Documente Academic
Documente Profesional
Documente Cultură
Draft
c VICTOR E. SAOUMA,
SPRING 1996
Contents
1 INTRODUCTION 1{1
1.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 1{1
1.2 Finite Element Discretization : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 1{2
1.3 Generalized Nodal Quantities : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 1{3
1.4 Computer Programs : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 1{3
1.5 Examples of applications : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 1{4
1.6 Chracteristic Matrix : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 1{4
2 PRELIMINARY CONSIDERATIONS 2{1
2.1 Notation : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 2{1
2.1.1 Engineering Notation : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 2{1
2.1.2 Dyadic/Vector Notation : : : : : : : : : : : : : : : : : : : : : : : : : : : : 2{2
2.1.3 Indicial/Tensorial Notation : : : : : : : : : : : : : : : : : : : : : : : : : : 2{2
2.2 Integral Theorems : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 2{3
2.2.1 Integration by Parts : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 2{4
2.2.2 Green-Gradient Theorem : : : : : : : : : : : : : : : : : : : : : : : : : : : 2{4
2.2.3 Green Theorem : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 2{4
2.2.4 Gauss-Divergence Theorem : : : : : : : : : : : : : : : : : : : : : : : : : : 2{4
2.2.5 Gauss-Green Theorem : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 2{5
2.3 Elliptic, Parabolic and Hyperbolic Equations : : : : : : : : : : : : : : : : : : : : 2{5
E 2-1 Seepage Problem;[?] : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 2{6
E 2-2 Difusion Problem;[?] : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 2{7
E 2-3 Wave Equation, [?] : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 2{9
3 REVIEW OF ELASTICITY 3{1
3.1 Stress : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 3{1
3.1.1 Stress Traction Relation : : : : : : : : : : : : : : : : : : : : : : : : : : : : 3{2
3.2 Kinematic/Strain : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 3{3
3.3 Equilibrium; Momentum Principle : : : : : : : : : : : : : : : : : : : : : : : : : : 3{7
3.4 Stress-Strain Relations in Elasticity : : : : : : : : : : : : : : : : : : : : : : : : : : 3{8
3.4.1 Plane Strain : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 3{10
3.4.2 Axisymmetry : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 3{10
0{2 CONTENTS
3.4.3 Plane Stress : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 3{10
Draft
3.5 Pore Pressures : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 3{11
3.6 y Field Equations for Thermo- and Poro Elasticity : : : : : : : : : : : : : : : : : 3{11
3.7 y Basic Equations of Anisotropic Elasticity : : : : : : : : : : : : : : : : : : : : : 3{13
3.8 Coordinate Transformations : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 3{16
3.8.1 y Plane Stress-Strain Compliance Transformation : : : : : : : : : : : : : : 3{17
3.9 Strain Energy Density : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 3{17
3.10 Summary : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 3{17
4 SCALAR FIELD PROBLEMS 4{1
4.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 4{1
4.1.1 Heat Transfer : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 4{1
4.2 Derivation of the Diusion Problem : : : : : : : : : : : : : : : : : : : : : : : : : 4{3
4.2.1 Simple 2D Derivation : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 4{3
4.2.2 yGeneralized Derivation : : : : : : : : : : : : : : : : : : : : : : : : : : : : 4{4
4.2.3 Boundary Conditions : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 4{6
4.3 Summary : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 4{6
5 DIRECT STIFFNESS METHOD 5{1
5.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 5{1
5.2 [ID] Matrix : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 5{2
5.3 LM Vector : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 5{3
5.4 Assembly of Global Stiness Matrix : : : : : : : : : : : : : : : : : : : : : : : : : 5{3
E 5-4 Global Stiness Matrix Assembly : : : : : : : : : : : : : : : : : : : : : : : 5{4
5.5 Skyline Storage of Global Stiness Matrix, MAXA Vector : : : : : : : : : : : : : : 5{6
5.6 Augmented Stiness Matrix : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 5{9
E 5-5 Direct Stiness Analysis of a Truss : : : : : : : : : : : : : : : : : : : : : : 5{14
5.7 Computer Program Flow Charts : : : : : : : : : : : : : : : : : : : : : : : : : : : 5{19
5.7.1 Input : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 5{19
5.7.2 Element Stiness Matrices : : : : : : : : : : : : : : : : : : : : : : : : : : : 5{20
5.7.3 Assembly : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 5{20
5.7.4 Decomposition : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 5{22
5.7.5 Load : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 5{22
5.7.6 Backsubstitution : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 5{23
5.7.7 Reactions : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 5{23
6 VARIATIONAL AND ENERGY METHODS 6{1
6.1 Variational Calculus; Preliminaries : : : : : : : : : : : : : : : : : : : : : : : : : : 6{1
6.1.1 Euler Equation : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 6{1
6.1.1.1 Problem Statemen : : : : : : : : : : : : : : : : : : : : : : : : : : 6{1
6.1.1.2 Derivation : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 6{2
6.1.1.3 Solution : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 6{4
6.1.2 Boundary Conditions : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 6{5
Victor Saouma Finite Elements
CONTENTS 0{3
6.2 Work and Strain Energy; Denitions : : : : : : : : : : : : : : : : : : : : : : : : : 6{6
Draft
6.2.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 6{6
6.2.2 Internal Strain Energy : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 6{7
6.2.3 Potential of External Work : : : : : : : : : : : : : : : : : : : : : : : : : : 6{8
6.2.4 Virtual Work : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 6{8
6.2.5 | Potential Energy : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 6{8
6.3 | Variational Statement of Potential Energy Principle : : : : : : : : : : : : : : : 6{10
6.4 Rayleigh-Ritz Method : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 6{13
7 INTERPOLATION FUNCTIONS 7{1
7.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7{1
7.2 Shape Functions : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7{1
7.3 C 0 : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7{2
7.3.1 Truss element : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7{2
7.3.2 Generalization : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7{3
7.3.3 Constant Strain Triangle Element : : : : : : : : : : : : : : : : : : : : : : 7{4
7.3.4 Further Generalization: Lagrangian Interpolation Functions : : : : : : : : 7{5
7.3.5 Rectangular Bilinear Element : : : : : : : : : : : : : : : : : : : : : : : : : 7{6
7.3.6 Solid Rectangular Trilinear Element : : : : : : : : : : : : : : : : : : : : : 7{8
7.4 C 1 : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7{9
7.4.1 Flexural : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7{9
7.4.2 C 1 : Hermitian Interpolation Functions : : : : : : : : : : : : : : : : : : : : 7{12
7.5 Characteristics of Shape Functions : : : : : : : : : : : : : : : : : : : : : : : : : : 7{12
7.6 Natural Coordinate System : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7{12
7.6.1 Straight Line : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7{14
7.6.2 Triangular Coordinates : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7{14
7.6.3 Volume Coordinates : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7{16
7.6.4 Interpolation Functions : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7{16
8 FINITE ELEMENT DISCRETIZATION and REQUIREMENTS 8{1
8.1 Discretization : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 8{1
8.2 General Element Requirements : : : : : : : : : : : : : : : : : : : : : : : : : : : : 8{4
8.3 Discretization Error and Convergence Rate : : : : : : : : : : : : : : : : : : : : : 8{4
8.4 Lower Bound Character of Minimum Potential Energy Based Solutions : : : : : 8{5
9 \SIMPLE" FINITE ELEMENTS 9{1
9.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 9{1
9.2 Rod Elements : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 9{1
9.2.1 Truss Element : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 9{1
9.2.2 Beam Element : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 9{2
9.3 Plane Stresss/Strain Elements : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 9{3
9.3.1 Constant Strain Triangle : : : : : : : : : : : : : : : : : : : : : : : : : : : 9{3
9.3.1.1 Cartesian Coordinate System : : : : : : : : : : : : : : : : : : : : 9{3
Victor Saouma Finite Elements
0{4 CONTENTS
9.3.1.1.1 Strain-Displacement Relations : : : : : : : : : : : : : : 9{3
Draft
9.3.1.1.2 Stiness Matrix : : : : : : : : : : : : : : : : : : : : : : 9{4
9.3.1.1.3 Internal Stresses : : : : : : : : : : : : : : : : : : : : : : 9{4
9.3.1.2 Natural Coordinate System : : : : : : : : : : : : : : : : : : : : : 9{4
9.3.2 Bilinear Quadrilateral : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 9{6
9.3.3 Element Assesment : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 9{6
9.3.3.1 Pascal's Triangle : : : : : : : : : : : : : : : : : : : : : : : : : : : 9{6
9.3.3.2 CST : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 9{7
9.3.3.3 BiLinear Element : : : : : : : : : : : : : : : : : : : : : : : : : : 9{8
9.3.3.4 Equilibrium and Compatibiliy in the Solution : : : : : : : : : : 9{9
10 ISOPARAMETRIC ELEMENTS 10{1
10.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{1
10.2 Element Formulation : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{3
10.2.1 Isoparametric Bar Element : : : : : : : : : : : : : : : : : : : : : : : : : : 10{3
10.2.2 Bilinear Isoparametric Element : : : : : : : : : : : : : : : : : : : : : : : : 10{5
10.2.2.1 Jacobian Operator : : : : : : : : : : : : : : : : : : : : : : : : : : 10{9
E 10-6 Jacobian Operators, [?] : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{10
10.2.3 Isoparametric Biquadratic Element : : : : : : : : : : : : : : : : : : : : : : 10{12
10.2.3.1 8 Noded, Serendipity Element : : : : : : : : : : : : : : : : : : : 10{13
10.2.3.2 9 Noded, Lagrangian element : : : : : : : : : : : : : : : : : : : : 10{13
10.2.3.3 Variable Element : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{15
10.2.4 Triangular Elements : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{19
10.2.5 pascal Triangles : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{19
10.2.6 Formulation (Uneditted) : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{19
10.3 Numerical Integration : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{21
10.3.1 Newton-Cotes : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{22
10.3.1.1 Newton-Cotes Quadrature for n = 2 and n = 3 : : : : : : : : : : 10{23
10.3.2 Gauss-Legendre Quadrature : : : : : : : : : : : : : : : : : : : : : : : : : : 10{24
10.3.2.1 y Legendre Polynomial : : : : : : : : : : : : : : : : : : : : : : : 10{26
10.3.2.2 Gauss-Legendre Quadrature for n = 2 : : : : : : : : : : : : : : : 10{26
10.3.3 Rectangular and Prism Regions : : : : : : : : : : : : : : : : : : : : : : : : 10{27
10.4 Stress Recovery; Nodal Stresses : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{27
10.5 Nodal Equivalent Loads : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{29
10.5.1 Gravity Load : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{29
10.5.2 Traction Load : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{30
10.5.3 Initial Strains/Stresses; Thermal Load : : : : : : : : : : : : : : : : : : : : 10{32
10.6 Computer Implementation : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{33
10.7 MATLAB Code for Shape Functions : : : : : : : : : : : : : : : : : : : : : : : : : 10{34
List of Figures
3.1 Stress Components on an Innitesimal Element :
:: : : : : : : : : : : : : : : : : 3{1
3.2 Stress Traction Relations : : : : : : : : : : : : :
:: : : : : : : : : : : : : : : : : 3{2
3.3 Strain Denition : : : : : : : : : : : : : : : : : :
:: : : : : : : : : : : : : : : : : 3{3
3.4 Equilibrium of Stresses, Cartesian Coordinates :
:: : : : : : : : : : : : : : : : : 3{7
3.5 Transversly Isotropic Material : : : : : : : : : : :
:: : : : : : : : : : : : : : : : : 3{14
3.6 Coordinate Systems for Stress Transformations :
:: : : : : : : : : : : : : : : : : 3{16
3.7 Fundamental Equations in Solid Mechanics : : :
:: : : : : : : : : : : : : : : : : 3{18
4.1 Flux vector : : : : : : : : : : : : : : : : : : : : ::: : : : : : : : : : : : : : : : : 4{2
4.2 Flux Through Sides of Dierential Element : : : :: : : : : : : : : : : : : : : : : 4{4
4.3 *Flow through a surface , : : : : : : : : : : : : : :: : : : : : : : : : : : : : : : : 4{5
4.4 Fundamental Equations of Heat Flow : : : : : : :: : : : : : : : : : : : : : : : : 4{7
5.1 Example for [ID] Matrix Determination : : : : : :: : : : : : : : : : : : : : : : : 5{5
5.2 Flowchart for Assembling Global Stiness Matrix : : : : : : : : : : : : : : : : : : 5{5
5.3 Example of Bandwidth : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 5{7
5.4 Numbering Schemes for Simple Structure : : : : : : : : : : : : : : : : : : : : : : 5{7
5.5 Beam Element : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 5{10
5.6 ID Values for Simple Beam : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 5{10
5.7 Flowchart for the Skyline Height Determination : : : : : : : : : : : : : : : : : : : 5{21
6.1 Variational and Dierential Operators : : : : : : : : : : : : : : : : : : : : : : : : 6{2
6.2 Strain Energy and Complementary Strain Energy : : : : : : : : : : : : : : : : : : 6{7
6.3 Single DOF Example for Potential Energy : : : : : : : : : : : : : : : : : : : : : : 6{9
6.4 Graphical Representation of the Potential Energy : : : : : : : : : : : : : : : : : : 6{10
7.1 Axial Finite Element : : : : : : : : : : : : : : : : :: : : : : : : : : : : : : : : : : 7{2
7.2 Constant Strain Triangle Element : : : : : : : : : : : : : : : : : : : : : : : : : : : 7{4
7.3 Rectangular Bilinear Element : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7{7
7.4 Solid Trilinear Rectangular Element : : : : : : : : : : : : : : : : : : : : : : : : : 7{8
7.5 Flexural Finite Element : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7{9
7.6 Shape Functions for Flexure of Uniform Beam Element. : : : : : : : : : : : : : : 7{11
7.7 Natural Coordinate System Along a Straight Line : : : : : : : : : : : : : : : : : 7{13
0{2 LIST OF FIGURES
7.8 Natural Coordinate System for a Triangle : : : : : : : : : : : : : : : : : : : : : : 7{15
Draft
9.1 Linear Triangular Element Subjected to Pure Bending : : : : : : : : : : : : : : : 9{7
9.2 Interelement Continuity of Stress : : : : : : : : : : : : : : : : : : : : : : : : : : : 9{9
9.3 Interelement Continuity of Strain : : : : : : : : : : : : : : : : : : : : : : : : : : : 9{10
10.1 Two-Dimensional Mapping of Some Elements : : : : : : : : : : : : : : : : : : : : 10{1
10.2 Actual and Parent Elements : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{2
10.3 Iso, Super, and Sub Parametric Elements : : : : : : : : : : : : : : : : : : : : : : 10{3
10.4 *Three-Noded Quadratic Bar Element : : : : : : : : : : : : : : : : : : : : : : : : 10{4
10.5 Four Noded Isoparametric Element : : : : : : : : : : : : : : : : : : : : : : : : : : 10{6
10.6 Dierential Element in Curvilinear Coordinate System : : : : : : : : : : : : : : : 10{8
10.7 Cross Product of Two Vectors : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{9
10.8 *Elements with Possible Singular Jacobians : : : : : : : : : : : : : : : : : : : : : 10{10
10.9 Serendipity and Lagrangian Quadratic Quadrilaterals : : : : : : : : : : : : : : : 10{12
10.10Serendipity Isoparametric Quadratic Finite Element: Global and Parent Element 10{13
10.11Shape Functions for 8 Noded Quadrilateral Element : : : : : : : : : : : : : : : : 10{14
10.12Shape Functions for 9 Noded Quadrilateral Element : : : : : : : : : : : : : : : : 10{16
10.13Nodal Numbering for Isoparameteric Elements : : : : : : : : : : : : : : : : : : : 10{17
10.14Pascal Triangles for Quadrilateral and Triangle Elements : : : : : : : : : : : : : : 10{18
10.15Newton-Cotes Numerical integration : : : : : : : : : : : : : : : : : : : : : : : : : 10{22
10.16Gauss-Legendre Integration over a Surface : : : : : : : : : : : : : : : : : : : : : : 10{27
10.17Extrapolation from 4-Node Quad Gauss Points [?] : : : : : : : : : : : : : : : : : 10{28
10.18Gravity Loading : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{30
10.19Traction Load in Isoparametric Elements : : : : : : : : : : : : : : : : : : : : : : 10{31
10.20Traction Load in Contiguous Isoparametric Elements : : : : : : : : : : : : : : : : 10{32
13.1 Arbitrary 3D Vector Transformation : : : : : : : : : : : : : : : : : : : : : : : : : 13{2
13.2 3D Vector Transformation : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 13{2
13.3 *Patch Tests : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 13{6
13.4 Eigenvectors Corresponding to a) Non-Zero and b) Zero Eigenvalues for a Square
Bilinear element : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 13{9
13.5 Independent Displacement Modes for a Bilinear Element : : : : : : : : : : : : : : 13{10
13.6 Hourglas Modes in Under-Integrated Quadratic Element : : : : : : : : : : : : : : 13{11
13.7 Rectangular Bilinear Element Subjected to Bending : : : : : : : : : : : : : : : : 13{12
13.8 Displacements Associated with Incompatible Modes for the Q6 Element : : : : : 13{13
13.9 Side Displacements Induced by Drilling d.o.f. : : : : : : : : : : : : : : : : : : : : 13{15
13.10Hybrid : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 13{16
14.1 Finite Element Formulation : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 14{1
14.2 Stresses in a Plate : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 14{2
14.3 Free Body Diagram of an Innitesimal Plate Element : : : : : : : : : : : : : : : 14{3
14.4 Displacements in a Plate : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 14{4
List of Tables
3.1 Number of Elastic Constants for Dierent Materials : : : : : : : : : : : : : : : : 3{15
4.1 Selected Examples of Diusion Problems : : : : : : : : : : : : : : : : : : : : : : : 4{1
4.2 Comparison of Scalar and Vector Field Problems : : : : : : : : : : : : : : : : : : 4{7
6.1 Essential and Natural Boundary Conditions : : : : : : : : : : : : : : : : : : : : : 6{6
7.1 Characteristics of Beam Element Shape Functions : : : : : : : : : : : : : : : : : 7{11
9.1 Interpretation of Shape Functions in Terms of Polynomial Series (1D & 2D) : : : 9{6
9.2 Polynomial Terms in Various Element Formulations (1D & 2D) : : : : : : : : : : 9{6
10.1 Shape Functions for Variable Node Elements : : : : : : : : : : : : : : : : : : : : 10{17
10.2 Shape Functions, and Natural Derivatives for Q8 Element : : : : : : : : : : : : : 10{17
10.3 Shape Functions and Derivatives for T6 Element : : : : : : : : : : : : : : : : : : 10{21
10.4 Weights for Newton-Cotes Quadrature Formulas : : : : : : : : : : : : : : : : : : 10{23
10.5 Integration Points and Weights for Gauss-Quadrature Formulaes Over the Inter-
val [,1; 1] : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10{25
10.6 Natural Coordinates of Bilinear Quadrilateral Nodes : : : : : : : : : : : : : : : : 10{28
12.1 Comparison of Scalar and Vector Field Problems, Revisited : : : : : : : : : : : : 12{14
13.1 Full and Reduced Numerical Integrations for Quadrilateral Elements : : : : : : : 13{10
13.2 Bilinsear and Exact Displacements/Strains : : : : : : : : : : : : : : : : : : : : : 13{12
14.1 Comparison of Governing Equations in Elasticity and Plate Bending : : : : : : : 14{13
14.2 Integration Rules for Mindlin Plate Elements : : : : : : : : : : : : : : : : : : : : 14{18
0{2 LIST OF TABLES
Draft
Chapter 1
INTRODUCTION
1.1 Introduction
1 Role of the course
Matrix \Rod" Element Structural Engineering
Finite Element \Continuum" Elements Mechanics/Field Problems
1D
1. Truss/spring
2. Beam (Bernouilli/Timoshenko)
3. Interface
2D (quadrilateral/triangular)
1. Plane Stress/Strain
2. Axisymmetric
3. Plate Bending (Mindlin/Kirscho)
4. Interface
3D
1. Shell (described by middle surface nodes)
2. Thick shell (described by top and bottom nodes)
3. Brick elements (tetrahedra, hexahedra)
Mass elements (1D, 2D, 3D)
Special
1. Crack
2. Innite
Approximation
Chapter 2
PRELIMINARY
CONSIDERATIONS
2.1 Notation
1 Equations of elasticity are expressed in terms of tensors, where
A tensor is a physical quantity, independent of any particular coordinate system yet
specied most conveniently by referring to an appropriate system of coordinates.
A tensor is classied by the rank or order
A Tensor of order zero is specied in any coordinate system by one coordinate and is a
scalar.
A tensor of order one has three coordinate components in space, hence it is a vector.
In general 3-D space the number of components of a tensor is 3n where n is the order of
the tensor.
2 For example, force and a stress are tensors of order 1 and 2 respectively.
3 To express tensors, there are three distinct notations which can be used: 1) Engineering; 2)
indicial; or 3) Dyadic.
4 Whereas the Engineering notation may be the simplest and most intuitive one, it often leads
to long and repetitive equations. Alternatively, the tensor and the dyadic form will lead to
shorter and more compact forms.
2.1.1 Engineering Notation
In the engineering notation, we carry on the various subscript(s) associated with each coordinate
axis, for example xx ; xy .
2{2 PRELIMINARY CONSIDERATIONS
2.1.2 Dyadic/Vector Notation
Draft
5 Uses bold face characters for tensors of order one and higher, ; . This notation is indepen-
dent of coordinate systems.
6 Since scalar operations are in general not applicable to vectors, we dene
A+B = B+A (2.1-a)
AB = ,BA (2.1-b)
A = Axi + Ay j + Az k (2.1-c)
AB = jAjjBj cos(A; B)
= A
x Bx + Ay By + Az Bz (2.1-d)
i j k
A B =
Ax Ay Az
(2.1-e)
Bx By Bz
grad A = rA = i @A +
@x @y j @A + k @A
@z (2.1-f)
div A = rA = i @x + j @y + k @z@ (iAx + jAy + kAz )
@ @
1. If there is one letter index, that index goes from i to n. For instance:
8 9
>
< a1 >
=
ai = ai = b a1 a2 a3 c = > a2 >
i = 1; 3 (2.2)
: a3 ;
assuming that n = 3.
2. A repeated index will take on all the values of its range, and the resulting tensors summed.
For instance:
a1ixi = a11 x1 + a12 x2 + a13 x3 (2.3)
Victor Saouma Finite Elements
2.2 Integral Theorems 2{3
3. Tensor's order:
Draft
First order tensor (such as force) has only one free index:
ai = ai = b a1 a2 a3 c (2.4)
Second order tensor (such as stress or strain) will have two free indeces.
2 3
D11 D22 D13
Dij 64 D21 D22 D23 7
5 (2.5)
D31 D32 D33
A fourth order tensor (such as Elastic constants) will have four free indeces.
4. Derivatives of tensor with respect to xi is written as ; i. For example:
@ @vi = v @v @Ti;j
@xi = ;i @x i i;i @xij = vi;j @xk = Ti;j;k (2.6)
Usefulness of the indicial notation is in presenting systems of equations in compact form.
