apart from reproducing a cutting process on a kinematic and kinetic level, thermal
aspects have to be considered, too. Analogously to the mechanical state and
interactions, a thermal state and thermal interactions are defined. To each particle i
with mass mi and heat capacity ci a temperature Ti is assigned, whose change is
governed by
ΔQi = ci mi ΔTi (6)
where Qi is the internal energy. A finite change of internal energy ΔQi = ΔQj is in-
duced, if two particles i and j interact mechanically, i. e. are in contact. The ex-
change of energy is governed by Fourier’s law, see [19],
λAij
Q =
δ ij
(
Ti − T j ,) (7)
and depends on the heat conductivity λ of the material, the distance of the particles
δij and the cross-section Aij, which the heat flows through. For a definition of the
flux surface, see [23].
14.3.2.5 Setup of the DEM Orthogonal Cutting Model
For the orthogonal cutting model, a setup as depicted in Figure 14.15 is chosen.
The tool has no dynamics of its own. Constraints are used for both feed and depth
of cut h. The workpiece has length l and height H.
h
l
z
H
x
Fig. 14.15 Model setup for orthogonal DEM cutting simulation
An illustration of the model is shown in Figure 14.16.The rake angle α = 8° and

clearance angle γ = 4° are fixed. The tool surface is described by triangle elements
interacting with the spheres of the workpiece model based on the Hertz force law,
see [11]. The forces acting on each triangle element are accumulated on a common
reference frame and can thus be interpreted as cutting and passive force. For the
temperature interaction, all dissipative force components, e. g. plastic force com-
ponents, are interpreted such that they increase the internal energy and thus the
temperature. Depending on the material and the model, only a fraction 0≤β≤1 of
the plastic work, which is called Taylor-Quinney coefficient, causes a rise in
temperature.
Fig. 14.16 DEM cutting model
14.3.2.6 Comparisons of the DEM Model with Experimental Results
To assess the quality of the DEM model, experimentally obtained results for both
cutting forces and temperatures generated during cutting are compared with the
numerical results. For the temperature interaction it is assumed that plastic defor-
mation is dissipated with a Taylor-Quinney coefficient β = 0.4 for aluminum and
β = 0.9 for steel. The parameters of the interaction are chosen as
Erod = 1.8·1011 N/m2, εmax = 0.16, ε0.2 = 0.003, k = 0.01·Erod, and r = 0.04 mm. The
experimental and simulated cutting forces are compared in Figure 14.17 for dif-
ferent depths of cut h ranging from h = 0.2 mm to h = 0.4 mm.
0 250
0.4
0.3 200
temperature [K]
−100
0.2 0.43
force [N]
150
−200 0.36
100 0.23
−300 0.17
50
−400 0
0 1 2 3 0 1 2 3 4
displacement [mm] time [ms]
Fig. 14.17 Cutting force for Aluminum Fig. 14.18 Temperature increase for
AlMgSi1.0 h = 0.2 mm
The experimental results are shown as constant levels with dashed lines and
with the color-coding indicating the depth of cut. It can be seen that for
h = 0.2 mm there is good agreement in the steady state regime. For larger depths of
cut, agreement is slightly worse but still good. The temperature increases generat-
ed during cutting are shown in Figure 14.18 for different vertical positions meas-
ured from the top of the workpiece indicated in the legend in [mm] for
h = 0.2 mm. In the chip at a depth of 0.17 mm, a rise of 200 K can be reported.
This fits quite well to the results reported by [20]. For Ck45, the forces of the
steady state regime are shown in Figure 14.19. The simulated cutting force agrees
well with the experimental results, which are not available for depths of cut great-
er than h = 0.3 mm. The evolution of the simulated passive force, which decreases
and increases again with increasing depth of cut for 0.1 mm < h < 0.3 mm, is due
to discretization effects. The decrease for h ≥ 0.3 mm is due to the chip pressing
against the rake face of the tool. The temperature increase in the workpiece for
h = 0.5 mm is shown in Figure 14.20 with the vertical position of measurement in-
dicated in the legend in [mm], assuming a Taylor-Quinney coefficient of β = 0.9.
The measured temperature increase for a depth of cut of 0.3 mm is ΔT = 340 K.
This is significantly lower than the simulated temperature increase in the chip,
which is shown for the vertical positions 0.04 mm and 0.43 mm and is found to be
around ΔT = 450 K. An explanation for this behavior is a too high value for β and
too much compressive plastic deformation in the chip. For vertical positions dee-
per within the workpiece, far better agreement can be seen, e. g. at 0.7 mm and
1.0 mm. The temperature curve for 0.56 mm illustrates how the temperature peaks,
when the tool passes, but the heat quickly distributes in the rest of the workpiece.
See [21, [22], [23]for further results.
cut sim
800 passive sim
cut exp
600 passive exp
force [N]
400
200
0
0.1 0.2 0.3 0.4 0.5
depth of cut [mm]
Fig. 14.19 Cutting force and passive force for Ck45

700
0.04 sim.
600 0.43 sim.
500 0.56 sim.
temperature [K]
0.69 sim.
400
1.0 sim
300 0.30 exp.
200 0.70 exp.
1.0 exp.
100
0
0 2 4 6 8
time [ms]
Fig. 14.20 Temperature increase for Ck45
An illustration of the cut off chip is shown in Figure 14.21. The chip is cohe-
sive and clearly separated from the rest of the workpiece.
Fig. 14.21 Illustration of cut-off chip
14.3.2.7 Orthogonal Cutting with Tool Dynamics
To expand the applicability of the so-far discussed approach some constraints of

the tool are removed and additional dynamics are introduced into the system. The
mounting of the tool is changed giving it a rotational degree of freedom, as illu-
strated in Figure 14.22. This allows for a vertical movement of the tool tip and a
rotational movement of the overall tool. The tool is coupled with a rotational
spring-damper combination. The dynamics of the tool is integrated into a separate
simulator, a program to model multi-body systems. Both particle and multi-body
simulator are coupled in a co-simulation approach, where state variables as well as
kinetic quantities are exchanged at fixed time steps. See [18] for a more detailed
explanation.
z Ct
mt I t we
x
be
Fig. 14.22 Model setup for orthogonal DEM cutting simulation
0 150
mm]
mm]
−50 100
−2
−2
force [N] / Δ z [10
force [N] / Δ z [10
−100
50
−150 passive force

0
Δz
tool
−200 cut force
Δz −50
tool
−250 0 1 2 3 4 5
0 1 2 3 4 5
displacement [mm] displacement [mm]
Fig. 14.23 Cutting force for elastic tool Fig. 14.24 Passive Force for elastic tool
support support
The model parameters are unchanged, except Erod = 5.9 1010 N/m2. The para-
meters determining where the tool is fixed are chosen as we = 0.02 m and
be = 0.04 m, i. e. both passive and cutting force bring about a torque on the tool.
The cutting force for a normalized workpiece width of 1 mm and the depth of cut
are both shown in Figure 14.23, indicated in the legend by ‘cut force’ for the force
and ‘Δztool’ for the change of vertical position of the tool tip. It can be seen that
starting from an initial depth of cut of 0.4 mm the cutting depth continuously in-
creases with continuous tool feed. A similar behaviour can be seen in Figure 14.24
for the passive force. A steep rise in passive force as the tool enters the workpiece
is followed by a continuous increase as the cutting depth increases. The tool tip
moves deeper and deeper into the workpiece due to the moment acting on the tool.
As the length of the workpiece is chosen as l = 4 mm both passive and cutting
force vanish for a tool displacement exceeding 4 mm.
14.4 Conclusion
This contribution illustrates that the proposed material model based on the Dis-
crete Element Method allows to capture essential characteristics of elasto-plastic
solids. The separation of material as shown for a Charpy impact test and in cutting
processes is included without effort. The comparison of simulated and experimen-
tally obtained cutting forces demonstrates that good agreement can be reached,
both on a qualitative as well as a quantitative level. An extension of the model
such that the thermodynamics of heat conduction can be included has been suc-
cessfully implemented. Comparing the temperatures generated during cutting
shows that, for an appropriate choice of the Taylor-Quinney coefficient, good
agreement between simulation and experiment can be reached. A further en-
hancement of the cutting process model, such that dynamics of the tool can be in-
cluded, is made based on a co-simulation approach. There, the tool dynamics is
accounted for in a separate multi-body simulator. The entirely different and physi-
cally sound behavior of the system once the constraints are removed shows the in-
creased applicability of the approach for modeling the interaction between tool,
workpiece and machine.
References
[1] Tsai, L.-W.: Robot Analysis – The Mechanics of Serial and Parallel Manipulators.
John Wiley & Sons, New York (1999)
[2] Merlet, J.-P.: Parallel Robots. Kluwer Academic Publishers, Dordrecht (2000)
[3] Preumont, A.: Mechatronics – Dynamics of Electromechanical and Piezoelectric Sys-
tems. Springer, Dordrecht (2006)
[4] Heisel, U., Storchak, M., Stehle, T.: Einfluss der Umgebungstemperatur beim ortho-
gonalen Zerspanen. wt Werkstattstechnik Online 100, 89–98 (2010) (in German)
[5] Frohmüller, R., Knoche, H.-J., Lierath, F.: Aufbau und Erprobung von Temperatur-
messeinrichtungen durch das IFQ im Rahmen des Schwerpunktprogramms Spanen
metallischer Werkstoffe mit hoher Geschwindigkeit. Spanen metallischer Werkstoffe
mit hohen Geschwindigkeiten Kolloquium des Schwerpunktprogramms der DFG,
108–115 (1999) (in German)
[6] Körtvelyessy, L.V.: Thermoelement Praxis. Vulkan Verlag, Essen (1981) (in Ger-
man)
[7] Müller, B.: Thermische Analyse des Zerspanens metallischer Werkstoffe bei hohen
Schnittgeschwindigkeiten. Dissertation, RWTH Aachen (2004) (in German)
[8] Heisel, U., Storchak, M., Stehle, T., Korotkih, M.: Temperaturbestimmung in den
Zerspanzonen. wt Werkstattstechnik Online 100, 365–370 (2010) (in German)
[9] Heisel, U., Krivoruchko, D.V., Zaloha, V.A., Storchak, M.: Cause Analysis of Errors
in FE Prediction of Orthogonal Cutting Performances. In: 10th CIRP International
Workshop on Modeling of Machining Operations, pp. 141–148 (2007)
[10] Heisel, U., Kryvoruchko, D.V., Zaloha, V.A., Storchak, M., Emelyanenko, S., Seli-
vonenko, S.N.: Finite Element Analysis of Cutting Force Dynamics. In: Proceedings
of the 11th CIRP Conference on Modeling of Machining Operations, September 16-
18, pp. 163–170 (2008)
[11] Fleissner, F.: Parallel Object Oriented Simulation with Lagrangian Particle Methods.
Schriften aus dem Institut für Technische und Numerische Mechanik der Universität
Stuttgart, Bd. 16. Shaker Verlag, Aachen (2010)
[12] Geradin, M., Rixen, D.: Mechanical Vibrations. John Wiley & Sons, Chichester
(1998)
[13] Lankarani, H.M., Nikravesh, P.E.: A Contact Force Model with Hysteresis Damping
for Impact Analysis of Multibody Systems. Journal of Mechanical Design 112, 369–
376 (1990)
[14] Liu, K., Gao, L., Tanimura, S.: Application of Discrete Element Method in Impact
Problems. JSME International Journal, Series A 47, 138–145 (2004)
[15] Fleissner, F., Gaugele, T., Eberhard, P.: Application of the Discrete Element Method
in Mechanical Engineering. Multibody System Dynamics 18, 81–94 (2007)
[16] Gaugele, T., Fleissner, F., Eberhard, P.: Simulation of Material Tests using Meshfree
Langrangian Particle Methods. Proceedings of the Institution of Mechanical Engi-
neers, Part K: Journal of Multi-body Dynamics 222(K4), 327–338 (2008)
[17] DIN10045, Kerbschlagbiegeversuch nach Charpy. Deutsches Institut für Normung
e.V., Berlin (1991)
[18] Gaugele, T., Eberhard, P., Storchak, M., Heisel, U.: A Discrete Material Model used
in a Co-simulated Charpy Impact Test and for Heat Transfer. In: Proceedings of the 1.
International Conference on Process Machine Interactions, pp. 361–369 (2008)
[19] Tychonoff, A., Samarski, A.: Differentialgleichungen der mathematischen Physik.
Deutscher Verlag der Wissenschaften, Berlin (1959) (in German)
[20] Jaspers, S., Dautzenberg, J.: Material Behavior in Metal Cutting: Strains, Strain
Rates, and Temperatures in Metal Cutting. Journal of Materials Processing Technolo-
gy 121, 123–135 (2002)
[21] Gaugele, T.: Application of the Discrete Element Method to Model Ductile, Heat
Conductive Materials. Dissertation, University of Stuttgart (2011) (submitted)
[22] Eberhard, P., Gaugele, T.: Quasi-static and dynamic properties of separable continua
based on Lagrangian Particle Methods. In: Proceedings of the International Confe-
rence on Particle-Based Methods, Fundamentals and Applications (Particles 2009),
pp. 398–401 (2009)
[23] Gaugele, T., Eberhard, P., Fleissner, F.: Particle Methods Used to Model Cutting
Processes Including Heat Conduction. In: Proceedings of the International Confe-
rence on Particle-Based Methods, Fundamentals and Applications (Particles 2009),
pp. 142–145 (2009)
Chapter 15
Surface Generation Process with Consideration
of the Balancing State in Diamond Machining
C. Brandt, A. Krause, J. Niebsch, J. Vehmeyer, E. Brinksmeier,

P. Maass, and R. Ramlau
Abstract. In order to manufacture optical components or mechanical parts with

high requirements regarding surface quality, diamond machining is frequently ap-
plied. Nevertheless, to achieve the desired surface quality, the understanding of the
surface generation process and its influencing parameters is highly important. One
crucial parameter is the residual unbalance of the main spindle. As the residual un-
balance affects the process and vice versa, the investigation of the process-machine
interaction is necessary. In this paper results of experimental work and mathematical
modelling of diamond machining under varying balancing states are presented. The
experiments show the connection between unbalances and resulting surface quality;
the mathematical model provides the possibility to simulate the surface quality for
given unbalances distributions. Furthermore, regularization techniques in order to
solve the inverse problem of computing the optimal balancing state for a given or
desired surface quality are presented.
15.1 Introduction
Ultraprecision diamond machining is mainly used to manufacture optical parts or
microstructures with form deviations of less then one micron and surface rough-
ness in the range of a few nanometers. To obtain these accuracies the requirements
regarding machine tool accuracy, process control and environmental conditions are
exceptional.
But even if all these requirements are met, unintended vibrations either induced
by the environment or by the process itself can result in insufficient part quality. Usu-
ally external vibrations can be kept low by appropriate foundations of the machine
tool. In contrast, process induced vibrations can occur because of misaligned tools
or workpieces, unsymmetrical workpieces, unsymmetrical workpiece mounting or
inhomogeneous workpiece materials. All these effects lead to non-homogeneous
mass distribution for the rotational axis of the system and therefore to vibrations
when under rotation. If existent these process induced vibrations can lead to an
330 C. Brandt et al.
increased form deviation and an increased surface roughness depending on the ex-
citing frequency. Therefore, precision balancing is crucial for ultraprecision ma-
chining processes [7]. In this chapter only ultraprecision turning will be considered,
nevertheless basic balancing principles will apply for other processes using rotating
spindles as well. The balancing process in ultraprecision machining is done man-
ually and it generally takes more than one iteration step to get the best balancing
condition.
The goal of the project is to get a deeper understanding of the interaction be-
tween balancing and the surface generation process in ultraprecision machining.
Therefore, a process-machine model has been developed which allows to predict the
surface topography based on the balancing state considering the interaction between
the machine tool structure and the machining process. In a first step, we developed
an interaction model for the experimental platform, see Section 15.3. A structure
model was developed to determine the vibrations for given force and moment dis-
tributions induced by unbalances as well as forces from the cutting process. This
model can also be utilized to determine unbalance distributions or balancing weights
from vibration data measured by sensors during idle spindle speed. Additionally, an
analytical process model for describing the cutting forces during the cutting pro-
cess has been developed. The cutting forces were related to the surface structure of
the workpiece. Since the cutting forces influence the vibrations of the machine and
vice versa, both sub-models have to be merged into a mechanical-dynamical model.
The complete model will enable us to determine the effective cutting forces and
displacements related to the workpiece surface. This model also allows to compute
the surface quality and in the future a necessary balancing state to achieve a needed
surface quality.
15.2 Balancing in Ultraprecision Diamond Machining

Balancing procedures are commonly used wherever rotating parts or spindles (re-
ferred to as rotors) are not allowed to exceed a certain eccentricity. To specify a
balanced or unbalanced rigid rotor respectively, the balance quality grade G is used.
The balance quality grade represents the tolerable track speed (mm/s) of the center
of gravity: G = ezul ω where ezul is the tolerable eccentricity and ω the angular ve-
locity. For example G 6.3 stands for a tolerable track speed of 6.3 mm/s. Within the
DIN ISO 1940-1 quality grades from G 6.3 down to G 0.4 are considered for con-
ventional machining [1]. These values are summarized in a diagram (Fig. 15.1). The
diagram allows for instance to obtain a tolerable specific unbalance for a needed
G value and the operating speed n. Additionally, these G values are standardized
for characteristic machine tool constructions where the mass of the rotor is within
a defined range of the total mass [13]. To show the difference between balancing
in conventional machining and ultraprecision machining the diagram has been ex-
tended by the G value range which was set as a target for the project described in
this article. The named target range for ultraprecision machining is G 0.04 or less
(depicted area, Fig. 15.1).
15 Balancing in Diamond Machining 331
Fig. 15.1 Balance quality

grades
15.2.1 Single-Plane Balancing

For ultraprecision machine tools rigid rotor behavior can be assumed for the main
spindle, which means the rotor will not or only insignificantly change its form or
unbalance state while running with operational speed or slower. In principle every
unbalance state for a rigid rotor can be compensated with two balancing planes
[13, 18]. In general these unbalances can be classified into three different types:
static unbalance, moment unbalance and dynamic unbalances.
A static unbalance occurs when a single unbalance is set to the radial plane of
a fully balanced rotor (Fig. 15.2). Hence, the center of gravity will be shifted away
from the shaft axis. When the spindle starts to rotate the unbalance will cause a cen-
trifugal force and therefore a vibration of the system. This kind of unbalance can
be compensated by either removing material from the spindle in the direction of the
unbalance or by adding a counterweight in the opposite direction of the unbalance.
The unbalance can be detected without continuous rotation of the spindle. The com-
pensation procedure is called single-plane-balancing (static balancing), because the
compensation is done in a single plane only [13]. Strictly speaking the above exam-
ple is only valid for a two dimensional system. But, although all real systems are
three dimensional, some may be treated as two dimensional [4].
A moment unbalance is caused by two identical unbalances, regarding their ab-
solute value, which are located opposite to each other in different radial planes
(Fig. 15.3). This unbalance cannot be detected without rotating the spindle. Under
rotation the two occurring unbalances will cause opposing centrifugal forces in dif-
ferent planes and therefore a moment to the spindle will occur [13]. The described
Fig. 15.2 Static unbalance

of a planar rotor [5]
Fig. 15.3 Moment unbal-

ance [5]
effect will be referred to as moment unbalance. As well as a static unbalance, a mo-

ment unbalance is mainly theoretical or will be used to describe a rotor or spindle
where the main influence is given by the moment unbalance. The depicted two kinds
of unbalances are necessary to describe and understand unbalances which occur in
real spindle systems. These unbalances are called dynamic unbalance and can be
described as a combination between a static unbalance and a moment unbalance.
Fig. 15.4 shows an example of dynamic unbalance. The center of gravity for the
workpiece and the counter weight could not be aligned within the balancing plane.
Therefore, the overlapping workpiece lead to a moment unbalance for the rotating
spindle.
Fig. 15.4 Dynamic un-

balance with single-plane
balancing (static balancing)
[4]
15.2.2 Dual-Plane Balancing

The described process of single-plane balancing is only capable of compensating
static unbalances. Furthermore, for most balancing setups a moment will be induced
because of the axial distance between balancing plane and unbalance. The general
balancing condition for most workpieces is called dynamic unbalance. It can be
regarded as a combination between static unbalance and a moment unbalance.
This dynamic unbalance can be compensated with a second balancing plane
within which a compensatory moment can be applied, see Fig. 15.5. With two bal-
ancing planes it is possible to compensate for all unbalance situations that may
occur, as long as the spindle can be considered as rigid [13]. Nevertheless, a second
balancing plane is increasing the complexity of the system and therefore, the effort
to balance a system will be increased too.
Fig. 15.5 Dual-plane-

balancing principle [4]
15.2.3 Ultraprecision Machine Tools

For most ultraprecision machine tools balancing is a manual, iterative and time con-
suming process with the goal to minimize the occurring vibrations. This process has
to be carried out whenever the workpiece is changed, because a small difference
in position can lead to a significant unbalance. Depending on workpiece size and
mounting situation (centric or eccentric), a rough balancing is done by applying a
counter weight. The fine balancing is done by applying set screws to the chuck.
15.2.4 Balancing Dependent Surface Generation

Because of the high requirements for the machined surfaces in ultraprecision ma-
chining it is necessary to understand to which extent the surface generation process
is influenced by the different process parameters. For this chapter the surface gen-
eration will be observed and investigated with regard to the balancing state of the
main spindle. As mentioned before balancing is one crucial factor for ultraprecision
machining but up to now it has not been investigated to which magnitude an occur-
ring unbalance influences the surface topography or the surface generation process
respectively.
To characterize the surface topography of the generated surfaces different mea-
surement devices are used. The form is measured by using an optical flatness tester,
where it is possible to assess the whole surface of the chosen samples. Roughness
measurements are taken by white light interferometry and by tactile profilometry (s.
Chap 1, Sect. 4 Workpiece analysis).
15.2.5 Process Forces with Regard to Unbalances

During machining an unbalance will lead to a deflected workpiece and therefore
change the depth of cut and consequently the process forces periodically. As the
process forces also act on the tool, the tool will be deflected as well. It can be
estimated, that the cutting force will deflect the tool opposite to the cutting direction,
but the tool will also be moved away from the surface and therefore reduce the
effective depth of cut. In consequence, the cutting force will be reduced and the tool
will move back in cutting direction and towards the surface as well. Depending on
the occurring forces, the damping properties of the tool and the unbalance, the tool
will either start to vibrate itself or it will reach a state of equilibrium.
Due to the interaction between the workpiece and the tool the surface of the
workpiece will be distorted. For an unbalanced and therefore vibrating workpiece
the engaged tool will act as a damper and reduce the induced vibrations. Due to this
process machine interaction it is necessary to take the process forces into account
with regard to the unbalance state and the resulting surface topography . But, be-
cause of size effects process models applied in conventional machining cannot be
applied. Previous investigations showed that in diamond turning the forces increase
for higher cutting velocities [11] whereas in conventional machining the forces de-
crease at higher cutting velocities because of heating and softening of the workpiece
material [14].
To assess the process forces during ultraprecision machining measurements have
been carried out using a triaxial dynamometer for low forces. The dynamometer is
mounted beneath the tool holder, on top of the test stands z-axis.
15.3 Experimental Setup

Commercially available ultraprecision machine tools are designed for manual single-
plane balancing and therefore, it is not possible to compensate for dynamic un-
balances. But, if a second balancing plane would be applied to an ultraprecision
machine tool a rise in workpiece quality relating to surface roughness and form
accuracy can be assumed.
To investigate the influence of dual-plane balancing in ultraprecision diamond
machining a test stand has been designed and built. Balancing experiments have
been carried out to investigate the influence between surface topography and corre-
sponding balancing situation. As the main goal of the investigations is to identify
and characterize balancing dependent effects during ultraprecision machining the
experimental procedure has been kept as simple as possible to reduce the influ-
ence of other effects. For this reason face turning was chosen as machining pro-
cess and plane discs (Fig. 15.6) were machined. As a result only one axis has to
be moved during machining. For all experiments an aluminum alloy (AlMg3) was
used a workpiece material which is commonly used for ultraprecision machining
processes.
Fig. 15.6 Machined planar

sample
Fig. 15.7 Balancing experiment platform (test stand), cross sectional view of the main spin-
dle (courtesy: Kugler) [5]
15.3.1 Test Stand

The test stand (Fig. 15.7, left) is based on an ultraprecision turning lathe, which
contains an automatic dual-plane balancing system. This system consists of two in-
dependent balancing actuators, so called balancers, both attached to the spindle. The
first balancer is located in front of the air bearing near the chuck, the second is lo-
cated behind the air bearing near the motor (Fig. 15.7, right). Attached to the spindle
housing are two vibration sensors located in same radial plane as the correspond-
ing balancers. Their signal is send to a control unit for analyzing and interpreting
the vibration data. The control unit calculates a balancing solution according to the
analyzed vibration data and sends a control impulse to each balancer. Because each
adjustment that is made to a balancer is simultaneously influencing both balancing
planes, the control unit is not able to calculate a single step solution, but has to ap-
proximate iteratively the best solution. This procedure demonstrates the complexity
of dual-plane balancing. For the functional principle of a single balancer and the
customizable machine tool properties, see [4, 5, 6].
15.4 Structure Process Interaction Model

In order to improve the balancing process with mathematical methods, a simulation
environment of the whole process has to be developed. The crucial point hereby
is the connection between unbalances and surface topography. Therefore, a model
which considers the interaction between machine structure and cutting process has
been built up. In the following subsections the separate parts of this model are pre-
sented, namely the machine model in 15.4.1, the process model in 15.4.2, the cou-
pling in 15.4.3 and the visualization of the surface topography in 15.4.4.
15.4.1 The Structure Model

The experimental platform from Fig. 15.7 has a complex structure which is difficult
to model in full detail. Thus, simplifying assumptions have been made that enable
us to handle the model mathematically but also make sure that the simplified model
is still a good approximation of the reality.
15.4.1.1 System Matrices
First, we have divided the platform into components which are illustrated in Fig.
15.8, namely the rotating part of the spindle, the spindle casing, the rotating part
of the motor, and the motor casing. The spindle rotor and casing are connected by
an air bearing with two spherical calottes. We have modelled this bearing by two
spring-damper elements although in a first attempt any damping in the springs was
neglected. The motor bearings are also modelled as spring-damper elements. Spin-
dle and motor are connected by a coupling that can compensate misalignments in
axial and radial directions as well as torsion. It is also modelled as a spring element.
Spindle and motor are supported by a granite base that is assumed to be rigid. The
joints to the granite base are modelled as firm spring elements. Additionally, Figure
15.8 shows the coordinate system. In operation, the spindle rotates counterclockwise
around the z-axis.
Secondly, we have developed a vibration model for each part of the machine sep-
arately. If we would consider only unbalances as possible reasons for vibrations it
would be sufficient to allow vibrations in radial directions x and y only. Since unbal-
ances cause harmonic vibrations and we assume isotropic bearings, the vibrations
in x and y direction are the same except for a phase shift of π /2. Nevertheless, the
forces from the cutting process act in all three directions. Thus each point along the
z-axis, i.e. each beam element of infinitesimal length ∂ z, has the following degrees
of freedom (DOF): the displacement u, v, w in x, y, z direction, the torsion angle βz ,
and the rotational angles βx , βy . We collect the DOF in a vector u = u(z,t). The
computation of u is based on an energy formulation, the so called Principle of Vir-
tual Displacements, see [8], Chapter 5. This principle is equivalent to equilibrium
conditions from which a partial differential equation for u(x,t) can be derived. Start-
ing from the energy formulation, we use the Finite Element Method (FEM) for the
discretization in the space variable x. We arrive at a system of ordinary differential
equations (ODE) in time of the form
Mü(t) + Su(t) = p(t), (15.1)

where M denotes the mass matrix, and S the stiffness matrix. In case of damping in
the system a third term Du̇(t) on the left hand side with a sparse damping matrix
D would be added. For the discretization, the considered parts of the platform are
divided into elements with nodes at each end. The movement of each point between
the nodes is described by ansatz functions scaled with the movement of the nodes.
Considering the boundary and transition conditions between the end node of one
element and the first node of the next element we will get system matrices M and
S for each part of the platform. This procedure is well known and we have mainly
followed [8] in order to derive the matrices. We have used the following partition:
1. Spindle rotor with the coupling: 36 elements.
2. Spindle casing: 30 elements.
3. Rotor part of the motor: 3 elements.
4. Motor casing: 2 elements.
Fig. 15.8 Modeled parts of

the test stand
As introduced above, each node has 6 degrees of freedom (DOF).

The DOF of each of the parts specified above are collected in vectors
usp−rot , usp−cas , um−rot , um−cas . A discretization in Ni elements in our model
leads to Ni + 1 nodes and thus 6 · (Ni + 1) DOF in the model of the ith
part. The vectors of DOF are subject to equation (15.1) with mass matrices
Msp−rot , Msp−cas , Mm−rot , Mm−cas and stiffness matrices Ssp−rot , Ssp−cas , Sm−rot ,
Sm−cas . If we collect all DOF in one vector
u = (uTsp−rot , uTsp−cas , uTm−rot , uTm−cas )T , (15.2)
we get a block diagonal structure for the entire mass and stiffness matrix of the
dimension 450 × 450.
Additionally, we have to consider the bearing elements that connect the DOF
of the corresponding nodes in the connected parts via a stiffness parameter C and
damping parameters D. This results in additional diagonal and in off-diagonal ele-
ments of the entire stiffness matrix which are collected in sparse matrices Cair for
the air bearing, Cm for the motor, and Cc for the coupling of motor and spindle:
⎛ ⎞ ⎛ ⎞
Ssp−rot 0 0 0 Cair −Cair 0 0
⎜ 0 Ssp−cas 0 0 ⎟ ⎜ 0 ⎟
S=⎜ ⎟ + ⎜ −Cair Cair + Cc −Cc ⎟.
⎝ 0 0 Sm−rot 0 ⎠ ⎝ 0 −Cc Cc + Cm −Cm ⎠
0 0 0 Sm−cas 0 0 −Cm Cm
(15.3)
The damping matrix D will have the same structure as the additional sparse ma-
trices in the stiffness matrix. Stiffness values for the coupling between motor and
spindle were provided by the manufacturer. For the air bearing between spindle
and spindle casing we had to rely on an inspection record that stated measurement
Fig. 15.9 Comparison of

model and measurement
data for given unbalance.
The two sensors are fixed
in y-direction on the casing
above the two balancers [2]
values for the axial and radial stiffness of the bearing in the steady state. We mention
again, that we have neglected the damping in a first attempt mainly because of the
lack of information. The identification of damping coefficients will be part of future
investigations.
15.4.1.2 Model Adjustment

Although we have carefully modelled the elements with respect to their geometry
and physical properties, we still made simplifications. Therefore, the model may
not fit reality immediately. The most uncertain values are the stiffness parameters
for the bearings, in particular those from the air bearing. A modal analysis of the ex-
perimental platform was carried out in order to adjust the model eigenfrequencies to
the eigenfrequencies of the platform but the analysis turned out to be faulty. So far,
the only reliable information we have of the real platform are the vibrational data
for certain defined unbalance settings in a frequency range of 5 · · · 50 Hz. For a good
model, the measured data has to fit the computed data reasonably well. With the
pre-chosen model parameters we were not able to achieve this data fit immediately.
Hence, we have changed the stiffnesses for all bearings and monitored its effect on
the vibrational response. It turned out that the stiffness of the air bearing changed the
vibrational behavior most notably, i.e. a reduction of its stiffness produced a desired
eigenfrequency at 15 Hz. The influence of other bearings on the lower eigenfre-
quencies was insignificant. A comparison of the data produced by the model with
the measurements can be seen in Fig. 15.9. Although this first attempt to optimize
the model according to the sparse information was successful, it is still questionable
if the model reflects the machine correctly in all necessary aspects. It is planned to
carry out further modal analyses in the near future. With these data available, a more
reliable model can be generated. Meanwhile, we have used this model for further
computations and tests, in particular the combination with the force model.
15.4.1.3 Solution of the Vibration Equation in the Presence of Unbalances

If only unbalances in our machine tool are considered, the right hand side p =
punb (t) of equation (15.1) has harmonic entries depending on the angular velocity ω .
In practice, the revolution speed n in rpm is given, therefore we have ω = 260π n. An
unbalance is modelled as a mass Δ m displaced from the shaft by a vector r = rei·φ
where φ is the angle to a given zero position. If the displaced mass rotates with an
angular velocity ω it induces a centrifugal force of an absolute value F :
F = ω 2 b, with b := Δ mr.
The projection of this force onto the x- and y- axis yields
Fx = ω 2 b sin(ω t + φ ) = ℑ(ω 2 beiφ eiω t ),

Fy = ω 2 b cos(ω t + φ ) = ℜ(ω 2 beiφ eiω t ),
where ℑ and ℜ denote the imaginary and real part of a complex number. Those
forces only apply to the displacement DOF in x-direction, and y-direction. All the
other DOF are not affected. Therefore, the sub-vector pk of p containing the entries
for the DOF of the k-th node, k =, · · · , N, has the form
⎛ ⎞
ℑ(ω 2 bk eiφk eiω t )
⎜ ℜ(ω 2 bk eiφk eiω t ) ⎟
⎜ ⎟
⎜ 0 ⎟
⎜
pk = ⎜ ⎟
0 ⎟
⎜ ⎟
⎝ 0 ⎠
0
We split punb (t) = (pk )k into part with sin and cos entries only in order to apply the
right hand side ansatz for each of the parts to solve (15.1):
punb = ℑ(q1 eiω t ) + ℜ(q2 eiω t ), with q1,2 = (q1,2

k )k (15.4)

q1k = ℑ(ω 2 bk eiφk eiω t ), 0, 0, 0, 0, 0

q2k = 0, ℜ(ω 2 bk eiφk eiω t ), 0, 0, 0, 0 .
j
Inserting the equation uunb (t) = u j eiω t , j = 1, 2 and its second derivative in (15.1)
yields
uunb = u1unb (t) + u2unb(t),
= ℑ((−ω 2 M + S)−1 q1 eiω t ) + ℜ((−ω 2M + S)−1 q2 eiω t ). (15.5)
The solution of (15.1) is the sum of the particular solution uunb and the general
solution of the homogeneous equation with right hand side zero. After a certain
time of rotation with a constant angular velocity and no other forces than those from
unbalances the homogeneous solution will die out due to small damping effects.
Hence, in this case we have the solution uunb .
15.4.2 The Process Model

The submodel for the machining process consists of two parts: a force model to
simulate the actual cutting force and a model for the actual process parameters as
well as the tool tip position on the workpiece surface. Actual parameter means that
the parameter is time dependent in contrast to the given constant input parameters at
the test stand. Figure 15.10 shows a schematic diagram of the considered diamond
face turning process. In face turning the tool is moving along the x-axis with a feed
velocity v f and cuts the workpiece with a depth of cut a p at its front face. The acting
force can be split in three components, the cutting force Fc in negative y-direction,
the thrust force Ft in z-direction and the feed force Ff in negative x-direction. In
the considered process, the cutting velocity vc (t) = 2π n(r − d(t)) is decreasing with
time since the rotational speed n is constant but the travelled distance d is increasing.
Therefore, we need a force model that includes the cutting velocity vc in addition to
the depth of cut a p and the feed rate f defined as the distance the diamond tool is
travelling during one revolution.
Fig. 15.10 Scheme of the

considered face turning
process
15.4.2.1 The Force Model

As there are a lot of standard force models for conventional cutting processes, the
development of force models for micro cutting is an actual research topic, because
several so called size effects occur like cutting edge effect, minimum chip thickness
and ploughing effects. See [15] for an overview about size effects in cutting opera-
tions. New force models have been developed, like the recently proposed slip-line
force model for micro turning with edge tool [9] including strain and temperature
effects. The critical chip thickness and micro ploughing effects are also examinated,
see for example [12].
In ultra-precision turning the situation is exceptional. The cutting parameters like
depth of cut and feed rate are in the range of some micrometers, which is possible
due to the use of diamond tools with much sharper cutting edges than conventional
(carbide) tools and the application of ultraprecise machine tools. Therefore, some
of the mentioned size effects do play only a subordinate role in diamond cutting
and the mentioned conventional models are not applicable to ultra-precision turning
experiments. Nevertheless, the forces in the experiments show a typical behavior
for micro machining and the used workpiece material (aluminium alloy AlMg3),
i.e. the thrust force is the dominant force component and the forces are in the range
of one Newton or less.
Another class of force models for micro turning are modified Kienzle models,
see [14]. In [10] an ansatz is presented to calculate an undeformed chip thickness h
which permits to consider tools with different tool nose geometries. Another modi-
fied Kienzle ansatz is proposed in [17] where the specific cutting force kc is repre-
sented as a product of functions including cutting velocity vc , friction μ , uncut chip
thickness h and cutting edge radius rβ , i.e.
kc = f1 (h) f2 (vc ) f3 (rβ ) f4 (μ ), (15.6)
β −β
with functions f1 (h) = ch−m and f2 = α1 vc 1 + α2 vc 2 . We picked up this idea and
the form of the functions f1 and f2 to model the specific cutting force as product
of functions gi depending on the depth of cut a p , the feed rate f and the cutting
velocity vc , i.e. to model the specific force in the form
kc = g1 (a p )g2 ( f )g3 (vc ) . (15.7)
Similar equations hold for the specific thrust and feed forces k p and k f . The func-
tions gi (i = 1, 2, 3) and the model constants therein are determined with help of
force measurements with different cutting conditions, see Table 15.1 for details.
Table 15.1 Experimental conditions for diamond turning experiments
Parameter value range

rotional spindle speed n = 800 . . . 1500 rev/min
feed rate f = 4 . . . 12 mm/min
depth of cut a p = 2 . . . 14 μ m
tool nose radius rε = 760 μ m
workpiece material AlMg3
The measurements of the specific forces over the depth of cut and feed rate are
illustrated in Fig. 15.11 forces together with the fitted curves g1 and g2 (solid line)
which had been determined to be of the form
−mia
gi1 (a p ) = cia a p (i = t, f , c) (15.8)
and
gi2 ( f ) = cif f −m f
i
(i = t, f , c) . (15.9)
The function g3 has been determined to be of form of the function f2 , i.e.
βi −β2i
gi3 (vc ) = α1i vc 1 + α2i vc (i = t, f , c) . (15.10)
Fig. 15.11 Measured specific forces over depth of cut (left hand side) and feed (right hand
side). The measured thrust force is represented by circles and the measured cutting force by
x-marks. The fitted curves g1 and g2 given by equation (15.8) and (15.9) are plotted as solid
line [3]
Fig. 15.12 Measured spe-

cific forces over cutting
velocity. The measured
thrust force is represented
by circles and the measured
cutting force by x-marks.
The fitted curve g3 given by
equation (15.10) is plotted
as solid line [3]
The function g3 and the measurements of the specific forces over cutting velocity are
shown in Fig. 15.12. Using the specific cutting forces we are now able to calculate
the specific cutting force components via the usual relationship for the forces by
Kienzle (see [14]):
Fc = kc Ac , Ft = kt Ac , F f = k f Ac , (15.11)
where Ac denotes the cross sectional area of cut which can be approximated by
Ac = a p f .
15.4.2.2 Simulation of the Tool Path

The second part of the process model consists of the simulation of the actual process
parameter and the tool path. Basically, the description of the tool tip position on the
surface is given by the movement of the tool, the deflections δi (i = x, y, z) of the tool
as well as the deflections Δi (i = x, y, z) and tilting βi (i = x, y, z) of the workpiece,
i.e.
x(t) = −r + v f t −δx (t) − Δx (t) , (15.12)

y(t) = −δy (t) − Δy (t) , (15.13)

z(t) = −a p + 2l1h δx2 + δy2 +δz (t) − Δz (t) − x(t) tan (βy (t)) , (15.14)
where lh denotes the length of the tool holder. The angle βy denotes the rotation of
the workpiece around the y-axis. This tilt of the workpiece can influence the actual
depth of the cut a p which is described in equation (15.14). Since the deflections
affect the actual tool tip position and we now define time dependent process param-
eters, namely the actual depth of cut
a p (t) = −z(t) (15.15)
and
vx (t) = d(t)
˙ , with d(t) = x(t) + r . (15.16)
The actual feed rate f is given by f (t) = n−1 vx (t). In contrast to the given constant
input parameters of the test stand, the time dependant parameters are called “actual
parameters”.
These actual process parameters are plugged into the equation (15.11) for the
force components using the equations for the specific forces (15.7). Using the actual
forces, we are now able to determine the deflections of the tool holder which are
proportional to the forces
Ff (t) Fc (t) Ft (t)

δx (t) = , δy (t) = , δz (t) = . (15.17)
kex key kez
Here kei denotes the corresponding stiffness in the direction i (i = x, y, z). Since
all force components have the same structure we get for all three spatial directions
( j = x, y, z) the deflection

cia a p (t)−ma c̃if f (t)−m f
i
−1
α1i vc (t)β1 + α2i vc (t)−β2 a p (t) f (t)
i i i
δ j (t) = kei
(15.18)
with (i = f , c,t) for ( j = x, y, z). Derivatives of the equations for the position
(15.12)-(15.14) and deflections (15.18) combined with equation (15.11),(15.15) and
(15.16) for the actual parameters deliver a system of ordinary differential equations
of the form
δ̇x (t) = v f − vx (t) − Δ̇x (t) , (15.19)

1
δ̇y (t) = k̇c (t)a p (t)vx (t) + kc (t) [ȧ p (t)vx (t) + a p(t)v̇x (t)] , (15.20)
nkey
1
δ̇z (t) = k̇t (t)a p (t)vx (t) + kt (t) [ȧ p (t)vx (t) + a p(t)v̇x (t)] , (15.21)
nkez

δ̇x (t) − n−1kex −1 v (t) k̇ (t)a (t) − k (t) ȧ (t)
x f p f p
v̇x (t) = −1
, (15.22)
n−1 kex k f (t)a p (t)
1
ȧ p (t) = − δx (t)δ̇x (t) + δy (t)δ̇y (t) − δ̇z (t) + Δ̇x (t) ,
lh
β̇y (t)
+ (r − d(t)) − vx (t) tan (βy (t)) , (15.23)
cos2 (βy (t))
˙ = vx (t) ,
d(t) (15.24)

f f f f

−m
k̇ f (t) = −caf c̃ f a p (t)−ma vx (t) f α1 β1 vc (t)β1 −1 − α2 β2 vc (t)−β2 −1
f f f f f

f f f f
−ma −1 −m f f −m f −1 −ma
+ ma a p (t)
f
ȧ p (t)vx (t) + m f vx (t) v̇x (t)a p (t)
f f

α1f vc (t)β1 + α2f vc (t)−β2 (15.25)

k̇c (t) = −cca c̃cf a p (t)−ma vx (t)−m f α1c β1c vc (t)β1 −1 − α2c β2c vc (t)−β2 −1
c c c c

+ mca a p (t)−ma −1 ȧ p (t)vx (t)−m f + mcf vx (t)−m f −1 v̇x (t)a p (t)−ma
c c c c

α1c vc (t)β1 + α2c vc (t)−β2 ,
c c
(15.26)

k̇t (t) = −cta c̃tf a p (t)−ma vx (t)−m f α1t β1t vc (t)β1 −1 − α2t β2t vc (t)−β2 −1
t t t t

mta a p (t)−ma −1 ȧ p (t)vx (t)−m f + mtf vx (t)−m f −1 v̇x (t)a p (t)−ma
t t t t

