Anda di halaman 1dari 310

LINEAR, STATIC FINITE ELEMENT

ANALYSIS

W. J. Anderson
Professor of Aerospace Engineering
The University of Michigan
Ann Arbor, MI 48109

January 25 , 1994

@Copyright 1994
All Rights Reserved
William J. Anderson
Ann Arbor, Michigan
3

ACKNOWLEDGMENTS

The author is indebted to the many students and associates who have contributed
concepts, examples and encouragement for the creation of this book. In particular,
Prof. Joe Eisley has stressed the energy approach and use of finite elements in the
Aerospace curriculum. Prof. Richard Scott has helped teach the material in summer
conferences. Industry associates have encouraged FEA teaching and use of commercial
FEA programs at the university: Don Dewhirst, Louie Nagy, George Campbell, Jerry
Joseph, S. C. Wang, Gordon Willis, Tom Tecco, and others.

Any such textbook rests on the achievements of other, earlier authors. This book
has been influenced most by William Weaver, Chandrakant Desai, John Abel, 0. C.
Zienkiewicz, Uri Kirsch, Ken Huebner and Tom Hughes, as well as a host of researchers.

Some examples have been taken from the author's doctoral students, including John
Russell, Myung Suh, Howard Gans, and Jungsun Park. Other students have pitched
in and helped to find errors. Their efforts are all appreciated.
Contents

1 INTRODUCTION & HISTORICAL REVIEW 1


1.1 DEFINITION OF "FINITE ELEMENT" 1
1.2 EXAMPLES OF FINITE ELEMENTS 1
1.3 THE DISCRETIZATION PROCESS . . . . . . . . 2
1.4 HISTORICAL REVIEW . . . . . . . . . . . . . . . 4
1.5 EXAMPLES OF FINITE ELEMENT SOLUTIONS . . 6
1.6 EXAMPLES OF RESEARCH USING F.E.A. . . . . . 9
1.7 STIFFNESS VS. FLEXIBILITY FORMULATION . . 14
1.8 EQUILIBRIUM APPROACH VS. ENERGY APPROACH 15
1.9 DISPLACEMENT METHOD VS. FORCE AND HYBRID METHODS 16

2 DERIVATION OF FINITE ELEMENTS 18


2.1 COMMON IDEAS FOR STRUCTURAL FINITE ELEMENTS . 19
2.1.1 Interpolation . . . . . . . . 19
2.1.2 Material Law . . . . . . . . 20
2.1.3 Strain-Displacement Law .. 21
2.2 SPECIFIC METHODS . 22
2.2.1 Equilibrium . . . . . . . . . 22
2.2.2 Virtual Work . . . . . . . . 23
2.2.3 Minimum Potential Energy 23
2.2.4 Minimum Complementary Energy . 24
2.2.5 Hybrid Methods (Reissner) . . . 24
2.3 GALERKIN METHOD (CLASSICAL) 24
2.3.1 Bubnov-Galerkin Method . . . 26
2.3.2 Petrov-Galerkin Method . . . . 26

3 LINE ELEMENT & ASSEMBLY PROCESS 29


3.1 CREATION OF LINE ELEMENT BY EQUILIBRIUM . . 29
3.2 ASSEMBLY OF LINE ELEMENTS . . . . . . . . . . . . 31
3.3 EMBEDDING THE LINE ELEMENT INTO A 3-D SPACE . 34
3.4 LINE ELEMENT WITH VARYING AREA . 36
3.4.1 False Start . . . . . . . . . . . . . . . . . . . . . . . . . 38

4
CONTENTS 5

3.4.2 Proper Approach . . . . . . . . 39


3.5 HOMEWORK . . . . . . . . . . . . . . 39

4 CONSTANT STRAIN TRIANGLE 43


4.1 PLANE STRESS . . . . . . . . . . . . . . . . . . . 43
4.2 CONSTANT-STRAIN TRIANGLE . . . . . . . . . 44
4.3 COMMENTS . . . . . . . . . . . . . . . . . . . .. 48
4.4 EXAMPLE OF TURNER TRIANGLE . . . . . . . 48
4.5 ASSEMBLY OF TRIANGULAR ELEMENTS . . . . . 49
4.6 EXAMPLE: MEMBRANE WITH IN-PLANE LOAD . 51
4.7 HOMEWORK . . . . . . . . . . . . . . . . . . . . . . . 53

5 INTERPOLATION 59
5.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . 59
5.2 ONE-DIMENSIONAL , LINEAR INTERPOLATION . . . . . 63
5.3 ONE-DIMENSIONAL QUADRATIC INTERPOLATION. 66
5.3.1 Summation Form . . . . . . . . 66
5.3.2 Product Form . . . . . . . . . . . . . . 67
5.4 PROOF OF NORMALIZATION (1-D) . . . . 68
5.5 HOMEWORK . . . . . . . . . . . . . . . . . . 68

6 VIRTUAL WORK AND POTENTIAL ENERGY THEOREMS 72


6.1 BACKGROUND . . . . . . . . . . . . . . . . 72
6.1.1 Stress-Strain Law (Constitutive Law) . . . . . . . . . . . . 72
6.1.2 Strain- Displacement Law . . . . . . . . . . . . . . . . . . . 74
6.1.3 Equilibrium Position . . . . . . . . . 74
6.1.4 Virtual Displacement . . . . . . . . . 74
6.1.5 Work . . . . . . . . . . . . . . .... 76
6.1.6 Virtual Work . . . . . . . . . . . . . 76
6.2 VIRTUAL WORK THEOREM .. 76
6.3 POTENTIAL ENERGY THEOREM . 77
6.4 HOMEWORK . . . . . . . . . . . . . . 79

7 DERIVATION OF A FINITE ELEMENT BY VIRTUAL WORK 84


7.1 DERIVATION FOR POINT LOADS . . . . ...... . 84
7.2 EQUIVALENT NODAL LOADS . . . . . . . . . . . . . . . . . . 86
7.3 PRESTRAIN AND PRESTRESS . . . . . . . . . . . . . . . . . . 87
7.4 REVIEW OF EQUIVALENT NODAL LOAD CONCEPT . . . . 88
7.5 ALTERNATE STRAIN ENERGY EXPRESSIONS 89
7.6 HOMEWORK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
.

CONTENTS 6

8 DERIVATION OF A LINE ELEMENT BY VIRTUAL WORK 96


8.1 LINE ELEMENT. ENERGY FORMULATION . . . . . . . . . . 96
8.2 EXAMPLE OF EQUIVALENT NODAL LOADS . . . 98
8.3 HOMEWORK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

9 BEAM ELEMENT 100


9.1 CLASSICAL, PLANAR FLEXURE . . . . . . . . . 100
9.2 FEA MODEL CONCEPTS FOR FLEXURE . . . . 103
9.3 EXAMPLE OF PLANAR BENDING OF BEAM . 106
9.4 EMBEDDING BEAM INTO 3-D . . . . . 109
9.4.1 Flexure in xz Plane . . . . . . . . . . . . . . . . . . 109
9.4.2 Axial Motion . . . . . . . . . . . . . . . . . . . . . 110
9.4.3 Torsion . . . . . . . . . . . . . . . . . . . . 110
9.4.4 Combined Beam Stiffness Effects in 3-D . . 110
9.5 THIN-WALLED BEAMS. WARPING. . . . . . . . 112
9.6 DEEP BEAMS. SHEARING DEFORMATION. . . 112
9.7 END RELEASE CONCEPT . . 112
9.8 BEAM OFFSET . . . . . . . . . . . . . . 113
9.9 HOMEWORK . . . . . . . . . . . . . . . . 114

10 STRESSES IN AUTO-HAUL TRAILER 123


10.1 INTRODUCTION . . . . 123
10.2 PHYSICAL MODELING . . . . . 125
10.2.1 Geometry . . . . . . . . . . 125
10.2.2 Materials . . . . . . . . . . . . . . . . . . . 125
10.2.3 Loads . . . . . . . . . . . . . . . . 126
10.3 FINITE ELEMENT MODELING . 126
10.4 ANALYTICAL RESULTS 128
10.4.1 Deflections . . . . . . . . 128
10.4.2 Stresses . . . . . . . .. 129
10.5 EXPERIMENTAL RESULTS . 130
10.6 CONCLUSIONS . . . . . . . . . . . . . . . . . . . . . . 130

11 COMBINING DIFFERENT ELEMENTS 133

12 LINEAR STATIC EQUATION SOLVERS 137


12.1 GENERAL . . . . . . . . . . . . . . . . . . . . . . . . 137
12.2 LINEAR STATIC STRESS ANALYSIS . . . . . . . . . 137
12.3 LINEAR, STATIC EQUATION SOLVERS .. 138
12.3.1 Direct Methods . . . . . . ..... . . ........ 138
12.3.2 Iterative Methods . . . . . . . . . . . . . .. 138
12.4 STORAGE CONCEPTS . . . . . . . . . . . . . . . 139
12.5 ADVANTAGE OF TRIANGULAR FORM FOR SOLUTION 139
CONTENTS 7

12.6 TRIANGULAR DECOMPOSITION . . . . . . . . . . . . . . . . . . . 140


12.7 GAUSS ELIMINATION . . . . . . . . . . . . . . . . . . . . . . . . . . 140
12.8 PROGRAMMING GAUSS ELIMINATION . . . . . . . . . . . . . . . 141
12.9 HOMEWORK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

13 SYMMETRY ARGUMENTS 149


13.1 GENERAL CONCEPT . . . . . . . . . . . . . . . . . . . . . . . . . . 149
13.2 REFLECTIVE PLANES . . . . . . . . . . . . . . . . . . . . . . . . . . 149
13.3 CYCLIC SYMMETRY . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
13.4 DIHEDRAL SYMMETRY . . . . . . . . . ... ...... ....... 154
13.5 HOMEWORK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154

14 REMOVAL OF RIGID BODY MODES 163


14.1 DEFINITION OF RIGID-BODY MODES . . . . . . . . . . . . . . . . 163
14.2 HOW TO REMOVE RIGID BODY MODES . . . . . . . . . . . . . . 164
14.3 EXAMPLE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
14.4 EFFECT OF RIGID BODY REMOVAL ON DEFORMED PLOTS .. 166
14.5 EFFECT OF SYMMETRY ARGUMENTS ON RIGID BODY MODES 167
14.6 MECHANISMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
14.7 NODAL SINGULARITIES. . . . . . . . . . . . . . . . . . . . . . . . . 168
14.8 HOMEWORK. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

15 MODIFICATION OF EQUILIBRIUM EQUATIONS 173


15.1 PROBLEM STATEMENT . . . . . . . . . . . . . . . . . . . . . . . . . 173
15.2 PARTITIONING . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
15.3 "PENALTY" METHOD APPROACH .................. 174
15.4 SOLVING IN PLACE . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
15.5 STATIC CONDENSATION . . . . . . . . . . . . . . . . . . . . . . . . 175
15.6 SUBSTRUCTURING . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
15.7 COMMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
15.8 HOMEWORK . . .' . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

16 BANDWIDTH CONCEPTS 181


16.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
16.2 BANDWIDTH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
16.3 WAVEFRONT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
16.4 "POINTER" METHODS FOR SPARSE MATRICES . . . . . . . . . . 184
16.5 DIFFICULTIES WITH CPU TIME FOR SOLID ELEMENTS . . . . . 185
16.6 COMMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
16.7 HOMEWORK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
16.8 REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
CONTENTS 8

17 THIN PLATES 196


17.1 PHYSICAL DESCRIPTION . . . . . . 196
17.2 HISTORICAL COMMENTS . . . . . . 198
17.3 CLASSICAL THIN-PLATE THEORY 198
17.3.1 In-Plane Forces . . . . . . . . . 199
17.3.2 Out-of-Plane Forces and Moments . 200
17.3.3 Boundary Conditions . . . . . . 202
17.3.4 Stress Recovery . . . . . . . . . . . 203
17.4 FINITE ELEMENT FORMULATION . . 204
17.5 MELOSH, ZIENKIEWICZ AND CHEUNG RECTANGULAR PLATE
ELEMENT . . . . . . . . . . . . . . . . . . . . 205
17.6 COMBINED BENDING AND STRETCHING . 209
17.7 THIN SHELLS . 210
17.8 THICK SHELLS 211
17.9 HOMEWORK. . 211

18 COORDINATE SYSTEMS 216


18.1 DEFINITIONS . . . . . . 216
18.1.1 Grid Point Coordinate System . 217
18.1.2 Element Coordinate System 218
18.1.3 Material Coordinate System . . 218
18.1.4 Global Coordinate System . . . 218
18.2 COORDINATE TRANSFORMATIONS 219
18.2.1 Aborted Assembly in Element Coordinates . 220
18.2.2 Cartesian Displacement Transformation in 2-D . 220
18.2.3 Force Transformation in 2- D . . 222
18.2.4 Stiffness Transformation in 2-D 222
18.2.5 Example: Rod Element in 2-D . 222
18.3 HOMEWORK . . . . . . . . . . . . . . 224

19 THERMAL STRESS 226


19.1 EQUIVALENT NODAL LOAD . . . . . . . . . . . 226
19.2 THERMAL STRAIN IN VARIOUS SITUATIONS 227
19.3 EXAMPLE: TWO-NODE LINE ELEMENT . 228
19.3.1 Constrained Element . . . . 229
19.3.2 Freely Expanding Element . 230
19.4 HOMEWORK . 231
19.5 REFERENCES . . . . . . . . . . . 233

20 GAUSS INTEGRATION 235


20.1 NUMERICAL INTEGRATION . . . . . . 235
20.2 INTERPOLATION . . . . . . . . . . . . . 236
20.3 TRAPEZOIDAL INTEGRATION RULE . 236
CONTENTS 9

20.4 SIMPSON'S INTEGRATION RULE 237


20.5 GAUSS INTEGRATION . . . . . . . 237
20.5.1 Symmetric Form . . . . . . . 237
20.5.2 One-Term Gauss Integration Formula . 239
20.5.3 Two-Term Gauss Integration Formula . 240
20.5.4 Gaussian Formulas with Many Sample Points . 242
20.6 EXAMPLE OF GAUSSIAN INTEGRATION IN ONE DIMENSION 243
20.6.1 Nonsymmetric Gaussian integration . . 244
20.7 GAUSS INTEGRATION IN 2-D . 245
20.7.1 Best Theoretical Procedure 245
20. 7.2 Practical Procedure . . . . . 245
20.7.3 Comments . . . . . . . . . . 246
20.8 RADAU AND LOBATTO INTEGRATION 246
20.9 Finite Element Applications . . . . 247
20.10HOMEWORK . . . . . . . . . . . . 247

21 ISOPARAMETRIC ELEMENTS 250


21.1 INTERPOLATING INDEPENDENT VARIABLES 250
21.1.1 Review of Interpolation of Functions . 250
21.1.2 Interpolation of Independent Variable . 251
21.2 ISOPARAMETRIC LINE ELEMENT .. 251
21.3 ISOPARAMETRIC QUADRILATERAL 253
21.4 HOMEWORK . . . . . . . . . . . . . . . . 257

22 SOLID ELEMENTS 262


22.1 TYPICAL TYPES OF SOLID ELEMENTS 262
22.2 ELASTICITY THEORY FOR SOLIDS .. 263
22.3 CONSTANT-STRAIN TETRAHEDRON . . . 264
22.4 HEXAHEDRON . . . . . . . . . . . . . . . . 265
22.5 EXAMPLE- THE MSC/NASTRAN HEXA ELEMENT .. 267
22.6 PENTAGONAL SOLID ELEMENT .. 268
22.7 HOMEWORK . . . . . . . . . . . . . . . . . . . . . . . . . 269

23 CASE STUDY: HYDRAULIC VALVE 272


23.1 PHYSICAL PROBLEM . . . . . 272
23.2 COST AND TIME ESTIMATES . 273
23.3 PHYSICAL MODELING . . . . . 273
23.4 FINITE ELEMENT MODELING . 275
23.5 RESULTS . . . . . . . . . . . . . . 276
23.6 TECHNICAL CONCLUSIONS .. 278
23.7 MANAGEMENT CONCLUSIONS 278
23.8 ADDITIONAL , LATER RESULTS 279
23.9 SCALING LAWS . . . . . . . . . . 279
CONTENTS 10

23.9.1 Pressure Loading . . . . . . . . . . . . . . . . . . . . . . . . . . 279


23.9.2 Gravitational and Inertial Loading . . . . . . . 280

24 MULTIPOINT CONSTRAINTS 282


24.1 OVERVIEW. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
24.2 NONWORKING CONSTRAINT FORCES. . . . . . . . . . . . . . . . 283
24.3 MATHEMATICAL RELATIONS . . . . . . . . . . . . . . . . . . . . . 284
24.4 MATRIX FORM OF MPC RELATIONS . . . . . . . . . . . . . . . . . 285
24.5 SAMPLE PROBLEM AND SOLUTION . . . . . . . . . . . . . . . . . 287
24.5.1 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . 287
24.5.2 Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
24.6 HOMEWORK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290

25 RIGID BODY ELEMENTS 292


25.1 OVERVIEW. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
25.2 RIGID BAR ELEMENT . . . . . . . . . . . . . . . . . . . . . . . . . . 292
25.3 GENERAL RIGID BODY ELEMENTS . . . . . . . . . . . . . . . . . 293
25.4 LOAD DISTRIBUTION BY RIGID ELEMENTS . . . . . . . . . . . . 294

26 STEADY STATE HEAT CONDUCTION 296


26.1 OVERVIEW. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
26.2 HEAT CONDUCTION VARIABLES . . . . . . . . . . . . . . . . . . . 297
26.3 DERIVATION OF EQUATIONS . . . . . . . . . . . . . . . . . . . . . 297
26.4 LINE ELEMENT IN HEAT CONDUCTION . . . . . . . . . . . . . . . 302
26.5 HEAT CONDUCTION EXAMPLE: SOLDERING WITH HEAT SINK 303
26.6 HOMEWORK. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
I

List of Figures

1.1 Line element. (Truss, beam, electrical resistor, pipe.) 2


1.2 Two-dimensional element. (Membrane, plate, shell.) 2
1.3 Solid element . . . . . . . . . . . . . . . . 2
1.4 Brachistochrone solution, due to Leibnitz. 5
1.5 Closed curve in 3-D space. . . . . . . . . . 5
1.6 Finite element mesh, due to Schellbach. . . 5
1. 7 Load ramp. Welded aluminum plate and beams. (Courtesy of Brooks
& Perkins Corp.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.8 Hydraulic valve. Gray cast iron. 2500 psi internal pressure. (Courtesy
of HYDRECO, Inc.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.9 Micro-Winchester disk drive chassis. Aluminum casting. (Courtesy of
Irwin-Olivetti, Inc.) . . . . . . . . . . . . . . . . . . 8
1.10 Heat conduction in a truck brake system. . . . . . . 8
1.11 Aircraft with trailing cable 3353 m (11,000 ft) long. 9
1.12 Family of balloon shapes. . . . . . . . . . . . . . 10
1.13 Finite element mesh for fully inflated balloon. . . . 10
1.14 Added mass coefficent for high altitude balloon 11
1.15 Finite element mesh for compressor blade. Plate elements. 11
1.16 Original shape of valve spring retainer. 12
1.17 Optimal shape of valve spring retainer. . . . . . . . . . . 12
1.18 Two wedges being pressed together . . . . . . . . . . . . 13
1.19 Baseline design with theoretically infinite stress at corner 13
1.20 Optimized shape of wedge, with uniform stress . 14
1.21 Finite element man to the rescue! . . . . . . . 14

2.1 Interpolation of f(x) over quadrilateral region. 20


2.2 An initially stretched string under lateral loads. 25

3.1 Two-node, constant area line element.. . . . . . 29


3.2 Equilibrium of external nodal forces with internal stresses. 30
3.3 Two line elements to be joined. . 31
3.4 Fictitious pin at node connection. . . . . 32
3.5 Line element in 3-D space . . . . . . . . 34
3.6 Local coordinate system for line element. 35

11
LIST OF FIGURES 12

3.7 Stretch in line element due to lateral deformation t. 35


3.8 Varying-area line element. . . . . . . . . . . . 37
3.9 Failed attempt to satisfy internal equilibrium. 38
3.10 Satisfaction of nodal equilibrium. . . . . 39
3.11 Spring element. Nodal displacement. . . 40
3.12 Triangles and line elements in assembly. 40
3.13 Line element with varying area. 41
3.14 Truss structure. . . . 41
3.15 Plane strain problem 42

4.1 Elasticity in 3-D . . . 43


4.2 Elasticity in 2-D . Plane stress illustrated. (Thin sheet.) 44
4.3 Plane stress. Constant strain (Turner) triangle. . . . . 45
4.4 Triangle element embedded in sheet . . . . . . . . . . . 46
4.5 Calculation of nodal loads from equilibrium arguments 46
4.6 Local coordinate system for developing stiffness of Turner triangle 49
4. 7 Assembly of triangle elements . 50
4.8 Assembly of stiffness terms . . . 50
4.9 Membrane under in-plane load . 51
4.10 Simple finite element mesh . . . 51
4.11 Alternate mesh . . . . . . . . . 52
4.12 Quarter-problem solved, by symmetry. 52
4.13 Constant strain triangles 53
4.14 Bracket (right triangle). 54
4.15 Constant strain triangle 54
4.16 Constant strain triangle 55
4.17 Finite element model of cantilever beam. 56
4.18 Two-dimensional quadrilateral. 57

5.1 Linear interpolation in one dimension. 59


5.2 Quadratic interpolation in one dimension. Information given at internal
point. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.3 Cubic interpolation in one dimension. Information about function and
its first derivative given at end-points. 60
5.4 Interpolation of two-dimensional scalar field. 61
5.5 Interpolation of two-dimensional vector field. 61
5.6 Interpolation of three-dimensional field. . . 62
5. 7 Numerical integration by trapezoidal rule. 63
5.8 Sine table. . . . . . . . . . . . . . . . . . . 63
5.9 Linear interpolation in sine table. . . . . . 64
5.10 Linear interpolation in one dimension. General case . . 64
5.11 Shape functions for linear interpolation in one dimension .. 65
LIST OF FIGURES 13

5.12 Contribution of each shape function to total functional value. Linear


interpolation. . . . . . . . . . . . . . . . 65
5.13 Quadratic interpolation in line element. . . . . . . . . . . . . . . . . . . 67
5.14 Sum of shape function. . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.15 Proof by contradiction. Attempt to pass nontrivial polynomial of degree
N-1 through N zero values. 69
5.16 Four-noded line element. 70
5.17 Constant strain triangle. . 70

6.1 Stress-strain law (nonlinear elastic). 73


6.2 Strain-displacement law (general case). 73
6.3 Load-deflection relation (general case). 73
6.4 Configuration of a body under load. . . 75
6.5 Virtual displacement. . . . . . . . . . . 75
6.6 Strains during a virtual displacement. . 75
6. 7 Virtual strain energy (increment in strain energy due to virtual strain). 78
6.8 Line element at equilibrium. 79
6.9 Man at end of diving board. 80
6.10 Line element assembly. . . . 81
6.11 Assembly of line elements. . 82
6.12 Constant-strain triangle in plane stress. 83

7.1 Typical finite element . . . . . . . . . . 84


7.2 Variables and transformations in FEA. 85
7.3 Distributed loads. Deflection exaggerated. 86
7.4 Stress-strain law. Initial stress and strain. 87
7.5 Plane stress pentagonal element with first scheme for shape functions 90
7.6 Plane stress pentagonal element with second scheme for shape functions 90
7. 7 Line element; assembly of line elements. 93
7.8 Constant strain triangle loaded by gravity . 93
7.9 Quadrilateral element in plane strain . . . . 94
7.10 Triangle with tip displacement in x direction 95

8.1 Line element. . . . . . . . . . . . . . . 96


8.2 General shape functions (hypothetical). 97
8.3 Distributed load. . . . . . . . . . 99
8.4 Equivalent nodal loads. . . . . . . 99
8.5 Load distribution on line element. 99

9.1 Straight, slender, prismatic bar in xy plane. Coordinate x axis passes


through beam centroidal axis. . . . . . . . . . . . . . . . . . . . . . . . 100
9.2 Side view of beam curved to circular arc. Deflection greatly exaggerated. 101
9.3 End view of beam cross-section looking toward origin. . . . . . . . . . . 102
LIST OF FIGURES 14

9.4 Side view of plane sections grid. !xy = 0. . . . . . . . . 102


9.5 Bottom view of beam with plane sections grid. lzx = 0. 103
9.6 Finite element model for pure flexure of beam. . . . . .103
9. 7 Shape functions for two noded beam in pure flexure. . 106
9.8 Beam with uniform load. . . . . . . . . . 107
9.9 Equivalent nodal loads. . . . . . . . . . . . . . . . . . 108
9.10 Rotations at nodes of example problem. . . . . . . . 108
9.11 Coordinates for 3-D beam including axial and torsional motion. 109
9.12 Stiffness for beam in 3-D including axial motion and torsional motion. . 111
9.13 v
Vector defining plane 1, and axis Ye . . . . . 111
9.14 Side view of deep beam. . . . . . . . . . . . . 112
9.15 Side view of shear deformation in deep beam. 113
9.16 Beam with slider mechanisms at ends. 113
9.17 Offset vectors for locating beam shear center away from nodes A, B. 114
9.18 Beam built in at both ends. . . . . . . 114
9.19 Angled beam before and after loading. 115
9.20 Clamped-clamped beam. 115
9.21 Loaded uniform beam. 115
9.22 Beam with center load. 116
9.23 Beam-rod system. . . . 117
9.24 Cantilever beam. . .. 117
9.25 Prescribed deflections. 118
9.26 Centrally loaded beam. 118
9.27 Prescribed deflections. 119
9.28 Linear load distribution. 119
9.29 Concentrated load. 120
9.30 Distributed load. . . . . 120
9. 31 Weight lifter. . . . . . . 121
9.32 Beam with midspan support 122

10.1 Loading empty autohaul trailer onto rail flatcar 123


10.2 Two loaded autohaul trailers on rail flatcar . . . 124
10.3 Conventional sidewall trailers and lowboy trailers 124
10.4 Track system . . . . . . . . . . . . . . . . . . . . 127
10.5 Method for raising (above) and holding (below) the auto 127
10.6 Sidewall, made of cold rolled steel tubing 128
10.7 Scheme for modeling the support system . . . . . 128
10.8 Finite element mesh including supports . . . . . . 129
10.9 Boundary conditions at rear wheels: "whiffletree" 130
10.10Loaded trailer. Corner lift . . . . . . . . . . 130
lO.llLoaded trailer. Front impact. . . . . . . . 131
10.12Loaded trailer. Front impact plus gravity. 131
LIST OF FIGURES 15

10.13Loaded trailer. Rear impact plus gravity 131


10.14High stress region due to nonaligned joint 131
10.15Properly aligned joint . . . . . . . . . . . . 132

11.1 Structure made of casting and sheet metal. 133


11.2 Finite element mesh, with solids and plates. 134
11.3 Transition between solids and plate, modeled with beams. 134
11.4 Transition modeled by extending plate into solid. 134
11.5 Finned, hollow cylinder. . . . . . . . 135
11.6 Cross-section AA of finned cylinder. . . . . . . . . 135

13.1 Points lying on normal to a plane and equidistance from it. 150
13.2 Space frame with yz reflective plane. . . . . 151
13.3 Solid element model with yz reflective plane. 152
13.4 Symmetric loading. . . . . . 152
13.5 Antisymmetric loading. . . . . . . . . . . . . 153
13.6 Body with cyclic symmetry. . . . . . . . . . 153
13.7 Segment from body with harmonic zero loads. 153
13.8 Dihedral symmetry. . . . . . . . . . . . . . . 154
13.9 Segment from body with dihedral symmetry. 154
13.10Beam with plane of symmetry at y=O. . . . 155
13.11Square cut-out in round tube. . . . . . . . . 155
13.12Equilateral triangle sheet with centered hole. 156
13.13Pierced Bars. . . . . . . . . . 157
13.14Plate supported by rigid bar. 157
13.15Symmetric beam . . . . . . . 158
13.16Quadrilateral with one reflective plane of symmetry 159
13.17Quadrilateral element with two planes of reflective symmetry. 1S9
13.18Quadrilateral with load at node 3. . 160
13.19Load with triangular distribution. 160
13.20Bracket with vertical load. 161

14.1 Desk telephone. . . . . . . 163


14.2 Rectangular sheet with circular hole under tension. Two-dimensions. 165
14.3 Crude FE mesh for 2-D problem. . . . . . . . . . . . . . . . . . 166
14.4 Deformed and undeformed body with left boundary constrained . 167
14.5 Deformed body centered on undeformed hole. . . . . . . . . . . . 167
14.6 Fortuitous removal of rigid body modes by symmetry constraints. 167
14.7 Disk. . . . . . 169
14.8 Half disk. . . . . . . 169
14.9 Tetrahedron. . . . . . 170
14.10Plane stress problem. 170
14.11Beam suspended between two guided supports. 171
LIST OF FIGURES 16

14.12Teeter-totter with two children. 172

15.1 Sheet metal under point loads .. 177


15.2 Identical line elements under point loads. 178
15.3 Assembly of line elements to be solved "in place" 179
15.4 Finite element model of membrane, before stiffening . 180
15.5 Stiffened membrane . . . . . . . . 180

16.1 Bandwidth and semi-bandwidth. . 182


16.2 First nodal numbering scheme. (Compact notation.) . 182
16.3 Second nodal numbering scheme. (Compact notation.) 183
16.4 First element numbering scheme for wavefront solver . 184
16.5 Second element numbering scheme for wavefront solver 184
16.6 "Skyline" storage. . . . . . . . . . . . . . 185
16.7 Comparison of 1-,2- ,3-dimensional space. 186
16.8 Two-dimensional problem. . 187
16.9 Two candidate FE problems 187
16.10Plane strain problem. . 188
16.11Plane stress problem. 188
16.12Axisymmetric body . 189
16.13Plate problem. 190
16.14Plane stress problem. 190
16.15Plate problem. 191
16.16Plane stress problem. 192
16.17Ring finite element model. 193
16.18Body with symmetric plane. 193
16.19"Stretched body" with symmetric plane. 194
16.20FE mesh for ship . . . . . . 194
16.21Reflective symmetry for ship . . . . . . . 195

17.1 Cylindrical curvature and twist in plates. 197


17.2 Anticlastic curvature in plate; bending in beam 197
17.3 Plate and grillage with same amount of material. 198
17.4 In-plane force resultants for an infinitesimal plate element 199
17.5 Out-of-plane shear, moment and twist resultants. . . . . . 200
17.6 Boundary conditions at right edge. . . . . . . . . . . . . . 202
17.7 An increasing twist along an edge, and how it causes negative shear
distribution. . . . . . . . . . . . . . . . . . . . . . . . . . . 203
17.8 Cancellation of vertical components of shear due to twist. . 203
17.9 Quadrilateral and triangular plate elements. . . . . 204
17.10Degrees of freedom in MZC rectangle. Pure flexure. . . . 205
17.11 Degrees of freedom in bending and stretching. . . . . . . 209
17.12Degrees of freedom for combined bending and stretching. 210
LIST OF FIGURES 17

17.13Faceted surface made from flat plate elements. . . . . . . . . . . . . . . 210


17.14Force balance at intersection of plates. . . . . . . . . . . . . . . . . . . 211
17.15Deformation of thick shells as modeled by 8 and 20 noded solid elements.212
17.16Plate under central load . . . . 212
17.17Plate problem. . . . . . . . . . 214
17.18Square plate with central load. . 214

18.1 Rectangular and cylindrical coordinate systems 216


18.2 Spherical and curvilinear coordinate systems . . 217
18.3 Assortment of element coordinate systems. . . . 218
18.4 Definition of material coordinates in typical element: QUAD4. 219
18.5 Example of "global" coordinates in NASTRAN. . . . . . . . . 219
18.6 Attempted assembly of two trusses. . . . . . . . . . . . . . . . 220
18.7 Relation between element coordinates and global coordinates. 221
18.8 Two-dimensional truss assembly, with basic coordinates shown. . 223
18.9 Inclined truss element. . . . . . . . . . . . . . . 223
18.10Comparison of horizontal and vertical stiffness. . 224
18.11 Truss element 225