For instance:
xi = cij zj (2.7)
this simple compacted equation (expressed as x = cz in dyadic notation), when expanded would
yield:
x1 = c11 z1 + c12 z2 + c13 z3
x2 = c21 z1 + c22 z2 + c23 z3 (2.8-a)
x3 = c31 z1 + c32 z2 + c33 z3
Similarly:
Aij = BipCjq Dpq (2.9)
A11 = B11 C11 D11 + B11 C12 D12 + B12 C11 D21 + B12 C12 D22
A12 = B11 C11 D11 + B11 C12 D12 + B12 C11 D21 + B12 C12 D22
A21 = B21 C11 D11 + B21 C12 D12 + B22 C11 D21 + B22 C12 D22
A22 = B21 C21 D11 + B21 C22 D12 + B22 C21 D21 + B22 C22 D22 (2.10-a)
@x d
= , x (2.14)
@x ,
10 Alternatively
Z Z Z Z Z Z Z Z
div qd
= qT :nd, , (r)T qd
(2.17)
,
19 Note:
1. The Laplace equation (r2 u = 0) is associated with the equilibrium problem, .
2. The Heat equation (Hut , K 2 r2 u = 0) corresponds to exponential decay. Also referred
to as Diusion equation (
uid
ow through porous media, irrotational
uid
ow, Saint
Venant torsion of elastic bars...).
3. The Wave equation (utt , K 2 r2 u = 0) corresponds to harmonic motion
Victor Saouma Finite Elements
2{6 PRELIMINARY CONSIDERATIONS
20 This classcation is established when solving Eq. 2.20 using the method of characteristics
Draft
because it is then observed that the character of the solutions is distinctly dierent for the three
categories of equations.
stands on permeable soil. Formulate the dierential equation governing the steady-state seepage
of water through the soil and give the corresponding boundary conditions.
Solution:
1. For a typical element of widths dx and dy (and unit thickness), the total
ow into the element
must be equal to the total
ow out of the element. Hence we have
@q y @q x
qy , qy + @y dx + qx , qx + @x dy = 0 (2.21-a)
, @q
@y
y dydx , @qx dxdy = 0
@x (2.21-b)
2. Using Darcy's law, the
ow is given in terms of the total potential (water elevation),
qx = ,k @
@x qy = ,k @
@y (2.22)
3. It may be noted that this same equation is also obtained in heat transfer analysis and in the
solution of electrostatic potential and other eld problems.
4. The boundary conditions are no-
ow boundary conditions in the soil at x = ,1 and x =
+1,
@ = 0; @ =0 (2.24)
@x x=,1 @x x=+1
at the rock-soil interface,
@ = 0 (2.25)
@y y=0
and at the dam-soil interface,
@ (x; L) = 0 for , h x + h (2.26)
@y 2 2
21 In addition, the total potential is prescribed at the water-soil interface,
(x; L)jx<,(h=2) = h1 ; (x; L)jx>,(h=2) = h2 (2.27)
The dierential equation in Eq. 2.23 and the boundary conditions in Eq. 2.24 to Eq. 2.27
dene the seepage
ow steady-state response.
1. The element force equilibrium requirements of a typical dierential element are, using
d'Alembert's principle,
A + A @
@x dx , A = A @ 2 u dx
@t2 (2.35)
2. The constitutive relation is
= E @u
@x (2.36)
3. Combining Eq. 2.35 and Eq. 2.36 we obtain
@2u = 1 @2u (2.37)
@x2 c2 @t2
4. The element interconnectivity requirements are satised because we assume that the dis-
placement u is continuous, and no additional compatibility conditions are applicable.
5. The boundary conditions are for t > 0
u(0; t) = 0 (2.38-a)
@u
EA @x (L; t) = F0 (2.38-b)
(2.38-c)
and the initial conditions are
u(x; 0) = 0 (2.39-a)
@u (x; 0) = 0 (2.39-b)
@t
With the above two sets of conditions, the formulation of the problem is complete, and Eq.
2.37 can be solved for the displacement response of the rod.
21Observations
Example 2-1 From the rst example we observe
1. The unknown state variables (or their normal derivatives) are given on the boundary.
These problems are for this reason also called boundary value problems, where we
should note that the solution at a general interior point depends on the data at every
point of the boundary.
2. A change in only one boundary value aects the complete solution; for instance, the
complete solution for depends on the actual value of h1 .
3. Elliptic dierential equations generally govern the steady-state response of systems.
Victor Saouma Finite Elements
2.3 Elliptic, Parabolic and Hyperbolic Equations 2{11
Example ??, 2-3 1. Comparing the governing dierential equations given in the three ex-
Draft
amples we note that in contrast to the elliptic equation, the parabolic and hyperbolic
equations Eq. 2.31 and Eq. 2.37 include time as an independent variable and thus
dene propagation problems. These problems are also called initial value problems
because the solution depends on the initial conditions.
2. We may note that analogous to the derivation of the dynamic equilibrium equations
of lumped-parameter models, the governing dierential equations of propagation
problems are obtained from the steady-state equations by including the \resistance
to change" (inertia) of the dierential elements.
3. The parabolic and hyperbolic dierential equations Eq. 2.31 and Eq. 2.37 would
become elliptic equations if the time-dependent terms were neglected. In this way
the initial value problems would be converted to boundary value problems with
steady-state solutions.
4. The solution of a boundary value problem depends on the data at all points of the
boundary. However, in propagation problem, the solution at an interior point may
depend only on the boundary conditions of part of the boundary and the initial
conditions over part of the interior domain.
Chapter 3
REVIEW OF ELASTICITY
3.1 Stress
1 A stress, Fig 3.1 is a second order cartesian tensor, ij where the 1st subscript (i) refers to
ti = nj ij (3.2)
2 3
11 12 13
b t1 t2 t3 c = b| n1 n{z2 n3 c} 64 21 22 23 7
5 (3.3)
direction cosines 31 32 33
| {z }
stress tensor
where n is a unit outward vector normal to the plane.
4 Note that the stress is dened at a point, and a traction is dened at a point and with respect
to a given plane orientation.
5 When expanded in cartesian coordinates,, the previous equation yields
8 Substituting the rst and third equation into the second one, we obtain
ds02 = dx
|
2 + dy 2 + dz 2 +2dudx + 2dvdy + 2dwdz + du2 + dv 2 + dw2
{z }
(3.8)
ds2
du = @u
@x dx + @u dy + @u dz
@y @z (3.9-a)
@v @v @v
dv = @x dx + @y dy + @z dz (3.9-b)
dw = @w
@x dx + @w dy + @w dz
@y @z (3.9-c)
10 Substituting this equation into the preceding one yields the nite strains
( " #)
ds0 2 , ds2 = 2 @u + 1 @u 2 + @v 2 + @w 2 dx2
@x 2 @x @x @x
( " 2 2 2 #)
+ @v 1 @u
2 @y + 2 @y + @y + @y @v @w dy2
( " 2 2 2 #)
+ @w 1 @u
2 @z + 2 @z + @z + @z @v @w dz2
+ @v + @u + @u @u + @v @v + @w @w dxdy
2 @x
@y @x @y @x @y @x @y
+ 2 @x + @u
@w + @u @u + @v @v + @w @w dxdz
@z @x @z @x @z @x @z
+ 2 @w @v @u @u @v @v @w @w
@y + @z + @y @z + @y @z + @y @z dydz (3.10-a)
11 We observe that ds0 2 , ds2 is zero if there is no relative displacement between A and B (i.e.
rigid body motion), otherwise the solid is strained. Hence ds0 2 , ds2 can be selected as an
appropriate measure of the deformation of the solid, and we dene the strain components as
ds0 2 , ds2 = 2"xxdx2 + 2"yy dy2 + 2"zz dz 2 + 4"xy dxdy + 4"xz dxdz + 4"yz dydz (3.11)
where
ij = 2"ij (i 6= j ) (3.19)
2. If the strains are given, then these strain-displacements provide a system of (6) nonlinear
partial dierential equation in terms of the unknown displacements (3).
3. "ik is the Green-Lagrange strain tensor.
4. The strains have been expressed in terms of the coordinates x; y; z in the undeformed state,
i.e. in the Lagrangian coordinate which is the preferred one in structural mechanics.
5. Alternatively we could have expressed ds0 2 , ds2 in terms of coordinates in the deformed
state, i.e. Eulerian coordinates x0 ; y0 ; z 0 , and the resulting strains are referred to as the
Almansi strain which is the preferred one in
uid mechanics.
6. These nonlinear equations are essential to describe structural behavior under large defor-
mation and for analysis of stability problems (Chapter 15).
7. In most cases the deformations are small enough for the quadratic term to be dropped,
the resulting equations reduce to
"xx = @u
Draft
@x (3.20)
"yy = @v
@y (3.21)
"zz = @w
@z (3.22)
xy = @x + @u
@v
@y (3.23)
xz = @w @u
@x + @z (3.24)
yz = @w +
@y @z
@v (3.25)
or
"ij = 12 (ui;k + uk;i) (3.26)
= Lu (3.28)
or 2 @ 3
8
> "xx 9
> @x 0 0
> >
0 @ 0
>
>
>
> "yy >
>
>
>
6
6 @y
78
7
u
9
0 @z@ 7< x
> > 6 7> >
<
"zz =
=
6
6
0 7 u
=
"xy @ @ 7> y
@x 0 (3.29)
> > 6 >
>
>
"xz
>
> 6 @y
@
7:
zu ;
>
>
>
>
>
>
>
>
6
4 @z 0 @x@ 7
5 | {z }
:
"yz ;
0 @z@ @y@ u
| {z } | {z }
L
14Starting with the set of forces acting on an innitesimal element of dimensions dx1 dx2 dx3 ,
Fig. 3.4 and writing the summation of forces, will yield
15 Alternatively, we may start from the momentum principle which states that the time rate of
change of the total momentum of a given set of particles equals the vector sum of all external
forces acting on the particles of the set, provided Newton's Third Law applies.
Z Z
d Z
td, + bd
= dt vd
(3.31)
,
LT + b = 0 (3.35)
or 8 9
>
> xx >
>
2 @ 00 @y@ @z@ 0
3>>
>
> yy >
>
>
>
8
bx
9
@x >
<
zz >
= >
< >
=
6
4 0 @y@ 0 @x@ 0 @z@ 7
5
xy + by =0
0 0 @z@ 0 @x@ @y@
>
> >
> >
: b >
; (3.36)
>
>
>
> xz
>
>
>
>
z
| {z }> > | {z }
: ;
LT yz b
| {z }
18 Expanding
@xx + @xy + @xz + bx = 0
@x @y @z
@yx + @yy + @yz + by = 0 (3.37)
@x @y @z
@zx + @zy + @zz + bz = 0
@x @y @z
20The (fourth order) tensor of elastic constants Dijklhas 81(34 ) components however, due to
the symmetry of both and , there are at most 36 9(92,1) distinct elastic terms.
Victor Saouma Finite Elements
3.4 Stress-Strain Relations in Elasticity 3{9
21 For the purpose of writing Hooke's Law, the double indexed system is often replaced by a
Draft
simple indexed system with a range of six:
62 =36
z }| {
k = Dkm "m k; m = 1; 2; 3; 4; 5; 6 (3.39)
22 For isotropic bodies (elastic properties independent of reference system used to describe it),
it can be shown that the number of independent elastic constants is two. The stress strain
relations can be written in terms of E and as:
E ij , E ij kk
"ij = 1+
ij = E "ij + ij "kk
(3.40)
1+ 1,2
where ij is the kroneker delta and is equal to 1 if i = j and to 0 if i 6= j . When the strain
equation is expanded in 3D cartesian coordinates it would yield:
"xy = E1 [xx , (yy + zz )]
"yy = E1 [yy , (zz + xx )]
"zz = E1 [zz , (xx + yy )] (3.41)
"xy = 1+E xy
"yz = 1+E yz
"zx = 1+E zx
23 Alternatively, we can derive expressions for the stress in term of the strain from the preceding
set of equations. This would yield:
8
> xx
9
>
2
1 1, 1, 0 0 0 38 > "xx
9
>
> > 6 1 0 0 0 > >
>
>
>
> yy
>
>
>
> 6 1 ,
1 ,
7>
7 >
>
> " yy
>
>
>
>
>
<
zz = E (1 , ) 6 1, 1,
>
= 6
6 1 0 0 0 7
7
>
<
" zz
>
=
(1 + )(1 , 2 ) 6 0 0 0 1 ,2 0 0
7
7 " (3.42)
>
> xy >
> 6 2(1 , nu ) 7 >
> xy >
>
>
yz >
> 1,2 >
"yz >
>
4 0 0 0 0 0 75 >
>
> > 6 > >
>
>
:
>
>
; 2(1,nu) >
>
:
>
>
;
zx 0 0 0 0 0 1
2(1
,
,
2
nu
)
" zx
| {z } | {z } | {z }
D
where is the shear modulus and
xy = 2"xy .
24 In terms of Lam e constant we would have
ij = "iiij + 2"ij (3.43)
"ij = 21 ij , ij 3+2
kk
where
= E
(1+ )(1,2 ) (3.44)
= E
2(1+ )
Victor Saouma Finite Elements
3{10 REVIEW OF ELASTICITY
25 In terms of initial stresses and strains
Draft
3.4.2 Axisymmetry
27 In solids of revolution, we can use a polar coordinate sytem and
"rr = @u
@r (3.47-a)
u
" = r (3.47-b)
"zz = @w
@z (3.47-c)
@u
"rz = @z + @w
@r (3.47-d)
= (1 + )(1 , 2 ) 6
6 1, 0 7
7
" (3.48)
>
>
>
:
rz
>
>
>
;
6
4 1, 0 7>
5>>
:
rz
>
>
>
;
0 0 0 1,2
| {z
2 }
D
3.4.3 Plane Stress
29If the longitudinal dimension in z direction is much smaller than in the x and y directions,
then yz = xz = zz = 0 throughout the thickness. Again, substituting into Eq. 3.42 we
i.e simply removing the hydrostatic pressure component from the total stress.
32 In the absence of thermal/initial stresses and pore pressures, the eld equations are written
as 0
ij;j + bi = 0 Equilibrium (3.52)
0
ij nj , t^i = 0 Natural B.C.
where bi and t^i are the body forces and surface tractions respectively.
33 These equations will form the basis of the variational formulation of the nite element
method.
34 To account for the eect of thermal/initial stresses and pore pressures we seek to modify Eq.
3.52, in such a way that bi and t^i are replaced by b0i and t^0i .
35 Thermal strains are caused by a change in temperature with respect to the stress-free con-
dition.
36 In Thermo- or poro-elasticity problems, the thermal strains and pore pressures are treated
as initial strains and stresses, respectively.
Victor Saouma Finite Elements
3{12 REVIEW OF ELASTICITY
37 In porous media, the total stress is equal to the sum of 0 and p. This last term is the
Draft
pore pressures and acts only in the voids of the material, and the eective stresses 0 act only
on the skeleton of the material.
38 It must be noted that the pore pressures p being considered in this discussion and throughout
the remainder of this course are the steady state pore pressures; excess pore pressures resulting
from dilatant behavior in the skeleton of the material are not considered.
39 Step by step generalization of the eld equations:
Clearly, in the absence of initial stresses ij = ij0 and in the absence of both initial strains and
stresses ij = ij0 .
41 The general stress-strain relationship obtained by substituting the constutive law into the
eective stress principle is
ij = Dijkl ("kl , "0kl ) + ij0 (3.62)
42 When thermal strains and pore pressures are considered in combination the constitutive law
is dened as a simple combination of Equations ?? and ??
43 The equilibrium equation and natural boundary conditions, respectively, can be rewritten in
terms of the eective stresses ij0
0
: ij;j + b0i = 0 Equilibrium (3.64)
,t : ij nj , t^0i = 0 Natural B.C.
0
where
b0i = bi , p;i , Dijkl T;i kl (3.65)
t^0i = t^i + p ni + T Dijkl ni kl
3.7 y Basic Equations of Anisotropic Elasticity
44 An alternate form of Eq. 3.38 is
"i = aij j (3.66)
where the indecis i and j go from 1 to 6. In the most general case this would yield 36
independent
constants aij , [?], however by virtue of symmetry (aij = aji ) this reduces to 21 (7)(6)
2 . If the
the material has one plane of elastic symmetry, then there would 13 independent constants;
If it has three mutually orthogonal planes of elastic symmetry, then we would say that it is
The total number of coecients for dierent materials is summarized in Table 3.1.
however, of the 12 elastic constants, only 9 are independent because the following relations
E1 21 = E2 12 (3.70-a)
E2 32 = E3 23 (3.70-b)
E3 13 = E1 31 (3.70-c)
Note that the preceding equations are written for the principal directions of elasticity, x, y,
and z in terms of the principal elastic constants (as opposed to constants in equations for
an arbitrary system of coordinates).
Whereas very few natural or man-made materials are truly orthotropic (certain crystals as
topaz are), a number of others are transversely isotropic. Transversely isotropic have through
every point a plane in which all directions are equivalent with respect to elastic properties (such
as in laminates, shist, quartz, roller compacted concrete, etc...). For transversely isotropic solids
in 3D, we have
"x = a11 x + a12 y + a13 z
"y = a12 x + a11 y + a13 z
"z = a13 (x + y ) + a33 z (3.71)
xy = 2(a11 , a12 )xy
yz = a44 xy
xz = a44 xz
and 0
a11 = E1 ; a12 = , E ; a13 = , E 0 ; a33 = , E10 ; a44 = , 10 (3.72)
3.10 Summary
47 Fig. 3.7 illustrates the fundamental equations in solid mechanics.
Essential B.C.
ui : ,u
Body Forces
?
Displacements
bi ui
?
Equilibrium
?
Kinematics
LT + b = 0 = Lu
? ?
Stresses Constitutive Rel. Strain
ij
- ij = Dijkl "kl
"ij
6
Nonessential B.C.
ti : ,t
Chapter 4
SCALAR FIELD PROBLEMS
4.1 Introduction
1 Scalar eld problems are encountered in almost all branches of engineering and physics. Most
of them can be viewed as special forms of the general Helmholtz equation given by
@ k @ + @ k @ + @ k @ + Q = c @ (4.1)
@x x @x @y y @y @z z @z @t
where (x; y; z ) is the eld variable to be solved.
2 Table 4.1 illustrates selected examples of the diusion equation.
div(Dr) + Q = 0
Equation D Q q Constitutive law
Heat Flow Temperature Thermal Heat Heat Fourrier
T conductivity supply/sink
ux q = ,DrT
Fluid
ow through Piezometric Permeability Fluid Volume Darcy
porous media head h Coecients supply
ux q = ,Dr
Diusion ion Permeability Ion Ion Fick
concentration coecients supply
ux q = ,Dr
Saint-Venant Prandtl's 1 Rate of Hooke
Torsion stress function twist 2
Conduction: takes place when a temperature gradient exists within a material and is governed
by Fourier's Law, Fig. 4.1 on ,q :
qx = ,kx @T
@x (4.2)
@T
qy = ,ky @y (4.3)
where T = T (x; y) is the temperature eld in the medium, qx and qy are the componenets
of the heat
ux (W/m2 or Btu/h-ft2 ), k is the thermal conductivity (W/m.o C or Btu/h-
ft-o F) and @T @T
@x , @y are the temperature gradients along the x and y respectively. Note
that heat
ows from \hot" to \cool" zones, hence the negative sign.
Convection: heat transfer takes place when a material is exposed to a moving
uid which is
at dierent temperature. It is governed by the Newton's Law of Cooling
q = h(T , T1) on ,c (4.4)
where q is the convective heat
ux, h is the convection heat transfer coecient or lm
coecient (W/m2 .o C or Btu/h-ft2 .o F). It depends on various factors, such as whether
convection is natural or forced, laminar or turbulent
ow, type of
uid, and geometry of
the body; T and T1 are the surface and
uid temperature, respectively. This mode is
considered as part of the boundary condition.
Radiation: is the energy transferred between two separated bodies at dierent temperatures
by means of electromagnetic waves. The fundamental law is the Stefan-Boltman's Law of
Thermal Radiation for black bodies in which the
ux is proportional to the fourth power
of the absolute temperature., which causes the problem to be nonlinear. This mode will
not be covered.
4 Consider a solid through which there is a
ow q of some quantity, mass, heat, chemical, etc...
5 The rate of transfer per unit area is q
qx qx + -@q@xx dx
6
- Q dy
?
6
qy
-
dx
,
vnd, =
div vd
(4.14)
12Finally, we dene the specic heat of a solid c as the amount of heat required to raise a unit
mass by one degree. Thus if is a temperature change which occurs in a mass m over a time
t, then the corresponding amount of heat that was added must have been cm, or
Z
I3 = cd
(4.17)
4.3 Summary
14 The fundamental equations of heat
ow are illustrated in Fig. 4.4
and the analogy between scalar and vector problems is shown in Table 4.2
? ?
Temperature
Balance
r:q = Q T
? ?
Flux Vector Constitutive Rel. Gradient
- q = ,DrT
q rT
6
Natural B.C.
qn : ,q [ ,c
Scalar Vector
Conduction Solid Mechanics
T u State variable
q Flux vector
qn = qT n t = n Flux at boundary
div q + Q = 0 L +b =0
T Balance/Equilibrium
rT = Lu \kinematic"
q = ,DrT = D Constitutive Law (Fourrier/Hooke)
T on ,T u on ,u Essential BC
qn on ,q [ ,c t on ,t Natural BC
Chapter 5
DIRECT STIFFNESS METHOD
5.1 Introduction
1 The direct stiness method, covered in Advanced Structural Analysis is brie
y reviewed in
this lecture.
2 A slightly dierent algorithm will be used for the assembly of the global stiness matrix.
3 Some of the prescribed steps are further discussed in the next sections.
1. At input stage read ID(idof,inod) of each degree of freedom for every node such that:
(
ID(idof; inod) = 01 if unrestrained d.o.f.
if restrained d.o.f. (5.1)
2. After all the node boundary conditions have been read, assign incrementally equation
numbers
(a) First to all the active dof
(b) Then to the other (restrained) dof.
(c) Multiply by -1 all the passive dof.
Note that the total number of dof will be equal to the number of nodes times the number
of dof/node NEQA.
3. The largest positive global degree of freedom number will be equal to NEQ (Number Of
Equations), which is the size of the square matrix which will have to be decomposed.
6 For example, for the frame shown in Fig. 5.1:
1. The input data le may contain:
5.3 LM Vector
7 The LM vector of a given element gives the global degree of freedom of each one of the element
degree of freedom's. For the structure shown in Fig. 5.1, we would have:
bLMc = b ,10 ,11 4 5 6 7 c element 1 (2 ! 3)
bLMc = b 5 6 7 1 2 3 c element 2 (3 ! 1)
bLMc = b 1 2 3 ,12 8 9 c element 3 (1 ! 4)
5.4 Assembly of Global Stiness Matrix
8 As for the element stiness matrix, the global stiness matrix [K] is such that Kij is the force
in degree of freedom i caused by a unit displacement at degree of freedom j .
9 Whereas this relationship was derived from basic analysis at the element level, at the structure
level, this term can be obtained from the contribution of the element stiness matrices [Ke ]
(written in global coordinate system).
10 For each Kij term, we shall add the contribution of all the elements which can connect degree
of freedom i to degree of freedom j , assuming that those forces are readily available from the
individual element stiness matrices written in global coordinate system.
11 Kij is non-zero if and only if degree of freedom i and degree of freedom j are connected by
an element or share a node.
12 There are usually more than one element connected to a dof. Hence, individual element
stiness matrices terms must be added up.
Victor Saouma Finite Elements
5{4 DIRECT STIFFNESS METHOD
13 Because each term of all the element stiness matrices must nd its position inside the global
Draft
stiness matrix [K], it is found computationally most eective to initialize the global stiness
matrix [KS ]NEQANEQA) to zero, and then loop through all the elements, and then through
each entry of the respective element stiness matrix Kije .