α1t vc (t)β1 + α2t vc (t)−β2 .
t t
(15.27)
i
Hereby, we defined c̃if = cif nm f , because of the relation vx (t) = n f (t) and equation
(15.9). The cutting velocity is calculated by vc (t) = 2π n (r − d(t)) and the vibrations
Δi , i ∈ {x, y, z}, as well as the tilt angle βy of the workpiece and their derivatives are
determined by the structural submodel. The system of differential equations will
be solved numerically with the MATLAB solver ”ode15i”. The stiffness values kei
(i = x, y, z) are determined by the geometrical dimensions of the tool holder and its
elasticity module (material: steel, E = 210 kN/mm2). The resulting actual forces Fi
are calculated by the deflections δ using equation (15.17). The numerical results are
presented in Sect. 15.4.5.
15.4.3 Coupling of the Submodels

In the last two sections both submodels have been presented, the structure model and
the process model. Solving the ODE (15.1) lead to the vibrations u of all elements
of the discretization. The first six entries are related to the workpiece and are input
parameters for the system of differential equations (15.19)-(15.27), i.e.
(Δx , Δy , Δz ) = (u1 , u2 , u3 ) and βy = u5 . (15.28)
The process model computes the actual forces and deflections of the tool which act
at the tool tip position on the workpiece. Therefore, these forces add up to the forces
of the unbalances in the right hand side of equation (15.1). The resulting additional
load vector has the form
pcut (t, u(t)) = (Ff , Fc , Ft , 0, Mc , Mt , 0, · · · , 0)T , with M{t,c} = F{t,c} × ra ,
where ra (t) = r − d(t) denotes the radius of the workpiece minus the already trav-
elled distance of the tool on the workpiece, cf. Fig. 15.10. Thus, a coupled system
of ODEs
Mü(t) + Su(t) = punb + pcut (t, F(t, w), Mt (t, w), Mc (t, w)) ,
ẇ(t) = g (t, w(t), ẇ(t), u(t)) ,
has to be solved, where the function g is given by the system of differential equa-
tions (15.19)-(15.27) and the corresponding variables are collected in the vector
w = (δx , δy , δz , vy , a p , d, kt , k f , kc ). In order to solve the non-linear coupled system
approximately, we have employed a time step algorithm and assumed the forces and
moments from the cutting process to be constant during a small time interval, i.e.
pcut (t) = pcut (ti ) for t ∈ [ti ,ti + Δ t]. Now,
u(t) = A(punb + pcut (ti )), t ∈ [ti ,ti + Δ t] ,
where A describes the solution operator of (15.1). The resulting deflections

(Δx , Δy , Δz ) from (15.28) are plugged into the force model and we compute
(Ff (ti+1 ), Fc (ti+1 ), Ft (ti+1 )) = B(u(ti + Δ t)) = B(δx , δy , δz )
at ti+1 = ti + Δ t. Here B denotes the solution operator for solving (15.19)-(15.27)

and use (15.17) afterwards. Again, we assume the cutting forces to be constant over
the next time interval [ti+1 ,ti+1 + Δ t]. This routine is repeated until the end of the
desired time interval t ∈ [t0 ,tend ] is reached, for which the coupled system should be
solved.
15.4.4 Surface Visualization

The visualization of a three-dimensional representation of a surface is an intuitive
but powerful and flexible technique in surface characterization and comparison. Sur-
faces produced in face turning can be represented as a continuous function S(x, y)
describing the surface height over the (x, y)-plane. For visualization and other digi-
tal processing, (x, y) is an element of a discrete support set consisting of regular or
irregular arranged points. For objective characterization, three dimensional surface
parameters can be derived from the surface function S, see [16] for more details.
We make investigations to study the surface at two scales. We are interested in the
form deviation of the global workpiece but also in the roughness structure. During
the development it turns out to be more efficient to decompose large surfaces into
two scales and to simulate both scales separately. This is done not only for the
save of computation time but also for avoiding alias effects which may occur when
Fig. 15.13 Radial sur-

face sections under ideal
conditions and for a simula-
tion, both with a p = 1mm,
f = 0.5mm, rε = 0.76mm
large surfaces at fine discretization are sampled down. Consequently, we calculate

S on a small rectangular sub domain at fine discretization to visualize the roughness
profile and to determine the local surface parameters. Therefore, a kinematic surface
simulation is developed in the following subsections. The global form is provided
by S, evaluated on selected tool tip positions.
15.4.4.1 Simulation of Surface Generation

The ideal kinematic surface resulting from face turning can be described by a set of
radial sections. Each radial section Sk is built by repetitions of the edge geometry in
intervals of feed per revolution. If a round nosed tool of radius rε is used, we can
formulate each radial section by
2
x+kf
Sk (x) = rε − rε − x + (k − 1/2) f −
2 f − a p, x ∈ [0, r], k ∈ [0, 1),
f
where k2π denotes the section’s angle. An ideal radial section S0 is shown in Fig.
15.13. The surface profile will change in two ways with respect to vibrations of the
workpiece and deflections of the tool. Firstly, the repetitions will not be necessarily
equidistant and the turning grooves will differ in their depth. Secondly, the geome-
try of each channel will change because tool and workpiece will not be positioned
orthogonal to each other. Figure 15.13 shows also a simulation result for a disturbed
process.
In the following, a discrete surface generation model, which is coupled with the
process model, is presented to provide the surface function S depending on vibra-
tions of the workpiece and tool deflection. The central component of the model is
the parameter to state operator
Φ :P→ Ê4×4, (15.29)

Fig. 15.14 Coordinate sys-

tem for dynamic surface
simulation; rotational mo-
tion of workpiece and tool
which maps kinematic information, provided by the structure model, to homoge-

neous matrices used to formulate the tool trajectories.
15.4.4.2 Calculation of the Tool Trajectories

Basically, the interacting objects, workpiece and tool, are assumed to be solid such
that they are configured in the coordinate system for any time point t by a rotation
and a translation. Thus, the tool and workpiece trajectories are given by affine linear
mappings ΦT and ΦW . In computer graphics homogeneous matrices are used to
perform such transformations. Therefore, we introduce a homogeneous coordinate
system, which is built from cartesian coordinates by
(x, y, z) → (x, y, z, 1). (15.30)
The characteristic motions in turning processes are the rotation of the workpiece
around the z-axis, which is expressed by
⎛ ⎞
cos(φ ) − sin(φ ) 0 0
⎜ sin(φ ) cos(φ ) 0 0 ⎟
ΦW = RZ (φ ) = ⎜⎝ 0
⎟,
0 1 0⎠
0 0 01
and the translation of the tool along the x-axis, which is expressed by
⎛ ⎞
1 0 0 r − f Nt
⎜0 1 0 0 ⎟
ΦT = T ((r − f Nt, 0, lh − a p)) = ⎜ ⎟
⎝ 0 0 1 lh − a p ⎠ .
000 1
These two operators describe the movement of the tool and of the workpiece under
ideal conditions. The positioning and alignment errors are included into the model
by applying further translations and rotations. Figure 15.14 shows the tilt angles βx
and βy of the workpiece. The moment acting around the x-axis has only a marginal
effect to surface generation, and therefore it is neglected in our model. The rotation
rate is assumed to be constant. This leads to the operator for the movement of the
workpiece:
ΦW = T (Δ )RY (βy )RZ (φ ). (15.31)
The tool inclination is changed appropriate to the deflection provided by the force
model by applying multiple rotations, which leads to
ΦT = T ((r − f Nt, 0, lh − a p))R(δx , δy ).
So far, the movement of the tool is described by ΦT and that one of the workpiece
by ΦW . For the material removal algorithm the relative position of the tool to the
workpiece is compulsory. ΦW and ΦT give absolute positions, but we can carry over
the workpiece operation to the tool using
Φ := ΦW−1 ΦT . (15.32)
The diamond tool, or more precisely its edge geometry, can be parameterized as
ε (ϕ ) = (rε cos ϕ , 0, −rε sin ϕ − l + rε , 1)T , ϕ ∈ [0, π ], (15.33)
with tool radius rε and length lh . Thus, the relative tool trajectories are given by
Φε (ϕ ) for ϕ ∈ [0, π ]. In case of ideal conditions, i.e. no vibration, no displacement
and no deflection, (15.32) can be written as
⎛ ⎞
cos(−φ ) − sin(−φ ) 0 (r − f Nt) cos(−φ )
⎜ sin(−φ ) cos(−φ ) 0 (r − f Nt) sin(−φ ) ⎟
Φ =⎜ ⎝
⎟.
⎠ (15.34)
0 0 1 lh − a p
0 0 0 1
In Fig. 15.15 the ideal and disturbed tool tip locus is visualized. For the ideal case the
trajectories of the tool tip are given by applying the tool tip ε (π/2) = (0, 0, −l, 1)T to
(15.34), leading to (x, y, z, 1)T = ((r − f Nt) cos(−φ ), (r − f Nt) sin(−φ ), −a p , 1)T ,
i.e. a spiral with decreasing radius appropriate to the feed rate located parallel to the
workpiece surface at height −a p .
15.4.4.3 Material Removal Process

The material removal takes place under ideal conditions, i.e. a homogeneous,
isotropic material and an ideal sharp tool in the sense of a cutting edge rβ = 0
are assumed. Consequently, the material passed by the cutting edge will be removed
completely, in particular no ploughing or elastic recovery is considered. Therefore,
the swept volume of the moving edge
Φε (ϕi ), 0 = ϕ0 < · · · < ϕn = π
is used to update the surface function S for points, which are passed by the cutting
edge.
Fig. 15.15 Relative tool tip locus for ideal conditions (left) and for an oscillating surface
(right); a p = 1mm, f =0.5mm
15.4.5 Numerical Simulation Results

Following Sections 15.4.4.2 and 15.4.4.3, a surface generation model for micro turn-
ing processes has been implemented. The model is able to simulate the global form
and roughness structure of the machined surface depending on the balancing state of
the machine. As mentioned in 15.4.2, the differential equations system for the pro-
cess model is solved numerically. The ODE (15.1) of the structure model is solved
numerically, too. A detailed description how to reformulate and solve the problem is
proposed in [2]. All algorithms are implemented in the mathematical programming
environment MATLAB. First, we tested the algorithm with different time steps Δ t.
The experiments showed that for higher frequencies the time resolution has be cho-
sen smaller. In particular, for the rotational speeds in our simulation we have to use
time steps equal or less than 1 ms.
We then tested the algorithm for several parameter settings. As expected, the
presence of unbalances mainly affects the deflection or vibration amplitudes in ra-
dial direction x and y. Nevertheless, the deflection in z direction is affected, too.
We can also observe quantitative effects for unbalance distributions of different
magnitude. Here, we will only present one example with the parameters defined
in Table 15.2, setting 1. We have used two different sets of unbalance distributions
f1 = ([22.4 gmm, 63◦ ], [4.5 gmm, 243◦], [4.7 gmm, 2◦ ]) and f2 = ([22.4 gmm, 63◦],
[0.45 gmm, 243◦], [0.47 gmm, 2◦ ]). The first position corresponds to the workpiece,
the second and the third to the balancer planes.
Figure 15.16 shows the development of the deflection of the workpiece in di-
rection of the spindle over time for both unbalance settings. The vibration in radial
direction as well as the development of the depth of cut are shown in Fig. 15.17.
Fig. 15.18 presents the thrust force Ft and the cutting force Fc . The higher unbal-
ance distribution f1 causes vibrations with bigger amplitudes.
Table 15.2 Parameter setting for the simulations
Parameter setting 1 setting 2 setting 3

rotational spindle speed n = 25 Hz n = 25 Hz n = 20 Hz
feed rate f = 5.33 μ m/rev f = 500 μ m/rev f = 8.33 μ m/rev
depth of cut ap = 5 μ m ap = 5 μ m ap = 5
tool nose radius rε = 760 μ m rε = 760 μ m rε = 760 μ m
workpiece radius r = 30 mm r = 5 mm r = 30 mm
tool holder length lh = 25 mm lh = 25 mm lh = 25 mm
time step Δ t = 1 ms Δ t = 0.1 ms Δ t = 1 ms
Fig. 15.16 Deflection of the workpiece in z-direction for f1 and f2 ; entire time interval (left),
and detail (right)
Fig. 15.17 Deflection of the workpiece in y-direction for f1 and f2 and depth of cut
The output of the process machine interaction model can also used to visualize
the machined surface. In a first step, surface simulations for simple oscillations of
type
βy (t) = ∑ ki sin(pi ω t), ω = n2π , (15.35)

i
Fig. 15.18 Simulated thrust and cutting force
are carried out in order to verify the basic function of the surface model. It is clear
that frequencies pi n in the signal βy can be recovered in the surface structure because
the workpiece is rotating with angular speed ω . In Fig. 15.19 two example surfaces
for different oscillations are shown.
Fig. 15.19 Simulated surfaces for oscillations of type (15.35)
In a second step, the robustness of surface generation is evaluated with the help
of the implemented surface model. This is an important step, because all input os-
cillations are result of numerical solved differential equations. Figure 15.20 shows
exemplarily a simulated surface for an oscillation βy of (50 + ε )Hz, where ε denotes
a small disturbance. When no disturbance exists (ε = 0) and a rotation frequency of
50Hz is assumed, the resulting surface is an inclined plane. With increasing degree
of disturbance the form error grows rapidly, which is demonstrated in Figure 15.20
for ε = 10−4 , 10−3 , 10−2. For ε = 10−4 a first distortion appears, which is growing
intensively for increasing disturbance. Finally, for ε = 10−2 the oscillation is heavily
asynchronous to the rotation leading to a completely wrong visual impression.
Model problems with reduced workpiece diameter and high feed speeds are con-
sidered for the benefit of short calculation time for all sub-models. The process
parameters of setting 2 in Table 15.2 are used in three model problems with dif-
ferent unbalance configurations, see Table 15.3. The related surfaces are plotted in
(a) ε = 10−4 (b) ε = 10−3 (c) ε = 10−2
Fig. 15.20 Simulated surfaces for an oscillation of (50 + ε )Hz; n = 50Hz, r = 30mm
Fig. 15.21 (a)–(c). In case (a), where no unbalances are set, a wavy surface can be
detected. This corresponds to the power spectrum of βy , where three different fre-
quencies can be identified. In case (b) and (c), where unbalances for the workpiece
or balancer planes are set, only the rotation frequency can be identified in the power
spectrum. The result is an inclined surface.
Table 15.3 Unbalance configuration for the model problems of setting 2
no. unbalance workpiece unbalance balancer planes 1 and 2

(a) no no
(b) no ([4.5 gmm, 243◦ ], [4.7 gmm, 2◦ ])
(c) [22.4 gmm, 63◦ ] ([4.5 gmm, 243◦ ], [4.7 gmm, 2◦ ])
(a) No Unbalances (b) Plane 1&2 set (c) Unbalance at workpiece

and plane 1&2 set
Fig. 15.21 Simulated surfaces for model problems
In a last step, surfaces under practical relevant conditions concerning workpiece

dimension and process parameters are computed. For a workpiece diameter of 60
mm and a feed rate of 5.33 μ m more than 5.600 overlapping channels generate the
resulting surface. To realize simulation results with reasonable expense, a two scale
model is compulsory. In analogy to the experiments, a set of simulations for different
rotation rates (n = 1200, . . ., 1800 [min−1 ]) and feed rates ( f = 5.33, . . ., 8.33 [μ m])
with two different balancing states are carried out. In the unbalanced case, only the
(a) Global form (b) Roughness profile
Fig. 15.22 Simulated surfaces for setting 3, Table 15.2, unbalanced (1.823g at the
workpiece)
(a) Global form (b) Roughness profile
Fig. 15.23 Simulated surfaces for setting 3, Table 15.2, best possible balancing state
rotation frequency can be identified in the oscillations βx , βy , βz . The displacements

and deflections are relatively small compared to the dominant oscillation βy and
the resulting surface is an inclined plane. The global form and a small rectangular
domain of 150 μ m × 50μ m are plotted in Figure 15.22. In the case of best possible
balancing state, βy indicates one frequency, which is close to the rotation frequency,
e.g. 48.96Hz for n = 50Hz. It can be assumed that this deviation is of numerical
nature. The simulated surface is highly defective, see Figure 15.23, and corresponds
to the scenario shown in Figure 15.20 (c).
15.5 Solving the Inverse Problem for Balancing

So far, we presented a process-machine interaction model which can compute the
surface topography of the workpiece with a given unbalance distribution and input
process parameters. Mathematically spoken, we have derived an operator A which
maps the vector p containing the unbalances distribution and the process parameters
to a surface S, i.e. we have the following operator equation
A(p) = S . (15.36)
This equation is called forward formulation of our model. Assuming now that for a
given surface S, we are interested do determine the necessary balancing state in or-
der to obtain the surface S, i.e. to solve equation (15.36) with respect to p. However,
measurement devices typically have limited precision and we assume that unprecise
and noisy measurement data Sδ are available, which fulfill ||S − Sδ || ≤ δ , where
δ denotes the measurement precision. Usually this operator is not continuously in-
vertible, which means that for given noisy data Sδ with a data error of the function
pδ = A−1 (Sδ ) might be an arbitrary bad approximation of the true unbalance dis-
tribution p. Problems with those properties are referred to as being ill-posed. In this
case, least square techniques, where pδ is computed as the minimizer of Ap−Sδ 2 ,
are unstable. The computation of pδ can be stabilized by using the regularization
methods presented in Chap. 3, i.e. by minimizing the so-called Tikhonov-functional
pαδ = min Ap − Sδ 2 + αΨ (p) (15.37)

p
instead. The penalty term Ψ (p) acts as a stabilizer and prevents large values of
Ψ (p).Typical choices of Ψ are, e.g.
1/p
Ψ (p) = p p := ∑ |pi | p
, 0 < p ≤ 2. (15.38)
i
In a first attempt we invert the structural submodel, i.e. given a vibration u we deter-
mine the unbalance distribution p causing the vibrations. This results are presented
in Chap. 3. Future work is dealing with the inversion of the full forward problem
(15.36).
15.6 Experimental Results

Face turning experiments were conducted to show to which extent an unbalance
affects the resulting surface topography. To further assist the investigation and un-
derstanding of the process-machine-interaction force measurements were carried
out during machining.
There are two main effects which result from an unbalance. First, the spindle on
which the unbalance is acting is deflected and performs a tumbling motion which
leads to a differing tool path. Second, the unbalance induces a vibration to the whole
machine tool structure where the magnitude of this vibration depends on the damp-
ing properties of the machine tool. These effects have different influences with re-
gard to the process, surface topography and process forces.
15.6.1 Form Deviation

The initial experiments showed a trend to higher form deviations (peak to valley,
s. Fig. 15.24) for unbalanced workpieces or workpieces which were machined with
an additional unbalance. In this context unbalanced workpiece means a workpiece
without any balancing procedures prior to machining. An additional unbalance is a
weight added to the workpiece by the use of setscrews to amplify balance induced
effects.
Fig. 15.24 Form deviations for balanced (left) and unbalanced (right) workpiece [11]
The influence of an unbalance in terms of form deviation can be seen in Fig.

15.25 and Fig. 15.26. The diagrams show the form deviation versus the depth of cut
and feed. Additionally, the cross sectional area of cut Ac is given for each experi-
ment series. In correspondence to previous experiments workpieces with unbalance
show a larger form deviation in general. Larger cross sectional areas of cut show
similar form deviations for workpieces with and without unbalances compared to
smaller cross sectional areas of cut. It is assumed that the engagement of the tool
is damping the unbalance induced vibration and the tumbling motion of the work-
piece respectively and therefore leads to similar values of form deviation compared
to machined workpieces without unbalance.
15.6.2 Surface Roughness

The initial experiments could not show a clear dependency between unbalance and
surface roughness as large deviations within the roughness data prevented any clear
correlation [5]. Additionally, the roughness values are still within the limit of 10 nm
Ra (i.e. optic quality) in the majority of the cases for machining with unbalance.
Fig. 15.25 Form deviation vs. depth of cut (with and without unbalance) [11]
Fig. 15.26 Form deviation vs. feed (with and without unbalance) [11]
From Fig. 15.27 and Fig. 15.28 it can be seen that the difference between ma-
chining with and without unbalance are only marginal. The roughness values are
slightly higher for machining with unbalance but for the observed cutting parame-
ters no strong influence can be observed.
Possibly the test stand is working at its limit in terms of accuracy as it has a lower
stiffness compared to current ultraprecision machine tools. Another possibility for
the low influence of unbalances is assumed in the process kinematic, as for turning
the surface normal and the direction of the force generated by the un-balance are
perpendicular to each other. For that reason the workpiece will not be deflected di-
rectly in direction of the tool which would cause a periodical change for the depth of
cut. Therefore, the current investigations will be extended by ultraprecision milling
experiments. For these experiments the normal of the generated surface and the di-
rection of the unbalance induced forces and with that the motion of the tool will be
the same.
Fig. 15.27 Surface roughness vs. depth of cut (with and without unbalance) [11]
Fig. 15.28 Surface roughness vs. feed (with and without unbalance)
15.6.3 Force Measurements

For the used workpiece material (AlMg3) the measured forces showed a character-
istic order in terms of their magnitude. The influencing effect of the unbalance is
always largest for the thrust force, followed by the cutting force and the feed force
with the lowest values.
Figure 15.29 shows the components of the resulting forces versus the depth of
cut. The dashed line represents the experiments where an unbalance was applied,
the continuous line represents the experiments in balanced state.
As expected, the forces rise with an increase for the depth of cut. From these
experiments it can be seen that there is a significant difference between balanced
and unbalanced workpieces only for the thrust force, whereas the cutting force and
the feed force are almost equivalent for balanced and unbalanced states. As the
thrust force is depending on the friction of the chip on the rake face, it is estimated
that the thrust force will have a specific value for a balanced workpiece. In addi-
tion, if the workpiece is unbalanced, it will be deflected and perform a tumbling
Fig. 15.29 Process forces Fi vs. depth of cut [11]
Fig. 15.30 Changed clear-

ance angle because of tum-
bling motion [11]
motion additionally to the rotation. The tumbling motion will tilt the workpiece pe-
riodically with respect to the tool. The tilting will not change the friction between
chip and rake face, but the angle between the workpiece surface and clearance face
will change periodically. For balanced machining this angle will be equivalent to
the clearance angle. For unbalanced machining if the workpiece is tilting towards
the clearance face, this angle will decrease and induce a larger frictional load on the
clearance face. With this additional frictional load the increased thrust force for un-
balanced workpieces can be explained. The cutting force would have been expected
to show a significant rise for unbalanced machining, because of the change in the
depth of cut. But at this stage this phenomenon cannot be explained.
A similar behavior for the force components can be recognized for an increasing
feed, again only the thrust force is influenced by the unbalance (see [11]).
15.7 Summary and Outlook

Up to now only single plane balancing has been used for ultraprecision machin-
ing processes although a secondary balancing plane seems promising with regard
to an optimized surface quality. Utilizing a test stand with dual-plane-balancing

capabilities machining experiments have been conducted to show the dependence
between balancing quality and surface generation.
Additionally, a modelling approach of the interaction between machining process
and machine tool structure has been developed. Two submodels (structure model
and process model) are combined in a nonlinear way, and the resulting interaction
model is solved numerically by a time step algorithm. The solution is used as a
basis to determine the surface topography and the surface quality by using a surface
simulation program which is still under development. The setup of the mathematical
model is supported by the experimental data.
The mathematical model of the dependency between unbalances and surface to-
pography will enable us to predict the surface quality of a workpiece for a given
balancing state of the machine as well as to compute the balancing state which is
necessary at least for a given surface quality. Additionally, the necessary balancing
weights can be determined efficiently from vibrational measurements at the casing
of the machine. This will reduce time to setup the machine for the cutting process
with a desired accuracy.
The experimental investigations showed an influence of the balancing quality
with respect to the form deviation of the machined workpieces. The form deviation
rose for machining with an additional unbalance. But for the same parameters the
form deviation decreased if the cross sectional area of cut was increased, which was
accounted for with a higher damping of the engaged tool. Although it was assumed
for unbalanced machining to have an impact on surface roughness, machining with
an additional unbalance showed only marginal differences compared to balanced
machining. Due to its adjustment possibilities the test stand is not as stiff as stan-
dard ultraprecison machine tools, we assume that the lower stiffness is responsi-
ble for difficulties in showing a clear dependency between unbalance and surface
roughness.
Surprisingly only the thrust force was considerably affected by the unbalances.
Cutting force and feed force did not show any clear effects for unbalanced machin-
ing. This behavior has not been anticipated, least for the cutting force. The lack of
influence with respect to the cutting force cannot be explained at this stage.
As the project is still in progress it is planned to investigate circumferential
milling as an additional ultraprecision machining process. For this process the di-
rection of the unbalance induced centrifugal force and the direction of the surface
normal are the same. Therefore, any motion of the tool in radial direction will di-
rectly affect the surface generation. Consequently the influence of unbalances on the
surface topography will be larger than for turning.
Acknowledgements. The presented research has been funded by the German Research
Foundation DFG within the Priority Program 1180.
References
1. DIN ISO 1940-1 mechanical vibration - balance quality requirements for rotors in a
constant (rigid) state - part 1: Specification and verification of balance tolerances
2. Brandt, C., Niebsch, J., Ramlau, R., Maass, P.: Modeling the influence of unbalances
for ultra-precision cutting processes. ZAMM - Journal of Applied Mathematics and Me-
chanics/Zeitschrift für Angewandte Mathematik und Mechanik 91(10), 795–808 (2011),
doi:10.1002/zamm.201000155
3. Brandt, C., Niebsch, J., Vehmeyer, J.: Modelling of ultra-precision turning process in
consideration of unbalances. In: 13th CIRP Conference on Modeling of Machining Op-
erations (2011)
4. Brinksmeier, E., Gläbe, R., Krause, A.: Precision balancing in ultraprecision diamond
machining. In: Laser Metrology and Machine Performance 8, Lamdamap 2007, vol. 8,
pp. 262–269 (2007)
5. Brinksmeier, E., Krause, A.: Dual plane balancing for diamond machining processes.
In: 3rd International Conference High Performance Cutting (HPC), Dublin, pp. 517–528
(2008)
6. Brinksmeier, E., Krause, A.: Surface generation in ultraprecision diamond machining
utilising dual-plane-balancing. In: International Conference on Process Machine Inter-
actions, Hannover, pp. 335–342 (2008)
7. Brinksmeier, E., Riemer, O.: Deterministic prodcution of complex optical elements. In-
ternational Journal of Production Engineering and Computers - Special Issue on CAPP
and Advances in Cutting Technology 4(5), 63–72 (2002)
8. Gasch, R., Knothe, K.: Strukturdynamik Bd. 2: Kontinua und ihre Diskretisierung.
Springer, Berlin (1989); IX, 336 S: graph. Darst
9. Jin, X., Altintas, Y.: Slip-line field model of micro-cutting process with round
tool edge effect. Journal of Materials Processing Technology (2010) (in press),
doi:10.1016/j.jmatprotec.2010.10.006
10. Köhler, J.: Berechnung der Zerspankräfte bei variierenden Spanungsquerschnittsformen.
Ph.D. thesis, Leibniz Universität Hannover (2010)
11. Krause, B.: Process forces in diamond machining with consideration of unbalances. In:
CIRP PMI, Vancouver (2010)
12. Malekian, M., Park, S., Jun, M.: Investigation of critical chip thickness and micro plough-
ing forces. In: Proceedings of the 2nd International Conference on Process Machine In-
teractions, Vancouver, Canada, June 10-11 (2010)
13. Schneider, H.: Auswuchttechnik. Springer (2007)
14. Tönshoff, H.K., Denkena, B.: Spanen: Grundlagen, 2., erw. und neu bearb. aufl edn.
Springer, Berlin (2004); XXIV, 417 S.: Ill., graph. Darst
15. Vollertsen, F., Biermann, D., Hansen, H., Jawahir, I., Kuzman, K.: Size effects in man-
ufacturing of metallic components. CIRP Annals - Manufacturing Technology 58(2),
566–587 (2009), doi:10.1016/j.cirp.2009.09.002
16. Dong, W.P., Blunt, L.: Three-dimensional surface topography, 2nd edn. Penton Press
(2000), http://www.sciencedirect.com/science/book/
9781857180268, XXII, 285 S
17. Weber, M., Autenrieth, H., Kotschenreuther, J., Gumbsch, P., Schulze, V., Lohe, D.,
Fleischer, J.: Influence of friction and process parameters on the specific cutting force and
surface characteristics in micro cutting. Machining Science and Technology 12, 474–497
(2008), doi:10.1080/10910340802518728
18. Zhou, S., Shi, J.: Active balancing and vibration control of rotating machinery: A survey.
The Shock and Vibration Digest 33, 361–371 (2001)
Chapter 16
Modeling and Simulation-Based Optimization
of a Turning Process
R. Britz, T. Maier, F. Schwarz, H. Ulbrich, and M.F. Zaeh
Abstract. Today, the productivity of machine tools is limited by the interactions

between machine and process. A method to predict these limits is presented here
using simulation model and an appropriate optimization algorithm. Therefore, a
short overview of the used theories in multi-body dynamics, cutting processes and
the mechanical model of the turning lathe is given. Furthermore, a modular cutting
force model as well as the coupling between process and structure is introduced.
For optimization, it is necessary to develop an objective function, where the quali-
ty and the productivity of the processes have to be represented. Finally, results of
the optimization process are shown.
16.1 Introduction
Interactions between machine structures and manufacturing processes, such as vi-
brations due to unsteady cutting, limit the process flow. With a detailed prediction
of these interactions it is possible to optimize operation charts, machine parame-
ters, path parameters and further details. The goal of the SPP1180 project "Para-
metric modeling, prediction and optimization of interaction between machining
process and structure using multi-body systems and implicit filtering" is to build a
simulation model of an existing turning lathe and to optimize the manufacturing
process with implicit filtering.
In this work, a facing process is used as test case. The machining operation runs
as follows: The turning tool moves to the point, where the machining starts. At
this point, the exact time of contact between the turning tool and the workpiece is
detected and an additional connection between cutting edge and workpiece is es-
tablished. This process is modeled by a modified unilateral contact augmented by
a cutting model.
In Section 16.2, an overview of the mechanical model of the turning lathe is
given as well as a model of the coupling between process and structure. For opti-
mization it is necessary to develop an objective function, where the quality and the
productivity of the processes have to be represented. The objective function is de-
veloped and presented in Section 16.3. Simulation results are utilized to calculate
the new surface and its characteristics, which criteria for an assessment of the
362 R. Britz et al.
machining quality are obtained from. These results, combined with the productivi-
ty of the process, are used as objective function variables. Optimized parameters
for the operation are detected with respect to the dynamic limits of the working
machine. Finally, results of the optimization process are shown and compared to
measurements.
16.2 Modeling
16.2.1 Multi-body Dynamics with Smooth Contacts

A brief overview is given for the equations describing the dynamics of systems
with bilateral and unilateral contacts including friction; a detailed description can
be found in [1]. The main focus is the formulation of contact problems. There are
two different ways to describe systems containing contacts: the associated force
laws can be functions of the system state or set-valued. Due to the important influ-
ence of local compliance in the contact model of the cutting process, the unilateral
contact formulation utilizes a functional force law; see [2]. This is in contrast to
purely rigid contact models, which are subject of current research.
Classically, contacts in multi-body dynamics are described as spring and dam-
per elements, minimizing penetration of the bodies coming into contact. Leading
to continuous time-dependent state variables, the resulting systems are called
smooth. The entire modeling uses MBSim, an open source multi-body simulation
software initiated at the Institute of Applied Mechanics [3], and available at
http:\\mbsim.berlios.de .
16.2.2 Equations of Motion

First attention is focused on multi-body systems without contact interactions. The
basic equations of motion are
M ( q )u = h(u , q , t ) . (16.1)
The mass matrix M depends on the vector of generalized position q . The vector
h holds all smooth external and gyroscopic forces depending on q , the genera-
lized velocities u and time t . For constrained systems, (Eq. 16.1) must be ex-
tended by the contact forces:
M ( q )u = h (u , q , t ) + Wλ . (16.2)
The constraint matrix W =[WN , WT ] provides the transformation from the space
of constraints to the configuration space. The vector λ = [λ N , λT ] holds values of
the normal and tangential contact forces.
16 Modeling and Simulation-Based Optimization of a Turning Process 363
16.2.3 Contact Kinematics

For all contacts including bilateral and unilateral, rigid and flexible models, a uni-
tary framework describes the kinematics of the associated contact points rc . These
potential contact points on the contours need to be determined and lead to the gap
distance g N and the relative velocities g N and g T . This information is used for
the calculation of generalized force directions W as well as for the evaluation of
force laws. Contours are described by a position vector r = r (q, s ) , the contour
normal n = n(q, s ) and the associated tangents, t1 = t1 (q, s ) =(∂r / ∂s ) and t2 re-
spectively, which depend on the generalized position of the associated body and
contour parameter s . Potential contact points are identified by the necessary con-
tact condition
 rT t 
f (q, s )=  DT 1  =0 . (16.3)
r t 
 D 2
The difference vector rD = rC1 − rC 2 between the potential contact points on both
bodies is perpendicular to both tangents.
Fig. 16.1 Contact between two bodies
As an example, the evaluation of the contact kinematics between a circle and a

general contour, as illustrated in Figure 16.1, is pointed out briefly. A detailed
description can be found in [1]. The contour parameter for potential contacts is
calculated by solving (16.3). Using the contour parameter s , which refers to po-
tential contact points, the distance between the two bodies results from
g = rD n . Based on the contact points, the Jacobians
N 2
J = (∂rCi / ∂q )T = (∂rCi / ∂u )T between the system-space and generalized coor-
dinates of the two bodies are developed. The generalized force directions are
364 R. Britz et al.
WNT = n1T J C1 + n2T J C 2 , (16.4)
WTT = t1T J C1 + t 2T J C 2 . (16.5)
WN and WT are used for projecting normal and tangential contact reactions
in (16.2).
16.2.4 Single-Valued Force Laws for Contacts

A common approach to modeling-compliant unilateral contacts is to take into ac-
count contact elasticity by using spring and damper elements to penalize penetra-
tion in normal direction. Therefore, λ N is decomposed into forces resulting from
the elastic and damping influence, λ Nc and λNd respectively. The normal contact
force λ N is zero, if the bodies are separated. Using linear spring-damper ele-
ments, the force law is written as
λNc = −cg N if g N < 0 , else λNc = 0 . (16.6)
For the damping value λNd , a linear correlation is applied for g N < 0 :
λNd = −dg N . (16.7)
Fig. 16.2 Linear and non-linear force law [2] Fig. 16.3 Set valued and regularized
form of COULOMB’s law [2]
Equivalent force laws are used for bilateral contacts without the distinction of
case in (Eq. 16.6). Friction is considered in closed unilateral contacts. Using the
normal force λ N and relative tangential velocity g T , the scalar force λT of planar
Coulomb friction is given by:
g T =0  λT ≤ μ λ N , (16.8)
g T
g T ≠0  λT =− μ λN . (16.9)
g T
This set-valued description is not suited for ordinary differential equation (ODE)
solvers, which are used for time-integration during simulation. Since high uni-
directional tangential velocities without sign-change are expected during the
cutting process, stick-slip transitions do not have to be taken into account, thus
enabling the regularization of (Eq. 16.9) and ODE time integration. One possibili-
ty is to smooth the set-valued force law of Coulomb as follows
 −
g T 
g T  
g T < ε  λT =− μ 1 − e  λN ,
v ref
(16.10)
g T  
 
g T
g T > ε  λT =− μ λN . (16.11)
g T
In contrast to (Eq. 16.8) and (Eq. 16.9), (Eq. 16.10) and (Eq. 16.11) do not permit
real sticking characterized by g T =0 and i. g. λT ≠ 0 . (Eq. 16.10) gives only
λT =0 for vanishing tangential velocities at the contact point. Both Coulomb fric-
tion laws, exact and regularized, are depicted in Figure 16.3.
16.2.5 Machine Model

The analyzed turning lathe has a classical design, see [2]. The spindle is embedded
in the headstock, which is fixed to the machine foundation. To realize the move-
ment of the turning tool a composite of a longitudinal carriage and a top slide is
used. In simulation, all components are modeled as rigid bodies with 6 degrees of
freedom. The physical representation of a rigid body is reduced to mass, rotational
inertia with respect to the point of reference and, if required, rigid geometric ele-
ments for contact descriptions. If the force application point is constant, it is poss-
ible to describe its position by a constant vector in the local coordinate system of
the respective body. Otherwise, the geometry is described by a contour parameter
(see Section 16.2.3), using interpolation methods [4, 5]. Arbitrary loads, including
connections between bodies by spring-damper elements, can be applied. Spring-
damper elements describe the connections between the bodies. Experiments are
made on a lathe, which has linear re-circulating ball bearings for the tracks. Vary-
ing levels of stiffness between the bear on and uplift loads can be observed, as in-
dicated in the manufacturers’ technical data sheets, see Figure 16.2. These tend to
result in eigenfrequency responses at the point of change in stiffness, see [2]. The
drives are modeled as kinetic excitations.
366 R. Britz et al.
16.2.6 Modular Cutting Force Model
16.2.6.1 Analytical Model
In order to determine the dynamic cutting forces at the workpiece and the tool cen-
ter point, depending on the process parameters, the applied materials and the tool
geometry, a modular cutting force model was set up. Please note that a detailed
description of the force model can be found in [6] and [7].
It is based on an analytical approach to describe the cutting forces. As there is
no formula for the general description of all turning processes, a modular structure
is chosen. Part formulations for the dynamic material behavior and non-linear fric-
tion at the tool chip interface can be selected, depending on the actual cutting
process. According to [8], the total forces in cutting FC, in feed FF and in passive
direction FP with respect to the side cutting edge angle ψ can be described as
follows:
′ + FCP
FC = FCS ′ , (16.12)
FF = (FFS ′ ) ⋅ cosψ + (FPS

′ + FFP ′ ) ⋅ sinψ
′ + FPP , (16.13)
FP = (FFS ′ ) ⋅ sinψ − (FPS

′ + FFP ′ ) ⋅ cosψ
′ + FPP . (16.14)
The prime symbol labels forces with respect to the cutting edge plane (see Fig.
16.4), while the indices S and P mark components resulting from shearing and
ploughing. [9] states that the shearing forces can be calculated by (Eq. 16.15) -
(Eq. 16.17) with the shear yield stress τs, the uncut chip thickness h, the width of
cut b, the rake, the chip flow and the inclination angle α, η and i as well as the
mean friction angle β and the shear angle Φ:
τ bh cos(β − α ) + tanitanηsinβ
′ = s ⋅
FCS , (16.15)
sinΦ
cos 2 (Φ + β − α ) + tan 2ηsin 2 β
τ s bh sin (β − α )
′ =
FFS ⋅ , (16.16)
sinΦcosi
cos (Φ + β − α ) + tan 2ηsin 2 β
2
τ bh cos(β − α )tani + tanηsinβ

′ = s ⋅
FPS . (16.17)
sinΦ
cos 2 (Φ + β − α ) + tan 2ηsin 2 β
One solution for the determination of the ploughing forces is based on edge coef-
ficients per unit K’, which can be determined by cutting tests and extrapolating the
measured forces to an uncut chip thickness of zero [9]
′ = K CP
FCP ′ ⋅ b , FFP
′ = K FP
′ ⋅ b , FPP
′ = K PP
′ ⋅b . (16.18, 16.19, 16.20)
Furthermore, a strictly analytic approach for the ploughing forces, based on metal
cutting simulations (see Sub-section 16.2.6.2), is described in [10].
Fig. 16.4 Machining variables and cutting force directions for tools with a side cutting edge
angle [7]
16.2.6.2 Metal Cutting Simulation
Additionally, a metal cutting simulation based on the Finite Element Method

(FEM) is set up (see Fig. 16.5) to determine the required parameters τs, β and Φ.
The used commercial pre-processor MSC.Mentat provides the possibility to define
models automatically using python scripts. Based on the considered cutting
process, a model of 2D orthogonal cutting is created, with tool and workpiece
modeled as deformable bodies. The thermo-mechanically coupled Lagrangian
numerical problem is handled by the MSC.Marc solver. Based on the work of
[11], the contact of tool and chip is considered thermally ideal. While the distant
parts of workpiece and tool remain at ambient temperature the heat transfer of free
surfaces to the ambiance is neglected.
Apart from the model generation, an appropriate material law has to be selected
for the defined cutting process. The determination of work material flow stress in
the primary shear zone is of high importance for the calculation of the cutting
forces. Thereby, the flow stress is mostly affected by micro-structural conditions,
temperature, strain and strain rate. The cutting force model modularly includes
different constitutive material laws to cover different lattice structures (see also
[12], [13]):
• Body-centered cubic materials

• Face-centered cubic materials
• Hexagonal closest packed materials
368 R. Britz et al.
Fig. 16.5 Coupling of the analytical and numerical model [7]
Additionally, the law of Johnson-Cook can be selected, which is often imple-

mented in commercial FEM software. Furthermore, a modified version of John-
son-Cook by [14] allows the consideration of the influence of blue brittleness on
the flow-stress curve.
Finally, an adequate description of the friction at the tool-chip-workpiece inter-
face is needed, as the friction conditions as well as the tool wear determine the
heat generation in the secondary and tertiary deformation zone. Due to high con-
tact pressures, temperatures and chip velocities the interfacial friction is not conti-
nuous [15]. On the rake face, the normal stress decreases from maximum to zero
from the tool tip to the point, where the chip separates from the tool. The shear
stress divides into a sticking and a sliding region. The mean frictional angle β can
be calculated directly, if measurement data for normal and frictional stress distri-
butions is available [7]. Otherwise, constant shear friction is chosen for the stick-
ing region while Coulomb friction is applied to the sliding region and β has to be
derived from the FEM simulation. Therefore, the resulting stress distribution of
the metal cutting simulation has to be evaluated in the deformation zones. In
addition to β, the shear yield stress τs can be calculated from the stress distribution.
The shear angle Φ can be determined from the chip formation. In addition, the
point of material separation can be derived from the FEM results and used for the
analytic determination of the ploughing forces [10].
As the temperature in the deformation zone plays a major role for an accurate
simulation of the cutting process, an experimental setup has been developed to de-
termine the temperature of a real cutting process [6]. A high-speed pyrometer
(Kleiber KGA 740-LO) provides the possibility of capturing the temperature while
machining. To get results for both the workpiece and the tool flank face, a special
test piece with milled holes has been developed (see Fig. 16.6). While the temper-
atures of the workpiece in decreasing distance to the tool can be observed at the
beginning, the temperature of the tool is measured at the moment, when the tool
penetrates the milled cavity. Afterwards, these results were used to calibrate the
FEM metal cutting simulation.
Fig. 16.6 Schematic representation of the temperature measurement setup [6]
Finally, all necessary parameters are transferred to the analytical cutting force
model so that it is possible to calculate the forces on workpiece and tool center
point resulting from the defined cutting process.
16.2.6.3 Comparison of Measurement and Simulation
The validation of the modular cutting force model was performed by a comparison
of simulated and experimentally-determined cutting forces. Therefore, facing tests
370 R. Britz et al.
were carried out at a turning center while capturing the resulting forces by a 3-
component tool holder dynamometer (Kistler 9121) carrying a tool with a side cut-
ting edge angle ψ of 45° and an inclination angle i of zero. To avoid the influence
of machine vibrations the process was driven within stable cutting conditions.
The results for the machining of AISI 1045 can be found in Figure 16.7. The
simulated forces resulting from a model with the constitutive material law for
body-centered cubic materials and an additional calculation for blue brittleness
(see [7]) and the measured forces show high agreement in the area above 90 %.
An even higher accuracy can be achieved by calibrating the model with meas-
ured temperature and force values. At this, the model updating is performed by
iteratively optimizing the friction parameters. As reliable frictional and normal
stress distributions can be currently found only by experimental determination (see
[15]), it is of high importance to improve the theoretical description of interfacial
friction in future research.
Fig. 16.7 Comparison of measured and simulated forces, exemplarily in cutting direction
(AISI 1045, v = 160 m/min, b = 2.83 mm) [7]
In conclusion, the modular cutting force model is available for the calculation
of the cutting forces on workpiece and tool center point. Hence, the parameterized
analytical force model can be implemented into the Machine Process Interaction
simulation.
16.2.7 Coupling of Cutting Force Model and Multi-body