19.1 Line element . 226


19.2 Heated line element under load 227
19.3 Aluminum link . . . . . . . . . 228
19.4 Stress-strain curves for heated and unheated case 229
19.5 Combined loading . . . . . . . . . . . . . . . . . . 230
19.6 Loads on constrained line element . . . . . . . . . 230
19.7 Stress-strain and load-displacement for constrained line element 230
19.8 Free expansion of aluminum link . . . . . . . . . . . . . . . . . . 231
19.9 Stress-strain and load-displacement for freely expanding line element 231
19.10Beam with thermal and mechanical loads . . 232
19.11Plane stress triangle under thermal load. . . 233
19.12Constant strain triangle under thermal load. 234

20.1 Area under a curve. . . . 235


20.2 Interpolation . . . . . . 236
20.3 Trapezoidal integration. 237
20.4 Simpson's rule integration. 238
20.5 Unsymmetric and symmetric domains. 238
20.6 An arbitrary function to be integrated with one Gauss term. 239
20.7 One term integration of linear function. . . . . . . 240
20.8 One term integration of higher degree polynomial. 240
20.9 Cubic polynomial on symmetric domain. 244
20.10Two-dimensional domain. 245
20.11Regular pattern of Gauss points. 246
LIST OF FIGURES 18

20.12Radau integration, with sample point at left end. 247


20.13Lobatto integration with sample points at both ends. 247
20.14Examples of Gauss point locations. 248

21.1 Shape functions for line element. 250


21.2 Mapping the line element onto the parent element. 252
21.3 Two-dimensional quadrilateral. 253
21.4 Mapping to the parent element. . . . 254
21.5 The shape function N1 (x,y). . . . . . 255
21.6 Gauss integration in two dimensions. 257
21.7 Integration of strain energy term in two dimensions. . 257
21.8 Two-dimensional finite element 258
21.9 Isoparametric element in 2-D 258
21.10Double unit square . . . . . . . 259
21.11Eight-noded quadrilateral plate element . 259
21.12Euler-Bernoulli beam, Turner triangle and plane stress quadrilateral. 260
21.13Parabolic (quadratic) line element. 261

22.1 Solid elements. . . . . . . . . . . . 262


22.2 Vector relations for elastic finite elements. 263
22.3 Laminated material. . . . . . . . . . . . . 263
22.4 Constant strain tetrahedron. . . . . . . . . 264
22.5 Volume coordinates defined on tetrahedron. 265
22.6 Zero energy deformation mode for hypothetical element solid element 266
22.7 Hour-glass deformation mode in solid element. . . . . . 266
22.8 Mapping of solid hexahedron onto the parent element.. 267
22.9 Gauss point arrays for 20 noded HEXA element . . . . 268
22.10Mapping the wedge element on the right angled prism. 268
22.11PENTA element reduced to plate-like dimensions. . . . 269
22.12Cube of elastic material. . . . . . . . . . . . . . . . . . 270
22.13Forces required to cause displacement at one node only. 270

23.1 Sketch of three bore hydraulic pump. . . . . . . . . . . 272


23.2 Three fixed reflective planes. . . . . . . . . . . . . . . . 274
23.3 Use of rigid body technique for moving reflective plane. 274
23.4 Fixed (top, right side) and moving (left side) reflective planes. 275
23.5 Mesh for hydraulic valve segment. 276
23.6 Side view of FE mesh. . . . . . . . 276
23.7 Top view of FE mesh. . . . . . . . 277
23.8 Side view, showing elliptical holes. . 277
23.9 High-stress regions at mold line. . . 278
23.10Stresses at bottom of fluid channel. 278
23.11 Column-like region to be straightened. 279
LIST OF FIGURES 19

23.12Internally pressurized hydraulic valve 280


23.13Geometrically scaled aircraft . 281

24.1 Joint between truss and plate 283


24.2 Hyperplane in displacement space 283
24.3 Nonworking constraint force . . . 284
24.4 Two plates to be welded, with reduced, spring model 290

25.1 Rigid bar element. . ............ . 293


25.2 Rigid element with distributed master d.o.f.. 293
25.3 Rigid element with one master node. . .. 294
25.4 Rigid element used for load redistribution. . 294

26.1 Relations between thermal stress and heat conduction problems 296
26.2 Interchangeability of operators . . . . . . . . . . . . . . . . . . . 297
26.3 Heat fluxes near a point (a), a surface (b), and at grid points (c,d) 297
26.4 Vectors in the structural stress problem . . . . 298
26.5 Vectors in the heat conduction problem . . . . 298
26.6 Body exposed to thermal boundary conditions 299
26.7 Finite element mesh for thermal problem 300
26.8 A single, typical thermal finite element . . . . 300
26.9 Line element in heat conduction . . . . . . . . 303
26.10 Heat sink (pliers) used while soldering diode lead 304
26.11 Physical layout, with nodes numbered . . 304
26.12Expected time variation of temperature . . 306
26.13Constant flux triangle . . . . . . . . . . . 306
26.14Quadratic heat element under center load 307
26.15Constant flux triangle subjected to heat flux at boundary . 307
26.16Aluminum rods exposed to thermal loading . . . . . . 308
26.17Tetrahedral heat conduction element. . . . . . . . . . 308
26.18Heat conduction quadrilateral subjected to edge flux. 309
26.19Rod pressed against wall and subjected to heat flux. . 309
List of Tables

10.1 Load cases for trailer study. . ............. . 129

13.1 Table 1. Symmetry constraints for structural problems. 151


13.2 Table 2. Symmetry constraints for elasticity problems. 151

16.1 Wavefront for first numbering scheme. 184


16.2 Wavefront for second numbering scheme. 185

20.1 Abscissae and Weights for Gauss Integration 243

20
Chapter 1

INTRODUCTION &
HISTORICAL REVIEW

1.1 DEFINITION OF "FINITE ELEMENT"


The finite element method has become a valuable tool for modeling structural,
mechanical, thermal, fluid and electrical systems. An element is defined as follows :

FINITE ELEMENT: A hypothetical subdivision of a structure or system,


possessing a regular shape which can be analyzed.

The finite element technology requires :

a) development of individual elements, often with classical mechanics concepts .

b) assembly of elements into a structure or system.

c) solution of the assembly, involving modern methods of computing.

The user must have a grasp of classical mechanics and numerical analysis in order to
use finite elements effectively.

1.2 EXAMPLES OF FINITE ELEMENTS


Typical finite elements are shown in Figs . 1.1-1.3. The element has specific points
called "nodes" or "grids" which occur primarily at vertices but occasionally appear
along edges, on the faces of the element, or even in the interior. Nodes

1) define the geometry of the element,

2) provide points at which physical information can be entered and retrieved

3) provide degrees of freedom for accuracy in computation.

1
CHAPTER 1. INTRODUCTION & HISTORICAL REVIEW 2

0"'-...L...--)_ _ _ _ _ . . . )

Figure 1.1 : Line element . (Truss, beam, electrical resistor, pipe.)

Figure 1.2: Two-dimensional element . (Membrane, plate, shell.)

Figure 1.3: Solid element

To develop such elements, one can assume the mathematical form of the internal
displacement field (or temperature, flow velocity, etc.) and develop a relation between
forces at nodes and displacements at nodes. This is called the displacement 1nethod
and is the most common approach in finite elements. An important part of the theory
is concerned with the changes in the field variables across the interelement boundaries .

Many different element types are created and entered into a finite element "library"
within a computer program. The user then creates models of complicated systems by
using combinations of element types.

1.3 THE DISCRETIZATION PROCESS


The finite element method has become a well-known tool for helping engineers and
scientists model complicated systems. The method has both practical and theoretical
CHAPTER 1. INTRODUCTION & HISTORICAL REVIEW 3

advantages. It allows the solution of bodies with strange shapes, with difficult boundary
conditions and loads, and made of many materials. The user of finite elements develops
intuition about mechanics , because one is forced to think in terms of matrix mappings
for all phenomena. After becoming comfortable with these transformations, one gains
an ability to organize the logic of mechanics.

Many physical systems are discrete from the outset, e.g. truss systems, space
frames, electrical and hydraulic circuits. These systems have been well handled by
engineers in the past 50 years and are called "networks." Networks can usually be
described by ordinary differential equations or algebraic equations.

Some physical systems, on the other hand, are continuous in nature, e.g. plates,
shells, fluid flow and electrical fields. These systems have been handled with difficulty
in the past, and are called "field problems." Field problems require the use of partial
differential equations. The goal of many approximate engineering theories, includ-
ing finite elements, is to convert field problems into something resembling a network
problem, i.e. to discretize the field.

The process of discretization can be done either in global or local way. The potential
energy method and Galerkin's method can represent the field variables by a series of
global functions, each of which must satisfy certain boundary conditions. On the other
hand, these same methods can model the field variables only on small regions (finite
elements). This local approach has advantages in terms of reducing the complexity of
the assembled set of matrix equations. Many terms in the "stiffness" matrix are zero,
and boundary conditions are more easily applied.

The finite element method has become the most successful method for discretization
of a broad range of field problems. It is particularly dominant for problems described
by elliptic partial differential equations, which are typical in structural mechanics.
Many commercial finite element analysis (FEA) codes are available to solve an amazing
variety of engineering problems.

The finite difference method is still the favored method in many flow and thermal
problems, particularly for problems described by hyperbolic differential equations. The
method is uniquely efficient for cases where an interface appears due to shock waves or
where there is a fluid-solid interface. Several commercial codes are available for finite
difference solutions to fluid and thermal problems.

Global approximate methods are still preferred by many people. Many engineers
feel that a global method called the "boundary element" method has a bright future.
This method is particularly efficient for solving for elastic response in solid bodies and
for problems with infinite domains. Only a few commercial computer programs are
available.

"Distorted" models were widely used in the past. An example was the use of frame
CHAPTER 1. INTRODUCTION & HISTORICAL REVIEW 4

elements to model stress and deflection in a concrete slab. Some reviewers believe
distorted modeling was a step on the way to the development of finite element methods.
The author, having tried to do such modeling in an auto-crash project, feels it was a
dead-end technology which may have delayed serious finite element development.

Modern finite element methods are based on true modeling, such as use of solid
elements to represent a solid body. We still, however, use flat plate elements in modeling
curved shells. This exception is not a serious case of distorted modeling because the
effects of curvature are included through the interaction of bending and stretching at
element boundaries.

Occasionally, one encounters in nature a system which is partially discrete and


partially continuous. The finite element method (unlike the finite difference method)
has no problem with the modeling of such a mixed system. Beam, truss and spring
elements are often combined with plate and solid elements.

1.4 HISTORICAL REVIEW


The following dates stand out in finite element history:

1696 Gottfried Leibnitz. Brachistochrone problem.


1851 Karl Schellbach. Minimum surface area.
1943 Richard Courant. Torsion of noncircular shaft.
1940's John Argyris. Aircraft structures.
1956 Turner, Martin, Clough and Topp. Aircraft structures.

Leibnitz [1) used a discrete method to solve the classical "brachistochrone" problem
to determine the curve of fastest descent for a bead traveling on a frictionless wire from
point A to point B (Fig. 1.4). The path is determined by points such as B and L, with
the further addition of intermediate points such as D. This is an exa1nple of a one-
dimensional problem, that is, the vertical coordinate of the point D is dependent only
on the horizontal coordinate.

Schellbach [1] studied a simple closed curve in 3-D space (Fig. 1.5) and wanted to
know the minimum-area surface enclosed by the curve. By looking at the curve from
above (Fig. 1.6), one sees the domain in the x,y plane. Schellbach divided the domain
into triangular regions and minimized the area of exactly half the number of triangles
(shaded). Although he minimized only the "half-area", the answer was correct for the
entire problem. This is an example of a 2-dimensional problem, where the z coordinate
of the surface depends on the x,y coordinates.

Courant [2) worked on the torsion of noncircular shafts in 1943, doing the first
problem in the modern finite element sense. Argyris [3) worked on aircraft structural
CHAPTER 1. INTRODUCTION & HISTORICAL REVIEW 5

Figure 1.4: Brachistochrone solution, due to Leibnitz.

z(x,y)

Figure 1.5: Closed curve in 3-D space.

0
y_2
0 x_2 x_l xo xl x2
--
X

y_l
Yo
yl
y2
y3
y4

!y
Figure 1.6: Finite element mesh, due to Schellbach.
CHAPTER 1. INTRODUCTION & HISTORICAL REVIEW 6

methods in the British aircraft industry in the late 1940's and 1950's. The field was not
really "ripe" at that time, however, because the digital computer had not yet matured.

The paper which has had the largest practical impact on the field of finite elements
was "Stiffness and Deflection Analysis of Complex Structure" by Turner, Clough, Mar-
tin and Topp [1] which appeared in the September, 1956 issue of Journal of Aeronautical
Sciences. They developed two-dimensional elements to model sheet material in aircraft
box beams. A result was the constant strain triangle and a rectangular element . The
constant strain (Turner) triangle survives today and is still an excellent tool to illustrate
the theory.

This early work on modeling structures was primarily supported by the aircraft
industry, which had very difficult geometries. Other researchers, however, were busy
working with naturally discrete systems, and concentrated on supporting technology
such as assembly of stiffness matrices, equation solvers and eigenvalue extraction. One
should therefore distinguish between the unique modeling aspects in finite elements,
in contrast to the supporting methodology developed in other contexts . Some of the
supporting work dates back to the 1800's (Lord Rayleigh) and to the 1930's (Southwell) .

As civil structural geometries became more complex and earthquake resistance be-
came important, civil engineers turned to finite element models in the 1960's with such
ferocity that at the present time, more of the public domain computer programs and
text books are due to civil engineers than any other group of engineers .

Workers in field problems such as hydrodynamics, electricity and magnetism became


serious about finite elements in the late 1960's and early 1970's. Each of these groups
had already developed tremendous capability in numerical analysis and only needed to
add the modeling capability of finite elements.

New frontiers at present are in acoustics, fluid mechanics and electromagnetic the-
ory. The finite element method provides a new way to look at these problems.

1.5 EXAMPLES OF FINITE ELEMENT SOLU-


TIONS
A series of finite ele1nent projects will be briefly illustrated to show different ele-
ments embedded in different spaces. In each case, an analysis was proposed, executed
and documented with a product either encouraged or rejected for production.

The load ramp in Fig. 1. 7 is used to load trucks or railroad cars . The ramp typi-
cally goes from ground level to truck bed. Fork lift trucks drive up and down it . The
structure is welded aluminum with steel grating on the bed. The picture is a "wire-
CHAPTER 1. INTRODUCTION & HISTORICAL REVIEW 7

frame" drawing showing the center line of all beam elements and the outline of all plate
elements, with the grating absent. Such computer graphics is extremely important in
developing the model to detect errors and to visualize deflection results. Preproces-
sors can also provide hidden-line removal to give enhanced clarity for plate and solid
structures.

The result for this project was to reduce the aluminum weight by 10% while main-
taining the same load capacity and to certify the ramp to meet governmental military
performance criteria. The computer code used was SAP IV. The load ramp is in
production.

Figure 1.7: Load ramp. Welded aluminum plate and beams. (Courtesy of Brooks &
Perkins Corp.)

A hydraulic valve is made of gray cast iron and pressurized at 2500 psi (Fig. 1.8).
It was desired to increase the operating pressure to 3500 psi. The valve has many
intricate passageways, and after much thought, 1/12 of the body was modeled using
planes of symmetry. The results confirmed the location of high stress regions and led
to a better understanding of the stresses. Three-dimensional elements (solids) were
used. Both SAP6 and MSC /N ASTRAN were used for the analysis.

A cast aluminum mainframe for a Micro-Winchester disk drive is shown in Fig. 1.9.
The casting originally had a natural frequency in bending at 350 hz and this was raised
to 500 hz through redesign. The casting was modeled by beam and plate elements.
The codes used were SAP6 and MSC /N ASTRAN. Vibration modes were animated by
MOVIE.BYU. The design was produced in small quantities but the business venture
failed.

Transient heat conduction in a truck brake system is modeled in Fig. 1.10. The
brake rotor and hub are axisymmetric bodies but have been modeled using pie-shaped
solid elements. The goal was to determine temperature at a bearing location. The
design is currently in production. The code used was MSC /N ASTRAN.
CHAPTER 1. INTRODUCTION & HISTORICAL REVIEW 8

Figure 1.8: Hydraulic valve. Gray cast iron. 2500 psi internal pressure. (Courtesy of
HYDRECO, Inc.)

Figure 1.9: Micro-Winchester disk drive chassis. Aluminum casting. (Courtesy of


Irwin-Olivetti, Inc.)

cold {

Figure 1.10: Heat conduction in a truck brake system.


CHAPTER 1. INTRODUCTION & HISTORICAL REVIEW 9

1.6 EXAMPLES OF RESEARCH USING F.E.A.


Several research projects will be described, some of which are doctoral dissertation.
A combination of commercial and personal computer programs were used.

A long trailing cable is towed by an airplane for use as an antenna, or possibly with
a capsule on the end to retrieve astronauts downed on an ice floe (Fig. 1.11). The
airplane flies in a circular orbit . John Russell [5] studied this problem using specially
developed nonlinear line elements. Ten of these "one-dimensional" elements were used
to model the cable in three-dimensional space. The cable elements were subjected to
inertial, elastic, gravitational and aerodynamic forces. The nonlinearity was due to
large deflections from the initial, vertical direction . The results showed the intricate
shape the cable takes during rotation, and revealed both static "jumps" and flutter
instabilities.

Figure 1.11 : Aircraft with trailing cable 3353 m (11,000 ft) long.

Bodies such as submarines and balloons which are immersed in a fluid experience
an effect called "added mass" when they accelerate. These bodies carry some of the
external fluid with them as they move. It is important for dynamic studies, such
as ballast drops in balloons, to understand the total inertial force . Calculation of
added mass of high-altitude research balloons has been done with a combination of
finite element, boundary element and acoustic technology [6]. A family of external
shapes (Fig. 1.12) were studied, corresponding to different inflation ratios . The balloon
structure was meshed using I-DEAS (Fig. 1.13), and the rigid body modes of the balloon
were calculated with MSC /N ASTRAN. The results for added mass are stated in a ratio
of added mass/displaced mass of air (Fig. 1.14). The added mass of the inflated balloon
CHAPTER 1. INTRODUCTION & HISTORICAL REVIEW 10