14 The assignment of the element stiness matrix term Kij e (note that e, i, and j are all known
since we are looping on e from 1 to the number of elements, and then looping on the rows and
columns of the element stiness matrix i; j ) into the global stiness matrix KklS is made through
the LM vector (note that it is k and l which must be determined).
15 Since the global stiness matrix is also symmetric, we would need to only assemble one side
of it, usually the upper one.
16 Contrarily to Matrix Structural Analysis, we will assemble the full augmented stiness matrix.
2 3 8 9
1 3 11 17 24 32 41 68 >
>
>
1 >
>
>
6
6 2 5 10 16 23 31 40 67 7
7 >
>
>
> 2 >
>
>
>
6 7 > >
6
6
4 9 15 22 30 39 66 7
7
>
>
>
>
4 >
>
>
>
6
6 6 8 14 21 29
48 5638 65 7
7
>
>
>
>
>
6 >
>
>
>
>
6
6 7 13 20 28
47 5537 64 7
7
>
>
>
>
7 >
>
>
>
K = 66
6
6 12 19 27
46 5436 63 7
7
7 MAXA =
<
12 =
6
18 26 35
45 53 62 77 >
>
>
18 >
>
>
6
6 25
44 5234 61 7
7
>
>
>
>
>
25 >
>
>
>
>
6
6 33
43 51 60 7 7 >
>
>
>
33 >
>
>
>
6
6
6
42 50 59 7
7
7
>
>
>
>
>
42 >
>
>
>
>
4 49 58 5 >
>
>
>
49 >
>
>
>
57 :
57 ;
Thus, to locate an element within the stiness matrix, we use the following formula:
30For internal book-keeping purpose, since we are assembling the augmented stiness matrix,
we proceed in two stages:
1. First number all the unrestrained degrees of freedom, i.e. those on ,t .
2. Then number all the degrees of freedom with known displacements, on ,u , and multiply
by -1.
31 Considering a simple beam, Fig. 5.5 the full stiness matrix is equal to
2
v1 1 v2 2 3
V1 12EI=L 3 6EI=L2 ,12EI=L 6EI=L2
3
[K] = M 16 6EI=L
6 2 4EI=L ,6EI=L2 2EI=L 777 (5.9)
V2 4 ,12EI=L3 ,6EI=L2
6
12EI=L3 ,6EI=L2 5
M2 6EI=L2 2EI=L ,6EI=L2 4EI=L
This matrix is singular, it has a rank 2 and order 4 (as it embodies also 2 rigid body motions).
32 We shall consider 3 dierent cases, Fig. 5.6
4. Ktt is inverted
2 3
L3 =3EI ,L=6EI 6EI=L2 ,6EI=L2
6 7
6
6 ,L=6EI L=3EI 6EI=L2 ,6EI=L2 7
7
4 6EI=L 2 6EI=L 2 12EI=L3 ,12EI=L3 5
,6EI=L ,6EI=L ,12EI=L3 12EI=L3
2 2
Solution:
1. Determine the structure ID matrix and the LM vector for each element Initial ID matrix
Node" 1 2 3 4 5 #
ID = 01 0 1 0 0
0 1 0 0
Final ID matrix
Node" 1 2 3 4 5#
ID = 1 2 ,9 4 6
,8 3 ,10 5 7
LM vectors for each element
2 3
Element 1 1 ,8 4 5
6 1
Element 26 ,8 2 3 777
Element 36
6 2
6
3 4 5 77
6 4
Element 46 5 6 7 77
6 ,9
Element 56 ,10 4 5 777
6 2
Element 66 3 6 7 77
6
Element 74 2 3 ,9 ,10 5
Element 8 ,9 ,10 6 7
Victor Saouma Finite Elements
5{16 DIRECT STIFFNESS METHOD
2. Derive the element stiness matrix in global coordinates
Draft
2 3
c2 cs ,c2 ,cs
[K ] = EA 6 cs
6
s2 ,cs ,s2 777
L 64 ,c2 ,cs c2 cs 5
,cs ,s2 cs s2
where c = cos = x2 ,L x1 ; s = sin = Y2 ,L Y1
Element 1 L = 200 , c = 1620,0 = 0:8, s = 1220,0 = 0:6,
EA
L = (30;000)(10)
20 = 15; 000
2
1 ,8 4 5 3
1 9600 7200 ,9600 ,7200
[K1 ] = , 8666 7200 5400 ,7200 ,5400 777
4 4 ,9600 ,7200 9600 7200 5
5 ,7200 ,5400 7200 5400
Element 2 L = 160 ; c = 1; s = 0; EA
L = 18; 750.
1
2
,8 2 33
1 18; 750 0 ,18; 750 0
[K2 ] = , 8666 0 0 0
2 4 ,18; 750 0 18; 750
0 777
05
3 0 0 0 0
Element 3 L = 120 ; c = 0; s = 1; EA
L = 25; 000
22
3 4 5 3
2 0 0 0 0
3 60
6
[K3 ] = 464 0 25 ; 000 0 ,25; 000 777
0 0 0 5
5 0 ,25; 000 0 25; 000
Element 4 L = 160 ; c = 1; s = 0; EA
L = 18; 750
2
4 5 6 73
4 18; 750 0 ,18; 750 0
[K4 ] = 56664 ,180; 750 0 0 0 777
6
0 18; 750 05
7 0 0 0 0
2
,9 ,10 4 5 3
,9 9600 ,7200 ,9600 7200
[K5 ] = , 10666 ,7200 5400 7200
4 4 ,9600 7200 9600
,5400 777
,7200 5
5 7200 ,5400 ,7200 5400
Element 6 L = 200 ; c = 0:8; s = 0:6; EA L = 15; 000
2
2 3 6 7 3
2 9600 7200 ,9600 ,7200
[K6 ] = 36664 ,7200
6
5400 ,7200 ,5400 777
9600 ,7200 9600 7200 5
7 ,7200 ,5400 7200 5400
Element 7 L = 160 ; c = 1; s = 0; EA
L = 18; 750
2
2 3 ,9 ,10 3
2 18; 750 0 ,18; 750 0
[K7 ] = 3,9 664 ,180; 750 0 0 0 777
6
0 18; 750 0 5
,10 0 0 0 0
Element 8 L = 120 ; c = 0; s = 1; EA
L = 25; 000
2
,9 ,10 6 7 3
,9 0 0 0 0
[K8 ] = ,1066 0 25; 000 0 ,25; 000 77
6 7
6 4 0 0 0 0 5
7 0 ,25; 000 0 25; 000
3. Assemble the augmented global stiness matrix in kips/ft.
1 2 3 4 5 6 7
12 9; 600 + 18; 750 ,18; 750 0 ,9; 600 ,7; 200 0 0 3
2 9; 600 + (2)18; 750 7; 200 0 0 ,9; 600 ,7; 200
36 5; 400 + 25; 000 0 ,25; 000 ,7; 200 ,5; 400 7
ktt = 46
4
18; 750 + (2)9; 600 7; 200 , 7; 200 ,18; 750 0 7
5
5 SYMMETRIC 25; 000 + 5; 400(2) 0 0
6 18; 750 + 9600 7200
7 25; 000 + 5; 400
,8 ,9 ,10 3
12 0 + 7; 200
2 0 ,18; 750 , 18; 750 0
36 0 0 0 7
ktu = 44 ,7; 200
6 ,9; 600 7; 200 7
5 ,5; 400 7; 200 ,5; 400 5
6 0
7 ,25; 000
Victor Saouma Finite Elements
5{18 DIRECT STIFFNESS METHOD
,8 ,9 ,10
Draft
,8 0 + 5; 400
kuu = ,9 9; 600 + 18; 750 0 , 7; 200 + 0
,10 ,7; 200 + 0 0 + 5; 400 + 25; 000
4. convert to kips/in and simplify
8 0 9 2 2; 362:5 ,1; 562:5 0 ,800 ,600 0 0 38 u1 9
>
> 0 >
> 3; 925:0 600 0 0 ,800 ,600 >
> u2 >
>
< 0 = 6 2; 533:33 0 ,2; 083:33 ,600 ,450 7< v3 =
0 =6 3; 162:5 0 ,1; 562:5 0 7 u4
>
> , 100 >
>
4 Symmetric 2; 983:33 0 0 5>
> v5 >
>
: 50 ; 2; 362:5 600 : u6 ;
0 2; 533:33 v7
v2 >
>
>
u2 >
>
>
:
v2 ;
( )
n o1 h i ,0:0223 n o
v1 = 15;000 0:8 0:6 ,0:8 ,0:6 0 = 52:1 Compression
v2 12 ,0:8 ,0:6 0:8 0:6 ,0:0102 ,52:1
( ),0:0856
n o2 h i ,0:0233 n o
v1 = 18;750 1 0 ,1 0 0 = ,43:2 Tension
v2 12 ,1 0 1 0 0:00433 43:2
" , 0 : 116 #
n o3 h i 0:00433 n o
v1 = 25;000 0 1 0 ,1 ,0:116 = ,63:3 Tension
v2 12 0 ,1 0 1 ,0:0102 63:3
( ,0:0856 )
n o4 h i ,0:0102 n o
v1 = 18;750 1 0 ,1 0 ,0:0856 = ,1:58 Tension
v2 12 ,1 0 1 0 ,0:00919 1:58
n o5 h ,0: i n o n o
v1 = 15;000 ,0:8 0:6 0:8 ,0:6 ,0:0102 = 54:0
v2 12 0:8 ,0:6 ,0:8 0:6 ,0:0856 ,54:0 Compression
n o6 h i ,0:116 n o
v1 = 15;000 0:8 0:6 ,0:8 ,0:6 ,0:00919 = ,60:43 K Tension
v2 12 ,0:8 ,0:6 0:8 0:6 ,0:0174 60:43 K
n o7 h i n o
v1 18;750 1 0 ,1 0 ,0:116 6:72
v2 = 12 ,1 0 1 0 0 = ,6:72 Compression
( 0 )
n o8 h i 0 n o
v1 = 25;000 0 1 0 ,1 0 = 36:3 Compression
v2 12 0 ,1 0 1 ,0:00919 ,36:3
,0:0174
8 9
>
>
>
1 >
>
>
>
>
>
>
>
2 >
>
>
>
>
>
>
< 4 >
>
=
MAXA = > 6 >
>
>
>
>
10 >
>
>
>
> >
>
>
>
>
15 >
>
>
>
:
20 ;
Victor SaoumaFigure 5.7: Flowchart for the Skyline Height DeterminationFinite Elements
5{22 DIRECT STIFFNESS METHOD
34 Once the MAXA vector has been determine, then term K (i; j ) in the square matrix, would be
Draft
stored in KK(MAXA(j)+j-i) (assuming j > i) in the compacted form of fKg.
1. Initialize the vector storing the compacted augmented stiness matrix to zero.
2. Loop through each element, e, and for each element:
(a) Determine its stiness matrix [Ke ]
(b) Dene the LM array of the element
(c) Loop through each row and column of the element stiness matrix, and add the
contributions of the element to the structure's stiness matrix (note that we assemble
only the upper half).
5.7.4 Decomposition
35 Decompose the global stiness matrix Kt only. This is a subset of the augmented stiness
matrix. Since the matrix is both symmetric and positive denite, the matrix can be decomposed
using Cholesky's method into: [K] = [L][L]T . Should a division by zero occur, or an attempt
to extract the square root of a negative number happen, then this would be an indication
that either the global stiness matrix is not properly assembled, or that there are not enough
restraint to prevent rigid body translation or rotation of the structure.
5.7.5 Load
36 Once the stiness matrix has been decomposed, than the main program should loop through
each load case.
1. Initialize augmented the load vector (of length NEQA) to zero.
2. Read number of loaded nodes. For each loaded node store the non-zero values inside the
load vector (using the ID matrix for determining storage location).
3. Loop on all loaded elements:
(a) Read element number, and load value
(b) Compute the nodal equivalent load
(c) Using the LM vector, add them to the nodal load vector.
37 Backsubstitution is achieved by multiplying the decomposed stiness matrix with the load
vector. The resulting vector stores the nodal displacements, in global coordinate system, cor-
responding to the unrestrained degree of freedom.
5.7.7 Reactions
38 Reactions are determined from
xxxx (5.11)
Chapter 6
VARIATIONAL AND ENERGY
METHODS
6.1 Variational Calculus; Preliminaries
From Pilkey and Wunderlich book
u(x)
C
B
u(x)
du
A dx
x
x=a x=c x=b
6.1.1.2 Derivation
4 We seek the function u(x) which extremizes .
5 Letting u
~ to be a family of neighbouring paths of the extremizing function u(x) and we assume
that at the end points x = a; b they coincide. We dene u~ as the sum of the extremizing path
and some arbitrary variation, Fig. 6.1.
u~(x; ") = u(x) + "(x) = u(x) + u(x) (6.3)
where " is a small parameter, and u(x) is the variation of u(x)
u = u~(x; ") , u(x) = "(x) (6.4-a)
= u~ , u (6.4-b)
and (x) is twice dierentiable, has undened amplitude, and (a) = (b) = 0. We note that
u~ coincides with u if " = 0
6 It can be shown that the variation and derivation operators are commutative
)
d
dx (u)0= u~0 (x; ") , u0 (x) d
du
7 Furthermore, the variational operator and the dierential calculus operator d can be simi-
larly used, i.e.
(u0 )2 = 2u0 u0 (6.6-a)
Victor Saouma Finite Elements
6.1 Variational Calculus; Preliminaries 6{3
(u + v) = u + v (6.6-b)
Draft
Z Z
udx = (u)dx (6.6-c)
u = @u @u
@x x + @y y (6.6-d)
however, they have clearly dierent meanings. du is associated with a neighbouring point at
a distance dx, however u is a small arbitrary change in u for a given x (there is no associated
x).
8 For boundaries where u is specied, its variation must be zero, and it is arbitrary elsewhere.
The variation u is said to undergo a virtual change.
9 To solve the variational problem of extremizing , we consider
Z b
(u + ") = (") = F (x; u + "; u0 + "0 )dx (6.7)
a
12 The fundamental lemma of the calculus of variation states that for continuous (x) in a
x b, and with arbitrary continuous function (x) which vanishes at a and b, then
Z b
(x) (x)dx = 0 , (x) = 0 (6.13)
a
13 Thus,
@F , d @F = 0 (6.14)
@u dx @u0
14 This dierential equation is called the Euler equation associated with and is a necessary
condition for u(x) to extremize .
15 Generalizing for a functional which depends on two eld variables, u = u(x; y ) and v =
v(x; y) Z Z
= F (x; y; u; v; u;x ; u;y ; v;x; v;y ; ; v;yy )dxdy (6.15)
There would be as many Euler equations as dependent eld variables
8
< @F , @ @F , @ @F + @ 22 @F + @ 2 @F + @ 22 @F = 0
@u @x @u;x @y @u;y @x2 @u;xx @x@y @u;xy @y @u;yy (6.16)
: @F , @ @F , @ @F + @ 2 @F + @ 2 @F + @ 22 @F = 0
@v @x @v;x @y @v;y @x @v;xx @x@y @v;xy @y @v;yy
16 We note that the Functional and the corresonpding Euler Equations, Eq. 6.1 and 6.14, or
Eq. 6.15 and 6.16 describe the same problem.
17 The Euler equations usually correspond to the governing dierential equation and are referred
to as the strong form (or classical form).
18 The functional is referred to as the weak form (or generalized solution). This classication
stems from the fact that equilibrium is enforced in an average sense over the body (and the
eld variable is dierentiated m times in the weak form, and 2m times in the strong form).
19 It can be shown that in the principle of virtual displacements, the Euler equations are the
equilibrium equations, whereas in the principle of virtual forces, they are the compatibility
equations.
20 Euler equations are dierential equations which can not always be solved by exact meth-
ods. An alternative method consists in bypassing the Euler equations and go directly to the
variational statement of the problem to the solution of the Euler equations.
21 Finite Element formulation are based on the weak form, whereas the formulation of Finite
Dierences are based on the strong form.
22 We still have to dene . The rst variation of a functional expression is
)
F = @F u + @F0 u0 Z b @F @F
= 0
@u u + @u0 u dx
R@u
b @u (6.17)
= a Fdx a
23 We have just shown that nding the stationary value of by setting = 0 is equivalent to
nding the extremal value of by setting d(d"") "=0 equal to zero.
24 Similarly, it can be shown that as with second derivatives in calculus, the second variation
2 can be used to characterize the extremum as either a minimum or maximum.
6.1.2 Boundary Conditions
25 Revisiting the integration by parts of the second term in Eq. 6.10, we had
Z b 0 @F
dx = @F b , Z b d @F dx (6.19)
a @u0 @u0 a a dx @u0
We note that
1. Derivation of the Euler equation required (a) = (b) = 0, thus this equation is a state-
ment of the essential (or forced) boundary conditions, where u(a) = u(b) = 0.
@F0 = 0 at x = a and b.
2. If we left arbitrary, then it would have been necessary to use @u
These are the natural boundary conditions.
26 For a problem with, one eld variable, in which the highest derivative in the governing
dierential equation is of order 2m (or simply m in the corresponding functional), then we have
Essential (or Forced, or geometric) boundary conditions, involve derivatives of order zero
(the eld variable itself) through m-1. Trial displacement functions are explicitely required
to satisfy this B.C. Mathematically, this corresponds to Dirichlet boundary-value problems.
Nonessential (or Natural, or static) boundary conditions, involve derivatives of order m
and up. This B.C. is implied by the satisfaction of the variational statement but not
explicitly stated in the functional itself. Mathematically, this corresponds to Neuman
boundary-value problems.
27 Table 6.1 illustrates the boundary conditions associated with some problems
The time-rate of change of the total energy (i.e., sum of the kinetic energy and the
internal energy) is equal to the sum of the rate of work done by the external forces
and the change of heat content per unit time:
d
dt (K + U ) = We + H (6.21)
where K is the kinetic energy, U the internal strain energy, W the external work, and H the
heat input to the system.
32 For an adiabatic system (no heat exchange) and if loads are applied in a quasi static manner
(no kinetic energy), the above relation simplies to:
We = U (6.22)
* *
U0 U0
A A
U0 U0
A A
Nonlinear Linear
U = U0 d
(6.25-b)
36To obtain a general form of the internal strain energy, we start with the general stress-strain
equation ??
= D( , 0 ) + 0 (6.26)
where D is the material stiness matrix; is the strain vector due to the displacements u; 0
is the initial strain vector; 0 is the initial stress vector; and is the stress vector.
Victor Saouma Finite Elements
6{8 VARIATIONAL AND ENERGY METHODS
37 The initial strains and stresses are the result of conditions such as heating or cooling of a
Draft
system or the presence of pore pressures in a system.
38 The strain energy U for a linear elastic system will then be dened as
Z Z Z
1
U = 2 Dd
, D0 d
+ T 0 d
T T (6.27)
where
is the volume of the system.
6.2.3 Potential of External Work
39 The Potential of external work W in an arbitrary system is dened as
Z Z
def
We = uT bd
+ uT ^td, + uP (6.28)
,t
where u are the displacements, b is the body force vector; ^t is the applied surface traction
vector; ,t is that portion of the boundary where ^t is applied, and P are the applied nodal
forces.
40 Note that the potential of the external work is dierent from the external work itself (usually
by a factor of 1/2)
6.2.4 Virtual Work
41We dene the virtual work done by the load on a body during a small, admissible (continuous
and satisfying the boundary conditions) change in displacements.
Z
Internal Virtual Work Wi = , d
(6.29)
Z
Z
External Virtual Work We = ^tud, + bud
(6.30)
,t
k= 500 lbf/in
43 Note that in the potential the full load is always acting, and through the displacements of
its points of application it does work but loses an equivalent amount of potential, this explains
the negative sign.
44 The theorem of stationary potential energy can e derived from the principle of virtual work
(see Matrix notes), and states
Of all kinematically admissible deformations (displacements satisfying the given
boundary conditions), the actual deformations (those which correspond to stresses
which satisfy equilibrium) are the ones for which the total potential energy assumes
a stationalry value.
or
@ = 0 (6.33)
45As an illustrative example (adapted from Willam, 1987), let us consider the single dof system
shown in Fig. 6.3. The strain energy U and external work W are given by
U = 12 u(Ku) = 250u2 (6.34-a)
We = mgu = 100u (6.34-b)
Thus the total potential energy is given by
= 250u2 , 100u (6.35)
20.0
Energy [lbfin]
0.0
External Work
Strain Energy 20.0
Total Potential Energy
47 It can be shown that the minimum potential energy yields a lower bound prediction of
displacements.
= UZ , We
def Z
(6.39-a)
We = def
uT bd
+ , uT ^td, + uP (6.39-b)
Z
t Z Z
U = 12 T Dd
, T D0d
+ T 0d
(6.39-c)
the functional for the general form of the potential energy variational principle is obtained
Z Z Z Z Z
= 12 T Dd
, T D0 d
+ T 0 d
, uT bd
, uT ^td, , uP (6.40)
,t
49 A variational statement is obtained by taking the rst variation of the variational principle
and setting this scalar quantity equal to zero.
50 The variational statement for the general form of the potential energy functional (i.e. Equa-
tion 6.40) is
Z Z Z Z Z
= T Dd
, T D 0 d
+ T 0 d
, uT bd
, uT ^td, = 0 (6.41)
,t
which is the Principle of Virtual Work.
51 We rewrite the strain-displacement relations in terms of a linear dierential operator L
= Lu (6.42)
where L is a linear dierential operator and u is the displacement vector.
52 Since the dierential operator L is linear, the variation of the strains can be expressed in
terms of the variation of the displacements u
= (Lu) = Lu (6.43)
Z
I,
Z
(Lu)T D0 d
= u G(D0 )d, , uT LT (D0 )d
T (6.47)
Z I, Z
(Lu)T 0 d
= uT G0 d, , uT LT 0 d
where G is a transformation matrix containing the direction cosines for a unit normal vector
such that the surface tractions t are dened as t = G and the surface integrals are over the
entire surface of the body @
.
56 Substituting Equation 6.47 into Equation 6.44, and using Eq. 6.26, the variational statement
becomes Z
= , uT fLT [D( , 0 ) + 0 ] + bgd
Z
+ uT fG[D( , 0 ) + 0 ] , ^tgd, = 0 (6.48)
,
57 Since u is arbitrary the expressions in the integrands within the braces must both be equal
to zero for to be equal to zero. Recognizing that the stress-strain relationship (i.e. Equation
6.26) appears in both the volume and surface integrals, the Euler equations are
LT + b = 0 on
(6.49)
G , ^t = 0 on ,t
where the rst Euler equation is the equilibrium equation and the second Euler equation denes
the natural boundary conditions. The natural boundary conditions are dened on ,t rather
1. Be continous.
2. Must be admissible, i.e. satisfy the essential boundary conditions (the natural boundary
conditions are included already in the variational statement. However, if also satisfy
them, then better results are achieved).
3. Must be independent and complete (which means that the exact displacement and their
derivatives that appear in can be arbitrary matched if enough terms are used. Further-
more, lowest order terms must also beincluded).
In general is a polynomial or trigonometric function.
1. cji can either be a set of coecients with no physical meanings, or variables associated
with nodal generalized displacements (such as de
ection or displacement).
2. If the coordinate functions satisfy the above requirements, then the solution converges
to the exact one if n increases.
3. For increasing values of n, the previously computed coecients remain unchanged.
4. Sine the strains are computed from the approximate displacements, strains and stresses
are generally less accurate than the displacements.
5. The equilibrium equations of the problem are satised only in the energy sense = 0
and not in the dierential equation sense (i.e. in the weak form but not in the strong one).
Therefore the displacements obtained from the approximation generaly do not satisfy the
equations of equilibrium.