Simulation Model
16.2.7.1 Cutting-Specific Contact Extensions
The simulation uses the cutting model mentioned above, which reduces the cutting
process to a single force between the chisel and the workpiece. The force applica-
tion point on the chisel is always constant and thus modeled as a point. Since the
force application point on the workpiece depends on the position of the chisel and
the angular position of the workpiece, it is modeled as a contour with switchable
radius, see Figure 16.9. Furthermore, the orientation between chisel and workpiece
has to be known to calculate the forces with (16.15) - (16.17). The position is af-
fected by the machine vibrations and must be determined at every time step; see
[19]. For that, the cutting edge and workpiece contour are augmented with vectors,
as illustrated in Figure 16.8.
Fig. 16.8 Calculation of current cutting angles[19]
The vectors describe the direction of the blade and the cutting face with respect
to the chisel. Using them together with the frenet trihedron of the contour and the
axial direction of the workpiece, all angles influencing the cutting forces can be
determined. For example, the angle κ is calculated from the bi-normal bw and
the direction of the blade sm , as follows:
π −bw sm
κ = − . (16.21)
2 bw sm
To figure out the sign of the angle we evaluated the scalar product with the normal
of the frenet trihedron nw :
sgn(κ ) =sgn( nw sm ) . (16.22)
Additional angles are calculated accordingly. Note that these operations are made
with respect to an absolute coordinate system.
Up to now, equations have been formed to compute the radial penetration and
all cutting angles. Penetration in axial direction and the discretization of the work-
piece, which yield an elliptical contact contour of the workpiece, are determined
by imperfections of the workpiece or clamping of the workpiece during the cutting
process. To this end, the oval contour of the workpiece is discretized during the in-
itialization, as shown in Figure 16.9.
372 R. Britz et al.
Fig. 16.9 Workpiece discretization
The radius of the ellipse is described in polar coordinates and is added to the
face side of the workpiece. The origin of the contour corresponds to the axis of ro-
tation. Before the cutting process starts, κ is set to 45°. During the machining, the
cut width b depends on time and increases. The cutting forces are a function of
chip width b and depth h , so b and h have to be evaluated in each time step.
For this purpose the initialized, previous and current contours and the length of the
workpiece are stored. The gap between workpiece and chisel is calculated from
the contact kinematics between the previous contour and the position of the cut-
ting edge. Using the contour parameter s , the current chip depth is estimated and
the current contour radius is stored. Also, the current position of the cutting edge,
the angle κ and the chip width in radial direction brad are calculated using the ra-
dius of the initialized contour. The chip width in axial direction, bax results from
the length of the workpiece minus the cutting edge position. The values brad and
bax are compared and the smaller one is used to calculate the actual chip width b
in the current state.
16.2.7.2 Kinematics of the Cutting Process
The model for the contact between turning tool and workpiece introduces the inte-
raction between the cutting process and the machine: one of the important pheno-
mena in process-structure-interaction is the separation of turning tool and
workpiece during the cutting process. The theory presented in the preceding sec-
tions allows the description of this behavior. The contours of the bodies are de-
scribed by a point (turning tool) and an oval contour (workpiece). For modeling
the cutting process, the gap between turning tool and workpiece is needed, which
is calculated from (16.3). With the achieved contour parameter s , the distance or
penetration results from g N = rSWS n1 . In addition, if penetration is detected with the
known position vector r , the new surface is calculated and recorded at every time
step. The values of the radius are recorded in a table and interpolated during the
next revolution. The solver needs to detect transitions and distinguish between the
possible configurations; see [19]. For this purpose, three operation states are de-
fined according to Figure 16.10:
a) Open contact: turning tool and workpiece are separated and the gap is positive.
b) Closed contact, no cutting: the turning tool contacts the workpiece without cut-
ting. Thereby, no chip is taken from the workpiece. At the point of contact,
normal and tangential forces are exchanged as for a standard frictional con-
tact. The penetration is smaller than a defined minimal penetration ε.
c) Closed contact, cutting: the penetration exceeds ε. Cutting starts and the contact
forces are overlaid by the cutting forces. During the cutting process, the new
surface resulting from the process needs to be recorded.
a) Open contact b) Closed contact c) Cutting

Fig. 16.10 Contact situations [19]
The numerical simulation accurately detects the transition times between the
operation states by event-based integration. It uses a root function to monitor the
contact states of the cutting model. This includes any overruns and drops below
the minimal penetration level, which describes the beginning and end of cutting.
16.3 Optimization
The aim of the project is to determine the optimal parameter set for an operation,
i. e. detect a set, which satisfies the economic aspect of the production. The com-
prehensive model of process and structure is used to predict the result of the ma-
chining; see [20]. With this information, the optimization algorithm is able to
determine the best parameter set.
16.3.1 Analysis of the Optimization Problem

In the following, the mathematical description of the optimization problem needed
for process optimization is introduced. Based on this mathematical formulation, a
suitable optimization algorithm is selected. An objective function is defined de-
pending on given optimization parameters p , which quantify the optimization
target:
g ( p ) = g (q ( p ), u ( p ), λ ( p )) . (16.23)
The objective is to minimize the target function g by finding optimal parameters

p limited by upper and lower bounds for each component of p . This leads to a
constrained optimization problem:
374 R. Britz et al.
min g ( p), B = { p ∈ R m : li ≤ pi ≤ ui ∀i = 1,..., m} . (16.24)

p∈B
The work of [16] shows that even optimization problems concerning rather simple
mechanical systems can result in target functions, which are non-smooth, have
many local minima and may chaotically depend on the optimization parameters.
Based on these experiences, the optimization algorithm IFFCO [17] is chosen, an
implementation of the implicit filtering method [18] for problems with bound
constraints.
16.3.2 Optimization Process

Finally, manufacturing has to satisfy economic objectives as well as the necessary
quality standards. This simple assertion defines the objective function for process
optimization - a compromise between productivity and quality. Therefore, the time
base simulation results have to be interpreted and weighted. In this work, only one
cut is optimized by changing setting values, e. g. the feed per revolution. Con-
versely, environmental conditions, including the structure of the machine, the
length of the workpiece and the workpiece clamping are not part of the optimiza-
tion. With these restrictions, the material removal rate Q expresses the productivi-
ty of the process. Using the parameters, the feed per revolution f , cutting depth
a p , and cutting velocity vc obtained from the simulation results, material remov-
al rate is evaluated with:
Q = vc a p f . (16.25)
On the other hand, the assessment of quality is not straight forward since it de-
pends on forces, deflections, vibrations, geometry of the tool, etc. In this work, the
objective function considers only deflections and the tool geometry to assess the
new surface and forces to value the tool wear.
The simulation records contact and cutting forces as well as the trace of the
turning tool on the workpiece. The objective function evaluates some specific val-
ues: first, the mean offset Δz of the turning tool path from the desired path,
representing the static deflection during the cut. Another important factor is the
theoretical roughness of the produced surface. Using the trace upon the workpiece
and the radius of corner it can be calculated as follows: a set of points in radial di-
rection is chosen within a small region around the x axis and used as supporting
points, see Figure 16.11a. Using the radius R and feed per revolution f , the
piecewise surface contour in radial direction is calculated, see Figure 16.11b:
 2 
 r  f f
zi ( x) =zi 0 + R1 − 1 −   ; for − <r< . (16.26)
  R  2 2
 
Next, we calculate the commonly-known center line average Ra and average sur-
face roughness Rz along this contour. Note that these values are only theoretical
estimates of surface quality. In real processes, surface roughness is also influenced
by other effects like abrasion thermal stress and so on. However, the trend of qual-
ity is determinable and adequate for optimization.
(a) Points of turning tool trace (b) Piece-wise surface calculation
Fig. 16.11 Calculation of the theoretical roughness [20]
The cutting force indicates the stress on the turning tool tip and the workpiece
surface. The ratio of the mean values of cutting force FC and normal force FN on
the cutting edge gives an advice of the abrasion. The objective function is eva-
luated using the simulation results. Using factor a we are able to weight produc-
tivity and quality:
R R Δz FN   Q0 
g = (1 − a)  a + z + +  + a  . (16.27)
 a0
R R z0 Δ z 0 FC   Q 
While the chosen objective function was successfully applied to find parameter
settings, which guarantee a good compromise, the real target is different: we want
to determine a parameter setting, which realizes the maximum productivity at a
minimum quality necessary to satisfy our technical requirements. Hence, factor a
would have to be known a priori. However, by introducing penalty terms, if the
surface quality is below a certain boarder, the optimizer can be utilized in a differ-
ent way: We want to get a minimum quality, therefore, the best parameters for the
maximum productivity have to be generated by the optimizer. At first, the virtual-
ly-produced surface is interpreted. If the minimum quality boarder is crossed, a
penalty is added to the objective function. If not, the objective function is calcu-
lated as given in (16.27).
376 R. Britz et al.
16.4 Optimization Results

To demonstrate the potential of the introduced method a facing process is opti-
mized using the simulation of a combined milling and turning machine with con-
ventional features. The cutting force model is qualified for the machining of work
material AISI 1045 with TiN-coated carbide reversing [7]. A continuous chip is
adopted. The plate has a tool cutting edge angle of 45° and a rake angle of 0°. The
cutting rate is 160 m/min. We searched for the optimal parameter of feed per revo-
lution f and cutting depth a p to finish a workpiece. Starting from the values
f = 0.1 mm
rev
, a p = 0.2mm , We obtained the optimized parameters f = 0.081 mm
rev ,
a p = 0.46mm after 53 steps. While theoretical surface roughness is preserved
during the optimization process, the productivity increases by about 70 % and the
ratio of normal and cutting force decreases by about 60 %, see Figure 16.12.
Fig. 16.12 Change of quality, force ratio and productivity during the optimization process
These results show that the simulation tool is able to find an optimal setup for
smooth finishing, regarding the used model.
These results are compared with measurements on the machine unit. Since the
number of experiments is limited only certain machine settings could be realized.
Therefore, some parameters, which strongly affect surface roughness, tool stress or
productivity, were chosen, e. g. the settings for the cut at the beginning and at the
end of the optimization. Table 16.1 lists parameter values used in the experiments.
In Figure 16.13, the theoretical roughness is compared to the measured rough-
ness during the optimization process. The difference between simulation and ex-
periment increases with a decreasing feed per revolution because the influence of
the chisel geometry decreases and the influence of the abrasion on the workpiece
increases. The abrasion effect is not modeled; the lower boarder of the box con-
straint must be modified. The failure increases rapidly, if both cutting parameters
go to zero.
Table 16.1 Parameter values for experimental verification
Step 1 2 5 7 8 9 11 30 40 52
Infeed 2.15 0.2 0.2 0.68 0.2 0.34 0.34 0.49 0.43 0.46
[mm]
Feed 0.1 0.57 0.34 0.10 0.21 0.06 0.18 0.07 0.07 0.08
[mm/rev]
As mentioned in Sub-section 16.3.2, the predicted quality is better than the

measured one. However, the trend of predicted roughness is close to the measured
trend and this is sufficient for the optimization.
Fig. 16.13 Comparison of the measured roughness and the theoretical roughness. The trend
of the produced roughness is predictable.
The predicted and the measured forces are shown in Figure 16.14. The mean
value of the predicted force is also close to the measured force.
Fig. 16.14 Comparison of the mean values of the measured forces and the predicted forces
The comparison shows that the simulation sufficiently predicts product quality
and tool stress. This allows the determination of an optimal cut setting, which sa-
tisfies the demands on productivity and quality.
378 R. Britz et al.
In general, the result is influenced by the machine. The developed tool is able
to optimize the process with respect to the machine. This is figured out in [20]. In
this article, two virtual machines with different stiffnesses are used. A rough ma-
chining process is optimized for both machines. The optimization tool finds dif-
ferent parameter settings on the machines, indicating that the optimization tool
figures out a parameter setting with respect to the process-structure interaction.
16.5 Conclusion
A tool for the optimization of turning processes has been introduced, which in-
cludes the model of the turning machine, the process and the target function. The
capability of the optimization environment is illustrated by means of examples. The
tool obtains the optimal parameter settings for a smooth finishing process with the
results of the simulation having been successfully verified by experiments.
References
[1] Pfeiffer, F., Glocker, C.: Multi-body Dynamics with Unilateral Contacts. John Wiley
& Sons, New York (1996)
[2] Britz, R., Ulbrich, H.: Lathe: Modeling and Coupling of process and structure. In:
Proccedings of the 1st International Conference on Process Machine Interactions, pp.
231–238 (2008)
[3] Förg, M., Zander, R., Ulbrich, H.: A framework for efficient simulation of spatial
contact problems. In: ECCOMAS (2007)
[4] Shabana, A.A.: Dynamics of Multi-body Systems. Cambridge University Press,
Cambridge (1998)
[5] Ulbrich, H.: Maschinendynamik. Teubner Verlag, Wiesbaden (1996)
[6] Zaeh, M.F., Schwarz, F.: Consideration of Tool and Workpiece Temperatures in a
Modular Cutting Force Model. In: Proceedings of the 1st International Conference on
Process Machine Interactions, pp. 353–360 (2008)
[7] Zaeh, M.F., Schwarz, F.: Modeling and Simulation of Process and Structure Interac-
tions Considering Turning Operations. In: Proceedings of the ASME International
Manufacturing Science and Engineering Conference (2009)
[8] Oxley, P.L.: The mechanics of machining: an analytical approach to assessing machi-
nability, Chichester u.a., Horwood (1989)
[9] Altintas, Y.: Manufacturing Automation. Cambridge University Press, Cambridge
(2000)
[10] Schwarz, F.: Simulation der Wechselwirkungen zwischen Prozess und Struktur bei
der Drehbearbeitung. iwb-Forschungsberichte Nr. 242. Herbert Utz Verlag, München
(2010)
[11] Yen, Y., Jain, A., Altan, T.: A finite element analysis of orthogonal machining using
different tool edge geometries. J. Mat. Proc. Technol. 146, 72–81 (2004)
[12] Zerilli, F., Armstrong, R.: Dislocation-mechanics-based constitutive relations for ma-
terial dynamics calculations. J. Appl. Phys. 61, 1816–1825 (1987)
[13] Zerilli, F.: Dislocation-mechanics-based constitutive equations. Metallurgical and
Material Transactions 35A, 2547–2555 (2004)
[14] Sartkulvanich, P., Koppka, F., Altan, T.: Determination of flow stress for metal cut-
ting simulation – a progress report. J. Mat. Proc. Technol. 146, 61–71 (2004)
[15] Özel, T.: The influence of friction models on finite element simulations of machining.
Int. J. Machine Tools and Manufacture 46, 518–530
[16] Neumann, L., Ulbirch, H.: Optimisation of a CVT-Chain. In: Proceedings of the 6th
World Congresses of Sturctural and Multidisciplinary (2005)
[17] Choi, T.D., Eslinger, O.J., Gilmore, P.A., Kelly, C.T.: User’s Guide to IFFCO. Center
for Research in Scientific Computation, North Carolina State University (2001)
[18] Kelley, T.C.: Iterative Methods for Optimization. Frontiers in Applied Mathematics.
SIAM, Philadelphia (1999)
[19] Britz, R., Ulbrich, H.: Simulation of Facing Processes of Profiles with Hexagon Cross
Section. In: Proceedings of the ASME International Manufacturing Science and En-
gineering Conference (2010)
[20] Britz, R., Ulbrich, H.: Modeling and Simulation-based Optimization of a Facing
Process. In: Proceedings of the 2ndt International Conference on Process Machine
Interactions (2010)
Part IV
Forming
Chapter 17
Advanced Forming Process Model - AFPM
K. Großmann, A. Hardtmann, H. Wiemer, L. Penter, and S. Kriechenbauer
Abstract. This chapter discusses methods of modeling and simulating metal form-
ing processes and explains their application in product design, production and
process planning. In today’s Finite Element (FE)-based forming analysis, major
effects on the forming process are being neglected. Based on the analysis of the
elastostatic press and tool properties, a conventional FE process model was ex-
tended with the most dominant elastostatic influences. It is shown that complex
elastic systems, such as die cushions and tool guidance, are quite easily imple-
mented in FE process simulations by using discrete elements and other reduction
methods, recently introduced in commercial simulation software. The benefit of
the Advanced Forming Process Model (AFPM) is demonstrated by an experimen-
tal verification. Servo mechanical presses enable the manufacturers to establish
high-speed processes in sheet metal forming. There, the dynamic press behavior
has a much larger influence on the forming process than it has in the relatively
static conventional deep drawing. As an example, a highly dynamic forming
process is simulated and explained in the following.
17.1 Introduction
Established as a powerful tool for designing sheet metal processes, the Finite Ele-
ment Method (FEM) supports product development as well as the planning of
production process and tool manufacturing.
Current process models merely contain the blank and the interface between
blank and tool, the so-called die-workpiece interface [1]. Commonly, the blank is
meshed with shell elements with simple constitutive equations. The die-workpiece
interface is usually represented by the Coulomb friction law and a contact algo-
rithm with constant coefficients. In present FE-simulations, the forming tool and
machine are modeled as rigid, i. e. the effects of their elastic properties on the
forming process are ignored. In reality, both forming tool and press clearly affect
the forming process as the requirement of die spotting during the tool try-out de-
monstrates. One way to minimize time and effort in tool try-outs is using more ac-
curate simulation methods during the tool planning stage.
New press technologies, such as servo-mechanical presses, enable faster
processes with remarkable accelerations. Higher accelerations require light parts
in motion. Lighter tools and press parts are less stiff and would therefore deform
384 K. Großmann et al.
stronger under process load. Conversely, these elastic deformations influence the
process conditions and therefore the final part shape significantly. Conventional
FE process simulations do not show the impact of elastic deformation on the
process.
Therefore, process simulations are required, which take elastic tool and press
properties into account and enable manufacturers and press designers to compute
the process-machine interactions. A better quality and efficiency control of metal-
forming processes already during the tool’s planning stage require interactions be-
tween the subsystems machine, tool and workpiece to be included in a complete
forming model.
The objectives of this chapter are
• introducing a holistic yet simple approach to address elastostatic deformations
of press and tool in FE process simulation and
• showing a way to simulate highly dynamic forming processes.
17.2 Ways of Model Advancement
Holistic process modeling of forming operations entails the coupling of the three
subsystems, press, tool and workpiece. Concepts as to how to connect the “work-
piece” model (the common metal forming model) to the press model are presented
in [4, 14]. The approaches are differentiated according to their way of integration:
• Offline coupling (non-reactive) is process characterization within the machine
model based on process force progressions. The force curve is calculated by
means of an FE-workpiece model. The data are transferred to the press model
via an ASCII-file [5]. Among others, this coupling method is applicable to ana-
lyze the operating performance of the machine and to detect the load on assem-
blies. In [5, 6], the ram deflections of a multi ram press were estimated. In [7],
the effects of ram tilting on the deep drawing process were analyzed.
• Integrated coupling stands for the extension of the workpiece model with a
press model in the FEA-environment. The integrated FE-model realizes direct
interaction between the process load and tool position as a result of the press
behavior. The concept describing the press by a reduced structure representing
process relevant effects is exemplified in [2, 3].
• By coupling discrete models (co-simulation) the machine effects are depicted
in independent press models. Advantageously, the complex influences of the
machine behavior (drive, guidance system, frame, etc. [9]) are modeled in de-
tail by Multi-Body-Simulation (MBS) while the workpiece description is made
by FEM. Thus, two simulation tools (MBS and FEM) connected by simulator
coupling are to be applied. The simulator coupling organizes the exchange of
data and synchronizes the different solution algorithms. A complete forming
model by coupling discrete simulation models is exemplified in [8]
17 Advanced Forming Process Model - AFPM 385
To design the forming process only the press behavior directly affecting the
process is relevant. Interactions between assemblies inside the press are less im-
portant. Hence, the integrated solution realized in [2, 3] is adequate. Furthermore,
applying the integrated approach avoids the independent major problem of coupl-
ing different simulation software.
17.3 Development of Advanced Forming Process Model

The acting loads during the process cause misalignments of the tool components
depending on the machine stiffness. The consequences are
• Form deviations of the sheet metal part,
• Surface quality variations,
• Change of adjusted blank holder forces,
• Increasing tool wear.
In conventional models of the forming process, the press influence is neglected.
Here, forming process models enhanced by implementing elastic properties of
machine and tool are called Advanced Forming Process Models (AFPM).
17.3.1 Press Model

The elastostatic behavior of the press is important when comparing different ma-
chines. The operational accuracy of the press results from the total deformations
of all mechanical parts. Accuracy defining properties of the press based on its de-
formation and deflection have significant influence on the sheet metal forming
process. In German standard DIN 55189 [10], these properties are determined as
accuracy parameters of presses. For the time being, only static properties are inte-
grated in the model:
• vertical total stiffness ctotZ,
• horizontal total stiffness ctotX and ctotY,
• resistances against tilting (tilting stiffness) cα and cβ.
The resulting stiffness parameters combine all static properties of the press assem-
blies each with regard to one degree of freedom in the press’ coordinate system.
In the following, the above-mentioned press properties are added to the AFPM
step by step. Currently, tool components are handled as rigid bodies. In a first step,
it is therefore sufficient to fix the tool model on a rigid model of ram. A rigid ram
model has the advantage of allowing principle stiffnesses of the press to be direct-
ly attached to the centre of gravity of the ram, as shown in Figure 17.1a. The prin-
ciple stiffness describes the elastostatic force-displacement behavior of the press
in X, Y or Z-direction or around the corresponding axis.
Fig. 17.1 Advanced Forming Process Model (AFPM) a) extended by a press model consi-
dering elastic mounting of the rigid ram in the COG, b) extended by a press model sepa-
rated in drive and guidance springs for the application of an elastic ram (example: O-frame
press with 4-point configuration)
In the next step, the forming process model is enhanced by modeling the tools
as elastic. This approach demands the tool to be elastically embedded into the ma-
chine model. The press ram and the table of the press have to be modeled as elas-
tic. As in real press structures, the elastic ram model’s joints to the press structure
are positioned at the locations of press drive and ram guidance.
Figure 17.1b illustrates the AFPM upgraded with spring elements, which act in
horizontal (X and Y) direction and in vertical (Z) direction. According to the real
press drive system, they are connected to the corner nodes of the ram model. The
following properties, which influence the press behavior, define the position and
parameterization of the spring elements:
• total stiffnesses of presses,

• type of drive system (1, 2, and 4-point configuration),
• press structure (C-frame and O-frame, ram, table of the press),
• guidance (with and without clearance).
Models of different press structures are shown in Figure 17.2. The stiffness val-
ues of the horizontal and vertical springs are derived from force-displacement
curves measured as stated by standard DIN 55189 [10]. The measuring proce-
dure to determine the deflection of the press under static load is explained in
Section 1.2.1.
Fig. 17.2 Scheme of models for different press structures
The principle of moments enables the conversion of the measured values of tilt-
ing stiffnesses cα, cβ and total vertical stiffness ctotZ to spring constants cX, cY and
cZ. For this purpose, geometrical parameters as ram size zram and the distances xdr
and ydr between force transmitting drive elements are necessary.
Following this principle, the tilting stiffnesses and the vertical total stiffness
were converted to stiffness parameters of the horizontal and vertical spring ele-
ments for all model structures in Figure 17.2. The transformation functions have
been summarized in Figure 17.3.
Fig. 17.3 Transformation functions
In case there are no measured values for the total stiffnesses available, refer-
ence values from [11] could be used alternatively. There typical ranges of vertical
total stiffnesses and tilting stiffnesses, classified by the press’ nominal force, can
be found. The ram height zram can be deduced from characteristic dimensions of
the press’ workspace. In the following example, the parameters for a press with a
nominal force of 1000 kN are used.
17.3.2 Die Cushion Model

An elementary influence on contact and friction conditions in the interface be-
tween die and blank holder results from the blank holder itself. According to the
usual practice, the blank holder is represented by a constant load applied to the
center of the rigid blank holder surface. However, in reality, the distributed load
on the blank holder surface results from the equilibrium between process load,
which is variable in time and location, and blank holder deflection/deformation.
Demonstrating this essential effect on drawn parts in the forming simulations de-
mands the extension of the model by the deformation and deflection of the blank
holder. Preliminary investigations [12] have led to the following approach.
The blank holder shape under process load depends on its elastic properties. Its
position is determined by the elastic properties of:
• the press (drive, blank holder ram, ram guidance, frame) in case of a double-
action press or
• the die cushion (drive, die cushion plate, guidance) in case of a single-action
press and
• the tool guidance.
To model the blank holder bearing in case of a double-action press the prefixed
press models are applicable. In order to realize the blank holder bearing of a die
cushion the model structures were successively developed, first to apply rigid and
then elastic blank holders.
Fig. 17.4 Model structure of the die cushion for applying an elastic blank holder
The die cushion model for an elastic blank holder is shown in Figure 17.4. The
stiffnesses of die cushion drive, guidance, pressure pin and pin guidance are
represented by substitutional springs. These spring elements are arranged on the
coupling nodes of the elastic blank holder as they are in the real structure. The
stiffness of the spring element was parameterized according to [13].
17.3.3 Demonstration Example

The following example visualizes how die cushion and blank holder affect the
deep drawing process of a rectangular tub. The simulation results of the process
models with integrated die cushion model are shown in Figure 17.5. It illustrates
sheet thickness and flow in forming process models with different levels of ab-
straction. For comparison, in Figure 17.5, the conventional forming process model
(rigid machine, tool and die cushion model) is displayed. Due to a constantly dis-
tributed load across the contact surface the tub possesses a symmetric sheet thick-
ness distribution and a symmetric sheet metal flow. In contrast, applying the
AFPMs the eccentric tool location results in ram tilting (see Figure 17.5b) or in
ram and blank holder tilting (see Figure 17.5c) respectively. Tilting leads to
asymmetric thickness distribution and asymmetric sheet metal flow. In this exam-
ple, the effect of the blank holder deflection to the drawn part is stronger than the
tilting of the ram.
Fig. 17.5 Sheet thickness reduction and material flow, O-frame press with 1-point
configuration a) Rigid press and tool, die cushion supported rigidly, b) Elastic press, die cu-
shion supported rigidly, rigid tool, c) Elastic press, die cushion supported elastically, rigid
tool
The deformations and surface pressures on the blank-holder according to the

results in Figure 17.5 show different distributions corresponding to the interac-
tions between the die cushion, machine and process and justify the sheet metal
flow as shown. In all calculations, the resulting blank holder force was the same.
17.3.4 Modeling Effect of Tool Guidance

The tool guidance has a substantial influence on the deflection behavior of the
tool-active elements. The exclusive consideration of the machine stiffnesses in the
advanced forming processing model is not sufficient to improve the quality of
the simulation results. Without considering the reinforcing effect of the tool guid-
ance the production of sheet metal parts cannot be guaranteed simulatively so that
the design of the parts would have to be changed unnecessarily at this.
The tool guidance can also be modeled with discrete elements. Figure 17.6
shows the reduction of the tool guidance (a) to a beam model (b) and further into a
discrete model with spring elements (c).
The stiffness of the tool guidance in x and y-direction has the characteristic of a
beam and is dependent on the free length of the tool guidance post (lTG). Since lTG
depends on the current press stroke the stiffness of the spring element is coupled
with the z-coordinate of the ram.
During the entire process, nodes N1, N2, N3 and N4, used to couple ram, BH
and bolster plate on the tool guidance, stay aligned along an axis perpendicular to
the surface of the bolster plate. The change of the ram’s instantaneous center of
rotation with stroke is automatically implemented in the FE simulation.
Fig. 17.6 Implementation the elastic behavior of tool guidance into FE process simulation:
a) section view press, b) tool guidance as beam model, c) tool guidance modeled with
spring elements
The path-dependent stiffnesses of spring elements can be calculated from the

bending stiffness of the pillars and contact stiffness of the guide bushing, which
are to be gathered from the catalogue data of the manufacturers [9]. Therefore, a
special material model for (discrete) beam models has to be developed to imple-
ment this reduced model of tool-guidance in the AFPM.
17.4 Experimental Verification

The experimental device was developed to demonstrate the effects of the tool’s
and press’ deflections on the deep-drawing process. Tool and press deflections
were required to be adjustable without any constructional changes of the press, i.
e. the variation of tool and press deflection had to be realized only by changing the
tool design.
17.4.1 Experimental Equipment

The experimental validation of the modeling concept was conducted on the single-
action hydraulic press Wanzke HPV 160, which is available at the IWM. In order
to determine the static tilting behavior and stiffness behavior of the press, defined
moment was applied by using a press-testing device. The curves obtained in the
test, see Figure 17.7, represent the tilting behavior of the ram about the X-axis and
Y-axis. These characteristic tilting curves result in the tilting stiffness cα, and cβ as
well as in the vertical total stiffness ctotZ. These values were converted into the
stiffness parameters of the horizontal springs cX±B and cY±A as well as the vertical
spring cZ into the FE model of an O-frame press with a 1-point driving system.
Fig. 17.7 Tilting behavior and press model with non-linear compression springs
Since the approach to using linear-elastic springs did not correspond to the real
tilting and stiffness behavior of the examined press, the press model is refined
with non-linear spring characteristics. In order to correctly consider the tilting be-
havior, which is dependent on the direction of the rotation, only non-linear com-
pression springs were attached to the corner nodes of the ram, see Figure 17.7.
The stiffness properties of the experimental tool were modified by implement-
ing disc springs in the die to be able to vary the total stiffnesses in the press-tool
system (Fig. 17.8).
In order to determine the relevant process characteristics it is necessary to in-
stall appropriate sensors. For recording the deflection, behavior incremental dis-
placement encoders with a measuring resolution of 1µm were attached to the ram
guidance. Furthermore, inductive displacement sensors were attached to the tool to
measure the distance between die and die bearing plate. Thus, the tilting of the die
provoked by using different disc spring configurations was measurable. Magneto-
strictive displacement sensors were used to determine the tilting and the vertical
resilience of the blank holder.
Fig. 17.8 Modification of the total stiffness and the tilting stiffnesses in the experimental
tool by means of variable disc springs
Fig. 17.9 Measurement system
The punch force was measured with four load-cells placed under the punch.
The press control systems delivered blank holder and ram forces. All input signals
were recorded with PC measuring boards. The measuring results were visualized
using the software Diadem 8.0 by National Instruments. An overview of the mea-
suring system is shown in Figure 17.9.
17.4.2 AFPM of the Experimental Equipment

In the initial FE simulation setup with non-linear springs, the rigid blank holder
was not allowed to tilt. In the simulation, a different tilting behavior was observed
due to an asymmetrical tool installation in the press but no correlations with the
deep drawing failures in the experiment were found. Therefore, it was necessary to
set up a more comprehensive forming process model (Figure 17.10) in LS-DYNA
containing the elastic properties of the HPV 160 press including die cushion,
forming tool, and disc springs.
Chapter 17.4.1 describes how to parameterize the horizontal and vertical
springs, which are attached to the ram corners (cx±B, cy±A, cz). Decoupling the die
from the ram by means of disc springs causes the ram stiffness to have only little
influence on the total stiffness of the experimental set-up. The elastostatic proper-
ties of both die cushion and tool guidance were included by modifying the spring
characteristics of the blank holder model (cx±B,NH; cy±A,NH; cz,NH)
Fig. 17.10 Advanced Forming Process Model considering the elastic properties of the expe-
rimental equipment.
The blank was meshed with shell elements with 5 integration-points over the
thickness. Adaptive meshing was applied. The elastic-plastic material behavior of
DC04 and H260B was modeled using Reihle’s law and parameters from [15] to
describe the flow stress curve. The friction between sheet metal and tool compo-
nents was modeled with the classical Coulomb-Friction-Model and a constant fric-
tion coefficient of 0.1. In order to improve performance and stability of the
simulation the relative tool travel was realized by moving the punch, which is
fixed in reality, instead of the ram.
17.4.3 Comparative Assessment of the Simulation Results

The predictions of the simulation and the experimental results match fairly well,
as indicated by the correlating material draw-in at the four edges of the remaining
flange, see Figure 17.11.
By including non-linear stiffness of machine and tool guidance into the simula-
tion model, the blank holder and the die can deflect according to the applied load
(tilting, resilience). This has a direct impact on the acting normal forces and there-
fore on the frictional forces as well as their distribution on the interface between
sheet and tool.
Fig. 17.11 Comparison of the simulation results with the experimental results - remaining
flange widths for different blank holder forces
Comparing the flange widths MP1 and MP3 on the long sides of the rectangular
cup indicates an asymmetric material draw-in. Due to ram tilting about the x-axis
the material draw-in at MP1 is more strongly restrained than at MP3. Figure 17.11
shows a good agreement between simulation and experiment for these particular
measuring points. This demonstrates that the advanced process model obtains
closer-to-reality results than the classical deep-drawing simulation.
The sheet metal thickness distribution, for example, at bottom corners of the
rectangular cup, reflects the different conditions in the contact surfaces between
sheet metal and tool. There, the substantially larger blank holder pressure (here at
the left flange side) causes a larger thickness reduction.
In order to avoid critical states in workpiece and tool as well as to achieve the
desired workpiece quality, it is important to be aware of the accurate state of de-
formation and stress during the forming process. The state of deformation and
tensile stress was experimentally determined for selected specimens in order to
evaluate the FE simulation results. Therefore, a square line grid was applied on the
blank surface and then scanned by means of photogrammetry. The image
processing returns the xyz-coordinates of the grid nodes.
The deformation and stress analysis as well as the data visualization and the
date storage were realized with the PC software AutoGrid® [16].
The simulation results can be examined based on the determined total or equiv-
alent strains. Computed results and experimentally determined results show a
good agreement. The state of stress for this particular case can be modeled with
high accuracy. This is the pre-condition for an identification of deep-drawing fail-
ures by analyzing the plane strain state with a Forming Limit Diagram (FLD)
Furthermore, the prediction of the wrinkles and cracks using an FLD was con-
firmed by the experimental results. The Forming Limit Curve (FLC) was generat-
ed according to the European standard ISO 12004 [17].
Fig. 17.12 Comparison between the results of the simulation and the experiment - analysis of the
drawing failures using a FLD
The results obtained with the advanced forming process model (Fig. 17.12)
show that the computation of wrinkling and the prediction of cracks match the ex-
perimental test results. Comparing these findings to the results of a classical deep-
drawing simulation clearly shows that experimentally observed deep-drawing
failures could not be predicted without including elastic properties of machine and
cushion into the FE-process model.
17.5 Software Prototype for FE Simulation Set-Up

Setting up a holistic process model with elastic press and tool properties is not an
easy thing to do. The presented methods to obtain an advanced process model
were developed with the open FEM software LS-DYNA. For modeling elastic
boundary conditions given by the machine, discrete elements, such as springs and
dampers were applied. These element types are usually not available for sheet
metal process models in commercial user-friendly software packages, such as
AutoForm, eta/DYNAFORM or PAM STAMP.
In order to facilitate the set-up for the user a software prototype “xForm” was
developed. The software concept enables the user to choose from a selection of
press and drawing cushion structures to implement them into the FE forming
analysis.
Figure 17.13 shows the set-up interface for a new press configuration in xForm.
The process model was divided into three components: blank, tool and machine
(press including die cushion), which are stored to the database of the program. At
present, the geometries of these components can only be loaded directly from
already existing LS-DYNA keyword files.
The data base editor allows the user to define necessary component-specific in-
formation, e. g. frame design and drive configuration as well as the vertical and
tilting stiffness of the press and die cushion.
An interpreter obtains material properties from keyword files and assigns them
to the components in the database. The component “tool” contains all the geome-
try and material data of the die, blank holder and punch. The non-linear material
parameters, such as flow curve or anisotropic values, are assigned to the meshed
blank.
In the modeling module, the three components of the database are linked with
each other. While auto-positioning the components the contact formulations are
defined. Subsequently, boundary conditions, such as ram stroke (drawing path)
and blank holder force as well as simulation-specific parameters such as time-
step-size control or adaptive mesh refinement, need to be assigned to the model.
The model editor facilitates the change of model components. Once the advanced
process model is set-up the program translates the model into an LS-DYNA key-
word file. Finally, the file is sent to the LS-DYNA-solver for computation.
The software prototype xForm shows that extending the FE process model with
machine properties can be realized comfortably. xForm demonstrates how a com-
mercial pre-processor should be arranged to make the methods available to a
broader user group. In further development - then based on a higher programming
language – the following extensions have to be implemented:
- CAD interface,
- mesh generator to mesh-imported CAD structures,
- spring-back compensation module including the tool elasticity,
- post-processor module for visualization and evaluation of the simulation
results,
- users’ help.
Fig. 17.13 GUI of xForm for an easy FE simulation set-up
17.6 Compensation for Elastic-Static Deformations

In order to account for elastic tool deformations in FE forming analysis, [18] discusses
four different approaches, namely static condensation, flexible rigid bodies, tied coarse
and fine meshes and full 3D discretization. Due to its simplicity, coupling coarse brick
elements with fine shell mesh on the surface was used for the investigations.
Fig. 17.14 Methodology to compensate for elastic deformations

Since the simulation of the elastostatic machine and tool behavior is feasible,
error compensation for these effects seems to be the next logical step. Therefore, a
methodology was developed counteracting elastostatic press and tool deformations
by rectifying the tool’s surface before it is built. The surfaces of the tooling after
springback compensation are fed into the FE simulation. The FE model, enhanced
with important elastic properties, calculates the elastic deformation of the tool’s
surface. In the next step, the deformed surface is extracted and the deviation is
added as offset to the original surface (Fig. 17.14)
The experiments were conducted on an experimental set-up with allows for ex-
tensive die deformation [19]. Figure 17.15 displays experimental and simulation
results for deep drawing with and without compensated die surface. The flange
outline (dashed line) shows that the material draw-in with compensated die sur-
face is very close to the draw-in predicted for deep drawing with an ideally stiff
ram. This indicates that the compensation for elastic press and tool deformations is
feasible.
Fig. 17.15 Application of compensation method to S-Rail tooling
17.7 Approach for Elasto-Dynamic FE Process Modeling

Spindle-driven servo presses enable the manufacturers to establish high-speed
processes in sheet metal forming. Furthermore, the advantage of complete free-
dom of press stroke opens doors to perfectly adjust the stroke to the process or to
establish completely new processes in sheet forming, such as so-called “Synchro”
deep drawing [20].
Conversely, high dynamical processes result in new requirements for the design
of the press drive and the structural components in the force flow. Conventional
quasi-static process simulations are obviously not suitable to calculate process pa-
rameters and reaction forces under dynamic conditions. Therefore, new methods
to simulate and predict process conditions for dynamic operations need to be es-
tablished [9].
The principle of “Synchro“ deep drawing was developed at the IWU-Chemnitz-
Dresden and is explained in [20]. In this process, the blank holder (BH) is sub-
jected to a blank holder force (BHF) oscillating between zero and the maximum
value. The die is attached to the ram, which follows a pre-defined displacement
curve in such a way that the part is drawn, when no BHF is applied. When the die
stops, the BHF rises and the wrinkles in the flange area of the part are flattened
out. Preliminary experiments have shown that the process has potential to achieve
larger drawing depths. The synchronization and optimization of the individual
displacement and force curves is a task for the FE process simulation.
Fig. 17.16 FE simulation of “Synchro” deep drawing [20]
Figure 17.16 shows the FE simulation of “Synchro” deep drawing. The die dis-
placement curve and the BHF oscillation serve as input parameters. Since it is not
possible to apply both displacement and force boundary condition to one rigid
body simultaneously, the BHF movement away from the sheet is realized by ap-
plying a negative BHF. The resulting BH displacement is limited to the desired
maximum distance between BH and sheet by using the option of a rigid body
stopper [21]. The rigid body stopper is synchronized with the die displacement.
Mass scaling is common practice in quasi-static FE simulations to achieve reason-
able CPU times [21].
When simulating “Synchro” deep drawing with explicit FE codes, mass scaling
is not an appropriate measure. Since the dynamic behavior is dependent on the
moved masses and their accelerations, the parts in motion must be modeled with
real masses, which are exposed to the real accelerations. Even selective mass scal-
ing of the sheet metal, which is very light compared with the press ram, is
not feasible since the oscillation of the BH excites the blank to vibrate. The
combination of a long process and high-frequency oscillations leads to extremely

long CPU times. Nevertheless, simulations and experiment with a simple round
cup were in good agreement.