16

14

12
~

a
~ 10

GJ 8
-
N
~ 6
~
0 4
z
2

~~~--~--~--~
0 2 4 6 8
NORMALIZED RADIUS

Figure 1.12: Family of balloon shapes.

Figure 1.13: Finite element mesh for fully inflated balloon.


'

CHAPTER 1. INTRODUCTION & HISTORICAL REVIEW 11

is greater in the vertical (on-axis) direction than lateral direction because of the relative
bluntness that way.

0.7
E-<
z
~
u 0.6
&
~
vertical
(on axis)
~
~
0
u
<t::
0
~
0.5
&
~
~ horizontal
0 ~
u
<t::
~ 0.40.0 0.2 0.4 0.6 0.8 1.0
P::::

~
INFLATION FRACTION

Figure 1.14: Added mass coefficent for high altitude balloon

Optimization of turbine and compressor blades was studied by Gans and Anderson. [9]
A rotating blade was modeled with finite element plate elements (Fig. 1.15) and a
specially written set of DMAP statements was used with MSC /N ASTRAN to meet
frequency constraints while minimizing weight. It was found possible to raise frequen-

Figure 1.15: Finite element mesh for compressor blade. Plate elements.

cies by 30% with little increase in weight by redistributing the metal thickness of the
blade.

The "geometric strain" method for optimization was developed by Suh and Anderson[7]
for shape optimization. An example problem was a valve spring retainer (Fig. 1.16), a
component of an automobile engine. Using an optimality criterion, the authors were
able to reduce the weight of a titanium valve spring retainer by 30% while keeping
CHAPTER 1. INTRODUCTION & HISTORICAL REVIEW 12

Figure 1.16: Original shape of valve spring retainer.

the same fatigue life. The proposed new shape (Fig. 1.17 differed substantially from

Figure 1.17: Optimal shape of valve spring retainer.

existing shapes.

Optimization was done for the shape of two wedges in contact, by Park and
Anderson[8]. Two wedges are shown in Fig. 1.18. The bottom one is stationary and
the top one in pressed into it with a pressure of 300 MPa. The upper surface of the
top wedge is fixed in size, but the two sides are free to be resized.

The geometric strain method changes the singular stress in the baseline design
(1.19) into the nonsingular stress field shown in Fig. 1.20. The method gives a rather
uniform stress, near the 300 MPa along the free face of the upper wedge.

Finally, on the light side, the author has conjured up a hero for all analysts in need
of solutions to tough problems. Finite-Element Man rushes to solve the next problem
(Fig. 1.21 )! On the serious side, there actually are a lot of biomechanics projects being
CHAPTER 1. INTRODUCTION & HISTORICAL REVIEW 13

300MPa

Figure 1.18: Two wedges being pressed together

Figure 1.19: Baseline design with theoretically infinite stress at corner


CHAPTER 1. INTRODUCTION & HISTORICAL REVIEW 14

33l.
~ 307.
::s
284.
6" 26l.
en- n8.
00
~ 21. 4.
......
.: 191.
c!j
.& 168.
u~
t:: 1At;;
J. ' lv .

~ 121.

ss
~
~
98.
71 .

::s
5. 1.

Figure 1.20: Optimized shape of wedge, with uniform stress

done with F .E .A . Analysis of bone, blood vessels, and tissue are routinely done.

Figure 1.21 : Finite element man to the rescue!

1.7 STIFFNESS VS. FLEXIBILITY FORMULA-


TION
Modern structural finite element theory is based on the use of stiffness rather than
fiexibilities. An element is viewed as a conductor of forces from one node to another
and the stiffness is a measure of how much force can be conducted per unit nodal
CHAPTER 1. INTRODUCTION & HISTORICAL REVIEW 15

displacement .
{P} = [I<]{u}
The loads P i and displacements Ui appear in matched pairs where their product
creates an energy. The P i and U i can be pairs of force/translation or moment/ angle.
The term ]{ij refers to the force Pi developed at the i-th degree of freedom due to a
unit displacement at only the j-th degree of freedom Uj.

The flexibility of a structural system is characterized by C ij, which refers to the


displacement at the i-th degree of freedom due to a unit force at only the j-th degree
of freedom u j.
{ u} = [C]{P}
The flexibility matrix is the inverse of the stiffness (when neither is singular) :

[C] = [K]- 1

In comparing the two ways to characterize a structural system, one finds that:

1) Flexibilities are far easier to measure in a laboratory;

2) Stiffnesses are easier to handle theoretically.

The emergence of stiffness as the major analytical tool has partially isolated the ex-
perimentalist from the theoretician.

A major computational advantage of stiffnesses is their more "local" nature- the


stiffness matrix has many zero terms while the flexibility matrix is fully populated.
Also, stiffnesses have an additive nature in the assembled system- stiffnesses represent
paths of force that can be added.

1.8 EQUILIBRIUM APPROACH VS. ENERGY


APPROACH
For some simple finite elements such as the line element (Chapter 3) and the Turner
triangle (Chapter 4), one can determine the stress fields by equilibrium and can develop
elements by using equilibrium arguments both for internal and external balance of
forces . For most elements, however, the stress field is statically indeterminate and one
does not have the foggiest idea about load paths. In these cases, energy methods are
required, such that proper energy balance ensures the correct stress field . We will start
our study of finite elements using simple equilibrium arguments and will then progress
to an energy balance for the remaining elements.
CHAPTER 1. INTRODUCTION & HISTORICAL REVIEW 16

1.9 DISPLACEMENT METHOD VS. FORCE AND


HYBRID METHODS
The emphasis in this book and in almost all commercial FEA codes is on the
displacement method, where the displacement field is assumed and the virtual work
theorem, potential energy theorem or Galerkin's method is used. An alternate for-
mulation based on complementary energy is called the force method. A combination
of the potential energy and complementary energy ideas is called the hybrid method.
The hybrid method is used in special elements such as those for a crack tip. This text
will concentrate on the use of the displacement method and the virtual work approach.
Alternate methods are well described by Zienkiewicz [10] and Hughes. [11]
Bibliography

[1] Williamson, F., "A Historical Note on the Finite Element Method," Int. Jour. for
Num. Method in Engr. ) 1980, pp. 930-934.

[2] Courant, R., "Variational Methods for the Solution of Problems of Equilibrium
and Vibration," Bull. Am. Math. Soc.) vol. 49, 1943, pp. 1-23.
[3] Argyris, J. H., "Energy Theorems and Structural Analysis," Butterworth, Great
Britain, 1960. Reprinted from Aircraft Engr., 1954-1955.
[4] Turner, M. H., Clough, R. W., Martin, H. C. and Topp, L.I., "Stiffness and
Deflection Analysis of Complex Structures," Jour. of Aero. Sci.) vol. 23, 1956, pp.
805-823.

[5] Russell, J. J. and Anderson, W. J., "Equilibrium and Stability of a Circularly


Towed Cable Subject to Aerodynamic Drag," Jour. of Aircraft, vol. 14, no. 7,
July 1977, pp. 680-686.
[6] Anderson, W. J., and Shah, Gourang, "Added Mass of High-Altitude Balloons,"
Proceedings of the AIAA International Conference on Balloon Technology, Albu-
querque, NM, October, 1991.
[7] Suh, Myung, and Anderson, W. J., "Application of Geometric Strain Method
to Shape Optimization of Spring Retainer," Proceedings of First International
Conference on Computer Aided Design of Structures, Southampton, UK, June,
1989.

[8] Park, Jungsun, "Selected Problems in Structural Mechanics," Ph. D. Dissertation,


The University of Michigan, Ann Arbor, MI, 1993.
[9] Gans, H. D., and Anderson, W. J., "Structural Optimization Incorporating Cen-
trifugal and Coriolis Effects," AIAA Journal, December, 1991.

[10] Zienkiewicz, 0. C., and Taylor, R. 1., "The Finite Element Method," Edition 4,
McGraw-Hill, New York, 1988.

[11] Hughes, T. J. R., "The Finite Element Method," Prentice-Hall, Inc. New Jersey,
1987.

17
Chapter 2

DERIVATION OF FINITE
ELEMENTS

Before embarking on development of specific finite elements, let's spend a chapter


about derivation of finite elements in general. After this chapter, we will get busy with
specific cases.

Finite elements can be constructed by a variety of methods . Historically, the


method was conceived using an equilibrium approach, where forces and displacements
were found for simple elements. This was feasible for statically determinate elements
in which the researcher could identify load paths. It was soon found that energy meth-
ods worked better for statically indeterminate elements. In the past 20 years, energy
methods (virtual work, potential energy, complementary energy) have dominated the
derivation of new elements.

Recently, a fundamental problem in element behavior, called "locking" has been


found, where the elements are too stiff. This seems to be an inherent problem in the
energy methods. Some researchers are using one of the methods of weighted residuals
(Petrov-Galerkin) to minimize locking. Weighted residual methods, in general, attempt
to minimize the mathematical error in the problem statement over the volume of the
element.

Let us classify the different methods that dominate structural element derivation:

Equilibrium
displacement method
force method

Energy Balance
virtual work
minimum potential energy
minimum complementary energy

18
CHAPTER 2. DERIVATION OF FINITE ELEMENTS 19

hybrid (Reissner) methods

Weighted Residuals
Galer kin (classical) method
Bubnov-Galerkin method
Petrov-Galerkin method

The most universal of the three categories above is the method of weighted resid-
uals, which applies to a broad class of problems, whether structural, electrical, fluid
or thermal. All of the methods above depend on certain common building blocks,
including:

choice of element
underlying theory
shape
number of grids
interpolation method
material law
strain-displacement law

In addition, the energy and weighted residual methods require integration over the
element volume, and hence a choice of an integration approach . The most common
approach at present is a type of numerical integration called Gaussian integration.

2.1 COMMON IDEAS FOR STRUCTURAL FI-


NITE ELEMENTS
2.1.1 Interpolation
Interpolation is a means of estimating functional values interior to a region where
the function is known on specific points around the boundary. In a one-dimensional
exa1nple, this is done when a person uses a table of values for the sine function, and
must interpolate for a value at an angle lying between the tabular values . (The use of
electronic calculators has eliminated the need for this interpolation!) In two dimensions
(Fig. 1), the functional value can be found from the values at the four grid points.

There are many ways to interpolate. One of the most efficient ways is to express
the functional value u(x,y,z) as an additive sum of "shape functions" N (x,y,z) with
coefficients ui :
(2.1)
where the shape functions are specially created. The shape functions have a unit value
at a "home" node, and a zero value at all other nodes. For a point interior to a
CHAPTER 2. DERIVATION OF FINITE ELEMENTS 20

f(x,y)

Figure 2.1: Interpolation of f(x) over quadrilateral region.

finite element, the shape functions provide the emphasis on the various grid values,
emphasizing those grids that are nearest to the point (x,y,z) of interest.

Shape functions are chosen differently for each type of element. Polynomials are
used in preference to transcendental function. Beams use cubic polynomials, plates use
cubics, solids use linear or quadratic polynomials.

Engineers often have trouble with the concept of an interpolating polynomial,


whereas mathematicians use them with more ease. The author's belief is that en-
gineers are more comfortable with extrapolation, in which we project a local state
(elevation at a point on a hiking path, say) to a remote position by using a combi-
nation of a functional value and a derivative. The interpolation discussed in Eqn. 2.1
uses no derivatives, however.

2.1.2 Material Law


The material (constitutive) law brings in the physical behavior of the material.
Common engineering materials such as steel and aluminum have a substantial region
of behavior that is linear. Engineering design practice often restricts the performance
of the material to that linear range. As a result, linear material laws are very important
in structural mechanics, as in contrast to fluid mechanics, for instance. The following
form can be used for linear materials, as well as a quasi-linear description of nonlinear
behavior (where the matrix [G] depends on the strains):

{a}= [G]{E}
CHAPTER 2. DERIVATION OF FINITE ELEMENTS 21

where {a} is the stress vector, with components

ax
ay
az
{a}=
T xy
Ty z
T zx

and { t:} is the strain vector, with components

tx
ty
tz
{t:} = {xy
{yz
{zx

and [G] is the generalized Hooke's law. For the case of linear, isotropic material such
as metal, plastic, etc.:

A+2G A A 0 0 0
A A+2G A 0 0 0
A A A+2G 0 0 0
[G]=
0 0 0 G 0 0
0 0 0 0 G 0
0 0 0 0 0 G
where the Lame' constant A is defined:

A= vE
- (1+v)(1-2v)
Only two material constants are needed for the isotropic case, E and v or alternatively,
A and v.

2.1.3 Strain-Displacement Law


For both small and large strain, the Green strain tensor relates strain and displace-
ment. In the engineering notation, Green's strain tensor is:

- ou 1 [( ou )2 ( ov )2 ( ow )2]
Ex - OX + 2 OX + OX + OX

_ -ov + -1 [(ou)
ty - - 2 + (ov)
- 2 + (ow)
- 2]
oy 2 oy oy oy
CHAPTER 2. DERIVATION OF FINITE ELEMENTS 22

= aw ~[(au)2 (av)2 (aw)2]


tz az + 2 az + az + az
l x = au + av + (au au + av av + aw aw)
y ay ax ax ay ax ay ax ay
I z = av + aw + (au au + av av + aw aw )
y az ay ay a z ay a z ay a z
I zx = aw + au + (au au + av av + aw aw)
ax az az ax az ax az ax
For many engineering problems, with small strains and displacements, the above
law reduces to the linear strain-displacement equation:
8
Ex 8x 0 0
a
0 0
Ey
tz 0
ay
0 8 { u(x,y,z)}
{c} = a __
8z
0
v( x,y, z )
lxy 8y 8x
8 8 w( x, y,z)
lyz 0 8z ay
l zx __ a
8z 0 ax

The strain-displacement law is given the symbol [D].

The full Green strain tensor is necessary for buckling problems , forming problems
and other large-strain, large-displacement problems .

2.2 SPECIFIC METHODS


2.2.1 Equilibrium
Equilibrium is often thought of in terms of Newton's second law:

{F} = [M]{u}

where the forces {F} are external forces. In finite element studies of elastic bodies, a
more detailed version is to be used , where one seeks a balance between externally ap-
plied loads and internally developed structural forces. In discrete form , the equilibrium
equation will become

[M]{u} + [B]{u} + [I<]{u} = {F} external

The internal structural forces [K] { u} play a major role . Finding the matrix [K] is
perhaps the greatest challenge in structural finite element theory. A second challenge
is to determine a discrete form of the external forces, i. e., changing from external
pressures to discrete forces at grids.
CHAPTER 2. DERIVATION OF FINITE ELEMENTS 23

The laws of equilibrium can be directly applied in many static problems. If the
finite element is statically determinate, the engineer can identify load paths and make
assumptions about force balances. For more complicated elements, such as solids , it is
almost impossible to figure out candidate forces at the nodes. One can, however, use
an energy balance or a form of Galerkin's method, to be discussed below.

2.2.2 Virtual Work


In general , one wants to find a balance of forces between external loads and inter-
nally generated structural forces. The statement of virtual work shows that the sum of
work done by external forces and internal forces during a virtual displacement equals
zero:
8Wexternal + 8Winternal = 0
A modified version of this statement (depending on a proof that the increment of
internal work is equal to the negative of the corresponding increment of strain energy
8U) shows that the increment of external work equals the increment of strain energy
during a virtual displacement:
8Wexternal = 8U
This equation yields immediate, useful results for developing the equilibrium laws for
finite elements, as well as for determining the equivalent grid loads.

2.2.3 Minimum Potential Energy


For systems that possess energy-conserving materials (that don't dissipate energy
during deformation) and that possess conservative external loads (that don't create
work on a closed cycle of motion) , the virtual work theorem reduces to the principle
of minimum potential energy. In this case, the external forces can be represented by a
work potential W :
{F} external = -{\7} W
and during a virtual displacement from equilibrium, we find that:

{V}(U + W) = {0}

(i. e. the potential energy is "stationary" ).

The potential energy theorem is useful for generating elements, and for proving
convergence to the correct answers . One of the early breakthroughs in finite elements
was when Melosh realized that the finite element displacement method was governed by
potential energy ideas, and that successive sets of finite element models with "nested"
grids converge to the proper answer "from below" in terms of displacements , stresses
and strains. (The proof was limited to particularly nice elements which have sides that
conform to neighboring elements. )
CHAPTER 2. DERIVATION OF FINITE ELEMENTS 24

2.2.4 Minimum Complementary Energy


Because the potential energy methods converge from below on stress, strain and
displacement, there has been interest in developing complementary energy models for
finite elements . These have been somewhat successful, but have not been incorporated
into the major commercial finite element codes to date. A more successful approach,
to be discussed next, was an attempt to average the overshooting and undershooting
effect of the two minimization theorems.

2 .2.5 Hybrid Methods (Reissner)


By incorporating some of the best features of both minimum potential and comple-
n1entary energy theorems, Reissner introduced an energy functional that is stationary
at the equilibrium position. The approach is named the "hybrid" finite element for-
mulation. Convergence to the correct stresses, strains and displacements is faster than
either of the minimization theorems. These methods have proved very useful in some
special-purpose situations, such as stresses near crack-tips.

2.3 GALERKIN METHOD (CLASSICAL)


The classical Galerkin method [1] has been used since 1915. It was used for a wide
variety of structural problems, and helped in some fields such as aeroelasticity, where
it is valid even though the problem is not energy conserving. The method was clarified
by Duncan, Fraser and Collar in a series of British reports [2]. For energy-conserving
problems, and when "weighting functions" are carefully chosen, the method is the same
as the Rayligh-Ritz approach (minimum potential energy) in structural mechanics .

We will use a simple example to explain the differences in the several Galerkin
methods [1] . Consider a simple string (Fig. 2) spanning 0 S x S 1, under constant
initial tension T . The differential equation and a pair of typical boundary conditions
are:
8 2w(x)
T(x) 8x2 + f(x) = 0 (0 S X S 1)
w(O) = wo
aw
ax (1) = el
If the loading f( x), the elevation at the left end, and the slope at the right end are
so given, there will be a unique answer for the deflected shape of the string. This is
calied the strong form of the problem.

Galerkin proposed that an approximate answer to the problem could be found by


assuming a series of functions for w( x) and minimizing the error found in the differential
CHAPTER 2. DERIVATION OF FINITE ELEMENTS 25

T T
______.

w(x) f f(x)
~t
I
~ ~I
/
Lle1
X
0 ~ 1

Figure 2.2: An initially stretched string under lateral loads.

equation. His original idea was to choose functions that exactly satisfied the boundary
conditions, so that there was no error to study on the boundary.

The assumed functions could be taken to be a set of functions Wn (x) satisfying


homogeneous boundary conditions, and a single function w0 ( x) to satisfy the nonho-
mogeneous boundary conditions:
N
w(x) = wo(x) + 2: CnWn(x)
n=l

In this way, the constants Cn can be adjusted to give the least error over the domain
(0,1) of the problem.

How does one minimize the error in the differential equation over the domain? The
error at any point x will be called t:( x) and should be driven to zero:

8 2w(x)
T(x) ax 2 + f(x) _ t:(x) ~ 0 (0::; x::; 1)

One might try to make this error zero at certain points (this is a different method
than we want, called collocation). Rather, let us use weighting functions Wn and use
N of them to emphasize various parts of the domain. We will integrate the error using
each weighting function:

fl a2w(x)
Jo [T(x) ax 2 + f(x)]Wn(x)dx = 0 (n=l,2 .. N)

or, in simpler terminology:

l E(x)Wn(x)dx = 0 (n = 1,2 .. .N)

This leads to N algebraic equations in the N unknowns Cn


CHAPTER 2. DERIVATION OF FINITE ELEMENTS 26

The method is referred to specifically as Galerkin's method if the weighting func-


tions Wn (x) are taken to be the same as the assumed displacement functions Wn (x).
This means that the error is weighted in the same manner as the displacements are
assumed. It is this approach that Frazer et. al. [2] could show equivalent to Rayleigh-
Ritz in energy-conserving problems, because the integrand in the error minimization
has terms that are exactly the strain energy and the external work.

One of the problems with the classical Galerkin method was the need to choose
trial functions that satisfy all homogeneous boundary conditions (both on the function
and the derivative, in our example). This makes it difficult to create trial functions,
particularly in two and three dimensions, and limits the generality of the method. This
requirement is relaxed in the next method discussed.

2.3.1 Bubnov-Galerkin Method


The Bubnov-Galerkin method [1], adds a term to the error minimization. Again,
we look at the string problem, with differential equation and boundary conditions:

8 2w
T(x) a2 x + f(x) =0

w(O) = Wo

aw
ax (1) = el
This time, we will allow trial functions that satisfy only the lower order (geometric)
homogeneous boundary conditions, but which can violate the higher order (natural)
homogeneous boundary conditions. The variational statement will contain a boundary
term that will tend to drive the error in the equation at the boundary to zero, as well
as in the interior.
[ 1 8 2w(x)
Jo [T( x) ax 2 + f( x )] Wn(x)dx + w(l)B1 0 (n=l,2 .. N)

In this method, the weighting functions Wn are taken to be the same set of functions
as the trial functions Wn. Again, N algebraic equations for theN unknowns en are found.
The Bubnov-Galerkin method also can be shown to be the same as the Rayleigh-
Ritz method for energy-conserving systems, and can be shown to be equivalent to
virtual work methods for general problems. The method described next will relax this
condition on weighting functions.

2.3.2 Petrov-Galerkin Method


There are cases, such as the "locking" problem in structural mechanics and flow
direction effects in fluid mechanics, where it helps to use weighting functions other than
CHAPTER 2. DERIVATION OF FINITE ELEMENTS 27

the trial functions. Although trial functions were the best choice for weights in energy-
conserving problems, there is good reason to generalize the method. Many researchers
feel that this method is the "wave of the future."
Bibliography

[1] Galerkin, B. G., Vestnik In z henerov, Vol. 1, 1915 (pp. 897-908) . For a discussion
of this paper, see Sokolnikoff, I. S., Mathematical Theory of Elasticity, McGraw-
Hill Book (pp . 413-416) .

[2] Frazer, Duncan and Collar, Great Britain Aircraft Research Council Reports and
Memoranda# 1798, 1848 and 1888.

[3] Hughes, T. J . R., The Finite Element Method," Prentice-Hall, Inc. New Jersey,
1987 (pp . 2-9) .

28
Chapter 3

LINE ELEMENT & ASSEMBLY


PROCESS

3.1 CREATION OF -LINE ELEMENT BY EQUI-


LIBRIUM
A two-node constant area line element is to be created (Fig. 3.1) . The element has
two nodal degrees of freedom u 1 and u 2 with associated forces f1 and f2 . The general
rule is that the product of nodal force and displacement must yield energy.

Figure 3.1: Two-node, constant area line element.

This simple element can illustrate several important features of finite element the-
ory. Since we have not yet developed a general energy approach, we must use equilib-
rium ideas instead. For instance,

!2 = - fl (3.1)
from equilibrium in the x direction. The one-dimensional stress-strain law is
O"x(x) = E Ex(x) (3.2)
and the one-dimensional strain-displacement law is
d
Ex (x) = dx u(x) (3.3)

29
CHAPTER 3. LINE ELEMENT & ASSEMBLY PROCESS 30

where u( x) is the displacement field.

If we assume that strain Ex is constant in the element , we can approximate the


strain by
6.L
E~- (3.4)
L
where the change in length is dependent on lit and u 2 :

Ex = --L- (3.5)

This turns out to be exact for a constant area element. It also implies (with Eqn. 3.2)
that stress is constant.

From equilibrium at the nodes (Fig. 3.2),

(3.6)

J2 = axA (3.7)
Rewriting this in terms of strain and finally displacement, one has

f~------ IDf2
_____. cr XA cr XA _____.
---------

Figure 3.2: Equilibrium of external nodal forces with internal stresses.

ft = -AE(u2- ut)/ L (3.8)

!2 = AE(u2- ut)/ L (3.9)


which can be put in matrix form:

(3.10)

This is the fundamental relation in finite element theory, the load-deflection law. The
matrix relating load and deflection (including the scalar factor) is defined to be a

l
stiffness [k]:
[k] = AE [ 1 -1 (3.11)
L -1 1
We were fortunate to obtain this answer by equilibrium methods. We were even
more fortunate in that the answer is exact , for the constant area element.
CHAPTER 3. LINE ELEMENT & ASSEMBLY PROCESS 31

Note that the determinant of the stiffness matrix for an element is zero, i.e., the
matrix is singular. This is always true for an element (not an assembly) and indicates
the presence of rigid body modes, which are discussed later. Rigid body modes are
needed for individual elements to move to their final position in a deformed structure.

The stiffness matrix has been developed using an "element" coordinate system, i.e.
an underlying coordinate system with origin at the left end of the particular element.
We have not actually made use of the special location of the origin, however , and
would have obtained the same result if another origin were used. It is generally true
that stiffness matrices are invariant under a translation of a Cartesian coordinate origin.
This is fortunate, because when one needs to assemble a number of elements, one must
use a common, or "basic", coordinate system that holds for all elements, and no single
system can be specially located for each element. In higher dimensional spaces, where
rotations of coordinate systems are possible, it is found that stiffnesses do depend on
the orientation (rotation) of the coordinates.

3.2 ASSEMBLY OF LINE ELEMENTS


The assembly process for finite element theory is straightforward and powerful. A
good way to visualize the process is to look at two distinct line elements and then to
join them by a fictitious 1 pin as in Fig. 3.4.

L
~ i
+____. fu_.
L....--.....:u,-:---------l-
t-+
-
u3e t~
4e
le 2e

Figure 3.3: Two line elements to be joined.

{ j~ } = E~~, [ ~ 1 ~1 ]{ ~~: } (3 .12)

{ j: } = El:2 [~1 ~1 ]{ ~:: } (3.13)

The way to assemble is to create a matrix problem large enough so that each element
can be "embedded" in it. Realizing that upon a assembly one has u 3 e = u 2 e and that
1
It is important to recognize at the outset that the assembly of elements is actually a welding of the
elements together, rather than pinning them together. The current artifice is used to get a feeling for
force equilibrium at the joint. After this discussion, the student must visualize nodal connections as
welded joints, unless one is actually talking about pin-ended trusses or unless certain "end releases" are
applied. The latter happens in the case of slotted joints or actual pins in beam and plate structures.
CHAPTER 3. LINE ELEMENT & ASSEMBLY PROCESS 32

Figure 3.4: Fictitious pin at node connection.

there will be only 3 independent coordinates, one writes

U=[
3 ~~
]{~:: } (3.14)

3x3 3

The equations for the first element can be written, by adding a trivial equation:

(3.15)

Likewise, the second element is described by

0
E2A2/ L2 (3.16)
-E2A2/ L2
Equations 3.15 and 3.16 are of the mathematical form

{X}= [A]{ z} (3.17)

{Y} = [B]{ z} (3.18)


and can be added, using the distributive nature of matrix multiplication and addition:

{X}+ {Y} = ([A]+ [B]) {Z} (3.19)


"-----..----'
a ssem bl ed sti ff ness

Some people say "stiffnesses just add," and this process is what is really meant.

In our example:
CHAPTER 3. LINE ELEMENT & ASSEMBLY PROCESS 33

The stiffness matrix has assembled nicely and the only displacements which appear are
the independent degrees of freedom. The force vector can be simplified by looking at
the force on the fictitious pin at the joint (Fig. 3.4). There is in general an external
force P on the pin, which satisfies

p = !2 + !3 (3.21)

hence we define the new symbols [I<] and { u}:


assem bled sti j f ne ss
~

{ ~ } = [ K ]
~
{ ~:: }
(3.22)

independent d. o.f.

Fortunately, the force vector consists of only externally imposed loads! This is a
wonderful situation, because the finite element method then suppresses the internal
forces and the need to know them. This is comparable to Lagrange's equations in
analytical dynamics, which also suppress internal forces of constraint and allow easier
solution.

One finally defines the external load vector symbol { P} and writes:

{P} = [K]{u} (3.23)


...._,_..,
ext erna lly applied loads

where the forces and displacements are typically renumbered to remove redundant
degrees of freedom:

This simple matrix Eqn 3.23 is the heart of static finite element solutions. The equa-
tion still needs to be modified before solution since the assembled stiffness matrix is
"singular" at this stage. This will be discussed later in a chapter on modification of
equations.

A similar assembly process can be done in electrical and fluid circuits and field
problems in many branches of science and engineering. For each class of problem, the
law used for assembly is appropriate to the variables , such as Kirchhoff 's current law
in electrical circuits and conservation of energy in heat conduction problems.
CHAPTER 3. LINE ELEMENT & ASSEMBLY PROCESS 34

3.3 EMBEDDING THE LINE ELEMENT INTO


A 3-D SPACE
The preceding development is for a 1- D element embedded in a 1- D space. The line
element , however , is often used as a pin-ended truss member in a 3-D assembly (space
truss). It is instructive to look as the process of creating the expanded, 3-D stiffness
matrix.

Consider the line element in Fig. 3.5. An element coordinate system is created at
the left end as in Fig. 3.6. Previous work gives us the relation between axial motion
and forces in the element system of coordinates. The stiffness is to be a 6 x 6 matrix
relating the 3 translations at each node:

ll Z

Figure 3.5: Line element in 3-D space

f xl EA/L k12 k13 -EA/L k1s k16 Uxl


jyl k21 k22 k23 k24 k2s k26 Uyl
Jz1 k31 k32 k33 k34 k3s k36 Uzl
< (3 .24)
f x2 -EA/L k42 k43 EA/L k45 k46 Ux 2
jy2 ks1 ks2 ks3 ks4 kss ks6 Uy2
Jz2 k61 k62 k63 k64 k6s k66 Uz2
~ ~
6 6x6 6

The remaining entries in the matrix will be found from physical reasoning. The
line element has stiffness only along its axis and will support no lateral (shear) forces .
This means that a displacement in the yz or zz directions at either end will generate no
forces in the Yl or zz directions:

k ij = 0 (i = 2' 3' 5' 6 j = 2,3,5,6) (3.25)


CHAPTER 3. LINE ELEMENT & ASSEMBLY PROCESS 35

and we have
EA/L -EA/L
0 0 0 0
0 0 0 0
[ k
] local
-EA/L
0 0
EA/L
0 0
(3.26)

0 0 0 0

(L,O,O)
/

k_------ - - - - - - -

Figure 3.6: Local coordinate system for line element.

In the linear theory of elasticity, for a line element, displacement in the yz and z z
directions will generate no net stretching of the element. This is shown in Fig. 3. 7
where a line element is subjected to a small lateral displacement in the z 1 direction,
at the right end, say. The deformed length of the element is the hypotenuse of the
triangle

~E
L
Figure 3. 7: Stretch in line element due to lateral deformation c.

Ldeformed JL2 + c2
1 1
L[1 + 2(c/L) 2 + S(c/L) 4 + ...] (3.27)

When terms of order ( c/ L )2 and higher are dropped, the deformed length is unchanged
to first order in c/L:
Ldeformed ~ L (3.28)
Therefore
k ij = 0 (i = 1' 4 j = 2, 3, 5, 6) (3.29)
CHAPTER 3. LINE ELEMENT & ASSEMBLY PROCESS 36

and
EA/L 0 0 -EA/L 0 0
0 0 0 0
0 0 0 0
[ k
] local
-EA/L 0
0
0 EA/L 0 0
0 0 0
(3 .30)

0 0 0 0
Finally, linear elastic stiffness matrices must be symmetric (Maxwell's reciprocity
theorem) :
kij = kji (i = 1 - t 6, j = 1 - t 6) (3.31)
and therefore 2

EA/L 0 0 -EA/L 0 0
0 0 0 0 0 0
0 0 0 0 0 0
[ k
] local
-EA/L
0
0
0
0 EA/L 0 0
0 0 0 0
(3 .32)

0 0 0 0 0 0
This matrix is widely used for truss, line and spring elements in F. E . programs .
To be used in an assembly however, one must transform the matrix from the element
(local) coordinate system to a basic coordinate system. This is done automatically for
the user by all general F . E. programs. See Chapter 18 for details on the element-basic
coordinate transformation.

3.4 LINE ELEMENT WITH VARYING AREA


Equilibrium methods have been completely successful in our work to this point.
Now we will consider a slightly more complicated element which will cause some trouble
in interpreting the stiffness matrix (Fig. 3.8) . The two-noded, varying-area line element
again has

(3 .33)
We again have the stress-strain law

(3 .34)
and the strain-displacement law
2
Perhaps some readers can see this result by inspection , but the logic used here is important .
Problem 1 at the end of this chapter is recommended as a test of your comprehension of the stiffness
matrix .
CHAPTER 3. LINE ELEMENT & ASSEMBLY PROCESS 37

Figure 3.8: Varying-area line element.

du
Ex=- (3.35)
dx
For this element, we avoid assuming constant strain a priori, because we know that
it cannot be so. (We have a constant force transmitted through a varying area, which
must lead to varying stress and strain.) A general way to attack such a problem (and
the reason the method is called the "displacement" method) is to assume a displacement
field such as

u(x) = q1 + q2x + q3 x 2 + q4 x 3 + ... (3.36)


displacement function
3
or, equivalently, to interpolate the displacement between the nodes using

(3.37)

where the functions of x on the rhs are called "shape functions", or "interpolation
functions."

When there are only two nodal coordinates u 1 and u 2 , it can be shown that the
displacement function can have only two independent constants.

(3.38)

which is a linear displacement model and leads to


d
Ex(x) dx (ql + q2x) (3.39)
q2 (constant strain model) (3.40)
and we are back to a constant strain model whether we like it or not! We are left
with an apparent paradox. The assumed displacement field causes a uniform stress
but results in a nonuniform axial force through the element (because of its varying
area).
3
It is possible to describe the internal displacement field either by displacement functions or by
shape functions. Each has its advantages. There is a unique relation between the two. See Chapter
5 for further discussion on interpolation.
CHAPTER 3. LINE ELEMENT & ASSEMBLY PROCESS 38

3.4.1 False Start


If we attempt to define nodal loads to satisfy local equilibrium, we might have a
free body diagram as in Fig. 3.9, where

Figure 3.9: Failed attempt to satisfy internal equilibrium.

-o-x(O)At (3.41)
du
-E-(O)At (3.42)
dx
and
(3.43)
The shape function in Eqn. 3.37 is useful to represent the internal displacement so
that duj dx above can be calculated:

(3.44)

(3.45)
In matrix form, we have:

(3.46)

but this is unacceptable , because

1) Nodal equilibrium is violated, !2 -1- !1


2) The stiffness matrix is not symmetric and a perpetual motion machine has been
created!

The only redeeming feature of the above development is that the internal equilib-
rium of the element has been satisfied. It has been at the expense of global equilibrium,
however. The nodal forces do not equilibrate.
CHAPTER 3. LINE ELEMENT & ASSEMBLY PROCESS 39

Figure 3.10: Satisfaction of nodal equilibrium.

3.4.2 Proper Approach


An alternate, and correct , philosophy is to maintain nodal (global) equilibrium and
to violate internal equilibrium. For instance, one can imagine some average cross-
sectional area (Fig. 3.10) such that

-axA average (3.47)


u2- ul
- E( L ) A average (3.48)

u2- ul
J2 = E( L )Aaverage (3.49)
This leads to
(3.50)

Surprisingly, this is a satisfactory model of the varying area line element , even
though internal equilibrium of the element is violated! The displacement field leads
to a stress field which is nowhere in equilibrium in the interior of the element. This
is a common feature of elements to be developed by the displacement finite element
method , using virtual work and potential energy principles.

3.5 HOMEWORK
P roblen1 1
This question is intended to test your knowledge of the stiffness matrix , its definition
and physical interpretation. A linearly-elastic coil spring is to be treated as a single
finite element (Fig. 3.11). Nodal degrees of freedom are the axial displacement u 1 , u 3
and the rotations u 2, u 4 . The spring is 200 mm long and 40 mm in outside diameter.
It is n1ade of steel wire of 1 mm diameter.

Two experiments have been carried out to provide stiffness data. The left end of
the spring is clamped firmly. A displacement u 3 of 1 mm causes forces j 1 = - 100 N
and j 2 = -50 Nmm (u 4 = 0) A rotation u 4 of 1 radian causes forces j 1 = -50 N and
CHAPTER 3. LINE ELEMENT & ASSEMBLY PROCESS 40

Figure 3.11: Spring element. Nodal displacement.

!2 = -100 Nmm (u 3 = 0). (Assume the spring remains within the linear range during
this rotation.)

A) Using elastic symmetry, geometric symmetry and equilibrium concepts, construct


the stiffness matrix for this spring.

B) If two such identical spring elements are joined in series, find the stiffness matrix
for the assembled system. Use letters to represent any stiffness components not found
in A).

Problem 2

Assemble the stiffness matrix for the planar structure in Fig. 3.12 using symbols
~ and # to represent the stiffness terms of the triangles and line elements, respec-
tively. Number nodes as you wish. Use "compact" notation where each term in the
displacement and force vector represents 2 d.o.f. at a node.

1\ 1\

Figure 3.12: Triangles and line elements in assembly.

Problem 3
Find the exact stiffness matrix [k] for a two-node line element (Fig. 3.13) with
varying area A = A 0 (1 +Ex/ L). Use an equilibrium method. You will not get a
polynomial for the exact displacement field.

Problem 4

A truss structure is assembled as shown in Fig. 3.14. The nodes are numbered 1-3.
The elements are numbered with Roman numerals. Suppose the structure has been
assembled:
CHAPTER 3. LINE ELEMENT & ASSEMBLY PROCESS 41

Figure 3.13: Line element with varying area.

II

Figure 3.14: Truss structure.

F1 kn k12 k13 k14 k15 k16 ul


F2 k22 k23 k24 k25 k26 u2
F3 k33 k34 k35 k36 u3
(3 .51)
F4 k44 k45 k46 u4
Fs (symm) kss ks6 us
F6 k66 U6

a) What sign does k43 have? Why?


b) What sign does (k33k44- k5 4) have? Why?
Problem 5

A plane strain problem is modeled with 3 rectangular and 2 triangular elements as


shown in Fig. 3.15. Using compact notation, how many stiffness terms in the assembled
stiffness matrix are zero? Count both k ij and k ji terms in your answer.

Problem 6

Relate the constants q1 and q2 in the displacement function (Eqn. 3.38) to the nodal
displacements u 1 and u 2 , assuming shape functions such as in Eqn. 3.37. You must
recognize that u(O) u 1 and u(L) _ u 2 to succeed.
CHAPTER 3. LINE ELEMENT & ASSEMBLY PROCESS 42

1 2
._------~~------~
3

4 5~6
7 8 9

Figure 3.15: Plane strain problem


Chapter 4

CONSTANT STRAIN
TRIANGLE

4.1 PLANE STRESS


The three-dimensional theory of elasticity involves concepts as sketched in Fig. 4.1.
Stress and strain are actually tensors, but we will use a vector characterization and
will be content to deal with the linear transformations (mapping) of vectors. The
stress-strain and strain- displacement laws are 6x6 and 6x3 matrices, respectively. The
equations of equilibrium and boundary conditions are typically differential equations in-
volving the stresses and/or displacements. The compatibility relations are needed when
mapping from strains to displacements because there are 6 equations in 3 unknowns
and one would not get an answer otherwise (an overdetermined system) . Boundary
conditions can be given in terms of stresses or displacements.

strain-
dis pl.
< />i uy~
~ luzj
/
/ ------
77
-'.-'1
- .......
'
( "/ "
' ______ _....... .... ,_)
compatibility
........

boundary conditions

Figure 4.1: Elasticity in 3-D

For the specialized case of plane stress (Fig. 4.2), one has zero stress normal to the

43
CHAPTER 4. CONSTANT STRAIN TRIANGLE 44

plane, o-z = 0, and the strain in that direction is found to be

boundary conditions

{ crxps~~~~~- {Ex}<~~~~r
(jy
'txy
y
Yxy
)
{ux}u
Y

equilibrium equations >

Figure 4.2: Elasticity in 2-D. Plane stress illustrated. (Thin sheet.)

(4.1)

The transformations used to relate stresses, strains and displacements in plane


stress are:

~ x
Y
} = [ 1:~2 ~2 ~
l-v 2
1
1-v2
] { Ex
Ey
}
(4.2)
{
Tx y 0 0 G { xy

{~:.}=[1, f]{~:} (4.3)

Note that the stress-strain law involves constants, whereas the strain-displacement law
involves derivatives.

4.2 CONSTANT-STRAIN TRIANGLE


The constant-strain triangle and a companion rectangular element were first de-
veloped to study the sheet metal structure of aircraft [1 J. Let us consider the case
where a triangular element has been cut from sheet (Fig. 4.3). Imagine nodal forces
and displacements as shown.

Stiffness matrices relate nodal forces to nodal displacements. Our job is to find two
linear transformations missing in Fig. 4.3. One needs only then to proceed leapfrog-
fashion with four successive transformations to relate nodal displacements to nodal
forces.
CHAPTER 4. CONSTANT STRAIN TRIANGLE 45

Figure 4.3: Plane stress. Constant strain (Turner) triangle.

The unknown transformation between nodal forces and internal stresses can be
found by simple equilibrium arguments. 1 First, strains are assumed constant.

a (4.4)
Ey b (4.5)
{xy c (4.6)
The triangle is then considered to be embedded in a uniform sheet of material under
constant stress in one direction as in Fig. 4.4 (The other stresses can be considered
later and the results superimposed.) The exploded diagram shows the triangular ele-
ment with concentrated loads at the midpoint of each side. This is a possible type of
concentrated loading that would place the triangular element in equilibrium. A better
approach is to break each concentrated load into two equal parts and then to apply
half to each nearby node (Fig. 4.4). It can be seen that

for instance. One can carry this process out to find the other forces for this loading and
then to include O"x and Txy stresses. The result is a linear mapping (transformation)
from stress to nodal loads:
{!} == []{ (]"} (4.7)
where stands for "equilibrium."
1
This step is usually not possible in finite element derivations. The line element, simple beam ele-
ment and Turner triangle are the main exceptions. The element must be either statically determinate
(line element) or so simple that this relation can be guessed (Turner triangle).
CHAPTER 4. CONSTANT STRAIN TRIANGLE 46

t t t
cry

tttttttttt

Figure 4.4: Triangle element embedded in sheet

Figure 4.5 : Calculation of nodal loads from equilibrium arguments


CHAPTER 4. CONSTANT STRAIN TRIANGLE 47

We now solve for the linear relation between generic displacements and nodal dis-
placements . This is more difficult conceptually because an intermediate step develops.
First of all, the strain-displacement relations are integrated:

Ux(x,y) J Exdx

j adx
ax+ f(y) (4.8)

uy(x,y) J Eydy

J bdy
by+ g(x) (4.9)
When the unknown functions f(y) and g( x) are substituted into the shear equation

aux 8uy
-+-=c (4.10)
8y ax
one obtains
f'(y) + g'(x) = c (4.11)
The derivatives f' (y) and g' ( x) must therefore be constants . Arbitrarily choosing
f '(y)- A yields g'(x) =C-A.
One can summarize the results in matrix form :
a
b

{ :: } = [ ~
0 0 y 1
y X -X 0 n c
A
B
c
(4.12)

{u} = [cf>]{q} (4.13)


This doesn't seem to help because we have merely defined a new vector {q} of constant
coefficients (three are the specified constant strains; the other three are constants of
integration). It is easy to relate these constants to the nodal displacements, however:

u1 ux(xl, Yl) = ax1 + Ay1 + B


u2 uy(xl, Yl) = by1 + (c- A)x1 +C

(4.14)
CHAPTER 4. CONSTANT STRAIN TRIANGLE 48

In matrix form
{u} = [H]{q} (4.15)
and solving for {q} ,
{q} = [H] - 1 { u} (4.16)
Now, put all these results together:

{!} []{ 0"}


[] [stress-strain] {E}
[] [G] [strain-displacement] {u}
[] [G] [D] [] { q}
{!} [E][G] [D][] [H]- 1 { u} (4.17)
[k]
{!} [k]{u} (4.18)

One can also define a strain matrix:

[B] [D][][H]- 1 (4.19)

and a "shape function" matrix (to be discussed later) :

[N] _ [][H]- 1 (4.20)

4.3 COMMENTS
1. This equilibrium approach works for this case, but is difficult to extend to more
complicated elements. Energy methods must be used instead.

2. Turner, et al., did not use a global cartesian coordinate system, but rather used
a local system. We will use a slightly different local system with the x axis lined
up with the base of the triangle, as in Fig. 4.6.

4.4 EXAMPLE OF TURNER TRIANGLE


The stiffness matrix for the Turner triangle is found most easily by using a local
coordinate system (Fig. 4.6) . One defines
CHAPTER 4. CONSTANT STRAIN TRIANGLE 49

Figure 4.6: Local coordinate system for developing stiffness of Turner triangle

and obtains the following matrix:


..1&_ ~ _..1&_ -~
X 21 X21 X 21
0 -v
x ?T
~ ~ ~ - .El.E2.... 0 ~
X 21 Y 3X21 X 21 Y 3X21 Y3
Eh _..1&_
X2 1
-~
X 21
..1&_
X 21
~
0 v
X
[k] = 2(1 - v2 ) -~ - .El.E2.... ~ ~ 0 ~
X 21 Y 3X21 X2 1 Y 3X2 1 Y3
0 0 0 0 0 0
-v -~ v ~
0 ~
Y3 Y3 Y3
x2
_2_ .E2_ _.E2_ -~
Y 3X21 X 21
- .El.E2....
Y 3X 21 X21 Y3
0
.E2_ ..1/.L _.EL _..1/.L -1
X21 X 21 X 21 X21
0
x2
Gh - .El.E2.... _..E_L __::1__ .EL ~
0
+ 2
Y 3X2 1
_.E2_
X 21
_..1&_
Y 3X21
.El...
X 21
J!.L
Y3
1 0
(4.21)
X 21 X 21 X 21 X 21
-~ -1 ~
1 ~
0
Y3 Y3 Y3
0 0 0 0 0 0

Modern symbolic languages such as Mathematica and Maple ease the problem of an
analytical inversion of [H].

4.5 ASSEMBLY OF TRIANGULAR ELEMENTS


A plate is cantilevered from a wall (Fig. 4. 7), with load P applied. Consider a
3-element representation of the plate using the constant-strain triangle. The assembly
has 5 nodes and 10 degrees of freedom. For each d.o.f. , one must specify either a force
or a displacement.

Each element in the model has

{/} = [k]{u}
where [k] is a 6x6 matrix. The stiffness matrix is assembled for the structure using
symbols for the element stiffnesses as shown in Fig. 4.8.
CHAPTER 4. CONSTANT STRAIN TRIANGLE 50

Figure 4. 7: Assembly of triangle elements

XXXX XX
XXXX XX
xxA~~~ LiA\
xx&~~~ ~A
~~~~DD131[li
~~~~DD~!Zl
DDDDDD
DDDDDD
xx~Afli~DD(il[jj
xx~Afli~DDI&J[t!

Figure 4.8: Assembly of stiffness terms


CHAPTER 4. CONSTANT STRAIN TRIANGLE 51

4.6 EXAMPLE: MEMBRANE WITH IN-PLANE


LOAD
A rectangular sheet of aluminum (Fig. 4.9) is clamped on all four boundaries . A
load of 10,000 N acts at the center. Find the displacement of the membrane at the
point of load application, using a simple FE model.

Figure 4.9: Membrane under in-plane load

Because the load lies in the plane, the problem can be solved as a plane stress
problem. It will be assumed that stresses are not high enough to buckle the sheet (this
is a serious problem, but is nonlinear and beyond our ability to study at present).