6. Since the continuous system is approximated by a nite number of coordinates (or d.o.f.),
then the approximate system is stier than the actual one, and the displacements obtained
from the Ritz method converge to the exact ones from below.
Chapter 7
INTERPOLATION FUNCTIONS
7.1 Introduction
1 Application of the principle of total potential energy (or virtual displacement) requires an
assumed displacement eld. This displacement eld can be approximated by interpolation
functions written in terms of:
1. Unknown polynomial coecients most appropriate for continuous systems, and the
Rayleigh-Ritz method
y = a1 + a2x + a3 x2 + a4 x3 (7.1)
A major drawback of this approach, is that the coecients have no physical meaning.
2. Unknown nodal deformations most appropriate for discrete systems and Potential En-
ergy based formulations
y = u = N1 u1 + N2 u2 + : : : + Nnun (7.2)
2 For simple problems both Eqn. 7.1 and Eqn. 7.2 can readily provide the exact solutions of the
d4 y = q for
exure), but for more complex ones, one
governing dierential equation (such as dx 4 EI
must use an approximate one.
7.3 C 0
7.3.2 Generalization
9 The previous derivation can be generalized by writing:
( )
u = a1 x + a2 = b x 1 c aa1 (7.14)
| {z } 2
[p] | {z }
fag
where [p] corresponds to the polynomial approximation, and fag is the coecient vector.
10 We next apply the boundary conditions:
( ) " #( )
u1 = L0 11 a1 (7.15)
u2 a2
| {z } | {z }| {z }
fug [L] fag
following inversion of [L], this leads to
( ) " #( )
a1
a2 = L1 ,L1 10 u1
u2 (7.16)
| {z } | {z }| {z }
fag [L],1 fug
Victor Saouma Finite Elements
7{4 INTERPOLATION FUNCTIONS
11 Substituting this last equation into Eq. 7.14, we obtain:
Draft
( )
u=b
|
(1 , Lx ) Lx
{z
c} uu12 (7.17)
[|p][{zL],1} | {z
fu g
}
[N]
13Note that in some cases L,1 is not always possible to obtain, and that in others there may
be considerable algebraic diculties for arbitrary geometries. Hence, we shall introduce later
on Lagrangian and Hermitian interpolation functions.
7.3.3 Constant Strain Triangle Element
14 Next we consider a triangular element, Fig. 7.2 with bi-linear displacement eld (in both x
16 We then multiply the inverse of [L] in Eq. 7.22 by [p] and obtain:
u = N1u1 + N2 u2 + N3 u3 (7.23)
where
N1 = x 1y (x2 y3 , xy3 , x2 y + x3 y)
2 3
N2 = x 1y (xy3 , x3y) (7.24)
2 3
y
N3 = y
3
We observe that each of the three shape functions is equal to 1 at the corrsponding node, and
equal to 0 at the other two.
17 The same shape functions can be derived for v :
v = N1 v1 + N2 v2 + N3 v3 (7.25)
18 Hence, the displacement eld will be given by:
8 9
>
> u1 >
>
( ) "
>
>
#>
>
>
v1 >
>
>
>
>
u = N01 N0 N02 N0 N03 N0
<
u2 =
(7.26)
v 1 2 3 >
>
>
v2 >
>
>
>
>
>
>
:
u3 >
>
>
>
;
v3
19 The element is refereed to as Constant Strain Triangle (CST) because it has a linear dis-
placement eld, and hence a constant strain.
7.3.4 Further Generalization: Lagrangian Interpolation Functions
20 In our earlier approach, the shape functions were obtained by:
1. Assumption of a polynomial function: u = bpcfag
Victor Saouma Finite Elements
7{6 INTERPOLATION FUNCTIONS
2. Application of the boundary conditions fug = [L]fag
Draft
3. Inversion of [L]
4. And nally [N] = [p][L],1
21 By following these operations, we have in eect dened the Lagrangian Interpolation Func-
tions for problems with C 0 interelement continuity (i.e continuity of displacement only).
22 The Lagrangian interpolation denes the coecients ([N] in our case) of a polynomial series
representation of a function in terms of values dened at discrete points (nodes in our case).
For points along a line this would yield:
Qm+1
(x,x )
Ni = Qmj=1+1;j6=i(xi ,xjj ) (7.27)
j =1;j 6=i
28 Similarly
u43 = xx2,,xx u4 + xx1 ,,xx u3
2 1 1 2
a , x x
= 2a u4 + 2a u3+ a (7.31)
29Next, we interpolate in the y direction along 1-4 and 2-3 between u12 and u43 . Again, we
use Eq. 7.27 however this time we replace x by y:
u = yy2 ,,yy u12 + yy1 ,,yy u43 (7.32)
2 1 1 2
= b 2,b y a 2,a x u1 + b 2,b y x 2+a a u2 + y 2+b b a 2,a x u4 + y 2+b b x 2+a a u3
= (a , x4)(abb , y) u1 + (a + x4)(abb , y) u2 + (a + x4)(abb + y) u3 + (a , x4)(abb + y) u4
| {z } | {z } | {z } | {z }
N1 N2 N3 N1
30 One can easily check that at each node i the corresponding Ni is equal to 1, and all others
to zero, and that at any point N1 + N2 + N3 + N4 = 1. Hence, the displacement eld will be
where
Ni = (a x)(b8abc
y)(c z) (7.35)
7.4 C 1
7.4.1 Flexural
32 With reference to Fig. 7.5. We have 4 d.o.f.'s, fug41 : and hence will need 4 shape functions,
N1 to N4 , and those will be obtained through 4 boundary conditions. Therefore we need to
assume a polynomial approximation for displacements of degree 3.
v = a1 x3 + a2 x2 + a3 x + a4 (7.36)
dv = 3a x2 + 2a x + a
= dx (7.37)
1 2 3
where = xl .
N1 = (1 + 2 3 , 3 2 )
N2 = x(1 , )2
N3 = (3 2 , 2 3 )
37 Hence, the shape functions for the
exural element are given by: N4 = x(2 , ) (7.43)and
are shown in Fig 7.6.
38 Table 7.1 illustrates the characteristics of those shape functions
0.8
N1
0.6
N3
N2
N4
0.4
N
0.2
0.0
0.2
0.0 0.2 0.4 0.6 0.8 1.0
(x/L)
=0 =1
Function Ni Ni;x Ni Ni;x
N1 = (1 + 23 , 32 ) 1 0 0 0
N2 = x(1 , )2 0 1 0 0
N3 = (32 , 2 3 ) 0 0 1 0
N4 = x( , )
2 0 0 0 1
39 For problems involving the rst derivative of the shape function, that is with C 1 interelement
continuity (i.e continuity of rst derivative or slope) such as for
exure, Hermitian interpolation
functions rather than Lagrangian ones should be used.
40 Hermitian interpolation functions are piecewise cubic functions which satisfy the conditions
of displacement and slope (C 0 , C 1 ) continuities. They are exensively used in CAD as Bezier
curves.
= a1 L21 + a2 L22 + a3L23 + a4 L1 L2 + a5L2L3 + a6 L3 L1 = a01 + a02 x + a03 y + a04 x2 + a05 xy + a06 y2
(7.60)
64 The shape functions for the Linear Triangle are simply the area coordinates,
N1 = L1 N2 = L2 N3 = L3 (7.61)
and each one is equal to unity at one node, zero at the others, and varies linearly.
65 The shape functions for other elements can be obtained from the Lagrangian interpolation
function used earlier Eq. 7.27
Qn
(L , L )
lk = Qni=0;i6=k(L , Li )
n (7.62-a)
i=0;i6=k k i
= (L (L,,LL)(0 )( L , L1 ) (L , Lk,1 )(L , Lk+1) (L , Ln) (7.62-b)
k 0 Lk , L1 ) (Lk , Lk,1 )(Lk , Lk+1 ) (Lk , Ln )
66 Hence, denoting a typical node i by three numbers q; r and s corresponding to the position
of coordinates L1i ; L2i and L3i , we can write the shape functions in terms of three Lagrangian
interpolation functions
Ni = lqq (L1 )lrr (L2 )lss (L3 ) (7.63)
67 Using this formula, the shape functions for higher order elements are
Quadratic triangle
Ni = Li(2Li , 1) Corner nodes (7.64)
Nj = 4Li Lk Midside nodes, i; j; k are on the same side
Cubic triangle
Ni = 12 Li(3Li , 1)(3Li , 2) Corner nodes
Nj = 92 LiLk (3Li , 1) Midside nodes, i; j; k are on the same side (7.65)
N10 = 27L1 L2 L3 Internal node
Chapter 8
FINITE ELEMENT
DISCRETIZATION and
REQUIREMENTS
8.1 Discretization
Adapted from Reich 1993
u = Nue (8.1)
4 The virtual displacements u at a point inside the element can also be dened in terms of
the shape functions N and the nodal virtual displacements ue for the element
u = N u e (8.2)
8{2 FINITE ELEMENT DISCRETIZATION and REQUIREMENTS
5 In order to discretize the volume integral in Equation 6.45
Draft
Z Z Z Z Z
= (Lu)T D(Lu)d
, (Lu)T D0 d
+ (Lu)T 0 d
, uT bd
, uT ^td, = 0
,t
(8.3)
dening the virtual strain energy for the element due to the nodal displacements u, the strains
at a point inside the element are expressed in terms of the nodal displacements ue using
Equation 8.1
= Lu = LNue (8.4)
and the virtual strains at a point inside the element are expressed in terms of the nodal
virtual displacements ue using Equation 8.2
= (Lu) = LNue (8.5)
B def
= LN (8.6)
8 In order to discretize the volume integrals in Equation 8.3 dening the virtual strain energy
for the element due to the initial strains 0 and stresses 0 , Equations 8.5 and 8.6, which dene
the virtual strains at a point inside the element in terms of the nodal virtual displacements
ue , are substituted into the integrands
Z Z
(Lu)T D 0 d
= ue
T BT D0d
(8.10)
e
e
Z Z
(Lu)T 0 d
= uTe BT 0 d
(8.11)
e
e
Victor Saouma Finite Elements
8.1 Discretization 8{3
9 Dening the initial force vector f0e as
Draft
Z Z
f0e = BT D0d
, BT 0d
(8.12)
e
t
10 In order to discretize the volume integral dening the work done by the body forces and the
surface integral dening the work done by the surface tractions in Equation 8.3, Equation 8.2
is substituted into the integrands
Z Z
uT bd
= uTe NT bd
(8.14)
e
e
Z Z
uT ^td, = uTe NT ^td, (8.15)
,t ,t
the sum of the internal and external virtual work due to body forces and surface tractions is
Z Z
uT bd
+ uT ^td, = uTe fe (8.17)
e ,t
12 Having obtained the discretization of the various integrals dening the variational statement
for the TPE variational principle, it is now possible to dene the discrete system of equa-
tions. Substituting Equations 8.7, 8.13, and 8.17 into Equation 8.3 and rearranging terms, the
discretized Principle of Virtual Work is
uTe Keue = uTe fe + uTe f0e (8.18)
13 Since uTe is an arbitrary (i.e. non-zero) vector appearing on both sides of Equation 8.18,
the discrete system of equations can be simplied into
16 Completness can be assessed through the patch test which will be discussed later.
17 Zero Strain Energy: For structural problems, there should be zero strain energy when the
element is subjected to a rigid body motion. Recall that in the stiness matrix formulation, the
matrix is singular as it embodies not only the force displacement relations, but the equilibrium
equations also. To each of those equations, corresponds a rigid body mode which can be detected
by an eigenvalue analysis (more about this later).
Chapter 9
\SIMPLE" FINITE ELEMENTS
9.1 Introduction
1 Having rst introduced the method of virtual displacements, than the interpolation (or shape)
functions [N], which relate internal to external nodal displacements, and nally having applied
the virtual displacement method to nite element in chapter ??, we derive the siness matrices
of some simple elements.
8 Using the shape function for the beam element from Eq. 7.43, and noting the change of
integration variable from dx to d , we obtain
8 6 9
>
> L 2 (2 , 1) >
>
Z 1<
[k] = EIz
, 2 (3 , 2) =
L 6 (2 , 1) , 2 (3 , 2) 6 (,2 + 1) , 2 (3 , 1)
6 (,2 + 1) L 2 L L 2 L Ld (9.6)
> L2
0> >
> |{z}
:
, L (3 , 1)
2 ; dx
or
2
v1 1 v2 2 3
12 We note that because we have 3 u and 3 v displacements, the size of [B] and [u] are 3 6
and 6 1 respectively.
13 Dierentiating the shape functions from Eq. 7.24 we obtain:
,1 1
2 3
8 1 9
>
> "xx >
> 6 x 0 x 0
|{z} 0 0 7
6 | {z 2} 2
>
>
>
>
>
"1yy >
>
>
>
>
6
6 @N1
|{z}
@N2
@N3
@x
7
78
7> 1u 9
>
>
xy
1 >
> 6 @x @x 7> >
>
>
>
>
>
> "2xx
>
>
>
>
>
6
6
0 0 0 x3 , x2 , ,x3 1 7>
7>
>
u
> 2
>
>
>
>
>
>
<
"2yy
=
=
6
6
6
x2 y3 x2 y3 y3 7<
7
7
u 3
=
(9.11)
>
>
>
xy
2 >
>
>
6
6
| {z } | {z } |{z}
@N1 @N2 @N3 7>
7>> v 1 >
>
>
>
>
>
"xx
>
>
> 6
x , x , ,x3 1
@y @y @y 7>> v
> 2
>
>
>
, x1 1
>
> 3 >
> 6 7> >
>
> >
> 6 3 2
0 7: v ;
3
>
>
>
> "3yy >
>
>
>
6
6 x y | x{z2y3} |{z}
2 3 y3 | {z 2}
x 2 |{z}
@N3
7
7 | {z }
:
xy
3 ; 4 | {z }
@N1 @N1 @N3 @N1
|{z}
@N2 @x
5 u
f g
| {z } @y @y @x @x @x
fg | {z }
[B]
2 3
, x12 0 x3 ,x2
x2 y3
6 1
x2 0 , x,2xy33 7 " #2 , x12 1 0 0 0 0
3
Z 6
6 0 0 1 7
7 E 1 0 x2
y3 1 x3 ,x2 , x,2xy33 1
= 6
6 0 x3 ,x2 , x12
7
7 1 , 2 0
1, )
4
x3 ,x2
0 0 0 x2 y3 y3 5 tdxdy
(9.12)
6
4 0
x2 y3
, x,2xy33 1 7
5|
0 0
{z 2 } | x2 y3
, ,x3
x2 y3 y3 {z , x2
1 1 1
x2 0 |dvol{z }
1
x2 }
0 0 [D] [B]
|
y{z
3 }
[B]T
2 3
y32 + x23,2 ,y32 2, x3 x23,2 x2 x3,2 ,y3 x3,2 x3y3 + y3 x3,2 ,x2y3
6 ,y32 , x3 x3,2 y 3 + x 3 , x 2 x 3 y 3 x 3 , 2 + x 3 y 3 ,x3 y3 x2 y3 7
6 7
6 x 2
=
6 ,y3 x3,2x3 , 2 , x2 x 3 x 2
2 , x 2 y3 x 2 y 3 0 7
6 y3 x3,2 + x3 y3 ,x2 y3 y32 + x23,2 , y32 , x3 x3,2 x2 x3,2 7
7
4 x3 y3 + y3 x3,2 ,x3 y3 x2 y3 ,y32 , x3 x3,2 y32 + x23 ,x22x3 5
,x2y3 x2 y3 0 x2 x3,2 ,x2 x3 x2
where = 1,2 , = 1+2 ,
= 2(1,ET
2 )x2 y3 , x3,2 = x3 , x2 , and y3,2 = y3 , y2 .
9.3.1.1.3 Internal Stresses 15 Recall from Eq. ?? that f g = [D] [B]fug hence for this
particular element we will have:
8 9
>
> u1 >
>
8
> x
9
>
2
1 0
32
, x12 1 0 0 0 0
3>>
>
>
>
u2 >
>
>
>
>
< = E x2 x , x , x
<
u3 =
y = 1, 6
1 0 76
0 0 0 x2 y3 , x2 y3
3 2 3 1 7
y3 >
> > 24 1, )
54
x3 ,x2 , ,x3
5
v1 >
xy 0 0 y3 {z , x2 0
: ; 1 1 1 >
> >
>
| {z } | {z
2 }|
x2 y3 x2 y3 x2 >
>
}>
>
:
v2 >
>
>
>
;
f g [D] [B] v3
| {z }
fu g
8 9
>
> u1 >
>
2
,y3 y3 0 x3,2 ,x3 x
3>>
>
> u2 >
>
>
>
2 ><
u3 >
=
= 64 ,y3 y3 0 x3,2 ,x3 x 7
2 5> v1 (9.13)
x3,2 ,x3 x2 ,y3 y3 0 >
>
>
>
> v2
>
>
>
>
>
>
>
: >
;
v3
where = (1,E2 )x2 y3
9.3.1.2 Natural Coordinate System
16From Eq. 9.12 we note that the element must be positioned in such a way that neither x2
nor y3 is equal to zero.
Victor Saouma Finite Elements
9.3 Plane Stresss/Strain Elements 9{5
17 A palliative to this problem is to use a formulation based on natural coordinate systems.
Draft
18 The displacement eld is given by
8 9
>
>
u1 >
>
( ) "
>
>
#>
>
>
v1 >
>
>
>
>
u = L01 L0 L02 L0 L03 L0
<
u2 =
(9.14)
v 1 2 3 >
>
>
v2 >
>
>
>
>
>
>
:
u3 >
>
>
>
;
v3
19 The strains are given by = LNu = Bu using Eq. 9.16
= (L1 ; L2 ; L3 ) (9.15-a)
@ = @ @L1 + @ @L2 + @ @L3 (9.15-b)
@x @L1 @x @L2 @x @L3 @x
@ = @ @L1 + @ @L2 + @ @L3 (9.15-c)
@y @L1 @y @L2 @y @L3 @y
@L1
@x = y223A ; @L
@x
2 = y31 ; @L3 = y12 ;
2A @x 2A (9.16)
@L 1 = x232A ; @L 2 = x13 ; @L3 = x21 ;
@y @y 2A @y 2A
we obtain 2 3
y23 0 y31 0 y12 0
[B] = 21A 64 0 x32 0 x13 0 x21 7
5 (9.17)
x32 y23 x13 y31 x21 y12
20 The constitutive matrix is given by
2 3
1 0
[D] = 1 6
1, 2 4 1 0 7
5 Plane Stress
0 02 1,2
1 0 3 (9.18)
(1, )
= E (1, ) 6 1 0 7
Plane Strain
(1+ )(1,2 ) 4 1, 5
0 0 1,2
2(1, )
21 Finally, the element stiness matrix is obtained from Eq. 8.8
Z
Ke = BT DBd
(9.19)
e
Table 9.1: Interpretation of Shape Functions in Terms of Polynomial Series (1D & 2D)
Element Terms # of Nodes
(terms)
Linear a1 , a2 2
Quadratic a1 , a2 , a4 3
Bi-Linear (triangle) a1 , a2 , a3 , 3
Bi-Linear (quadrilateral) a1 , a2 , a3 , a5 4
Bi-Quadratic (Serendipity) a1 , a2 , a3 , a4 , a5 , a6 , a8 , a9 8
Bi-Quadratic (Lagrangian) a1 , a2 , a3 , a4 , a5 , a6 , a8 , a9 , a13 9
Table 9.2: Polynomial Terms in Various Element Formulations (1D & 2D)
24 A complete polynomial contains all the terms above a certain line in the Pascal triangle.
25 The more terms are included, the higher the accuracy. For instance u = a1 + a2 x + a3 y + a5 xy
is more accurate than u = a1 + a2 x + a3 y, however the rate of convergence is unchanged (because
Victor Saouma Finite Elements
9.3 Plane Stresss/Strain Elements 9{7
Draft
29 Imposing nodal displacements consistent with theory (u1 = ,u, u2 = u, and u3 = 0) and
similar descriptions for v, and determining the strains from = Bu for the CST element, we
obtain
1
"xx = 0 "yy = 0
xy = c , 4l2 u c (9.24)
which are clearly wrong. Hence, we conclude the CST element can not properly perform in
bending problems (As is expected from a constant strain element behavior when used in a linear
strain problem).
Victor Saouma Finite Elements
9{8 \SIMPLE" FINITE ELEMENTS
30 This deciency will later be addressed by including an additional internal \drilling" degree
Draft
of freedom.
31 Zero energy mode can be veried by considering for example the strain "xx which is given by
" = 1 (,y u + y u )
xx x2 y3 3 1 3 2 (9.25)
since u1 = u2 for a rigid body motion, we see that "x = 0, similar conditions can be found for
"yy and "xy
9.3.3.3 BiLinear Element
Flexure Under
exure, the top and bottom sides remain straight giving rise to erroneous
strains.
Inclined edges: From Pascals's triangle, the displacement eld for a bilinear rectangular ele-
ment is given by
u = a1 + a2 x + a3 y + a5 xy (9.26)
which is consistent with the element shape functions derived earlier (Eq. 7.33)
32 We note that whereas there is a quadratic term (a5 xy) convergence is governed by the
full linear expansion, and that this last term may cause parasitic eects.
33For an element with an inclined boundary given by y = ax + b, insertion of this
expression into Eq. 9.26 yields
u = a1 + a3 b + (a2 + a3 a + a5 b)x + a5ax2 (9.27)
hence, u varies quadratically with x along the inclined boundary. However, this bound-
ary is uniquely dened by only two nodal displacements u1 and u2 , thus the quadratic
expression is not uniquely dened and we may have dierent variation along the side from
one element to the adjacent one.
34 Continuity across the element not being satised, the element is non-conforming.
35We conclude that the bilinear quadrilateral element must always have its four edges
parallel to the coordinate system.
36 This severe restriction will be alleviated in the next lecture by Isoparametric element
formulations.
Equilibrium of Nodal Forces: Those are automatically satised by denition (Ku , f = 0).
Compatibility at Nodes: Elements connected to one another have the same displacements
(along corresponding dof) at the connecting node.
Interelement Continuity of Stress: is usually not satised. For example consider Fig. 9.3
if node 4 is the only one displaced (all others remaining xed), then there will be a
discontinuity of xx across element boundary 2-3.
Interelement Compatibility: Most elements have this requirement satised. If an element
does not then it is labeled as incompatible or nonconforming, Fig. 9.3
Equilibrium Inside Elements: is not always satised. In general satisfaction of equilibrium
at every point demands relations between dof which do not necessarily result from Ku ,
f =0
Compatibility Inside an Element: is satised as we only require that the element displace-
ment be continuous and single-valued.
Chapter 10
ISOPARAMETRIC ELEMENTS
10.1 Introduction
1 We have previously examined simple nite elements, in this lecture we shall distort those
simpler elements into others of more arbitrary shape.
2 Correspondingly, the natural coordinates will be distorted into new curvilinear sets when
plotted in a cartesian x; y; z space, Fig. 10.1.
Local coordinates
(-1,1) (1,1)
(-1,-1)
(1,-1)
L 2=0 3
1
L 2=0 3
L1 =0
L 3=
1
0
L1 =0
2
L 3=
0
5 Nodal displacements at any point inside the element can be written in terms of the nodal
10{2 ISOPARAMETRIC ELEMENTS
vj
Draft
(-1,1) (1,1)
uj
mapping
x,u
Actual Element Parent Element
6 When elements are also distorted, the coordinates of any point can also be expressed in terms
of nodal coordinates
8 9
x1 >
< >
=
x = N1 x1 + N2 x2 + = N > x2
~ ~ ~ ~
= Nx
: .. >
;
8
. 9
y >
< 1 >
=
y = N~1 y1 + N~2 y2 + = N~ > y2 ~
= Ny (10.3)
: .. >
;
8
. 9
>
<
z 1 >
=
z = N~1 z 1 + N~2 z2 + = N~ > z2 ~
= Nz
: .. >
;
.