Based on conventional FE process modeling and on an analysis of the relevant ef-
fects of the press, tool and die cushion properties on the forming process an ad-
vanced modeling method was developed.
The simulation results obtained by applying the AFPM and a conventional
model of forming process (without elastic press, tool and die cushion properties)
were compared. The comparison shows that relevant elastostatic effects of the
press, tool and die cushion on the forming process can be visualized with the ad-
vanced process model.
It is shown that the simulation matches the experimental results obtained on an
example tool-press configuration for drawing rectangular cups.
A complete forming process model is available to consider different configura-
tions of presses and die cushions already in the process-planning stage. Now, it
seems to be possible to reduce the effort of tool try-out and ramp-up by using FE
simulations.
In order to facilitate the set-up of an AFPM for the user the software-prototype
xForm was developed. xForm enables the user to choose from a selection of press
and die cushion structures to implement them into the FE forming analysis.
In the field of elastostatic process simulation, the experimental validation of the
tool guidance model will be part of further investigations. An asymmetric tooling
will be used to show the effect of the tool guidance on the final part shape by
means of FE simulation and experiment.
Regarding dynamical responses of the press to a certain process a modal analy-
sis of the press structure without the process model will be conducted. From natu-
ral frequencies and mode shapes of the press parts, reduced dynamical sub-models
will be derived. Subsequently, the sub-models will be connected and appropriate
damping parameters need to be determined.
References
[1] Roll, K., Wiegand, K.: Eine Möglichkeit zur Berücksichtigung der Werkzeugelasti-
zität bei der Blechformsimulation. In: Proceedings of Neuere Entwicklungen in der
Blechumformung Leinefelden-Echterdingen (Stuttgart), MAT INFO Werkstoff-
Informationsgesellschaft mbH, Frankfurt (2006)
[2] Großmann, K., Hardtmann, A., Wiemer, H.: Simulation des Blechumformprozesses
unter Berücksichtigung des statischen Verhaltens der Pressmaschine. ZWF Zeitschrift
für wirtschaftlichen Fabrikbetrieb 101(10), 600–605 (2006)
[3] Großmann, K., Wiemer, H., Hardtmann, A., Penter, L.: Stand der Simulation von
Umformprozessen mit den elastischen Einflüssen aus Maschine und Werkzeug. In:
Proceedings of EFB-Colloquia Neue Wege zum wirtschaftlichen Leichtbau, pp. 185–
198. EFB, Hannover (2007) (in Fellbach)
[4] Großmann, K., Wiemer, H.: Maschine, Werkzeug und Prozess – ein ganzheitliches
System. In: Proceedings of 6th Sächsische Fachtagung Umformtechnik, TU Dresden
(1999)
[5] Großmann, K., Wiemer, H., Thoms, V., Schirmacher, F.: Modellierung des Blechum-
formprozesses in der Anlagensimulation auf Grundlage von Ergebnissen der FEM-
Prozeßsimulation. EFB-Forschungsbericht No. 178. EFB, Hannover (2001)
[6] Großmann, K., Wiemer, H.: Simulation of the Process Influencing Behaviour of
Forming Machines. In: Steel Research International, vol. 76(2/3), pp. 205–210. Ver-
lag Stahleisen GmbH, Düsseldorf (2005)
[7] Schirmacher, F., Wiemer, H.: Kopplung von Prozess- und Anlagensimulation – ein
wesentlicher Teilschritt auf dem Weg zum Virtuellen Umformen. In: Proceedings of
EFB-Colloquia Prozessoptimierung in der Blechverarbeitung, pp. 11.1–11.18. EFB,
Hannover (2001)
[8] Brecher, C., Schapp, L., Paepenmüller, F.: Gekoppelte Simulation von Presse und
Massivumformprozess. Werkstattstechnik Online 96(7/8), 526–529 (2006)
[9] Wiemer, H.: Stand und Möglichkeiten der Systemsimulation von mechanischen
Pressmaschinen. State and Potentials of Forming Press Simulation. Dissertation, TU
Dresden (2004)
[10] German Standard DIN 55189, Ermittlung von Kennwerten für Pressen der Blechve-
rarbeitung bei statischer Belastung. Beuth Verlag, Berlin (1988)
[11] VDI-Guideline VDI 3145, Pressen zum Kaltmassivumformen. Blatt 1: Mechanische
und hydraulische Pressen. VDI-Verlag, Düsseldorf (1985)
[12] Großmann, K., Wiemer, H., Hardtmann, A., Penter, L.: Faster to Sound Parts by Ad-
vanced Forming Process Simulation - Advanced Forming Process Model Including
the Elastic Effects of the Forming Press and Tool. In: Steel Research International,
vol. 78(10/11), pp. 825–830. Verlag Stahleisen GmbH, Düsseldorf (2007)
[13] Pahl, K.J.: Elastische Wechselwirkungen im Ziehapparat einfach wirkender Pressen.
VDI Verlag, Düsseldorf (1994)
[14] Großmann, K., Wiemer, H.: Möglichkeiten der erweiterten Modellierung von Um-
formprozessen zur Berücksichtigung der Wechselwirkungen mit Werkzeugen und
Maschine. In: Proceedings of Neuere Entwicklungen in der Blechumformung. MAT
INFO Werkstoff-Informationsgesellschaft mbH, Frankfurt (2008)
[15] Thoms, V., Voelkner, W., Süße, D.: Methoden zur Kennwertermittlung für Blech-
werkstoffe. EFB-Forschungsbericht No. 187. EFB, Hannover (2002)
[16] Feldmann, P.: Application of strain analysis system AutoGrid® for evaluation of
formability tests and for strain analysis on deformed parts. In: Proceedings of the
IDDRG Conference Györ, Hungary, pp. 21–23 (2007)
[17] European Standard, EI 12004, Metallic materials - Sheet and strip - Determination of
forming limit curves. CEN European Committee for Standardization, Brussels (2007)
[18] Haufe, A., Roll, K., Bogon, P.: Sheet metal forming simulation with elastic tools in
Ls-Dyna. In: Proceedings of Numisheet, pp. 743–748 (2008)
[19] Großmann, K., Wiemer, H., Hardtmann, A., Penter, L., Kriechenbauer, S.: Static
Compensation for Elastic Tool and Press Deformations during Deep Drawing. Prod.
Eng. Res. Devel. 4(2-3), 157–164 (2010), doi:10.1007/s11740-010-0216-7
[20] Großmann, K., Wiemer, H., Neugebauer, R., Mauermann, R.: Simulationsun-
terstützung für das Tiefziehen auf Servospindelpressen. In: Proceedings of EFB-
Colloquia: Bauteile der Zukunft. EFB, Hannover (2010)
[21] Hallquist, J.O., et al.: LS-DYNA Keyword User’s Manual, vol. I, Version 971, Li-
vermore Software Technology Corporation, Livermore California (2007)
Chapter 18
Consideration of the Machine Influence
on Multistage Sheet Metal Forming Processes
B.-A. Behrens, A. Bouguecha, R. Krimm, T. Matthias, and M. Czora
Abstract. Metal forming processes are highly affected by the properties of the
forming machine. In multistage processes, the force path curve of each stage is in-
fluenced by the surrounding stages. This research focuses on the interaction be-
tween the forming machine and multistage processes with respect to the quality of
the workpiece. The accuracy of the numerical results could be increased by coupl-
ing the process simulation with the machine simulation. Further influencing fac-
tors, such as different friction conditions during forming and the elasticity of the
tool, were considered in the simulation. In addition, the thermal properties of the
press were investigated and considered in the machine simulation. Hence, a better
correlation between the numerical results and the experiment could be achieved.
18.1 Introduction
Metal forming processes, such as deep drawing, place high demands on the prop-
erties of the forming machine. Even slight scatters or inaccuracies in the ram mo-
tion can lead to imperfections of the workpiece. The characteristic deformation of
the forming machine and the tool system under load has a crucial influence on the
result. To simulate forming processes, the use of the finite elements method
(FEM), where local influences on the process (e. g. the lubrication or inhomoge-
neous material properties) can be included in the simulation, is state of the art. In-
fluences between the machine and the forming process have not been taken into
account yet. In metal forming simulations, an ideal rigid forming machine is as-
sumed. Thus, differences between simulation and reality occur. Hence, for many
applications it is not possible to manufacture parts with the required accuracy right
from the beginning so tools have to be set up for each machine in a try-out phase.
In order to reduce the start-up time for new metal forming tools it is necessary
to consider the properties of the forming machine in the FE process simulation.
Different approaches for a coupled simulation to consider the interaction between
forming machine and forming process are known. A classification is drawn be-
tween offline-coupling, model integration and co-simulation.
In offline-coupling, the process simulation and the machine simulation are cal-
culated separately. After calculating the process, the resulting force path curve is
404 B.-A. Behrens et al.
transferred into the machine simulation as input data for a new calculation. Subse-
quently, the correction of the ram movement as a result of the machine simulation
serves as a boundary condition for the FEM simulation, if a new calculation by the
two models is performed by means of an iterative loop. In several loops, an ap-
proximation to the real workpiece dimensions can be achieved [1]. The most im-
portant advantage of offline coupled simulations is that both simulations are
implemented in their common software environment. Due to this, both simulations
can benefit from the tools, which are available in the typical software environ-
ment. The disadvantage of this approach is the fact that no dynamic interaction
can be simulated and that the calculation time increases due to the repeated
calculation.
An approach, where the machine and the forming process are simulated in a
common simulation environment, is called model integration [2, 3]. In [4], the
machine structure was reduced and integrated into the FEM process model by
means of elasticity conditions. The tilt and the vertical deflection as well as the de-
formation of the die cushion and the ram were taken into consideration by means
of spring elements. This approach resulted in an improvement of the accuracy of
the forming simulation. However, if the machine properties and the drive control
are examined, this method is limited.
In co-simulation two independent, sufficiently detailed models for the forming
machine and the forming process, which exist in different simulation environ-
ments, are used [5, 6]. These models exchange data after each time-step. Thus, it
is possible to consider dynamic influences within the simulation, which occur dur-
ing the forming process. In [7], a coupled simulation for bulk metal forming
processes was realized. Due to the consideration of the machine properties the ac-
curacy of the forming simulation could be increased. Furthermore, the results of
the forming simulation were compared using machine models with different de-
grees of detail. Even with the least detailed machine model the simulation results
of the workpiece dimensions could be approximated to reality by means of a co-
simulation in comparison to a separate process simulation.
Within this project, an offline-coupling was implemented. The machine model
is based on a phenomenological approach. The process simulation is modeled in a
classic FEM environment. The simulation of single processes is state of the art.
This project deals with the simulation of multistage processes, where interactions
between each stage have to be considered additionally.
18.2 Machine
Experimental investigations were carried out on a high-speed press with a space
rod drive (fig. 18.1). The advantage of a space rod drive is the independent ad-
justment of the ram position and the length of the stroke. The connection rods of a
conventional mechanical press can only be moved in two directions. The connec-
tion rods of a space rod drive have one more degree of freedom. Thus, the kine-
matics of the press result in a slower course of motion shortly before reaching the
bottom dead centre (bdc). This is especially advantageous for deep drawing
processes. The bdc marks the reversal point of the ram.
18 Consideration of the Machine Influence 405
Fig. 18.1 High-speed press Wanzke STA63
18.2.1 Approach
The method to describe the elastic properties of forming machines is based on the
results of experimental tests using a phenomenological approach. The deflections
of the forming machine under process load are measured and reproduced using an
elasticity matrix with variable elements as machine model. The required informa-
tion for the model is provided by means of a single measurement of the press,
based on a static press measurement similar to the standard DIN 55189 [8]. Due to
this fact the model is applicable for many metal forming machines after carrying
out this measurement.
The relative total shift between the ram and the bolster plate results from de-
formations of the connection rods, the ram guide, the press frame, the bolster plate
and the ram as well as the partially non-linear elastical characteristics of bearings
and guidances. This leads to a shift of the position of the bottom dead centre. The
deformations of the press components, except the ram and the bolster plate, can be
summed up as deflection of the ram in relation to the bolster plate including the
tilting. The deflection and the tilting are first calculated in the machine simulation
assuming a rigid bolster plate and ram. The relative total shift is calculated in a
further step by superposing the deflection with the deformation of the bolster plate
and the ram (Fig. 18.2). For employing the principle of superposition it has to be
assumed that superposed loads cause summable deformations. This principle is
applicable as long as there is a linear correlation between the loads and the result-
ing deformations [9]. This approval is valid for very stiff bodies as the ram and the
bolster plate.
Unloaded Loaded
Clamping plates Clamping plates Superposition

rigid elastic
Deflection Tilting Deformation
Ram
Ram { + Bolster plate

=
Bolster plate Deflection per stage
Relative total shift/
Tilting and deflection Cumulative deformation displacement
Modelling:
Stages Stages Stages
Fig. 18.2 Superposition of deflection and deformation
The number of stages of multistage forming processes is an adjustable parame-

ter in the machine model. A load at a single stage not only causes a shift at the
stage itself but also at all other stages.
18.2.2 The Parameterization of the Machine Simulation

18.2.2.1 Press Measurement
The machine model requires the properties of the press as input data. In order to
provide these data it is necessary to determine the properties under a defined load.
The measurement procedure is based on a static press measurement similar to
DIN 55189. Therefore, a load is applied between the ram and the bolster plate via
a hydraulic cylinder. The hydraulic cylinder is placed at the position, where the
stages will be mounted. The total shift of the press is determined at the positions
of the stages by means of displacement transducers. Fig. 18.3 shows the measure-
ment setup of the press for a three-stage process, where the first stage is loaded.
Stage 1 is loaded first. The shift due to the load occurring at the stage itself and
at the other stages is recorded. Afterwards, the procedure for the second and the
third stage is executed in the same way.
The results of each measurement are path-force curves of the ram at each stage.
These are the input data for the machine model, which describe the properties of
the press. Fig. 18.4 exemplarily illustrates the total shift at stage 1 in relation to the
position of the application of load. The figure shows that a load at stage 1 causes
the highest total shift at stage 1, whereas a load at stage 2 of the same amplitude
causes a smaller total shift at stage 1. Correspondingly, a load at stage 3 causes an
even smaller total shift at stage 1. By means of this data the influences of the stag-
es are described. These influences decrease with an increasing distance to the
related stage.
1. Load in stage 1
Loading unit Measurement

of the total shift
in all stages
Hydr. cylinder
Pressure
measurement Hand pump
Fig. 18.3 Measurement setup
1.0 Load position
0.9 Stage 1
0.8 Stage 2
Total shift at stage 1 [mm]
F
0.7
F
0.6
0.5
0.4 Stage 3
0.3
F
0.2
0.1
0.0
0 50 100 150 200 250 300
Load [kN]
Fig. 18.4 Total shift at stage 1 dependent on a load at stage 1, stage 2 and stage 3
18.2.2.2 Investigation of the Temperature-Influence
During the start-up period of a metal forming production line, temperature

changes arise on the components of the press as well as on the metal forming
tools. Changes in the temperature of the press components have an influence on
the properties of the press and therefore on the quality of the manufactured prod-
ucts. To predict such effects by means of a coupled simulation it is essential to ex-
tend the machine model with respect to temperature effects. Hence, the influence
on the static properties of the press due to changes of the temperature was ex-
amined. It could be shown that the position of the bdc is affected by temperature
changes.
To determine the temperature curve of the press, temperature measurements
were executed during the operating time of the press up to a nearly steady thermal
condition. Apart from the temperature curve of the press, its influence on process
parameters, such as forming force and the position of the bottom dead centre, was
investigated. In order to quantify and consider the thermal influence in the ma-
chine model, parts were formed after different operating intervals and the force
path curve of these metal forming processes was captured for the different thermal
conditions at each operating interval. Therefore, piezo-electric load cells were in-
tegrated into the tool stages and thermocouples were applied to relevant press
components (Fig. 18.5). According to Fig. 18.5, for a permanent number of
strokes of 90 rpm the system reaches a nearly steady thermal condition after eight
hours of operation.
Stroke rate: 90 rpm

Press operating time: 8 hours
Oil tank left Oil tank right Workpiece: axisymmetric / material DC04
Measuring equipment:
- Temperature thermocouples
- Stage forces piezoelectric load cells
- Ram stroke incremental displ. transducers
Con-rod left Con-rod right Results: Temperature

Results: Temperatureprofile
course
Ram 38.0
35.3
Temperature [°C]
Guidance left Guidance right

32.7
30.0
27.3
24,7
Reference
22.0
0 1 2 3 4 5 6 7 8 9
1.Stage 2.Stage 3.Stage Press operating time [h]
Deep drawing Contouring Embossing
Con-rod left Oil tank left Guidance left Ram
Con-rod right Oil tank right Guidance right
Fig. 18.5 Experimental setup and temperature curve
In order to provide a general comparison between the force-time curve and the
temperature-time curve, the temperature of the right connection rod is shown in
Figure 18.6. The right connection rod has a direct influence on the embossing
stage as the process force at the embossing stage is mainly transferred by the right
connection rod. Between the temperature-time curve at the right connection rod
and the curve of the embossing force at different operating times correlations can
be identified. Both measurement curves have a high gradient at the beginning,

which decreases with time. Thus, the curves converge towards a certain steady
state. This is related to the fact that the punch not only forms the sheet metal but it
also pushes against the bottom of the die. Thus, the force at stage 3 increases as
the expansion of the connection rod leads to a higher force on the bottom of the
die, which possesses a high stiffness and a correlation can be seen (Fig. 18.6, bot-
tom right).The values of the process force at different operating times at the deep
drawing stage and at the contouring stage do not change with the measured in-
crease of temperature (Fig. 18.6). These values mainly depend on the material
characteristics of the sheet metal. The increase of the temperature for the deep
drawing stage and the contouring stage results in a shift of the bdc since the dies
are open in forming direction. As a consequence, the forming path increases and
affects the workpiece geometry.
Stage 1 Stage 2 Stage 3
Deep-drawing
Deep-drawing Contouring
Contouring Embossing
Embossing
Deep-drawing force Temperature of the right con-rod

Contouring force Embossing force
62.50 77.6 38.00
Deep drawing force [kN]
35.0
Embossing force [kN]
Contouring force [kN]
61.75 36.75
Temperature [°C]
77.4 32.8
61.00 77.2 35.50 30.7
60.25 77.0 34.25 28.5
59.50 76.8 33.00 26.3
58.75 76.6 31.75 24.2
58.00 76.4 30.50 22.0
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
Press operating time [h] Press operating time [h]
Fig. 18.6 Maximum forces at the forming stages with increasing operating time
The differences of the position of the bdc due to the increase of the temperature
of the press can be measured by means of displacement transducers. The changes
of the bdc are affected by the expansion of the press components due to the in-
creased temperatures. The initial condition of the “cold” operating state is taken as
a reference. The difference between the bdc and the reference bdc defines the total
shift caused by changes in the temperature profile of the machine. Fig. 18.7 illu-
strates the position of the bdc with respect to the operating time for the stages deep
drawing, contouring and embossing. It was measured by incremental displacement
transducers.
Deep drawing stage Contouring stage Embossing stage
a 76.30 b
Position of the bdc [mm]

76.25
22.5°C
22.5°C 36°C 34.5°C
76.20
76.15 27°C
76.10 36°C 38°C
76.05
76.00
0 1 2 3 4 5 6 7 8 9
a Press operating time [h] b
Fig. 18.7 Development of the shift of the bottom dead centre (bdc) related to the operating
time of the press
Fig. 18.7 shows that the shift of the bdc of all three stages is similar and con-
verges exponentially with time. The highest gradients of the shift curve in position
of the bdc occur during the start-up period of the press. As the operation time in-
creases the gradients of the shift curve in position of the bdc decrease until a near-
ly steady thermal condition is reached. This result corresponds to the temperature
measurements at the press components (Fig. 18.5). Thus, for each stage a shift of
the position of the bdc in forming direction occurs and converges towards a cer-
tain value with increased operating time.
Differences in the position of the bdc between initial condition at the start-up
and the steady thermal condition of the press after eight hours occur at all three
stages. At the deep drawing stage, the difference is 0.08 mm, 0.05 mm at the con-
touring stage and 0.03 mm at the embossing stage. Although these differences
mainly result from the temperature increase they also depend on the process. The
differences in the shift of the position of each bdc in relation to the position of the
bdc in “cold condition” are 0.1 % (deep drawing), 0.06 % (contouring), 0.04 %
(embossing). Thus, for the machine model a temperature coefficient of 0.04 % for
the steady thermal condition of the press was implemented. This value improves
the accuracy of the simulation of the press for all investigated processes by use of
the minimal requirements for the embossing stage. Therefore, no further settings
are required by the user.
18.3 Modeling of a Three-Stage Process

For the experiments, a modular tool set for a three-stage forming process was de-
signed. The three single tools contain forming processes, which are common in
multistage sheet metal forming: these are deep drawing, contouring and emboss-
ing (Fig. 18.8). For the deep drawing stage and the contouring stage spring-
actuated blank holders are applied. For the final forming of the part all forming
stages have to be run through. Every single stage causes a characteristic force path
curve. If several individual tools are mounted under one ram, the curves change
for each stage since the press deforms and the single stage gets influenced due to
Fig. 18.8 Design of the examined forming stages deep drawing, contouring and embossing
forces of other stages, which are applied simultaneously. The stages can be
mounted into the press and operated individually as well as together.
The design of the three-stage tool system was supported by a process simula-
tion. Therefore, simulation models based on the CAD data of the tools were gen-
erated and numerical experiments by means of the finite elements method were
executed. Fig. 18.9 illustrates the three simulation models for the forming stages
deep drawing, contouring and embossing.
Fig. 18.9 Finite elements models of the three forming stages

The tools are designed to form sheet metals with a thickness of 1.0 mm. The re-
sults of the simulation show that this design leads to the required utilization of 80
% of the nominal press force, if a DC 04 deep drawing steel is used. Within the
simulation, the friction was set to a constant value of μ = 0.16. In the progression
of the project, a new friction law to consider the dependency of the friction on the
smoothing pressure and the contact pressure was applied.
In a further step, the drawing die was modeled elastically. The deep drawing
stage is illustrated in Fig. 18.10. The punch and the blank holder were still mod-
eled with rigid shell elements. Only the die was modeled with elastically-solid
elements to prevent an increasing simulation time. The elastic properties of the die
were taken into account in order to simulate the varying local contact conditions.
This allows an accurate calculation of the contact pressure.
Forming
direction Punch
(rigid)
Blank holder
(rigid)
z Die
x (elastic)
y
Fig. 18.10 Finite elements model of the deep drawing stage with the elastic die
For an improved description of the friction between the tool and the workpiece
during a sheet metal forming process the evolution of the surface conditions of the
workpiece has to be considered. The influence of the changing surface topology
on friction conditions has to be included in the FE simulation. The contact pres-
sure leads to a smoothing of the surface topology. The smoothed surface results in
a reduced friction coefficient. For this reason, a concept for the description of a
technical surface structure was used [11]. A workpiece with different smoothed
regions is shown in Fig. 18.11.
Strong smoothing
Weak smoothing
Fig. 18.11 Drawn cup with different degree of smoothed regions
A friction law µ = (PE, PK), where the friction coefficient µ is a function of the
smoothing pressure PE and the contact pressure PK, is implemented in
ABAQUS/Explicit via a user subroutine. This friction law considers the local sur-
face pressure as well as the transformation history of the sheet metal surface.
The difference in contact pressure at the contact surface leads to a different
component smoothing. The high pressures at the die radius (Fig. 18.12) contribute
to different friction conditions during the forming process.
35.0
Contact pressure [MPa]
15.0
0.0
Fig. 18.12 Contact pressure distribution at the die
18.4 Coupling of the Machine Simulation and the Process

Simulation
The “offline-coupling” method mentioned above was applied to couple the ma-
chine simulation and the process simulation. The two simulation environments run
sequentially and the entire cycle iterates. Fig. 18.13 illustrates the progression of
the coupled simulation between the process simulation in ABAQUS and the ma-
chine simulation in Matlab.
2. Step
Time-depending
Initial condition
course of the
Approximated
forming force
time-depending
course of motion 1. Step 3. Step
of the ram Process Machine
(forming machine
simulation simulation
ideal rigid)
Abaqus
Matlab
4. Step
New
cycle Convergence
criterion
reached? Correction of
Break the ram stroke
Fig. 18.13 Progression of the coupled simulation
A cycle of the coupled simulation contains four sequentially running steps. The
ram motion at each stage, which is necessary for the required part depth assuming
an ideal rigid press, serves at first as boundary condition for the process simula-
tion. The process simulation calculates each forming stage one by one and then
provides the related coordinates for the nodes of the mesh of the tool punch, the
applied node forces and the related forming times in a text file (ASCII format).
The text file with the information about the force-time curves of the punch is sent
to the machine simulation in Matlab in a second step. By means of this input data
the relative shift between the bolster plate and the ram can be determined by the
machine simulation. The output of the machine simulation consists of way-time
curves for the process simulation, considering the deflection and the deformation
of the machine components. This result is then sent back to the process simulation
and serves as new input data for the process simulation. This process is repeated
until a certain convergence criterion is reached. The convergence criterion consists
of the comparison of the difference between the forming paths of two succeeded
iterations. This value was set to 0.01 mm. Fig. 18.14 illustrates a simulation,
where the convergence criterion was reached after three iteration cycles.
Calculated part height U [mm]

Convergence criterion: |U -U |≤ 0,01 mm
n+1 n
Deep drawing
26,01 mm
26,01 mm
25,99 mm
26,75 mm
26,00 mm
25,00 mm
26,85 mm
24,61 mm
26,20 mm
24,62 mm
24,61 mm
26,23 mm
Contouring
Embossing
0 1 2 3 Iterations
Fig. 18.14 Results after each iteration cycle of a coupled simulation
The data exchange is executed by means of a Matlab function and an external,

self-running file. The calculation time of the coupled simulation with ABAQUS
6.7 and Matlab on an Intel Xeon 3.4 GHz processor with 2 GB RAM takes about
20 hours including the enhancements mentioned above (friction model, elastic die
model, temperature factor) of the enhanced coupled simulation.
18.5 Validation of the Coupled Simulation

In order to verify the results of the coupled enhanced simulation they were com-
pared with experimental results (Fig. 18.15) and the results of the previous
coupled simulation regarding geometrical properties (Table 18.1). In contrast to to
the previous simulation, the enhanced coupled simulation considers the elastic die,
the improved friction law and the temperature influence of the machine. The re-
sults of the previous coupled simulation model are presented in [12].
Fig. 18.15 Experimental and simulated height of the part after the deep drawing stage
The height of the formed parts was measured with the optical measurement sys-
tem GOM ATOS II. In Table I, the results are compared and it is shown that the
results of the coupled enhanced simulation could be improved in comparison with
the enhanced decoupled and the previous coupled simulation. Using an elastic tool
model, an improved friction law and the temperature factor leads to a more realis-
tic simulation but the simulation time is more than six times higher than the pre-
vious coupled simulation.
Table 18.1 Comparison of the experimental and the simulated results
– – Simulation – Part height

time – [mm]
– [min]
– – – Stage1 – Stage2 – Stage3
– Experiment – – 24.44 – 25.92 – 25.86
– Previous coupled – 170 – 24.61 – 26.01 – 25.99
simulation
– Decoupled en- – 390 – 24.90 – 26.10 – 26.10
hanced simulation
– Coupled enhanced – 1240 – 24.55 – 25.98 – 25.92
simulation
18.6 Conclusion
The numerical simulation of forming processes is an established method for the
analysis of common metal forming processes. By this method, the interactions be-
tween the process and the machine are not taken into consideration. Within the
experiments in the frameworks of this project it was shown that the machine has
an impact on the process and that this impact also affects the geometry and the
quality of the manufactured parts.
By means of the described coupling of the process simulation with a machine
simulation it became possible to take into consideration the influences of the ma-
chine. These influences mainly result in the shift of the bottom dead centre caused
by the deformation and the deflection of the ram as well as the tilting of the ram.
To improve the accuracy of the dimensions of the simulated workpieces these
influences have to be considered in the boundary conditions of the process simula-
tion. Therefore, an offline-coupling was used and validated by the results of expe-
rimental data. It was shown that by means of this offline-coupling the simulation
results approached the real geometrical dimensions of the workpieces.
With an increasing degree of detail of the models and an enhancement by fur-
ther influencing parameters, more realistic results with a higher compliance were
reached. However, these lead to higher calculation times and to the necessity of
further computing resources. It could be shown that the total shift of the bottom
dead centre for the analyzed workpiece is less than 1 mm. Particularly the thermal
influence of the machine on the process is very small. The differences of the posi-
tion of the bottom dead centre concerning a change in the temperature profile of
the machine range around a few hundredths of a millimeter.
Due to the fact that the influence of a forming machine on a forming process is
nominal, the coupled simulation has especially advantages for the tool set-up in
forming machines for tools including embossing processes. The tolerances of the
height of an embossing punch are often limited to hundredths of a millimeter.
References
[1] Bäcker, V., Klocke, F., Wegner, H., Timmer, A., Grzhibovskis, R., Rjasanow, S.:
Analysis of the Deep Rolling Process on Turbine Blades using the FEM/BEM-
Coupling. In: IOP Conf. Series: Materials Science and Engineering, vol. 10 (2010)
[2] Großmann, K., Hardtmann, A., Wiemer, H., Penter, L.: An Advanced Forming
Process Model including the Interactions between Machine, Tool and Process. In:
Proceedings of 1st International Conference on PMI, Hannover, Germany, pp. 125–
132 (2008)
[3] Bogon, P., Roll, K.: Ein Ansatz zur Berechnung und Kompensation der elastischen
Werkzeugdeformation bei Ziehwerkzeugen. In: 30. EFB-Kolloquium Blechverarbei-
tung 02./03. März, Bad Boll (2010)
[4] Großmann, K., Wiemer, H.: Feasibilities of Advanced Forming Process Modelling
Considering the Interactions with Dies and Machines. In: Proceedings of Conference
New Developments in Sheet Metal Forming Technology, Fellbach bei Stuttgart,
Germany, pp. 311–350 (2008)
[5] Lohse, H., Marthiens, O., Helduser, S., Matthias, T., Behrens, B.-A.: Reglerauslegung
für hydraulische Tiefziehpressen. In: Ölhydraulik und Pneumatik. 3/2010/März, pp.
S.2–9. Vereinigte Fachverlage GmbH, Mainz (2010)
[6] Behrens, B.-A., Marthiens, O., Matthias, T., Helduser, S., Lohse, H.: Antriebs- und
Prozessoptimierung hydraulischer Tiefziehpressen mit Hilfe der gekoppelten Simula-
tion. EFB-Forschungsbericht NR. 291, Europäische Forschungsgesellschaft für
Blechverarbeitung (2009) ISBN 978-3-86776-325-7
[7] Schapp, O.: Gekoppelte Simulation von Maschine und Prozess in der Massivumfor-
mung. Dissertation, RWTH Aachen (2008)
[8] NN, DIN 55189, Ermittlung von Kennwerten für Pressen der Blechverarbeitung bei
statischer Belastung. Beuth-Verlag, Berlin (1988)
[9] Pestel, E.: Technische Mechanik. Band 1. BI Wissenschaftsverlag (1988) ISBN: 3-
411-03153-0
[10] Krimm, R.: Berechnung der lastabhängigen Maschinenauffederung zur Verkürzung
der Anlaufzeit neuer Transferwerkzeugsätze. Dissertation, Leibniz Universität Han-
nover (2006) ISBN: 3-939026-07-7
[11] Behrens, B.-A., Sabitovic, A.: Modelling of friction in deep drawing considering re-
versible sheet changes. In: Proceedings of the IDDRG Conference, Olofström, Swe-
den, pp. 125–132 (2008)
[12] Behrens, B.-A., Matthias, T., Czora, M., Poelmeyer, J., Ahrens, M.: Improving the
accuracy of numerical investigations of multistage sheet metal processes by coupling
a process FE analysis with the machine simulation. In: Proceedings of 1st Interna-
tional Conference on Process Machine Interaction, Hannover, Germany, pp. 133–138
(2008)
Chapter 19
Optimization of Tool and Process Design for the
Cold Forging of Net-Shape Parts by Simulation
T. Kroiß and U. Engel
Abstract. As a result of the research work in this project, a comprehensive ap-

proach is presented for the consideration of the interactions between process, tool
and machine in the FE-based design of cold forging tools and processes: The ap-
proach comprises an efficient determination of the deflection characteristic of
press and tooling system and its subsequent condensed modeling in combination
with the FE simulation of a cold forging process. Then, based on a set of simula-
tions, a parametric process model is developed. It permits an optimization of the
values of influencing parameters to achieve high workpiece accuracy considering
the interactions. By acquiring and, afterwards, applying knowledge on the process
behavior, the required number of simulations for the parametric process model and
the optimization can be reduced considerably. The approach can be completed by
using the parametric process model for estimating scatter and uncertainties of tar-
get values depending on those of the influencing parameters.
19.1 Introduction
Cold forging offers the potential to manufacture, in large quantities, parts with a
high dimensional accuracy and surface quality as well as high strength due to
work hardening. Thus, the industry increasingly demands high strength cold
forged parts, which need no machining after the forming process. This kind of net-
shape production can help to save machining resources and material costs. For this
reason, achieving Net-shape accuracy has been an important target in recent re-
search activities on cold forging [1].
However, the high flow stress of the workpiece materials results in high form-
ing loads. As a consequence, the whole system, which consists of press and tool-
ing system, is subjected to considerable displacements and elastic deflections,
which can cause deviations in the workpiece dimensions. These interactions be-
tween process, press and tooling system make it difficult to achieve a high work-
piece accuracy solely based on process simulation [2]. Hence, achieving high
workpiece accuracy without extensive adjustment at the production start requires
the consideration of the elastic behavior of the whole system at an early stage, i. e.
within the phase of laying out the process in the FE simulation.
420 T. Kroiß and U. Engel
This can be done with a combined FE modeling of the cold forging process and
the characteristic of press and tooling system. Based on such a comprehensive, in-
tegrated simulation model, it would be possible to determine an adjusted process
and tool design so that the displacements and deflections of press and tooling sys-
tem are compensated. However, to permit a systematical determination of the ap-
propriate influencing parameters without try-out on the level of FE simulation, the
simulation model should be transformed into a parametric model of the entire
process behavior. This can be the base for a computer-aided optimization of the
values of parameters with influence on the workpiece dimensions so that high
workpiece accuracy can be achieved considering the interactions.
In the following, an approach is presented for the consideration of the men-
tioned interactions in the tool and process design by FE simulation through a
combined modeling of process, press and tooling system. To obtain a quick and, at
the same time, exactly enough combined modeling, tailored methods for the de-
termination and modeling of the entire deflection characteristic were developed in
a comprehensive approach. Built up on a set of simulations with the integrated
model under variation of several influencing parameters, a parametric process
model including an optimization algorithm was developed. To reduce the neces-
sary number of simulations for the generation of the parametric process model,
process knowledge was acquired with regard to the specific behavior of common-
ly used cold forging processes. The approach was developed for cold forging
processes in combination with stroke-controlled and force-controlled presses
respectively.
19.2 Characteristics of Press and Tooling System

For the production of cold forging parts, stroke-controlled as well as force-
controlled presses are commonly used. Apart from the deflection of the tooling
system under process load, the elastic behavior of the press frame itself can have
an influence on the accuracy of the cold forging process. The modeling of these
deflections and their influence on the workpiece dimensions has been subject of
several recent research projects in bulk and sheet metal forming [2-5]. On a
stroke-controlled press, these deflections lead directly to a reduced actual punch
stroke and, in turn, to deviations in the workpiece dimensions. In contrast, the def-
lection of the frame of a hydraulic press in force-controlled mode can be compen-
sated by the hydraulic control. However, the deflections of the tooling system
itself affect the workpiece dimensions in both cases. On force-controlled presses,
mechanical end stops additionally influence the effective punch position and thus
the workpiece dimensions. The press parameter tilting stiffness is neglected in the
following investigations as the presses are loaded centrically and a pillar die set is
used, which avoids tilting.
The measurement of the deflection characteristic of a press can be carried out
according to DIN 55189 [6]. This standardized measurement is carried out with a
static load and the ram of the stroke-controlled press at the bottom dead center
(BDC). Apart from this method, further approaches for measuring the press
19 Optimization of Tool and Process Design 421
behavior were developed [7, 8]. Stroke-controlled presses usually show a non-
linear behavior with a clearance due to initial effects of the presses’ drive system
and a nearly constant stiffness at higher load. However, the deflection not only
affects the press itself but also the tooling system up to the punch and the pre-
stressed die. The deflection of these components leads to a change of the punch
position influencing the workpiece dimensions. To represent this behavior in the
FE process simulation, the deflection of all these mentioned press and tool com-
ponents has to be modeled.
19.3 Methodical Approaches for the Determination and

Modeling of Deflection Characteristics
19.3.1 Tooling System with Stroke-Controlled Press

To determine the deflection characteristic of the whole system consisting of tool-
ing system and stroke-controlled press and to integrate it into the FE process simu-
lation, the following methodical approach was developed. As a basic condition it
is known that actually only the axial stiffness is relevant as the press is loaded cen-
trically and a pillar die set is used to avoid tilting. The two major elements sub-
jected to deflection are the stroke-controlled press and the tooling system. The
press characteristic is determined once only in an experiment as it remains nearly
constant over the time, apart from wear, etc. The characteristic of several available
presses can be stored in a data base. In contrast, the tooling system mounted on the
press can change frequently. To avoid experimental investigations on various tool-
ing systems and to consider the tooling system characteristic anyhow before its
production in the phase of laying out the forming process, its deflection characte-
ristic is determined by simulation. For this purpose, the tooling system is split up
into assemblies. These components are modeled with elastic bodies in the FE si-
mulation and then loaded according to the real forming process. In the case study
in Section 19.4.1, the components of the tooling system are reduced to 2D axially-
symmetric models in the FE simulation program simufact.forming. Based on the
simulation results, load-deflection curves, which usually show a nearly linear cha-
racteristic, are generated. Their gradient represents the constant stiffness of the re-
spective assembly. A single value as substitute quantity can then be calculated for
the whole tooling system from the stiffnesses of the assemblies. The stiffness of
the press and the whole tooling system can also be combined to an equivalent
stiffness. This stiffness is incorporated into the FE process simulation as a spring
element. The press clearance is considered as a constant shift of the punch. Thus,
the combined modeling is implemented in one single simulation environment. The
condensed modeling by spring elements only causes a negligible increase in com-
putation time. In Section 19.4.1, the approach is carried out for a full forward ex-
trusion process on a stroke-controlled press.
19.3.2 Tooling System on Force-Controlled Press

For the determination of the deflection characteristic of the tooling system with
mechanical end stops on a force-controlled press and its modeling in the FE
process simulation, a methodical approach, similar to the FE-based analysis of the
tool assemblies on a stroke-controlled press, was developed. To be independent of
experimental investigations the deflection characteristic is also determined by FE
simulation. However, the setup is different:
The tooling system investigated in the case study in Section 19.4.2 is illustrated
in Figure 19.1. It is based on a pillar die-set. The BDC of the press is limited with
two steel end stops. At the end of the forging process, the top plate of the die-set
touches the end stops with a certain force depending on press force and forming
load. Modeling the tooling system characteristic requires the consideration of two
different stiffnesses for the end stops and the components in the flow of the punch
force respectively. Hence, the tooling system is split up into two superior assem-
blies, which are investigated separately: the mechanical end stops loaded by the
entire press force and the tool components loaded by the forming force.
top tool plate

load cell
punch end stop
end stop die
bottom tool plate
Fig. 19.1 Tooling system with steel end stops on force-controlled press [9]
Furthermore, the tool components affected by the forming load are split up into
the top and the bottom tool plate. Again, modeling the mentioned components in
the FE simulation and loading them according to the real forming process leads to
a constant stiffness for each assembly. Both a reduced 2D axially-symmetric mod-
el and two 3D models were built in the case study in Section 19.4.2 using the FE
simulation program simufact.forming. The stiffness of the end stops can be direct-
ly incorporated into the FE process simulation as a spring element. For the tool
components in the flow of the forming force, an equivalent value as substitute
quantity is calculated based on the stiffnesses of the top and bottom assemblies.
This substitute stiffness is incorporated into the FE process simulation as a spring
element, too. In Section 19.4.2, the approach is carried out for a backward can ex-
trusion process on a force-controlled press.
19.4 Case Studies
19.4.1 Full Forward Extrusion on Stroke-Controlled Press

Experimental Determination of Press Characteristic
The investigations were carried out on a stroke-controlled press Komatsu-

Maypress MKN 63/6A. The press stroke is 60 mm, the cycle time is 0.75 s. The
press has an open C-frame. The load-deflection curve of the press in axial direc-
tion was determined according to DIN 55189-1 [6]. Load and deflection were
measured with a load cell and a travel sensor respectively. The maximum test load
was 375 kN. Six load cycles were performed. The resulting curve is illustrated in
Figure 19.2 a. It has a non-linear behavior with an initial clearance due to the
presses’ drive system and a nearly constant stiffness at higher load. The Figure
contains all six measurements showing a very low scatter. The press stiffness is
Cpress = 265 kN/mm. An extrapolation of the linear behavior at high loads to the
intersection with the ordinate results in a clearance of 0.4 mm [10].
Experimental FE simulation
setup models
2.0
top tool plate
mm
1.6 load cell
1.4
Vertical deflection
1.2 punch adapter for

1.0 load cell
0.8 Cpress = 265 kN/mm
0.6
0.4 prestressed die
0.2 clearance = 0.4 mm
insulating plate bottom tool plate
0
0 50 100 150 200 250 300 kN 400 support ring
Force
a. b.
Fig. 19.2 Deflection characteristic of Komatsu-Maypress MKN 63/6A (a); Tooling system
– Experimental setup / FE simulation models of top and bottom assembly (b) [10]
FE Modeling of Tooling System Characteristic
The tooling system for the full forward extrusion process in consideration is illu-
strated in Figure 19.2 b. The left side of the figure shows the experimental setup.
The tooling system is supported by two bars and a ring at the center. The support
ring avoids deflection of the bottom tool plate. The pillar die-set contains a load
cell to get the forming loads in the experimental investigations. The die for the full
forward extrusion process, pre-stressed by two shrink rings, has a nominal initial
diameter of 12 mm, which is reduced to 6 mm after the shoulder. The die shoulder
has an angle of 45°. To determine the load-deflection curves in simulation, the
system is divided into two assemblies, which include the parts mounted on the top
and the bottom plate of the tooling system. Figure 19.2 b right shows the elements
of the simulation models. The two FE models contain all parts of the tooling sys-
tem in the flow of the forming force.
The simulation model of the top includes the tool plate, the load cell and the
adapter for the load cell. The model at the bottom contains the tool plate, the sup-
port ring, the support for the ejector and an insulating plate (E = 2,550 N/mm2).
The simulation models are 2D axially-symmetric. This simplification is feasible as
the plates are solely compressed due to contact with axially-symmetric parts with-
out bending. The two assemblies are loaded through rigid bodies, which have the
shape of the punch head (top) and the pre-stressed die (bottom).
The stiffnesses of the punch and the pre-stressed die are not included in this
calculation as they are considered as elastic bodies in the FE process simulation.
The gradient of the displacement of the rigid contact bodies vs. the load represents
the stiffness of the respective assembly. The resulting linear load-displacement
curves are shown in Figure 19.3. The assembly of the top tool plate has
a stiffness of Ctool_top = 1,851 kN/mm, the bottom plate a stiffness of
Ctool_bottom = 5,336 kN/mm. As the two stiffnesses behave like two serially ar-
ranged spring elements, their reciprocals add up. This leads to a stiffness of
Ctool = 1,374 kN/mm for the whole tooling system.
600
Ctool_bottom = 5336 kN/mm
kN
400 Ctool_top = 1851 kN/mm
300
Force
200
C tool = 1374 kN/mm
100
0
0 0.05 0.10 0.15 0.20 mm 0.30
Deflection
Fig. 19.3 Determination of stiffness of tooling system on stroke-controlled press

by load-displacement curves [10]
The stiffnesses of the tooling system and the press are serially arranged as well.
Hence, the reciprocals of Ctool = 1,374 kN/mm and Cpress = 265 kN/mm add up to
the reciprocal of the whole system. This leads to a stiffness for the whole system
of Cpress_tool = 222 kN/mm [10].
Integrated FE Simulation Model
The deflection characteristic was incorporated into the FE simulation of the full
forward extrusion process (Fig. 19.4 a). The full forward extrusion process was
modeled 2D axially-symmetric in the FE simulation program simufact.forming. It
is a combined simulation of workpiece, die and the pre-stressing system of the die.
Thus, the model includes the influence of the radial deflection of the die on the
workpiece dimensions. The punch is also an elastic body. Its stroke (60 mm) is
applied by a rigid body at the top of the elastic punch.
Fig. 19.4 Full forward extrusion on stroke-controlled press – Integrated FE simulation

model (a); photo of workpiece with specified target values (b) [10]
Cpress_tool was incorporated into the FE process simulation as a spring element. It

is located at the bottom of the pre-stressed die reducing the actual punch position
when loaded. The press clearance is considered with a constant shift of 0.4 mm.
The punch is moved away from the die by this value [10].
Experimental Validation
To validate the integrated FE model, full forward extrusion experiments on the