Two possible mesh layouts are shown in Figures 4.10 and 4.11. The first has the
advantage for this simple hand calculation that only the interior four elements will be

Figure 4.10: Simple finite element mesh

stressed. Furthermore, by symmetry, one can study a single quadrant of the problem,
with one-quarter of the load applied. The geometry of such a quarter-model is shown
in Fig. 4.12. The equilibrium equation is:

ul 2500
kn k12 k13
0 F2
k21 k22 k23
0 F3
k31 k32 k33 (4.22)

knn
0 Fn
CHAPTER 4. COJVSTA JVT STRA.I1V TRIA N GLE

Figure 4.11: AI tern ate mesh

(0' 500) (1000' 500)


3e
----------------------~

y
X 2500N
1._~----~~----------~ 2
(0 '0) (1000 '0)

Figure 4.12: Quarter-problem solved, by symmetry

The only nonzero coordinate is u 1 . The relevant equation for it is:

k11 u 1 = 2500 (4.23)

[ Ghx~ ]u1 +[ Ehy 3 - ]u1 = 2500 (4.24)


2y3x21 2(1 - v 2)x21
(80, 000)(10)(1000) 2 ] [ (208,000)(10}(500) ]u =
[ 2500 (4.25)
2(500)(1000) Ul + 2(1 - 0.3 2 )(1000) l

The displacement of the midpoint of the membrane u1 is:

2500 N
(4.26)
1, 371,000 Njmm
= 0.00182 mm (4.27)

The alternate mesh proposed in Fig. 4.11 can also be used. It is found that the
exa.ct same displacement is predicted at the center of the membrane. This is quite
unusual-two meshes with such different character typically give very different results.
It does turn out that the stress fields for the two models are substantially different ,
however. Notice that the stresses in the outer four elements in the first mesh (Fig. 4.10)
are zero, which is a strong idealization.
CHAPTER 4. CONSTANT STRAIN TRIANGLE 53

4. 7 HOMEWORK
Problem 1
Three constant-strain Turner triangles are assembled in two dimensions (Fig. 4.13).
Using symbols X, ~ and 0 for the element stiffness terms, show the assembled stiffness
matrix, in compact notation.

Figure 4.13: Constant strain triangles

Problem 2
A bracket is shown in Fig. 4.14 with vertical load at the tip. If the bracket is modeled
with a single constant strain (Turner) triangle, how much does it deflect vertically at
the tip? The bracket has properties

2
E 2.0680 x 10 5 MPa (30.00 x 10 6 lb/in )

v 0.3
E
G
2(1 + v)

Problem 3
Develop the equilibrium matrix which relates internal stresses to nodal forces for
the Turner triangle. This is the [] matrix mentioned in Equation 4. 7. Use the basic
coordinate system in Fig. 4.15.

Problem 4

Consider the constant strain, plane stress (Turner) triangle in Fig. 4.6. If the 3rd
node (at the top) is moved horizontally one unit to the right , and if the other degrees
of freedom are constrained, how much force is required to keep node 3 from moving
vertically?
CHAPTER 4. CONSTANT STRAIN TRIANGLE 54

lOOmm

lOOON

Figure 4.14: Bracket (right triangle).

Figure 4.15: Constant strain triangle


CHAPTER 4. CONSTANT STRAIN TRIANGLE 55

Problem 5
A) A constant strain triangle is shown in Fig. 4.16. The top node is moved 1 mm
(0.0393 in) to the right while the other nodes are fixed. If the element is made of steel
and is 10 mm (0.393 in) thick, how much strain energy is stored in the triangle due to
this deformation?

E 206, 800. MPa (30.00 x 10 6 psi)


G 79,550. MPa (11.54 x 10 6 psi)
v 0.3

lOOmm IOOmm
I

1
1

.
X

Figure 4.16: Constant strain triangle

B) If the triangle in part A) were actually cut from steel and loaded at one corner
to produce the same nodal displacement, would the stored strain energy be less than,
equal to, or greater than the value found for the finite element?

Problem 6
What are the appropriate nodal force and displacement vectors {!} and { u} in a
plane stress solution of a cantilever beam shown in Fig. 4.17? Insert as many known
components as possible.

Side question: What is a fundamental problem in such a finite element model of a


beam where only one constant strain element is used through the depth?

Problem 7
The triangle of Turner, Clough, Martin and Topp was originally developed using:
a) energy ideas
b) equilibrium ideas
c) set theory
d) integral equations
CHAPTER 4. CONSTANT STRAIN TRIANGLE 56

I.

X 0
f 1 I lOOON
(225 lb)

Figure 4.17: Finite element model of cantilever beam.

Problem 8

The Turner triangle is important because of its


a) accuracy
b) linear strain field
c) historical significance
d) use in beam theory

Problem 9

The Turner triangle has linear interpolation functions . This means that
a) no stresses can be recovered
b) the element is more accurate than 6-noded triangles
c) the a x stress is constant throughout the element
d) the a x and a y stresses equal each other throughout the element
e) none of the above.

Problem 10

An attempt to separate the shear and direct stress effects in the Turner triangle
stiffness matrix
a) is successful
b) fails because shear and direct stress effects are always coupled
c) is not useful
P roblem 11

Give the relation between displacement and shear strain in the plane stress quadri-
lateral in Fig. 4.18.
{ xy = [?){?}
CHAPTER 4. CONSTANT STRAIN TRIANGLE 57

L X
\ ____________.7
Figure 4.18: Two-dimensional quadrilateral.
Bibliography

[1] Turner, Clough, Martin and Topp, "Stiffness and Deflection Analysis of Complex
Structures," Journal of the Aeronautical Sciences) Vol. 23, September 1956, pp .
805-823.

58
Chapter 5

INTERPOLATION

5.1 INTRODUCTION
Consider the following definitions:

Interpolation is the process of estimating functional values within a do-


Inain bounded by points where the functional values are known.

Extrapolation is the process of estimating functional values external to


a domain where functional values are known.

Interpolation once was a common task for engineers. Logarithm and trigonometry ta-
bles were given in point-wise fashion and were interpolated for practical use. Engineers
have a natural intuition for linear interpolation in one dimension (passing a straight
line through two points and finding interior values as in Fig. 5.1). The typical engi-

f(x)

Figure 5.1: Linear interpolation in one dimension.

neer gets weak in the knees, however, when he/she is required to do "higher-order"
interpolation, or when two-dimensional or three-dimensional cases are encountered.
Higher-order interpolation requires fitting the interior of a region by polynomials of
quadratic and higher degree. In this case, the region may contain known points in the

59
CHAPTER 5. INTERPOLATION 60

interior (Fig. 5.2) and/or may involve information about derivatives at the end points
(Fig. 5.3).

f(x)
.....-- -- __.

Figure 5.2: Quadratic interpolation in one dimension. Information given at internal


point.

f(x)

A
~'--1 -... __ _ - /
,---

Figure 5.3: Cubic interpolation in one dimension. Information about function and its
first derivative given at end-points.

In two dimensions, interpolation of a scalar function can be interpreted as a surface


generated over a domain (Fig. 5.4). The polynomial interpolation in Fig. 5.4 involves
a quadratic function of the form

(5.1)

Often in two dimensions, one must interpolate a vector function such as the dis-
placement field:
{u} = { Ux } = { ql + q2x + q3y + q4xy } (5.2)
uy qs + q6x + q7y + qsxy
in Fig. 5.5. One separately interpolates the Ux and uy components. The formulas for
interpolating the two functions are identical, but the functional values at the nodes are
not, of course.
CHAPTER 5. INTERPOLATION 61

f(x,y)

X
Domain of problem

Figure 5.4: Interpolation of two-dimensional scalar field.

Figure 5.5: Interpolation of two-dimensional vector field.


CHAPTER 5. INTERPOLATION 62

In three dimensions, one interpolates vectors as in Fig. 5.6. If this is difficult to


visualize, one can resort to a mental image of a scalar problem, such a heat conduction,
where temperature is a function of position in a 3-D body.

Figure 5.6: Interpolation of three-dimensional field.

Hamming [1] suggests that interpolation uses information about the function f( x)
in the form of:

1. function values only


2. values of derivatives
3. differences of function values
4. arbitrarily placed samples

In the discussion above, we have used the first 3 types of information. The 4th type is
novel and will be discussed in a later chapter on Gaussian integration.

Interpolation leads naturally to numerical integration. Consider the area under a


curve in one dimension, as in Fig. 5. 7. The use of such straight-line interpolation is
called integration by the trapezoidal rule. This is a very important concept in theory
and practice. The common Riemann integral is defined in calculus as the limit of a
process of trapezoidal integration where Ai is the area under one such segment and a
sequence of refined subdivisions is carefully prescribed.

1
X2

Xl
f( x )dx = lim 2::::
N-+ oo -1 N
~- '
Ai (5.3)

When quadratic interpolation is used to numerically integrate the area under a curve,
the method is called Simpson's rule.
CHAPTER 5. INTERPOLATION 63

f(x)

Figure 5. 7: Numerical integration by trapezoidal rule.

5.2 ONE-DIMENSIONAL, LINEAR INTERPO-


LATION
Consider the function sin( xi) as given in a crude table for Xi = 0, 10, 20, 30, ... , 90
and as plotted in Fig. 5.8. Suppose one desires a value of sin(24). A linear interpola-
tion (Fig. 5.9) gives
!(24) = !(20) + 0.4[!(30) - !(20)) (5.4)

f(x)
,......_ ..... __..._ ....
,.. ,..JIY-

/" "'
tl

I'
/
{
I

0 20 40 60 80 90 X

Figure 5.8: Sine table.

This form is immediately obvious. One can rewrite it by collecting terms :

!(24) = 0.6f(20) + 0.4f(30) (5.5)

The expression now interpolates by weighting the endpoint values by constants that
reflect the distance to the end points . This form is far more versatile than Equation 5.4,
and unfortunately for engineers, harder to appreciate.1
1
Interpolation using weighted values of the function taken at special points seems to be a hard
concept for engineers to grasp, yet mathematicians have no trouble with it.
CHAPTER 5. INTERPOLATION 64

f(x)

f(30)

f(20)

Figure 5.9: Linear interpolation in sine table.

The general case of linear interpolation in one dimension (Fig. 5.10) is

(5.6)

This is rewritten
(5.7)
where the Ni ( x) are called shape functions, or interpolation functions (Fig. 5.11). The
shape functions are linear polynomials and have the ability to exactly interpolate any

f(x)

I I

X X X
1 2

Figure 5.10: Linear interpolation in one dimension. General case.

linear function. In Fig. 5.12, the contribution of each shape function toward interpo-
lation of f( x) is shown.

The same expression for linear interpolation is obtained if one starts with a Taylor
senes

df 1 d2 f 2
f( x ) = f( xi) + dx lx1 ~ X + 2dx 2 lx1 (~x) +... (5.8)
and ignores the higher order terms in ~ x, as well as approximating the derivative by
differencing:
CHAPTER 5. INTERPOLATION 65

f(x)

1.0

X X
1 2

Figure 5.11: Shape functions for linear interpolation in one dimension.

f(x)

Figure 5.12: Contribution of each shape function to total functional value. Linear
interpolation.
CHAP TER 5. INTERP OLATION 66

f( x) f( xi) + (j( x2)- f(xi) )( x _ xi)


X 2 - XI

( X 2 - X )j(xi) + ( X - XI )j(x2)
X 2 - XI X2- XI

NI(x)f(xi) + N2(x)j(x2) (5.9)


The shape functions NI (x ) and N 2 ( x ) are defined to have unit values at the "home"
nodes and zero values at other nodes :

(5.10)

They also have a unit sum throughout the domain:

(5.11)

This normalization generalizes to higher-order, one-dimensional interpolation with


polynomials .

5.3 ONE-DIMENSIONAL QUADRATIC INTER-


POLATION
The primary task in interpolation is development of proper shape functions . In
finite element displacement theory, polynomials are used almost exclusively. This has
an advantage that error in solutions is more easily determined because the derivative
of the field variable is also a polynomial of specific degree. This is not true with
transcendental functions , as often used in global methods such as Galerkin's method.

One can construct shape functions in two ways, summation form or product form .
In each case, one applies a condition for unit value at the "home" node and zero value
at other nodes :

(5.12)

5.3.1 Summation Form


Consider the three-noded line element in Fig. 5.13. Use an element coordinate
system with origin at the left end. Assume

NI(x)=a+bx+cx 2 (5.13)

Applying the "defining" Eqn. 5.12, we will pass the curve NI (x) through the 3 points
in Fig. 5.13.
NI(O) = 1 =a
CHAPTER 5. INTERPOLATION 67

f(x) __f(x) - - /
~--
/

X
0 L/2 L

Figure 5.13: Quadratic interpolation in line element.

N1(Lj2) = 0 =a+ bL/2 + cL 2 /4 (5.14)


N1(L) = 0 =a+ bL + cL 2
This yields 3 equations in 3 unknowns which can be solved for a,b, and c, yielding

a=1

b= -3/ L (5.15)
c = 2/ L2
and hence
N1(x) = 1- 3(x/L) + 2(x/L) 2 (5.16)
Similar work gives
N2(x) = 4(x/L)- 4(x/L) 2 (5 .17)
N3(x) = -(x/L) + 2(x/L) 2 (5.18)
Notice how the shape functions are nondimensionalized in terms of the ratio ( x / L).
The quadratic, one-dimensional interpolation has a unit normalization over the
entire domain:

(5.19)

5.3.2 Product Form


A second way to create shape functions is to factorize the shape function, including
enough factors to cause zeroes at nodes, as well as an arbitrary constant to give unit
value over the home node. For quadratic interpolation (Fig. 5.13), assume
CHAPTER 5. INTERPOLATION 68

N1 = C(x- L/2)(x- L) (5 .20)


Applying the unit value condition of Eqn. 5.12, we have

N1(0) = C( -L/2)( -L) (5.21)

Hence
(5 .22)
and
2
N1(x) L 2 (x- L/2)(x- L)
1- 3(x/L) + 2(x/L) 2 (5.23)

Similarly, one obtains N 2 ( x) and N 3 ( x) as given above.


For the line element, one can more easily do higher order interpolation (cubic,
quartic, quintic, etc) by using the product form than the summation form. The latter
leads to simultaneous equations which are difficult .

5.4 PROOF OF NORMALIZATION (1-D)


We found for linear and quadratic interpolation that the sum of the shape functions
had an interesting normalization over the entire domain

(0::; X::; L) (5 .24)

A short proof of this relation for shape functions of all polynomial degree in one di-
mension follows .

The shape functions for a domain with N nodes are polynomials of degree N-1.
Because of the normalization at the home nodes, the sum of the shape functions I: Ni( x)
must have unit value at the nodes (Fig. 5.14). If we form the polynomial P(x) =
I: Ni ( x) - 1, it is of degree N - 1 and has zero crossings at the nodes as in Fig. 5.15.
The fundamental theorem of algebra implies that a nonzero polynomial of degree of
N-1 can have at most N-1 roots (zero crossings), therefore P(x) is identically zero.
Q.E.D .

5.5 HOMEWORK
Problem 1.
If one wishes to interpolate a field variable in one dimension, and wishes to use four
nodes:
CHAPTER 5. INTERPOLATION 69

f(x)
I:N.(x)
1
1 ~

P(x) = I: N .(x) - 1
----- 1

Figure 5.14: Sum of shape function.

P(x)
=C~:::Ni(x))-1 / Impossible!
,...'( ~~~..' ... ),. ....... ,~
X

Figure 5.15: Proof by contradiction. Attempt to pass nontrivial polynomial of degree


N-1 through N zero values.
CHAPTER 5. INTERPOLATION 70

a) a cubic polynomial is needed


b) a quartic polynomial is needed
c) derivatives of the function must be used

Problem 2.

Propose a shape function N 1 (x) for the 4-noded line element in Fig. 5.16.

--
L/3 2L/3 L

Figure 5.16: Four-noded line element.

Problem 3.

Does a column in the shape function matrix [N] represent internal displacement
fields due to a unit nodal displacement at a specific d.o.f.?
Problem 4.
Develop the shape functions for the three-noded, constant strain triangle shown in
Fig 5.17.

y
(1,2) (3,2)

~--~~-------------x
(1,0)

Figure 5.17: Constant strain triangle.


Bibliography

[1] Hamming, Numerical Methods for Scientists and Engineers, McGraw- Hill Book
Co., New York, 2nd Edition, 1973, Chapters 14-19.

71
Chapter 6

VIRTUAL WORK AND


POTENTIAL ENERGY
THEOREMS

6.1 BACKGROUN D
A finite element often has a number of nodes, making several load paths possible.
In structural terms, such an element is statically indeterminate, i.e., it is not possible
to determine the load paths and stresses from equilibrium arguments alone. Energy
concepts such as virtual work and potential energy are very helpful in analyzing such
ele1nents. These methods enforce displacement compatibility within the element and
unravel the indeterminacy.

Much of the finite element theory in solid mechanics is concerned with the stress-
strain law (constitutive law), the strain-displacement law, and the load-deflection re-
lation. These are symbolically sketched in Figs. 6.1- 6.3, where a scalar version of the
vector quantities is used. These sketches are an artifice, but are useful in sorting out
concepts. The unloaded, undeflected structure is always available as a reference; there-
fore the load-deflection relation passing through the origin in Fig. 6.3 is the general
case.

Many finite element problems are linear, where each of the relations in Figs. 6.1-
6.3 is a straight line. Also, most finite element problems have no prestress { a 0 } or
prestrain {Eo} such that the lines in Figs. 6.1- 6.3 would all pass through the origin.
Let us not assume linearity yet; but rather, retain the general nonlinear case through
the discussion on virtual work.

6. 1.1 Stress-Strain Law ( Const itutive Law)


The general form of the nonlinear, elastic stress- strain law is of the form

72
CHAPTER 6. VIRTUAL WORI( AND POTENTIAL ENERGY THEOREMS 73

Figure 6.1: Stress-strain law (nonlinear elastic).

Figure 6.2: Strain-displacement law (general case).

Figure 6.3: Load-deflection relation (general case).


CHAPTER 6. VIRTUAL WORJ( AND POTENTIAL ENERGY THEOREMS 74

{a-} =function( { E}) (6.1)


A linearized version of this law, valid in the neighborhood of a reference point I
(Fig. 6.1) is
{a-} = [G] ({ E} - {Eo} ) + {a-o} (6.2)
The standard practice is for the reference point I to correspond to the origin in Fig. 6.3.
In this case, {a-0} and {Eo} are prestress and prestrain present in the unloaded structure.
Prestress and prestrain are zero in most structural problems.

6.1.2 Strain-Displacement Law


The relation between internal strain and nodal displacements is sketched in Fig. 6.2.
This is a geometric relation (kinematic) and is not a function of the material, but does
depend on the total strain involved. Most engineering structures remain in the linear
range.

6.1.3 Equilibrium Position


If a body is loaded with static loads {P}, it deforms and reaches a static equilibrium
at a point labeled II in Fig. 6.3. The shaded area represents the work done by the
external loads on the body, if the loads are slowly applied.

6.1.4 Virtual Displacement


A virtual displacement { ~u} is an infinitesimal displacement from position II, in
a 1nanner consistent with geometric constraints (Figs. 6.4, 6.5). The body moves to
position III and undergoes an increment of strain energy density { ~U} shown as the
shaded area in Fig. 6.6. The stresses during this virtual displacement are assumed to
remain constant. The strain energy due to virtual strain is

~U = Jfv (stress)(~strain)dV
,___,_, "--v---' (6.3)
II II-+!II

The amount of work done by the external forces (shaded area in Fig. 6.5) is:

~W = Lfi~uei (6.4)
'-v-'
Il-+!II

The energies in Equations 6.3 and 6.4 will be equal only when the material is
nondissipative, i.e., elastic.
CHAPTER 6. VIRTUAL WORJ( AND POTENTIAL ENERGY THEOREMS 75

p
j

Figure 6.4: Configuration of a body under load.

Figure 6.5: Virtual displacement.

Figure 6.6: Strains during a virtual displacement.


CHAPTER 6. VIRTUAL WORl( AND POTENTIAL ENERGY THEOREMS 76

6.1.5 Work
Work is defined most fundamentally in an incremental manner:
~W - (force) (~displacement) (6.5)
One must be on guard to never define work as a product of force times displacement
and then to take an increment:

Beware Ill
... [
~W
= Pu+ ~Pu
=WP~u l Fa1se Ill
...

Furthermore, the work done by a force on a body involves the distance the body moves,
rather than the distance the force moves. This is very important with respect to sliding
forces.

6.1.6 Virtual Work


This is the work done, whether by external or internal forces, during a virtual
displacement, i.e., in moving from state II to state III.

6.2 VIRTUAL WORK THEOREM


In mechanics, there are three fundamental statements of mechanical equilibrium
which may be used interchangeably. These are Newton's laws of motion (principally the
second law, F = rna), Hamilton's principle and the virtual work theorem, all of which
can be used for statics and dynamics problems. The virtual work theorem applies to all
of mechanics, but we will use a version valid for nonrelativistic, nonthermal problems .
VIRTU AL WORK THEOREM: A body is in equilibrium if and only if the sum
of all virtual work done during an arbitrary virtual displacement is zero.

~ Winternal
~~
+ ~ Wexternal =0 (6.6)
II -+ III II -+ III
This is a general statement for bodies represented by Fig. 6.5 and Fig. 6.6, which may
have dissipative nonlinear materials. An important subcase occurs when the material
of the body is nondissipative, e.g., nonlinear elastic. The work done by the internal
forces (i .e., stresses and strains) can be shown to be the negative of the strain energy
increment by an involved mathematical proof not done here.

~Winternal
~
= -~U
'-v--"
(6.7)
II -+ III II -+ III

Inserting Eqn. 6.7 into Eqn. 6.6, one obtains:

~ U = ~ Wexternal
..___, ~
(6.8)
II -+ III II -+ III
CHAPTER 6. VIRTUAL WORJ( AND POTENTIAL ENERGY THEOREMS 77

MODIFIED VIRTUAL WORK THEOREM: A nonlinear) elastic body is in equi-


librium if and only if the increment of work done by external forces equals the
change in strain energy during an arbitrary virtual displacement .

Concluding Comments:

1. The theorems to this point don't imply the existence of either a function W or a
function U which mean anything. It only means one can calculate the increments
of energy, i.e., the shaded areas in Fig. 6.5 and Fig. 6.6 for the case of energy-
conserving materials.

2. Equation 6.8 may be intuitive to some readers because, for a non-dissipative


material, the work done by the external forces during a real displacement I ---+
II equals the energy stored in the body. This intuitive idea is a scalar idea.
Equation 6.8 is a vector concept valid for arbitrary displacement and is much
more powerful than a conservation of energy idea.

6.3 POTENTIAL ENERGY THEOREM


The potential energy theorem can now be derived. This theorem is less general
than virtual work and is restricted to linear elastic systems and to energy-conserving
external force fields .

For the potential energy theorem, we need to use a variational operator 8. We will
change our viewpoint of the Ll symbol used above through the use of a definition:

8{u} = {Llu} (6.9)


"'-....--' '--v--"
II I I - III

In other words, we pass from an incremental usage of Ll to an operator form 8. The


symbol 8 will retain the concept of a virtual displacement, i.e., a small change from
the II position, with forces and stresses remaining unchanged. The use of 8 { u} implies
that a single valued function u exists.

If the external force field is conservative (energy-conserving), it can be derived from


a potential:

{F} = -{V}W (6 .10)


where { F} is the external force vector, { \7} is a generalized gradient operator which
acts on all displacement-like quantities, and W is the work potential.

Putting all of our "work" ideas together:


CHAPTER 6. VIRTUAL WORJ( AND POTENTIAL ENERGY THEOREMS 78

~Wexternal (Forces) x (~displacements )


----.....--
II
~
II II
v

-+ III
(Forces) x 8(displacements )
~ -v
II II
8(Forces) x (displacements )
~
II II

8(~) (6.11)
II

where we now admit the existence of a force potential W and use an operator 8 which
acts only on displacement-like quantities. Also, since the system is linearly elastic
(Fig. 6. 7), we can write symbolically

Figure 6.7: Virtual strain energy (increment in strain energy due to virtual strain).

II
6.U
~
-+ III
lvr (stress
II
)(6.strain )dv
"-v--' ~
II -+ III

lvr (stiffness)(strain
'-v--'
)(~strain
~
)dv
II II -+ III

lvr (stiffness)(strain
'-v--'
II
)8(strain )dv
'-v--'
II

8~ (stiffness) (strain ) 2 dv
r
2 lv '-v--'
II
8( u )
'-v-'
(6.12)
II

This, of course, implies the existence of a strain energy function U. The operator 8 acts
on only strains. We now combine results of Eqns. 6.8, Eqn. 6.11 and Eqn. 6.12 to get
CHAPTER 6. VIRTUAL WORI( AND POTENTIAL ENERGY THEOREMS 79

- 5( w
~
)- 5( u )= 0
~
(6.13)
II II

u + w)
5( ~~ =0 (6 .14)
II II
or
~=0
I
(6 .15)

where V is called "potential energy." (Warning: In some physics courses, the same
words are used to denote strain energy- do not confuse the potential energy V in a
structural system with strain energy U.)

Potential Energy Theor em: The potential energy of a mechanical system made of
linearly elastic elements and exposed to conservative forces is "stationary" (has
zero first variation) at the static equilibrium configuration.

Concluding Comment :

If one generalizes to systems with even mild nonlinearities, the question of stability
of this equilibrium position becomes important . The second variation of the potential
energy then is needed. This is a more advanced topic.

6.4 HOMEWORK
Problem 1
Virtual strain energy is defined as the change (increment) of strain energy during
a virtual displacement . Consider a line element with EA/ l = 104 N /mm under a 1000
N tensile load as shown. After the line element is in static equilibrium (Fig. 6.8), a
virtual displacement

. ()
1000 N ' - " - - - - - - --
-;-----.
------- 1000 N

Figure 6.8: Line element at equilibrium.

0.001 }
{ilu} = { -0.001 mm

is imposed. What is the resulting virtual strain energy?


CHAPTER 6. VIRTUAL WORK AND POTENTIAL ENERGY THEOREMS 80

Comments: Remember that work is defined as {increment of displacement }T {force}.

Problem 2
A 200-lb. man is poised (motionless) at the end of a thin, flexible diving board
(Fig. 6.9). If the end of the board is given a virtual displacement downward of 1 in,
how much virtual work is done on the board by gravity (the man's weight)?

%~r--------------------~
%~------------------~]
~

Figure 6.9: Man at end of diving board.

How much work is done by the internal elastic forces during the same 1" displacement?

Problem 3
Near equilibrium, a virtual displacement causes no virtual work. Does potential
energy change much with the same virtual displacement?

Problem 4
Is the work potential as it appears in the potential energy theorem equal numerically
to the work done by the external loads?

Problem 5 (Multiple choice)


The potential energy theorem
a) uses a strain energy measured from some initial reference position
b) uses a work function defined from some initial reference
c) requires a variation of displacement quantities
d) all of the above

Problem 6 (Multiple choice)


The virtual work theorem requires
a) an infinitesimal, virtual displacement
b) an energy conserving system
c) that one account for work done by external forces on the body in question
d) the first and third answers above
CHAPTER 6. VIRTUAL WORK AND POTENTIAL ENERGY THEOREMS 81

Proble1n 7 (Multiple choice)

Virtual work is the work done


a) during a displacement from deformed to undeformed states
b) by external forces
c) by internal forces
d) during a virtual displacement

Problem 8

Two line elements are joined as shown in Fig. 6.10.

1 2 3

Figure 6.10: Line element assembly.

E1A1/ L 1 = 1000 Njmm

E2A2/L2 = 3000 Nfmm


A 6000 N force is applied at the middle node and the assembly comes to rest at
equilibrium. A virtual displacement ~u 2 = +0.001 mm then takes place.

A) What is the equilibrium displacement at node 2?

B) What is the strain energy at equilibrium?

C) What is the virtual work due to external forces during this virtual displacement?

D) What is the virtual strain energy due to the virtual displacement?


Problem 9.

A structural system has the following equation of equilibrium, expressed in N, mm,


sec units:

The system is at equilibrium. It is then given a virtual displacement:

~u 1 } = { 0.001 }
{ ~u2 -0.002

A) How much virtual work is done by the external forces?

B) How much virtual strain energy is stored?


CHAPTER 6. VIRTUAL WORJ( AND POTENTIAL ENERGY THE OREMS 82

C) How much total virtual work is done?


D) What was the potential energy of the system at equilibrium? (Be sure to use the
structural mechanics meaning of "potential energy" and not the physicists' meaning.)

Problem 10.
A two link assembly is loaded as shown in Fig. 6.11 . The left end is pinned. The
live load at the left end is directly reacted by the pin support . The stiffnesses of the
two links are:

500 N/mm

1000 N/mm

A) Set up the finite element equations for the assembly.


B) Put the equations in standard form for finite element solution.
C) Solve for the displacements.
D) Calculate the potential energy of the system at equilibrium.

4000 N 400 N 600 N


44---74~r-----~G~---~--~
7!7))7 CD
Figure 6.11 : Assembly of line elements.

Problem 11.
Consider the equilateral triangle in Fig. 6.12. It is made of steel, has sides with
length 200 mm, is 10 mm thick and is clamped on the bottom edge. A load of 1000 N
acts vertically at the upper vertex. Use a one-element model to answer the following
questions .

E = 207 , 000 MPa v = 0.3


A) Find the deflections at the upper vertex due to the load.

B) While the system is in equilibrium under the load, a virtual displacement of the tip
in the horizontal direction is given: u 5 = 0.00001 mm. What is the virtual work done
by the external force?
CHAPTER 6. VIRTUAL W ORK AND P OTENTIAL ENERGY THEOREMS 83

C) While the system is in equilibrium under the load, a different virtual displacement
of the tip in the vertical direction is given: u 6 = -0.001 mm. What is the virtual strain
energy during this virtual displacement?

////////

Figure 6.12: Constant-strain triangle in plane stress.


Chapter 7

DERIVATION OF A FINITE
ELEMENT BY VIRTUAL WORK

7.1 DERIVATION FOR POINT LOADS


Consider a single linearly-elastic finite element with concentrated external forces
(Fig. 7.1) . Three standard displacement configurations are shown. Use the modified
form of the virtual work theorem to develop the element stiffness. For the remainder
of our linear work , we will drop the distinction between ~ and 8 and use only 8 to
indicate both incremental and operational procedures .
8{u} {~u} (7.1)
The flow chart for unknowns is given in Fig. 7.2.

Figure 7.1: Typical finite element

The modified virtual work theorem is

8Wext 8U 0 (7 .2)
'-v---" "-....-"
II -+ III II -+ III

{8u}T { j} - { {8E} T { 0"}


dV 0 (7.3)
"-v--' "-v-" Jv '-v---" "-v-"
II -+ III II II -+ III II

84
CHAPTER 7. DERIVATION OF A FINITE ELEMENT BY VIRTUAL WORI( 85

1
[G] [D] [Hf[<t>]
[f] [cr] <= [] <= [u] <= [q] <= [u]

[k]
~
[B]

Figure 7.2: Variables and transformations in FEA.

The virtual strain is found from the expression for strain

{c} = [B]{u} (7.4)


by operating on both sides by 8, where in a linear system, [B] is not a function of u:

{8t} = [B]{8u} (7.5)


Hence , with {a} = [G][ B]{ u}, the virtual work theorem becomes

{8u}T {/}- fv {8u}T[B]T[G][B]{ u}dV = 0 (7.6)

The nodal displacements {8u }T and {u} do not depend on the integration variables and
can be factored out:

(7. 7)

Since {8u }T is an arbitrary vector, the vector enclosed in the large parentheses must
be zero. This leads to
(7.8)

Defining the integral portion to be the stiffness matrix, we have

{/} = [k]{u}. (7.9)


This approach has defined the stiffness matrix without the need of an equilibrium
matrix.

Several forms of the stiffness matrix follow:

[k] fv[Bf[G][B]dV (7.10)

l[Nf[Df[G][D][N]dV (7.11)

[H] - 1 T !v [] T [D] T [G] [D] [>] dV[ H] - 1 (7.12)


CHAPTER 7. DERIVATION OF A FINITE ELEMENT BY VIRTUAL WORJ( 86

The last version is particularly good for analytical integration. The form to be
chosen depends on the starting point:

1. newer methods start with an assumed shape function [N], hence Eq. 7.11 is used.

2. older methods start with assumed displacement function [</>], hence Eq. 7.12 is
used.

7.2 EQUIVALENT NODAL LOADS


Suppose an element has acting on it several kinds of distributed loads (Fig. 7.3).
These can be volumetric, surface or line loads. In each case, a contribution to the virtual

II
I

Figure 7.3: Distributed loads. Deflection exaggerated.

work will be made by a force moving the body through an increment of displacement.
Although the force and the spatial dimensions vary, the relevant displacement in each
case is {8u} = [N]{8u}. Hence if
X(x,y,z) volume load
T(x,y,z) surface load
( x,y,z) line load
then
DWext 8U = 0 (7.13)
II .- III II .- III

becomes
{8u}T {f} + f {8u}T {(x,y,z)}df + f {8u}T {T( x,y,z) }dS
'-..---""-..-" Jf. ' - . . - - - " ' - . . - " " Js ' - . . - - - " ' - . . - " "
II-+!II II II-+III II II-+III II

+ f { 8u} T {X (X' y' z)} dV - f { 8E} T { (J} dV = 0 ( 7.14)


lv ..____,~ lv ~..___,
II-+III II II-+III II
CHAPTER 7. DERIVATION OF A FINITE ELEMENT BY VIRTUAL WORI( 87

This is rewritten

{8uV({f}+ j)Nf{L}d + f.[Nf{T}dS

+ /)Nf{X}dV- fv[Bf[G][B]dV{u}) = 0 (7.15)

All integrals can be evaluated once the internal displacement fields (and hence [N], []
and [H]) have been assigned. Each integral is given a name:

{f}c := ft [N]T {}df Equivalent nodal load due to line load


{f}T := fs[N]T {T}dS Equivalent nodal load due to surface load
{f}x- fv[N]T{X}dV Equivalent nodal load due to volume load

Finally, because {8u} in Eq. 7.15 is arbitrary, the equation of equilibrium becomes:

{!} + {f}c + {f}x + {f}T = [k]{u} (7.16)

7.3 PRESTRAIN AND PRESTRESS


Many physical problems have an initial strain or initial stress . Initial strain (pre-
strain) can be caused by thermal expansion. In nonlinear problems, one often intro-
duces a prestrain or prestress as an artifice to aid the solution. In all of these cases , a
linear stress-strain law is used to represent a portion of the stress-strain curve in the
neighborhood of the reference (Eo, a 0 ) point (Fig. 7.4).

(J

cro
I
I
I
I
I
/
~
0

Figure 7.4: Stress-strain law. Initial stress and strain.

{a} = [G]( { E} - {Eo}) + {a o} (7.17)


The virtual work then gives an additional increment of strain energy:

8U J{8E} {a} dV
T (7.18)

j ([B] {8u}) T ([ G] ({ E} - {Eo}) + {a o}) dV (7.19)


CHAPTER 7. DERIVATION OF A FINITE ELEMENT BY VIRTUAL WORI\ 88

which can be expanded

OU = {OuV .fv[Bf[G]{E}dV-{OuV fv[Bf[G]{Eo}dV+{Ouf fv[Bf{~o}dV (7.20)

After combining this 8U with the energy balance in the previous section (Eqn. 7.15)
and recognizing that {8} is arbitrary:

{J}+ fv[Nf{X}dV + DNf{T}dS+ 1[Nf{ }d+

+ .fvrsJT[G]{Eo}dV- j [Bf{~o}dV = .fvrsJT[GJ[BJdV{u} (7.21)

This gives a complete set of equivalent nodal loads in the equilibrium equation:

[k]{u} = {!} + {f}x + {f}r +


'-v--' '-....--"
body forces surfa ce for ce s

{j} C
'-....--"
+ "-v-'
{f}uo + {j}fo
'-v--'
(7.22)
li n e loads pre stress pr estrain

7.4 REVIEW OF EQUIVALENT NODAL LOAD


CONCEPT
The virtual work theorem has led to

[k]{u} = {f} + fv[N]T {X}dV + fs[N]T {T}dS


+ 1[Nf{}d + .fvrsf[G]{Eo}dV- fv[Bf{~o}dV (7.23)

We will define

{f}x - fv[N]T {X}dV (7.24)


'-v--'
body for ces

{f}r - fs[NJT {T}dV (7.25)


'-....--"
sur fa ce forc es

{!}c - 1[N]T{}df (7.26)


'-....--"
li n e f or ces

{f}fo - .fv[Bf[G]{ Eo}dV (7.27)


'-v--'
prestrain

{!} lTQ - - fv[B]T {ao}dV (7.28)


"-v-'
pr estre ss
CHAPTER 7. DERIVATION OF A FINITE ELEMENT BY VIRTUAL WORK 89

7.5 ALTERNATE STRAIN ENERGY EXPRES-


SIONS
Consider the strain energy in a linear, elastic structure at equilibrium, state II.
If the structure were at zero stress and strain at state I, then the increase in energy
absorbed equals the work done:

~u
(7.29)
r-n I - II

~ J{V {o- }dV = ~ {u} {j}


E T (7 .30)

~{u}T j [B]r[GJ[B]dV{u} = ~{uV[k]{u} (7.31)

[k]

At equilibrium, strain energy takes any of the 4 forms (on either side) of Eqns. 7.30 -
7.31. In general, strain energy must be greater than or equal to zero for any displace-
ment{u}:
(7.32)

If a nonzero set of displacements {u} exists for which the strain energy is exactly zero,
then {u} is a rigid body mode and the stiffness matrix [k J is called positive semi-
definite. If there are no rigid body modes, mechanisms or grid point singularities, the
[k J matrix is positive definite.

7.6 HOMEWORK
P roblem 1 .
A regular pentagon element has been developed for plane stress problems (Fig. 7.5) .
Each side is of the same length and each interior angle is equal. It is desired to discuss
the shape functions that would be appropriate for such an element.

A) If one uses a product form for the shape function, one could use the equations for
the lines f1 (x,y) and f2 (x,y) in Fig. 7.5 in the form:

could use the lines f3 (x,y) and f4 (x,y) in Fig. 7.6 in the form :
CHAPTER 7. DERIVATION OF A FINITE ELEMENT BY VIRTUAL WORJ( 90

Figure 7.5: Plane stress pentagonal element with first scheme for shape functions

Figure 7.6: Plane stress pentagonal element with second scheme for shape functions

The element has a distributed load load on the side directly opposite node 1. For
the two cases, which of the 5 nodes would have positive, zero and negative equivalent
nodal loads? Your goal will be to construct a table with plus, zero and minus sign
entries. You will want to construct sketches of the candidate shape functions and
figure out the general sense of the equivalent nodal loads.

Node Case A Case B


1
2
3
4
5
CHAPTER 7. DERIVATION OF A FINITE ELEMENT BY VIRTUAL WORJ( 91

Problem 2.
Suppose one wishes to work with displacement functions (rather than shape func-
tions) for the pentagon in the previous problem.

A) Propose a displacement function for the x displacement u( x,y).

B) Interpret each term in the displacement function for u(x,y).

Problem 3
A hypothetical structural element with stiffness [k] has a set of loads applied which
gives an equilibrium displacement {u}. A subsequent virtual displacement { 8u} is
g1ven. Choose the closest answer to how much virtual work is done by the external
forces.

{u} = { l} mm {8u} = { ~0.01 } mm

(a) -5 N mm
(b) 0 N mm
(c) 10 N mm
(d) 50 N mm
(e) 100 N mm.
Problem 4
Equivalent nodal loads:
(a) can be derived using virtual work;
(b) are simpler than lumped load ideas;
(c) allow shorter computer solution times;
(d) do not work for distributed loads.

Problem 5
Equivalent nodal loads:

(a) are always smaller than lumped loads;


(b) depend on the shape functions;
(c) are found by differentiation;
(d) are easier to calculate than lumped loads;
(e) none of the above.
CHAPTER 7. DERIVATION OF A FINITE ELEMENT BY VIRTUAL WORJ( 92

Problem 6

Generalized coordinates:

(a) must have the same dimensions as physical coordinates


(b) must be the same in number as physical coordinates
(c) must not be used for beams
(d) are used in shape functions
(e) cannot cause rigid body motion.

Problem 7
Energy approaches in F.E. have advantages in:

(a) development of equivalent nodal loads


(b) development of elements with complicated shape functions
(c) both of the above.

Problem 8

If a material is orthotropic, one must account for this in the:


(a) stress-strain law
(b) strain-displacement law
(c) equivalent nodal loads due to volume loads
(d) all of the above.
Problem 9
Vector force equilibrium at a node is described by:

(a) a single equilibrium equation in "detailed" notation


(b) a single equilibrium equation in "compact" notation
(c) a column of the stiffness matrix
(d) summation of external forces at the node.
Problem 10
In the basic theory of finite elements, an element with n nodes always has:
(a) n generalized coordinates in detailed notation;
(b) 2n shape functions in detailed notation;
(c) three displacement fields: ux(x, y) , uy( x , y ), uz(x , y );
(d) n nodal displacements in compact notation.

Problem 11.

A hypothetical three-noded line element is to be created (Fig. 7. 7, top). It will be


used to model a chain on a chain saw (Fig. 7. 7, bottom). The material is steel and is
CHAPTER 7. DERIVATION OF A FINITE ELEMENT BY VIRTUAL WORI\ 93

of thickness h. Find the stiffness component k 11 in symbolic form.

Figure 7. 7: Line element; assembly of line elements.

Problem 12.

A constant strain (Turner) triangle is used in an orientation shown in Fig. 7.8 such
that gravity loads the triangle with 50 N /mm 2 . Find the equivalent nodal loads on
node 2 only. The shape functions are:

[N(x )]=[(1-x-y) 0 x 0 y OJ
'y 0 (1 -X- y) 0 X 0 y


1
2
t t
(0,0) (1,0)

Figure 7.8: Constant strain triangle loaded by gravity

Problem 13.
CHAPTER 7. DERIVATION OF A FINITE ELEMENT BY VIRTUAL WORI\ 94

The plane strain quadrilateral shown has been developed using the following dis-
placement functions:

u(x,y) q1 + q2x + q3y + q4xy + qsx 2y 2


v(x,y) q6 + q1x + qsy + qgxy + q10x 2y 2

A) Write down the [</>(x, y)] matrix, which relates internal displacements to generalized
coordinates.

B) Write down the [H] matrix, which relates generalized coordinates and nodal coor-
dinates, for the specific element shown.

(-1,1) (1,1)
4 3
y

5 1Lx
(-1,~-A 2

~
Figure 7.9: Quadrilateral element in plane strain

Problem 14.

A triangle has been made of 3 mm thick aluminum sheet (Fig. 7.6). The sides are
100 mm each (an equilateral triangle). Loading has been placed on the system such
that the tip of the triangle has moved to the right (positive x direction) 1 mm. There
is no vertical translation of this point.

Consider a model of the triangle consisting of a single Turner (constant strain)


triangle. If the tip of the triangle is given a virtual displacement of 0.001 mm in
the + x direction, what is the virtual work done by the external forces during this
displacement?

E = 68,950 MPa 1J = 0.3


CHAPTER 7. DERIVATION OF A FINITE ELEMENT BY VIRTUAL WORI( 95

lmm

Figure 7.10: Triangle with tip displacement in x direction


Chapter 8

DERIVATION OF A LINE
ELEMENT BY VIRTUAL WORK

8.1 LINE ELEMENT. ENERGY FORMULATION


A concrete example is needed to tie down the preceding theory. Consider the
constant area, two-node line element (Fig. 8.1).

~A

Figure 8.1: Line element.

We need to find the operators [G), [D), and [N] in the expressions for stiffness :

(8 .1)

and for line load:


{!} e .n.l., = t[ Nf {(X )}dx (8 .2)
The shape must satisfy:

(8.3)

At this point, the shape functions might have the general shape sketched in Fig. 8.2.
They should be the simplest polynomials meeting the conditions of Eqns . 8.3, and these

96
CHAPTER 8. DERIVATION OF A LINE ELEMENT BY VIRTUAL WORI( 97

N(x)
N (x) N/x)
1

~
--"', ' /
/
~
\ ./
\ /
>/
,...-- \
// '\
v \.
0 ...__
--- r
X

Figure 8.2: General shape functions (hypothetical).

would be linear functions:

{u(x)} [N1(x) N2(x)] { ~~ } (8.4)

{u(x)} = [1-x/L x/LJ{ ~~} (8.5)


[N]

In one-dimensional elasticity, Hooke's law is

[G)= [E]
and the strain displacement matrix is

[DJ = [d~J

The intermediate matrix [B] is useful for calculations:

[B] [D][N] (8.6)


d
= [dx][1- x j L xj L]
= [-1/L 1/L]
The stiffness matrix is:

[k] = A t[BjT[GJ[B]dx

= A {L { - 1/L } [E]{ -1/L 1/L}dx


lo 1/L

= At { ~%L }{-E/L EjL}dx


CHAPTER 8. DERIVATION OF A LINE ELEMENT BY VIRTUAL WORl( 98

2 2
[L [ EIL -EIL ]
A Jo -EIL 2 EIL 2 dx

AE [ 1
L -1
-1
1
l
which is the same result as obtained from equilibrium arguments , and is exact for the
constant area case at hand. The equivalent nodal line load becomes

{f} e.n.l., = t[Nf {.C( x )}dx


load per unit length /

{j} e.n. l. ,L rL {
(1 - XI L) (X) } dx
lo (xiL) (x)
foL (1 - xI L ) ( x) dx }
{ foL( xl L ) (x )dx (8.7)

8.2 EXAMPLE OF EQUIVALENT NODAL LOADS


Choose a specific case, such as in Fig. 8.3 where (x ) 0 (xiL) 2 . The equivalent
nodal load is

{j} e.n .l. c

(8.8)

Does this correctly represent the total running load on the element?

Total Load

(8.9)

This checks. The equivalent nodal loads are statically equivalent to the original dis-
tributed load.
CHAPTER 8. DERIVATION OF A LINE ELEMENT BY VIRTUAL WORJ( 99

Figure 8.3: Distributed load.

______________________~)+---+
(~~--+

Figure 8.4: Equivalent nodal loads.

8.3 HOMEWORK
Problem 1.
Find the equivalent nodal loads for the load cases shown in Fig. 8.5 for a two-noded
line element .

The dimensions on 0 are force/length and the dimension on is force. The force
0 is a constant running load at the centerline of the line element and is sketched above
the line only for clarity.

Hint: For the concentrated load, use the Dirac delta function 8(x- x 0 ), which takes
zero values away from the reference point x 0 but has the unit integral
+ oo
/_- oo 8(x- x 0 )dx = 1

Lo{l~ ~I L
< > < >

~L/2:1
L
LL~L/3 j
Figure 8.5: Load distribution on line element.
Chapter 9

BEAM ELEMENT

9.1 CLASSICAL, PLANAR FLEXURE


Consider the straight, slender, prismatic bar shown in Fig. 9.1. It is constrained
to move only in the xy plane. Lateral forces p( x) act along the beam and external
shearing force V and moment M act at the ends of the beam. There are no axial forces
or torsional forces. We will study this beam as an example of a 3-D stress field and
will simplify the problem through geometric arguments related to the slenderness of
the beam. The displacement field usually written as lu J lux, uy, uzJ will instead be
written as lu J = lu, v, w J. Of particular interest is the displacement of the centroidal
axis. If the coordinate system is aligned with this beam axis, the deflection v(x, 0, 0)
is the relevant quantity and is denoted by v( x) .

Figure 9.1: Straight, slender, prismatic bar in xy plane. Coordinate x ax1s passes
through beam centroidal axis .

For a right-handed coordinate system, the classical beam theory for deflections
in the xy plane gives similar sign conventions as for finite element theory. We will
therefore use the xy plane.

The Euler- Bernoulli- N avier approach for flexure is to study the deflection of the
centroidal axis under the assumptions :

1) linear, elastic material


2) small deflections and small slopes
3) plane sections initially perpendicular to the beam axis remain plane and perpen-

100
CHAPTER 9. BEAM ELEMENT 101

dicular to the axis after deformation


4) stress and strain do not depend on the z coordinate.
The strain-displacement relations for the beam come from consideration of bending
to a circular shape (Fig. 9.2). The "plane sections" assumption and the common angle

Figure 9.2: Side view of beam curved to circular arc. Deflection greatly exaggerated.

!J..B for the wedges at distances of R and R-y from the centroidal axis yield

l(y) -l(O)
Ex(x, y) =
l(O)
(R - y )!J..B - R!J..B
R!J..B
y
(9.1)
R
The curvature ~ is found from differential geometry to be, for an arbitrary curve
in the xy plane:
1
(9.2)
R [1 + (d~~x) )2](3/2)
Since the slope ~~ is small (assumption 2)

(9.3)

Using this in Eqn. 9.1, we have

(9.4)

This is the crucial strain component. As seen, it is related to the distance y from the
centroidal axis and the curvature of the centroidal axis. Due to zero axial stress along
it, the centroidal axis will be renamed the "neutral" axis.

Because of the small lateral dimensions of the beam and because surface stresses
CHAPTER 9. BEAM ELEMENT 102

Figure 9.3: End view of beam cross-section looking toward origin.

(Fig. 9.3) are zero or small/ one argues that the following stresses are negligible
throughout the cross-section:
ay(x, y) ~ 0 (9.5)
az(x, y) ~ 0 (9.6)
Tyz(x, y) ~ 0 (9.7)
The plane sections assumption leads to deformations as in a side view of the beam
(Fig. 9.4); by definition
/xy = 0 (9.8)

Figure 9.4: Side view of plane sections grid. /xy = 0.

The plane sections assumption also causes the deformation field seen in the bottom
view in Fig. 9.5:
/zx = 0 (9.9)

Vanishing of the 5 components of stress and strain in Eqns. 9.5-9.9 has an interesting
effect on the strain energy density in the beam:

1
u 2[axEx +Py Ey +P z tz + Txy'}{'y +YC'z /yz + TzxYzJ (9.10)
1
(9.11)
1
The typical student of mechanics will worry about the stress (say O'y) directly under a distributed
load and ask how it can be neglected . Surface pressures are several orders of magnitude smaller than
axial stresses, however , as can be seen by example.
CHAPTER 9. BEAM ELEMENT 103