7 Shape function matrices [N] and [N~ ] are functions of ; ; and . Hence, three possible cases
arise, Fig. 10.3
Geometric Displacement
Nodal Nodal
Config. Config.
Isoparametric
Subparametric
Superparametric
1 3 2
x,u
L
= ( (,, )(1 )( ,,2))(( ,,k,1 )()( ,,k+1 ) )((, ,n ) ) (10.5-b)
k 1 k 2 k k,1 k k+1 k n
For a three noded quadratic element 1 = ,1, 2 = +1, and 3 = 0. Substituting, we obtain
the three shape functions
( ,2 )( ,3 )
N1 ( ) = (1 ,2 )(1 ,3 ) = (,(1,,1)(
1)( ,0)
,1,0) = 2 ( , )
1 2
( ,1 )( ,3 ) ( +1)( ,0)
N2 () = (2 ,1 )(2 ,3 ) = (1+1)(1 ,0) = 12 ( 2 + ) (10.6)
( ,1 )( ,2 ) ( +1)( ,1)
N3 () = (3 ,1 )(3 ,2 ) = (0+1)(0 ,1) = 1 , 2
Hence,
x() = bNcb x1 x2 x3 cT and u( ) = bNcb u1 u2 u3 cT (10.7)
where
bNc = b 12 (2 , ) 12 (2 + ) 1 , 2 c (10.8)
11 The strain displacement relation is given by Eq. 8.4, = Lu = LNue = Bue , and the
dierential operator L is equal to dxd . For this one dimensional case, this reduces to
8 9
17 We observe that B, in general, contains terms in both the numerator and denominator, and
hence the expression can not be analytically inverted. Furthermore, the limits of integration
are now from -1 to +1, and we shall see later on how to numerically integrate it.
10.2.2 Bilinear Isoparametric Element
18 We have previously derived the stiness matrix of a rectangular element (aligned with the
coordinate axis), this formulation will generalize it to an arbitrary quadrilateral shape.
19 Fig. 10.5 illustrates the element
x,u
Figure 10.5: Four Noded Isoparametric Element
25 Considering the local set of coordinates ; and the corresponding global one x; y, the chain
rules would give
( ) " #( ) ( ) ( )
@Ni @x @y @Ni @Ni @Ni
@
@Ni = @ @
@x @y
@x
@N i or @x
@N i = [J],1 @
@Ni (10.24)
@ @ {z @ @y @y @
| }
J
26 Thus B, Eq. 10.23-a is obtained from
2 3 2 3
@Ni (;) 0 @Ni (;) 0
4 6
X @x @
B(; ) = 6 0 @Ni (;) 7
7 = J,1 64
6
0 @Ni (;) 7
7 (10.25)
4 @y 5 @ 5
i=1 @Ni (;) @Ni (;) @Ni (;) @Ni (;)
@y @x @ @
Victor Saouma Finite Elements
10{8 ISOPARAMETRIC ELEMENTS
27 Expanding the denition of the Jacobian
Draft
8 9 " #( ) " #
< @Ni (;) = @x @y @Ni 4
X @Ni :x @Ni :y
@ = @ @ @x = @ i @ i (10.26-a)
: @Ni (;) ; @x @y @N i @Ni :xi @Ni :y
@ |
@ {z @ }
@y i=1 @ @ i
J 2 3
" @N1 @N2 @N3 @N3 # x1 y1
6 x2 y2
6 7
= @ @ @ @ = 64 x 7
(10.26-b)
@N1 @N2 @N3 @N3 3 y3 7
5
@ @ @ @
x4 y4
2 3
" # x1 y1
= 14 , (1 , ) (1 , ) (1 + ) ,(1 + ) x2 y2 777(10.26-c)
6
6
,(1 , ) ,(1 + ) (1 + ) (1 , ) 6
4 x3 y3 5
x4 y4
29 From calculus, if and are some arbitrary curvlinear coordinates, Fig. 10.6, then
y
6
B
@y d C
ds d
@
e
@y d
A:
@
d
r
O @x d @x d
-
x
@ @
6C=AB
B ,
,
, jBj sin
,,
, -
A
C = AB (10.29-a)
= jAjjBj sin k
(10.29-b)
i j k
Ax Ay 0z = (|AxBy {z
=
, BxAy}) k (10.29-c)
Bx By 0
Area
jCj = jAjjBj sin (10.29-d)
hence, the dierential area dxdy is then equal to the length of the vector resulting from the
cross product of drds and is equal to
" #
@x @x
d(area) = dxdy = det @ @ dd
@y @y (10.30)
|
@ {z @ }
J
31 Finally determine the element stiness matrix from
Z Z Z 1 Z 1
[k]88 = [B]T
83 [D]3x3 [B]38 tdxdy = [k] = [B]T [D][B]tjJjdd (10.31)
,1 ,1
detJ = dx:dy
d:d sin
(10.33)
Thus we observe that if is small or close to 180o , then det J will be very small, if the angle is
greater than 180 o , the determinant is negative (implying a negative stiness which will usually
trigger an error/stop in a FE analysis).
33 In general it is recommended that 30o < < 150o .
34 The inverse of the jacobian exists as long as the element is not much distorted or folds back
upon itself, Fig. ?? in those cases there is no unique relation between the coordinates.
35 It can be easily shown that for parallelograms, the Jacobian is constant, whereas for nonpar-
allelograms it is not.
36 In general J is an indicator of the amount of element distorsion with respect to a 2x2 square
one. Some times it is constant, others it varies within the element.
y
1
2 y 1
1
x 2
x
4 1 3/4
3 4
3 6 4
2
y
1
2
60
x 1
3 6 4
Solution:
The coordinates are given by Eq. 10.4, the shape functions by Eq. 10.19, and the Jacobian by
Eq. 10.24.
Element 1:
x = 14 (1 , )(1 , )x3 + 41 (1 + )(1 , )x4
+ 41 (1 + )(1 + )x1 + 14 (1 , )(1 + )x2 (10.34-a)
= 14 (1 , )(1 , )(,3) + 14 (1 + )(1 , )(3)
+ 14 (1 + )(1 + )(3) + 14 (1 , )(1 + )(,3) (10.34-b)
y = 14 (1 , )(1 , )y 3 + 14 (1 + )(1 , )y 4
+ 14 (1 + )(1 + )y 1 + 14 (1 , )(1 + )y2 (10.34-c)
= 14 (1 , )(1 , )(,2) + 14 (1 + )(1 , )(,2)
+ 14 (1 + )(1 + )(2) + 14 (1 , )(1 + )(2) (10.34-d)
" #
[J] = 30 02 (10.34-e)
Element 2:
p p
x = 14 (1 , )(1 , )(,(3 + 1=(2 3))) + 14 (1 + )(1 , )(3 , 1=2 3))
Victor Saouma Finite Elements
10{12 ISOPARAMETRIC ELEMENTS
p p
+ 14 (1 + )(1 + )(3 + 1=2 3)) + 14 (1 , )(1 + )(,(3 , 1=2 3))) (10.35-a)
Draft
Element 3:
x = 41 (1 , )(1 , )(,1) + 14 (1 + )(1)
+ 14 (1 + )(1 + )(1) + 14 (1 , )(,1) (10.36-a)
y = 14 (1 , )(1 , )(,3=4) + 14 (1 + )(1 , )(,3=4)
+ 14 (1 + )(1 + )(5=4) + 14 (1 , )(1 + )(1=4) (10.36-b)
" #
1 4 (1
[J] = 4 0 (3 + ) + ) (10.36-c)
4 4
7 7
3 3
8 8 9
6 6
1 1
5 5
2 2
a b
39In this formulation, the x2 y2 term is missing and to the 8 terms in the assumed polynomial
expansion correspond the 8 nodes (4 corner and 4 midside).
Figure 10.10: Serendipity Isoparametric Quadratic Finite Element: Global and Parent Element
41 The shape functions for the corner and midisde nodes are shown in Fig. 10.11.
10.2.3.2 9 Noded, Lagrangian element
42 If we were to follow a similar procedure to the one adopted to extract the bilinear shape
fuctions, we would obtain 9 shape functions, which must in turn correspond to 9 (rather than
8) nodes.
1 1
0.8
0.5
0.6
0
0.4
0.5
0.2
1 0
1 1
0.5 1 0.5 1
0 0.5 0 0.5
0 0
0.5 0.5
0.5 0.5
1 1 1 1
0 1 0.1
1
0.2
0.8
0.8
0.6
0.6 0.6
0.4
0.5
0.4 0.4
0.9
0.2 0.2
0.4 0.4
44 All the quadratic terms are present, hence there are 9 terms in the polynomial expansion,
and the 9th node will correspond to an internal node.
45 The shape functions in this case can be directly obtained from the Lagrangian interpolation
functions, yielding
Ni = 41 (1
,
+ i) (1 + i ) (i + i , 1) i = 1; 2; 3; 4
Ni = 12 1 , 2 (1, + i) i = 5; 7 (10.40)
Ni = 12 (1 + i ) 1 + 2 i = 6; 8
N9 = (1 , 2 )(1 , 2 ) i=9
46 The last shape function is often called bubble function
47 Those shapre function dier slightly from those of the serendipity element.
48 Q9 elements perform much better than the Q8 if edges are not parallel or slightly curved.
49 The shape functions for the corner and midisde nodes are shown in Fig. 10.12.
10.2.3.3 Variable Element
50 Based on the above, we can generalize the formulation to one of a quadrilateral element with
variable number of nodes.
51 This element may have dierent order of variation along dierent edges, and is quite useful
to facilitate the grading of a nite element mesh.
52 In its simplest formulation, it has four nodes, and has a linear variation along all sides, and
in the most general case it is a full quadratic element.
53 The shape functions may than be obtained from Table 10.1. Note that these shape functions
are for the hierarchical element in which the corner nodes are numbered rst, and midside ones
after, Fig. 10.13.
0.5 0.8
0.6
0
0.4
0.5 0.2
0
1
1 0.2
1
0.5 1 0.5 1
0 0.5 0.5
0
0 0
0.5 0.5
0.5 0.5
1 1 1 1
1 1
0.8 0.8
0.6 0.6
0.6
0.4 0.4
0.6
0.2 0.2
0.2
0 0
1 0.1
0.8 0.4
1
0.6 0.6
0.8 0.2
0.4
0.8
0.6 0.2
0.4 0
0.9
0.2 0.2
0.4
0 0.7
1
0.6 0.5
0.5 1
0 0.5
0.8 0.3
0
0.5
0.5
1
1 1 1 0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1
i fi fi; fi;
1 4 , , , , , 1)
1 (1 )(1 )( ,
1 (2 + )(1 )
4 ,
1 (1 )(2 + )
4
2 4 , , , 1)
1 (1 + )(1 )( , ,
1 (2 )(1 )
4 ,
1 (1 + )(2 )
4
3 4 , 1)
1 (1 + )(1 + )( + 1 (2 + )(1 + )
4
1 (1 + )(2 + )
4
4 4 , , , , 1)
1 (1 )(1 + )( ,
1 (2 )(1 + )
4 , ,
1 (1 )(2 )
4
5 , ,
1 (1 2 )(1 )
2 ,(1 , ) , ,1 (1 2 )
2
6 ,
1 (1 + )(1 2 )
2
1 (1 , 2 )
2 ,(1 + )
7 ,
1 (1 2 )(1 + )
2 ,(1 + ) 1 (1 , 2 )
2
8 , ,
1 (1 )(1 2 )
2 , 12 (1 , 2) ,(1 , )
Constant term 1
Linear terms Linear element
Quadratic terms 2 2 Quadratic element
Cubic terms 3 2 2 3 Cubic element
Quartic terms 3 3
Serendipity Quadrilateral Elements
Constant term 1
Linear terms Linear element
Quadratic terms 2 2 Quadratic element
Cubic terms 3 2 2 3 Cubic element
Quartic terms 3 2 2 3
Quintic terms 3 2 2 3
Sixtic terms 3 3
Lagrangian Quadrilateral Elements
Constant term 1
Linear terms Linear elements
Quadratic terms 2 2 Quadratic element
Cubic term 3 2 2 3 Cubic element
Triangle Elements
Figure 10.14: Pascal Triangles for Quadrilateral and Triangle Elements
f3;x = b3 f3;y = a3
2A 2A
Then the ith part of the B matrix becomes:
2 3
fi;x 0
Bi = 6
4 0 fi;y 7
5 (i = 1; 2; 3) (10.46)
fi;y fi;x
Victor Saouma Finite Elements
10{20 ISOPARAMETRIC ELEMENTS
The stinesses are obtained from:
Draft
K = BT E BAt (10.47)
As before, we will approach this isoparametric element by rst examining its straight-sided
parent. Nodal displacements for either element are:
q = fq1 ; q2; ; q12 g = fu1 ; v1 ; ; v6 g (10.48)
Geometric interpolation functions for the parent element are linear, as is the case for any
triangle with straight sides. Quadratic displacement shape functions may be written in natural
coordinates as: 6 6
X X
u = fiui v = fi vi (10.49)
i=1 i=1
where
f1 = (21 , 1)1 f4 = 41 2
f2 = (22 , 1)2 f5 = 42 3 (10.50)
f3 = (23 , 1)3 f6 = 43 3
Derivatives for strain-displacement relationships in this subparametric triangle are found by
substituting the expressions for 1 ; 2 and 3 from Eqs. xxx into the quadratic functions of Eqs.
xxx and dierentiating with respect to x and y. Then the stiness matrix for the element is
obtained from: Z
K = t BT E B dA (10.51)
A
Wherever x and y appear in this integral, we substitute Eqs. xx, and we use area integrals to
nd the explicit results in Table xx.
For the isoparametric triangle T 6 in Fig. xx, we take the geometric interpolation functions
to be: 6 6
X X
x = fi xi y = fi yi (10.52)
i=1 i=1
where fi are given by Eqs. xx. Thus, the edges of the element become quadratic curves, as
indicated in the gure. Because the natural coordinates are curvilinear, the Jacobian matrix is
required. Thus,
6
X 6
X
J11 = @x = fi;1 xi J12 = @y = fi;1 yi
@1 @1
i=1 i=1
(10.53)
6
X 6
X
J21 = @x = fi;2 xi J22 = @y = fi;2 yi
@2 @2
i=1 i=1
i fi fi;1 fi;2
58 As shown above, F = BT :D:B for nite element stiness matrix evaluation, and each element
is integrated individually.
Rb
59 The integration of a F ( )d is essentially based on passing a polynomial P ( ) through given
values of F ( ) and then use ab P ( )d as an approximation.
R
Z b Z b
F ()d P ()d (10.57)
a a
-1 0 1
63It can be shown that this method permits exact integration of polynomial of order n , 1,
and that if n is odd, then we can exactly integrate polynomials of order n. Hence we use in
general odd values of n,
10.3.1.1 Newton-Cotes Quadrature for n = 2 and n = 3
For
R1
n = 2 over [,1; 1], we select equally spaced points at 1 = ,1 and 2 = 1 to evaluate
,1 P ( )d
P2
P () = i=1 li ( )F (i )
l1 () = ,2 = 1 (1 , )
1 ,2 2
l2 () = ,1 = 1 (1 + )
Z2 ,1 2 Z
l1()d = 21
1 1
W1(2) = (1 , )d = 1
Z , 1 Z,11
l2 ( )d = 12
1
W2(2) = (1 + )d = 1
,1 ,1
Z 1 Z 1 2
X
F ()d P ()d = Wi(2) F (i) = F (,1) + F (1)
,1 ,1 i=1
which is the trapezoidal rule
For n = 3 over [,1; 1], we select equally spaced points at 1 = ,1 2 = 0, and 3 = 1, to
Z 1 Z 1 n
X
F ()d P ( )d = Wi(n)F (i ) (10.61)
,1 ,1 i=1
however in this formulation both Wi(n) and i are unknowns to be yet determined. Thus, we
have a total of 2n unknowns.
66 At the integration points P (i ) = F (i ), however at intermediary points the dierence can
be expressed as
F ( ) = P () + | ()(0 + 1 {z+ 2 2 + }) (10.62)
0 at =i ;i=1;2;;n
since we want the left side to be exactly equal to P ( ) at the integration points, we dene
() = ( , 1 )( , 2 ):::( , n ) (10.63)
as a polynomial of order n, to be equal to zero at the integration points i .
67 i should be appropriately selected in orer to eliminate the gap between F ( ) and P ( ) at
intermediary points.
Wi(n) = ,1 1 li ()d
R
Weights R (10.69)
Gauss Points ,1 1 ( ) j d = 0 j = 0; 1; ; n , 1
71 Integration points i and weight coecients Wi(n) are given in Table 10.5, (Bathe 1982).
n i Wi(n) Error
1 1 F (2) ( )
p0 p 2 3
2 ,
p
1= 3 1=p 3 1 1 135 F ( )
1 (4)
3 , 3=5 0 3=5 5=9 8=9 5=9 15;750 F ( )
1 (6)
Table 10.5: Integration Points and Weights for Gauss-Quadrature Formulaes Over the Interval
[,1; 1]
72 The solutions (Gauss integration points) are equal to the roots of a Legendre polynomial
dened by
L0 () = 1
L1 () = (10.70)
2 ,1 k ,
Lk () = k Lk,1() , k Lk,2 () 2 k n
k 1
and the n Gauss integration points are determined by solving Ln ( ) = 0 for its roots i ,
i = 0; 1; ; n , 1.
73 The weighting functions are then given by
2(1 , i )
Wi(n) = [nL
2
(10.71)
( )]2
n,1 i
= + 0.6
= +1/ 3
= 0.6
= 1/ 3 9 = + 0.6
= +1/ 3 6
4
2 3
5 8
2
1 1
3
4 7
= 1/ 3 = 0.6
78 In the isoparametric formulation, nodal stresses are very poor, and best results are obtained
at the Gauss points.
79 To evaluate nodal stresses, two approaches:
As expected, the sum of each row is equal to one, and for stresses we replace by xx, y y,
and xy
81 As we know, dierent nodal stresses will be obtained from adjacent elements. To obtain a
single value we can either take
1. Unweighted average of all the nodal stresses.
2. Weighted average of nodal stresses based on the relative sizes of the elements as determined
from their area through det(J ).
89 First we determine the components of the distributed loads in the x and y directions by
considering the forces acting on an incremental length dS of the loaded edge, Fig. 10.19:
dPx = (pt dS cos , pndS sin ) = (pt dx , pndy) (10.86)
dPy = (pndS cos , ptdS sin ) = (pndx , pt dy)
"0xx = (1 + )T (10.95-a)
Victor Saouma Finite Elements
10.6 Computer Implementation 10{33
"0yy = (1 + )T (10.95-b)
Draft
xy
0 = 0 (10.95-c)
zz0 = ,ET (10.95-d)
meshc(X,Y,N1)
print -deps2 shap8-1.eps
c=contour(X,Y,N1);
clabel(c)
print -deps2 shap8-1-c.eps
meshc(X,Y,N8)
print -deps2 shap8-8.eps
c=contour(X,Y,N8);
clabel(c)
print -deps2 shap8-8-c.eps
N1=N1-0.25*N9;
meshc(X,Y,N1)
print -deps2 shap9-1.eps
c=contour(X,Y,N1);
clabel(c)
print -deps2 shap9-1-c.eps
N8=N8-0.5*N9;
meshc(X,Y,N8)
print -deps2 shap9-8.eps
c=contour(X,Y,N8);
clabel(c)
print -deps2 shap9-8-c.eps
meshc(X,Y,N9)
print -deps2 shap9-9.eps
c=contour(X,Y,N9);
clabel(c)
print -deps2 shap9-9-c.eps
Chapter 11
WEIGHTED RESIDUAL
METHODS
Adapted from Ottosen (1992)
11.1 Introduction
1 Thus far, the nite element method was derived from a variational principle through the
Rayleigh-Ritz method.
2 In Elasto-Statics, the variational principle used was the minimization of the total potential
energy, and the resulting weak formulation is equivalent to the equation of equilibrium (strong
form).
3 In many branches of physical sciences, a variational principle may not be available, and all
what is available is a dierential equation and its associated boundary conditions. Note that a
variational formulation is not possible because the order of the dierential equation is odd.
4 It is for those problems that the weighted residual methods are employed.
Z
(11.2)
Positive Denite (L[u])ud
> 0
where u and v are any functions which satisfy essential and natural boundary conditions.
11.2.1.1 Application to 1D Axial Member
8 As an example, let us consider the steady state response of a bar. The governing dierential
equation and boundary conditions are
Governing Dierential Equation ,EA @@x2 u2 = 0
Essential Boundary conditions ujx=0 = 0 (11.3)
Natural Boundary conditions u
EA x x=L = P
Hence
L = ,EA @x@ 22 ; = u; g = 0; (11.4)
B1 = 1; q1 = 0; B2 = EA x ; q2 = P ;
9 To determine whether the operator L is symmetric and positive denite, we consider P = 0,
and integrating by part twice1 we obtain
l
Z l @ 2u @u
Z l
,EA @x2 vdx = , EA @x v + EA @x@u @v dx (11.5-a)
0 0 0 @x
l l
@u
@v
Z l
= , EA @x v + EAu @x , EA @x
@ 2 v udx (11.5-b)
2 0
0 0
But since the boundary conditions are u = v = 0 at x = 0 and @u=@x = @v=@x = 0 at x = l,
then we only have
,EA @@xu2 vdx = ,EA @x @ 2 v udx
Z l 2 Z l
2 (11.6)
0 0
hence the operator is symmetric.
We can also conclude that the operator is positive denite from Eq. 11.5-a
2
Z l @ 2u Z l
@u
,EA @x2 udx = EA @x dx (11.7)
0 0
Z b Z b
1 From Eq.2.11: u(x)v0 (x)dx = u(x)v(x)jba , v(x)u0 (x)dx
a a
R = L( a) + g = L( )a + g (11.16)
18 In this method, the weight function w is based on Dirac's delta function (x , xi ) and thus
19 The residual is set to be exactly equal to zero at n points. This is achieved by adopting for
the weight function w the Dirac's delta function (x , xi ) where
(
(x , xi ) = 1 if x = xi (11.19-a)
0 otherwise
Z 1
(x , xi)dx = 1 (11.19-b)
,1
i.e. the weight functions are equal to the trial functions. Since the trial functions are orthogonal
to the residuals, then Eq. 11.15 becomes
Z b T
Rdx = 0 (11.22)
a
LT + b = 0 (11.24)
where = Lu and
8 9
>
> xx >
>
2 @ 0 @y@ @z@ 0
0
3 >
>
>
> yy >
>
>
>
8
bx
9
@x >
<
zz >
= >
< >
=
LT = 64 0 @y@ 0 @x@ 0 @z@ 75 ; = > xy >
; b=> by >
; (11.25)
0 0 @z@ 0 @x@ @y@ >
>
>
>
> xz
>
>
>
>
>
: bz ;
>
: >
;
yz
xy
,
,
@y
+ wx xz nz d, , @w
Z Z Z
x d
+ wxbxd
= 0 (11.28-a)
@z xz
,
x x
Z
Z
wz tz d, , @w z @w z @w z
Z
,
,
Z
@wx + @wy + @wz +
xx @y yy @z zz
@x
@wx + @wy + @wx + @wz + @wy + @wz d
= 0 (11.32-a)
@y @x xy @z @x xz @z @y yz
5 Note similarity with the derivation of the principle of virtual work.
6 (R vi;i d
= R vi ni d,)
,
36 But since the L operator is symmetric, from Eq. 11.2, the above equation can also be written
as
Z Z Z
wT (L)d
= wT td, + wT bd
(11.36)
,
which is the weak form of the dierential equation of equilibrium subjected to the traction
(natural) boundary conditions.