Komatsu press were compared with simulation results regarding the workpiece
length (Fig. 19.4 b). The length directly depends on the actual punch position. In
addition, simulations were performed with a rigid FE process simulation model
without the characteristic of press and tooling system. To evaluate the influence of
different forming loads and, as a consequence, different deflections on the work-
piece lengths, the experiments were carried out with the two steels C35 (1.0501)
and 20MnB4 (1.5525), which noticeably differ in their flow stresses. C35 has an
about 20 % higher flow stress than 20MnB4. In the FE simulations, the flow stress
data of the two workpiece materials was used, too. The friction factor for the si-
mulation was identified numerically with the help of the measured forming loads
in BDC. The cylindric specimens have a length of 12 mm and a diameter of
11.6 mm. In the experiments, 16 specimens of 20MnB4 and 20 of C35 were
formed at room temperature. They were coated with zinc phosphate and molybde-
num disulfide.
The workpiece lengths for the experiments and simulations are shown in
Figure 19.5. The average lengths of the experiments are 22.81 mm (+0.11 mm,
-0.15 mm) for 20MnB4 and 21.73 mm (+0.18 mm, -0.21 mm) for C35. The differ-
ence in length between 20MnB4 and C35 is 1.08 mm. The workpiece lengths re-
sulting from the integrated FE simulation model are 22.83 mm for 20MnB4 and
21.75 mm for C35. Hence, the difference in length between the two workpiece
materials is also 1.08 mm. The deviation compared to the experiments is only 0.02
mm for both materials. Only a small difference in average forming loads in the
BDC of 3 kN (20MnB4) and 6 kN (C35) can be observed compared to the expe-
riments. The process simulations with the rigid model lead to similar forming
loads. However, the resulting workpiece lengths of 25.90 mm (20MnB4) and
25.57 mm (C35) differ considerably from the experiments. The respective devia-
tions in length are 3.09 mm (20MnB4) and 3.84 mm (C35). The pure process si-
mulation (0.33 mm) cannot sufficiently represent the difference in length between
the workpiece materials compared to the real process (1.08 mm). The integrated
FE simulation model is able to accurately represent the interactions between forg-
ing process, tool and machine with influence on the workpiece dimensions. The
modeling of the whole characteristic with a shift and a substitute stiffness leads to
a reduction of the deviation between experiment and simulation from up to 3.84
mm (17.7 %) to 0.02 mm (0.1 %) compared to the rigid simulation [10].
27
20MnB4 25.90 C35 Experiment
mm 25.57
Integrated FE
25
Workpiece length
simulation model
24 FE process simulation
23 22.81 22.83
22 21.73 21.75
21
0˜
149 kN 146 kN 152 kN 205 kN 199 kN 208 kN
Fig. 19.5 Workpiece lengths depending on workpiece materials and forming force in BDC;
comparison of experiments to Integrated FE model and pure process simulation [11]
19.4.2 Backward Can Extrusion on Force-Controlled Press

The tool and punch layout as well as the resulting workpiece geometry of the
backward can extrusion example process are shown in Figure 19.6 a. The punch
has an undercut so that the workpiece does not touch the punch behind the form-
ing zone. The punch diameter in the forming zone is 8.5 mm. The die consists of
the full forward extrusion die with a backing device.
Fig. 19.6 Backward can extrusion process (a); Integrated FE simulation model of backward
can extrusion process on force-controlled press (b) [9]
FE Modeling of Tooling System Characteristic
For the determination of the load-deflection curves in simulation, the investigated

tooling system is divided into three assemblies, which include the parts mounted
on the top and the bottom plate of the tooling system, respectively, as well as the
end stops. The FE simulations were performed in simufact.forming. The two FE
models of top and bottom tool plate contain the parts of the tooling system in the
flow of the forming force. The simulation model at the top (Fig. 19.7 a) includes
tool plate, load cell, load cell adapter and spacer plates. To reduce the simulation
effort, a 2D axially-symmetric FE model was used. The model at the bottom con-
tains tool plate, spacer plate and two bars, which the tool plate is mounted on
(Fig. 19.7 b). The bolted joints between tool plate and bars are considered with
boundary conditions in simulation. As the bars cause a bending of the bottom tool
plaste when loaded, a 3D FE model was created for this assembly. The 3D model
of the end stops consists of top and bottom tool plate, end stops and the bars as
well (Fig. 19.7 c). The three assemblies are loaded through rigid bodies with the
shape of the punch head (top), the pre-stressed die (bottom) and the press ram re-
spectively (end stops) in the effective direction of the press.
The stiffness of punch and die is not included in the simulations as they are
considered as elastic bodies in the FE process simulation. The gradients of the
load versus the displacements of the rigid contact bodies in press direction lead to
the following stiffnesses for the three assemblies. The stiffness of the end stops re-
sulting from the gradient of the linear curve is Cstop = 2,970 kN/mm. Figure 19.8
shows the stiffness of the assemblies of top and bottom tool plate. The top
tool plate has a stiffness of Ctool_top = 2,189 kN/mm, the bottom plate a stiffness
of Ctool_bottom = 4,218 kN/mm (curve bent at one increment due to contact determi-
nation issues). The two latter stiffnesses behave like serially-arranged spring
Fig. 19.7 FE models of top tool plate (a), bottom tool plate (b) and end stops (c) [9]
elements. Therefore, their reciprocals add up. The resulting stiffness is

Ctool = 1,441 kN/mm for top and bottom tool plate. Ctool and Cstop, describing the
deflection characteristic of the whole tooling system, can be incorporated into the
FE simulation of the backward can extrusion process [9].
1400
kN
C tool_bottom = 4218 kN/mm
1000
800 Ctool_top = 2189 kN/mm
Force
600
400
200
C tool = 1441 kN/mm
0
0 0.05 0.10 0.15 0.20 0.25 mm 0.35
Deflection
Fig. 19.8 Determination of stiffness of tooling system on force-controlled press by

load-displacement curves [9]
Integrated FE Simulation Model
The backward can extrusion process was modeled 2D axially-symmetric in the FE

simulation program simufact.forming according to the full forward extrusion
process. The punch is modeled as an elastic body. Its stroke is applied by a rigid
body at the top of the punch. Ctool was incorporated into the FE process simulation
as a spring element, located at the bottom of the die. The end stop (Cstop) is mod-
eled with a spring element and a rigid contact body. The rigid punch hits the end
stops at the end of the forming process. After reaching a certain force limit when
compressing the end stops, the simulation is stopped and the workpiece is ejected.
The integrated FE simulation model is shown in Figure 19.6 b [9].
Experimental Validation
The results from the integrated FE simulation model were compared to backward
can extrusion experiments. To be able to check the correctness of both stiffnesses
Cstop and Ctool of the FE simulation model, the influencing parameters press force
(force on end stops) and forming force were varied in the experiment as well as in
the simulation. In the experiment, the press force is varied by the press control.
The forming force is varied by using two different workpiece materials 100Cr6
(1.3505) and 16MnCr5 (1.7131). Both materials noticeably differ in their flow
stresses, which leads to different punch forces. In the simulation, the forming
force is varied using flow stress data of the two workpiece materials. The punch
force in experiments is measured with a piezo-electric load cell Kistler 9091A.
The experimental investigations on the backward can extrusion process were
carried out on the hydraulic press Lasco TSP 100. Five workpieces were formed at
each parameter setting. The press was used in force-controlled mode at a punch
velocity of 50 mm/s. The press forces were set to 391, 612 and 867 kN for
16MnCr5 (potentiometer settings of 50, 75 and 100 %). The experiments with
100Cr6 were carried out at a press force of 612 kN. In the simulation, the press
forces were set as termination condition, too. The workpieces in the experiments
were coated with a phosphate conversion layer as lubricant carrier, which soap
was applied on for lubrication. The friction factor in the FE simulation was identi-
fied numerically. The experiments were carried out with tool and workpieces at
room temperature. The four parameter settings were also assigned to the integrated
FE simulation model. To show the improvement in simulation accuracy, addition-
al FE simulations with rigid tool (Ctool → ∞) and end stops (Cstop → ∞) / elastic
punch and rigid tool and end stops / rigid punch, respectively, were run. The pre-
stressed die remained elastic in all simulations. The accuracy of the simulation
models is evaluated using the workpiece dimension thickness at bottom
(Fig. 19.9). It is a characteristic for the actual punch position including the
deflections.
Figure 19.9 shows the workpiece dimensions resulting from the experiments
and the FE simulations. The integrated FE simulation model, considering the
whole deflection characteristic, achieves the highest accuracy compared to the ex-
periments for almost all parameter settings. However, at a press force of 867 kN
(16MnCr5), the simulation model with rigid tool and elastic punch shows a similar
deviation from the experiments (0.03 mm) compared to the entirely elastic model
(0.04 mm). The reason for this is the compensation of the tool deflection in the
flow of punch force by the increasing compression of the end stops at high press
forces. At low forces on the end stops due to low press forces, mainly the elastici-
ty of the tool in the flow of the forming force is crucial for the total deflection.
Hence, the deviation of the rigid models compared to the experiments is consider-
able at a low press force of 391 kN (0.20 and 0.31 mm). The entirely elastic model
shows a significantly lower deviation in this case (0.03 mm). The simulation mod-
els with rigid tool or additionally rigid punch lead to the same workpiece dimen-
sions at different press forces because the increasing compression of the end stops
at higher press forces is not represented.
4.2
Simulation
Experiment Tool elastic, punch elastic
mm
Tool rigid, punch elastic
4.0
Thickness at bottom
Tool rigid, punch rigid
3.8
3.6
0
16MnCr5 16MnCr5 16MnCr5 100Cr6
391 kN 612 kN 867 kN 612 kN
Fig. 19.9 Thickness at bottom of backward can extruded part – comparison between expe-
riment and simulation models depending on workpiece material and press force [9]
The high flow stress of 100Cr6 leads to a higher forming force, a higher punch
deflection and thus to a higher remaining thickness at bottom and a smaller depth
of can. The integrated FE simulation model shows the highest accuracy for this pa-
rameter setting, too (0.05 mm of deviation). However, even the simulation with ri-
gid tool and punch results in a bigger thickness at the bottom for 100Cr6 compared
to 16MnCr5 due to the radial elastic deformation of the pre-stressed die. An elastic
modeling of the punch already helps to increase the simulation accuracy compared
to a rigid punch as its long and thin shape contains a high compliance [9].
19.5 Process Modeling and Optimization

The basic approach for the calculation and analysis of the parametric process
model is presented using the case study full forward extrusion on the stroke-
controlled press. Based on the presented integrated FE simulation model, a set of
simulations varying influencing parameters, such as the die dimensions, is run.
Data points are generated as the base for an analysis of the parameter interactions
and the calculation of a parametric model of the entire process behavior. The
process behavior is modeled for several target values depending on different in-
fluencing parameters. It can be used to determine an adjusted combination of val-
ues of influencing parameters in a subsequent optimization step to achieve the
specified target values.
Table 19.1 Variation levels of influencing parameters related to center point; linear primary
effects on target values, calculated by parameter variation from low to high level
Variation levels Effect on Effect on Effect on
Influencing
related to DT1 DT2 LT
parameters
center point in mm in mm in mm
DI1 ± 0.30 mm 0.60 0.00 -2.55
DI2 ± 0.30 mm 0.00 0.60 -2.27
II ± 50 % -0.12 -0.08 0.96
SI ± 0.50 mm 0.00 0.00 2.97
The following three target values were defined to specify the dimensions of the
full forward extruded part (Fig. 19.4 b): workpiece length LT = 22 mm, starting di-
ameter DT1 = 12 mm and final diameter DT2 = 6 mm. Four influencing parameters
with considerable influence on these target values were identified. These are
available for the following optimization of the target values: The die diameters
above (DI1) and below (DI2) the die shoulder, the punch position SI and the relative
interference fit II of the pre-stress rings of the die. The dimensions of the billet are
constant. To be able to consider quadratic process behavior each of the influencing
parameters was varied on three levels based on a center point. To detect all poten-
tial interactions a full factorial experimental design was applied. This leads
to an amount of 34 = 81 different parameter combinations and, thus, 81 simula-
tions with the integrated FE model. The spring stiffness in the model is
Cpress_tool = 222 kN/mm. The press clearance of 0.4 mm is included as a constant
shift of the punch. The friction factor is m = 0.08. Based on a sensitivity analysis,
the variation range for the influencing parameters was determined as shown in
Table 19.1.
The data points for the three target values permit the calculation of the linear
primary effects and interactions. Only the low and high variation levels of the pa-
rameters are required for these calculations. The primary effects (Table 19.1)
show a dependency of the final workpiece length LT on all influencing parameters.
A change of the die diameters affects LT due to the constancy of volume. II also
contributes to all target values by influencing the effective die diameters. The pri-
mary effects of the die diameters (DI1,2) on the workpiece diameters (DT1,2) nearly
agree with the variation range of the influencing parameters from the low to the
high level (0.6 mm). Varying one die diameter has no visible influence on the oth-
er one. SI shows no noticeable influence on the workpiece diameters in the inves-
tigated variation range.
0.5
0.42
0.33
mm
0.25
0.3 0.20
Interaction
0.2 0.09
0.1
0.01
0 LT
DI1*DI2 D *I DT2
I1 I DI1*SI D *I DT1
I2 I DI2*SI
I*S
I I
Fig. 19.10 Absolute values of the interactions between two influencing parameters related
to the target values; calculation based on parameter variation on low and high level [10]
The dependency of an influencing parameter effect on a target value on the lev-

el of one or several other influencing parameters is called interaction. Interactions
between two influencing parameters are shown in Figure 19.10 for the three target
values. The interacting parameters are marked with an “*”. No distinct
interactions can be observed concerning the workpiece diameters DT1,2. In con-

trast, considerable interactions can be found regarding the workpiece length. In
addition to the interactions, the workpiece length LT is influenced quadratically by
the die diameters via the cross-sectional areas. Hence, a quadratic relation was
chosen for the approximation of the full factorial design. Thus, to consider the in-
teractions and the quadratic process behavior without specific process knowledge,
a comprehensive full factorial design with three variation levels is necessary:
Based on the data points resulting from 81 simulation runs, the calculation of the
process model as well as the parameter optimization are realized with the Leven-
berg-Marquardt algorithm, a method, which combines the Gauss-Newton algo-
rithm and the method of gradient descent [12]. The function of the process model
is adjusted to data points according to a least squares fitting problem. For the op-
timization, the fitted process model is used. The algorithm is implemented in
MATLAB 7.3 [10].
19.5.1 Full Factorial Experimental Design

The quadratic process model reaches high accuracy. The relative maximum errors
are 0.06 % for DT1, 0.02 % for DT2 and 0.93 % for LT compared to the 81 data
points calculated with the integrated FE simulation model. Based on this quadratic
process model, the influencing parameters were optimized. A deviation of 10-10
mm between the specified workpiece dimensions and the target values resulting
from the optimized process model parameters was set as a stop criterion. As both
the die diameters DI1,2 and the relative interference fit II influence the resulting
workpiece diameter, several parameter combinations can lead to the same work-
piece diameters. To get a unique solution II was kept constant. The accuracy of the
optimized influencing parameters was validated in a FE simulation with the opti-
mized combinations. A comparison of its results with the specified workpiece di-
mensions shows a good correlation and deviations in the range of 0.5 % for LT and
smaller than 0.05 % for the workpiece diameters DT1,2 [10].
19.5.2 Process Modeling with Reduced Number of Simulations

Despite its high accuracy, the parametric process model based on a full factorial
design requires many simulation runs. A reduced number of variant simulations
are desirable. The discussion of effects and interactions already showed the poten-
tial for a process modeling with less data points. Thus, the process modeling and
optimization with a reduced number of simulations makes use of the following
process knowledge, found in the investigations on the full forward extrusion
process:
• With respect to the workpiece diameters, the die diameters and the in-
terference fit show an almost linear behavior without any interactions
• The workpiece diameters are almost independent of the punch position
in the limited variation range of the punch stroke
Based on these conclusions, the following two approaches were developed.
Sequential Approach
The first approach is a sequential optimization of workpiece diameters and work-

piece length respectively. In a first step, a linear process model for the workpiece
diameters is created depending on the die diameters DI1,2 and the interference fit II.
The influencing parameters are varied in a sensitivity analysis (only one parameter
varied at the same time) based on a center point and two variation levels. In an op-
timization, the parameter settings to achieve the specified workpiece diameters are
calculated. DI1,2 and II are adjusted in the integrated FE simulation model accord-
ing to the optimization results. Then, based on this model, the punch position is
varied on a low and a high level. Finally, the value for the punch position is calcu-
lated to reach the specified target value for LT.
Compared to the full factorial investigation with 34 = 81 variant simulations,
the computational effort could be reduced to 7 + 3 = 10 simulations. Seven simu-
lations are needed for the modeling of the diameters, the remaining three simula-
tions refer to the variation of the punch position. Leaving out the center point of
the variation levels would permit a further reduction by 2 simulations. To validate
the approach it was carried out for the full forward extrusion process. In the opti-
mization step, II was again kept constant. The workpiece dimensions resulting
from the subsequent FE simulation with optimized parameter settings show a high
accuracy. The deviations range from 0.01 % for the workpiece diameters DT1,2 to
0.06 % for the length LT. Despite its high accuracy, this sequential procedure with
a separate consideration of workpiece diameters and length does not lead to a
comprehensive model of the process behavior. This, for example, makes it diffi-
cult to investigate the influence of the scatter of influencing parameters on the tar-
get values. To permit a comprehensive modeling with a reduced number of
simulation runs the following approach was developed [10].
Standardized Length
The approach uses the specific process characteristics found above. In addition,
process knowledge concerning the geometric relations of the workpiece dimen-
sions is applied. The objective of this analytic approach is the elimination of the
influence of the die diameters (DI1,2) and the interference fit II on the workpiece
length LT, related to a reference point. This corresponds to the calculation of a
standardized workpiece length, which only depends on the punch position SI. Af-
ter this, according to the independency of the workpiece diameters of the punch
position, the required punch position to achieve the specified workpiece length can
be determined by linear interpolation. In Figure 19.5.2, the resulting lengths of the
81 simulation runs are plotted. The systematic permutation of the influencing pa-
rameters in this full factorial experimental design leads to a periodic behavior of
the workpiece lengths. In the full factorial investigation, the punch position is va-
ried every 27 simulations. Hence, the 81 single standardized lengths should form a
step function with the steps at the changes in the punch position.
According to the observed linear behavior without interactions concerning the
workpiece diameters, the required die diameters DI1,2 and interference fit II to
achieve the specified workpiece diameters can be determined by means of their
linear combination. In the investigated quasi-stationary cold forging process, the

effect of the die diameters and the interference fit on the length is almost entirely
caused by the displacement of a certain workpiece volume depending on the
change in the actual die diameters. Via the cross-sectional areas, this leads to a qu-
adratic influence on the workpiece length. In a previous realization of this ap-
proach, the quadratic behavior was considered by an iterative smoothing [11]. In
this actual implementation, the quadratic influence of cross-sectional areas is con-
sidered from the beginning.
Thus, the basic approach is to identify the workpiece volume affected, related
to a reference point and depending on the value of the respective influencing pa-
rameter (DI1,2, II). Subsequently, depending on the actual values for die diameters
and interference fit, the actually affected workpiece volumes and, thereafter, the
change in workpiece length resulting from the parameter values can be deter-
mined. This change in length is subtracted from the specified target value of the
workpiece length. The remaining difference between the length of the reference
point and this standardized length is caused by the punch position. This permits a
determination of the required punch position to achieve the specified workpiece
length LT.
-0.5 mm Punch pos. = 0 mm 0.5 mm

26
FE simulation results
mm
22 Ls_low
Length
Ls_high
20 Ls
18
Calculation of standardized
length Ls with 7 data points
0
1 9 18 27 36 45 54 63 72 81
Consecutive numbering of FE simulations
Fig. 19.11 Workpiece lengths resulting from FE simulations based on full factorial experi-
mental design; standardized length Ls calculated based on 7 data points
Seven data points marked with grey dots in Figure 19.11 are required as the ba-
sis for the analytic approach: Based on a reference point at the low punch level
(No. 9), the interference fit II and the two die diameters DI1,2 are varied separately
between the minimum and the maximum variation level. The influenced work-
piece volume is calculated using the change in workpiece length and the actual
workpiece diameter below the die shoulder. The die diameters are then also varied
on the high punch level (reference point No. 63) because the workpiece volume
affected by the diameters depends on the proportionate length above and below
the die shoulder and, with that, on the punch position.
In the present case study, the 81 workpiece lengths resulting from the simula-
tion runs with of the full factorial design are the specified target values for LT. De-
pending on the respective values of die diameters and interference fit, a back
calculation from the influenced workpiece volume to the change in workpiece
length is made for each of the 81 lengths. The change is subtracted from the target
length for the high and low punch level. This leads to Ls_high and Ls_low respec-
tively (Fig. 19.11). To consider the dependency of the influenced workpiece vo-
lume on the punch position the two curves are weighted depending on the punch
position. The resulting standardized length Ls in Figure 19.11 clearly shows the
required step characteristic. The die diameters and the interference fit to reach the
specified workpiece diameters are determined by adding up the effects, since no
relevant interactions were observed concerning the diameters.
To assess the accuracy of the presented approach it is compared to the simula-
tion results of the full factorial experimental design. The workpiece lengths
represent the target values. The application of the standardized length leads to the
corresponding approximated values for the punch position. Ideally, the resulting
punch position should match the values set in the FE model. However, there are
some deviations. The workpiece lengths corresponding to the respective punch
position were calculated in a reverse calculation for comparison with the 81 simu-
lation results. In addition, for a comprehensive evaluation, the full factorial
approximation and a linear process model neglecting the interactions are also
compared. The linear model is based on a single variation of each influencing pa-
rameter on a low and a high level from a center point. It requires eight simulation
runs compared to seven needed for the standardized length. The function values of
the process models are compared to the workpiece lengths resulting from the si-
mulation. Accuracy is evaluated using the absolute value of the maximum error
and the sum of squared errors.
0.6 3.0 Quadratic approximation

2.96
0.50 (81 simulations)
mm mm2
Sum of squared errors
Linear model (8 simulations)

2.0
Absolute value
0.4
of max. error
Standardized length
0.3 1.5 (7 simulations)
0.2 0.19 1.0
0.1 0.06 0.5

0.26
0 0.03
0
Fig. 19.12 Errors of the three process models compared to the FE simulation results based
on the full factorial experimental design for target value workpiece length
Figure 19.12 shows the results of this comparison. As expected, the full factori-
al approach (81 points) has the lowest deviations from the original data points.
The maximum error is 0.06 mm, the sum of squared errors is 0.03 mm2. The linear
process model based on the addition of effects has considerably higher error
values of 0.50 mm and 2.96 mm2. Obviously, neglecting the interactions is not va-
lid. The standardized length represents the real behavior in a much better way.
Here, the maximum error is 0.19 mm, the sum of squared errors is 0.26 mm2. This
means a reduction by 62 % and 91 %, respectively, compared to the linear process
model, which even requires one more simulation run. Thus, the presented ap-
proach includes a comprehensive modeling of the entire process behavior with an
extremely reduced number of required simulation runs. It has the potential to be
applied to further quasi-stationary cold forging processes.
19.6 Conclusions and Outlook

The presented approach provides a comprehensive procedure for an optimized FE-
based tool and process design considering the interactions between process, tool
and machine in cold forging. The approach includes the entire sequence from the
determination of the deflection characteristic to the calculation of a parametric
process model including an optimization of the values of influencing parameters
to achieve high workpiece accuracy. Approaches for the determination of the def-
lection characteristic consisting of tooling system and stroke-controlled press as
well as a tooling system on a force-controlled press were developed. The tooling
system characteristic was determined by FE simulations of tool assemblies to
avoid experimental investigations. The deflection characteristics were incorpo-
rated into the FE simulations of cold forging processes as spring elements and a
constant shift of the punch to consider non-linear press effects. The experimental
validation by two case studies, a full forward extrusion on a stroke-controlled and
a backward can extrusion process on a force-controlled press, shows a significant
increase in simulation accuracy. Concerning the full forward extrusion process,
the deviations in length are reduced from up to 17.7 % with the rigid simulation to
0.1 %. Thus, the integrated FE simulation model is able to represent the entire
process behavior including the interactions between process, press and the tooling
system. With the help of a set of simulations varying the influencing parameters,
such as the die dimensions, data points were generated as the basis for a parame-
tric model of the process behavior and a subsequent optimization. A quadratic
approximation based on a full factorial investigation then lead to maximum devia-
tions below 1 %. To reduce the required number of simulations for the process
model compared to a full factorial experimental design, two approaches were de-
veloped using process knowledge from an analysis of process behavior including
effects and interactions. A sequential procedure helps to reduce the number of va-
riant simulations from 81 to 10 for the example process full forward extrusion
without affecting the accuracy. In a second approach, an analytic back calculation
was developed to determine the punch position required for the specified work-
piece length depending on the die dimensions. With this standardized length, the
number of simulations could be reduced from 81 to 7 for the example process full
forward extrusion. Compared to a simple linear process model the sum of squared
errors was reduced by 91 %. This comprehensive description of the process beha-
vior, like the parametric process model based on the full factorial design, permits a
consideration of the scatter of influencing parameters to predict the achievable
workpiece tolerances as well as the required tolerances in the whole system. By

means of Monte-Carlo simulations, for example, the frequency density of target
values can be estimated depending on the scatter of influencing parameters.
Moreover, in future, a closed-die lateral cold extrusion process will be investi-
gated. Due to a severely growing forming force with increasing die filling, differ-
ent deflection characteristics and interactions are expected. The developed
comprehensive procedure is the basis for these supplemental investigations.
References
[1] Balendra, R.: Net-Shape-Forming: State of the Art. J. Mater. Process Technol. 115,
172–179 (2001), doi:10.1016/S0924-0136(01)00812-3
[2] Kroiss, T., Völkl, R., Engel, U., Geiger, M.: Modeling of Process-Tool-Machine Inte-
ractions in Cold Forging. In: Proc. 1st Int. Conf. on PMI, Hannover, Germany, pp.
125–132 (2008)
[3] Behrens, B.A., Matthias, T., Czora, M., et al.: Improving the accuracy of numerical
investigations of multi-stage sheet metal processes by coupling a process FE analysis
with the machine simulation. In: Proc. 1st Int. Conf. on PMI, Hannover, Germany,
pp. 133–139 (2008)
[4] Brecher, C., Schapp, L., Paepenmüller, F.: Gekoppelte Simulation von Presse und
Massivumformprozess. Berücksichtigung (nicht-)linearen Maschinenverhaltens in der
Umformsimulation. wt Werkstattstechnik Online 96, 526–529 (2006)
[5] Großmann, K., Wiemer, H., Hardtmann, A., Penter, L.: Faster to sound parts by ad-
vanced forming process simulation – advanced forming process model including the
elastic effects of the forming press and tool. Steel Res. Int. 78, 825–830 (2007)
[6] DIN 55189, Machine tools – determination of the ratings of presses for sheet metal
working under static load. Beuth, Berlin (1988)
[7] Chodnikiewicz, K., Balendra, R., Wanheim, T.: A new concept for the measurement
of press stiffness. J. Mater. Process Technol. 44, 293–299 (1994), doi:10.1016/0924-
0136(94)90442-1
[8] Arentoft, M., Wanheim, T.: A new Approach to determine Press Stiffness. CIRP
Ann. 54(1), 265–268 (2005), doi:10.1016/S0007-8506(07)60099-7
[9] Kroiß, T., Engel, U., Merklein, M.: Simulation-Based Determination of Deflection
Characteristic of Tooling System and its Modeling in FE Simulation of Cold Forging
Process. Steel Res. Int (special ed. Met. Form 2010) 81, 318–321 (2010)
[10] Kroiß, T., Engel, U., Merklein, M.: Process Modeling In Cold Forging Considering
The Process-Tool-Machine Interactions. In: Proc. 10th Int. Conf. Numer. Methods
Ind. Form Process, AIP Conf. Proc., vol. 1252, pp. 312–319 (2010)
[11] Kroiß, T., Engel, U., Merklein, M.: Modeling of the Behavior of a Cold Forging
Process Considering the Deflection of Tooling System and Press. In: Proc. 2nd Int.
Conf. on PMI (2010) ISBN: 978-0-9866331-0-2
[12] Spellucci, P.: Numerische Verfahren der nichtlinearen Optimierung. Birkhäuser, Ba-
sel (1993)
Chapter 20
Interaction Effects between Strip and Work
Roll during Flat Rolling Process
S. Puchhala, M. Franzke, and G. Hirt
Abstract. During the flat rolling process (cold or hot), the strip flatness and
thickness profile are highly influenced by the interaction effects between strip and
work rolls. To understand and analyze these effects a new modeling concept was
developed. Within this concept, the tool simulations are separated from the
process simulation. With the help of an automatic coupling module, the influences
of the tool effects are realized within the process simulation. With this modeling
concept, three types of interaction phenomena are studied and validated using ex-
periments: elastic roll effects during the cold rolling process, work roll thermal ef-
fects during the hot rolling process and tribological effects (abrasive wear) on the
process simulation. It was also shown that, compared to the single FE model,
this modeling concept is relatively faster and suitable for large 3D models without
losing the quality of the predicted results.
20.1 Introduction
During the flat rolling process, the dimensional tolerances and the mechanical-
metallurgical properties of the rolled products become more stringent. Therefore,
computational models are being used to investigate the physical aspects, such as
thermo-mechanical events of the strip, mechanical deformation and thermal evolu-
tion of the work rolls, and tribological effects, such as wear. Conventional numer-
ical modeling of such processes assuming plane strain conditions (2D) may not be
sufficient since the contact pressure distribution along the strip width may have a
non-uniform profile due to the elastic deformation of the work rolls. Depending on
the roll elastic deformation, strip flatness profile changes, i. e. crowns, will origi-
nate (see Fig. 20.1). The modeling of such processes using a conventional single
FE model and considering all the above-mentioned effects, will not only compli-
cate the computation (deformable – deformable contact situation has to be
handled) but also affect the computation time. This situation is even more compli-
cated when contact between elastic tools (in the case of backup rolls) has to be
considered.
Significant effort was devoted to the study of mechanical and thermo-
mechanical effects in the strip and work rolls using the numerical methods.
Particularly, A. Kainz [1] simulated a skin pass cold rolling process (in 2D) consi-
dering elastic effects of the work rolls using the commercial FE program
440 S. Puchhala, M. Franzke, and G. Hirt
ABAQUS/EXPLICIT. He stated that, for the computation of 22 ms (150 mm strip

length) of real process time, simulation required several days. On the other hand,
T.H.Kim, S.M.LEE [2] presented an integrated FE process model, where the strip
profile was analyzed using 3D implicit simulations. However, the mechanical con-
tact computations (at strip-work roll and work roll-backup roll interface) are han-
dled using analytical models and coupled with individual FE simulation models.
Several other researchers [3, 4, and 5] have presented the modeling of cold rolling
processes considering elastic work rolls using single FE models in 2D configura-
tions. A fully-coupled 3D FE simulation of the cold rolling process (considering
the roll flattening and bending effects) is yet to be realized.
Fig. 20. 1 Strip profile quality problems during cold rolling process
For the modeling of a hot rolling process, Tseng [6] simulated steady state heat
transfer in the work roll using a two-dimensional Finite Difference Method. Using
an empirical approach, he included the heat generation term due to plastic defor-
mation and friction between roll and strip. On the other hand, Hsu and Evans [7]
conducted a two-dimensional FE analysis of the rolling process considering a
steady state heat transfer condition. C.G.SUN and S.M.HWANG [8,9] presented
integrated models considering the thermo-mechanical effects both in roll and strip
for 2D and 3D. They presented a coupled model for the hot rolling simulation,
where they considered the steady state heat transfer using 2D models for the rolls.
They also considered the elastic deformation of the work rolls. LEE and Hwang
[10] presented a similar model for the computation of the thermal effects for the
work rolls. These models have been validated with the experimental results and
are good for the estimation of the thermal behavior of the rolls during a hot rolling
process. The only untouched area in those models is the analysis of bi-directional
interaction effects between strip and rolls.
At the same time, during cold and hot rolling processes, the tribological events,
particularly abrasive wear of the work rolls, directly affect the rolled strip surface
quality and may hinder the flatness profile [11, 12]. Many investigations have
been carried out to understand the wear mechanism of the rolls during rolling
processes [13,14,15,16]. All of these studies are related to the modeling of wear
using analytical methods or empirical models. As far as we know, offline calcula-
tions of precise local work roll wear during hot or cold rolling have not yet been
noticed.
20 Interaction Effects between Strip and Work Roll during Flat Rolling Process 441
The objective of this work is to develop a validated simulation tool for the
analysis of the interaction effects between the work rolls and the strip during a flat
rolling process. For that, a novel FE modeling concept was developed considering
three different aspects, which influence the strip quality, i. e. elastic effects of the
work rolls during a cold rolling process, thermo-mechanical effects of the work
rolls during a hot rolling process and work roll tribological effects (mainly abra-
sive wear) during cold and hot rolling processes. In the following sections, each of
these aspects is discussed in detail. In the next section, a detailed description of the
de-coupled modeling concept (applicable for the modeling of all three aspects
mentioned above) is explained. The details of abrasive wear models and their im-
plementation are also discussed within the next section. In the section “modeling
and analysis of a cold rolling process with elastic rolls”, the cold rolling experi-
mental setup and the optical deformation measuring method is explained. Later,
the FE modeling of the cold rolling process using the new modeling concept is de-
scribed; and the simulation and experimental results are given in a sub-section. In
the section “modeling and analysis of hot rolling process”, an identical structure as
mentioned above is maintained. In the following section, the modeling of work
roll wear is described. At the end, a detailed discussion on the overall work (mod-
eling concept and the results) is given.
20.2 Concept and Implementation

A numerical simulation strategy was developed using FEM to analyze the interac-
tion effects during rolling processes. Within this concept, the computation of the
flat rolling process simulation is separated into two simulation models. In this con-
text, the process simulation model consists of the workpiece (modeled using elas-
tic-plastic material description for cold rolling and thermo-mechanical description
for hot rolling) and the work rolls were modeled as rigid bodies (see Fig. 20.2). On
the other hand, a different simulation model is considered for the tools (pure elas-
tic model in case of cold rolling and thermo-mechanical model in case of hot roll-
ing). These two separate models are coupled using an automatic-update module.
This module uses a mapping table (where the FE surface mesh information of both
rigid and elastic tools is saved) to update the geometrical information in both si-
mulation models. For the communication and compatibility between two simula-
tion models, an identical surface mesh is to be used in both models. Further details
of the concept, such as data transfer, tool activation and mesh compatibility (Do
Elastic Tool Simulation DOELTOSIM module), are described in the [19].
Based on this modeling concept, two other technical aspects were implemented:
thermo-mechanical tool effects and wear prediction. For the modeling of a hot
rolling process, consideration of the tool thermo-mechanical effects are very im-
portant since a large amount of the energy is conducted into the work rolls, when
they come in contact with the hot strip. Here, the computation of roll thermo-
mechanical effects is separated from the process simulation model. At each
pre-defined time step, the computation of tool effects is activated. In case of 10
Sec total simulation time, if the pre-defined time step size is 2 Sec, the tool simu-
lation is activated 5 times. As the thermal effects are transient, unlike the elastic
simulation model, the thermo-mechanical tool computation is carried out simulta-

neously until the next activation takes place. As the process and tool simulation
models run concurrently, the tool simulation results should be readily available
once the process simulation reaches the activation time-step. For that, a novel con-
cept of rotating boundary conditions was developed to avoid the accumulation of
large rotations of the Lagrange mesh of the work roll.
FE-Model for
FE-Model for
process simulation
elastic tool simulation
(With rigid tool)
LARSTRAN/SHAPE DOELTOSIM LARSTRAN/SHAPE
Deformation*
damping
Increment
Contact-pressure
Elastic-plastic Pure elastic

material model material model
Mapping table
Fig. 20. 2 FEM-FEM coupling concept
Within this concept, the work roll is fixed in all DOF (Degrees Of Freedom)
and the boundary conditions, such as heat flux at roll-strip contact, cooling
zones are defined as rotating (with the rolling velocity) boundary conditions (see
Fig. 20.3).
Fig. 20.3 Concept of rotating boundary condition

The conductive heat flux (at roll - strip contact zone) is calculated using the
state variables (contact pressure and strip temperature) provided by process simu-
lation (Eq. 20.1).
q = hgap (Tstrip − Troll ) (20.1)
The gap conductance hgap is defined as a function of contact pressure, roll rough-
ness and roll hardness. Detailed explanation of this model is given in the section
“modeling and analysis of hot rolling”. In the same way, the convective heat
fluxes (due to water cooling and air cooling) are calculated using Equations 20.2
and 20.3.
q = hwater (Troll − Twater ) (20.2)
q = hair (Troll − Tair ) (20.3)

For the application of conductive heat flux, the contact angle θ is necessary and
approximated using the contact length ( ld′ ) and deformed roll radius ( r ′ ) as
shown in Equation 4.
ld′
tan θ = with
r′
(20.4)
F
r ′ = r (1 + C H and ld = r ′Δh
b Δh
The last aspect considered to analyze the interaction effects is work roll wear. It is
known that the rolled strip profile will have the roll gap contour, i. e. if the work
rolls are worn, the roll gap contour will no longer be uniform. Hence, it is impor-
tant to analyze the wear progress of the work rolls during the rolling process. To
analyze the local abrasive wear of the work rolls, three different wear models were
selected and are coupled to the FE process simulation similar to the concept
shown in Figure 20.2. The first model is the basic Archard model [15], where the
wear depth is a function of normal forces FN, hardness of the softer surface H , the
sliding distance s and wear coefficient k (Eq. 20.5). The basic assumption of this
model is that the wear rate is independent of the apparent contact area; and it also
makes no considerations of variations with time. Moreover, this model is
widely used since it provides a good estimation for the order of magnitude with
less effort.
V F (20.5)
=k N
s H
The second model is an empirical model, which is a modified version of Oike’s
equation [18], which is tailored for the hot rolling process (Eq. 20.6). This model
was developed and is being used for the on-line observation of wear for the finish-
ing stands of a hot strip mill at TATA steel, India. Within the equation, the most
important process parameter is the contact pressure P and other parameters, e. g.
wstrip is strip width, ld is contact length, ∆h is the thickness reduction, Lstrip is the
length of the rolled strip and R is the radius of the work roll. The empirical
constants α, β and γ were determined using a stochastic optimization technique,
i. e. the summation of wear error (difference between maximum measured wear

and predicted wear) should be minimum. For the rolling of high carbon steels with
30 % thickness reduction, the wear coefficients were predicted as 1.0E-10, 0.188
and 0.11 respectively.
β
Lstrip  P 
W ( x) = α   (ld Δh )γ (20.6)
2πR  wstrip ld 
Within the last wear model, energy dissipation is introduced into the Archard’s
wear model (K.E. Nürnberg [14]) leading to reasonable qualitative wear predic-
tion. By applying the Coulomb law to the Archard equation (Eq. 20.7) the dissipa-
tion rate is defined as the product of shear stress τ (MPa) to the sliding velocity ν
(mm/s). From Equation 20.8, the rate of wear and wear work Z can be directly re-
lated to the frictional dissipation rate. At the same time, the accumulated wear
work is directly proportional to the dissipated energy. Finally, the wear model at
element level can be written as a function of the accumulated wear work Zacc and
the contact pressure (as shown in Eq. 20.9).
 pv dt = H  μτ v dt
k k
w= t t t (20.7)
H
– D = τ t vt Z = μ D  (20.8)
k (Z acc,e ) n
– we (r ) =
H  pe (r , t )vslid (r , t )dt where Z acc = Z
r =1
e (20.9)
20.3 Modeling and Analysis of Cold Rolling Process

In general, the cold rolling process is carried out at the final stages of the rolling
process, i. e. no other forming process step is conducted, which has an impact on
the thickness profile in a corrective manner. On the other hand, the thickness pro-
file has a substantial influence on the following forming processes, i. e. deep
drawing process. The thickness profile directly depends on the contact pressure
distribution along the roll gap. Due to elastic deformation of the work rolls an in-
homogeneous contact pressure distribution may develop.
Table 20.1 Process and geometric parameters for the cold rolling process
Parameter Value
Initial strip thickness 1.67 mm
Exit strip thickness 1.65, 1.48, 1.22, 1.20 mm
Work roll radius 300 mm
Work roll velocity 200 mm/sec
Friction coefficient 0.3
Strip tension (front and back) 30 kN
To obtain a deep understanding of the plastic flow of the strip and elastic de-
formation of the work rolls, implicit FE simulations were performed utilizing the
newly-developed concept. As described in the implementation section, the process
simulation model consists of a strip with elastic-plastic material description; and
the rolls are considered as rigid bodies. Four different thickness reduction cases
were simulated. Since the contact length for each case is not higher than 10 mm, at
least 20 elements (with 0.5 mm length for both strip and roll) are necessary for the
detailed analysis of plastic flow within the roll gap (Fig. 20.4).
Fig. 20.4 FE mesh of the strip and work rolls for cold rolling simulations
To validate the simulation results obtained by the new modeling concept and to
understand the underlying elastic-plastic strip material flow and elastic work roll
deformation, cold rolling experiments were conducted on the rolling mill installed
at IBF (Fig. 20.5 left). High-strength steel strips (ZStE340) with initial thicknesses
of 1.67 mm were rolled to four different thickness reductions (Tab. 1).
Fig. 20.5 Rolling mill installed at IBF (left), rolling mill deformation in rolling direction
(top right) and in thickness direction (bottom right)
The rolling was conducted at a relatively low velocity 500 mm/s (conventional
cold rolling velocity ranges 5 m/s up to 10 m/s) as the rolling mill limit is reached.
For each case, the rolling force and the exit-strip thickness were measured. Fur-
thermore, the elastic deformation of the whole mill was also measured using an
optical measuring method TRITOP [38]. As shown in Figure 20.5, coded (fixed
reference marks) and uncoded marks were placed on the rolling mill. At static
loading conditions, free-form photographs were taken using a high resolution
camera. With the help of the TRITOP software, the deformation of the uncoded
marks was measured (see Fig. 20.5). As expected, the maximum deformation
(0.12 mm) was observed in thickness direction whereas in the rolling direction the
maximum deformation was observed as 0.08 mm.
20.3.1 Experimental and Simulation Results of Cold Rolling

Process
The cold rolling process under consideration was simulated in 3D configuration
using a non-linear implicit FE solver LARSTRAN/SHAPE. To obtain reliable
steady state results, at least 150 mm strip length (~15 contact lengths) was simu-
lated. As expected, the rolling force predicted using an elastic work rolls model is
very close to the measurement (see Fig. 20.6). At the same time, the measured exit
strip thickness also matches the simulated results very well. For the analysis of the
strip thickness profile, the contact pressure distribution along the strip width was
examined. It is clear from Figure 20.6 (bottom left) that the contact pressure dis-
tribution has a non-uniform profile in the elastic work roll model, which may lead
to a non-uniform stress distribution within the roll gap. This directly influences the
exit strip thickness. As shown in Figure 20.6 (bottom right), the equivalent plastic
Fig. 20.6 Comparison of rolling force and exit strip thickness

strain along the strip width is not uniform (for high thickness reduction case),
which confirms the non-uniform exit strip thickness for a 28.2 % thickness reduc-
tion case. Whereas for a low thickness reduction case, the plastic strain along the
strip width is relatively uniform, this will lead to a uniform thickness reduction.
For the analysis of the plastic flow of the strip the comprehension of the shear
stress distribution is very important. As shown in Figure 20.7 (left), there is a small
“no-slip-zone” in the 28.2 % thickness reduction case. This means that there will be
no significant thickness reduction in that region, whereas in a 14 % thickness re-
duction case (Fig. 20.7 - right) there is no such zone. From the Figures, it is clear
that the plastic flow using the new modeling concept is identical to the single FE
model. It can also be observed that the rolling force in the new model is smaller
than in the single FE model, which may be due to a lower contact length in the new
model. This directly depends on the frequency of the activation of the elastic roll
model. This situation can be improved by increasing the activation of the tool simu-
lation model, which may lead to a higher computation time. A compromise has to
be found between these two phenomena to achieve reliable results. Detailed expla-
nation regarding the frequency of the activation of tool simulation is given in [20].
Fig. 20.7 Comparison of shear stress distribution (28.2 %, 11.3 %)
Any change in the work roll profile will directly affect the end strip quality. In
order to control the required output strip specifications a careful analysis of the
roll deformation is the key. Figure 20.8 shows the stress distribution in the strip
thickness direction for a 23 % thickness reduction case predicted using
LARSTRAN and ABAQUS [36].
Fig. 20.8 Comparison of elastic stress distribution in work rolls