~~~tllll-~
Figure 9.5: Bottom view of beam with plane sections grid. l zx = 0.

The only contribution to strain energy and to stiffness is through the axial stress
and strain. This is why some engineers view a beam in bending as if it were a bundle of
axial fibers with each fiber acting independently. The stress-strain law of importance
is hence that relating axial stress and strain:

(9.12)

The nonzero terms in Eqn. 9.10 include displacements Ey and c2 , which represent the
free contraction and extension in the lateral direction- a Poisson ratio effect. Although,
there are shearing stresses Tx y, the plane sections argument "locks" the shearing strain
and the shear stresses can do no work .

9.2 FEA MODEL CONCEPTS FOR FLEXURE


Flexure of the bar in Fig. 9.1 will be modeled by a two-noded finite element in as
sketched in Fig. 9.6. It is anticipated that the beam is a "structural" element which will
transmit both forces and moments at the nodes. Rather than use the general notation
lfJ - lf1 !2 j3 j4j, we will use lfJ - [VA, MA, VB, MB] to suggest the di1nensional
=
difference of the force resultants , and likewise [u] lvA, ()A, VB, BBJ .

vA'vA

SA, MA( _t....,_x_ _ _ _ _.t J 8B, MB

Figure 9.6: Finite element model for pure flexure of beam.

The vector unknowns for the beam can be given with the zero components denoted:

ax Ex
VA 0

!
f.y
u(x,y, z)
MA
VB
MB
) 0
Txy
0
tz
0
/yz
{ v(x)
w(x,y,z)
Tzx 0
CHAPTER 9. BEAM ELEMENT 104

The stiffness matrix will depend only on an integration of strain energy of the form
J{ 8c }T {a }dV. We have seen in Eqn. 9.11 that only ax and Ex survive and only the
constitutive law
{ax(X, y)} = [E]{ tx(X , y)} (9.13)
is important. Since Ex (and hence 8cx) is completely determined by the centroidal axis
deformation
d2
{cx(x,y)} = [-y dx 2 ]{v(x)} (9.14)
This is the only necessary strain-displacement relation. The flow chart simplifies to

To complete the finite element, we need to either find [] and [H]- 1 (the displace-
ment function approach) or find [N] (the shape function approach). We will use the
displacement function approach, which is a little more physical in nature, but will later
display the shape functions which are generated.

The displacement v( x) of the centroidal axis will be interpolated by a cubic:

(9.15)

The generalized coordinate q1 represents a rigid body translation (upward) and the
coordinate q2 represents a rigid body rotation about the left end. It is known that
these rigid body modes are needed to allow the element to "drift" to its proper spot in
a loaded structure, without causing stress. The term q3 leads to "constant strain" and
q4 leads to "linear strain" ; refering to the variation of strain in the x direction. The
constant strain term is needed for convergence to correct results (something like the
longest wavelength term in a Fourier series). In matrix form

(9.16)

The field variable {v(x )} creates not only the nodal displacements VA and VB , but
also controls nodal rotations:
BA =dxdv (0) (9.17)
CHAPTER 9. BEAM ELEMENT 105

dv
()B dx (L) (9.18)

This leads to the definition of [H]

(9.19)

The columns in H denote the nodal displacements and rotations corresponding to


rigid body translation, rigid body rotation about the left end, constant strain and
linear strain, respectively. Inversion of [H] yields

(9.20)

The shape function matrix is now found:

(9.21)

Each term in [N] gives the internal displacement field corresponding to a unit
nodal displacement (or rotation). These terms are shown in Fig. 9.7, where each takes
a unit value in one degree of freedom at its "home" node and zero values at all other
nodal degrees of freedom. We could have directly used this definition to construct
the shape functions. The direct construction of shape functions is a more modern
way to develop a finite element, but bypasses the physical intuition found from the
displacement functions.

Having found the necessary mappings for the beam element, we are ready to derive
the stiffness matrix and the equivalent nodal loads for line loads ( x). For stiffness,
the [B] matrix is:

[B] [D][N]
d2
[-y dx2][N]
-I 3
[-6L + 12x, -4L 2 + 6Lx, 6L- 12x, -2L 2 - 6Lx] (9.22)
CHAPTER 9. BEAM ELEMENT 106

Figure 9. 7: Shape functions for two noded beam in pure flexure .

The stiffness matrix is

[k] = t j j[Bf[G][B]dydzdx (9.23)

The [B] matrix in linear in y and the total integrand is quadratic in y, leading naturally
to the definition of area moment of inertia of the cross section:

(9.24)

Evaluation of the integral in Eqn. 9.23 , using the definition of I, leads to:

(9.25)

The equation of equilibrium for the beam is

[k] { U} = {f} + {J} enl , (9 .26)

The general expression for equivalent nodal loads:

can be evaluated in specific cases.

9.3 EXAMPLE OF PLANAR BENDING OF BEAM


Find the rotation at the ends of a uniform beam with pinned ends when subjected
to a line load of 10 N/mm (57.1 lb/in) , as shown in Fig. 9.8. Use a single element to
model the problem. The beam has properties:
CHAPTER 9. BEAM ELEMENT 107

EI 10 10 N mm 2 (3.48 x 10 6 lb in 2 )
L 1000 mm (39.37 in)
lJ 0.3

P(x)=lO N/mm (57.1lb/in)

f!!!!!!!f!!!!f
J'\
7771777
(\
77l777T
I
4 lOOOmm I
(39.37in)

Figure 9.8: Beam with uniform load.

Solution

Find [k]:

Now find the equivalent nodal loads.

{J}e.n.l.

Substantial moments result. (As the mesh is refined, however, forces vary as L and
moments vary as 1 2 , and moments disappear in the limit as L ---+ 0. This might be
expected.
CHAPTER 9. BEAM ELEMENT 108

5000 N 5000 N

cf-..u.____
833,000 N mm
___J

833,000 N mm
f)
Figure 9.9: Equivalent nodal loads.

ll
Now , proceed with the solution:

60~~ 4 x ~~~ -6~~~ 2 x ~~~


6 6

10 [
-12 -6000 12 -6000
IJ~0 ) = ~ VB~A )
+
~ 83~:5, ~~~
000
)
6000 2 X 10 6 -6000 4 X 10 6 OB 0 -833,000
The equations are uncoupled by writing the first and third equations:
60,000BA + 60 , 000BB VA+ 5000
-60 , oooeA- 60 , oooeB = vB + 5ooo
and the second and fourth equations:
4 X 10 7 0A
+2 X 10 7 BB 833,000
7 7
2 X 10 0A + 4 X 10 0B -833, 000
Solving for eA and eB in the latter equations:

} = { _ 0.0417 } radians
{ eOBA .
0 0417
These rotations are sketched in Fig. 9.10.

Figure 9.10: Rotations at nodes of example problem.

The external reactions VA and VB are then found:

{ ~; } = { =~~~~: }N.
The complete equation of equilibrium is:
equiva lent nodal loads

0.~417 ~:~:: )
5,000 )
[ ]{ l J ) J + J 833,000

l -0.0417 l 0 l 5,000
-833,000
nodal di splacement s external f orces (reactions)
CHAPTER 9. BEAM ELEMENT 109

The above numerical answer can be compared to classical Euler-Bernoulli- N avier


theory and is found to be exact . This means that the cubic shape functions are sufficient
to exactly model the uniform loading case for pinned ends. In fact, cubic shape functions
are also exact for any combination of concentrated loads at the nodes, including all
types of end conditions.

9.4 EMBEDDING BEAM INTO 3-D


We have so far studied pure flexure of a bar in two dimensions. Let us consider
more general behavior of a bar (frame member) in 3-D . This requires an element
coordinate system (Fig. 9.11) and addition of axial motion, torsion and bending in the
xz plane. The previous stiffness matrix relates u 2 , u 6 , u 8 and u 12 in this more general
notation. The xy plane in element coordinates is called plane 1 and the moment of
inertia opposing motion in that plane is called 11 . The relation for bending in the xy
plane is rewritten:

6L
4L 2
(symm)
-12
-6L 2
12 -6L~~
4L 2
ll (9 .27)

l:::x
Figure 9.11: Coordinates for 3-D beam including axial and torsional motion.

9.4.1 Flexure in xz Plane


The relation between centroidal axis and nodal angles in the xz plane is different
from the xy plane

dw(O)
us=---
dx
CHAPTER 9. BEAM ELEMENT 110

dw(L)
un = -
dx
and causes different shape functions. The stiffness becomes

-6L - 12
2
4L 6L
(symm) 12

9.4.2 Axial Motion


Degrees of freedom u 1 and u 7 are needed to describe axial motion of the beam.
Neglecting nonlinear interaction between axial and lateral motion, the same relation
holds as for the line element:

EA [ 1 -1
L -1 1
l[ l l
u
U7
1
= [ /1
!7
(9 .28)

This relation neglects nonlinear effects, such as in Euler column buckling.

9.4.3 Torsion
The torsional displacement of an element is described by

(9.29)

where J is the torsional stiffness constant and G is the shear modulus . The constant
J reduces to the polar moment of inertia only for certain compact cross-sections, such
as solid squares and circles and circular tubes .

9.4.4 Combined Beam Stiffness Effects in 3-D


Assembling all stiffness effects for flexure in two planes, axial and torsional motion,
one has the stiffness shown in Fig. 9.12 . This stiffness presumes that the planes 1 and
2 have been specially chosen as principal planes of inertia. In each of these planes,
the motion in the plane acts independently from the other plane. (The derivation of
flexure as a planar problem implied this, though nothing was said at the time!)

To uniquely prescribe plane 1, one can define a vector v centered at node A of the
beam (Fig. 9.13). Plane 1 passes through the axis of the beam and vector v. This also
establishes the local y coordinate axis. Plane 2 is then defined as passing through the
beam axis and orthogonal to plane 1. This then provides the local z coordinate axis.
CHAPTER 9. BEAM ELEMENT 111

EA EA
L --y;-
12Ei] 6Eh _12EJ1 6EI1
3 2 3 2
12Eh _6Eh _12Eh _6ff2
3 2 3
GJ GJ
L - L
_ 6EI2 4Eh 6Eh 12Eh
2 L 2 L
6Eh 4EI1 _6EI1 12EJ1
2 L 2 L
-EA
y L
EA
_12ffill _6f!1 12ffill _6f!l
_12ffih 6f!2 12ffih 6f!2
GJ GJ
-L L
_6Eh 12Eh 6EI2 4Eh
2 L 2 L
6Eh 12Ei) 6Ei) 4EJJ
2 L - 2 L

Figure 9.12: Stiffness for beam in 3-D including axial motion and torsional motion.

y
e

Figure 9.13: Vector v defining plane 1, and axis Ye


CHAPTER 9. BEAM ELEMENT 112

9.5 THIN-WALLED BEAMS. WARPING.


The theory to this point has dealt with compact, prismatic beams (bars). If we wish
to extend the results to thin-walled, open cross-section beams, new effects can occur.
The auto industry, for instance, uses many frame members made of rolled sheet metal
in the form of C sections or more complicated shapes . When such beams are subjected
to torsion, the cross-section will distort . Some codes, such as MSC / N ASTRAN, have
special elements to allow for this behavior. In MSC/NASTRAN the simple beam
element is called a BAR and the thin-walled element is called a BEAM . Users of
each program must determine how much detail is built into the beam elements in the
program.

9.6 DEEP BEAMS. SHEARING DEFORMATION.


When beams become relatively deep (Fig. 9.14) the bending stiffness in the xy
plane becomes so large that relatively speaking, the shear stiffness becomes smaller.
As a result, shear deformation (Fig. 9.15) is not negligible and the "plane sections"
assumption made in the Euler- Bernoulli beam does not hold . Some elements, such
as the BEAM element in the general purpose code SAPIV automatically include this
effect. Engineers refer to the beam as a Timoshenko beam when shear deformation is
included.

~A---------.
~
~
~A-------------~

/:
~ p

Figure 9.14: Side view of deep beam.

9. 7 END RELEASE CONCEPT


An important concept in practical problems is the "end release." Many times the
beam element is used as a mechanism in which an internal pinned joint or slider occurs.
The reader will recall that internal nodes in finite elements are by default considered a
continuous connection (welds) rather than pins . General purpose codes allow the ends
of beam elements to act as if pinned or sliding in a slot . Any or all of the 6 degrees of
freedom per end can be so treated.

I ~
CHAPTER 9. BEAM ELEMENT 113

Figure 9.15: Side view of shear deformation in deep beam.