37 The weight vector w is completely arbitrary. In the principle of virtual displacement, it
would correspond to the virtual displacement.
11.4.1.2 FE Discretization
38 Given that the displacement eld is written as
u = Nu (11.37)
where N are the shape functions, and u are the nodal known displacements. Note that this is
the application of Eq. 11.9-a
39 The Galerkin method uses the shape functions for the weight (functions Wi ) Hence, Eq.
11.13-a will be
w = cN (11.38-a)
and
Lw = LNc = Bc (11.39)
Victor Saouma Finite Elements
11.4 Applications of the Galerkin Method 11{9
40 Inserting the last two equations into the weak form, Eq. 11.36, yields
Draft
Z Z Z
cT
B T d
,
,
N T td, ,
N T bd
= 0 (11.40)
(Note similarity with Eq. 11.14). Since c is an arbitrary matrix, we conclude that
Z Z Z
BT d
= NT td, + NT bd
(11.41)
,
This equation is applicable to both linear and nonlinear systems. The right hand side repreent
the eect of the load.
41 Next we introduce a constitutive model, Eq. 3.62 which is rewritten in vector form
= D( , 0 ) (11.42)
44 Thus, using the Galerkin approach, we have recovered the same stiness formulation as the
one previously obtained from a Variational Principle.
45 Whereas, there may not be any clearer advantage in using the Galerkin approach when a
variational principle is available, this may be the only formulation possible in the absence of
the second.
Chapter 12
FINITE ELEMENT
DISCRETIZATION OF THE
FIELD EQUATION
Adapted from Ottosen (1992)
1 Having \proven" the validity of the Galerkin method by its ability to recover the variational
based formulation, we now turn our attention to a problem which (strictly speaking) could only
be handled by such an approach.
2 As a vehicle for such an application, we will consider the eld equation which was discussed
in Chapter 4.
4 Recalling (Eq. 4.5) that q = ,Dr, and that for steady state problems the right hand side
reduces to zero, and nally (for the mere sake of clarity) substituting by T , this equation
reduces to
div
|
q{z, Q} = 0 (12.2)
L+g
12{2 FINITE ELEMENT DISCRETIZATION OF THE FIELD EQUATION
5 The boundary conditions are given by
Draft
Essential, Temperature ,T T = g
Natural, Flux ,q qn = qt n = qf (12.3)
Natural, Convection Flux ,c qc = h(T , T1 )
6 We now apply Eq. 11.12 Z
w(div q , Q)d
= 0 (12.4)
Z Z Z Z
(rw)T qtdA = wqf td, + wqn td, , wQtdA (12.8)
A ,q ,T A
where along ,q the
ux (which may include concvection) is known but the temperature is
not, and along ,T , the temperature is known but
ux is not. This equation is valid for any
constitutive relation and is analogous to Eq. 11.36.
10 Applying Fourrier Law, we obtain
Z Z Z Z
(rw)T DrTtdA = , wqf td, , wqn td, + wQtdA (12.9)
A ,q ,T A
This is the weak form of our problem.
11 By extension, the weak form in 3D would be
Z I I Z
(rw)T DrTdV = wqf dS , wqndS + wQdV (12.10)
V Sq ST V
1R R H
A div qdA = , q
T nd, , R R
A( r)T qdA
Victor Saouma Finite Elements
12.2 FE Discretization 12{3
12.2 FE Discretization
Draft
rT = b @T @T cT (12.12)
@x @y
furthermore, the
ux qn at the boundary is given by qn = qT n.
14 Substituting
rT = rNT = BT (12.13)
where "
@Ni #
B = @N @x
i (12.14)
@y
15 Note, this denition of the B matrix for scalar eld problems, should not be confused with
the one for vector problems, Eq. ??
2 @Ni 0
3 2 @Ni 0 3
@x @
B = 64 0 @Ni 7
= J,1 64 0 @Ni 7
(12.15)
@y 5 @ 5
@Ni @Ni @Ni @Ni
@y @x @ @
12.2.1 No Convection
16 Substituting the previous set of equations, Eq. 12.9 reduces to
Z Z Z Z
(rw)T DBtdA T =, wqf td, , wqntd, + wQtdA (12.16)
A ,q ,T A
19 But since this expression should hold for any arbitrary cT matrix, we conclude that
KeT = fZbe + fQe (12.21)
Ke = ABT DBtdA (12.22)
Z Z
fbe = , NT qf td, , NT qntd, (12.23)
,q ,T
Z
fQe = , NT QtdA (12.24)
A
12.2.2 Convection
20 Next we consider the eet of the convection term, and thus
, = ,q + ,T + ,c (12.25)
The presence of the convection term aects only fb in Eq. 12.24 which now becomes
Z Z Z
fb = , NT qf td, , NT qntd, , NT qctd, (12.26)
,q ,T ,c
The outer temperature is T0 = 20o C , convection heat transfer takes place on the inner surface
of the wall with the
uid temperature T1 = 800o C , the lm coecient (or convection term)
h = 25 W/m2 :oC. We seek the temperature distribution across the wall.
Victor Saouma Finite Elements
12{6 FINITE ELEMENT DISCRETIZATION OF THE FIELD EQUATION
Solution:
Draft
2 ,1 B Bd(
T
)
= kL
2 Z 1
1 ,1 h
,1; 1 d
i
2 L ,1 1
" #
k 1
= L ,1 1 , 1 (12.37-a)
"
1 2 # "
1 2 #
20
K1 = 0:3 ,11 ,1 = 66:67 ,66:67 (12.37-b)
1 ,66:67 66:67
"
2 3 # "
2 3 #
K2 = 030 1 ,1 = 200:00 ,200:00 (12.37-c)
:15 ,1 1 ,200:00 200:00
Victor Saouma Finite Elements
12.3 Illustrative Examples 12{7
3 4 # " 3 4 #
Draft
"
K3 = 050 1 ,1 = 333:34 ,333:34 (12.37-d)
:15 1 1 , ,333:34 333:34
2
1 2 3 4 3
66:67 ,66:67
K = 664
6
, 66 : 67 66 : 67 + 200:00 ,200:00 7
Element Convection Matrix: Since convection occurs only at one point, 1, we just add the
convection term h = 25 to K1;1
2
1 2 3 4 3
66:67 + 25 ,66:67 0: 0:
6 ,66:67 266 : 67 , 200 : 00 0: 777
6
K = 64 0: ,200:00 533:34 ,333:34 5
0: 0: ,333:34 333:34
2
1 2 3 4 3
91:67 ,66:67 0: 0:
= 664 ,660::67 ,266 :67 ,200:00 0: 777
6
4. In this case, one of the essential boundary conditions is non-zero, hence following a
procedure similar to the one outlined in Eq. this reduces to
8 9
2 38 9 8 9
81:67 ,66:67 0: >
< T1 ? >
= >
< 20; 000: >
=
>
>
>
<
>
>
>
0 (12.41-a)
=
6
4 ,66:67 266:67 ,200:00 7
5 T2 ? = 0 , 20: > 0 >
0: ,200:00 533:34 >
: T3 ? >
; >
: 0 >
; >
>
:
>
,333:34 >
;
8 9
>
> >
>
> >
=
<
20; 000: =
(12.41-b)
>
>
>
0 >
>
>
:
6; 666:80 ;
We seek to determine the temperature distribution and amount of heat transfered from the n
to the air at 20o C where the lm coecient h=9 W/mo C.
23 Using a three element discretization, and assuming the tip of the n to be isolated.
Solution:
"
1 2 #
K1 = 0360 1 ,1 (12.44-b)
:033 ,1 1
"
2 3 #
K2 = 0360 1 ,1 (12.44-c)
:033 ,1 1
24 Substituting, the element convection stiness matrix (dropping the t term which cor-
responds to the third dimension) will thus be
Z Z L
Kec = ,c
hN Nd, = h NT Nd,
T
0
(12.46)
Given that we have two surfaces across which convection occurs, and that this term stems
from the convection heat loss in an element of thickness t, we rewrite the previous equation
as
Kc = 2th NT Ndx
Z L
(12.47)
0
Victor Saouma Finite Elements
12.3 Illustrative Examples 12{11
but, from Eq. 12.35-d dx = L2 d , thus
Draft
Kc = hLt
Z 1
NT Nd (12.48)
,1
Inserting the term from Eq. 12.45-d we nally obtain
" #
Kc = hL 2 1
3t 1 2 (12.49)
25 As for the conduction stiness matrix, the assembled convection stiness matrix will
be
1 2 3 432
2 1
(9)(3 : 33 10 ,2 ) 6
61 4 1
7
Kc = 3(10,3 ) 6
4
7
1 4 1 75 (12.50)
1 2
Element heat rate vector: Can be determined from
Z Z Z
fb = , N qf td, , N qntd, + T1 NT htd,
T T (12.51)
,q ,T ,c
8 9
> 1 >
(20)(9)(3:33 10 ),2 >
>
<
2
>
>
=
8 9
>
>
>
P1 ? > >
>
<
12; 000: = (12.54-b)
>
>
>
12; 000: >
>
>
:
6; 000: ;
2 38 9 8 9
11; 000: ,10; 700: 0: 0: >
> 235: >> >
> P1 ? > >
7>
6 ,10; 700:
6
22 ; 000 : , 10 ; 700 : 0: 7
<
T2 ? = 12; 000: >
>
= >
< =
(12.54-c)
6
4 0: ,10; 700: 22; 000: ,10; 700: 7
5>>
>
T3 ? >>
>
>
>
>
12; 000: >
>
>
0: 0: ,10; 700 11; 000: :
T4 ? ; :
6; 000: ;
As in the previous example, we rewrite this last equation as
2 38 9 8 9 8 9
22; 000: ,10; 700: 0: < T2 ? >
> = < 12; 000: >
> = < ,10; 700 >
> =
6
4 , 10 ; 700: 22; 000: , 10; 700 : 7
5 T 3 ? = 12 ; 000 : , 235 : 0 :
0: ,10; 700: 11; 000: > : T ? >
4
; >
: 6; 000: > ; >
: 0: > ;
8 9
< 2:514 10 >
> 6
=
= > 12; 000: > (12.55-a)
: 6; 000: ;
Solving, we obtain 8 9 8 9
< T2 >
> = < 208:77 >
> =
T = 194 :25 oC (12.56)
> 3 > > >
: T4 ; : 189:50 ;
Heat loss in the n can be computed from H = Pe H e where H e = h(Tavg , T1)As sand the
surface area of each element is As = 2(1 0:0333)m2
Victor Saouma Finite Elements
12.4 Comparison Between Vector and Scalar Formulations 12{13
Draft
Scalar Vector
Conduction Solid Mechanics
I Dierential
T
Equation
div q + Q = 0 L +b = 0 Balance/Equilibrium
T u State variable
rT = Lu \kinematic"
q Flux vector
qn = qT n t = n Flux at boundary
q = ,DrT = D Constitutive law
T on ,T u on ,u Essential BC
qn on ,q [ ,c t on ,t Natural BC
Z Z IIZ Weak Formulation
Z
wT (L)d
= wT td, BT d
= NT td, Discretized Weak Form
Z,
Z,
+ wT bd
+ NT bd
Chapter 13
TOPICS in STRUCTURAL
MECHANICS
13.1 Transformations
13.1.1 Vectors
1 We transform an arbitrary vector V, from (x; y; z ) coordinate system to (x0 ; y0 ; z 0 ), Fig. 13.1:
through the following transformation
8 9 2 38 9
>
< Vx0 >
= lx0x lx0y lx0z >
< Vx >
=
>
Vy0 >
= 4 ly0 x ly0 y ly0 z
6 7
5
>
Vy >
(13.1)
: Vz0 ; lz0x lz0 y lz0z : Vz ;
| {z }
,
where lij is the direction cosine of axis i with respect to axis j , Fig. 13.2, and thus the rows
of the matrix correspond to the rotated vectors with respect to the original ones corresponding
to the columns.
lij direction cosines of rotated axis i with respect to original axis j .
l1j = (l11 ; l12 ; l13 ) direction cosines of x
l2j = (l21 ; l22 ; l23 ) direction cosines of y
l3j = (l31 ; l32 ; l33 ) direction cosines of z
i.e. l11 = cos , l12 = cos , and l13 = cos
13{2 TOPICS in STRUCTURAL MECHANICS
Draft
5 From the strain denition in terms of displacement derivatives, and and collecting terms, we
obtain
2
l12 m21 n21 l1 m1 m1 n1 n1 l1 3
6 l2 m22 n23 l2 m2 m2 n2 n2 l2 77
6 2
6
0 = 6
6
l32 m23 n23 l3 m3 m3 n3 n3 l3 777
6 2l1 l2 2m1 m2 2n1 n2 l1 m2 + l2 m1 m1 n2 + m2 n1 n1 l2 + n2 l1 7 (13.7)
6 7
4 2l2 l3 2m2 m3 2n2 n3 l2 m3 + l3 m2 m2 n3 + m3 n2 n2 l3 + n3 l2 5
2l3 l1 2m3 m1 2n3 n1 l3 m1 + l1 m3 m3 n1 + m1 n3 n3 l1 + n1 l3
| {z }
,"
9 It should be noted that this operation is only needed for anisotropic constitutive models.
13.1.5 Stiness Matrix
10 From d0 = ,d and p = ,T p0 substituting
kd = p = ,T p0 = ,T k0d0 = ,T k0,d (13.13)
or
k = ,T k0 , (13.14)
where the subscripts r and c refer to retained and condensed d.o.f. respectively.
13 For substructuring, subscripts r and c refer to interface and interior d.o.f. Hence,
" #( ) ( )
krr krc dr = pr (13.16)
kcr kcc dc pc
or
k0 dr = p0 (13.17)
where
k0 = krr , krck,cc1 kcr (13.18-a)
p0 = p0r , krck,cc1 pc (13.18-b)
14 Noting the common form of the above two equations, we deduce that condensation may
also be viewed as a transformation from a d; p; k system to a dr ; p0 ; k0 through the following
transformation
k0 = ,Tc k,c (13.19-a)
p0 = ,Tc p (13.19-b)
where ( )
,c = ,kccI kcr
,1
(13.20)
15 k0 is a r by r matrix, whereae k was r+c by r+c (yet we still have to separately decompose
kcc which is c by c.
1. Insert the exact (analytical) solution for all the nodal displacement in d and verify that
Kij dj , pi = 0 (13.21)
2. Insert the exact (analytical) solution for all nodal displacements on the boundaries of the
patch and internal node displacement di is found from
di = Kii,1 (pi , Kij dj ) i 6= j (13.22)
3. Specify the minimum number of essential (displacement) B.C., and apply natural (trac-
tions) B.C. Solve for
Kd = p (13.23)
and compare with the exact solution.
where B is the statics (or equilibrium) matrix, relating external nodal forces to internal forces;
d is a
exibility matrix, and d,1 is its inverse or reduced stiness matrix. We note that the
stiness matrix is obviously singular, since the second \row" is linearly dependent on the rst
one (through [B]) and thus, its determinent is equal to zero.
23 The reduced stiness matrix, which is the inverse of a
exibility matrix, is not.
U = 12 buc fFg
= 12 buc[K] fug (13.25)
25 If we consider a system where the load fFg applied to each node is proportional to the
element nodal displacement fug by a factor , we have:
[K] fug = fug (13.26)
or
([K] , [I]) fug = 0 (13.27)
26 This is by denition an eigenproblem. There will be as many eigenvalues i as there are
degrees of freedom (or rows in [K]). To each eigenvalue i corresponds an eigenvector fugi .
27 If the eigenvectors are normalized such that:
1. In a rigid body motion, all nodes displace by the same amount, and there are no internal
strains. Hence in a rigid body motion the strain energy U (and thus corresponding )
must be equal to zero.
2. There should be as many zero eigenvalues as there are possible independent rigid body
motions (i.e. number of equations of equilibrium). For a two dimensional Lagrangian
element, there should be three zero eigenvalues, corresponding to two translations and
one rotation.
3. Too few zero eigenvalues is an indication of an element lacking the capability of rigid body
motion without strain.
4. Too many zero eigenvalues is an indication of undesirable mechanism (or failure).
5. Eigenvalues should not change when the element is rotated.
6. Similar modes (such as
exure in two orthogonal directions) will have identical eigenvalues
(for isotropic material).
7. When comparing the stiness matrices of two identical elements but based on dierent
formulations, the one with the lowest strain energy (tr[K] = i) is best.
29Hence, the element stiness matrix will have:
Order: corresponds to the number of degrees of freedom (i.e size of the matrix).
Rank: corresponds to the total number of linearly independent equations which is equat to
the order minus the number of rigid body motions.
Rank Deciency: would be equal to the total number of zero eigenvalues minus the rank.
30 For a square bilinear element the non-zero eigenvectors are shown in Fig. 13.4.
13.4.3 Order of Integration
13.4.3.1 Full Integration
31For numerically integrated elements, a question of paramount importance is the order of the
numerical integration. If the order of integration is too
6 7 8
Figure 13.4: Eigenvectors Corresponding to a) Non-Zero and b) Zero Eigenvalues for a Square
Bilinear element
High: it would certainly result in an exact formulation, which however may be computationally
expensive.
Low: Then we may have rank deciency.
Hence, we ought to properly select the order of numerical quadrature.
32 We recall that
34 Those orders of approximations are only valid for parallelograms, as the edge becomes in-
clined, then the error increases.
13.4.3.2 Reduced Integration
35 Reduced integration, i.e. using fewer Gauss points than required for an exact integration,
will cause the element to soften because higher order polynomial terms may vanish at the points
of lower order rule.
Table 13.1: Full and Reduced Numerical Integrations for Quadrilateral Elements
.
39 If the element is articially softened, then there will be spurious zero energy modes, and the
stiness matrix will be rank decient, i.e. its rank is less than the number of element d.o.f.
minus the number of rigid body modes.
40 The independent displacement modes of a bilinear element are shown in Fig. 13.5. First
4 3
1 2 3 4
1 2
5 6 7 8
Q8
13.4.3.3 Selective Reduced Integration
44 In selective integration, we integrate the dierent strain terms with dierent orders of inte-
gration.
45 More about this technique in subsequent chapter.
48 From this table we note that parasitic shear contributes to the strain energy stored in the
element (which is equal to PT :d). Hence for a given imposed displacement, the forces would
be larger and M1 > M2 .
49 Computing the ratio of strain energies of the two elements, we obtain
" #
M1 = 1 1 + 1 a 2 (13.31)
M2 1 + 1 + 2 b
Victor Saouma Finite Elements
13.5 Parasitic Shear/Incompatible Elements 13{13
Draft
Figure 13.8: Displacements Associated with Incompatible Modes for the Q6 Element
hence, as a=b increases, the element \locks" (a term which will be explained in later chapters).
13.5.2 Q6, The Solution
50 Comparing the displacement elds in Table 13.2 we observe that the bilinear element errs
from the exact solution by omitting the displacement mode associated with (1 , 2 ) and (1 , 2 ).
51 We can remedy to this situation by adding the missing modes as internal nodes
X
u = Niui + (1 , 2 )a1 + (1 , 2 )a2 (13.32-a)
X
v = Ni vi + (1 , 2 )a3 + (1 , 2 )a4 (13.32-b)
where the Ni are given by Eq. 13.30, and ai are nodeless d.o.f. The element now has 6 degrees
of freedom, and hence the element is referred to as Q6.
52 The B matrix would have to be correspondingly adjusted. and following its formulation, the
additional d.o.f. ai are eliminated through condensation (Sect. 13.2).
53 The Q6 element is incompatible (or nonconforming) due to the square distribution of u or v
along the edge which is dened by only two nodes, Fig. 13.8.
54 The softening in this case counters the inherent overstiness of the standard assumed dis-
placement formulation.
13.5.3 QM6, Further Enhancements
55 It can be shown that the Q6 element passes the patch test only if it is rectangular.
56 To remedy to this deciency, we consider that portion of the B matrix associated with the
nodeless d.o.f. and label it Ba , and then the corresponding consistent nodal load vector is
Z e
pa = ,
BTa 0 d
e (13.33)
13.7 Constraints
Chapter 14
PLATES
Adapted from Pilkey & Wunderlich
1 This chapter will cover the transverse deformation of plates. The approach followed will be
consistent with the nite element formulation, Fig. 14.1.
Fundamental Relations
Draft
?
Dierential Equation
?
Variational Formulation
?
Finite Element Discretization
3 Note that in plate theory, we ignore the eect of the membrane forces, those in turn will be
accounted for in shells.
Victor Saouma Finite Elements
14.1 Fundamental Relations 14{3
4 The equation of equilibrium is derived by considering an innitesimal element tdx dy subjected
Draft
to an applied transverse load pz . We would have to consider three equations of equilibrium,
Fig. 14.3:
16The 12(1Et,3 2 ) term is referred to as the
exural rigidity and is analogous to the
exural stiness
EI of a beam (if the plate has unit width, and = 0, then EI = Et3=12).
14.2 Kirchho
17 This theory, is primarily applicable to thin plates in which shear deformations can be ne-
glected.
14.2.1 Fundamental Relations
18In this formulation, shear deformations are neglected, and formulation is analogous to the
conventional Euler-Bernouilli beam theory.
Kinematic Relations Since shear deformations are neglected,
xz =
yz = 0 and thus the
last two relation of Eq. 14.10 reduce to
xx = ,w;x and yy = ,w;y (14.17)
and the rst three strains become
xx = ,z @@xw2 ; yy = ,z @@yw2 ; @2w ;
2 2
xy = ,2z @x@y (14.18)
or 8 9 8 9
>
< xx >
=
>
>
< , @x@222 >
>
=n o
yy = > , @y@ 2 w
>
: 2xy >
; : ,2 @ @
> >
>
; | {z } (14.19)
| {z } |
@x @y
{z }
u
L
Constitutive Relation The constitutive relation of Eq. 14.16 still applies
8 9 2 38 9
< Mxx >
> = Et 6 1 1
3 0 > < xx > =
M yy = 12(1 , 2 ) 4 0 0 0 75 > yy >
(14.20)
>
: M >
; 1, : 2 ;
xy 2 xy
| {z } | {z }| {z }
M D
(Note the similarity with the corresponding equations for beam
exure ddx2 M2 , Vx = 0)
14.2.2 Dierential Equation
19If we combine the kinematic with the constitutive relation, 14.19 and 14.20 respectively we
obtain 2
8 9 @ 22 + @ 22 3
>
< M xx =
>
Et3 66 @x@ 2 @y2 7 n o
>
M yy >
= , 12(1 , 2 ) 4 @y 2 + @ 7
@x2 5 w (14.22)
: M
xy ;
(1 , ) @x @y
@ @
20 Finally, we substitute the equilibrium equation, 14.21, into the previous one,
@4w + 2 @4w + @4w = pz (14.23)
@x4 @x2 @y2 @y4 Et3 2
12(1, )
or
r4 w = pz (14.24)
Et3
12(1, 2 )
Note the similarity with the corresponding equation for beams
@ 2 EI @ 2 w = p (14.25)
@x2 @x2 z
14.2.3 Stresses
21Combining the stress-strain relation of Eq. 14.14, with Eq. 14.18, and 14.22, the stresses
could be expressed in terms of the moments
xx = M xxz
t3=12 ; yy = M yy z
t3 =12 ; xy = M xy z
t3 =12
(14.26)
again we note the analogy with the
exural stress expression in beams = My
I . Using the three
dimensional equilibrium equation, Eq. 3.30:
@xx + @xy + @xz = 0
@x @y @z
Victor Saouma Finite Elements
14.2 Kirchho 14{9
@yx + @yy + @yz = 0 (14.27-a)
Draft
@x @y @z
@zx + @zy + @zz = 0
@x @y @z
integrating the rst equation yields
xz = ,
Z z @xx + @xy dz = Z t=2 12z @Mxx + @Mxy dz
t=2 @x @y z t3 @x @y
Z t=2
= 12
t3 zVxdz (14.28-a)
z
Finally
2 " # 2 " #
xz = 32Vtx 1 , 2tz and yz = 32Vty 1 , 2tz (14.29)
Thus the shear stress distribution across the plate thickness is parabolic (though they tend to
be very small compared to xy , and the peak shear stresses occur at the middle surface (z = 0)
where
xz jmax = 23 Vtx and yz jmax = 32 Vty (14.30)
Finally, it can be shown that zz varies cubically.