In both cases, the maximum compressive stress is about 360 N/mm², which is
considerably lower than the yield stress of 860 MPa of the roll material. For the
de-coupled simulation, a total of 200 coupled cycles were used and the computa-
tion time is up to 2 times faster than the single FE model. The computation time
for the single FE model was about 614 minutes, whereas with the new concept
(using DOELTOSIM module), it required only 316 minutes. From Figure 20.8 it
is clear that the developed modeling concept is reliable and faster. Further details
of the elastic roll influence on cold rolling process are explained in [19, 20].
20.4 Modeling and Analysis of Hot Rolling Process

In the hot rolling process, a large amount of heat energy is conducted into the
work rolls during the contact with the hot strip. At the same time, the temperature
distribution of the work rolls influences the forming characteristics of the strip. To
analyze these interdependent phenomena, lab scale experiments and numerical si-
mulations were conducted. Single-pass hot rolling experiments were conducted
using 1.4301 stainless steel strips with an initial thickness of 1.9 mm and 150 mm
width. The strip material flow curve was determined (using Eq. 20.10) for differ-
ent strain rates and temperature up to strip inlet temperature, i. e >1,050°C. The
strip material properties and rolling parameters are given in Table 2.
k f = K .ϕ ( A+ B.T ) .e C.T .ϕ D .e E.ϕ

K = 7140.64 N/mm², A = -0.26, B = 0.0004 1/°C, C = -0.0033 1/°C, (20.10)
D = 0.223, E = -0.62
The strips were heated up to 1,050°C and rolled with a relatively low velocity of
200 mm/sec to two different thickness reductions (21 % and 26 %). To get a sys-
tem level understanding, the rolling force was measured with high accuracy. To
understand the heat transfer phenomenon the analysis of strip temperature distri-
bution at the roll gap is an important aspect. It was not possible to measure the
temperature distribution of the strip within the roll gap, it was, however, measured
before and after the roll gap using pyrometers, as shown in the experimental setup
(see Fig. 20.9).
Fig. 20.9 Pyrometer setup for hot rolling experiments

The quality of the measured temperature using the pyrometer lies in the defini-
tion of the emission coefficient. The pyrometer was trained to obtain the precise
coefficient using an additional experiment, where a specimen connected with a
thermo-couple and the temperature was measured using both systems. Once both
systems were measuring the identical temperature, the pyrometer was used in the
real experiments. For the analysis of the strip temperature within the roll gap ex-
tensive numerical simulations were conducted.
In order to study the effects of the transient heat transfer phenomenon at the roll
gap a good understanding of the process parameters is critical. For that purpose,
numerical simulations were conducted using the newly-developed modeling con-
cept. Within this concept, the process simulation was modeled using thermo-
mechanical strip and the work rolls were modeled as rigid bodies. Although the
roll was modeled as a rigid body the solver will consider the temperature distribu-
tion of the roll for solving the heat transfer at the roll gap. The work roll tempera-
ture will be updated using an external work roll model.
Table 20.2 Hot rolling process parameters and material properties
Roll geometric parameters Strip geometric parameters

Diameter 300 mm Initial thickness 1.9 mm
Length 1,000 mm Exit thickness 1.5,1.4 mm
Velocity 200,400,1000 mm/s Width 150 mm
Roll material properties Strip material properties Units
Elastic 215E3 at 20 °C 198.340E3 at 20 °C N/mm²
modulus 169E3 at 600 °C 117.000E3 at 1,000 °C
Density 7.8E-9 at 20 °C 7.68E-9 at 500°C Kg/mm³
7.6E-9 at 600 °C 7.55E-9 at 1,000°C
Conductivity 48.0 at 20 °C 14.85 at 20 °C W/mm K
44.3 at 600 °C 27.35 at 1,000 °C
Specific heat 4.6E8 at 20 °C 4.55E8 at 20°C J/Kg K
5.9E8 at 600 °C 5.85E8 at 1,000 °C
For the detailed analysis of the heat transfer phenomenon at the roll gap, the de-
finition of gap conductance is most critical. Tseng [21, 22, 23] reviewed various
analytical models for the definition of gap conductance and mentioned two sophis-
ticated relations. The first is based on the analysis by Mikic [24] (Eq. 20.10), the
second was semi-empirically derived by Yovanovich et. al. [25] (Eq. 20.11). In
both Equations, kh is the harmonic mean thermal conductivity in W/mm K, Ra is
the combined arithmetic surface roughness in mm, P is the contact pressure and M
is the hardness of the softer material. For hot rolling, the analysis of thermal con-
tact conductance of the interface should include the conductance of oxide scale.
To serially link the oxide layer with interface contact conductance, the conduc-
tance in Equations 20.11 and 20.12 can be modified as shown in Equation 20.13,
where the underline signifies the contact conductance including the oxide scale
conductance, which equals k0/δ0. Here, k0 and δ0 are the thermal conductivity and
thickness of oxide scale respectively. The properties of the oxide layer during the
hot rolling of various grades of steels were studied in detail by the large number of
researchers [26, 27, 28]. Conventionally, high pressure jets are used to desiccate
between the stands so that the scale layer is kept relatively thin. A 10 μm oxide
layer is consistent with most measurements [27, 28]; and the hardness of the oxide
layer was observed as 980 MPa (HV100). A user sub-routine was developed (and
implemented in a model program for PEP) to define the gap conductivity accord-
ing to Equation 20.13, where the local contact pressure is extracted from the
process simulation. With this, a detailed analysis of heat transfer between work
roll and strip was conducted.
hc = 3800k h Ra−0.257 [P / (M + P )]0.95 + (k m / δ m )runc (20.11)

(20.12)
hc = 4200k h Ra−0.257 [P / M ]0.95 + (k m / δ m )runc
h c = hc (ko / δ o ) /[hc + (ko / δ o )]

(20.13)
Furthermore, it is important to analyze the thermo-mechanical behavior of the

work rolls during the hot rolling process. The work rolls are heated during the
contact with the strip by conduction, which is predominant in the hot rolling
process [29]. The heating of the rolls is countered by the proper cooling system,
which is normally provided by water sprinklers located at the entry and exit side
of the stand (Fig. 20.10). The rolling velocity and the angle the cooling (on both
sides) is provided at determine the temperature distribution and also the surface
stresses of the work rolls. To analyze these aspects a rotating boundary concept
was introduced. To determine steady state temperature distribution within the
work rolls various parameter variations were tested and qualitative results will be
presented in the next section.
Fig. 20.10 Boundary conditions for work rolls during hot rolling process
20.4.1 Experimental and Simulation Results of Hot Rolling

Process
Various hot rolling experiments and simulations were conducted to analyze the
heat transfer phenomenon between the hot strip and the work roll. Coupled ther-
mo-mechanical simulations were conducted using the non-linear implicit solver
LARSTRAN/SHAPE [37]. The computation of roll thermal effects is conducted
in a separate simulation model and coupled with the process simulation using the
newly-developed module DOHOTSIM. Transient 3D thermo-mechanical simula-
tions were conducted with the rolling process data given in Table 2. The rolling
force measured was about 750 kN and 850 kN for 21 % and 26 % thickness reduc-
tions respectively. Since the strip surface roughness was high, using a high-
friction coefficient (µ=0.6) and with the pressure-dependent gap conductance, the
predicted rolling force was about 735 kN and 837 kN for 21 % and 26 % thickness
reduction cases respectively. For the analysis of plastic flow in the roll gap, the
shear stress was analyzed under two different aspects.
Fig. 20.11 Strip shear stress and temperature distribution in hot rolling
The first one is the influence of elastic tool effects and the second the definition
of gap conductance. The shear stress as shown in Figure 20.11 (top left) is pre-
dicted for two different rolling velocities, both using rigid and elastic rolls for 21
% thickness reduction case. For a high velocity, low shear stress is predicted and
vice versa. This may directly depend on the strip temperature distribution. At
the same time, the influence of the elastic work roll on the plastic deformation of
the strip is considerably low. However, the influence of the definition of gap
conductance on the plastic deformation has a considerable effect (Fig. 20.11, top
right). On the entry side, the shear stress distribution has identical behavior in both
models whereas on the exit side, the pressure-dependent model predicts high shear
stresses. This may be due to the tendency in the strip temperature in both models.
Fig. 20.12 Surface temperature distribution of the work roll for first rotation (left) and after
100 sec (right)
For the clear understanding of the heat transfer between strip and work roll, a
pressure-dependent gap conductance was defined according to Equation 10. Fig-
ure 20.11 (bottom left) shows the predicted gap conductance (for two different
friction coefficients) for a node as it passes through the roll gap. From the Figure,
it is clear that the gap conductance has a profile identical to the contact pressure
distribution and reaches its maximum at the exit side of the roll gap, where the
normal compressive contact pressure has its maxima. Figure 20.11 (bottom right)
makes it clear that the strip temperature predicted using both models before and
after the roll gap matches well with the measured temperature (755 °C). An impor-
tant phenomenon can be observed here: the strip temperature along the roll gap
has a large deviation (pressure-dependent model predicts up to 230 °C lower than
the constant gap conductance model- see Fig. 20.11) in both models. Neglecting
this phenomenon during the modeling of a hot rolling process may lead to serious
problems since the metallurgical changes directly depend on the forming tempera-
ture and others [30, 31].
Fig. 20.13 Temperature distribution at surface and inside the roll (left - conventional cool-
ing, right - double cooling)
Fig. 20.14 Temperature distribution at the surface and inside the roll
A detailed analysis of the thermal evolution of the work rolls is very important
since the life and the wear of the work rolls can be controlled by improving the
cooling system [32, 33]. The proper cooling may reduce the induced thermal
stresses and, from that, the abatement of the thermal fatigue may be achieved. To
understand this phenomenon, a new rotating boundary model was developed and
implemented within the FE program PEP [39] and LARSTRAN. For the qualita-
tive numerical analysis, the hot rolling lab experiments were modeled assuming
different cooling strategies. Figure 20.12 shows the temperature evolution at the
surface of the roll for the first rotation (left) and after 10 seconds of rolling (rolling
velocity 1 m/sec) for three different cooling strategies (conventional, double cool-
ing and late cooling [33, 34]). The schematic diagram of the first two strategies is
shown in Fig. 20.13 and the late cooling means that the cooling starts at 130°).
From the first rotation curves for all cooling methods it can be observed that due
to insufficient cooling a small amount of heat (about 15°C) is retained in the work
rolls. After 100 seconds of rolling, the temperature distribution has a clear differ-
ence (up to 30 °C) between double and conventional cooling methods. The steady
state thermal evolution (Fig. 20.13) confirms that the thermal gradient produced
during one cycle is restricted to the surface of the roll. One more important phe-
nomenon can be observed in these figures, i. e. the temperature on the surface at
some angular positions is cooler than the core and in other locations hotter than
the core. This leads to tensile stresses at cooler locations and compressive stresses
at the hotter locations, which may cause the fire-cracks on the surface of the rolls.
Fig. 20.15 Penetration of temperature for different cooling methods

To understand the cooling efficiency and the resulting thermal effects the histo-
ry of the temperature penetration for various cooling strategies has to be ex-
amined. Figure 20.14 and 20.15 show the thermal evolution of the work rolls for
four different cooling methods at two different locations (9 and 16 mm inside).
From the Figures it is clear that the double cooling phenomenon slows down the
penetration of the thermal gradient and less heat is retained in the rolls after each
rotation.
20.5 Work Roll Wear Modeling and Simulation Results
The most important tribological effect, which influences the exit strip thickness
profile is work roll wear. During a cold rolling process, due to very high quality
tolerances, macroscopic wear of the work rolls may not be observed since the rolls
are changed frequently after a defined roll surface roughness. During hot rolling
process (particularly the continuous strip casting mill), however, the work roll
wear is more significant and mostly not uniform along the strip width [12,13].
Figure 20.16 shows a typical wear profile of a work roll [35], where the wear at
the edges of the strip is considerably higher than the middle. This might be be-
cause of the non-uniform temperature (particularly cooled edges) of the incoming
strip [34], which results in high contact pressure at the edges. The middle wear,
however, is relatively homogeneous. For the analysis of the wear profile, various
analytical wear models have been developed and are also being used for on-line
observations [11, 12, 13]. The major drawback of most of the models is that they
are not suitable for the prediction of local effects but useful to get the average val-
ue (along the strip width direction) of the wear [13, 14]. It is also necessary to un-
derstand the phenomenon of local wear for the better planning of the scheduling
and roll change.
Fig. 20.16 Typical work roll wear profile during hot rolling process (Stand 5)
Within this work, a decoupled modeling concept was introduced for the predic-
tion of local work roll wear during cold and hot rolling process. As described in
the concept and implementation section, three different wear models were imple-
mented to analyze the work roll wear effects on the rolling process. The models
were validated by the published data [13, 18]. In particular, Sikdar [18] published
hot rolling process information and the measured work roll wear for the finishing
stands. Within their work, the roll surface profile was measured before and after a
rolling schedule using a contact type profile gauge with a measuring accuracy of ±
1 μm. After rolling a selected number of strips (for the published case, 56 strips),
the worn work rolls were removed from the finishing mill and subjected to 2h of
water cooling followed by 3h of air cooling. This cooling schedule eliminated any
effects of thermal expansion on the work rolls. The other source used for the vali-
dation was published by Mohammed Tahir [13], where the wear of the work rolls
was evaluated for different roll materials for hot rolling finishing mills. According
to this input, various simulations were conducted to understand the work roll wear
using three different wear models and compared with the published data. The
work roll wear of the finishing stand (stand 4, top roll) of a tandem rolling mill
was further analyzed. The process conditions at this stand are provided in [18].
Three different strip widths (900, 700 and 500 mm) were simulated for the strip
length of 100 km each. The average simulated rolling force (for three strips) was
about 13.5 MN whereas the measured force ranges at about 12.5 MN [18].
For the analysis of wear using the empirical model (Eq. 6), the coefficients α, β
and γ were 1.0E-10, 0.188 and 0.11 respectively. To predict wear the normal pres-
sure is obtained from the process simulation and Equation 6 solved for each
surface element. As shown in Figure 20.17, for the work roll wear profile at the
centre, i. e. up to 500 mm, the wear is higher and uniform. The reason is that, for
all strip widths, the work roll is loaded at the centre. When using an empirical
wear model, the wear along the strip width is uniform; the reason for this could be
that the model considers only the normal contact pressure. Another cause may be
the inlet strip temperature, which is assumed to be uniform over the width, and
there is no considerable elastic deformation of the work rolls, which influence the
contact pressure profile. Figure 20.17 (right) shows the contact pressure distribu-
tion (for 700 mm strip width) before and after updating the worn roll profile with-
in the process simulation.
Fig. 20.17 Comparison of work roll wear for roll stand 4

Since the stresses due to friction are relatively high during the rolling process
any prediction neglecting these may lead to undesired results [17]. For that, an
energy-based wear model was implemented, which considers the frictional dissi-
pation rate. For each element, the accumulated wear work is estimated. The
required shear stress and the normal stresses are extracted from the process simu-
lation. The relative velocity in rolling direction is the difference of the roll velocity
and the strip velocity within the roll gap. Using this information, wear Equation 10
is solved for each surface element. As shown in Figure 20.17, the wear predicted
by the energy model is relatively local and higher than empirically estimated at the
middle, where high shear stresses may occur. The order of magnitude from both
models is very close to the measured wear profile. Further analysis and the valida-
tion of these models have to be conducted with real experimental data.
A new FE modeling concept was developed for the analysis of interaction effects
during the flat rolling process. Within this concept, the tool simulation is separated
from the process simulation and, using an automatic coupling module, the com-
munication between these two models was realized. The advantages of this model-
ing concept includes a short computation time (up to 2 times faster than a single
FE model in the case of an elastic tool model) compared to the single FE model,
suitable for the analysis of flatness problems (modeling of 3D), easy to handle
thermo-mechanical effects during the modeling of a hot rolling process and the in-
fluence of tribological effects (particularly abrasive wear of the work rolls) on the
strip quality can be analyzed. The simulation results of the cold rolling process
(with the careful selection of frequency of activation of the tool computation)
show good agreement with the single FE model and also the experiments. At the
same time, the modeling of the hot rolling process was also validated with the help
of lab experiments. The rolling force was clearly underestimated using a constant
gap conductance model, even with a high friction coefficient. When using the
pressure-dependent model, however, the predicted rolling force is very close to the
measured value. At the same time, the exit strip temperature and strip exit thick-
ness also matches well with the measurements. For the analysis of roll thermal
events for different cooling strategies, extensive numerical analysis was conducted
using the rotating boundary model. It was shown that the model is suitable and
easy to handle for the prediction of steady state thermal gradients within the work
rolls.
At the end, wear models were validated using published data. Since the history
of the rolled strip width is not mentioned within the publication, further analysis
and validation of the simulated results are necessary. Future work will consider
obtaining precise wear data from the industry to validate the developed models.
References
[1] Kainz, A., Krimplelstätter, K., Zemen, K.: FE Simulation of thin strip and temper
rolling processes. In: ABAQUS Austria User Conference (2003)
[2] Kim, T.H., Lee, W.H.: An integrated FE process model for the prediction of strip pro-
file in flat rolling. ISIJ 43, 1947–1956 (2003)
[3] Ohe, K., Kajiura, S., Simada, S., Mizuta, A., Morimoto, Y., Fujino, T., Anraku, K.:
Development of shape control in plate rolling. In: METEC Conference proceedings,
vol. 2, pp. 78–85 (1994)
[4] Buessler, P., Montmitonnent, P.: A review on theoretical analysis of rolling in Eu-
rope. ISIJ 31, 525–538 (1991)
[5] Guo, R.-M.: Development, verification and application of an optimal crown shape
control model for rolling mills with multiple control devices. In: AISE Conference
(1995)
[6] Tseng, A.A.: A numerical heat transfer analysis of strip rolling. ASME 106, 512–517
(1984)
[7] Hsu, C.T., Evans, R.W.: Finite element analysis on the hot rolling of steel. Adv. Tech.
Plasticity 2, 587–593 (1990)
[8] Sun, C.G., Hwang, S.M.: Prediction of roll thermal profile in hot strip rolling by the
Finite element method. ISIJ 40, 794–801 (2000)
[9] Sun, C.G., Hwang, S.M., Ryoo, S.R., Kwak, W.J.: An integrated FE process model
for precision analysis of thermo-mechanical behaviour of the rolls and strip in hot
strip rolling. Comp. Methods. Appl. Mech. Engg. 191, 4015–4033 (2002)
[10] Lee, J.H., Hwang, S.M., Park, H.D.: FE-based on-line model for the prediction of roll
force and roll power in hot strip rolling. ISIJ 40, 1013–1018 (2000)
[11] Jiang, Z.Y., Tieu, A.K.: Contact mechanics and work roll wear in cold rolling of thin
strip. Wear Journal 263, 1447–1453 (2007)
[12] Magne, A., Gaspard, C., Gabriel, M.: Wear behaviour of steels for hot working roll-
ing-mill rolls. CRM 57, 25–39 (1980)
[13] Mohammed, T., Widell, B.: Roll wear evaluation of HSS, HiCr and IC work rolls in
hot strip mill. Steel Research 74, 624–630 (2003)
[14] Kivilcim, E.N., Nürnberg, G., Golle, M., Hoffmann, H.: Simulation of wear on sheet
metal forming tools—An energy approach. Journal of Wear, 357–363 (2008)
[15] Archard, J.F., Hirst, W.: Wear of metals under un-lubricated conditions. Proceedings
of the Royal Society of London 236, 397–410 (1956)
[16] Byon, S.M., Kim, S.I., Lee, Y.: A semi-analytical model for predicting the wear con-
tour in rod rolling process. Journal of Materials Processing Technology 191, 306–309
(2007)
[17] Bowden, F.P., Tabor, D.: Friction, lubrication and wear: a survey of work during the
last decade. British Journal of App. Physics 17, 1521–1544 (1966)
[18] John, S., Sikdar, S., Mukhopadhyay, A.: Roll wear prediction model for finishing
stands of hot strip mill. Iron and Steel Making 33, 169–175 (2006)
[19] Franzke, M., Puchhala, S., Dackweiler, H.: Modeling of interaction effects between
strip and roll during flat rolling process. In: NUMIFORM Conference Proceedings,
vol. 908, pp. 1489–1494 (2007)
[20] Franzke, M., Puchhala, S., Dackweiler, H.: Modeling of interaction effects between
process and machine during flat rolling process. In: PMI Conference Proceedings
(2008)
[21] Tseng, A.A., Huang, C.H.: The estimation of surface thermal behavior of the working
roll in hot rolling process. Heat and Mass Transfer 18, 1019–1031 (1995)
[22] Tseng, A.: Thermal modeling of roll and strip interface in rolling process: part 1 – re-
view. Numerical Heat Transfer 35, 115–135 (1999)
[23] Tseng, A.: Thermal modeling of roll and strip interface in rolling process: part 2 – re-
view. Numerical Heat Transfer 35, 135–154 (1999)
[24] Mikic, B.B.: Thermal contact conductance: Theoretical considerations. International
Heat Transfer 17, 205–214 (1974)
[25] Yovanovich, M.M., De Vaal, J., Hegazy, A.: A statistical model to predict thermal
gap conductance between conforming rough surfaces. AIAA Paper No: 82-0888
(1998)
[26] Colas, R., Torres, M.: Modeling heat conduction through an oxide layer during hot
rolling of steel. ASME Manufacturing Science Engineering 68(2), 577–582 (1994)
[27] Browne, K.M., Dryden, J., Assefpour, M.: Modeling scaling and de-scaling in hot
strip mills. Recent Advances in Heat Transfer and Micro Structure Modeling for Met-
al Processing 67, 187–201 (1995)
[28] Ranta, H., Larkoila, J., Korhonen, A.S.: A study of scale effects during accelerated
cooling. In: Modeling of Metal Rolling Process Conference Proceedings, pp. 638–647
(1993)
[29] Tseng, A.A.: A numerical heat transfer analysis of strip rolling. ASME Journal of
Heat Transfer 106, 512–517 (1984)
[30] Sellars, C.M., McLaern, A.J.: Modeling distribution of microstructure during hot roll-
ing of stainless steel. Materials Science and Technology 8, 1090–1095 (1975)
[31] Militzer, M., Nakata, N.: Modeling of microstructure evolution during hot rolling of a
780 MPa high strength steel. ISIJ 45, 82–90 (2005)
[32] Tseng, A.A., Lin, F.H., Gunderia, A.S., Ni, D.S.: Roll cooling and its relationship to
roll life. Metallurgical Transactions 20A, 2305–2320 (1988)
[33] Sun, C.G., Hwang, S.M.: Prediction of roll thermal profile in hot strip rolling by the
FE Method. ISIJ 40, 794–801 (2000)
[34] Azene, Y.T., Roy, R., Farrugia, D., Onisa, C., Trumann, M.H.: Work roll cooling sys-
tem design optimization in presence of uncertainty. In: CIRP Design Conference Pro-
ceedings, pp. 57–65 (2009)
[35] Partender, E., Mitter, S.: Ein Modell zur Prognose des Arbeitswalzenverschleiß der
Breibandstraße in Linz. In: XXII Verformungskundlisches Kolloquium (2003)
[36] Simulia, http://www.simulia.com/products/abaqus_fea.html
[37] Lasso GmBh, http://www.lasso.de/index.php?id=28
[38] GOM, Optische 3D-Koordinatenmessmaschine
http://www.gom.com/de/messsysteme/
systemuebersicht/tritop.html
[39] Franzke, M.: Zielgrößenadaptierte Netzdiagnose und -generierung zur Anwendung
der Finite Elemente Methode in der Umformtechnik. Dissertation, RWTH Aachen
University (1999)
Chapter 21
Increase of the Dimensional Accuracy of Sheet
Metal Parts Utilizing a Model-Based Path
Planning for Robot-Based Incremental Forming
H. Meier, S. Reese, Y. Kiliclar, and R. Laurischkat
Abstract. The principle of robot-based incremental sheet metal forming is based

on flexible shaping by means of a freely programmable path-synchronous move-
ment of two tools, which are operated by two industrial robots. The final shape is
produced by the incremental infeed of the forming tool in depth direction and its
movement along the geometry’s contour in lateral direction. The main problem
during the forming process is the influence on the dimensional accuracy resulting
from the compliance of the involved machine structures and the spring-back ef-
fects of the workpiece. The project aims to predict these deviations caused by
compliances and carry out a compensative path planning based on this prediction.
Finite element analysis using a material model developed at the Institute of Ap-
plied Mechanics (IFAM) [1] has been used for the simulation of the forming
process.
21.1 Introduction
In the field of sheet metal forming, the production of prototypes or parts in low
batch sizes is being realized by high-priced conventional forming processes. The
lack of flexibility as well as high investment costs clearly point out the limits of
conventional sheet metal forming for these parts. Recently, different kinematics-
based sheet metal forming processes have been developed. In most cases, these
methods only require simple dies or supports ([2], [3]).
In this regard, the Chair of Production Systems (LPS) at the Ruhr-University of
Bochum has developed a robot-based incremental sheet metal forming process
(Roboforming) [4]. The Roboforming principle is based on flexible shaping by
means of freely programmable path-synchronous movements of two industrial 6-
axis robots driving universal workpiece-independent forming tools. Figure 21.1a
shows the machine setup, which consists of two KUKA KR360 robots and a frame
with a blank holder. The final shape is produced by the incremental inward motion
of the forming tool in depth direction and its movement along the contour in later-
al direction by driving either on parallel layers or on a helical path. The supporting
tool, also being realized with a simple geometry, holds the sheet on the backside
by moving synchronously along the outer contour, constantly on the same level or
460 H. Meier et al.
directly opposed to the forming tool. Thereby, a pre-defined gap between the two
hemispherical tools is being generated (Figure 21.1b).
Due to the high flexibility regarding the reachable geometries due to the robots’
kinematics and the tools’ simplicity, Roboforming provides a rather interesting
method for industrial prototyping and low batch size production. Apart from free-
form surfaces (Figure 21.1c), it is possible to form undercuts in a multi-step strat-
egy in a single clamping with the two 6-axis kinematics (Figure 21.1d).
Fig. 21.1 a) Roboforming setup, b) basic forming strategies, c) free-form surface and d) cy-
linder with undercut (97° wall angle) [5]
Nevertheless, up to now, the resultant geometries are showing significant dev-

iations from the planned geometries, especially when the tool path is planned on
the CAD model without any adaptations.
The forming strategy with its forming parameters has a main influence on the
process result. Depending on the material and the geometry being formed, para-
meters like the infeed, the forming velocity or the tool path especially affect the
part’s accuracy. Additionally, the sheet is subject to springback. While these two
effects emerge in the context of all incremental sheet metal forming processes,
Roboforming is furthermore subject to robot influences.
Since traditional applications of industrial robots consist of repetitive opera-
tions such as spot welding, the robot development of the last years has been fo-
cused on increasing the repeat accuracy, which nowadays is better than 0.1 mm
for certain robots. However, the absolute positioning accuracy of such a robot is
not better than several tenth of a millimeter. Furthermore, this accuracy is only
guaranteed under the condition that no external forces act on the robot effector.
For this reason, in Roboforming, process forces near the payload of the robot lead
to tool deviations of up to 3 mm.
Figure 21.2 exemplarily shows the process forces and tool deviations for the
forming of a frustum geometry. The forces show circumferential homogeneous
characteristics in all directions (radial, tangential and infeed) during the whole
forming process and are highly reproducible, as several tests have shown. The tool
deviation shows a contrary behavior (Figure 21.2d). While the absolute tool path
deviation for angles φ about 170 deg amounts to 1.1 to 1.3 mm, deviations of more
than 2.2 mm are measured in the opposite area (angle φ about 330 deg). This non-
linear stiffness of the robot structure is mainly caused by the interaction of the
21 Increase of the Dimensional Accuracy of Sheet Metal Parts 461
serial kinematic posture and the elasticities of its six joints. Each joint elasticity is
composed of elasticities and clearances in the gears and bearings, which are not
detected by the rotary encoders of the motors and therefore not compensated by
the cascaded control structure. To handle these deviations an entire model-based
method is developed, where no additional sensors or other modifications to the ro-
bot hardware are necessary. The implementation and experimental validation of
this model-based compensation method are described in this chapter.
Fig. 21.2 Measured process forces in a) radial, b) tangential, c) infeed direction and d) re-
sulting absolute tool path deviation for a frustum geometry (Ø90x40mm x 60°) produced
by roboforming with an incremental infeed of 1 mm. (material: DC06, thickness: 0.8 mm)
21.2 Coupled Process Structure Model

The objective of the current work is the prediction and compensation of the path
deviations in Roboforming resulting from the low stiffness behavior of industrial
robots. Therefore, the resulting forming forces in the sheet metal have to be com-
puted and applied to the robots’ kinematics. The reaction forces in the sheet metal
are currently determined using a finite element (FE) approach, while the simula-
tion of the robots’ kinematics is carried out by means of a multi-body system
(MBS) simulation. Figure 21.3 shows the general idea of the procedure.
In a first step, the process forces are computed in the finite element model as-
suming an ideal stiff robot by moving the tool along the planned reference path.
These forces, combined with the reference path, build the input for the multi-body
462 H. Meier et al.
system modeling the robot. Here, the actual tool path and the robot compliance is
obtained due to the previously computed forces. Once the actual path is known,
the path correction compensating the force-dependent deviations is executed and
the correction data-set is derived. It contains the path, which has to be driven to
reach the reference path. When compiled and transferred to the robot control unit,
the corrected tool path allows the production of a workpiece without robot-
influenced errors.
Fig. 21.3 Path generation concept including robot compliance compensation [5]
21.2.1 Process Model

With increasing availability of fast computer hardware, the finite element method
is used more and more efficiently to predict and simulate material and structural
behavior. First of all, the use of a purely isotropic hardening model (expansion of
the yield surface) is not sufficient. Therefore, a finite strain constitutive model,
which combines non-linear kinematic and isotropic hardening, based on the mul-
tiplicative split of the deformation gradient, developed in a recent work by Vladi-
mirov et al. [6], is used in this work. The papers Choi et al. [7], Dettmer & Reese
[8], Hakansson et al. [9], Menzel et al. [10], Svendsen et al. [11] and Wallin et al.
[12] are dealing with this subject. In order to reduce computational time a new ef-
ficient finite element technology is developed. Nevertheless, the simulation of Ro-
boforming remains a difficult task. The computation time is extremely high for
implicit simulations due to the complex contact behavior.
21.2.1.1 Material Modeling

The constitutive model is based on the multiplicative split Fp = Fpe Fpi of the plas-
tic deformation gradient into “elastic” and “inelastic” parts, F = Fe Fp being the
classical multiplicative split of F . As a result, a continuum mechanical extension
of the classical rheological model of Armstrong-Frederick kinematic hardening
[13] can be achieved. The Helmholtz free energy per unit of undeformed volume
ψ is additively decomposed into the three parts ψ = ψ e ( Ce ) + ψ kin ( C pe ) + ψ iso (κ ) .

The first part ψ e describes the macroscopic elastic material properties. The
second term ψ kin corresponds to the elastic energy stored in dislocation fields due
to kinematic hardening and vanishes, if the kinematic hardening is zero. The third
term represents elastic energy due to isotropic hardening, where is the isotropic
hardening variable. The Helmholtz free energy is a function of the elastic right
Cauchy-Green tensor Ce = FeT Fe = Fp−T C Fp−1 and the elastic part of the plastic
right Cauchy-Green tensor is defined as C pe = FpeT Fpe = Fpi−T C Fpi−1 . Inserting this
into the Clausius-Duhem inequality −ψ + S ⋅ (1 / 2 ) C ≥ 0 results in a relation for
the second Piola-Kirchhoff stress tensor S . The derivation of the material model
is suitably carried out in the intermediate configuration. For the numerical imple-
mentation of the constitutive equations it is, however, more appropriate to work in
the undeformed or reference configuration.
The set of constitutive equations of the model in the reference configuration is
summarized below [1]:
Stress tensors
∂ψ e −T ∂ψ kin −T
S = 2 Fp−1 Fp , X = 2 Fpi−1 Fpi , Y = C S − CP X , Ykin = C p X (21.1)
∂Ce ∂C pe
Evolution equations
Y DC p b 2
C p = 2 λ , C pi = 2 λ YkinD C pi , κ = λ (21.2)
Y ⋅ (Y )
D D T c 3
Yield function
Φ = Y D ⋅ (Y D )T −
2
3
(σ y − R ) , R = −Q (1 − e− βκ ) (21.3)
Kuhn-Tucker conditions
λ ≥ 0, Φ ≤ 0, λ Φ = 0 (21.4)
21.2.1.2 Finite Element Technology
Recent research has been focusing on the large deformation version of a new
eight-node solid-shell finite element based on reduced integration with hourglass
stabilization [14]. The major problem of low-order finite-elements, used to simu-
late thin structures as the sheet metal, is locking, a non-physical stiffening effect.
Therefore, in our recent solid-shell formulation the enhanced assumed strain
(EAS) as well as the assumed natural strain (ANS) concepts are implemented to
circumvent locking. To cure transverse shear-locking, the corresponding trans-
verse shear terms evaluated at locking-free sampling points are interpolated within
464 H. Meier et al.
the element domain. The same procedure applied to the transverse normal strain
cures curvature thickness locking. For the EAS concept, the derivation is based on
the well-established two-field variational functional
g1 (u , Ee ) =  S ( E ) : δ Ec dV + g ext = 0 (21.5)
B0
g 2 (u , Ee ) =  S ( E ) : δ Ee dV = 0 (21.6)
B0
which depends on the displacement vector u and the enhanced Green-Lagrange

strain tensor . In this enhanced strain tensor, the strain in thickness direction is
enriched linearly. is the compatible Green-Lagrange strain tensor. Hence, the
EAS concept simplifies to a scalar equation, which leads to an efficient and robust
element formulation in combination with the ANS concept. Further important key
points to improve the efficiency of the hourglass stabilization are different Taylor
expansions with respect to the shell director. The hourglass kernel depends on
the material behavior in the integration point. This new finite element technology
has been implemented into ABAQUS as user element sub-routine (UEL).
21.2.2 Robot Model including Compliance Compensation

Figure 21.4 shows the model structure for compensating the tool path deflections
caused by the process forces occurring at the tool. It consists of several specific
robot sub-models in combination with a controller. In a first step, the reference
position and reference orientation (xr, yr, zr, Ar, Br, Cr) of the tool are transformed
from the local workpiece coordinate system to the robot coordinate system (angles
ϕ1 … ϕ6 of the six axes) via an inverse rigid robot model. These robot coordinates
in combination with the process forces (Fx, Fy, Fz) build the input of the compliant
robot forward model, which computes the deflected position and orientation of the
tool (xd, yd, zd, Ad, Bd, Cd). The error between the reference and the deflected posi-
tion is minimized by a controller affecting the robot coordinates. Using these
coordinates (ϕ1`… ϕ6`) as the input of a rigid robot forward model allows the cal-
culation of the position and orientation, which compensate the process force-
dependent deflection.
xr, yr, zr, xc, yc, zc,

Fx, Fy, Fz
A r, B r, C r xd, yd, zd, Ac, Bc, Cc
Ad, Bd, Cd
∑
Rigid Robot Compliant Robot Rigid Robot

Controller
Inverse Model Forward Model Forward Model
∑ ’ ’
Fig. 21.4 Simplified structure of the robot model including compliance compensation
The complete model is set up in SimMechanics, which is a toolbox completely

integrated into the Matlab/Simulink environment. The design and implementation
of each sub-model is described in the following section.
21.2.2.1 Rigid Robot Inverse Model
The calculation of the joint angles necessary to reach a specified end-effector posi-
tion requires an inverse kinematic solution. In the implemented model, a numeri-
cal solution based on the Denavit-Hartenberg convention [15] for the description
of the kinematical structure is used. It makes use of the pseudo-inverse of the ma-
nipulator Jakobian to iteratively calculate the six joint angles of the robot for a
given position and orientation of the TCP. [16]
21.2.2.2 Compliant Robot Forward Model
The examined serial robot KUKA KR360 is an open-ended structure consisting of

six links, which are connected by rotational joints in series. The parts of the robot
with the largest impact on overall positioning accuracy have been identified to be
the elasticities in the joints and gears (compare [17], [18]). As the robot links can
be assumed to be stiff the mechanical behavior of the forming tool driving robot is
simulated by means of a multi-body system. The goal of the MBS simulation is to
calculate the deviation of the tool center point depending on the process forces at
the tool tip and the robots’ posture. The basic robot model is a tree-structured
MBS built up in SimMechanics. All kinematic and kinetic parameters of the robot
– such as lengths, masses, centers of masses and inertias of the links – are taken
from manufacturer data sheets or estimated. As the forming velocity is low a qua-
si-static computation is used and therefore motor-rotor inertias are not taken into
account.
To consider the robot’s compliance the standard stiff-joint model for the robot
is extended to flexible joints. The first five joints are each added by two
tilting elasticity elements and one elasticity element in the driven rotational direc-
tion. For joint six, only elasticity in tilting direction is considered because the
rotation-symmetric tool is not exposed to any torque in rotational direction. Thus,
the complete compliant robot model is described by 17 separate joint-stiffness
characteristics.
Figure 21.5 shows a sample joint able to simulate its forced motion as well as
its specific compliance characteristics, as modeled in SimMechanics. Therefore, a
massless dummy is added, which allows the division into a driven part and a com-
pliant part. The driven part connects the dummy with the previous robot link i
through the specified link-connecting revolute joint Joint_i_rigid. The angular
motion, which drives this joint, is calculated by the inverse model of the rigid ro-
bot. There is no spring stiffness on this part and its motion is not influenced by
forces. The compliant part consists of a spherical joint (Joint_i_flex) connecting
the dummy to the next link i+1. While the spherical joint allows rotations in all
three rotational degrees of freedom (DOF) a restoring torque is generated depend-
ing on the torsion angle and torsion velocity of the spherical joint for each of the
466 H. Meier et al.
Fig. 21.5 Compliant joint model
three DOF. The detail in Figure 21.5 exemplarily shows the design of the steady
damping and the torsion angle-dependent stiffness function for one tilting direc-
tion. Element Joint_i_stiffness_R3 considers the required stiffness characteristic as
shown exemplarily in Figure 21.6. All 17 stiffness characteristics were identified
through appropriate experiments described in chapter 21.2.2.4.
21.2.2.3 Rigid Robot Forward Model
The rigid forward model of the robot is the same as the compliant forward model
but without the compliant part of the joints and the required dummies.
21.2.2.4 Experimental Identification of Stiffness Parameters
In order to measure the stiffness characteristics of all six joints in the 3 rotational
DOF an optical 3-D motion and position measurement system was used. The Me-
tris K600 system measures a position and orientation of a marker, which is
attached to at least 3 infrared ray emitting LEDs by use of 3 CCD cameras at an
volumetric accuracy (+/-2σ) of 90 μm. The system also allows the measurement
of relative movements between two objects by preparing both objects with LEDs.
This feature was used to measure the torsion angle of each robot joint under a de-
fined load by measuring the relative movement between the previous and the fol-
lowing link. The load was applied by a pneumatic cylinder, which was mounted
on the effector of the measured robot. Supporting the pressurized cylinder by the
second robot, a defined static force with varying amplitudes was applied to the
measured robot. To apply high torques in a defined direction to each joint opti-
mized postures of the robot were used. This procedure has been significantly
assuaged due to the use of the second robot as the support can easily be located in
the workspace. The cylinder force was measured by means of a force torque sen-
sor (accuracy: ~5 % of full-scale load) mounted between the cylinder and the ro-
bots’ effector. The relevant torque was computed considering the distance
between the cylinder and the analyzed joint.
b)
a)
Fig. 21.6 Identification of the stiffness characteristic of axis 1 in rotational direction: Expe-
rimental setup and resulting stiffness characteristic Torque vs. Torsion Angle
Figure 21.6 exemplarily shows the experimental setup to identify the stiffness
characteristic of the first axis in rotational direction. The robot was set to its initial
posture to maximize the arm lever and therefore the applied torque and to realize
an exclusive deflection in the DOF of interest. Three LEDs were attached to the
carrousel to measure its deflection against the ground. Therefore, there was no
need for additional LEDs. At steps of 100 N, the force was increased up to a max-
imum of 1,400 N, decreased to 0 N, furthermore decreased up to -1,600 N and
increased up to 0 N again. Special care was taken to ensure that the brakes of the
robot remain unclamped during the experiment. The simplified characteristic
shown in Figure 21.6 was derived considering the arm lever of about 3,300 mm.
21.3 Experimental Model Validation

The validation of the entire process structure model was carried out separately for
each sub-model. To evaluate the accuracy of each model one typical forming
process was performed in an experiment, which provides all required measure-
ment data.
21.3.1 Experimental Setup

To validate the process structure model and the compensation approach an
experiment was executed to form the frustum geometry shown in Figure 21.7
(geometry: Ø90 mm x 40 mm x 60°, material: DC06, thickness: 0.8 mm). On the
468 H. Meier et al.
one hand, this geometry can be simply described analytical and allows an easy in-
terpretation of the experimental results. On the other hand, it represents the typical
forming strategy of geometries with Roboforming, where the effector mostly
moves with ring-like paths. As shown in Figure 21.7b, in the experiment, the ef-
fector moves along a helical path from outside to inside and at the same time
gradually increases the processing depth. The tool velocity v is set to 25 mm/s.
During the forming process, the actual TCP path was measured by an optical mea-
suring device, which detects the position of LEDs applied to the robot’s effector
(volumetric accuracy (+/-2σ): 90 μm). The process forces were measured using a
force-torque sensor located between robot effector and tool (accuracy: ~5 % of
full-scale load) (Figure 21.7a). Furthermore, the interpolated reference path and
the motion of each robot joint, measured by the rotary encoders and therefore not
including the elasticities of the axes, were logged by the robots’ control unit.
Fig. 21.7 a) Experimental setup, b) Helical tool path
The description of the tool path for the programming of the robot and for the
simulation model as well as the tool position measurement are based on the local
Cartesian coordinate system CS located in the centre of the geometry, as shown in
Figure 21.7b. For a better presentation of the results, the measured and simulated
paths were transformed to the basis ϕ, which describes the angle between each
point of the path and the starting point of the forming process. ϕ is set to 0° at the
starting point of the forming process and continuously increases along the helical
tool path. The total rotation of the path during the whole forming process was
about 19,000°, which equals to about 53 revolutions. Based on the independent
variable ϕ, the vectors z , x2 + y2 , and x 2 + y 2 + z 2 can be used to describe the
tool position.
21.3.2 Process Model Validation