An example of various end releases is given in Fig. 9.16. Here a beam is sliding in
a pinned arrangement at both left and right ends, with the coordinate system shown.
Consider the 6 d.o.f. at the left end, numbered consecutively 1 through 6. The double
pin arrangement allows unrestrained motion in x translation (1) , y translation (2), y
rotation (5) and z rotation (6) . At the right end, the released d.o.f. are x translation
( 1) and x rotation (4). Finite element programs will properly release the force (or
moment) in each of these d.o.f.

r4tf---i1
( crz=) 0
Figure 9.16: Beam with slider mechanisms at ends.

End releases play an important role in modeling the tracks that support automobiles
in auto- haul trailers. See the case study following for more details.

9.8 BEAM OFFSET


When a beam is used as a stiffener for sheet material, its shear center does not lie
in the neutral surface of the sheet, but rather is displaced a short distance. It is handy
for this case, and others, to have the finite element program provide the additional
nodes that define the beam in its offset location (Fig. 9.17) . The usual case is that the
user has provided node points A and B and wishes to shift the ends of the beam by
distances Wa and wb respectively without defining nodes at the beam ends.
CHAPTER 9. BEAM ELEMENT 114

Yelement
Zelement

B
Figure 9.17: Offset vectors for locating beam shear center away from nodes A, B .

9.9 HOMEWORK
P r oblem 1

Write out the assembled equations of equilibrium for a two-element representation


of the clamped-clamped beam shown in Figure 10.19 using numerical values for loads
and proper end conditions. Use symbols for element lengths L 1 and L 2 , moments of
inertia 11 , 12 and modulii E 1 , E 2 .

1000 N

i (225 lb)

Figure 9.18: Beam built in at both ends.

P r oblem 2

Two straight beams are welded together at an angle of 30, described by the in-
tersection of the two scribed lines representing the neutral axes (Fig. 9.19) . The body
is then loaded. A FE analysis predicts that a rotation of 10 should occur at node 2.
After loading, then, what is the angle f3 at node 2, as indicated by the intersection of
the same scribed lines?
(a) 20
(b) 30
(c) 40
CHAPTER 9. BEAM ELEMENT 115

Figure 9.19: Angled beam before and after loading.

(d) None of the above


(e) There is not enough information to tell.
Problem 3

A uniform beam is fixed at both ends as shown in Fig. 9.20. Using finite element
theory, and using two Euler-Bernoulli beam elements, find the rotation () at the center
of the beam due to a moment MB only.

~ ~ ~
% eB' M B ~

~I-2L~I
Figure 9.20: Clamped-clamped beam.

Problem 4

Given a slender, uniform beam as shown in Fig. 9.21, calculate the rotation ()B at
the right end if:

tVA VBt
eA ( 1.____ _ _ ____.1) 88
1000 mm
(39.37in)
~
Figure 9.21: Loaded uniform beam.

VA= 0
CHAPTER 9. BEAM ELEMENT 116

BA = 0
VB= 0
MB = 1.13 x 10 7 mm N (10 5 in lb)

~; = 175 N/mm (10 3 lb/in)


Problem 5
A slender elastic beam is clamped at both ends by rigid walls as shown in Fig. 9.22.
This support constrains both the deflections and rotations at the ends. A lateral
force of magnitude P2 acts at the center of the beam. Using known stiffness matrices ,
calculate the deflection at the center of the beam due to P2 Use two Euler- Bernoulli
beam elements. The beam has stiffness EI and length L.

Figure 9.22: Beam with center load.

Problem 6
A beam is clamped at both ends (Fig. 9.23). A rod extending from above helps to
support it. Find the deflection of the beam at the center when a 4450 N (1000 lb ).
load is applied as shown.
EI = 5. 74 x 10 12 N mm 2 (2.00 x 10 9 lb in 2 ) for the beam
EA = 2.89 x 10 7 N (6.50 x 106 lb) for the rod
(This problem appeared on the national Professional Engineer's Exam several years
ago. It required one hour to solve by conventional analysis, but is done in 20 minutes
by FEA.)
Problem 7
An Euler-Bernoulli beam (Fig. 9.24) has the following properties:

A = 6451 mm 2 (10 in 2 )
f zz = 6.243 X 10 6 (15 in 4 )
E = 2.07 x 10 5 N/mm 2 (30 x 10 6 lb/in 2 )
v = 0.3
L = 5080 mm (200 in)
CHAPTER 9. BEAM ELEMENT 117


t
1000 mm

1500
mm
4450 N (1000 lb)

Figure 9.23: Beam-rod system.

Figure 9.24: Cantilever beam.


CHAPTER 9. BEAM ELEMENT 119

v(L)

Figure 9.27: Prescribed deflections.

midpoint is deflected 0.4 mm upward and 0.1 rad clockwise (Fig. 9.27), what is the
vertical (shear) force on the left end of the beam?
Comments: Use finite element concepts. Do not derive any stiffness matrices or equiv-
alent nodal load vectors from first principles.
Problem 11

For the beam element based on Euler-Bernoulli theory and with displacement func-
tion v(x) = q1 + q2 + q3 x2 + q4 x3 , develop the equivalent nodal loads for the loading
shown in Fig. 9.28.

Figure 9.28: Linear load distribution.

Problem 12

A beam has a concentrated load Pat a point 21/3 from its left end (Fig. 9.29). What
fraction of this load should be applied to the right end of the beam as an equivalent
nodal load, if only one beam element is used?

(a) 0.75

(b) 0.71
(c) 0.66
(d) 0.60
(e) 0.50
CHAPTER 9. BEAM ELEMENT 120

Figure 9.29: Concentrated load.

Problem 13

A beam has a load over half its span, as shown in Fig. 9.30. If modeled by a single
element, find the vertical component of the equivalent nodal load on the left end.

200 N/mm

t!t!t!t!!J
I so mm I
Figure 9.30: Distributed load.

A 3226 mm 2 (5 in 2 )
E 6.90 x 10 4 N/mm 2 (10 7 psi)
v 0.3

Problem 14

A bar bell (for weight-lifters) is modelled as a beam using 3 two-dimensional Euler-


Bernoulli beam elements (Fig. 9.31). The points where the weight-lifter's hands are
applied are considered pinned. No axial displacements are considered. Consider motion
constrained to the xy plane so that only two degrees of freedom per node need to be
considered.
A) What characterizes the term k18 in the assembled stiffness matrix? (This term
related the vertical displacements at node 1 to the moment created at node 4.)

(i) it is negative, (ii) it is zero, (iii) it is positive, (iv) one cannot tell from the data
g1ven

B) How many of the 8 degrees of freedom are constrained due to boundary condi-
tions?
CHAPTER 9. BEAM ELEMENT 121

Figure 9.31: Weight lifter.

(i) two, (ii) three, (iii) four, (iv) six

C) From the sketch shown, try to relate k14 and k 58 .

(i) k14 = kss


(ii) k14 = -kss
(iii) k14 = kss = 0
(iv) one cannot tell from the data given

D) What is the relation between k 14 and k16 ?


(i) k14 = k76
(ii) k14 = -k76
(iii) k14 = k76 = 0
(iv) one cannot tell from the data given
Problem 15.
A cantilever beam has a knife edge (pinned) support at its midpoint (Fig. 9.9).
The end of the beam is loaded in a way as to cause a tip deflection of 1 mm and a tip
rotation of .01 radian counterclockwise. What is the vertical reaction of the support
on the beam? (i. e., what is the external load acting on the beam at its midpoint?)

EI/L 3 = 200. N/mm L = 1000. mm

Problem 16.

A beam is clamped at the left end and pinned at the right, as shown. The tip has
a moment of 106 N mm applied to it.
CHAPTER 9. BEAM ELEMENT 122

t
l
Figure 9.32: Beam with midspan support

A) What is the tip rotation?


B) What is the stored strain energy in the beam at equilibrium?
C) What is the potential energy (as used in the potential energy theorem) of the system
at equilibrium?

10 4 N/mm
10 3 N/mm
1000 mm
Problem 17.
A cantilever beam is subjected to an unknown tip force P and moment Mas shown,
and has come to static equilibrium. The equilibrium tip deflection is 0.5 inch and the
tip rotation is .01 radian counterclockwise.

A) What is the virtual work done by the external load during a virtual displacement
of .01 inch downward at the tip of the beam?
B) What is the virtual strain energy change during the same virtual displacement
as in part A) ?

E 30. x 10 6 psi
v 0.3
L 1000 mm
A 100 mm2
I 1. x 10 5 mm4
Chapter 10

STRESSES IN AUTO-HAUL
TRAILER

10. 1 INTRODUCTION
Large, open frame trailers (Figure 10.1) are used to transport new automobiles
from factory to dealerships. The particular trailer shown carries seven automobiles.

Figure 10.1: Loading empty autohaul trailer onto rail flatcar

The frame is made of rectangular steel tubing and is entirely welded. The sidewall
(Figure 10.2), is the basic structure to be stress and deflection analyzed. The rear
wheels, trailer suspension and auto support system are standard commercial products
and are not in question.

This particular trailer, Traffic Transport Engineering, Model M-183, is to be used for
railroad service, with two trailers per flatcar (Figure 10.3). The trailer must therefore
withstand railroad loads as well as over the road (highway) loads. The railroad loads

123
CHAPTER 10. STRESSES IN AUTO-HAUL TRAILER 124

Figure 10.2: Two loaded autohaul trailers on rail flatcar

lJ: I
00
\t)J lJ: I \t}J
00

'A~
00 00

Figure 10.3: Conventional sidewall trailers and lowboy trailers

include loading the trailers on the flatcars and the "humping" loads when flatcars are
coupled.

The trailer hitch is called a "fifth wheel" and consists of a kingpin on the trailer
which protrudes into a slotted plate on the truck tractor. The kingpin is an integral
part of the trailer. When the trailer is loaded on the flatcar, the kingpin is locked into
a stanchion on the rail car. This fixes the front end of the trailer; however, the rear
end of the trailer is free to roll on its wheels.
The railroad industry specifies that such trailers withstand standard humping loads
as generated in carefully controlled tests. The flatcar is rolled into a stationary, heavily
loaded gondola car, at four, six and eight mph. This impact is to be applied in both
fore and aft directions. Side loads are not considered.
Another type of loading on the trailer is due to lifting of the loaded trailer by its
CHAPTER 10. STRESSES IN AUTO-HAUL TRAILER 125

upper four corners and placing it on the flatcar. This causes localized corner loads on
the trailer not normally experienced in over-the- road usage.

The trailer is typified by an intricate placement of side members, apparently the


work of hours of thoughtful structural design. In fact, the members are positioned to
allow the driver who has loaded the autos to open the car door and climb off the trailer!
The trailer is almost symmetric from side-to-side, but has a few members which differ.

10.2 PHYSICAL MODELING


Physical assumptions in stress and deflection problems usually involve geometry,
materials and loads.

10.2.1 Geometry
The manufacturer has previous experience in designing the tracks carrying the
autos, the transverse members, the kingpin area and the wheel-suspension area. The
new feature is the sidewall. Because the trailer is loaded most heavily in the fore-
aft direction, and because of near symmetry, it will be assumed the sidewalls act
independently (because of near symmetry) and the left and right sidewalls will be
studied separately. Each sidewall model will be loaded with half the weight of the
autos.

The cross-sectional properties of the thin-walled steel tubes will be taken from
handbooks where possible and will be calculated on the basis of "square-cornered"
tubes otherwise. (The slight loss of material in the bend radius will be neglected.)

The individual tubes are beam-like in nature and can be considered as beams . The
structure will be therefore treated as a two-dimensional frame.

Before finite element methods existed, auto haul trailers had been stress analyzed
by assuming that the entire sidewall acted as a single Euler-Bernoulli beam. This
implied that plane sections perpendicular to a neutral axis before deformation remain
plane and perpendicular to that axis afterwards . The deflections found here will show
how oversimplified such an approach would be.

10.2.2 Materials
The sidewall is constructed of low-carbon welded steel tubing. The modulus of
elasticity is taken as E = 30 ,000,000 psi, and Poisson's ni.tio is v = 0.3. For stress and
CHAPTER 10. STRESSES IN AUTO-HAUL TRAILER 126

deflection studies, the yield stress is important only at the conclusion of the study for
failure analysis .

10.2.3 Loads
For over-the-road (highway) use of the trailer, the basic load is 1 g downward and
no dynamic effects are to be considered. This is a simplification of the true dynamic
environment on the road.

In the rail-haul case, the trailer is slowly lifted by its upper corners, involving again
a 1 g downward load with the upper corners of the frame constrained and the wheels
and kingpin free (unconstrained) .

When "humping" occurs at eight mph, severe dynamic response occurs. The details
are important because the wheels of one loaded automobile may be only two inches
from the windshield of the next! Unfortunately, it was not economically possible to
carry out a full dynamic analysis . An equivalent static loading is needed for stress
and deflection studies . After much thought and comparison with auto and aircraft
crash studies, it was decided to design to 2 g fore and aft for the humping loads .
In retrospect, this was reasonable, inasmuch as calculations show the trailer will flip
end-over-end at 2.8 g. The trailer surely need not be designed to withstand humping
loads high enough to overturn the trailer in this way; therefore, the value of 2 g seems
sufficient .1

The way the automobiles are suspended in the trailers is interesting. The wheels
are captured in sheet metal tracks which are initially raised by hydraulic jacks (Fig-
ure 10.4) . The ends of the tracks are pinned in place by a variety of linkages, and then
the hydraulic pressure is released (Fig. 10.5) . The system is statically determinant,
i.e., the system would collapse if the pin were pulled out (in the half-model).

10.3 FINITE ELEMENT MODELING


The general purpose program SAP6 was used with beam elements to model both
right and left sidewalls . The grid for the left sidewall is shown in Figure 10.6. Many
beam connections are, in fact, reinforced by gussets; but the gussets are neglected
in the finite element model. The stresses at the joints will therefore be somewhat
overestimated.

The automobile support system is modeled by connecting the automobile center of


gravity to the appropriate points on the sidewall by rigid, massless beams (Fig. 10. 7).
The entire sidewall and automobile support system is shown in Fig. 10.8. Two
1
The parties involved have now agreed that the 2 g loading is proper for these trailers.
CHAPTER 10. STRESSES IN AUTO-HAUL TRAILER 127

c~

{Q ~
~tracks __J

Figure 10.4: Track system

~pm /

Jack

Figure 10.5: Method for raising (above) and holding (below) the auto

computer runs were made for the left side wall, involving a total of five load cases
(Table I). The corner lift (Run 1) involves nodal constraints in the vertical direction
at the two upper corners and three horizontal constraints at upper corners to remove
rigid body motion in the horizontal plane (yaw, fore-aft translation and side motion).

The "on wheels" cases (Run 2) were given boundary conditions where both rear
wheels to remain in contact with the ground. In addition, the kingpin is fixed in space.

Later runs made on similar trailers have modeled the suspension system more accu-
rately, allowing more realistic load distribution on the rear wheels. This is not a trivial
modeling problem, partly because the air suspension system tends to equalize load on
the two rear axles. One way to model equal loads on the axles is to use a whiffletree
CHAPTER 10. STRESSES IN AUTO-HAUL TRAILER 128

Figure 10. 6: Sidewall, made of cold rolled steel tubing

Figure 10.7: Scheme for modeling the support system

(Fig. 10.9) .

The model of the right sidewall of the trailer required 168 elements, 123 nodes and
366 degrees of freedom . The half-bandwidth, after resequencing, was 36.

10.4 ANALYTICAL RESULTS


Both deflections and stresses were found for the sidewall and will be discussed in
turn.

10.4.1 Deflections
The deflections in the right sidewall for the load cases 1, 4 and 5 are given in
Figures 10.10- 10.13. The deflections are reasonable; there is less than an inch droop
at the trailer center in all load cases. An unexpected result (although obvious, in
retrospect) is that the front impact case is much more severe than the rear impact
case. Because the kingpin is fixed and the rear wheels can roll, the front impact puts
the trailer in compression. This causes a droop which is additive to the pull of gravity.
The rear impact case puts the trailer in tension and cancellation with gravity occurs.
CHAPTER 10. STRESSES IN AUTO-HAUL TRAILER 129

Figure 10.8: Finite element mesh including supports

Table 10.1: Load cases for trailer study.

Run Number Support System Load Direction


1 Corner lift 1g l
2 On wheels 1g l
3 " 2g +--
4 " 2g ---+
5 " 1g l 2g +--

The outdated assumption used before finite element methods of "plane sections" in
treating the entire sidewall as a single beam is seen to be highly inappropriate.

10.4.2 Stresses
The stresses found during over-the-road conditions (1 g downward, load case 2)
were less than 20,000 psi throughout the sidewall. Likewise, for the rail corner lift (1
g downward, load case 1) stresses were less than 20,000 psi throughout.

For the humping conditions, the rear impact at 2 g plus normal gravity caused
stresses which were under 20,000 psi throughout the sidewall. The frontal impact at
2 g plus normal gravity did cause stresses in the 30,000 to 40,000 psi range in several
members and one member was found to have 4 7,300 psi stress. This member was
a vertical tube being "pinched" between loads carried by cross members which were
eccentrically aligned (Fig. 10.14). A recommendation was made to have the joint
aligned (Fig. 10.15) so that large bending stresses are not generated in the vertical
member. 2
2
The reader should be reminded of the collapse of walkways in the Hyatt Regency Hotel in Kansas
City in 1981 in which such an eccentric joint failed under large tensile forces.
CHAPTER 10. STRESSES IN AUTO-HAUL TRAILER 130

Figure 10.9: Boundary conditions at rear wheels : "whiffletree"

Figure 10.10: Loaded trailer. Corner lift.

10.5 EXPERIMENTAL RESULTS


The results of the analysis were used as a preliminary check of the trailer design .
The final acceptance of the design was to be based on actual loading and humping tests
carried out by Southern Pacific. These tests were done in 1981; the rail-haul trailer
passed the tests in all regards.

10.6 CONCLUSIONS
Two prototype trailers were built in 1980, were tested in 1981 and have been under-
going service testing in California, carrying General Motors automobiles . The design
is under consideration for fleet service. It is likely that the trailer will be redesigned
and made larger before production occurs .
CHAPTER 10. STRESSES IN AUTO-HAUL TRAILER 131

Figure 10.11: Loaded trailer. Front impact.

Figure 10.12: Loaded trailer. Front impact plus gravity.

Figure 10.13: Loaded trailer. Rear impact plus gravity

high
stress t-----

Figure 10.14: High stress region due to nonaligned joint


CHAPTER 10. STRESSES IN AUTO-HAUL TRAILER 132

'~ \ 4

\ (

Figure 10.15: Properly aligned joint


Chapter 11

COMBINING DIFFERENT
ELEMENTS

A common difficulty in finite element modeling is that one part of a structure is best
modeled by solids and another part by plates (Fig. 11 .1). How does one join the two
types of elements? A false start is sketched in Fig. 11.2, where the plate elements will
"fall down," i.e., there is a rigid body rotation of the plates about the weld line. The
reason is that the underlying space has 6 d.o.f. per node whereas solids only contribute
stiffness in 3 d.o.f. and plate in 5 d.o.f.

weld line

Figure 11 .1: Structure made of casting and sheet metal.

The general problem is that of joining "elasticity" elements (plane stress and strain,
rods and solids) to structural elements (beams, plates and shells). The structural
elements have rotations and moments defined at nodal points. On the other hand,
elasticity elements provide no elastic stiffness opposing a rotation at a point . This
means a possible hinge is created at every intersection of an elasticity element and a
structural element .

Elasticity elements have only displacements and no rotations defined at a point.


An attempt some years ago was made to introduce the notion of rotational d .o.f. in
elasticity, the "coupled stress" theory, but this never became a general analysis tool.

A popular way to prevent hinging of structural elements at the joint is to provide


extra support by building a framework of beam elements such as the "H" arrangement

133
CHAPTER 11. COMBINING DIFFERENT ELEMENTS 134

soli~ plates
~=t==l r.~ .
Figure 11.2: Finite element mesh, with solids and plates.

shown in Fig. 11.3. These beam elements are taken to be relatively stiff compared with
the plate; indeed, rigid beam elements are sometimes used.

=/==~J--------7'_....__'-_
________v =/----.
Figure 11.3: Transition between solids and plate, modeled with beams.

Another method is to imbed the plate into the solid structure as in Fig. 11.4. This
method depends on the capability in FEA to superpose elements in the same space.
The discretization merely adds stiffness terms to nodal d.o.f., and there is no problem
in overlaying elements. Of course, a slight over-estimation of mass and stiffness may
occur.

~plate extension
( ~ into solid

Figure 11.4: Transition modeled by extending plate into solid.

One of the most widely quoted articles on the joint between plate and solid elements
is the paper by Feld and Sou dry. [1] In this work, the authors discuss several modeling
techniques which involve rigid elements, both of a bar and a general element type.
The recommended procedures involve use of 2 or more such rigid body elements that
carry the moment between the plate and the solid body. The work was inspired by the
intersection between turbine blades and the turbine disk.

One combination of elements is to be avoided - that of axisymmetric elements


with any other. The need would occur in a problem having axisymmetry with some
CHAPTER 11. COMBINING DIFFERENT ELEMENTS 135

exceptions such as the cooling fins in Fig. 11.5. The difficulty is that the axisymmetric
element is a "ring" element which extends around the axis but is modeled only in the
xz plane (say). The fins in this case do not extend around , and the use of a plate
element for the fin (Fig. 11.6) would cause misleading results. There are two ways to
model the cylinder. First, one can use an idealized axisymmetric model. You must use
"smeared" properties for the fin to represent stiffness and/ or thermal properties as an
average around the ring. This is particularly useful in heat conduction where the path
of conduction is accurately known. Second, one can take a pie-shaped wedge and use
either plates or solid elements alone.

Figure 11.5: Finned, hollow cylinder.

Figure 11.6: Cross-section AA of finned cylinder.

A table of stiffness contributions is given below to show the differences in elements.

Number of D.O.F.
Element Receiving Stiffness
Contribution Per Node
Line element (rod, spring, truss) 1
Plane stress (membrane) 2
Plane Strain 2
Axisymmetric 2 (but phantom stiffness
e
in direction)
Solid 3
Beam (bar) 6
Plate, Shell 5 or 6, depending on
formulation
Bibliography

[1) Feld, D. J. and Soudry, J. G., "Modeling the interface between shell and solid
elements," Proceedings of the 1983 MSC/NASTRAN User's Conference, The
MacNeal-Schwendler Corp, Los Angeles, CA, March 1983.

136
Chapter 12

LINEAR STATIC EQUATION


SOLVERS

12.1 GENERAL
Three general categories of equation solvers are used in structural finite element
work: 1) linear algebraic equation solvers for static stress and deformation 2) eigen-
value solvers for buckling and vibration 3) numerical integration in time for transient
vibration problems. We are interested in the first case at present, and the second and
third approaches will be discussed in later chapters.

The linear algebraic problem is

[K]{u} = {P}

where F is known and u is unknown. One can distinguish 4 different alternatives , called
the Fredholm alternatives:

1) Det[K] =I 0 P =I 0 Linear, static stress analysis .


2) Det[K] # 0 P = 0 Trivial case. No deflection.
3) Det[K] = 0 P = 0 Eigenvalue problem.
4) Det[K] = 0 P # 0 Pathological. Of no use in statics .
We are interested in the first alternative.

12.2 LINEAR STATIC STRESS ANALYSIS


Consider the equilibrium equation

[K]{u} = {P} (12 .1)

137
CHAPTER 12. LINEAR STATIC EQUATION SOLVERS 138

We assume that this equation is in a standard form where [I<] and { P} are known
and { u} is unknown, and [I<] is N x N, real, symmetric and positive definite. Positive
definite means that {u} [I<] { u} > 0 for all nonzero { u}. (Any displacement causes
positive strain energy.)

The solution for { u} involves methods related to inversion of [I<]:

(12 .2)
but inversion is not done in practice. An equivalent process is done, usually through
the use of "triangular decomposition" of [I<] as explained below.

12.3 LINEAR, STATIC EQUATION SOLVERS


Static stress and deflection equations can be solved by direct or iterative methods:

12.3.1 Direct Methods

Elimination (learned in high school for 2 or 3 equations)


Cramer's rule (learned as college freshman for 3 to 5 equations)
Gauss elimination (learned as college sophomore)
Wavefront (not often taught)
Dissection (not often taught)

12.3.2 Iterative Methods

Southwell relaxation (developed in the 1930's)


Gauss-Seidel
Jacobi
Chaotic random (used in parallel processing schemes)

The direct methods, except for Cramer's rule, are all related to Gauss elimination
but with differing order of elimination. Gauss elimination will be discussed in detail
later because it is the most popular static solver at present . Gauss elimination takes a
predetermined number of steps.

Iterative methods have not been used as much as direct methods in recent years
for linear problems . Iterative methods can take fewer steps to give the answer but are
not as reliable as direct methods. Some people predict iterative methods will be used
more as vector processors become more powerful.
CHAPTER 12. LINEAR STATIC EQUATION SOLVERS 139

12.4 STORAGE CONCEPTS


The cost of using an equation solver depends on the number of floating point oper-
ations (FLOPS), but depends even more on the occupancy of high-speed memory, by
a ratio of 2 to 1. The way the coefficients are held in memory is more important than
the details of the solver algorithm!

How much high-speed memory is required depends on the storage scheme:

pointer methods for sparse matrices

banded methods

wavefront method

The wavefront method of equation solution both a storage method and a solver. We
will discuss storage in a separate chapter on bandwidth concepts.

12.5 ADVANTAGE OF TRIANGULAR FORM


FOR SOLUTION
As motivation for the following section, consider the set of equations

(12.3)

This equation is in a triangular form and is particularly easy to solve from top to
bottom.

3u2 = 8- 2(7)

6u3 = 12- 4(7) - 5( -2) U3 = -1


If the equations had been in an inverted triangular form

(12.4)

they would have been equally as easy to solve, but from bottom to top.
CHAPTER 12. LINEAR STATIC EQUATION S OLVERS 140

12.6 TRIANGULAR DECOMPOSITION


If [I<] is an N x N symmetric, positive definite matrix, the equation

[K]{u} = {P} (12.5)


can be put in a triangular form by two factorization decompositions of [I<]
[K] = [L][L]T- [U]T[U] ( Cholesky factorization)
[I<] := [L1][D][L1]T := [UI]T[D][UI] (Gauss-Doolittle factorization, where L1
and ul have unit diagonal terms)
or by two additive decompositions:

[K] = [ K1 O] +[ ~~ 0
K 2]
Gauss-Seidel decomposition (12.6)
0 0 0 0

[K] - [ O Kt o ] + [ Ko2 o o I<o2] Jacobi decomposition (12 .7)

Factorization decompositions are typically used for direct solvers whereas additive
decompositions are used in iterative solvers. Factorization can require on the order of
n 3 floating point operations and is the expensive part of a direct solution.

Gauss- Doolittle factorization is also called the modified Cholesky decomposition,


or the LDU decomposition. In the latter case, versions for symmetric and asymmetric
matrices are distinguished by context.

12.7 GAUSS ELIMINATION


Consider Gauss elimination with a Gauss-Doolittle factorization. Banding will not
be considered; the shifting and truncating processes needed for banded matrices can
be added later.

Assume a linear, static stress problem in the standard form where all components
of { u} are unknown:
[I<] { u} = {P} (12 .8)
where [K] is real, symmetric and positive definite. For the moment, assume that [I<]
has already been successfully decomposed by a Gauss- Doolittle factorization .

(12.9)
CHAPTER 12. LINEAR STATIC EQUATION SOLVERS 141

Define X as an intermediate vector

(12.10)

and introduce it into Eqn 12.9 to get

(12 .11)

This is a triangular form which is simple to solve from the top down and is called
the "forward solution." One then uses the definition of {X} to find {u}.

[D)[L 1]T {u} = {X} (12 .12)


~ '-v--'
unknown known

Premultiplication by the diagonal matrix does not destroy the upper triangular form
of [L 1]T and the LHS is triangular

(12 .13)

This is easily solved from bottom to top, and is called "back substitution."

To summarize, Gauss elimination consists of decomposition, forward solution, and


back substitution.

12.8 PROGRAMMING GAUSS ELIMINATION


The details of this method are given because of its importance and the academic
value received for a modest amount of study. The first step is the decomposition of
[K]:

(symm)

Accounting for symmetry, one has a set of ~ ( n 2 +n) equations in ~ (n 2 +n) unknowns.
Try to unravel the problem. Look at the first column of [I<] :

kn = dn Idn = kn I (Algorithm 1)

(12.15)
CHAPTER 12. LINEAR STATIC EQUATION SOLVERS 142

The recursive formula for li 1 is

Itil = ki1/ dn I (i = 2, 3, ... , n) (Algorithm 2) (12.16)


In the remaining columns of [k], we start at the diagonal term

d22 = k22 - l21 dul21


which generalizes for other diagonal terms to

(Algorithm 3) (12.17)
For general terms neither in first column, nor on diagonal, one obtains:

or

This generalizes to

(Algorithm 4)

The decomposition must proceed in an orderly way, column by column from the
left in [I<], and from diagonal term to bottom. The algorithms apply to regions as
shown in Table 1.

TABLE 1. FORTRAN PROGRAM FOR DECOMPOSITON

SUBROUTINE DCOMP(N, K)
C THIS SUBROUTINE DECOMPOSES AN NxN SYMMETRIC,
C POSITIVE DEFINITE MATRIX [K] INTO THE
C PRODUCT [L1] [D] [L1]AT. THE MATRIX [L1] IS
C RETURNED TO THE MAIN PROGRAM IN THE LOWER LEFT
C PORTION OF [K] . THE MATRIX [D] IS RETURNED ON
CHAPTER 12. LINEAR STATIC EQUATION SOLVERS 143

C THE DIAGONAL OF [K] .


c
REAL K(100,100), SUM
INTEGER I, KK, K1, J, N
DO 50 J = 1, N
DO 50 I = J, N
SUM= K(I,J)
K1 = J - 1
IF (J .EQ. 1) GO TO 20
DO 10 KK = 1, K1
10 SUM= SUM- K(J,KK) * K(I,KK) * K(KK,KK)
20 IF (I .NE. J) GO TO 40
IF (SUM .LE. 0.0) WRITE (6,30)
30 FORMAT () DECOMPOSITION HAS FAILED. NEGATIVE TERM
1 ON MAIN DIAGONAL))
IF (SUM .LE. 0.0) RETURN
K(I,J) = SUM
GO TO 50
40 K(I,J) =SUM/ K(J,J)
50 CONTINUE
RETURN
END

The subroutine DCOMP(N ,K) decomposes a real, symmetric, positive definite nxn
matrix [K) into a diagonal matrix [D) and a lower unit triangular matrix [L 1 ):

The matrices [D) and [L 1 ) are stored on the diagonal and lower triangular part of
[K], respectively. The newly computed quantities dii and lij can be written on top
of the kij locations as they are calculated (1), thus avoiding the necessity of creating
another large matrix to store [D] and [L 1 ). The Fortran IV code for DCOMP is given
in Table 1. The number of Fortran statements required to decompose an arbitrarily
large matrix is only 17! 1

The forward solution will now be done. The recursive relation is

[LI)[D)[LI)T{u} = {P}

[L1]{X} = {P}
1
The programming style here is due largely to Weaver [1]. Many serious students have probably
independently found the trick of overlaying the new matrix terms on top of the old!
CHAPTER 12. LINEAR STATIC EQUATION SOLVERS 144

x1 p1

~~1
0 0 X2 p2

[
131
1
132
0
1
1 l X3

Xn
p3

Pn
(12.18)

X1 = P1,

x2 = P2 - 121x1,

X3 = p3 - 131 X1 - 132X2

I Xi = pi - I::~-:11 1ikXk I (i=1,2, ... ,n) (Algorithm 5)


Back substitution is

li1 X1
dn121 dn131 li2 X2

[ df d22
0
d22l32
d33
li3 X3
(12.19)
dnJ
lin Xn

dn-1,n-1 Un-1 + dn-1,n-11n,n-1 lin = Xn-1


Xn-1 - dn-1,n-11n,n-1 lin
Un-1 =
dn-1,n-1

The recursive form is:

(i=n,n-1, ... ,1) (Algorithm 6)

The subroutine SOLVE(N,K,P,U) solves the matrix equation

[K]{u} = {P} (12.20)

where the matrix [K] is real, symmetric and positive definite and has been decom-
posed by Gauss-Doolittle triangular factorization. The lower unit matrix [L 1 ] and the
diagonal matrix [D] are stored in the lower triangular and diagonal portions of [I<],
respectively. The Fortran IV code for SOLVE is given in Table 2.