14.2.4 Variational Formulation
22 Prior to the nite element discretization, we seek to obtain from the previously derived
relations a variational formulation of the problem.
23 The internal virtual work is given by
Z Z
Wi = , t dA = , t MdA (14.31)
A A
where M = D is obtained from Eq. 14.20, and = Lu is obtained from Eq. 14.19.
24 Accounting for the external virtual work, we obtain
Z Z Z
W = Wi + We = , ut (LT DL)udA + ut p V dA + utpd, = 0 (14.32)
| A {z } A ,p
| {z }
where pv contain the applied transverse loading pz , and p the edge loading ps.
25 Substituting for the LT DL term we obtain
2 38 9
1 0 >
> , @x@222 >>
,2 @x@ @y@ c 12(1Et, 2 ) 64 1
3 < =
kL = LT DL = b , @x@22 , @y@22 0 7
5 , @y@ 2 >
0 0 1, >
2
>
: ,2 @x@ @y@ >
;
Victor Saouma Finite Elements
14{10 PLATES
= K
Draft
" ! !#
@ 2 @ 2 + @ 2 + 2(1 , ) @ @ @ @ + @ 2 @ 2 + @ 2 (14.33-a)
@x2 @x2 @y2 @x @y @x @y @y2 @y2 @x2
and nally, noting that u = w, the virtual work becomes
Z Z
W = , w (KLwpz ) dA + ,p wps d
= 0 (14.34)
A ,
14.3 Reissner-Mindlin
26This theory, is primarily applicable to thick plates in which shear deformations can not be
neglected.
14.3.1 Fundamental Relations
27In this formulation, shear deformation is accounted for and is the extension of the Timo-
shenko's beam theory.
Kinematic: The kinematic relation of Eq. 14.12 still holds and is written in slightly modied
form
8 9 2 @ 3
< xx >
> @x 0 ( )
=
yy > = 4 0 @y 5
6 @ 7 xx (14.35-a)
>
: 2 ; @ @ yy
xy @y {z @x } | {z }
| {z } |
Lf
8 9
(
xz
) "
@ 1 0
#>
< w >
=
= @x
@ xx (14.35-b)
yz @y 0 1 >
: >
yy
;
| {z } | {z }
Ls | {z }
u
A f t2 A s
Z Z
+ ut pV dA + ut pd, = 0 (14.42)
A ,
kLf = Lf Df Lf
t (14.43)
kLs = LtsDsLs (14.44)
where ks = 5=6 is the shear correction factor
Z
W = b w xx yy c
A
2 @ 5t @ @ 5t @ @ 5t 5t @ 3(
@x 6 @x + @y 6 @y @y 6 6 @x w )
@ 5 t @ @ @ K (1, ) @ 5 t @ @ @ K (1, ) @
4 @y 6 @x K @x + @y 2 @y + 6 @x K @y + @y 2 @x 5 xx dA
5t @ @ K @ + @ K (1, ) @ @ K @ + @ K (1, ) @ + 5t yy
6 @x @y @x @x 2 @y @y @y @x 2 @x 6
( ) ( )
Z pz Z ps
b w xx yy c 0 dA + b w xx yy c 0 dA = 0 (14.45-a)
A 0 , 0
14.3.4 Summary
31 Table 14.1 summarizes some of the major equations governing plate bending, and contrasts
them with the equivalent elasticity ones.
Victor Saouma Finite Elements
2 @
Elasticity
3 Kinematic
Plate Theory (Kirchho/Mindlin)
Draft
8 @x 0 0
> "xx 9 0 @ 0 ( )
=
< "yy > 6 @y @ 7 ux xx
@ 22
, @x
"zz 0 0 @z 7 @
" xy =6
6 @ @ 0 7 uy yy = , @y22 w
>
: "xz >; 4 @y
@ @x @ 5 uz 2xy @ | {z }
"yz @z 0 @x | {z } | {z } ,2 @x
@ @y
u
| {z } 0 @ @ u
| {z }
@z @y L
| {z }
L #
Victor Saouma
" @
xx @x 0 n on o h @ i w
yy = 0 @ xx ;
xz = @x 1 0 xx
@ @y
@ yy
yz @
@y 0 1
2xy @y @x | {z } | {z } | {z
yy
| {z } | {z }
}| {z }
Ls u
14.3 Reissner-Mindlin
Lf
8 9 Equilibrium 9
" @ @ @ # > xx > " @ @ # 8 Mxx
@x 0 0 @y @z 0 < yy = bx @x 0 @y ,1 0 < Myy = 0 0
0 @
@y 0 @
@x 0 @
@z
zz =0 by 0 @
@y @
@x 0 ,1 Mxy + 0 = 0
@ @ @ > xy + bz @ @ : Vx ; pz 0
0 0 @z 0 @x @y : xz >
; | {z } 0 0 0 @x @y Vy
| {z } yz | {z } | {z }
LT | {z } b LT
M 9
" @
0 @ ,1 0
# 8 Mxx
= 0
@x @ @y
@
< Myy 0
0 @y 0 ,1
@x Mxy + 0 = 0
0 0 @x 0
@ @y@ : Vx ; pz 0
| {z }| Vy
{z }
LT =b LTf LTs c M
8 2 1 Constitutive
> xx 9 0 0 0 38 "xx 9
=
< yy > 1 0 0 0 > =
< "yy > Mxx 1 0 xx
zz = 6 1 0 0 0 7 "zz Myy =K 1 0 yy
xy 4 0 0 0
0 1,
> ;
: yz > 0 0 0 0
0 5 > "xy >
0 : "yz ; Mxy 0 0 2 2xy
zx 0 0 0 0 0
"zx | {z } | {z }| {z }
| {z } | {z }| {z } M D
D 8 9 2 38
xx 1 0 xx 9
< M
Myy = K 1 0 0 < yy =
=6 0 0 1, 7
Mxy 4 2 h i 5 2xy
: Vx ; 1 0 :
xz ;
Vy 0 0 1
yz
| {z } | {z }| {z }
M D
= E(1, ) 1,2
= 2(1 K= Et3 = 65 t
(1+ )(1,2 ) ; = 1, ;
, ) 12(1, 2 ) ;
Virtual Work Z Z
R R R
=
(Lu)T Dd
,
uT bd
, ,t
uT ^td, = 0 W = , w (KLwpz ) dA +
,p wps d
= 0
Z A Z , Z Z
W = ,K t kLf dA + 6ks (1t2, ) ut kLs udA + ut pV dA + ut pd, = 0
A A A ,
14{13
Finite Elements
Table 14.1: Comparison of Governing Equations in Elasticity and Plate Bending
14{14 PLATES
14.4 Finite Element Formulations
Draft
32Numerous elements have been proposed for plate bending. An ideal plate element should
have the following properties:
1. Formulation should be based on continuum mechanics and plate theory, the nodal d.o.f.
are the transverse displacement w, and sectional rotations xx and yy .
2. The element should be \numerically correct", and convergent. The element stiness
matrix must contain the three rigid-body modes and no rank deciency (spurious zero-
energy modes).
3. The element should not \lock" in thin plate analyses.
4. The predictive capability of the element should be insensitive to element geometric dis-
tortions.
33 Since plate structure is a special case of shell structures, and plates under large displacement
exhibit shell characteristics (through membrane actions), similar formulations can be employed
for both plates and shells.
14.4.1 Rectangular Element
14.4.1.1 Formulation
34 The stiness matrix, including shear deformation eects, for a rectangular element, Fig. ??
will be developed.
@ = 1 @ and @ = 1 @ (14.49)
@x a @ @y b @
thus substituting Eq. 14.35-a
2 1 @ 3
a @ 0
4 0 1 @ 5 (1 , )(1 , ) (1 , ) (1 , ) 0 0 0 0
1 @
b1 @
@ 0 0 0 0 (1 , )(1 , ) (1 , ) (1 , )
|
b @ {z a @ } | {z }
N
Lf
2 1, 1, 3
, a a a , a 0 0 0 0
=4 0 0 0 0 , 1,b , b 1, 5 (14.50-a)
, 1,b , b 1, , 1,a 1,a b b
, a
|
b b {z a }
B
39 We note that Lf = Lu (used in elasticity) for in plane deformation, and Df = 12t2 D, thus
the stiness matrix for bending can be formed directly from the stiness matrix for in-plane
deformation.
40 Upon substitution and integration we obtain
kf = 24(1Et, 2 ) 12
t 2
= 56 t (14.52-c)
we obtain
Z Z
K ut kLs udA = K (Lts u)t ILs udA
A Z A
= K (Lt Nu)t Ls NudA
AZ s
= Kut Bts Bs dAu (14.53-a)
A
Recalling that Bs = Ls N, and substituting from Eq. 14.35-b for Ls we obtain
Z
ks = Bts BsdA (14.54-a)
" #
Bs = B 11 B12
B21 0 B23
0 (14.54-b)
Note that the rst row and second row in the second equations correspond to
xz and
yz
respectively.
42 Upon substitution,
h i
B11 = , (1,a ) (1,)
a
a , a (14.55-a)
Victor Saouma Finite Elements
14.4 Finite Element Formulations 14{17
h i
B12 = (1 , )(1 , ) (1 , ) 1 , (14.55-b)
Draft
h i
B21 = , (1,b ) , b b (1,b ) (14.55-c)
h i
B12 = (1 , )(1 , ) (1 , ) (1 , ) (14.55-d)
43In order to facilitate the integration of ks , a one point Gauss integration = = 0:5 is used.
Upon substitution
" #
Bs =
c , 21a 21a 21a , 21a 1
4
1
4
1
4
1
4 0 (14.56)
, 21b , 21b 21b 21b 0 1
4
1
4
1
4
1
4
45 Through inspection of the previous equation, and noting that = 6ks (1t2, ) , we observe that
for very thin plates is very large, hence as t ! 0; ! 1. Hence, unrealistic u = 0 can be
obtained independently of the load. This phenomena is called locking (of the displacements).
46 Locking occurs also in thin beam elements when shear deformation eects are accounted for.
47 To alleviate the locking problems is to make Ks singular so that Ks is nite. This can
be accomplished through reduced numerical integration for ks , Table 14.2, which in turn will
reduce the rank of the matrix.
Integration Rule
Full Reduced Selective
kf ks kf ks kf ks
Bilinear 4 nodes, 12 d.o.f. 2x2 2x2 1x1 1x1 2x2 1x1
Quadratic 9 nodes, 27 d.o.f. 3x3 3x3 2x2 2x2 3x3 2x2
Serendipity 8 nodes, 24 d.o.f. 3x3 3x3 2x2 2x2 3x3 2x2
48The same phenomena occurs for beams. This is illustrated by evaluating the de
ection of a
cantilever beam, due to both
exural and shear deformation, under a point load
w = PL
3
+ PL = PL3 1 + 3EI (14.61)
EI ks A EI ks AL2
or 3
!
w = PL 1 + 3 ; = ks AL2 = k L 2 = O L2 (14.62)
EI EI sE r h2
Hence, as the slenderness ratio increases, shear deformation decreases. Furthermore, Euler
theory is recovered if the shear stiness ! 1, however shear locking can occur for small .
49 Shear locking eects are aggravated by large element distortions.
56 Hence
w = w1 + w2 = Nu wb (14.68)
where Nu and wb are dened in Eq. 14.64 and 14.67-a.
57 The unknown quantities w bi can in turn be expressed in terms of the nodal unknown dis-
@x and yy = , @y in Eq. 14.68.
placements by substituting w, xx = , @w @w
u = Nb uwb (14.69)
where u = b w1 x1 y1 w3 x3 y3 c thus wb = N
b ,1 u = Gu and
u
w = NuGu = Nu (14.70)
with w = N1 u1 + N2 u2 + N3 u3 where u1 = b wi xxi yyi cT
59 Upon substitution, it can be shown that
8 9T
>
< L1 + L21 L2 + L21 L3 , L1 L2 , L1 L23 >
=
N1 = > ,b3(L21L2 + 12 L1 L2L3) + b2(L21 L3 + 12 L1 L2L3) (14.71)
: c3 (L21 L2 + 21 L1 L2 L3 ) , c2 (L21 L3 + 12 L1 L2 L3 ) >
;
14.4.2.2 Nonconformity
61 Considering Fig. 14.8
6y 1
Draft
2 3
- x
2 3
64 This nonconforming element, nevertheless is a viable one as the 3 rigid body modes, and
constant strains are present. In addition, it passes the patch test.
65 Despite its nonconformity, this element is quite appealing because the deformation is ex-
pressed only in terms of the middle surface de
ection w.
14.4.3 Discrete Kirchho Triangle
66 From the previous formulation, it is apparent that it is dicult to formulate a compatible
triangular element with 9 d.o.f. using a single polynomial approximation for w.
67 A viable approach consists in starting with the thick plate, or Reissner-Mindlin plate theory,
by separating the d.o.f. for translation and rotations and making them independent one from
the other. These in turn are made continuous at the inter-element boundaries, thus ensuring
C0 continuity. Since the plate is very thin, the term corresponding to shear deformation in the
virtual work will be neglected. The shape functions will be designed to ensure compatibility,
thus the element will be conforming. Finally, the Kirchho plate theory assumptions will be
only introduced at discrete points along the element boundaries.
68 The starting point will be Eq. 14.40-c, Fig. 14.9
Z Z
,Wi = t Ltf Df Lf dA + ut LtsDs Ls udA (14.78)
A A
where the rst term corresponds to
exure, and the second one to the eect of shear which will
be neglected.
69 Since for bending, only the rotations appear in the formulation, the rotations can be approx-
imated by
X6 X6
xx = Ni xi; yy = Ni yi (14.79)
i=1 i=1
70 The shape functions Ni for the natural coordinates with quadratic expansions are
N1 = L1 (2L1 , 1) = (1 , L2 , L3 )(1 , 2L2 , 2L3 )
Victor Saouma Finite Elements
3
u
y Formulations
14.4 Finite Element 14{23
u
6 6 u
5
Draft
u
1 u4
u
2
- x
[0,60] N2 = L2 (2L2 , 1)
N3 = L3 (2L3 , 1)
N4 = 4L2 L3
N5 = 4L1 L3 = 4(1 , L2 , L3 )L3
N6 = 4L1 L2 = 4(1 , L2 , L3 )L2 (14.80-a)
71 For this element, so far we have a total of 12 dof, and we seek to formulate it in terms of only
9 dof, 3 at each corner node corresponding to wi ; ixx and iyy thus we need to impose certain
constraints for this reduction.
72 Hence, Kirchho assumptions will be selectively applied:
1. Zero transverse shear strains at the corners: or from Eq. 14.35-b
8 9
( ) "
@ #> w > ( ) ( ) ( )
xz 1 0 0 xx = , w;x
< =
yz = @x
@ 0 1 >
xx >
= 0 ) yy ,w;y (14.81)
@y : yy ;
76 Substituting the last three equation into Eq. 14.79 at each node to obtain xi and yi to nd
the shape functions results in
xx = Nxvi ; yy = Ny vi (14.89)
Victor Saouma Finite Elements
14.4 Finite Element Formulations 14{25
or
Draft
8 9
>
>
>
w1 >
>
>
>
>
>
>
>
1x >
>
>
>
>
>
>
>
>
>
1y >
>
>
>
>
( i ) " >
#>
> w2 >
>
>
= N x1 Nx2 Nx3 Nx4 Nx5 Nx6 Nx7 Nx8 Nx9
< =
xx
2x (14.90)
iyy N N N N N N N N N
y1 y2 y3 y4 y5 y6 y7 y8 y9 > >
| {z } | {z
>
}>
>
>
>
2y >
>
>
>
>
N >
>
>
> w3 >
>
>
>
> >
>
>
>
>
>
3x >
>
>
>
>
:
3y ;
| {z }
v
where
Nx1 = 1:5(a6 N6 , a5 N5 ); Ny1 = 1:5(d6 N6 , d5 N5 )
Nx2 = N1 , b5 N5 , b6N6 ; Ny2 = Nx3
Nx3 = ,c5 N5 , c6 N6 ; Ny3 = N1 , e5 N5 , e6 N6
Nx4 = 1:5(a6 N4 , a5 N6 ); Ny4 = 1:5(d4 N4 , d6 N6 )
Nx5 = N2 , b5 N6 , b6N4 ; Ny5 = Ny3 (14.91)
Nx6 = ,c6 N6 , c4 N4 ; Ny6 = N2 , e5 N6 , e4 N4
Nx1 = 1:5(a6 N5 , a5 N4 ); Ny1 = 1:5(d5 N5 , d4 N4 )
Nx2 = N3 , b5 N4 , b5N5 ; Ny2 = Nx3
Nx3 = ,c4 N4 , c5 N5 ; Ny3 = N3 , e4 N4 , e5 N5
1 2 1 2 1 2 1 2
and ak = , xlij2ij , bk = 4 xijl,ij2 2 yij , ck = 34 xijlij2yij , dk = , ylij2ij , ek = 4 yijl,ij2 2 xij and k = 4; 5; 6
corresponds to sides ij =23, 31, and 12 respectively.
77 With the shape functions thus dened, we can nally determine the stiness matrix from
the principle of virtual work (Eq. 14.40-c). Since shear deformations are neglected, we use only
the rst part.
Z
Wi = T Ltf Df Lf dA (14.92-a)
A | {z }
kLf
Z
= vi T NT Ltf Df Lf NdAvi (14.92-b)
ZA
= vi T BT Df BdA vi (14.92-c)
| A {z }
ke
(14.92-d)
Thus the stiness matrix can be evaluated
78 Major conclusions
1. Two theories: thin plates (Kirchho/Euler bernouilli), and thick plates (Mindlin-Reissner/Timoshenko),
without and with shear deformations.
2. Strong analogy between plates and beams.
3. If membrane actions are accounted for, lead to formulation of shell elements.
4. ) M , " ) .
5. If thick plate theory is applied to thin plates, will have shear locking problems.
6. Shear locking is alleviated by reduced integration of the shear components.
7. FOr thin plate theory, will have non-conforming elements if only 3 d.o.f. are adopted at
each node.
8. The DKT element alleviated this problem.
9. Reduced integration will alleviate problems arising from mesh distortions.
10. Shear locking is primarily an issue for 3 node and 4 node elements. Higher order ex-
pansions preform much better, whereby Lagrangian 6 and 16 node elements perform
signicantly better than their Serenditpity counterparts of 8 node and 12 node elements.
11. The usual 3 DOF/node Kirchho elements do not maintain normal slope continuity.
Chapter 15
GEOMETRIC NONLINEARITY
15.1 Strong Form
1 Column buckling theory originated with Leonhard Euler in 1744.
2 An initially straight member is concentrically loaded, and all bers remain elastic until buck-
ling occur.
3 For buckling to occur, it must be assumed that the column is slightly bent as shown in Fig.
15.1. Note, in reality no column is either perfectly straight, and in all cases a minor imperfection
P P
x
x and y are
principal axes
x
8 This last equation can be satised if: 1) A = 0, that is there is no de
ection; 2) kL = 0, that
is no applied load; or 3)
kL = n (15.6)
,
P = n 2 or
Thus buckling will occur if EI L
P = n L2EI
2 2
9The fundamental buckling mode, i.e. a single curvature de
ection, will occur for n = 1; Thus
Euler critical load for a pinned column is
Pcr = LEI
2
2 (15.7)
cr = E2
2
L
(15.8)
r
where I = Ar2 .
11 Note that buckling will take place with respect to the weakest of the two axis.
12 In the preceding approach, the buckling loads were obtained for a column with specied
boundary conditons. A second order dierential equation, valid specically for the member
being analyzed was used.
13 In the next approach, we derive a single fourth order equation which will be applicable to
any column regardelss of the boundary conditions.
14 Considering a beam-column subjected to axial and shear forces as well as a moment, Fig.
15.2, taking the moment about i for the beam segment and assuming the angle dxdv between the
axis of the beam and the horizontal axis is small, leads to
2
dM ( dx ) dV
M , M + dx dx + w 2 + V + dx dx , P dx dx = 0 dv (15.9)
15 neglecting the terms in dx2 which are small, and then dierentiating each term with respect
to x, we obtain
d2 M , dV , P d2 v = 0 (15.10)
dx2 dx dx2
16 However, considering equilibrium in the y direction gives
dV = ,w (15.11)
dx
17 From beam theory, neglecting axial and shear deformations, we have
d2 v
M = ,EI dx (15.12)
2
18Substituting Eq. 15.11 and 15.12 into 15.10, and assuming a beam of uniform cross section,
we obtain
EI d v4 , P d v2 = w
4 2
(15.13)
dx dx
19 P , the general solution of this fourth order dierential equation to any set
Introdcing k2 = EI
of boundary conditions is
v = C1 sin kx + C2 cos kx + C3 x + C4 (15.14)
20 If we consider again the satbility of a hinged-hinged column, the boundary conditions are
v = 0; v;xx = 0 at x = 0 (15.15)
v = 0; v;xx = 0 at x = L
Victor Saouma Finite Elements
15{4 GEOMETRIC NONLINEARITY
Draft
w(x)
P P
x
dx
y,u
w
M
V
P i V+ x dx
V
v P
x
dx
M+ M
x dx
i i P
P
j
P
j
dx
Figure 15.2: Simply Supported Beam Column; Dierential Segment; Eect of Axial Force P
Pcr = LEI
2
2 (15.17)
which was derived earlier using the lower order dierential equation.
21 Next we consider a column with one end xed (at x = 0), and one end hinged (at x = L).
The boundary conditions are
v = 0; v;xx = 0 atx = 0 (15.18)
v = 0; v;x = 0 atx = L
These boundary conditions will yield C2 = C4 = 0, and
sin kL , kL cos kL = 0 (15.19)
But since cos kL can not possibly be equal to zero, the preceding equation can be reduced to
tan kL = kL (15.20)
which is a transcendental algebraic equation and can only be solved numerically. We are
essentially looking at the intersection of y = x and y = tan x, Fig. 15.3 and the smallest
P , the smallest critical load is
positive root is kL = 4:4934, since k2 = EI
Pcr = (4:4934) 2 EI
2
EI = (15.21)
L2 (0:699L)2
Note that if we were to solve for x such that v;xx = 0 (i.e. an in
ection point), then x = 0:699L.
22 We observe that in using the higher order dierential equation, we can account for both
natural and essential boundary conditions.
15.1.3 Slenderness Ratio
23 For dierent boundary conditions, we dene the slenderness ratio
= lre
q
where le is the eective length and is equal to le = kl and r the radius of gyration (r = AI ).