The Roboforming process is simulated by means of a finite element analysis per-
formed in ABAQUS/Standard. In order to estimate the performance of the model
an experiment was conducted using a 220x220 mm² fully-clamped DC06 metal
sheet with a thickness of 0.8 mm. The forming tool has a tool tip radius of 6 mm.
The Young’s Modulus is 206,000 MPa and the Poisson’s ratio is 0.3. We used a
path obtained from the real-time path (optical measuring device Metris K600) of
21 Increase of the Dimensio
onal Accuracy of Sheet Metal Parts 4669
the robots. The x and y-ax xes describe the plane of the sheet. The z-axis is used foor
the thickness direction.
The time-dependent paath is given by x(t), y(t) and z(t). These coordinates oveer
time are translated into a user sub-routine to specify prescribed boundary conddi-
tion (DISP). At this, the influence
i of acceleration and velocity are neglected. Thhe
simulation of a complicatted forming process like Roboforming requires a realisttic
description of the materiaal behavior. As presented above, a large deformation elaas-
to-plastic material model with combined non-linear kinematic and isotropic haar-
dening is used to handle the springback effect of the sheet. It should describe thhe
Bauschinger and ratchetin ng effect, which belongs to the most characteristic phee-
nomena of the hardening behavior of metals. The Bauschinger effect is defined bby
the fact that straining in one
o direction reduces the yield stress in the opposite ddi-
rection, while ratcheting denotes the progressive increase of the mean strain as a
result of unsymmetrical stress cycles. Due to the fact that the contact behaviour is
very complex the solution n time is dominated by the number of contact iterationns,
which is relatively high in
n this simulation. Therefore, the contact formulation is aan
important aspect. Here, thhe forming and supporting tools are modeled as rigid boo-
dies. The interaction betwween them and the deformable sheet metal is defined bby
a penalty surface-to-surfaace contact method with finite sliding. The friction coefffi-
cient is 0.05. The resultss shown in this chapter are computed using an 8-nodde
continuum-shell element with reduced integration (SC8R). The equivalent plasttic
strain of the deformed sheet metal for an infeed of 20.04 mm can be seen iin
Figures 21.8a and b.
Fig. 21.8 a) Equivalent plasttic strain (FE-Simulation) b) Maximum infeed
The process forces fromm the simulation of the sheet metal forming applying thhe
measured tool path of thee experiment are compared to the process forces provideed
by the experiment. Due to o the lack of available experiments to fit the material pa-
rameters differences are to
o be expected. Moreover, the simulation of one compleete
forming process of up to 20 mm in the infeed direction is very time-consuming.
A more efficient way is required
r to apply the correct material parameters for thhe
simulation.
470 H. Meier et al.
21.3.3 Robot Model Validation

The validation of the robot model and the compensation approach was conducted
by use of the experiment described in Chapter 21.3.1. Furthermore, it was vali-
dated step-wise along its sub-models shown in Figure 21.4.
In the first step, the inverse model of the rigid robot was validated by compar-
ing the results of the axis-angle calculations of the simulation model and the robot
controller for arbitrary tool paths. It could be determined that the accuracy of the
simulation results only depends on the accuracy and properties of the selected ite-
ration and solver parameters. The simulation model agrees with the robot control-
ler at an accuracy of 1e-4 degrees or better.
In the next step, the accuracy of the unloaded robot was analyzed to obtain a
reference for the following validation steps. Therefore, the robot effector was
moved along the given helical tool path but without a sheet metal in the clamping
device. Figure 21.9 shows the experimental results as the position of the TCP ex-
pressed as vector x 2 + y 2 + z 2 vs. ϕ. Whenomparing the measured path (dashed
green) with the reference path (solid black), periodical oscillations with a maximal
amplitude of about 0.6 mm and a period of 360° can be detected. These deviations
are within the absolute path accuracy specification of the robot manufacturer.
They are caused by several reasons, as described in [19]. For example, the unstea-
dy course of the reference path, which is logged by the robot’s control unit, shows
errors occurring during the interpolation.
Fig. 21.9 Comparison of the reference path and the unloaded measured path to evaluate the
robots’ path accuracy. Comparison of the simulation results and experimental results for the
loaded robot to evaluate the robot model accuracy.
To validate the compliant robot model the experiment was repeated with the
same tool path but with a sheet metal placed in the clamping device. Based on the
measured process forces (Figure 21.2) during this experiment, a simulation was
executed to calculate the deviated path. Figure 21.9 shows the results of both the
loaded robot experiment (dotted blue) and the simulation of the loaded robot (dash
dotted red). Considering the result of the real robot, in the first half of the experi-
ment, the average deviation increases up to 1.1 mm and the oscillation amplitude
enlarges up to 1.4 mm for the loaded robot. During the second half of the experi-
ment, these deviations stay constant. These characteristics are caused by the
process forces in all directions, which increase up to the middle of the experiment
and the rotation of the forces in radial and tangential direction about Z at a fre-
quency of ϕ = 360°. Further investigations of the results (not shown in the
diagram) have shown that the portion of the deviation in radial and tangential
direction is noticeably higher than in infeed direction, which is caused by the
higher stiffness characteristics of the robot in infeed direction. In conclusion, the
comparison of the simulated path and the experimental path shows a very good
qualitative and quantitative agreement. The maximal difference is about 0.3 mm.
Hence, the compliant robot model shows a behavior close to reality.
For the validation of the compensation approach, the compensation path al-
ready calculated in the previous simulation was used in another forming experi-
ment. Figure 21.10 shows the calculated compensation path (dashed green) and
the resulting measured path of the experiment (dash dotted red). With reference to
Fig. 21.10 Comparison of the resulting tool path before and after compensation to evaluate
the performance of the compensation approach
472 H. Meier et al.
the reference path (solid black), the compensation path is nearly symmetrical to
the uncompensated path of the loaded robot (dotted blue). Considering the meas-
ured tool path of the final experiment it can be noticed that it matches the refer-
ence path very well. Not only the average deviations but also the oscillations have
been reduced in all directions. A decrease from 1.1 mm to 0.1 mm for the average
deviation and from 1.4 mm to 0.8 mm for the oscillation amplitude can be ob-
served. These remaining oscillations are mostly the results of the robot´s inaccura-
cy itself, as already shown in Figure 21.9.
21.5 Conclusion
The chapter shows that the presented model-based approach has a high potential to
significantly increase the geometrical accuracy in Roboforming. The demonstrated
experimental results document a high efficiency of the compensation method. In
order to complete the model-based approach further optimization and the integra-
tion of the FEM model are necessary. Due to the fact that the computation time is
still very high and path deviations still exist, different investigations still have to
be carried out. With the presented material model we can further investigate the
influence of the kinematic hardening on the compliance of the sheet metal and the
springback. This is expected to provide significant impact on the simulation re-
sults. Additionally, the contact algorithms have to be improved. Finally, the ap-
proach has to be tested with complex geometries.
References
[1] Vladimirov, I.N., Pietryga, M.P., Reese, S.: On the modelling of nonlinear-kinematic
hardening at finite strains with application to springback – comparison of time inte-
gration algorithms. International Journal for Numerical Methods in Engineering 75,
1–28 (2008)
[2] Douflou, J., Szekeres, A., Vanherck, P.: Force Measurements for Single Point Incre-
mental Forming: A Experimental Study. In: Proceedings of the 11th Int. Conference
on Sheet Metal, pp. 441–448 (2005)
[3] Jeswiet, J., Micari, F., Hirt, G., Bramley, A., Douflou, J., Allwood, J.: Asymmetric
Single Point Forming of Sheet Metal. In: Proceedings of the 55th CIRP General As-
sembly in Antalya, pp. 88–114 (2005) (in Antalya)
[4] Meier, H., Smukala, V., Dewald, O., Zhang, J.: Two Point Incremental Forming with
Two Moving Forming Tools. In: Proceedings of the 12th International Conference on
Sheet Metal, pp. 599–605 (2007)
[5] Meier, H., Laurischkat, R., Zhu, J.: A Model Based Approach to Increase the Part
Accuracy in Robot based Incremental Sheet Metal Forming. In: AIP Conference Pro-
ceedings, vol. 1315, pp. 1407–1412 (2010)
[6] Vladimirov, I.N.: Anisotropic material modelling with application to sheet metal
forming RWTH Aachen, Institut für Angewandte Mechanik, Technische Universität
Braunschweig, Dissertation (2009)
[7] Choi, Y., Han, C.S., Lee, J.K., Wagoner, R.: Modeling multi-axial deformation of
planar anisotropic elasto-plastic materials. part I: Theory, International Journal of
Plasticity 22, 1745–1764 (2006)
[8] Dettmer, W., Reese, S.: On the theoretical and numerical modelling ofArmstrong-
Frederick kinematic hardening in the finite strain regime. Computer Methods in Ap-
plied Mechanics and Engineering 193, 87–116 (2004)
[9] Hakansson, P., Wallin, M., Ristinmaa, M.: Comparison of isotropic hardening and ki-
nematic hardening in thermoplasticity. International Journal of Plasticity 21, 1435–
1460 (2005)
[10] Menzel, A., Ekh, M., Runesson, K., Steinmann, P.: A framework for multiplicative
elastoplasticity with kinematic hardering coupled to anisotropic damage. International
Journal of Plasticity 21, 397–434 (2005)
[11] Svendsen, B., Levkovitch, V., Wang, J., Reusch, F., Reese, S.: Application of the
concept of evolving structure tensors to the modeling of initial and induced anisotro-
py at large deformation. Computers & Structures 84, 1077–1085 (2006)
[12] Wallin, M., Ristinmaa, M.: Deformation gradient based kinematic hardening model.
International Journal of Plasticity 21, 2025–2050 (2005)
[13] Armstrong, F., PJ, CO: A mathematical representation of the multiaxial Bauschinger
effect, C.E.G.B. Report RD/B/N731, Berkeley Nuclear Laboratories, Berkeley, U.K
(1966)
[14] Schwarze, M., Reese, S.: A reduced integration solid-shell finite element based on the
EAS and the ANS concept – geometrically linear problems. International Journal for
Numerical Methods in Engineering 80, 1322–1355 (2009)
[15] Hartenberg, R.S., Denavit, J.: A kinematic notation for lower pair mechanisms based
on matrices. Journal of Applied Mechanics 77, 215–221 (1955)
[16] Corke, P.I.: A Robotics Toolbox for MATLAB. IEEE Robotics and Automation
Magazine 3(1), 24–32 (1996)
[17] Gerstmann, U.: Robotergenauigkeit: Der Getriebeeinfluss auf die Arbeits- und Posi-
tionsgenauigkeit, Dissertation, Universität Hannover (1997)
[18] Pham, M.T., Gautier, M., Poignet, P.: Identification of Joint Stiffness with Bandpass
Filtering. In: Proceedings of the IEEE Int. Conf. on Robotics & Automation, pp.
2867–2872 (2001)
[19] Beyer, L.: Genauigkeitssteigerung von Industrierobotern, Dissertation, Universität der
Bundeswehr Hamburg (2004)
Chapter 22
Gear Rolling Process
R. Neugebauer, U. Hellfritzsch, M. Lahl,

M. Milbrandt, S. Schiller, and T. Druwe
Abstract. The rolling process is an efficient alternative to currently exclusively

applied cut-ting processes for the production of high gears, particularly regarding
economic and ecological aspects. The manufacturing of involute gear profiles by
forming, specifically by rolling, has several advantages in comparison to cutting
methods, e. g. significantly shorter process times, no material loss and subsequent-
ly no chip disposal, strength increase in the forming zone and a high surface quali-
ty. Due to these characteristics gear forming will continue to gain relevance in
future gear manufacturing. The Chapter presents the efforts being made at the
Fraunhofer IWU Chemnitz to reach an advancement of the high gear rolling
process by improving the gearing qualities. It presents the investigation and analy-
sis of the interaction between tools, machine and forming process in gear rolling.
The results of measurements on the displacements of workpiece, tools and clamp-
ing device during the rolling process of a high gearing are displayed. The setup
and results of accompanying simulation of the forming process are also given.
22.1 Introduction
The gear rolling process with round tools is one example of cold bulk metal
forming technologies. Cold rolling of high gearings is a new technique with new
requirements on the rolling equipment and process. Initially, pitch-related ques-
tions are of primary importance. The profile depth and the pitch size of high gear-
ings are larger than those of stub tooth and normal gearings. This results in the
displacement extent of the rolling tool by about half of the initial pitch value and a
changing pitch value during the process (initial and calibration pitch). Due to the
occurring strain hardening during cold forming the material is characterized by an
enormous flow stress that generates high stress rates during the forming process.
Resulting effects are elastic deformations of the rolling tools and a significant dis-
placement of machine components. The process forces affect the structure of the
machine, cause deviations in the rolling kinematics and influence the forming
forces that need to be generated by the machine. This complex interaction thus re-
quires a combined consideration of the forming process and structure of the ma-
chine in terms of rolling process cause-and-effect relationships as well as reactions
of involved machine components. The forming result largely depends on these
factors (Figure 22.1).
476 R. Neugebauer et al.
Fig. 22.1 Influencing factors on the gear quality
Existing machine and process design procedures applied to high gearings insuf-
ficiently comply with the rolling standards DIN 3960, 3961 and 3962. The main
reason is the inadequate understanding of interactions and relations between ma-
chine structure / components and forming process [1]. The aim of the presented
investigations was to determine the influence of machine components and tool
displacements in the rolling process on the gear quality of rolled high gearings.
The results lead to a better understanding of the system machine-process-tool and
can be used to enhance the quality of rolled high gears by implanting them in the
tool, process and machine design operations.
22.2.1 Gear Rolling Technique

The gear rolling technique, used by the IWU Chemnitz, is a cross rolling method
applied for profile or thread geometries. The two round tools are toothed outside
with a symmetric contour. The forming process starts by reducing the distance of
the axes of the two tools, which rotate synchronously. The forming of the gear
geometry is realized by the rolling kinematic between the tool profile and the gen-
erated evolvent flanks of the gear. As is typical for the rolling of gears, the initia-
tion of rolling forces occurs in tangential direction, resulting in different states of
stress on the left and right flank. The difference also causes asymmetric material
flow results (Figure 22.2).
22 Gear Rolling Process 477
Fig. 22.2 Asymmetric stress conditions during the rolling process
The process of gear rolling with round tools is divided into three phases:
• Initial rolling phase: rolling of tool tip pitches into the pre-form-diameter of the
workpiece (WP)
• Penetration phase: penetration of the round tools into the workpiece until de-
manded parameters of gearing are reached (tip diameter and root diameter)
• Calibrating phase: calibration of the finished rolled gearing to achieve complete
and uniform tooth form filling and circularity (Fig. 22.3)
Fig. 22.3 Phases of the rolling process
By setting up a rolling process with controlled points of reversion (changing

rolling direction of both tools) it is possible to position the tip fold of each tooth
centrally and allow a calibration forming of the tip, thus realizing the complete
tooth form filling in the final calibrating phase. However, the optimized closing of
the tip fold has no influence on the load capacity of the gearing. By controlling the
rolling process in this way an asymmetric forming of flank areas can be avoided.
The round rolling technique allows a repeated over-rolling of the workpiece.

There is no limit to the number of revolutions, which theoretically can be regarded
as a forming process with infinite tool length. The complexity of the process con-
trol consists in the fine tuning of the adjustment of the workpiece pre-diameter,
tool and workpiece material, tool evolvent profile and the penetration stage.
Process results can produce gear qualities according to DIN 3960 [2], DIN 3961
[3] and DIN 3962 [4] with limited remaining variations of profile, tooth trace,
pitch and concentricity.
22.2.2 Range of Currently Rolled Gears and Achieved Qualities

The industrial manufacturing of high gearings, mainly for automotive gears, is
currently realized exclusively by cutting technologies. The manufacture by form-
ing, i. e. rolling methods, represents a scientific novelty in the international tech-
nological development. Figure 22.4 illustrates the current and future development
of profile cross rolling with round, outside-toothed tools.
Fig. 22.4 State of research for rolling of high gearings
The gear qualities are measured according to DIN 3962. This standard defines a
range of quality measures based on dimensions and accuracies for each relevant
gear variable. The technological limits in gear forming have been considerably ex-
tended. Ranges of gear types, formerly impossible to form, have been produced by
the round rolling technique. State of the art qualities for the rolling of stub tooth
gearings are qualities 7 to 8. First experimental investigations resulted in rolled
high gearings for automotive components with tooth height coefficient y > 2.7
with quality 9 to 10. The tooth height coefficient is defined by the following rela-
tion, where da and df are the tip and root diameters, mn is the normal module and hz
is the tooth height:
da − d f hz
y= = (22.1)
2mn mn
In order to be industrially applicable quality levels 7 to 8 have to be achieved for

the rolling of high gearings as well. The innovative character of manufacturing
high gearings by the cold rolling of massive billet requires the adaptation and im-
provement of machine concepts, the analysis of the interactions and reactions in
the system machine - tool - gear profile and the development of scientifically
based means and methods for the prediction of gear qualities. In Figure 22.5, an
example of a cold rolled gear (y = 2.47) with front area finishing is illustrated. The
right side column in Figure 22.5 shows the measurement curves of the relevant pa-
rameters profile, tooth trace and pitch. The qualities of the gear according to DIN
3962 are also given.
Fig. 22.5 Quality parameters of a rolled high gearing
In the past, the accuracy of the rolled gear wheels depended to a large extent on
the quality of the tools. Nowadays, conventional manufacturing processes can
produce gear qualities 5 or 6. Even quality 4 can be reached by profile grinding.
These qualities are sufficient for reaching gear qualities 7 or 8 in the rolling
process.
22.2.3 Interaction of Machine and Process

The rolling of high gearings at the Fraunhofer IWU Chemnitz is performed
with the rolling machine ‘PWZ Spezial’ (Fig. 22.6). The machine is conceptually
designed for applications in technical research and for the manufacturing of small
series in the sector of rolling worm geometries, special profiles and stub tooth
gearings. The PWZ consists of the machine base combined with two rolling slides,
which are supported and guided by needle roller bearing tracks. Two hydraulic cy-
linders inside the rolling slides generate the movement of feed and return stroke.
The motion control of both slides is based on an incremental length measurement
system.
Fig. 22.6 Components of the rolling machine
The presented investigation aimed at attaining knowledge about the machine

behavior during the rolling process. The machine concept was designed without
the consideration of manufacturing cold rolled high gears. The specific demands
of rolling high gears require the analysis of the behavior of this machine type.
Based on these insights the research and development of interactions between the
machinery, process and rolled part can be continued. The initial task in the expe-
riments consisted of the analysis of pitch relations in combination with the axial
positions of both tools at the process beginning. In Figure 22.7, the variation of
critical axial positions and tool displacement are shown.
Fig. 22.7 Tool displacement
Uneven numbers of teeth on the tool and gearing cause an unsymmetrical dis-
tribution of force impact points on the left and right side of the workpiece. In con-
sequence, left and right tool are horizontally displaced by about half pitch at the
tool tip diameter. The parameters of the gear used in the experiments are listed in
Table 22.1.
Table 22.1 Gear parameters
number of teeth zpart 20

normal module mn 3.45 mm
helix angle β 20°
pressure angle αn 24°
tip diameter da 84.2 mm

root diameter df 69.0 mm
tooth height hz 7.6 mm
tooth height coefficient y 2.2
material - 16MnCr5
Based on the value of the pitch at the tool tip diameter ptool,A, the initial pitch pa
is calculated from the billet pre-form diameter dV and tooth number zpart according
to the equation (22.2). The pre-form diameter results from the cross-section area
of the gearing with the parameters given in Table 22.1.
dV ⋅ π 73.2 mm ⋅ π
pa = ptool,A = = = 11.5 mm (22.2)
z part 20
Subsequently, the initial tool displacement can be calculated:
pa 11.5 mm
= = 5.75 mm (22.3)
2 2
In addition to the problem of tool displacement the problem of an inconstant pitch
is a significant factor. The pitch value changes during the rolling process in a non-
linear progression (Fig. 22.8).
Fig. 22.8 Pitch variable process [1]
The initial pitch pa is calculated by using the pre-form diameter dv, while the
calibration pitch pk is calculated by using the target root diameter df:
d f ⋅π 69 mm ⋅ π
pk = = = 10.84 mm (22.4)
z part 20
Due to the pitch difference between initial pitch (pa = 11.5 mm) and calibration
pitch (pk = 10.84 mm) the tool teeth are subjected to bending stress. This stress
furthermore affects machine components and the rolling kinematic. The aim of
the investigations was the identification of the character and magnitude of the
described problems. The displacements and elastic deformations of machine com-
ponents were measured. The ‘Faro Laser Tracker’ was used to detect and record
displacement and deformation. This 3D portable laser measuring system is used
for the measurement of devices and machinery. Hereby, a laser beam from the
tracker is focused onto a mirror reflector and projected back by this mirror. The
reflected light waves allow the determination of distances as well as horizontal
and vertical angles. Using the obtained data the exact 3D positions can be calcu-
lated. Another advantage of the laser tracker system is the ability to measure mo-
tion. The mirrors rotate and the tracker, driven by a highly sensitive servomotor,
follows the mirror movement [5]. Seven rotating mirrors were mounted on specif-
ic points of the machine for the tests (Fig. 22.9).
Fig. 22.9 Placement of the measuring points in the machine
Measuring points 1 and 2 were installed on the bearing blocks in front and be-
hind the tool to record the movement of the left tool axis under load. Furthermore,
the interaction effect with the left carriage could be analyzed by the rotating mir-
rors 6 and 7. To determine the tool displacement during the rolling process mirrors
were mounted in front of and behind the round rolling tools (1, 2 and 4). The inte-
raction between tool holder and rolling carriage could be measured by the rotating
mirrors installed parallel to the tool measuring points (5, 6 and 7). At the tips of
each clamping device a measuring point (3) was fixed in axial workpiece position
to determine the interaction between tools, workpiece, and workpiece holder. The
measuring points on the tool axes (measuring points 1 and 4) were used to detect
the relative positions between tools and workpiece during the rolling process. The
experiments were carried out with the tool rotation speed set of 12 revolutions per
minute and 4 reversion points. The graph in Figure 22.10 displays the progress of
the rolling force and tool path during the rolling process. The maximum recorded
force was 99.2 kN, which corresponds to a machine load of about 25 % (maxi-
mum load capacity 400 kN).
Fig. 22.10 Penetration and force path
22.2.4 Results
The following diagrams show the displacement of the measuring points (MP) dur-
ing the rolling process recorded by the laser tracker system. The beginning of the
process represents the zero-situation, i. e. the unloaded initial position of the ma-
chine system. During the rolling process the direction of the tool revolution re-
versed four times. At the points of reversion during the penetration phase the
entire machine system is briefly unloaded. Such reversion points correspond to the
points of sharp amplitude changes of the measurement curves.
Displacements in x-y plane (Fig. 22.11 and 22.12)

The measurement points on the tool axis as well as on the rolling slides were dis-
placed under load into negative y-direction. At the front bearing block (measuring
point 1 (MP 1) a higher load was detected. In consequence, the rolling carriage
(points 6 and 7) tilted around the z-axis.
Fig. 22.11 Displacement in the x-y plane (left: tool axle; right: rolling slide)
Fig. 22.12 Displacement in the x-y plane (left: initiation; right: displacement)
Displacements in x-z plane (Fig. 22.13 and 22.14)

The entire system was displaced in positive x-direction during the rolling process.
Both rolling slides and the tool holders (measurement points 1, 2 and 4) moved
parallel to each other. A strong displacement of the left rear-bearing block could
be noticed.
Fig. 22.13 Displacement in the x-z plane (left: tools; right: rolling slides)
Fig. 22.14 Displacement in the x-z plane (left: initiation; right: displacement)
Displacements in y-z plane (Fig. 22.15 and 22.16)

Both tools as well as the tips of the clamping devices were displaced out of the ini-
tial position in negative y-direction. Measuring points 1 and 3 were moving paral-
lel to each other during the forming process. The right tool (measurement point 4)
showed a stronger displacement from the axial position during the forming [6].
Fig. 22.15 Displacement in the y-z plane (both tools und clamping tips)
Fig. 22.16 Displacement in the y-z plane (left: initiation; right: displacement)
Because of the increase of available computing capacities and the latest findings in
the FEM simulation of incremental forming technologies an illustration of the de-
manding process kinematics became possible. For the selected reference gearing
of module mn = 3.45 mm and zpart = 20 a regular process containment with tools as
deformable geometries, the workpiece, the rolling kinematic and the boundary pa-
rameters was defined. The aim of simulating the rolling process was to calculate
the three-dimensional component material flow, the maximum achievable defor-
mation, the extent of workpiece strain hardening, the resulting orientation of ma-
terial fiber and the occurring tool and workpiece loads. The simulation was carried
out using the Transvalor Software Forge. The rolling machine itself was defined
as non-deformable. The measured displacements of tools, workpiece and machine
components shall be included into the simulation model in further projects. The
process was computed in an acceptable calculating time and could be verified by
practical investigations. The following tables provide an overview of setting pa-
rameters of the simulation.
Table 22.2 Parameters of simulated workpiece
numeric simulation of rolling process – workpiece

- C15/16MnCr5, soft annealed
- Young’s modulus: 200 kN/mm²
material
- Poisson ratio: 0.3
- cold forming according to Hansel-Spittel
- 4-knots tetrahedron-shaped elements
meshing - element size in forming zone: 1 mm
- element size in core zone: 9 mm
discretization
contact conditions - µ = 0.2 (friction coefficient)

tools – workpiece - m = 0.4 (friction factor)
thermal conditions - not defined
manipulator - defined to limit axial material flow
Table 22.3 Parameters of simulated round tools
numeric simulation of rolling process – rolling tools

- pure elastic material behavior
material - Young’s modulus: 200 kN/mm²
- Poisson ratio: 0.3
- deformable geometry
- 4-knots tetrahedron-shaped elements
meshing - element size in contact zone :4 mm
- element size in core zone: 20 mm
5
- approx. 1.9·10 elements
discretization
numeric simulation of rolling process – tool holders / drive

contact to - spindle diameter: 120 mm
tools - fixed connection
- feed 0.07 mm/s
kinematics - rotation speed 12 rpm with changes of direction
- reversion points after 11, 17, 23 and 28 workpiece revolutions
Figure 22.17 shows the stress distribution on the tool and deformed workpiece
in the normal section at the end of the penetration phase (left figure) and a picture
of the rolled component (right figure). A detailed validation and evaluation of the
quantitative simulation resulting in verifiability is intended to provide a compari-
son with the practice tests or with the experimental-numerical method for the de-
termination of the tool loads. The equivalent stress by v. Mises was used to
represent the three-dimensional stress state, showing high values of equivalent
stress. The results of the simulation lead to the selection of high performance
powder metallurgical steel with high strength and toughness as rolling tool materi-
al. Compared to the real material behavior the simulation shows a rotation kine-
matics-related volume increase in the forming simulation process in spite of
Runge-Kutta interpolation (RK2), which results in a faster "form filling" of the
tool gap with workpiece material in the simulation than in the real process.
Fig. 22.17 Stress distribution during calibration according to v. Mises
22.4 Evaluation of Interactions

The forces occurring during the rolling process, including the additional load
induced by changing pitch values, cause a displacement of rolling tools and sup-
porting components. The consequence is a deviation from the optimum rolling
kinematics resulting in profile, tooth trace and pitch deviations and thus in a di-
vergence from the geometrical workpiece contour defined by DIN 3962. As a con-
sequence of the tangential contact of the tool teeth and the workpiece as well as
the asymmetrical displacements of the rolling tools, the tool bearings and the
workpiece fixture, the tool tooth in the contact zone is subjected to a bending load.
This causes an elastic deformation of the tool tooth and an interference with the
next tooth, such that the following tool tooth pushes against the next workpiece
tooth too early. The rolling tools are highly stressed by dynamically oscillating
loads on the tool tooth during contact with the workpiece. Particularly the asym-
metrical displacements of the two rolling tools and the workpiece fixture cause a
high load on the tips of the clamping system. Consequently, the displacements and
deformations reduce gearing quality as well as tool life.
22.5 Conclusion
The demand for manufacturing methods, which ensure a high utilization of ma-
terial with a high quality of the final products and short production times, conti-
nuously pushes the improvement of existing and the development of new process
technologies. The manufacturing of transmission-typical gears by cold forming is
one new approach. Current industrial applications of gear rolling are in the range
of stub tooth gearing, tappet gearings and pinions in defined limits to maximum
module 1.6. First investigations demonstrated that a forming capacity of high
gearings can be obtained by adapting and optimizing the rolling process to the
specific gear parameters. The results of the trial series proved that tooth heights
can be manufactured, which so far have not been producible by cold cross rolling
processes. The main tasks of future research are to improve the tool and process
design regarding the accuracy of the rolling kinematic and the precision of the
pitch. The machine concepts of future rolling machines have to be adapted to and
optimized for high gear rolling as currently available machines are derived from
guidelines for the rolling of stub tooth gearings. Especially the stiffness and/or de-
sign of the workpiece holder and rolling tool axles have to be enhanced, regarding
the high loads in the process of high gear rolling. The first steps into the field of
cold forming of high gears have been taken but major research work on the analy-
sis of the interaction between process, machine and part is needed in order to de-
velop the technology to a reliable level for industrial application. From a scientific
point of view the detection of tool stresses within the process boundary conditions
provided the basis for further steps in a technology research and process optimiza-
tion. The knowledge of load parameters as well as their local and temporal occur-
rence allows the performance of direct discharges for parameter-adjusted tool and
process dimensioning. Adequate tool materials and heat treatments become select-
able. Thus, the FEM simulation is an implementation for shortening development
times and product innovations while maintaining the product quality requirements.
References
[1] Neugebauer, R., Hellfritzsch, U., Lahl, M.: Advanced process limits by rolling of heli-
cal gears. In: 11th ESAFORM Conference on Material Forming, Lyon (2008)
[2] DIN 3960, Definitions, parameters and equations for involute cylindrical gears and
gear pairs. Beuth Verlag GmbH, Berlin (1987)
[3] DIN 3961, Tolerance for cylindrical gear teeth, bases. Beuth Verlag GmbH, Berlin
(1978)
[4] DIN 3962, Tolerance for cylindrical gear teeth. Beuth Verlag GmbH, Berlin (1978)
[5] Neugebauer, R., Klug, D., Hellfritzsch, U.: Description of the interactions during gear
rolling as a basis for a method for the prognosis of the attainable quality parameters.
Prod. Eng. 1(3), 253–257 (2007), doi:10.1007/s11740-007-0041-9
[6] Neugebauer, R., Hellfritzsch, U., Lahl, M.: Interaction of Machine, Tool and Process
by Rolling of High Gears. In: Proceedings of the 1st International Conference on
Process Machine Interactions, Hannover (2008)
Chapter 23
Investigation of the Complex Interactions
during Impulse Forming of Tubular Parts
Fr.-W. Bach, M. Kleiner, and A.E. Tekkaya
Abstract. The expansion of tubes by direct application of gas detonation waves or

electromagnetic forming (EMF) is an alternative forming method for hollow-
section workpieces. In particular, the process can be used for typical hydro-formed
parts, car body or exhaust elements in the automotive industry, for example. The
introduced processes belong to the category of high speed forming methods and
provide typical advantages, such as higher achievable strains, compared to quasi-
static methods using high water pressure. Another advantage of these processes is
the avoidance of high contact forces by employing an “inertia-locked tool” system
due to the extremely short process time. To develop a controllable process it is es-
sential to gain a good knowledge of the interactions in the system. This can be
achieved by using simulations in combination with experimental investigations;
their results are the topic of this paper. Also included are special investigations of
the material behavior at high strain rates.
23.1 Introduction
Implementing lightweight construction concepts has been becoming more and
more important over the last years. Thereby, material as well as design aspects
have to be considered. The application of space frame structures from hollow pro-
files made of the typical lightweight metals as aluminum and magnesium alloys is
a common approach. However, to gain maximum weight reduction, it is necessary
to apply the most suitable material for each separate component [1]. Consequent-
ly, a material mix including high strength steels, (fiber-reinforced) polymers and
the already mentioned lightweight alloys results. Further weight reduction can be
achieved by combining the functions of different parts and thus eliminating sever-
al components completely. This constructive approach usually leads to a higher
complexity of the remaining components, which have to fulfill different and some-
times contradictory functions [2]. This concerns the geometry and material as-
pects. It could be necessary to combine the corrosion resistance and strength of
aluminum, steel or titanium with the high electrical and thermal conductivity of
aluminum or copper, for example, guaranteeing a thin-walled but stiff geometry at
the same time. Consequently, highly complex forming and joining tasks have to be
performed in modern production processes so that conventional technologies fre-
quently reach their limits. Considering conventional forming operations, such as
492 Fr.-W. Bach, M. Kleiner, and A.E. Tekkaya
quasi-static hydroforming alone, the formability might be too low to realize the
complex part geometry without wrinkling, buckling and/or cracking. When think-
ing of thermal joining technologies, problems might arise due to the high affinity
of aluminium, steel and titanium to the atmospheric gases oxygen, nitrogen
and hydrogen. Especially at elevated temperatures in the range of 300°C and
more, brittle intermetallic phases can be formed, which decrease the joint quality
significantly.
Here, high speed or impulse-forming technologies, such as explosive forming
by a gas detonation and electromagnetic forming (EMF), offer a high potential.
They can be applied for forming with and without a form defining tool as well as
for joining by forming, where the connections can be based on force-fit, form-fit
and impact-welded material joints. The processes use the energy density of a de-
tonating gas (forming by gas detonation) or pulsed magnetic fields (EMF) to acce-
lerate and form tubular workpieces without any mechanical contact to the tools
and machines, for example. Although the principle and the advantages of both
technologies have already been known for several decades no wide implementa-
tion into industrial series production has been realizable up to now. One important
reason for this deficit is the lack of a deep understanding of the process and, based
upon this, the lack of general and easily applicable design strategies for the
process dimensioning [3]. To fill these gap fundamental investigations considering
the complex interdependencies occurring during high-speed forming and the join-
ing of tubes by gas detonation and EMF, respectively, were carried out bearing in
mind aspects of forming technology, material science and process technology. The
results of this work are presented in this article, structured according to two differ-
ent sub-systems:
• Sub-system 1 deals with the acceleration and the free forming of a workpiece.
Here, the investigations focus on the interdependencies between the pressure
pulse and the workpiece deformation, which, in turn, retroact on the temporary
and local development of the pressure. These investigations were carried out
disregarding the special application (forming or joining) but considering the
two different acceleration technologies separately. In the case of forming by
gas detonation, the interdependencies between the reactive gas, the energy
source, and the working medium at the same time and the workpiece is rele-
vant, while in the case of EMF, the interdependencies between the tool coil, the
magnetic field and pressure, and the workpiece are considered.
• Sub-system 2 deals with interdependencies during the impact of the workpiece
and an according obstacle, which can be a form-defining tool or a joining part-
ner. Here, the impacting parameters (e. g. impacting velocity and angle), the
properties of the two components (workpiece and obstacle) and the formation
of the contact are analyzed. The evaluation of the result is carried out consider-
ing the special application. While in the case of forming against a form defining
tool, the form filling and the accuracy of the deformed part are decisive,
the formation of the joint and the transferable load are relevant in the case of
joining.
23 Investigation of the Complex Interactions during Impulse Forming 493
23.2 Sub-system 1: Acceleration and Free Forming of a

Workpiece
23.2.1 Electro-Magnetic Forming

In the EMF of tubular components, the energy density of pulsed magnetic fields is
used to exert radial forces to workpieces made of an electrically highly conductive
material and deform them within a few micro-seconds. The principle can be ap-
plied in two different variants depending on the setup of the tool coil and the
workpiece. In electromagnetic compression, the tube is positioned in a cylindrical
coil while in electromagnetic expansion the coil is surrounded by the workpiece.
23.2.1.1 Setup and Process Principle
In both variants, the setup can be represented by an oscillating circuit, where the
forming machine serves as the energy supply while the tool coil and the workpiece
represent the consumer load.
Fig. 23.1 Principle sketch of the electro-magnetic tube compression and expansion
In Figure 23.1, the forming machine is represented by its equivalent circuit dia-
gram consisting of the capacitor C, the inner resistance Ri and the inner inductance
Li. Closing the high current switch causes a sudden discharge of the capacitor and
thus a damped sinusoidal current I(t) flows through the tool coil so that a corres-
ponding magnetic field H(r,z,t) results. The time-dependent field induces a current
in the workpiece, which flows opposed to the coil current and shields the magnetic
field. Consequently, radial Lorentz’s forces act on the workpiece and deform it in
the direction away from the tool coil. These volume forces can be mathematically
transferred to virtual surface forces, the so-called magnetic pressure p(r,z,t).
Due to the short process times in the range of 100 µs or less in impulse forming
and joining, high demands have to be made on the measurement technique. During
EMF, the coil current is measured via a Rogowski probe; on this basis, the mag-
netic pressure can be calculated analytically or via a coupled structural mechanic
and electro-magnetic simulation [4]. The workpiece deformation can be measured

online using e. g. optical measurement techniques. To quantify the displacement
of particular surface points as a function of the time, a laser shadowing method
was applied. Depending on the regarded process variant the setup is adapted. In
tube compression, a linear laser shines axially through the workpiece. In case of
expansion, the laser beam is directed rectangularly to the tube axis and the tube
radius. In both cases, the transmitted light is received and transferred to an accord-
ing voltage signal by a position sensitive device (PSD) at the other side of the
workpiece. Due to the deformation of the tube the light is partly shadowed and the
voltage signal of the PSD changes accordingly. However, due to geometric restric-
tions in case of compression, only the deformation of the tube’s smallest cross-
section, typically located in the center of the coil, can be observed. Details about
this measurement method are described in [4].
23.2.1.2 Acting Loads and Workpiece Deformation
In EMF, significant interdependencies between the acting loads and the workpiece
deformation were identified. These become evident when comparing current
courses measured during a forming process and during experiments with the same
process parameters but with prevented workpiece deformation. Figure 23.2
presents such curves together with the pressure curves and the radial displacement
of the smallest cross-section measured during a tube compression.
The diagrams clearly indicate that, at the beginning of the process, the coil
current and the magnetic pressure increase accordingly while the workpiece re-
mains undeformed. Plastic deformation occurs as soon as the inertia is overcome
and the stresses in the workpiece reach the flowstress of the material. It is corre-
lated to an increasing gap width between tool coil and workpiece; consequently,
the inductivity of the consumer load (tool coil and workpiece) steadily increases.
Therefore, the amplitude and the frequency of the coil current are lower compared
to the reference experiment with prevented workpiece deformation. Thus, consi-
dering the direct comparison of the current measurements the beginning of the
workpiece deformation is indicated by the point of time, where the curves are no
longer collinear. After the saddle point in the current curve measured during the
forming experiment, the inductivity of the oscillating current remains constant.
This can indicate that the forming is completed. In the displacement curve, an
elastic radial oscillation of the tube can be recognized because the high kinetic
energy resulting from the high forming velocity cannot be dissipated spontaneous-
ly. [4] Under certain circumstances, the increasing gap volume between coil and
tube can cause the magnetic pressure to decrease faster than the current. Moreo-
ver, with time, the magnetic field penetrates the workpiece causing a further pres-
sure diminution. This penetration is faster, the lower the discharging frequency,
the thinner the workpiece wall and the lower the electrical conductivity of the
workpiece [5].
Fig. 23. 2 Interdependency of current course, magnetic pressure and radial displacement in
electromagnetic tube compression
23.2.2 Forming by Gas Detonation