TABLE 2. FORTRAN PROGRAM FOR SOLUTION


CHAPTER 12. LINEAR STATIC EQUATION SOLVERS 145

SUBROUTINE SOLVE(N, K, P, U)
C THIS SUBROUTINE SOLVES THE LINEAR ALGEBRAIC SET
C OF EQUATIONS [K]{U} = {P}
C WHERE [K] IS AN NxN, REAL, SYMMETRIC, POSITIVE
C DEFINITE MATRIX AND {U} AND {P} ARE VECTORS
C WITH N COMPONENTS. THIS SOLVER PRESUMES [K]
C HAS BEEN DECOMPOSED AND STORED AS IN "DECOMP".
c
REAL U(100), P(100), K(100,100), SUM
INTEGER KK, K1, K2, I, N
DO 20 I = 1, N
SUM = P(I)
K1 = I - 1
IF (I .EQ. 1) GO TO 20
DO 10 KK = 1, K1
10 SUM= SUM - K(I,KK) * U(KK)
20 U(I) = SUM
DO 40 I1 = 1, N
I = N - I1 + 1
SUM = U(I)
K2 = I + 1
IF (I .EQ. N) GO TO 40
DO 30 KK = K2, N
30 SUM= SUM- K(KK,I) * U(KK) * K(I,I)
40 U(I) = SUM/ K(I,I)
RETURN
END

12.9 HOMEWORK
Problem 1.
Given the equations:
2ul + u2 = 11
u1 = 2u 2 = 13
Carry out a Gauss-Doolittle triangular decomposition of the stiffness matrix. Find the
matrices in the form

but do not solve for the unknown displacements.

Problem 2.
CHAPTER 12. LINEAR STATIC EQUATION SOLVERS 146

A) Solve the following set of equations with the use of a Gauss-Doolittle decompo-
sition:

B) Solve the same equations by a Cholesky decomposition method.

Problem 3.

Solve the set of equations given in Problem 2 by using a Gauss-Seidel iteration.


How many iterations are needed to get 1% accuracy? The method converges slowly; a
computer program will be needed to get this convergence. Start with an initial guess:

{u}(o) = { J}

Problem 4

A) Carry out a formal solution of the equations:

2ul + 6u2 = 14
by the Cholesky decomposition. (Decompose the coefficient matrix into the product
of a lower triangular matrix and its transpose.) Show all of your steps. Describe the
procedure with proper terminology.

B) Attack the same set of equations with Gauss-Seidel iteration. Assume an initial
vector { ~ } and find the next vector iteration.

Problem 5.

The purpose of the decomposition of a real symmetric matrix before solution is to:
(a) reduce storage requirements
(b) allow iteration
(c) organize a sequential solution
(d) improve positive definiteness

Problem 6.

Decompose the following matrix using the Gauss- Doolittle method:

[k] = [ !]
~ ~
CHAPTER 12. LINEAR STATIC EQUATION SOLVERS 147

Problem 7.

Consider the following matrix [K] which has been proposed as a stiffness matrix in
a structural mechanics problem.

A)
[! ; lll
Is this a useful matrix for solution of the static equilibrium equations for the
problem? Discuss all details.

B) If the first coordinate, u 1 , is constrained to zero, decompose the relevant part of


the stiffness matrix by the LDLT decomposition method.

C) Use the Cholesky decomposition to decompose the same matrix. Discuss the
difference in the methods, and why the LDLT method is preferred in large structural
problems.

Problem 8.

Decompose the following matrix into triangular factors using the Gauss- Doolittle
decomposition.

Problem 9.

A set of 3 equations is given. The constraint u 1 = 0 is applied.


1) Show how to reduce the set of equations.
2) Use the modified Choleski (Gauss-Doolittle) decomposition to solve the reduced set
of equations. Follow the formal procedure (i. e., don't just use high-school elimination
of variables).
3) Recover the force of constraint.
Bibliography

[1] Weaver, William Jr., Unpublished lectures, Summer Conference on Finite Element
Analysis, Stanford University, Palo Alto, CA, 1969.

148
Chapter 13

SYMMETRY ARGUMENTS

13.1 GENERAL CONCEPT


Many physical problems have special points, lines or surfaces about which certain
repetitive p~operties occur. The analyst can exploit this symmetric nature by modeling
only one segment of the body, knowing that the repetitive nature will cause the same
results to hold over the entire body. This can save time and money, since computer
costs are often proportional to the cube of the size of the problem. Many problems can
be broken into segments ~ to ~ the size of the original body.

One special case of symmetry is axisymmetry, in which a 3-D problem collapses


into a 2-D space. The savings in modeling costs are very large. A xi symmetry depends
on geometry, materials, boundary conditions and loading. The same considerations
apply in other types of symmetry.
'

At present, practicing engineers often ignore symmetries of a body and its load,
and solve the entire system. This is a fail-safe approach, and graphical preprocessors
allow building of bodies with reflections and rotations, so the whole-body approach
is acceptable. If an experienced engineer is able to handle the boundary conditions
and/or understands how to ~se the cyclic symmetry capabilities in the NASTRAN
programs, the reduced problem should be solved.

13.2 REFLECTIVE PLANES


Discussions of reflective symmetry involve the independent variables (geometry,
b.c., materials, and loads) and the dependent variables (displacement, strain, stress
or temperature) . One defines symmetry of a variable with respect to the plane if the
variable takes on mirror image values on opposite sides of the plane.

Symmetry of a scalar variable with respect to a plane means that the points A and
B (Fig. 13.1) lying on the same normal to the plane and equidistant from it will have

149
CHAPTER 13. SYMMETRY ARGUMENTS 150

the same value of the variable. Antisymmetry of a scalar variable w.r.t. a plane means
that the points A and Bin Fig. 13.1 will have values of the variable which are opposite
in sign. The same plane can be a plane of symmetry for one variable and a plane of
antisymmetry for another variable.

A
/ B
-

/
Figure 13.1: Points lying on normal to a plane and equidistance from it.

Symmetry of vector and tensor variables w.r.t. a plane is a more complicated issue.
Displacement, force, strain, and stress variables require careful thought. For symmetry,
the variables must possess a zero derivative at the plane w.r.t. the normal distance
from the plane.

If a problem has a plane of symmetry for all independent and dependent variables,
the problem is said to be symmetric w.r.t. the plane.

If a problem has a plane of symmetry for geometry, materials and boundary con-
ditions, but loading is antisymmetric about the plane, then the problem is said to
be antisymmetric w.r.t. the plane. All dependent variables will turn out to be anti-
symmetric. This is analogous to the case where one negative number in a product of
numbers will cause the product to be negative.

A common situation in mechanics is where a plane of symmetry exists for geometry,


materials and boundary conditions, but the loading is asymmetric (neither symmetric
nor antisymmetric). One decomposes the load into symmetric and antisymmetric parts
and solves a symmetric problem and an antisymmetric problem. Each problem is half
the size of the original problem, because the cost of solution varies as the square or
cube of problem size, this is advantageous.

Modeling a segment of a symmetric structure requires a hypothetical cut at the


plane of symmetry. This causes two modeling difficulties. First, live loads on that
cut must be reduced (usually by half) to account for comparable loading on the image
segment, which is not modeled. Second the internal forces at the cut are brought out
into the open and must be considered as unknowns. Normally, in FEA, these internal
forces are suppressed.

I
CHAPTER 13. SYMMETRY ARGUMENTS 151

Proper b.c. must be applied at the cut surface. For structural members such as in
Fig. 13.2 and with a yz reflective plane, say, the b.c. at the cut plane of symmetry are
given in Table 1.

Table 13.1: Table 1. Symmetry constraints for structural problems.


SYMMETRIC B.C. ANTISYMMETRIC B.C.
CONSTRAINED ZERO CONSTRAINED ZERO
D.O.F. FORCES D.O.F. FORCES
1 (DX) 2 (FY) 2 (DY) 1 (FX)
5 (RY) 3 (FZ) 3 (DZ) 5 (MY)
6 (RZ) 4 (MX) 4 (RX) 6 (MZ)

Figure 13.2: Space frame with yz reflective plane.

An elasticity problem in 3-D with solid elements (Fig. 13.3) has boundary conditions
as shown in Table 2. As expected, these are subsets of the structural cases, above.

Table 13.2: Table 2. Symmetry constraints for elasticity problems.


SYMMETRIC B.C. ANTISYMMETRIC B.C.
CONSTRAINED ZERO CONSTRAINED ZERO
D.O.F. FORCES D.O.F. FORCES
1 (DX) 2 (FY) 2 (DY) 1 (FX)
3 (FZ) 3 (DZ)

Finally, it is not obvious how to decompose loads into symmetric and antisymmetric
components. Rather than ask whether the applied loads themselves are opposite in sign,
you must ask whether they tend to cause symmetric or antisymmetric displacements.
The equal and opposite loads in Fig. 13.4 tend to cause a symmetric displacement field
CHAPTER 13. SYMMETRY ARGUMENTS 152

Figure 13.3: Solid element model with yz reflective plane.

(a mirror image). For the direction normal to the plane, a symmetric displacement has
opposite signs on opposite sides. On the other hand, for the two directions parallel
to the plane, symmetric displacements (mirror image) have the same sign on opposite
sides. When one gets into trouble with such concepts, it can always be resolved by
considering what a mirror really does, or even getting out a mirror to demonstrate the
case!

The engineer can usually figure out the symmetric constraint list from basic logic.
For the antisymmetric case (Fig. 13.5), however, it is usually best to take the constraints
as the complement of the symmetric case.

Figure 13.4: Symmetric loading.

13.3 CYCLIC SYMMETRY


Many physical problems have an axis from which radiate surfaces which divide
the structure into segments which are identical (Fig. 13.6) . These bodies have cyclic
symmetry of a type where the boundary conditions are periodic. In the simplest case
where loading has the same periodicity as the geometric (so-called harmonic zero case),
the b. c. are repetitive. Figure 13.7 shows this case, where in cylindrical coordinates,
CHAPTER 13. SYMMETRY ARGUMENTS 153

Figure 13.5: Antisymmetric loading.

This relation is called a "multi-point constraint". Many engineers have solved har-
monic zero problems by the traditional method of isolating a segment and using many
multipoint constraints to relate displacements on the cut edges. One must also be
careful to remove rigid body modes, which is the Achilles tendon of cyclic symmetry
modeling.
For general aperiodic loading, the cyclic problem cannot be handled in the tradi-
tional way. One can use the N ASTRAN series programs for an automated version of
cyclic symmetry.[1],[2]

Figure 13.6: Body with cyclic symmetry.

/
{u} 1

23

Figure 13.7: Segment from body with harmonic zero loads.


CHAPTER 13. SYMMETRY ARGUMENTS 154

13.4 DIHEDRAL SYMMETRY


Many structures appear as in Fig. 13.8, where a series of reflective planes appear,
but in an alternating manner. As with cyclic symmetry, some problems have simple
loading which is periodic (including right and left reflections that are required) and can
be solved by the traditional method. This would involve isolating a segment 13.9 and
applying symmetric b.c. on each reflective plane. An example of a problem solvable in
this way is the uniform pressurization of all holes in the body.

For general (asymmetric) loading on a body with dihedral symmetry, one can resort
to the dihedral symmetry behavior in the N ASTRAN series of programs .

Figure 13.8: Dihedral symmetry.

Figure 13.9: Segment from body with dihedral symmetry.

13.5 HOMEWORK
P roblem 1
CHAPTER 13. SYMMETRY ARGUMENTS 155

An assembly has a plane of symmetry at y=O. A beam passes through this plane
(Fig. 13.10), with centroidal axis in the y direction. Using global coordinates, which
of the following degrees of freedom should be constrained at the plane of symmetry?

Figure 13.10: Beam with plane of symmetry at y=O.

(a) 2,4,6 (DY, RX, RZ)


(b) 1,3,5 (DX, DZ, RY)
(c) 2,5,6 (DY, RY, RZ)
(d) 1,2,3 (DX, DY, DZ)
(e) 3,5,6 (DZ, RY, RZ)
Problem 2
A two-dimensional problem involves a square cut-out in a round tube (Fig. 13.11).
The tube is to be internally pressurized and free on the outer surface. Using symmetry
arguments, what is the minimum fraction of the body you would have to model for a
solution?

tI

-+--m I
J

-,..-

Figure 13.11: Square cut-out in round tube.

Problem 3

A triangular sheet of metal has a hole punched as in Fig. 13.12, centered on the
centroid of the triangle. The sheet is put in compression as shown.

A) What is the smallest portion of the problem that can be solved, using symmetry?
B) If you solved one half of the problem, using symmetry, would you have to constrain
CHAPTER 13. SYMMETRY ARGUMENTS 156

Figure 13.12: Equilateral triangle sheet with centered hole.

any rigid body modes, in addition to confining the motion to be planar and to applying
symmetry conditions?

(1) no
(2) yes, constrain one rigid body mode
(3) yes, constrain two rigid body modes
(4) yes, constrain three rigid body modes
(5) one can't tell- it depends on which way the body is divided

C) If the body is clamped at the lower boundary and subjected to gravity in the
vertical direction, what is the smallest portion that can be solved, using symmetry?

Problem 4
Two hexagonal bars have been pierced by triangular holes (Fig. 13.13) and are
loaded on three sides by uniform distributed stresses. The polygons are regular and
are concentrically located such that their centroids coincide. Each represents a two-
dimensional, plane strain problem, with the only difference being the orientation of the
triangle.
(a) Which one( s), if any, of the problems can be solved by modeling a smaller
portion of the problem using symmetry? Sketch.
(b) If either or both can be solved using a smaller portion, give the type of boundary
conditions to be imposed on the cut edges to preserve symmetry.

Problem 5
A rectangular, thin elastic plate is suspended from its center by a rigid rod (Fig. 13.14).
The physical problem involved is to find the deflection of the plate due to gravity. It is
desired here, however, to ask only questions about boundary conditions and symmetry.

Suppose the plate is modeled by 4 rectangular finite elements as shown, and by sym-
metry, only 1/4 of the problem (one element) need be solved. Provide the conditions
CHAPTER 13. SYMMETRY ARGUMENTS 157

Figure 13.13: Pierced Bars.

Figure 13.14: Plate supported by rigid bar.


CHAPTER 13. SYMMETRY ARGUMENTS 158

to be placed on the following variables due to symmetry and boundary conditions.


Reduce the number of constraints to the minimum number needed to run a typical
finite element program.

Node 1: ui, u2, u3, B1, B2, ()3, PI, P2, P3, MI, M2, M3
Node 2: ui, u2, u3, B1, B2, ()3, PI, P2, P3, MI, M2, M3

Problem 6.
A beam-like bracket extends through a wall and is loaded symmetrically with re-
spect to the wall (Fig. 13.15). The geometry, material and displacement conditions are
symmetric with respect to the wall. Using the coordinate system shown, what variables
do you expect to take zero values at the wall? (Mention all displacement and force-like
components.)

..
z

v
Figure 13.15: Symmetric beam

Problem 7.
A hypothetical, two-dimensional, linear, elastic finite element has four nodes (Fig. 13.16).
It has a reflective plane in geometry and material, which passes through the x axis. If
nodes 1,2 and 4 are held stationary, and if node 3 is loaded with 200 Nat 45 as shown,
then the resulting horizontal displacement is 1 mm and the vertical displacement is 2
mm.

Find numerical values for as many stiffness terms kij as possible.

Problem 8.

A linear, elastic, plane-stress finite element has 4 nodes as shown in Fig. 13.17.
There are two reflective planes with respect to geometry and materials, passing through
the x and y axes.

A) Determine which terms in the stiffness matrix are zero.


CHAPTER 13. SYMMETRY ARGUMENTS 159

Figure 13.16: Quadrilateral with one reflective plane of symmetry

1 3

4
Figure 13.17: Quadrilateral element with two planes of reflective symmetry.

B) Consider the following mental experiment for the element. The nodes 1,2 and
4 are fully constrained. A load of 1000 N is placed on node 3 at 30 (Fig. 13.18),
and causes a horizontal displacement of 2 mm and a vertical displacement of 1 mm at
that node. Determine as many nonzero element stiffness terms as possible, from this
additional information.
C) What is the stored strain energy in the element under the loads and boundary
conditions in part b, above?

D) A total load of 1000 N in the x direction is applied in a triangular distribution on


one edge of the element as shown in Fig. 13.19. What are the equivalent nodal loads
that can replace the distributed load? Explain your logic.
Problem 9.

Consider the bracket formed from two beams, lying in the xy plane (Fig. 13.20).
The load is vertical. It is desired to use traditional reflective symmetry to allow solution
of a smaller problem. Explain the use of reflective symmetry, including:
1) proper boundary conditions at the reflective plane
2) how to handle the load (symmetric, antisymmetric components)
CHAPTER 13. SYMMETRY ARGUMENTS 160

Figure 13.18: Quadrilateral with load at node 3.

Figure 13.19: Load with triangular distribution.

3) how to recover results in the half of the problem which is not modeled.
CHAPTER 13. SYMMETRY ARGUMENTS 161

Figure 13.20: Bracket with vertical load.


Bibliography

[1] Anderson, W. J., "Cyclic Symmetry," Video Study Guide 1004, The MacNeal-
Schwendler Corp., Los Angeles, CA, 1988.

[2] MSC/NASTRAN Applications Manual, The MacNeal-Schwendler Corp., Los An-


geles, CA, 1991.

162
Chapter 14

REMOVAL OF RIGID BODY


MODES

14.1 DEFINITION OF RIGID-BODY MODES


Many elastic bodies can translate freely or rotate freely because there are not enough
constraints applied to fix the body to the earth. A typical desk telephone (Fig. 14.1)
is free to translate in two directions and to rotate about a vertical axis. Each such

Figure 14.1: Desk telephone .

distinct type of motion is called a rigid body mode . For linear elastic bodies:

A rigid body mode is a nonzero displacement field {u} which causes no


strain energy:
1
- {u} [K] {u} = o ( 14.1)
2 '-..-" '-..-"
!-+II !-+II

The opposite of a rigid body mode is an "elastic mode," i.e., a displacement field which
causes strain energy.
An equivalent definition for a rigid body mode, using an equilibrium argument, is :

163
CHAPTER 14. REMOVAL OF RIGID BODY MODES 164

A rigid body mode is a nonzero displacement field {ii} which causes no


elast ic forces:
[K]{ii} = 0 (14.2)
For such an equation to have a nonzero solution {u} one must have

det[I<] = 0 (14.3)

This is one of the four alternatives for linear algebraic equations .

Rigid body modes are legitimate in many analyses, particularly dynamic problems
where the overall motion of a body is important . Their presence causes trouble in static
stress analysis, however, because they prevent the solution of the equ ilibrium
equations :
[I<]{ u} = {P} ( 14.4)
The reason for failure is that det[K] = 0 and [I<] cannot be triangularly factorized .
One says that [K] is "singular".

The requirement for removal of rigid body modes is a nuisance in many stress
problems which could reasonably be stated as if the body were suspended in mid-air.
For instance, stress in a tire results primarily from the internal pressure and vehicle
weight and does not depend on which direction the car is traveling . Nevertheless, for a
unique solution, one must specify the orientation of the tire in space so as to not allow
rigid body modes .

In the theory of elasticity, there are:

6 rigid body modes in 3-dimensional space;


3 rigid body modes in 2-dimensional space;
1 rigid body mode in !-dimensional space.

General purpose programs such as MSC /N ASTRAN, MARC, ANSYS and ABAQUS
assume that the space is 3-D and assign 6 d .o.f. to each node unless the user states
differently. As a result, bodies have six rigid body modes in these programs .

Axisymmetric problems are a special case. In some codes, the elements are "rings"
that have only one rigid body mode, translation along the axis . Also, structural as-
semblies made of beams and plates can be constrained to lie in two dimensions. One
must be careful to decide how many modes have been removed.

14.2 HOW TO REMOVE RIGID BODY MODES


Most physical problems have symmetries or displacement boundary conditions that
fortuitously remove some rigid body modes. The remaining rigid body modes must be
removed by the analyst.
CHAPTER 14. REMOVAL OF RIGID BODY MODES 165

In finite element solutions, the user must artificially constrain degrees of freedom
to prevent rigid body translations and rotations . Every user must understand the logic
required to remove these modes. Generally, one constrains enough displacements to
prevent rigid body translation or rotation but not enough to constrain elastic defor-
mation between any nodes. Application of an artificial constraint at a node should not
introduce external force at the relevant degree of freedom. It is wise to check the forces
of constraint at the end of an FE solution to ensure that those forces are zero.

Every three-dimensional body should have at least six displacement components


specified (typically as zero) to remove six rigid body modes. Likewise, bodies con-
strained to move in two dimensions still require specification of three displacements .
Bodies constrained to move in one dimension require specification of one displacement.
One rigid body mode is removed for each properly constrained displacement .

14.3 EXAMPLE
Let us look at the stress around the circular hole in a sheet under uniform tension
(Fig. 14.2) . Note that the problem statement says nothing about orientation of the
sheet because orientation has no effect on stresses . The engineer, however, must ensure
that the system does not rotate or translate.

...__ ____.
...__ ____.
...__
...__
...__
...__
...__
...__
...__
&-x ____.
____.
____.
____.
____.
____.
____.

Figure 14.2: Rectangular sheet with circular hole under tension. Two-dimensions.

Suppose the problem were to be solved with the crude grid in Fig. 14.3. Assume
that body has been constrained to move in the xy plane. (This means that there are
constraints on the 3,4,5,6 degrees of freedom.) One could constrain the three in-plane
rigid body modes by setting to zero:

(a) horizontal displacement at node 4;


(b) vertical displacement at node 4;
(c) vertical displacement at node 8.

These constraints do not affect the elastic modes, whereas the following ones do and
are therefore unacceptable :
CHAPTER 14. REMOVAL OF RIGID BODY M ODES 166

Figure 14.3: Crude FE mesh for 2-D problem.

(a) horizontal displacement at node 4;


(b) vertical displacement at node 4;
(c) horizontal displacement at node 8.

The latter constraints incorrectly prevent any cumulative elastic strain in the x direc-
tion between nodes 4 and 8. Also, they do not prevent small rotations of the body
about node point 4 and are, therefore, a complete failure.

14.4 EFFECT OF RIGID BODY REMOVAL ON


DEFORMED PLOTS
Rigid body modes must be eliminated as discussed above in order to obtain static
stresses and displacements. There usually is some arbitrariness in constraining the
modes. An important secondary consideration is how the remaining elastic motion will
appear in plots of the deformed body. A careful choice of rigid body constraints will lead
to the most attractive and easily understood plots. This often involves constraining
motion so that the center of interest (center of a hole, high stress region, etc.) does
not move during loading.

The suggested nodal constraints for the body in Fig. 14.3 would cause superposed
undeformed and deformed plots as in Fig. 14.4. A better system of constraints would
involve zero values of

a) horizontal displacement at node 6


b) horizontal displacement at node 12
c) vertical displacement at node 5.

This would lead to the plot in Fig. 14.5 which shows the hole deformation more clearly.
CHAPTER 14. REMOVAL OF RIGID BODY M ODES 167

r- - - - - - - - - - - ------,
I
I

'0-,
I J

"- I
I

Figure 14.4: Deformed and undeformed body with left boundary constrained

r- - - - - - - - - - - - - - - - - - ,
I I
I I
I I

:
I
0 I
I I

Figure 14.5: Deformed body centered on undeformed hole.

14.5 EFFECT OF SYMMETRY ARGUMENTS


ON RIGID BODY MODES
If one recognizes two planes of symmetry in the example above, one can solve the
smaller problem in Fig. 14.6. In this case, symmetry requires constraints on:

Figure 14.6: Fortuitous removal of rigid body modes by symmetry constraints .

(a) horizontal displacement at node 10;


(b) horizontal displacements at node 12;
(c) vertical displacement at node 7;
(d) vertical displacement at node 8.

These symmetry conditions remove all rigid body modes, and further artificial con-
CHAPTER 14. REMOVAL OF RIGID BODY M ODES 168

straints are not needed. Symmetry also implies that the vertical forces on 10 and 12
are zero, and the horizontal force on 7 is zero. Node 8 is unique because it is a loaded
node on a plane of symmetry. The internal horizontal load on 8 due to the neighboring
mirror image is zero by symmetry, so the only the live load acts on 8. Finite element
programs assume loads are zero unless otherwise stated, therefore the zero shear force
condition on a cut reflective plane is obtained "for free."

14.6 MECHANISMS
In addition to the normal rigid body modes (six in number, at most), "mechanisms"
often exist in structures. These are rigid body modes of components such as linkages
or rotating shafts which are indeed free to move relative to the basic structure. A
specific example is the receiver on the desk phone (Fig. 14.1). Even if one clamps the
phone to the desk, one can still lift the receiver, which itself has six rigid body modes,
or mechanisms, relative to the phone body.

Mechanisms can be real, as in the case of the receiver above or rotation of a phono-
graph turntable. One must constrain such motion when doing a static stress study,
such as finding the force between the phonograph needle and the record. Mechanisms
can also be artificially introduced by joining plates and solids in a way so as to intro-
duce an unwanted hinge at what should be a weld line, or by connecting components
to the "drilling" d .o.f. in plates (see Chapter 17 for disucussion) .

14.7 NODAL SINGULARITIES


Nodes in finite element problems have from one to six degrees of freedom. Each
such degree of freedom will be singular (indeterminate) if there is no element connected
to it to provide elastic stiffness . One must carefully construct a finite element model
to prevent such nodal singularities. Some of the occurences are obvious, such as when
a space truss is created and the engineer does not restrain the rotational degrees of
freedom . Others are not so obvious, such as when a fiat plate is inclined at an angle,
and the direction normal to the plate has a singularity in the rotation vector (the
"drilling" degree of freedom).

Some finite element codes can automatically eliminate grid point singularities . The
search for the weakest direction in translation or rotation is an eigenvalue problem.
One can pose two such 3x3 eigenvalue problems (one each in translation and rotation)
at each grid, determine the singular direction and then "ground" it . If live loads are
attached to that direction, one should be sure that there is not a physical modeling
error to correct, not merely a chance mathematical singularity.

There is a current effort by software houses to remove the classic drilling singularity
CHAPTER 14. REMOVAL OF RIGID BODY MODES 169

in plate elements . The new plate elements will have to overcome user resistance to gain
acceptance, however.

14.8 HOMEWORK
P roblem 1

A plane stress problem involves a disk which is loaded by edge pressures as shown
in Fig. 14.7. It is desired to solve half of the problem as shown in Fig. 14.8.

Figure 14.7: Disk.

2 X

Figure 14.8: Half disk.

Using standard notation:

A) What degrees of freedom must be constrained over the whole problem to reduce
it to a two-dimensional space?

B) What additional physical conditions on force and displacement are valid on the
plane of symmetry?
CHAPTER 14. REMOVAL OF RIGID BODY MODES 170

C) What additional constraints are needed to prevent rigid body motion, after all of
the above conditions are met?

Problem 2
For a plane stress finite element constrained to the xy plane, there are how many
rigid body motions possible?

Problem 3
A solid, tetrahedron element (Fig. 14.9) is to be generated with displacement func-
tions. How many rigid-body modes does its have? How many straining modes does it
have?

z~
X

Figure 14.9: Tetrahedron.

Problem 4
A plane stress problem has been posed where all stress boundary conditions have
been given (Fig. 14.10). A twelve-element model has been proposed as shown, where
only degrees of freedom u 1 and u 2 have been constrained.

u
32
u
31

u ~ u
2 10

u
1

Figure 14.10: Plane stress problem.

A) How many additional degrees of freedom must be constrained to properly remove


rigid body modes?
CHAPTER 14. REMOVAL OF RIGID BODY MODES 171

B) Name all of the candidate degrees of freedom shown (9, 10, 31 and 32) that are
acceptable choices to remove the remaining degree( s) of freedom.

Problem 5
How many rigid body modes does an axisymmetric element have?
Problem 6.
A rigid bar element has degrees of freedom 1-6 at end A and 1-6 at end B. This
give a total of 12 d.o.f. Suppose a user wishes to use a rigid bar model such as the
RBAR model in MSC /N ASTRAN. The user specifies independent degrees of freedom
1,2 and 5 at end A (creating an "n" set there) and wants to use independent d.o.f. 1,2
and 6 at end B (creating more "n" set terms there). Will this cause trouble or is it
okay to have these variables specifying the independent degrees of freedom?
Problem 7.
Solve for the elastic deflections of the horizontal beam in Fig. 14.11 using finite
element concepts. The two ends of the beam are "guided," i. e., they are allowed no
rotation, but translate freely. The vertical flat plates that form the guide as well as
the rollers are considered made of rigid material.. Neglect the forces of gravity.

EI/L 3 = 10,000 N/mm

L = 1000 mm

"" 4000N /
/
'\P
\..h
'\lc
! P/
\P
'\ t 2000 N

~L -1....---L -1
Figure 14.11: Beam suspended between two guided supports.

Problem 8.

Consider the "teeter-totter" in Fig. 14.12. Two children, weighing 150 N and 250
N, are located on the device as shown. Show how you would make a finite element
CHAPTER 14. REMOVAL OF RIGID BODY MODES 172

model to find the deflection of the beam and the stress at the hinge. Describe in detail
the boundary conditions needed. The solution is to be exact.

250N
150N

Figure 14.12: Teeter-totter with two children.


Chapter 15

MODIFICATION OF
EQUILIBRIUM EQUATIONS

15.1 PROBLEM STATEMENT


Assuming that a finite element problem has been properly posed and the element
equilibrium equations have been successfully assembled, one has a set of equations
[K]{u} = {P} (15.1)
where [K] is the assembled stiffness matrix and { P} is the assembled vector of external
loads, plus equivalent nodal loads.

Mathematically, 3 possibilities exist:


1. All components of {u} are known and { P} is to be found.

2. All components of {P} are known and { u} is to be found.

3. A mixture of components of {u} and {P} are known, and the remainder of each
are to be found.
The first case is easy to solve because { P} can be found by direct multiplication.
This case never occurs in practice because displacements are not known at internal
nodes.

The second case is the desired form for standard equation solvers. It, too, rarely
occurs in practice because specification of some displacements usually occurs on the
boundary. There are a few problems, however, in which all boundary conditions are
on forces.

The third case, a "mixed" boundary value problem, is the usual finite element
situation. Typically 10% of the displacements and 90% of the forces are specified. The
dominance of force specification is because forces are typically zero at internal nodes. 1
1
Students sometimes do not notice that zero is a legitimate "known" number in thi~ situation .

173
CHAPTER 15. M ODIFICATION OF EQUILIBRIUM EQUATIONS 174

Our approach will be to convert the third case, which occurs in nature, into the
second case, which can be solved mathematically.

15.2 PARTITIONING
Partitioning is a method of reducing the number of equations studied to those where
only displacements are unknown. It proceeds more easily when there is a fortuitious
or pre-planned numbering of degrees of freedom so that all specified displacements are
left to the last (say):

(15 .2)

(The unknown quantities are in boldface type) . The equations are then partitioned,
where each submatrix follows the normal laws for matrix operations:

[I<n){ u l} + [I<12]{u2} = {P1} (15.3)

[I<21]{u1} + [I<22]{u2} = {P2 } (15.4)


Equation 15.3 can be solved for the unknown displacements { u 1}:

( 15.5)

Often the reactions at the supports (degrees of freedom where displacement is spec-
ified) are desired . The final step is then to use Eqn. 15.4 to directly solve for {P2 },
since the L.H.S . is known .

15.3 "PENALTY" METHOD APPROACH


One can solve the equations for a mixed boundary value problem, without regard
to the order in which they appear, by an artifice due to Payne & Irons. The set of
equations is modified by converting the known displacements into unknowns, and then
forcing them to the desired value by choosing a very large value on the diagonal.