24 le is the distance between two adjacent (ctitious or actual) in
ection points, Fig. 15.4
6.0
4.0
2.0
0.0
-2.0
-4.0
-6.0
-8.0
-10.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0
Figure 15.3: Solution of the Tanscendental Equation for the Buckling Load of a Fixed-Hinged
Column
15.2 Weak Form
15.2.1 Strain Energy
25 Considering a uniform section prismatic element, Fig. ??, subjected to axial and
exural
deformation (no shear), the axial strain-displacement relation is given by1
!
du
"x = dx , y dx2 + d 2v 1 dv 2 (15.22)
|{z} 2| {zdx }
Axial |Flexure
{z }
Large Deformation
26 we note that the rst and second terms are the familiar components of axial and
exural
strains respectively, and the third one (which is nonlinear) is obtained from large-de
ection
strain-displacement.
27 The Strain energy of the element is given by
Z
U e = 12 E"2x d
(15.23)
1 For an innitesimal segment of initial length dx, subjected to a vertical displacement v;x dx, ds = (1+ v 2 ) 21
2
;x
ds,dx v;x
(1 + 12 v;x2 )dx, the strain will be "= dx 2 . This corresponds to the Green-Lagrange strain which is
"x = u;x + 12 (u2;x + v;x2 + w;x2 ).
Victor Saouma Finite Elements
15.2 Weak Form 15{7
Draft
i.p.
l/4 i.p.
i.p.
<kl<l
kl=l kl= l l l
2
2
l
i.p.
l/4
i.p.
i.p.
l l<kl<
8
i.p. l
l
kl=21
kl=1
Pcr
Pcr Pcr
i.p.
l<kl<
8
i.p.
k=2 k=1
Figure 15.4: Critical lengths of columns
29 Noting that Z Z Z
dA = A; ydA = 0; y2 dA = I (15.25)
A A A
for y measured from the centroid, U e reduces to
2 !2 3
Z
U e = 12 4A dxdu 2 + I d2 v
dv 4 + A du dv 2 5 Edx (15.26-a)
+ A4 dx
L dx2 dx dx
4
We discard the highest order term A4 dx
dv in order to transform the above equation into a
linear instability formulation.
30 Under the assumption of an independent prebuckling analysis for axial loading, the axial
load Px is
Px = EA du
dx (15.27)
Thus Eq. 15.26-a reduces to
2 ! 3
U e = 21
Z
EA du
2
d2 v 2 + P dv 2 5 dx
dx + EI (15.28)
4 x dx
L dx2
31We can thus decouple the strain energy into two components, one associated with axial and
the other with
exural deformations
U e = Uae + Ufe (15.29-a)
2
EA du
Z
Uae = 12 dx dx (15.29-b)
L2 3
!
1 Z
d 2v 2
dv 2
Ufe = 2 4EI dx2 + Px dx 5 dx (15.29-c)
L
32Recall, from Eq. 6.15 that a functional in terms of two eld variables with higher order
derivatives of the form
Z Z
= F (x; y; u; v; u;x ; u;y ; v;x; v;y ; ; v;yy )dxdy (15.30)
There would be as many Euler equations as dependent eld variables, Eq. 6.16
8
@ 2 @F @ 2 @F @ 2 @F
@u , @x @u;x , @y @u;y + @x2 @u;xx + @x@y @u;xy + @y2 @u;yy = 0
< @F
> @ @F @ @F
@ 2 @F @ 2 @F @ 2 @F (15.31)
: @v , @x @v;x , @y @v;y + @x2 @v;xx + @x@y @v;xy + @y2 @v;yy = 0
> @F @ @F @ @F
Substituting into the Euler equation, and assuming constant Px , and EI , we obtain
d4 v , P d2 v = 0
EI dx (15.35)
4 x dx2
2 u1 v1 1 u2 v2 2 3
0 06 L0 0 06 0
6
6
6
0 5 10 0 , 5 10L 2 777
0 10L 152 L2 0 , 10L , L30 77
kg = PL (15.41)
6
6
6 0 06 0L 0 0 0 7
6
6
6
0 , 5 , 10 0 65 , 10L 777
4 0 10L , L302 0 , 10L 152 L2 5
K = Ke + Kg (15.44)
41 We assume that conservative loading is applied, that is the direction of the load does not
\follow" the de
ected direction of the member upon which it acts.
47 Alternatively, it can simply be argued that there is no unique solution (bifurcation condition)
to .
48 The lowest value of , crit will give the buckling load for the structure and the buckling
loads will be given by
Pcrit = critP (15.49)
49 The corresponding deformed shape is directly obtained from the corresponding eigenvector.
2
3
(1) l
5
6 4
l
(2)
8
9
Solution:
2
1 2 3 4 5 6 3
EA
L 0 0 , EA
L 0 0
6
6 0 12EI
L3
6EI
L2 0 , 12LEI3 6EI
L2
7
7
k1e =
6
6
6
0 6EI
L2
4EI
L 0 , 6LEI2 2EI
L 7
7
7 (15.50-a)
6
6 , EA
L 0 0 EA
L 0 0 7
7
6
4 0 , 12LEI3 , 6LEI2 0 12EI
L3 , 6EI 7
L2 5
0 6EI
L2
2EI
L 0 , L2
6EI 4EI
L
2
4 5 6 7 8 9 3
EA
L 0 0 , EA
L 0 0
6
6 0 12EI
L3
6EI
L2 0 , 12LEI3 6EI
L2
7
7
k2e =
6
6
6
0 6EI
L2
4EI
L 0 , 6LEI2 2EI
L 7
7
7 (15.50-b)
6
6 , EA
L 0 0 EA
L 0 0 7
7
6
4 0 , 12LEI3 , 6LEI2 0 12EI
L3 , 6EI 7
L2 5
0 6EI
L2
2EI
L 0 , L2
6EI 4EI
L
Similarly, the geometric stiness matrices are given by
2
1 2 3 4 5 6 3
0 0 0 0 0 0
60
6
6
5 10
L 0 , 65 10L 777
k1g = ,LP
6
60
6
L
10
2 L2
15 0 , 10L , L302 77 (15.51-a)
60 0 0 0 0 0 7
6 7
4 0 ,5 , 10L ,
6 6 0 6 L 7
5 10 5
0 10L , L302 0 , L
10 15 L
2 2
2
4 5 6 7 8 9 3
4 5 6 7 8 9
60
6
6
5 10
L 0 , 65 10L 777
k2g = ,LP , 10L , L302 77
6 L 2 L2
60 0
6
60
10
0
15
0 0 0 0 7 (15.51-b)
6 7
4 0 ,5 , 10L ,
6 6 0 6 L 7
5 10 5
0 10L , L302 0 , L
10 15 L
2 2
The stiness matrices Ke and Kg can now be assembled from the element stinesses. Elim-
inating rows and columns 2, 7, 8, 9 corresponding to zero displacements in the column, we
obtain
1 4 3 5 6
2 AL2 2
I 2 , I2
AL 0 0 0 3
6, I
6 AL
EI 2 ALI 0 0 0 777
Ke = L3 66 0
6
0 4L ,6L 2L2 77
2 (15.52)
4 0 0 ,6L 24 0 5
1 4 5 3 6
1 ,
0 0 0
4 , 2 ,
0 0 0
3 0 0 2 2 , 15 ,6,L + 10 2 + 30 = 0
(15.55)
5 0 0 ,6L + 10 12 2 , 5
,
0
6 0 0
2 + 30 0 4 2 , 15
Expanding the determinant, we obtain the cubic equation in
33 , 2202 + 3; 840 , 14; 400 = 0 (15.56)
and the lowest root of this equation is = 5:1772 .
We note that from Eq. 15.21, the exact solution for a column of length L was
Pcr = (4:4934) (4:4934)2 EI = 5:0477 EI
2
l2 EI = (2L)2 L2 (15.57)
and thus, the numerical value is about 2.6 percent higher than the exact one. The mathematica
code for this operation is:
(* Define elastic stiffness matrices *)
ke[e_,a_,l_,i_]:={
{e a/l , 0 , 0 , -e a/l , 0 , 0 },
{0 , 12 e i/l^3 , 6 e i/l^2 , 0 , -12 e i/l^3 , 6 e i/l^2 },
{0 , 6 e i/l^2 , 4 e i/l , 0 , -6 e i/l^2 , 2 e i/l },
{-e a/l , 0 , 0 , e a/l , 0 , 0 },
{ 0 , -12 e i/l^3 , -6 e i/l^2 , 0 , 12 e i/l^3 , -6 e i/l^2 },
P P
15
I=200
I=50 6
10
I=100
Solution:
The element stiness matrices are given by
2
u1 2 0 0 3
20 1; 208
k1e = 664 1;208
6
96; 667 777 (15.58-a)
5
u1 2
2
0 0 3
0:01 0:10
k1g = ,P 64 0:10 16 :00
6
6 777 (15.58-b)
5
2
0 2 0 3 3
k2e = 664 128 ; 890 64; 440 777
6
u1 2 3
(66:75) , P (0:026666) (1; 208:33) , P (0:1) (1; 678:24) , P (0:1)
(1; 208:33) , P (0:1) (225; 556:) , P (16:) (64; 444:4) , P (0) = 0 (15.60)
(1; 678:24) , P (0:1) (64; 444:) , P (0) (209; 444:) , P (9:6)
The smallest buckling load amplication factor is thus equal to 2; 017 kips.
(* Initialize constants *)
a1=0
a2=0
a3=0
i1=100
i2=200
i3=50
l1=10 12
l2=15 12
l3=6 12
e1=29000
e2=e1
e3=e1
(* Define elastic stiffness matrices *)
ke[e_,a_,l_,i_]:={
{e a/l , 0 , 0 , -e a/l , 0 , 0 },
{0 , 12 e i/l^3 , 6 e i/l^2 , 0 , -12 e i/l^3 , 6 e i/l^2 },
{0 , 6 e i/l^2 , 4 e i/l , 0 , -6 e i/l^2 , 2 e i/l },
{-e a/l , 0 , 0 , e a/l , 0 , 0 },
{ 0 , -12 e i/l^3 , -6 e i/l^2 , 0 , 12 e i/l^3 , -6 e i/l^2 },
{ 0 , 6e i/l^2 , 2 e i/l , 0 , -6 e i/l^2 , 4 e i/l }
}
ke1=ke[e1,a1,l1,i1]
ke2=ke[e2,a2,l2,i2]
ke3=ke[e3,a3,l3,i3]
(* Define geometric stiffness matrices *)
kg[l_,p_]:=p/l{
{0 , 0 , 0 , 0 , 0 , 0 },
{0 , 6/5 , l/10 , 0 , - 6/5 , l/10 },
{0 , l/10 , 2 l^2/15 , 0 , - l/10 , - l^2/30 },
{0 , 0 , 0 , 0 , 0 , 0 },
{0 , -6/5 , - l/10 , 0 , 6/5 , - l/10 },
{0 , l/10 , - l^2/30 , 0 , - l/10 , 2 l^2/15 }
50
80,000
6m 6m
Solution:
Using two elements for the beam column, the only degrees of freedom are the de
ection and
rotation at midspan (we neglect the axial deformation).
Victor Saouma Finite Elements
15.4 Geometric Non-Linearity 15{19
The element stiness and geometric matrices are given by
Draft
0 0 0 0 v1 2
2 3
0 0 0 0 0 0
60
6 222 ; 222 : 666 ; 666: 0 ,222; 222: 666; 666: 77
[K1e ] = 666 00 666;0666: 2; 6660; 666
6
0: ,666; 666: 1; 333; 333 777 (15.61)
6
0 0 0 7
7
4 0 ,222; 222: ,666; 666: 0 222; 222: ,666; 666: 5
0 666; 666: 1; 333; 333 0: ,666; 666: 2; 666; 666
0 v1 2 0 0 0
2 3
0 0 0 0 0 0
60
6 222; 222: 666; 666: 0 ,222; 222: 666; 666: 77
[K2e ] = 666 00 666;0666: 2; 6660; 666 00: ,6660; 666: 1; 3330; 333 777
6 7
(15.62)
6 7
4 0 ,222; 222: ,666; 666: 0 222; 222: ,666; 666: 5
0 666; 666: 1; 333; 333 0: ,666; 666: 2; 666; 666
0 0 0 0 v1 2
2
0 0 0 0 0 0 3
6 0 ,16; 000 ,8; 000 0 16; 000 ,8; 000 77
6
[K1g ] = 666 00 ,8;0000 ,640; 000 00 8; 000 16; 000 777
6
0 0 77 (15.63)
6
40 16; 000 8; 000 0 ,16; 000 8; 000 5
0 ,8; 000 16; 000 0: 8; 000 ,64; 000
0 v1 2 0 0 0
2
0 0 0 0 0 0 3
6 0 ,16; 000 ,8; 000 0 16; 000 ,8; 000 77
6
[K2g ] = 666 00 ,8;0000 ,640; 000 00 8; 000 16; 000 777
6
0 0 77 (15.64)
6
40 16; 000 8; 000 0 ,16; 000 8; 000 5
0 ,8; 000 16; 000 0: 8; 000 ,64; 000
Assembling the stiness and geometric matrices we get
v1 2
[K] = 412; 444: 0: (15.65)
0: 5; 205; 330
and the displacements would be
v1 = ,0:00012123 (15.66)
2 0
1,000
12 12
1 2
Solution:
15.5 Summary
? ? ?
2 2
dx , y
"x = du
2nd Order D.E. 4th Order D.E. dv dv
+ 12 dx
dx2
2 B.C. 4 B.C.
? ?
d2 y
, EIP = 0 U= 1R
2
E" d
2
dx2
v = ,A sin kx , B cos kx
? ?
d4 v
EI , P = w
dx4
d2 v
dx2 K= Ke Kg
v = C1 sin kx + C2 cos kx + C3 x + C4
?
P = (Ke + Kg )
?
jKe + Kg j = 0
Chapter 16
MATERIAL NONLINEARITIES
Adapted from Zienkiewicz
16.1 Introduction
50 Given that an element stiness matrix is given by
Z
K =
BT DBd
(16.1)
material nonlinearity stems from the non-linear stress-strain relation embodied in the material
constitutive matrix D.
51 Recalling that the strains are derivatives of the displacement, the stiness matrix becomes
itself a function of the displacements, K = K(u) and is thus non-linear.
52 Nonlinear problems must be analysed incrementally. For the sake of discussion, we will
assume that the analysis is driven by increments of loads P.
53 At the end of each load increment, internal forces must be in equilibrium with the external
ones.
Rn+1 R(un+1) = R(un+1) , P = 0 (16.2)
where R is the reaction vector (or internal forces) corresponding to the total displacement
NX
elem Z NX
elem Z
R=
e
BT d
=
e
BT D j d
(16.3)
e=1 e=1
P is the vector of externally applied force in the increment,
u is the displacement vector,
R the residual force vector.
54 For equilibrium to be satised, the vector of reactions (or internal forces) R must be equal
to the one of external ones P.
16{2 MATERIAL NONLINEARITIES
55 We start the analysis from an equilibrium conguration, at the end of increment n such that
Draft
u = un; Rn = 0; P = Pn (16.4)
and apply an increment of load Pn such that
Pn+1 = P , n + Pn (16.5)
and we seek to determine the corresponding changee in displacement
un+1 = un + un (16.6)
We will keep Pn reasonably small to capture the full nonlinear response.
6 n +1
Pn+1
6 6 ? ? u
1 6
KT1
Pn R1n+1 6
u1n u2n
Pn ? ? u - -
n
u1n - 2
un -
- u
61 In general the cpu time required for the extra iterations required by this method is less than
the one saved by the assembly and decomposition of the stiness matrix for each iteration.
62 It should be mentioned that the tangent stiness matrix does not necessarily have to be the
true tangent stiness matrix; an approximation of the true tangent stiness matrix or even the
initial stiness matrix will generally produce satisfactory results, albeit at the cost of additional
iterations.
R3n+1
Draft
P
R2n+1 R4n+1
6 n +1
Pn+1
6 6 ? ?? u
KT1
1 66
Pn R1n+1 6
u1n
Pn ? ? u -u-2n -
u3n
u1n2 -
un3 -
un -
- u
un u1n+1 u2 u3n+1
n+1
P R3n+1 R5n+1
Draft
R2 n+1 Rn+1
4
6 n +1
Pn+1
6 6 ? ? ?? u
1
K0 66
Pn R1n+1 6
6
? ?u
Pn
n - u4n
u-1nu-2nu-3n
-u1n 2
u-n3
un - - u
u2n+1 u4n+1
un u1n+1 u3n+1
63This method is a compromise between the rst two. First we seek two displacements by two
cycles of modied Newto-Raphson, then a secant to the curve is established between those two
points, and a step taken along it, Fig. 16.4.
P
R2n+1 R3n+1 R4n+1
Draft
6 n +1
Pn+1
6 6 ? ? ? u
1 e
6
K0
6
Pn R1n+1
e
6
Pn ? ?u
n
u-1n u2n -u3n -
-u1n u2n - 2
un-
- u
64 Subsequently, each step will be taken along a secant connecting the previous two points.
Hence, starting with
u1n = ,K,T 1 R1n+1 (16.15)
the secant slope can be determined
u1n
(K2S ),1 = , (R1 , (16.16)
n+1 Rn+1 )
2
and then
u2n = ,(K2S ),1 R2n+1 (16.17)
uin = ,(KiS ),1 Rin+1 (16.18)
65 This process can be generalized to (Ki ),1 = , uin (16.19)
S (Rin,+11 , Rin+1 )
Victor Saouma Finite Elements
16{6 MATERIAL NONLINEARITIES
16.3 Acceleration of Convergence, Line Search Method
Draft
From Reich (1993)
66 The line search is an iterative technique for automatically under- or over-relaxing the dis-
placement corrections uj so as to accelerate the convergence of nonlinear solution algorithms.
67 The amount of under- or over-relaxation is determined by enforcing an orthogonality condi-
tion between the displacement corrections uj and the residual loads Rj +1, which amounts to
forcing the iterative change in energy to be zero.
68 The displacement corrections are multiplied by a scalar value sk dening the amount of
under- or over-relaxation such that the total displacements uj +1;k are dened as
uj+1;k = uj + sk uj (16.20)
69 For k = 0 and k = 1, the values of sk are 0.0 and 1.0, respectively. Therefore, uj +1;0 = uj
and uj +1;1 = uj +1 . The orthogonality condition is quantied by a scalar value gk representing
the iterative change in energy, which is dened as
gk = uj Rj+1;k (16.21)
where
Rj+1;k = P , R(uj+1;k) (16.22)
are the residual loads at the end of solution iteration j and line search iteration k.
70 g k can be expressed as a function of sk (see Figure ??) and the object of the line search is to
nd sk such that gk is zero. An estimate of sk+1 such that gk+1 is zero can be computed using
a simple extrapolation procedure based on similar triangles
sk+1 = sk (16.23)
g0 g0 , gk
On rearranging terms, sk+1 is dened as
!
sk+1 = sk g0 , gkg0 (16.24)
71 As a preventative measure, sk+1 is assigned a value of 5.0 for all sk+1 > 5:0 so that unre-
strained over-relaxation is inhibited. Once sk+1 is estimated, uj +1;k+1, Rj +1;k+1 , and Rj +1;k+1
are computed for the next line search iteration, Fig. 16.5.
72 The line search terminates after three iterations or when
j g0 j 0:8 (16.25)
j gk j
and g0 gk 0:001 j g0 j. Smaller tolerances may be used to determine if the line search has
converged, Zienkiewicz & Taylor (1991) prefer to use 0.6, but Criseld (1979) concluded that
there was little advantange to be gained by doing such.
Victor Saouma Finite Elements
16.4 Other Control 16{7
g
Draft
g0
g1
s1 s2 s
74 In all preceding methods, iterations are performed until one or all of a variety of convergence
criteria are satisied. Relative convergence criteria are optionally enforced on the displacements,
loads, and/or incremental energy to dene the termination conditions.
75 The relative displacement criteria is dened in terms of the displacment corrections duj and
the updated incremental displacements uj +1 as
u = k kuuj +1k2k
j
(16.26)
2
16.6 Plasticity
Chapter 17
DYNAMIC ANALYSIS
Adapted from Kishen Chandra
17.1 Introduction
50 The equations of equilibrium governing the linear dynamic response of a system of nite
elements is given by [?]:
MU + CU_ + KU = R (17.1)
where M , C and K are the mass, damping and stiness matrices, R is the external load
vector and U; U_ and U are the acceleration, velocity and displacement vectors of the nite
element assemblage.
51 Mathematically Equation 17.1 represents a system of linear dierential equation of second
order, and in principle, the solution of the equations can be obtained by standard procedures
for the solution of dierential equatons with constant coecients.
52 However, the procedures proposed for the solution of general systems of dierential equations
can become very expensive if the order of the matrices is large - unless specic advantage is
taken of the special characteristics of the coecient matrices M, C and K.
53 In practical nite element analysis, we are therefore mainly interested in a few eective
methods, namely direct integration and mode superposition. These two techniques are closely
related and the choice for one method or the other is determined only by their numerical
eectiveness. In the present implementation the direct integration method is employed.
65 Similarly,
2 @ 22v jt @2v
t +1 , @t2 jti
2 @3v t
R2 = 2! @t3 jt=ti +2 ' 2! @t i
t
' t
(vi+1 , vi) (17.12)
and
are parameters representing numerical approximations.
66 Substituting Equations 17.11 and 17.12 into Equations 17.9 and 17.10 respectively, we obtain,
68 The scheme is explicit when is 0. When
is 1/2, this explicit form has the same numerical
properties as the central dierence method.
69 From Eqn. 17.13, we have
K = 1 t2 M +
t C + K
C]v + [ 1 M + (
, 1)C]v_ + [( 1 , 1)M + t (
, 2)C]v
fi+1 = fi+1 + [ 1 t2 M + t i t i 2 2 i
Appendix A
MISC.
A.1 Units & Conversion Factors
length, m (meter) 1 inch = 0.0254 m; 1 m = 39.37 inch
Force, N (Newton) 1 lb = 4.4482 N; 1 N = 0.22481 lb
Mass, Kg (kilogram) 1 lbm = 0.45359 Kg; 1 Kg=2.2046 lb
Density, Kg/m3 1 lbm/ft3 = 16.018 Kg/m3 ;1 Kg/m3 =0.062428 lbm/ft3
Temperature, T T o F=[(9/5)To C+32]
Acceleration, m/s2 1 in/s2 = 0.0254 m/s2 ;
Stiness, N/m 1 lb/in = 175.1 N/m
Stress, Pa = N/m2 1 psi = 6,894.8 Pa; 1 MPa = 145.04 psi
Work, energy, N-m=Joule 1 ft-lbf= 1.3558 J; 1 J = 0.73756 ft- lbf
Power, J/s=W
Heat Transfer
Convection coecient, h 1 Btu/h.ft2 .o F = 5.6783 W/m2 .o C
Heat, J 1 Btu=1055.06 J; 1 Btu = 778.17 ft-lb
Heat Source/Sink, Q W/m3 =
Heat
ux (q) 1 Btu/h.ft2 = 3.1546 W/m2
Specic heat, c 1 Btu/o F = 1,899.108 J/o C
Thermal conductivity, k 1 Btu/h.ft.o F = 1.7307 W/m.o C
Seepage Flow
permeability, k
Fracture
p Mechanics p
Stress intensity factor, K 1 MPa m=1.099 ksi in
Fracture energy GF 1 lb/in =.0057 N/m;
A{2 MISC.
A.2 Metric Prexes and Multipliers
Draft