23.2.2.1 Setup and Process Principle
The experimental setup used for investigation forming by gas detonation is illu-
strated in Figure 23.3 and explained in the following: A tube, which is fixed in the
testing apparatus designed by the SWL1 and located in the explosion-protected
area of the SWL, is filled with a stoichiometric mixture of hydrogen and oxygen
using a filling connector via mass flow controllers. The gas is then detonated ex-
ternally using a wire, which is ignited by a high voltage discharge. The detonation
wave, which is immediately formed, travels through the pipework, which bifur-
cates and then enters the detonation tube from two sides. The gas mixture in the
apparatus is now ignited, which a homogeneous conflagration produces over the
entire cross-section. A tubular-shaped probe, located in the middle of the appara-
tus, is installed co-axially to the workpiece and fixed at both ends of the testing
1
Stoßwellenlabor RWTH Aachen.
apparatus. Two pressure sensors, located at different positions in this probe, are
connected to the gas-filled space by radial holes and shielded from direct contact
with both the active components as well as the combustion gases by high tempera-
ture grease.
Fig. 23.3 Experimental setup used in forming by gas detonation [6]
To keep the apparatus' design simple, all the tested workpieces have an initial
diameter of 40 mm. In the course of the test setup, the tubes are slid to the right
end onto a sealing ring and fixed by means of a split sleeve. The tube's axial
movement is prevented by an O-ring seal at the left end. Four planar tool compo-
nents, which constrain the forming but still permit the formed region to be directly
observed, can be arranged around the workpiece. Moreover, the workpiece can be
replaced by a thick-walled tube and thereby enable pressure profiles to be meas-
ured without influencing the volume changes due to the forming process [6].
The optically accessible forming section can be photographed during the
process by a high-speed (HS) camera. The camera enables the workpiece's contour
to be investigated at different times during the forming. The DRELLO type HS
camera has a framing frequency of 10 MHz for 24 images. It is triggered by the
first pressure sensor due to the travelling pressure wave [6].
23.2.2.2 Acting Loads and Workpiece Deformation
The two typical, measured pressure profiles depicted in Figure 23.4 were recorded
during free and constrained expansion tests using Al 99.5 and thick-walled tubes
respectively.
Fig. 23.4 Pressure courses during free tube expansion and reference experiments with pre-
vented workpiece deformation [6]
An initial pressure p0 of 8.5 MPa was used for the tests. Owing to the strong
oscillations the measured pressure signals required smoothing. The zero point on
the time axis is defined by the initial signal of the first pressure sensor. The initial-
ly measured signal deflection of the pressure curve cannot be resolved in detail
due to the smoothing or the sensor signal's strong vibrational behavior. The theo-
retical maximum value for this initial pressure is 17 MPa. In the diagrams of the
first and second pressure sensors, the second, smaller rise in pressure is explained
by the partial reflection of the detonation wave at the sealing ring depicted in Fig-
ure 23.3. The third noticeable pressure rise after approx. 280 µs and 180 µs, re-
spectively, is generated by the reflection of the detonation wave at the end wall.
Consequently, the expansion of the tube leads to a faster and earlier drop in the
pressure profile based on the position of the detonation wave [6].
To characterize the forming sequence high-speed photographs were taken dur-
ing selected free expansion tests. The initial gas mixture's pressure, which deter-
mines the maximum pressure operating in the process, is limited to p0 ≈ 0.9 MPa
for the 40 x 2 mm Al 99.5 tubes. Exceeding this value leads to the specimen's fail-
ure by crack formation during the expansion process. The forming sequence is de-
picted in Figure 23.5 with the simultaneous reference to the pressure profiles. In
Figure 23.5a), an original HS-camera image showing the process status 121 µs af-
ter triggering by means of the first pressure sensor using the arrival of the detona-
tion wave is depicted. This geometry can be seen as the form's final contour. The
diagram in Figure 23.5b) shows various stages of the tube's expansion. Additional-
ly, the position of the wave front at 21 µs and 41 µs is plotted. The tube wall's
highest radial speed of 45 m/s is attained at the right hand end of the free-region
shortly before the second pressure sensor indicates the second large rise in pres-
sure (compare Fig. 23.4), which is explained by the waves’ reflections [6].
Further elevating the initial pressure p0 produces a still larger expansion, which
continues beyond the observation time (Fig. 23.5, experiment no. 33)). The final
diameter obtained is plotted on the graph's right ordinate.
Fig. 23.5 Results of online measurements of the expansion during forming by gas detonation
[6]
To be able to also track the effect of the higher pressures, the four planar tool
components are employed. In Figure 23.5c, the commencing phase of the expan-
sion is shown for the initial pressure of p0 = 1.13 MPa. The maximum radial speed
amounted to 65 m/s at this pressure. This value approximately corresponds to a
strain rate of 3,000 s-1[6].
By supplementing the described experimental investigations a numerical analy-
sis of the forming by gas detonation was realized. As a first step, an uncoupled
simulation strategy was applied in order to avoid problems due to changing boun-
dary conditions resulting from the displacement of the tube wall. Such problems
typically occur in fluid mechanical calculations, which are usually based on Euler
formulations. The shock tube simulation program Kasimir, developed by SWL,
was employed to determine the pressure-time profile [7]. With the aid of a Rie-
mann solver, this software solves the thermo-dynamic equations for shock waves
taking into account the chemical reaction. The results obtained using this program
were employed as time-dependent input parameters for the structural mechanical
simulation in Msc.Marc in combination with various user sub-routines developed
at the IUL2. In order to take into account the influence of the inertial masses dur-
ing the forming, an implicit Newmark time-integration scheme was employed to
solve the equations of motion, ignoring the damping effects [8]. An ALE strategy
using a second FE mesh was developed and implemented for interpolating the im-
ported data. In this way, the forces' time and spatially-dependent boundary condi-
tions were introduced into the structurally dynamic model. This mesh, which is
only necessary for the time and spatial interpolations in the first step of the
modeling, is simultaneously distorted by the tube's forming. For this purpose, an
2
Institut für Umformtechnik und Leichtbau, Technische Universität Dortmund.
implicit, quasi-static simulation is performed, based on the structural-mechanical

equations of motion, using a tim-saving, iterative sparse-matrix equation solver. In
Figure 23.6a, the meshing and the according boundary conditions are illustrated.
Fig. 23.6 Results of numerical investigations according to [6]
Figure 23.6c shows the results achieved by applying this simulation strategy to
a free tube expansion. After initiating the detonation wave, the tube's wall remains
in its initial position for about 20 µs until the forming begins. The reason for this
is the effect of the inertial masses and their associated forces, which counteract the
acceleration. The elastic-plastic deformation of the initially accelerated tube re-
gion ceases when the detonation wave reaches the end of the tube. It was also
possible to experimentally verify this time delay between the detonation wave and
the tube wall's movement.
The reflected wave re-accelerates the tube's wall, which is already moving. Fol-
lowing the wave's second reflection, the deformation process has terminated and
the final form has been obtained. The entire process lasts for approx. 200 µs, whe-
reby the quantitative data for the process sequence depends to a large extent on the
starting parameters. The numerically determined contour fits the measurement
well qualitatively (compare Figure 23.6 and Figure 23.8). The suggested modeling
including the simplification employed here, such as the decoupling strategy, can
therefore be considered as permissible.
A completely coupled simulation, possessing simultaneous and step-wise solu-
tions of both field problems, is necessary for a mathematically correct model of
the process and, in particular, of the interactions. However, the required effort for
this is disproportionate to the gain in knowledge with respect to the state of the
current investigations. For this reason, the fall in gas pressure due to the increase
in volume was also simply modeled by means of an exponential pressure drop.
However, the results obtained using this strategy showed no substantial influence
of the deformation process.
23.2.3 Comparison of the Workpiece Properties

to Quasi-Statically-Formed Reference Parts
In order to generate a reference for the investigated impulse forming processes,
apparatus for quasi-static forming were developed and constructed at the IUL in
Dortmund. For this purpose, a testing rig for internal high-pressure joining was
modified to be able to carry out the appropriate burst tests on the semi-finished
products used. During the testing, the specimen's expansion and internal pressure
can be continuously recorded via diametrically located inductive displacement
transducers and an appropriate pressure sensor respectively. Another method was
developed, which uses an elastomeric tool to quasi-statically expand tubes. This
tool enables tubular specimens to be radially expanded by means of axially com-
pressing a polyurethane mandrel (hardness Shore 90 A). This tool was developed
to minimize a possible influence on the thermographs, which are necessary to cha-
racterize the process and described in more detail below, by the cooling effect,
which could be noticeable due to the application of water pressure (hydroforming)
to the tube. The experiments for die-forming the tubular semi-finished products
were carried out in a Müller-Weingarten BZE 1000-30.1.1 type 10 MN hydraulic
press, available at the IUL, by using a hydro-forming tool with identical die-
inserts as those used for the gas detonation tests. The geometry of the quasi-
statically expanded tubes and of the specimens formed by gas detonation
was quantified by tactile measurements using the 3-dimensional coordinate mea-
surement machine Zeiss Prismo VAST HTG. Thereby, the roundness as well as
the wall thickness and the contour were determined (compare Figure 23.7 and
Figure 23.8).
Here, it is noticeable that for the gas detonation forming, the profile of the wall
thickness distribution can be recognized in the roundness measurements. This
phenomenon is already known from electromagnetically compressed tubes [9].
One explanation is the influence of the inertial forces as a consequence of the
process' high energy. This symmetric expansion of the specimen can also, at least
partially, explain the high strains attained. In contrast to this, the quasi-statically
formed workpiece shows a non-uniform distribution of wall thicknesses, which
indicates premature failure of flaws, such as the protruding tube's seam in the
semi-finished product. The roundness measurements also point to this effect since
bulging occurs in the sections, where the wall thickness has thinned out.
Fig. 23.7 Roundness and wall thickness profiles of free expansion tests a) gas detonation, b)
quasi-static water pressure (according to 6)
Figure 23.8 shows the contour and the according distribution of the hardness
measured for a specimen expanded by gas detonation.
It can be clearly recognized that the hardness increases with increasing defor-
mation. This effect corresponds to the strain hardening known from conventional
forming processes. Comparing the hardness values to according measurements for
a quasi-statically expanded tube shows good qualitative and quantitative agree-
ment. This indicates that the high strain rates occurring during forming by gas
detonation do not significantly influence the strain hardening behavior of this ma-
terial. This tendency again corresponds to similar experiments comparing quasi-
statically formed and electromagnetically compressed tubes [10].
Fig. 23.8 Workpiece contour and hardness increase [6]
However, the temperature gradient, the impulsive forces and the accordingly
high forming velocities lead to a specific deformation of the workpiece during
forming by gas detonation; consequently, the corresponding forming mechanisms
are expected to differ from conventional ones. Therefore, more fundamental ma-
terial investigations were carried out to provide more comprehensive knowledge
about the process specific influences on the material behavior.
High-speed tensile tests were carried out in a drop tower testing-rig at the IW3
to determine the influence of the strain rate on the material properties. By using
this drop tower it was possible to investigate high strain rates in round-bar speci-
mens. In this test, the specimen is accelerated using various falling test weights in
a 4 m high drop tower testing-rig with a maximum falling speed of 6.3 m/s. A test
weight impacts the anvil; and the resulting impulse is transmitted to the lower spe-
cimen's grips via four rods. In this way, the specimen is subjected to straining at a
high speed. The installed load cell (Hottinger 50kN, type U3) detects the resulting
load. In addition to this, specimens were successfully prepared in an appropriately
equipped metallurgical laboratory for subsequent investigations in scanning elec-
tron and light microscopes (see below).
Moreover, in collaboration with the IUL and SWL, it was possible to success-
fully carry out thermographic investigations of both the gas denotation procedure
as well as the comparative static procedures. This enabled the investigation of the
effects of the heat, generated during the method and forming. The thermal camera
employed for this purpose, ThermaCAM SC 3000 made by the company FLIR,
incorporated a high-speed facility.
To investigate these influences metallographic investigations were carried out
by the IW. The micro-structural changes as a result of deformation can be de-
scribed by a multiplicity of structure-relevant parameters. Dislocations and grain
boundaries have a substantial influence on the dynamics of the forming process
[11, 12]. For example, dislocations within a grain show a different behavior than
at or near grain boundaries. These differences are caused by the structure and type
of the dislocations [13]. Previous analyses of the dislocation movement and the
dislocation density changes as well as the analyses of grain boundary migration
took place under thermal activation or at low strain and strain rates. The present
results of the conducted analyses can serve as the basis for the description of the
plastic deformation mechanisms at high strain rates in the finite element simula-
tion. The structural changes were examined using aluminum samples. The
processes during plastic deformation are related to different dislocation-drag me-
chanisms [14]. These are influenced by the following factors: - thermally activated
dislocation movement energy of dislocation drag - impurity viscosity - phonon
viscosity - electron viscosity. The first three mechanisms are particularly relevant
in static tensile tests with the conventional universal testing machines (tension
speed of 2 µm/s to 1,000 mm/min). It is well-known from literature that the
mechanism of phonon viscosity is very important at tension speeds of 1 m/s to
1,500 m/s [15].
The processes at the grain boundaries are extraordinarily important as well as
the dislocation drag in grains for the measured characteristics of the process of gas
detonation. The mobility of grain boundary and its driving strength are the funda-
mental parameters of the grain boundary slip [16]. The origin of this driving
strength has different causes. For most cases, the strength results from the differ-
ences in the free energy of the adjacent grains. This difference can be caused by
the different surface energies, permeabilities and disorientations of the adjacent
grains or by different dislocation densities on both sides of the grain boundary
3
Institut für Werkstoffkunde, Leibniz Universität Hannover.
[17]. A grain boundary slip sets the deformation energy stored in the dislocations
free; and the energy of the complete system is minimized. One of the most impor-
tant procedures for the analysis of the dislocation and the grain boundary structure
is the light-microscope etching pit method. Numerous methods have already been
suggested for the determination of the defect structure in aluminum. At the IW, a
new etching agent and a special procedure were developed on the basis of an etch-
ing agent, recommended by Barret and Levenson [18]. When using this procedure
the dissolution speed does not depend on the crystallographic orientation so that
the etching speed is evenly distributed on specific crystal faces. This procedure
can be used for the determination of changes in the defect structure after the de-
formation of poly-crystalline aluminum. A further application is to simultaneously
corrode grain boundaries, inner-grain dislocation pits and grain boundary disloca-
tion pits, using the described etching agent in combination with a certain parame-
ter set for electrolytic polishing. The reproducibility of the metallographically
determined etching pit density can be determined by comparison with trans-
electron-microscopy (TEM) measurements of the dislocation density. The sample
material was poly-crystalline aluminum (99.5 %), which was tested in the form of
rolled strips of 1 to 4 mm thickness and 70 mm width and of cast slabs of 7 mm
thickness and 70 mm width. The samples for the tension tests were prepared from
strips and slabs, homogenized for 72 hours at a temperature of 580°C and cooled
down along with the furnace.
After the tension tests with different test speeds, the samples were examined by
light-microscopy parallel to the tension direction (longitudinal axis). The samples
for the etching pit method were electro-polished in electrolytes (20 % perchloric
acid, 10 % butylglycol and 70 % ethyl alcohol) by using a voltage of 25 V and a
flow rate of 5 for a duration of 10 s. The etching agent for this corrosion con-
tained: 10 % HF, 30 % H20, 45 % HCI and 15 % HNO3.
Fig. 23.9 Light optical and SEM investigations of the dislocation structure. The specimens
are pulled with different velocities
In Figure 23.9, examples of the etched micro-structure are shown. Different test
speeds were used for that. The etching pit arrangement corresponds to a disloca-
tion structure after deformation whereas the original grain boundaries are marked
by pit-free seams. The light microscopic and the REM investigations of the spe-
cimens of the tension tests show that two processes influence the microstructure of
the material. On the one hand, the dynamics of the grain boundaries cause an in-
crease of the grains and a frequency movement of the grain boundaries. On the
other hand, dislocation structures were formed, which, on their part, are able to
cause the development of a new structure. Both processes influence each other. A
selection of the results of the investigations according to this effect is shown in
Figure 23.9. Increasing pulling velocities cause a higher density of dislocations at
the grain boundaries and a growing deformation of the grains. Smaller pulling ve-
locities (2 mm/s) cause so-called dislocation fringes at the grain boundaries. An
accumulation of dislocations can be detected near this phenomenon. This causes a
significantly higher density of dislocations in this area than in the middle of the
grains. A further increase of the pulling velocity causes higher lengths at first and
then continuous devolutions of the walls of dislocations. The density of the dislo-
cations in those walls increases in relation to the degree of forming.
23.3 Sub-system 2: Impact of the Accelerated Workpiece

with an Obstacle
23.3.1 Model Experiment for High-Speed Joining

To provide a basic understanding regarding the significance and influence of sepa-
rate parameters describing the collision, a model experiment was built up by IW.
The setup is shown in Figure 23.10 and consists of a joining unit and a pneumati-
cally-driven acceleration unit. The principle is briefly described in the following:
To initiate the process the pressure chamber (10) is pressurized and a mass (7) is
positioned at the end of the acceleration tube (6), which is connected to the accele-
ration chamber (8). Opening a high-speed valve (9) pressurizes the acceleration
chamber and a section of the acceleration tube so that the mass is speeded up
through the tube and bounds to the dynamic support (5) carrying the dynamic join-
ing partner (4). The impact leads to a sudden relative movement (x) in the magni-
tude of some millimeters between the joining partners, resulting in a welding
process, provided that the chosen process parameters are suitable. Thereby, the
impact velocity of the mass, which is expected to be a decisive aspect, can be in-
fluenced via the dimensioning of the mass and/or via an adaptation of the initial
pressure in the pressure chamber. [19]
Fig. 23.10 Setup of the high-speed joining device [19]
The joining quality was evaluated using metallographic methods. The investi-
gations have shown that for velocities up to 25 m/s no metallic bonding resulted
whereas higher impact velocities led to local impact welding. Figure 23.11 shows
the contact area of specimens, joint with an impact velocity of approximately
30 m/s, approximately 100 m/s and approximately 130 m/s in representative mi-
crographs. It can be observed that increasing the impact velocity results in larger
welded areas. Statistic interpretations of the micro-structural data suggest a linear
correlation between this ratio of welded area and contact surface and the impact
velocity in the regarded span of values. [19]
Fig. 23.11 Influence of different impact velocities on the contact area of joined specimens
[19]
However, also the intactness of the microstructure in the contact area has to be
considered when evaluating the welding quality. As shown in Figure 23.14, micro-
fractures occur in the aluminum parallel to the contact surface, indicating a deteri-
oration of the joining quality, if a critical value of the impact velocity is exceeded.
Here, this is shown for an exemplary impacting velocity of 130 m/s. In conclusion,
it can be said that an optimum value for the impact velocity exists. It has to be
high enough to maximize the ratio of the welded area and the contact surface but
low enough to avoid micro-fractures. For the regarded material combination and
setup, this optimum lies between 100 m/s and 130 m/s. [19]
To identify the resulting phases in the joining area, polished and etched speci-
mens were analyzed by an energy dispersive x-ray (EDX) and micro-analysis; and
the element distribution in the welding area was quantified. It was shown in [19]
that in-between the titanium and the aluminum base material an extremely thin
(thickness < 1 µm) titanium-rich transition area is located, however, according to
[20], intersection layers of such small thickness can be regarded as uncritical.
23.3.2 Joining by Electromagnetic Forming

As already mentioned, the joints produced by EMF can be based on a dominating
interference-fit, a dominating form-fit or a metallic bonding. The resulting joining
mechanism depends on the properties of the joining partners and the process pa-
rameters. Provided that the joint is thoroughly designed, in principle, all three
joining mechanisms are suitable to achieve a transferable load comparable to the
strength of the parent material of the components to be connected. However, the
investigations presented here, focus on the manufacturing of magnetic pulse
welded joints, i. e. connections based on metallic bonding.
Experimental investigations considering the joining of aluminum tubes (materi-
al: Al99.9; Ø20 x 1 mm) to solid mandrels (material: TiAl6V4; Ø17 mm) were
carried out at the IUL using a forming machine (SMU1500) with a capacitance of
80 µF, an inner inductance of 75 nH, an inner resistance of 6.8 mΩ and a tool coil
with a nominal diameter of 20 mm, a length of 30 mm and a number of turns of
12. Considering the results of the experiments performed with the model experi-
ment described above, in the first tests, a capacitor charging energy of 500 J was
chosen because velocity measurements had proved that, in this case, an impacting
velocity in the range of 100 m/s is reached (see [19]). However, the resulting joint
was merely based on force-fit, i. e. an elastic plastic bracing of tube and mandrel;
and no welding occurred. Consequently, the charging energy was increased to
1,000 J and 1,500 J. Provided that the same gap width between the joining part-
ners is given, a higher charging energy will directly lead to a higher impacting ve-
locity and this, in turn, causes local welding of tube and mandrel.
The achieved weld quality was evaluated by means of micrographic investiga-
tions carried out by IW. As shown in Figure 23.12, the weld quality differs along
the axis of the tube. For both charging energies, in the middle of the compression
area (z≈0 mm), gaps between the joining partners can be identified although small
adhesions of material particles welded on each other can be found. With an in-
creasing charging energy, the length of this zone was reduced. A zone of high
welding quality is located adjacent to this area. For the charging energy of 1,500 J,
this zone continues up to the axial end of the joining area. Contrarily, in the case
of the specimen welded with a charging energy of 1,000 J, another region charac-
terized by a strongly inhomogeneous layer between the aluminum and the titanium
was found. This area extends to the end of the compression zone and, again, indi-
cates an inadequate joining quality.
Fig. 23.12 Micro-structural state of joints produced by EMF [19]
A possible explanation for the different required impacting velocities in case of

the model experiment and in case of EMF is the influence of the impacting angle.
In the model experiment, this angle is 5°. Contrarily, in EMF, the angle is much
higher and varies along the axis during the forming as indicated in Figure 23.13.
Fig. 23.13 Distribution of impact velocity and angle along the workpiece axis [19]
In the center (z = 0 mm) of the compression area, the highest workpiece

velocity (>> 100 m/s) occurs but the extremely high impact angle (90°) prevents
welding because the velocity is directed perpendicularly to the contact surface
without any parallel component. This combination seems to be unsuitable for the
generation of proper welds. For higher z-values, the absolute value of the velocity
and especially the perpendicular component decreases while the parallel compo-
nent of the velocity increases due to the decreasing impact angle. In the region,
where the high welding quality can be observed, favorable combinations of the
impacting velocity and the impacting angle were reached. This approach suggests
that not only high workpiece velocities are required for the generation of good
welding properties but matching conditions of local impact velocity and angle also
have to be granted. In order to determine such parameter constellations the model
experimental setup can be used as it allows a separate modification of the impact
velocity and impact angle.
23.3.3 Forming by Gas Detonation into a Die

To investigate the forming by gas detonation into a form defining tool a die was
installed in the fixture, as shown in Figure 23.14. Using the tooling components,
larger local strains are generated due to the local constraints of the tube's expan-
sion, where the formed part's corner sections are developed. Tests were carried out
with semi-finished products, which were furnished with a dot-matrix. This enabled
the strain distribution to be more precisely investigated. Following these tests, an
evaluation of the geometric changes was performed with the aid of the Argus
analytical optical system made by the company GOM, Braunschweig. These
achievable high strains (of approx. 30 % in the free sections) indicate a favorable
influence of the high forming speeds on the attainable formability during the gas
detonation. Based on these promising results, a more sophisticated die concept
was realized considering the process limits. The tool geometry can be varied by
means of different die inserts. Furthermore, the volume between the tube's wall
and the tool's contour can be evacuated using a vacuum pump in order to avoid the
influence of trapped air on the forming or diesel effects.
Fig. 23.14 Testing apparatus for die-forming tests
The tooling system employed permits reference tests to be carried out using
identical tool geometries in the technically comparable procedure of hydro-
forming (IHU). For the die-forming tests, the aluminum alloy (AlMgSi1) was se-
lected as the material. During the testing, the tool's internal contour was modified
by means of exchangeable tool inserts. The inserts' internal contour exhibited a cy-
lindrical form, whose diameter was increased step-by-step. It is shown that using
AlMgSi1, a maximum diameter of 48 mm can be obtained without cracking.
Assuming a homogeneous tangential strain, this increase in diameter represents a

strain of 20 %, which is a high value for this material. For this test, the workpiece
was subjected to an initial pressure of p0 = 4 MPa leading to a theoretical maxi-
mum pressure of approx. 80 MPa during the detonation.
During the tests, it was possible to detect a significant influence of the initial
pressure used on the form's quality and its surface finish. For the 47 mm diameter
die-insert at an initial pressure of p0 = 3.0 MPa, a coarse grain-like structure is vis-
ible at three locations on the workpiece's surface. This suggests incipient failure at
the extruded tube's weld seam. Elevating the initial pressure to p0 = 3.5 MPa con-
siderably improves the workpiece's entire surface quality. In these tests, it was
also possible to slightly improve the forming. Figure 23.15 shows a residual devia-
tion from the die-tool's contour, which can be attributed to spring-back and to the
different pressure ratios due to the formation of the detonation wave.
Fig. 23.15 Effect of initial pressure on the workpiece surface (left) and quality of the form-
ing for increasing initial pressure (right)
During the tests, failures also occurred due to crack formation when using the
large die-inserts. Owing to the cracking, the detonation gases escape into the eva-
cuated tool space and then leak out through the tool's mating surfaces. This pro-
duces combustion marks of up to 3 mm deep in the steel inserts or, as may be the
case, in the tool's mating surfaces.
During the static reference tests using the hydroforming test tool, it was not
possible to successfully carry out experiments using even the smallest tool-insert
with a 44 mm diameter. By using an Al 99.5 tube, it was possible to form a
workpiece to this maximum diameter although, here, failure had also been ob-
served already for slightly larger die-inserts. To guarantee comparability, no axial
forces were applied during the hydro-forming tests. The die-force was only used
for sealing.
23.4 Conclusions
To investigate the complex interdependencies during the impulse forming of tubu-
lar parts, complementary experimental and numerical investigations were carried
out considering two different sub-systems. Sub-system 1 considers the workpiece
acceleration. Here, electromagnetic forming and forming by gas detonation were

regarded as acceleration mechanisms. In sub-system 2, the impact of the work-
piece and an obstacle, i. e. a joining partner or a form-defining tool, is analyzed.
Focusing on sub-system 1 for the electromagnetic forming of tubes, the inter-
dependencies between the acting loads, characterized by the time-dependent
course of the coil current, and the magnetic pressure were characterized. Regard-
ing the gas detonation process, the general feasibility of this method as a forming
process was shown. Most notable are the extended reachable strains in tubular
profiles made of aluminum alloys AlMgSi1and Al99.5, compared to conventional
forming processes. Microscopic investigations of the crystallographic structure
have delivered first indications regarding the reasons for the observed effect. To
enable a process planning for this innovative process, further experimental and
metallurgical investigations have to be carried out to deduce concrete laws for the
material behavior. Further investigations will focus on the identification of process
limits for different alloys, the adaptation of tool construction to the process and the
advancement of numerical tools for a reliable virtual process design.
Considering sub-system 2, the investigations about EMF were focused on join-
ing and especially on impact welding. It was proven that, for the feasibility of this
special application, the adjustment of the impacting velocity and angle are deci-
sive. To determine suitable values for these process parameters a model high-
velocity joining setup was realized and the potential of welding specimens with
this device was proven. To transfer the results from these investigations to the
electromagnetic welding process the investigation results considering sub-system
1 can be used. Thus, both approaches will be exploited complementarily in order
to allow a target-oriented process dimensioning for the electromagnetic welding
technology.
References
[1] Lesemann, M., Sahr, C., Hart, S., Taylor, R.: SuperLIGHT-CAR – the Multi-Material
Car Body. In: Proceedings of the 7th LS-DYNA Anwenderforum, Bamberg (2008)
[2] Kleiner, M., Geiger, M., Klaus, A.: Manufacturing of Lightweight Components by
Metal Forming. In: Annals of the CIRP Manufacturing Technology, 53rd General As-
sembly of CIRP, vol. 52(2), pp. 521–542 (2003)
[3] Psyk, V., Risch, D., Kinsey, B.L., Tekkaya, A.E., Kleiner, M.: Electromagnetic form-
ing – A review. J. Mater. Process. Tech (2011),
doi: 10.1016/j.jmatproctec.2010.12.012
[4] Beerwald, C.: Grundlagen der Prozessauslegung und –gestaltung bei der elektromag-
netischen Umformung. Dr. Ing. Dissertation, TU Dortmund (2005)
[5] Kaden, H.: Wirbelströme und Schirmung in der Nachrichtentechnik. Springer, Berlin
(1959)
[6] Weber, M., Hermes, M., Brosius, A., Beerwald, C., Gersthteyn, G., Olivier, H.,
Kleiner, M., Bach, F.-W.: Process Investigation of Tube Expansion by Gas Detona-
tion. In: Proceedings of the International Conference on High Speed Forming
(ICHSF) 2006, pp. 161–174 (2006) ISBN 3-00-018432-5
[7] Esser, B.: Die Zustandsgrößen im Stoßwellenkanal als Ergebnisse eines exakten Rie-
mannlösers. Dissertation, RWTH Aachen (1991)
[8] Bathe, K.J.: Finite Elemente Methoden, 2nd edn. Springer, Berlin (2002)
[9] Psyk, V., Beerwald, C., Homberg, W., Kleiner, M.: Extension of Forming Limits by
Using a Process Combination of Electromagnetic Forming and Hydroforming. In:
Proceedings of the 8th International Conference on Technology of Plasticity, ICTP
(2005) ISBN 88-87331-74-X
[10] Psyk, V.: Prozesskette Krümmen – Elektromagnetisch Komprimieren – Innenhoch-
druckumformen für Rohre und profilförmige Bauteile. Dr. Ing. Dissertation, TU
Dortmund (2010) ISBN 978-3-8322-9026-9
[11] Gottstein, G., Shvindlerman, L.S.: Grain boundary migration in metals: thermody-
namics, kinetics, applications, vol. 385. CRC Press, Boca Raton (1999)
[12] Johnston, W.G., Gilman, J.J.: Dislocation Velocities, Dislocation Densities and Plas-
tic Flow in Lithium Fluoride Crystals. Journal of Applied Physics 30(2), 129 (1959)
[13] Gottstein, G.: Physik. Grundlagen der Materialkunde. Springer, Berlin (2001)
[14] Roos, A.: Fast-moving dislocations in high strain rate deformation. Groningen Uni-
versity Press (1999)
[15] Alshitz, V.I., Indenbom, V.L.: Dynamic dragging of dislocations. Soviet Physics Us-
pekhi 18(1) (1975)
[16] Mattissen, D.: Insitu Untersuchung des Einflusses der Tripelpunkte auf die Korngren-
zenbewegung in Aluminium, Dissertation, RWTH Aachen (2004)
[17] Czubayko, M.: Korngrenzenbewegung in Aluminium und Zink, Dissertation, RWTH
Aachen (1998)
[18] Vovk, V.T.: Gasexplosion als Werkzeug in der Fertigungstechnik. Habilitation, Mag-
deburg, Univ. (1999)
[19] Psyk, V., Gershteyn, G., Demir, O.K., Brosius, A., Tekkaya, A.E., Schaper, M., Bach,
F.-W.: Process Analysis and Physical Simulation of Electromagnetic Joining of Thin-
Walled Parts. In: Proceedings of the International Conference on High Speed Form-
ing, pp. 181–190 (2008) ISBN: 3-9809535-3-X
[20] Kreimeyer, M., Wagner, F., Zerner, I., Sepold, G.: Laser beam joining of aluminium
with titanium with the use of an adapted working head. In: DVS-Berichte, vol. 212,
pp. 317–321. DVS-Verlag, Düsseldorf (2001)
Author Index
Abele, E. 3, 245 Hirt, G. 439

Andres, M. 53, 285 Hömberg, D. 29, 203
Aurich, J.C. 3, 81
Kiliclar, Y. 459
Bach, Fr.-W. 491 Kirsch, B. 3, 81
Bauer, J. 245 Kleiner, M. 491
Behrens, B.-A. 3, 403 Klocke, F. 3, 29, 101
Biermann, D. 3, 121 Krause, A. 3, 329
Blum, H. 53, 121 Kriechenbauer, S. 383
Bouabid, A. 29, 81 Krimm, R. 403
Bouguecha, A. 403 Kröger, M. 143
Brandt, C. 53, 329 Kroiß, T. 3, 419
Brecher, C. 3, 29, 179
Brinksmeier, E. 3, 329 Lahl, M. 475
Britz, R. 361 Laurischkat, R. 3, 459
Löser, M. 3, 29, 225
Carstensen, C. 53, 143
Czora, M. 3, 403 Maaß, P. 53, 329
Mahr, F. 3, 265
Deichmueller, M. 29, 143 Maier, T. 361
Denkena, B. 3, 29, 143, 285 Maischak, M. 285
de Payrebrune, K.M. 143 Matthias, T. 403
Druwe, T. 475 Meier, H. 3, 459
Milbrandt, M. 475
Eberhard, P. 309
Neugebauer, R. 475
Engel, U. 3, 419
Niebsch, J. 53, 329
Esser, M. 179
Penter, L. 383
Franzke, M. 439 Pischan, M. 3, 245
Friedmann, M. 245 Puchhala, S. 439
Gaugele, T. 309 Rademacher, A. 53, 121

Großmann, K. 3, 29, 225, 383 Ramlau, R. 53, 329
Rasper, P. 3, 203
Hardtmann, A. 29, 383 Reese, S. 459
Heinisch, D. 3, 285 Reinl, C. 245
Heisel, U. 3, 309 Rott, O. 29, 203
Hellfritzsch, U. 475
Hemker, T. 245 Scheidler, A.V. 3, 121
Hermes, R. 3, 29, 179 Schiller, S. 475
514 Author Index
Schröder, A. 53, 143 Vehmeyer, J. 329

Schwarz, F. 361 von Stryk, O. 245
Shi, Y. 265 von Wagner, U. 265
Steinmann, P. 29, 81
Stephan, E.-P. 53, 285 Wegner, H. 101
Storchak, M. 3, 309 Weinert, K. 121
Weiß, M. 3, 29, 101
Tekkaya, A.E. 491 Wiedemann, S. 53, 143
Wiemer, H. 383
Uhlmann, E. 3, 203, 265
Ulbrich, H. 361 Zaeh, M.F. 361
Subject Index
a posteriori error estimate 54, 61, 64, cold rolling 475, 479
132 compensation 398
residual 58 geometrical approach 259
abrasive machining 39 model based 258
acoustic emission 19, 104 compliance 246
adaptive finite element method 132 compliance simulation 185
additional stability lobes 238 comprehensive spatial multiple degrees
AE sensor 19 of freedom (MDOF) model 265
AFPM 385 contact
analogous models 35 bilateral 364
analytical-empirical process model 110 contact length 114
arbitrary lagrangian eulerian approach contact pressure 452
31 contact problem 54
axes dynamic 64
elasticities 248 elasticity 54
friction 54
frictional 60
back-step test 11
control system 111
backward can extrusion 426
convergence criterion 414
balance quality grad 330, 331
coolant forces 106
balancing 72, 74, 330, 352
cooling system 450
dual-plane balancing 332, 334
coulomb's Law 125, 131
single-plane balancing 331, 332,
coupled model 110
358
coupled simulations 48
balancing plane 331, 332, 334
coupling 345
bending vibration 184
coupling of models 44
blank holder 388
coupling of the models 302
bulk forming 40
critical chip thickness 272
critical depth of cut, 232
Campbell diagram 279 cross rolling 476, 478, 489
chatter 47, 179 cutting edge radius 273
chatter criterion 206, 278 cutting force 253
chatter detection 206 action reaction law 216
chatter frequencies 278 assumptions 217
chip longitudinal section 113 comparison 369
chip thickness modulation 189 modular model 366
chisel process structure interaction 215
orientation 371 spectrum 221
circularity test 10 temperature 210
cold forging 419, 420 uncut chip thickness 215
cold forming 475, 487, 489 cutting force coefficients 36
516 Subject Index
cutting force model 36, 301 finite element models 31

cutting force variations 45 finite element simulation 128
cutting tool 143 finite elements (FE) 268
finite element method 43, 54, 336
damage cases 288 adaptive 54, 57, 60, 63
dampening 108 mixed 56
deflection characteristic 420, 422, 425, fixed frame 272
428 flank face contact 194
deformation efficiency 42 flexible multi-body simulation 30
deforming forces 42 force
delay differential equation 47 cutting 253, 366
design 411 nonlinear law 364
device’s eigenfrequency 18 ploughing 366
Dexel model 129, 218 set-valued 365
diamond machining 329 shearing 366
die cushion 388 force coefficients 36
direction cosine matrix 271 force measurements 357
directional coefficients 228 force model 340
displacement development 5 force ratio factor 116
dynamic compliance behavior 180 force transducer 16
dynamic cutting force 188 force-controlled press 422, 426, 429
dynamic cutting force model 196 forces in cold extrusion processes 18
dynamic cutting forces 38 form deviation 354, 355
dynamic force model 191 formed parts 415
dynamic machine tool behavior 7 forming limit curve 394
dynamic Machine Tool Behavior 7 forming process model 386
dynamic process behavior 187 fourier series 229
dynamic Process Modeling 399 frequency response function 8, 103,
dynamic weak points 197 241, 266
dynamical loads 18 friction coefficient 125
friction law 413
full forward extrusion 423, 430, 432
eigenfrequencies 103 full immersion cutting 277
eigenfrequency 124
eigenmode 112
gap conductance 449
elastic work roll 445
gear
elasticity 249
backlash 249
elasto-plasticity 60
gear qualities 476, 478, 479
multi yield 61
gear rolling 475, 476, 477, 489, 490
Euler-Bernoulli beam 268
geometric model 10
excitation frequencies 277
geometric-kinematical simulation 128
experimental modal analysis 9
geometry assessment 24
exponential cutting force models 37
grinding forces 113
grinding models 39
facing process grinding wheel
optimization 376 toroid 121
face turning 334, 340 grinding wheel contours 137
fast-fourier-transform 206
FEM
heat sources 12
metal cutting simulation 367
helicoidal mills 234
FEM simulation 404, 487, 490
helix angle 233, 274
finite element mode 112
Subject Index 517
high gearing 475, 476, 478, 479, 480, measurement errors 208
489 mechanical (tactile) techniques 24
high performance cutting 179 mechanistic cutting force models 36
high-speed press 404 meshing 487, 488
Hopf instability 277 metal cutting simulation
hot rolling experiments 448 FEM 367
hydrodynamic pressure 107 friction conditions 368
hysteresis 6 material laws 367
metal forming 40
inconel 718, 109 metal forming processes 403
incremental forming 487 metrological acquisition 188
infrared radiatio 21 micro cutting 265
instability islands 238 micro force model 342
integrated FE simulation model 425, micro milling machine tool 266
426, 428, 429, 432 micro sensors 21
interacting system 45 micro turning 341
interaction effects 441 mixed formulation 60
interferometry 23 microscopic approaches for grinding
inverse problem 54, 67, 69 models 40
milled surfaces 280
milling 227, 245
joint
milling system 45
actuated 248
modal analysis 182, 183, 266
virtual 248
robot 254
modal analysis 9, 124
kinematics 10 modal models 237
modal parameters 36
lagrange multiplier 60 mode shapes 269
laser interferometer measurement 12 model coupling 384
laser tracker system 483, 484 model integration 48
light fringe projection 24 multi body simulation 110
linear cutting force models 37 multi degree of freedom system 229
multi frequency solution 226
multi body models 30
machine behavior 480
multi body simulation 30
machine concept 479, 480, 490
multi body system 213
machine tool behavior 123
multi body system
machining tests 204
contact 363
macroscopic approaches for grinding
MBSim 362
models 40
multi body system
main spindle systems 180
dynamics 245
mass spring damper systems 35
multi dexel model 251
material behavior 488
material flow 476, 487
material laws NC-shape grinding 121
metal cutting simulation 367 Neo-Hookean model 33
material removal 114 net-shape production 419
calculation 245 newmark-scheme 34
material removal model 217 nonholonomic constrain 271
material removal process 348 nozzle 114
measurement numerical simulation 218
temperature 369 nyquist criterion 227
518 Subject Index
objective 373 radial immersion 242

offline-coupling 403 regenerative chatter 225
operation state 372 regenerative vibrations 179
optical techniques 24 regularization 68, 353
optimization 373 removal predictor 128
comparison 376 residual stresses 290
facing process 376 robot
forces 377 industrial 245
implicit filtering 361 stiffness 255
material removal rate 374 rolling force 446, 484
product quality 378 rolling machine 480, 487
productivity 374 rolling process. See Introduction
ordinary differential equation 344 rolling velocitiy 451
oscillator chains 268 rotating boundary conditions 442
rotating frame 272
parameter identification 54, 67, 133, rotor-dynamics 181
205, 218 roughness of technical surfaces 22
parametric process model 430, 432 runout 281
path deviation 246
path planning 137 SEM measurements 291
phenomenological approach 405 shear stress 447
piezo ceramic actuators 281 sheet-metal forming 40
piezoelectric Elements 16 signorini problem 65, 129, 131
piola-stress tensor 32 simulated stability chart 194
pitch difference 482 simulated surface 118
plastic flow of the strip 445 simulated surface structure 197
positioning accuracy 123 simulation 349
prediction simulation
surface contour 374 dynamics 250
workpiece contour 372 milling robot 257
press measurement 406 simulation strategy 441
process analysis 15 simulation techniques 48
process force 333, 409 single degree of freedom system 237
calculation 251 solution algorithm
process force model 130, 237 uzawa 56
process forces 16, 125, 246 specific force constants 275
process knowledge 432, 433 speed-stroke grinding 101
process models 36, 44, 340 spindle dynamics 299
process optimization 432 spindle-tool holder-tool
process parameter 343 system (STT) 267
process stability splitting of resonance peaks 184
machine effects 220 stability 45
supple workpiece structure 221 stability boundaries 225
thermal effects 210 stability lobe diagram 225, 265
uncut chip thickness 221 stability prediction 277
workpiece dynamics 209 stability simulation 38
workpiece temperature 211 static machine compliance 5
process temperatures 126 static machine tool behavior 4
pyrometry 20 stiffnesses 6
Subject Index 519
straight fluted 234 tool extraction 285, 286

strain gauges 16 tool grinding 143
strip profile 440 tool guidance 389
stroke-controlled press 421, 423 tool loads 488
structural analysis 3 tool trajectories 346
structure models 44 tool-workpiece-thermal element method
surface assessment 22 20
surface conditions 412 tooth height coefficient 478, 479, 481
surface generation 333 tooth passing frequency 232
surface profile 198 total deformation energy 42
surface roughness 330, 354, 356 transient heat transfer 449
surface topography 333, 334, 345, 354 tribological effects 441
surface visualization 345, 353 turning 334, 361
system matrices 336
ultra-precision machining 17, 341
temperature ultraprecision 333
measurement 369 ultraprecision turning 70
temperature distribution 20 unbalance 329, 339
temperature profile 15 dynamic unbalance 332
temperature-Influence 407 moment unbalance 331
test rig 182, 292 uncut chip thickness 46
the process dynamics 16
theoretical methods 41 variational inequality 56
theoretical surface roughness 273 vibration Equation 70, 336, 339
thermal damage 20 vibration model 336
thermal evolution 453 vibrations 329
thermal shrink-fit tool holders 285 v. Mises 488, 489
thermocouples 13, 20 vitrified bonded CBN 117
thermocouples and thermographic
cameras 21
waviness 22
thermoelectric and radiation
wear 454
measurement methods 20
wear models 443
thermo-elastic workpiece 213
wear work 444
thermography 20
white light interferometry 23
thermomechanical contact problem 131
workpiece
thermo-mechanical effects 440
orientation 371
thermo-mechanical tool computation.
workpiece analysis 22
siehe
workpiece material 109
thickness profile 446
three-stage tool system 411
Tikhonov-functional 69, 72 xForm 396
time domain simulations 48, 212
tool centre point 240 zeroth order approximation 226
tool clamping 268 zeroth-order-approximation method
tool displacement 480, 482, 483 277

Notes On Production Engineering

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Notes On Production Engineering

Încărcat de

Drepturi de autor:

Formate disponibile

Lecture Notes in Production Engineering

For further volumes:

Process Machine Interactions

ISSN 2194-0525 e-ISSN 2194-0533

c Springer-Verlag Berlin Heidelberg 2013

Printed on acid-free paper

The Priority Program 1180 “Prediction and Manipulation of Interactions between

Prof. Dr.-Ing. Berend Denkena Dr.-Ing. Ferdinand Hollmann

Part II: Grinding

Part III: Cutting

Part IV: Forming

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513

Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515

E. Abele, J. C. Aurich, B.-A. Behrens, D. Biermann, C. Brecher,

1.2 Structural Analysis of Machine Tools

1.2.1 Static Machine Tool Behavior

Fig. 1.1 Characterization of static machine compliance (based on [1])

A method to determine the deflection of presses under a static load is described

The standard describes the determination of the deflection in forming direction

1.2.2 Dynamic Machine Tool Behavior

1.2.2.1 Frequency Response Function

Fig. 1.4 Principle of the measurement of frequency response functions

1.2.2.2 Experimental Modal Analysis

Experimental modal analysis is a method for determining the dynamic characteris-

1.2.3 Measurement of Kinematics

1.2.3.1 Circularity Test

1.2.3.2 Back-Step Test

1.2.3.3 Measurement of the Machine Axes using Laser Interferometry

1.2.4 Thermal Machine Tool Behavior

conduction, convection or radiation caused by ambient heat flow. In addition, the

Table 1.1 Comparison of contact and non-contact measurement of temperature [12]

Contact measurement Non-contact measurement

To consider thermal convection effects the temperature of the environment is

Fig. 1.10 Temperature profile of a high speed stamping machine [9]

1.3 Process Analysis

1.3.1 Process Force Measurements

1.3.1.1 Force Transducer based on Strain Gauges

1.3.1.2 Force Measurement Based on Piezoelectric Elements

The piezoelectric effect is based on an interaction between electrical field strength,

1.3.1.3 Force Measurement in Ultra-Precision Machining

In ultra-precision machining, all process parameters are - compared to conven-

0 Cutting force 0,5

1.3.1.5 Force Measurement in Cold Extrusion

1.3.2 Acoustic Emission (AE) Measurements

1.3.3 Process Temperature Measurements

1.3.3.1 Investigation of the Temperatures in Cutting

Thermocouples have a relatively low time resolution and it is difficult to place

1.3.3.2 Investigation of the Temperatures in Grinding

An overview of different methods for temperature measurements is presented

1.3.3.3 Experimental Investigation of the Temperature in the Cutting Zones

To determine the temperature in the cutting zone, experimental investigations are

1.4 Workpiece Analysis

1.4.1 Surface Assessment

1.4.2 Geometry Assessment

Fig. 1.16 Measuring configuration of a white light fringe projection [38]

Fig. 1.17 Scanner system (a) and a scanned workpiece (b)

The determination of the resulting deviation is evaluated by comparing the

techniques can be used, depending on the special parameters to be determined.

C. Brecher, A. Bouabid, M. Deichmueller, B. Denkena, K. Großmann,

The prediction of process machine interactions requires models of the subsystems

using measured frequency response functions. The different approaches to model

2.2 Models of the Machine Behavior

2.2.1 Multi-Body Models