ORIGINAL PROBLEM

[
~:: ~:~ ~::
]{31 ]{32 ]{33
] { :: } = {
2.7
~~p~3~: } (15.6)
CHAPTER 15. M ODIFICATION OF EQUILIBRIUM EQUATIONS 175

MODIFIED PROBLEM

[ ~:: ~:~
]{31 ]{32 12
10
~::X ]{33
]{ ~~ } = { 10 ~~~~:
U3 12
2. 7X X ]{33
} (15. 7)

The modified equations are then solved in the usual way by versions of Gaussian
elimination or by iteration. To see the effect of the large constant 10 12 , separate out
the last equation from Eqn. 15.7:

(15.8)

Since stiffness matrices are diagonally dominant, i.e., K 33 is of at least the same size
as ]{31 and ]{32,
(15.9)
The value of u 3 can be made as close to 2. 7 as wanted by increasing the constant 10 12
even further.

In solving by this approach, one has inserted some dummy equations in the process
and therefore must lose (at least temporarily) some information from the original set
of equations . The information lost is the reaction force at each node, which can be
recovered by a final step using direct summation, e.g.
3
P3 = L 1<3juj ( Uj known from previous solution) (15.10)
j=1

15.4 SOLVING IN PLACE


An alternate method of solving in place is to extensively modify the stiffness matrix
as shown, using the same original matrix in the last section:

(15.11)

This method is exact . It involves more manipulation than the previous method, how-
ever. It is practical, and has been used in many smaller F .E . programs.

15.5 STATIC CONDENSATION


Every finite element solution must use a modification procedure similar to those
in the three previous sections to put the equations in the standard form for solution.
There is a further way to eliminate a portion of the problem and reduce the number
of equations. This is called static condensation. It really is Gaussian elimination in a
CHAPTER 15. MODIFICATION OF EQUILIBRIUM EQUATIONS 176

preferred order. The method described here is based on partitioning, which allows a
simple description.

Suppose one has a set of equations in which one is not interested in several degrees
of freedom, and is willing to suppress them, or "condense them out." Let {u 2 } represent
the undesired degrees of freedom, which have been fortuitously numbered as the last
degrees of freedom. All forces are assumed known.

(15.12)

undesired d.o.f. /

Partitioning the equations gives

[I<n]{ul} + [J<12]{u2} ={PI} (15.13)

[J<21]{u1} + [J<22]{u2} = {P2} (15.14)


One solves for { u 2 } from Eq. 15.14

(15.15)

and inserts it in Eq. 15.13, thereby eliminating { u 2 }:

(15.16)

Collecting terms, one can define new stiffnesses and forces:

(15.17)
[K] {P}

This defines a reduced problem involving fewer degrees of freedom, but more compli-
cated stiffness and force system:

[K]{ u1} = {P} (15.18)

Static condensation is not recommended for general use in static problems, because
it merely rearranges the order of solution in the Gauss elimination process. It is
important, however, for removal of internal nodes in elements and in cyclic symmetry
problems. It is very important in dynamics and will be reconsidered later under the
name Guyan reduction.
CHAPTER 15. MODIFICATION OF EQUILIBRIUM EQUATIONS 177

15.6 SUBSTRUCTURING
This is a method of dividing large structures into substructures for easier solution.
It is philosophically similar to static condensation, but involves more assumptions
about the partitioning of the body, boundary conditions, etc. We will not discuss the
methodology.

15.7 COMMENTS
1. Partitioning is used extensively in the N ASTRAN programs, for reducing the
number of d.o.f. It is not used as much in other codes.

2. The partitioning method for static condensation, substructuring, Gauss- Dolittle


decomposition, Cholesky decomposition and high school elimination methods are
all just variations on Gauss elimination using a preferred order for eliminating
variables .

15.8 HOMEWORK
P roblem 1

A rectangular sheet of metal has a square hole cut from its center as shown in
Fig. 15.1. The sheet is clamped at its left and upper edges and is free on its right and
lower edges. A 100 N force is applied downward at its lower right corner as shown.
The goal is to solve the two-dimensional stress problem without considering bending
or buckling effects .

l OO N l OON

Figure 15.1: Sheet metal under point loads.

Write down a physical description of all boundary conditions that need to be ex-
plicitly entered into a general purpose program.
CHAPTER 15. M ODIFICATION OF EQUILIBRIUM EQUATIONS 178

A hypothetical example is given in tabular form:


Node Displacement Force
27 1,4,6 Fz = 250N
(DX,RX,RZ)
38 Fy =-lOON
Problem 2
A) If the set of equations shown is to be solved by first modifying them with the
Payne and Irons method for solving "in place", how many equations must be solved
simultaneously after the modification?

B) If the set of equations shown is to be solved by partitioning, how many equations


must be solved simultaneously after the partitioning? What is the largest matrix that
must be inverted for a solution by partitioning?

1 3 7 9 2 ul 20
3 10 5 8 4 u2 30
7 5 40 2 4 10 p3 (15 .19)
9 8 2 100 20 12 p4
2 4 4 20 400 Us 15

Problem 3
Three identical line elements are joined as shown in Fig. 15.2. The left node u 1 is
constrained and the third node u 3 is moved 1 mm (.039 in) to the right . Set up the
equations of equilibrium for the 4 x 4 matrix problem, using the artifice of Payne and
Irons to modify the equations for solution "in place." Let k 1 = E 1 A 1 / L 1 , etc.

u= O
1
p =0 p =0
____. -l... 4__.
o) o)

Figure 15.2: Identical line elements under point loads.

(15 .20)

Express the equations in the simplest modified form, without solving for any of the
unknowns.
CHAPTER 15. MODIFICATION OF EQUILIBRIUM EQUATIONS 179

Problem 4
Solve the preceding problem by using partitioning. Take E 1 A1 / L 1
(i=1,2,3).
Problem 5
Do usual finite element problems involve a higher number of specified displacements
or of specified forces?

Problem 6.
The following set of equations is to be solved.

A) Show a mathematical modification to the equations that will allow solution by


the penalty method.

B) Sketch the physical analogy of what you have done mathematically in part A,
above. Use an assembly of the three line elements, showing the live loads as a reference.

u2 u3 u4
_ ____j-t-__..
~~----=:;-____:::rfL__.. _ ____.+ __..

Figure 15.3: Assembly of line elements to be solved "in place"

Problem 7.
A membrane has plan dimensions of 400 mm by 600 mm (Fig. 15.4). It will be
loaded only in the x,y plane. The membrane is modeled by four triangular, three-
noded elements. A rod (truss) with stiffness EA/1=10 4 N /mm is subsequently added
as a stiffener between nodes 1 and 3 as shown in Fig. 15.5. What size is the assembled
stiffness matrix? Show the explicit terms that the stiffening rod will contribute to the
assembled global stiffness matrix. Use a single, underlying coordinate system for the
whole problem.
CHAPTER 15. MODIFICATION OF EQUILIBRIUM EQUATIONS 180

1 2

Figure 15.4: Finite element model of membrane, before stiffening

Figure 15.5: Stiffened membrane


Chapter 16

BANDWIDTH CONCEPTS

16.1 INTRODUCTION
The computational difficulty in solving large sets of linear, algebraic equations de-
pends on the way the equations are coupled. This is seen by how fully populated the
matrix of coefficients is . Three concepts are important: "bandwidth," "wavefront," and
"pointer methods". Equation solvers usually require that one or two of these be consid-
ered in order to reduce storage costs and computer central processor unit (CPU) time .
Equation solvers of the banded or wavefront types require careful ordering of nodes
or elements, respectively. Equation solvers using pointers use sparse matrix methods
that only store nonzero stiffness terms. Fortunately, most large general purpose F .E.
codes and many preprocessors can resequence the node or element numbering so as to
minimize storage and CPU expense.

16.2 BANDWIDTH
Bandwidth refers to the width of the diagonal band made up of n nonzero terms
in a banded, sparse matrix (Fig. 16.1 ). Typically, by numbering the nodes properly,
the F .E . method leads to a stiffness matrix which has nonzero terms near the main
diagonal and zero elsewhere. The bandwidth is a dimensionless number indicating how
1nany terms wide this band is; no nonzero terms may fall outside the band. Bandwidth
can be given in terms of compact or detailed notation. The semi-bandwidth, B, is more
important in symmetric matrices and is the technical concept used exclusively in this
book:
SEMI-BANDWIDTH= ~(BANDWIDTH+ 1)
The semi-bandwidth includes the main diagonal term . In fact, for most structural
problems, authors will use the term bandwidth when semi-bandwidth is actually meant.
This seems to cause little trouble in practice. To expose the reader to the real world,
the terms will be used interchangeably in this text, but always refer to semi-bandwidth.

181
CHAPTER 16. BANDWIDTH CONCEPTS 182

....- -
"'~000
"'( 0 0 0

~ ~th ~ ~0
0

b an d w1"dt h , B
spm1-
0 ' /
0 0 0 0 "":"...__._:-~'"!l~t-"1
0 0 0 0 0
'- -

Figure 16.1: Bandwidth and semi-bandwidth.

Consider a physical problem with nodes numbered in two different ways (Figs. 16.2,
16.3). In each case, it can be shown that

B=(D+1)f (16.1)

1 2 3 4

sl
61 71 sl
-~X 0 X 0 X ~-
X 0
X 0 XX X X,o
0 X X X 0 X X X'\
0 X 0 X 0 0 X X ~
, X X X 0 X X 0 0
~XX 0 X X X 0
0 vX X X 0 X X X
~0 ~ X 0 0 X X
-

Figure 16.2: First nodal numbering scheme. (Compact notation.)

where D is the greatest difference in node number within an element, and f is the
number of D.O.F. per node. In the first case,

B = (5+ 1)(2) = 12

In the second case


B = (3 + 1)(2) = 8
Hence, the bandwidth depends on the way the nodes are numbered.

For Gaussian elimination types of solution, the CPU time required is proportional
to the size of the matrix times the semi-bandwidth squared:

(CPU TIME) ex (N)(B) 2


CHAPTER 16. BANDWIDTH CONCEPTS 183

1 3 5 7

21 41 61 sl

Figure 16.3: Second nodal numbering scheme. (Compact notation.)

In the example discussed, the running time for the second case would be

and the CPU time is cut by more than half by renumbering the nodes!

MSC /N ASTRAN, SAP6, and most public domain finite element codes are depen-
dent on bandwidth ideas. The numbering scheme for elem e nts is unimportant for
bandwidth ideas .

16.3 WAVEFRONT
Another type of equation solver, also based on Gaussian elimination, is the "wave-
front" approach. This was originally developed independently in England by Bruce
Irons(l) and in the United States by Robert Melosh. (2 ) In this approach, the element
numbering is important, rather than the node numbering . The goal is to hold as few
equations in high speed storage as possible, putting the remaining terms on disk or
other lower cost storage. The solution proceeds elem e nt by e lement . The equations
of equilibrium for a node are activated when an element is first formed which contains
that node. The equations of equilibrium for that node remain active until the last
element containing the node is processed. That node is then deactivated. The solution
is a Gaussian elimination in a preferred order.

Because storage occupancy on many computers (particularly virtual storage ma-


chines) varies with time, the frontal method can exploit that time dependence. A
measure of occupancy is the root mean square value of wavefront.
CHAPTER 16. BANDWIDTH CONCEPTS 184

Element Active Node No . of Active Nodes Nodes Dropped


1 1,2,3,8 4 1,8
2 2,3,4,5,6,7 6 4,6
3 2,3 ,5,7 4 2,3 ,5,7

Table 16.1: Wavefront for first numbering scheme.

An example is now given where the same problem is solved with two different
element numbering schemes (Figs. 16.4, 16.5) .

1 3 7 4

I
CD I
0 I
0 I
8 2 5 6

Figure 16.4: First element numbering scheme for wavefront solver

1 3 7 4

8 2 5 6

Figure 16.5: Second element numbering scheme for wavefront solver

The first numbering scheme has a wavefront with an rms value of 4. 76. The second
scheme has a smaller wavefront with an rms value of 4.0. Note that the numbering
scheme of the nodes is unimportant for wavefront ideas.

16.4 "POINTER" METHODS FOR SPARSE MA-


TRICES
Some computer programs store large matrices by interspersing integers (pointers to
row and column information) among the real numbers that make up the matrix. The
idea is to avoid storage of zero real numbers .

An example is for column-wise storage of a matrix, where the first several entries
in the data set are integers that indicate the type of matrix , overall size, etc. One
CHAPTER 16. BANDWIDTH CONCEPTS 185

Element Active Node No. of Active Nodes Nodes Dropped


1 1,2,3,8 4 1,8
2 2,3,5,7 4 2,3
3 4,5,6,7 4 4,5,6,7

Table 16.2: Wavefront for second numbering scheme.

Figure 16.6: "Skyline" storage.

can then follow by giving the column number of interest, the first row in which a
nonzero real number is encountered, and then the string of real numbers that represent
the nonzero terms . The occurence of the next integer will indicate a skip to the row
number indicated and is followed by a another string of real, nonzero numbers . In this
way, by sacrificing some storage space for pointers, one can avoid storage of any zero
numbers. This is particularly useful in dealing with banded stiffness matrices .

16.5 DIFFICULTIES WITH CPU TIME FOR SOLID


ELEMENTS
The run time for 3-D problems can be prohibitive if care is not taken. An example
which is rather scary is proposed by Zienkiewicz( 3 ) where comparable sized 1-D, 2-
D, and 3-D problems are discussed. Suppose each problem has 20 elements in each
direction as shown. It is known that CPU time T for the equation solver varies as:

T ex (Number of Equations)(Semi-Bandwidth?

The number of degrees of freedom per node increases from 1-D to 3-D and the
CHAPTER 16. BANDWIDTH CONCEPTS 186

1-D
3-D
~
1/
2
~~
1 1111111111 -D

Figure 16.7: Comparison of 1-,2-,3-dimensional space.

number of nodes and the bandwidth also increase strongly. Indeed:


T 1 _v const(21)(2) 2 = const(84)
T 2 _v const(882)( 46) 2 = const(1,866,312)
T3-D const(27, 783)(1392) 2
const(5.83 x 10 10 )
Even comparing the 2-D and the 3-D solution, we see that the 3-D solution requires
24,600 times as much computer time as the 2-D . This is a fantastic increase in com-
plexity as dimensions increase. The situation is not hopeless, however. Effort in the
direction of using fewer but more sophisticated elements is the proper approach. Ele-
ments with higher degree polynomial shape functions, as in p-convergent codes, seem
to be a part of the answer .

16.6 COMMENTS
1. Minimizing bandwidth (or wavefront) saves both on floating point arithmetic and
CPU time and on high speed storage requirements.

2. High-speed storage time typically represents 2/3 of the total cost of running F.E.
programs. At present, then, one should attempt to reduce high-speed storage as
the highest priority.

16.7 HOMEWORK
Problem 1
A two-dimensional problem is given in Fig. 16.8. What is the lowest bandwidth
you can achieve by optimal nodal numbering? Use detailed notation, i.e., give the half-
bandwidth in actual number of terms from the main diagonal in the stiffness matrix.

P roblem 2
An equation solver which uses Gauss- Doolittle decomposition is used to solve two
different problems as follows:
CHAPTER 16. BANDWIDTH CONCEPTS 187

Figure 16.8: Two-dimensional problem.

a) 150 nodes, 3-D space frame (beam elements) B = 75 (detailed notation)


b) 400 nodes, 2- D, plane stress B = 80 (detailed notation)
Compare the CPU time needed for solving Problems A and B.
Problem 3
Two problems as shown in Fig. 16.9 are to be solved by finite element methods. A
Gauss elimination equation solver is to be used. Discuss which problem will take more
CPU time for the equation solver.

A) Three-dimensional brick B) Two-dimensional quadrilateral


ellements. Eight nodes per brick. elements. Four nodes per element.

Figure 16.9: Two candidate FE problems

Problem 4

A plane-strain problem is modeled as shown in Fig. 16.10. What is the half band-
width of the given system, in detailed notation? How many zero terms will there be in
the stiffness matrix, in detailed notation, for the entire matrix?
Problem 5
CHAPTER 16. BANDWIDTH CONCEPTS 188

11
6
---------- I 0
3

2 e----~..,__ _ _..... 9
5
1

Figure 16.10: Plane strain problem.

A plane stress problem has the mesh shown in Fig. 16.11.

3 7
6
0 G)
2 8
5
CD 0
1 4 9
@)
10 11

12

13

Figure 16.11: Plane stress problem.

A) What is the semi-bandwidth of the assembled set of equations for the nodal
numbering given? Use detailed notation.

B) What is the optimum semi-bandwidth which could be obtained by renumbering


the nodes? Again use detailed notation.

C) For the given element numbering, what is the maximum wavefront for the
problem, in terms of number of active equations?
Problem 6

What is the half bandwidth (in detailed notation) of the stiffness matrix composed
of 6 beams connected end to end in 3-D? Bending, stretching and torsion are allowed.
a) Two
b) Six
CHAPTER 16. BANDWIDTH CONCEPTS 189

c) Ten
d) Twelve

Problem 7
Find the semi-bandwidth, In compact notation, for the axisymmetric body In
Fig. 16.12.

4
y 1
2
3

Figure 16.12: Axisymmetric body

Problem 8
The main reason that 3-D elements (solids) cost so much to run is which of the
following:
a) a solid body requires so many equations for its description
b) solid elements such as the tetrahedron must have complicated shape functions
c) computers can handle matrices K(I ,J) but have more trouble with matrices
K(I,J,L)
d) the bandwidth (or wavefront) is larger for 3-D bodies.
Problem 9
A plate problem has been solved using 6 four-noded elements as shown in Fig. 16.13.
It is desired to redo the problem with the same number of eight-noded elements. If a
Gauss elimination solution is used, by what factor will the computer CPU time increase
for the equation solving portion?

Problem 10
A plane stress study uses rectangular elements (Fig. 16.14). The nodes are shown.
A) How many nodal degrees of freedom are there?

B) What is the half bandwidth?

C) If you renumber , what is the optimum bandwidth?


CHAPTER 16. BANDWIDTH CONCEPTS 190

Figure 16.13: Plate problem.

6 8 10 15
3 ~----~----~------~----~

I
I
I
51
2 -- -- ~ - - - -+-------+ - - - - 14
8 11
I
I

1
d ~ 12
13

cut-out in metal

Figure 16.14: Plane stress problem.

D) If nodes 1 and 3 are restrained in the horizontal direction and node 2 is constrained
in the vertical direction, are rigid body motions in the plane possible?
Problem 11

An engineer has numbered the nodes for a F .E. study of a plate with hole as shown
in Fig. 16.15. Plate elements with 6 D.O.F. per node are used.
A) What is the current half bandwidth, in detailed notation?
B) Show a suggested nodal numbering scheme that cuts the bandwidth to below
130.

C) What is the half bandwidth of your suggested numbering system, in detailed


notation? Show the critical element which causes the maximum difference in node
numbers.

Problem 12

A stiffness matrix is given:


CHAPTER 16. BANDWIDTH CONCEPTS 191

84
83

12
--+-----1 74

99

Figure 16.15: Plate problem.

[k] = [ !]
~ ~
A) The nodes in the stiffness matrix have been resequenced to yield the stiffness
matrix as shown below. Using concepts from large matrix solutions, what reduction
in CPU time results from the resequencing?

[k] = [ ! ~ ~]
B) What reduction in matrix storage would result in a typical banded equation solver
(such as in SAP6) after resequencing? (Don't consider the pointer technique used in
N ASTRAN codes, which eliminates zeroes within the bandwidth.)
Problem 13.
A) In compact notation, what is the bandwidth of the plate problem shown below?

B) Again in compact notation, find the optimum bandwidth that results from renum-
bering the nodes.

C) Using the approximation developed in the notes, what is the ratio of CPU time
required to solve the problem in A) as compared to B), above?

D) For a small problem such as this, discuss the error involved in the approximation
used in part C) above.
CHAPTER 16. BANDWIDTH CONCEPTS 192

Problem 14.

Turner triangles are used to form the small structure, shown in Fig. 16.16. Use
detailed notation for all your discussion, below.

A) What is the bandwidth for the existing node numbering?


B) What is the optimum bandwidth possible using renumbering of the nodes? (You
must show your final scheme.)
C) If the problem were a large one, what would be the percentage reduction in CPU
between the two nodal numbering schemes?

1 2 3

~------~~------~6
4 5

7 8

Figure 16.16: Plane stress problem.

Problem 15.

The finite element model of the ring-like structure in Fig. 16.17 is to be solved by
a wavefront solver. The problem is plane stress.

A) What is the minimum rms wavefront in compact notation?


B) Give an optimum numbering scheme.
Problem 16.

I) Consider the body in Figure 16.18. It is clamped at the bottom surface, and has a
load in the y direction. It has a plane of reflective symmetry with respect to geometry,
material and displacement boundary conditions. The body is much shorter in the y
direction than in the other two directions. The beam is to be modeled with beam
elements and the solid by solid elements. There is an obvious problem of joining the
two types of elements together.

A) Discuss how you would join the beam and the solid bodies in a typical finite element
CHAPTER 16. BANDWIDTH CONCEPTS 193

Figure 16.17: Ring finite element model.

process.
B) Discuss thoroughly how you would exploit the plane of symmetry. Be sure to
include:
i) How does the bandwidth change from the whole model to the half model?
ii) How does the storage requirement for the stiffness matrix change?
iii) How does the CPU time change, considering this to be a large problem?
iv) What boundary conditions must you put on the cut plane of nodes that only are
connected to solids?
v) What boundary conditions must you put on nodes that are connected to both a
solid and a beam element?
II) Repeat the entire discussion for the body in Fig. 16.19 with a load in the z
direction. In this case, the body is much longer in the y direction than in the other
two directions.

z IV
X

Figure 16.18: Body with symmetric plane.


CHAPTER 16. BANDWIDTH CONCEPTS 194

z y
X

Figure 16.19: "Stretched body" with symmetric plane.

P roblem 17.

Figure 16.20: FE mesh for ship

16.8 REFERENCES
1. R. J. Melosh and R. M. Bamford, "Efficient Solution of Load-Deflection Equations,"
Jour. Structural Div . ASCE, Vol 95, 1969, pp . 661-676.
2. B. M. Irons, "A Frontal Solution Program," Int . Jour. Num. Meth. Engr., Vol. 2,
1970, pp . 5-32.
3. 0. C. Zienkiewicz, and R. L. Taylor, "The Finite Element Method," Third Edition,
McGraw-Hill, N.Y., 1988, pp. 89-90.
CHAPTER 16. BANDWIDTH CONCEPTS 195

Figure 16.21: Reflective symmetry for ship


Chapter 17

THIN PLATES

17.1 PHYSICAL DESCRIPTION


Plates are thin, flat structures such as metal sheet, plywood and window panes . If
the plate has thickness of 1/10 or less than its plan dimensions, the plate is "thin."
Plates which have thickness ratios of less than 1/10 are called "thick" and their shear
deformation becomes relatively more important. Neglect of the shear deformation in
thin plate theories causes error of approximately 2% for plates with 1/10 thickness
ratios. This is acceptable and as a result, classical, flexural thin-plate theory is widely
used.

The deflection of thin plates under lateral loading can be described by the deflection
and slopes of the plate's middle surface. Such a Kirchhoff- Love approach is comparable
to the Euler-Bernoulli beam theory. In principle, a finite element thin-plate element
should provide continuity of deflection and normal slope along interelement boundaries,
but this compatibility (C 1 continuity) is not always enforced.

Hughes [1] thoroughly reviews plate elements, particularly those including shear de-
formation (Reissner-Mindlin theories). He finds that the inclusion of shear deformation
is helpful, allowing such elements with only deflection continuity ( C 0 ) to outperform
Kirchhoff- Love elements with slope continuity ( C 1 ). In this chapter, we will present
the Kirchhoff- Love theory, however, and illustrate a simple element. This will serve
only as an introduction to plate elements. There are a bewildering number of plate
elements by now and the derivation of some of the better ones is proprietary and not
well documented.

Lateral loading on plates can cause deflections such as cylindrical curvature or twist
( Fig. 17.1). It is actually rather difficult to deform a flat plate to a cylindrical surface,
because it requires moments on the two sets of opposing edges to be in a specific ratio.
The reader's physical intuition may be corrupted by the ease with which one can roll
paper into a cylinder- this is possible only because large deflections cause midplane

196
CHAPTER 17. THIN PLATES 197

cylindrical curvature

~ twist

Figure 17.1: Cylindrical curvature and twist in plates.

forces that "tame" the bending deflections. Twist is often seen in nature, such as in
the warpage of drying wooden boards.

If one applies moments only at two opposing edges of a plate (Fig. 17.2), one obtains
"anticlastic" curvature. This shows the natural tendency of a plate to relieve inplane
stresses by having opposite curvature in the unloaded direction as the loaded direction
(a Poisson ratio effect). One can explain the effect by looking at the beam cross section
in Fig. 17.2 which is trapezoidal under an end moment. The material expands at the
upper (compression) side of the beam and contracts on the lower surface. A group

Figure 17.2: Anti clastic curvature in plate; bending in beam

of such deformed beams placed side by side give same the form as the anticlastically
deformed plate.

Is it more effective to use the same amount of material in a plate or in beams? A


mental experiment can be proposed (Fig. 17.3) where one enforces cylindrical curvature
CHAPTER 17. THIN PLATES 198

Figure 17.3: Plate and grillage with same amount of material.

on both a plate (with balanced moments) and an equivalent grillage of beams . The
solution would show that the plate is stiffer than the set of beams, by about 10%. This
is due to a factor 1/(1-v 2 ) in a plate flexural rigidity coefficient to be defined later.
There is hence no advantage to slotting a plate, although esthetic reasons sometimes
dictate grillages for park benches, etc.

For plates with thickness greater than 1/10 of the plan dimensions, shear defor-
mations become important and must be considered. As plates become thicker, shear
deformation becomes important because the bending stiffness grows proportionally as
thickness cubed whereas shear stiffness grows only as thickness to the first power.
Very thick plates are so stiff in bending that bending deformations disappear relative
to shearing deformation.

17.2 HISTORICAL COMMENTS


Plates have long been important structural components. Napoleon was interested
in better military installations, including bridges, and offered a 3000 franc award for
a satisfactory mathematical theory for the vibration of plates. [2] Sophie Germain won
the prize in 1815, with the correct fourth order differential equation. The proper
boundary conditions, however, were not found until 1850, by Kirchhoff, who used an
energy approach to show the consistent boundary conditions. Thin plate theory is
often called Kirchhoff- Love theory or Poisson- Kirchhoff theory[1].

17.3 CLASSICAL THIN-PLATE THEORY


We will briefly review the Kirchhoff- Love thin-plate theory. There are several com-
peting sign conventions available. The author prefers the one used in comercial finite
element codes (e.g. NASTRAN) which retains the old-fashioned beam concept that a
positive moment causes compression in the top fiber.
CHAPTER 17. THIN PLATES 199

17.3.1 In-Plane Forces


To show the force and moment resultants, we will use an infinitesimal (rather than
finite) element of material. The resultant forces found by integrating stresses through
the thickness of the plate are F x, F y, F xy and F yx (Fig. 17.4) . If there are no body
moments acting, one can set F xy = F yx. (Body moments are not common in structural

tL x
Figure 17.4: In-plane force resultants for an infinitesimal plate element

mechanics .)

The in-plane resultants are defined:


h/2
j-h/2
rJxdz (17.1)
h /2
j-h/2
rJydz (17.2)
h/2
j
-h/2
Txy dz (17.3)

The in-plane equations of static equilibrium are


aFx aFyx
-
ax
+ay
- 0 (17.4)
aFxy aFy
-
ax
+ ay
- = 0 (17.5)

This is basically a plane stress problem expressed in force resultants. One can show ,
by summing moments over a vanishingly small element, that F yx = F xy As a result,
only the variable F xy is retained .

Boundary conditions on this planar solution involve displacements and force resul-
tants. The variables
u(x,y), v( x ,y), Fx( x, y) , Fy( x ,y), and Fxy(x , y)
CHAPTER 17. THIN PLATES 200

are evaluated on the appropriate boundaries. Since the equations of equilibrium are
first-order differential equations , only one boundary condition per "edge" is needed.
Once this in-plane problem is solved and the resultant forces are known , the out-of-
plane problem can be solved.

17.3.2 Out-of-Plane Forces and Moments


The infinitesimal element in Fig. 17.5 shows a set of moments , twists and shears as
used in NASTRAN .

Figure 17.5: Out-of-plane shear, moment and twist resultants.

Mx -
jh/2 O"xzdz (17.6)
- h/2

My -
jh/2 O"y z dz ( 17.7)
-h/2

M xy -
jh/2 Txyz dz (17.8)
-h/2

Vx
jh/2 Txz dz (17.9)
-h/2

Vy
jh/2 Tyz dz (17.10)
-h/2
(17 .11)

Note that the bending moments on the edges away from the origin tend to cause
the element to curl with positive slope along the adjacent cut edge. Also, the twist on
the faces away from the origin tend to cause positive slope on that same cut edge. The
shears also are defined positive upward on the faces away from the origin. This is the
most friendly of all the possible definitions for plate resultants.
CHAPTER 17. THIN PLATES 201

One uses assumptions similar to Euler-Bernoulli beam theory to proceed.


a) Plane sections in the cross-section which are initially perpendicular to the middle
surface remain plane and perpendicular to that surface after deformation.
b) Stresses normal to the plate surface a zz are zero (in seeming contradiction to the
fact that plates are loaded in that direction!) .
c) The lateral deflection of the plate w( x,y,z) is characterized completely by deflection
of the middle surface w( x,y).
One can then use equilibrium or energy arguments to develop the set of differential
equations for the thin plate, where the lateral load p(x,y) is positive upward in the z
direction:
Eh 3 8 4w 8 4w 8 4w 8 2w 8 2w o 2w
12(1- v2) ( 8x4 + 2 8x2[)y2 + 8y4)- Fx 8x2 - 2Fxy oxoy- Fy [)y2 = p(x, y) (17 .12)

The equation can be simplified by defining the bending rigidity D:


Eh 3
D=---- (17.13)
- 12(1 - v 2 )
and the biharmonic operator:

(17.14)

which yields :
[)2w [)2w [)2w
4
DV w- Fx ox 2 - 2Fxy oxoy - Fy oy 2 = p(x, y) (17.15)
This equation describes the combined flexure and stretch of a thin plate, and is called
"small deflection" plate theory.

We have seen that there is an effect of the resultant in-plane forces on the out
of plane bending, but that the out-of-plane bending does not effect in-plane forces .
This "one-way" interaction is physically true when plate deflections w(x,y) are much
less than the thickness h. The theory is mildly nonlinear (one needs to consider the
structure in the deformed configuration when balancing forces, or alternatively, to carry
higher order terms in Green's strain tensor) . This interaction can lead to buckling of
the plate, when the resultant forces are compressive or when shear resultants are large.

If one chooses to study the pure flexure of a plate, which is a linear problem, one
drops the force resultants in Eqn. 17.12 to obtain:

(17.16)
This equation was first found by Lagrange in 1811 and is considered the "pure flexure"
case. It is the basis for the MZC rectangular plate element to be discussed later.
CHAPTER 17. THIN PLATES 202

17.3.3 Boundary Conditions


Consider the plate in Fig. 17 .6. We will apply boundary conditions along the edge

yf
br-----------,.
I
I

x=a~

__.
a x

Figure 17.6: Boundary conditions at right edge.

x = a. Three common boundary conditions are: clamped, simply-supported and free.


Clamped boundaries imply that the deflection w and slope ~; are zero (Table 1).
Simply-supported boundaries are those where the deflection w and the moment Mx
along the edge is zero. Free boundary conditions are more intricate and involve the
zero moment about the cut edge and a hybrid shear /twist condition (that Germain
and Kirchhoff worried about).

Table 1. Boundary conditions for plate, on x == a face.

Physical Quantity C SS F
Deflection w 0 0
Slope ow 0
ox
Moment D [ EJ2w V EJ2w]
EJx2 + EJy2
0 0
Shear D a ( a2w2 a2w]
- ox ox + 2oy
02
Twist D(1 - v) oxoy
w

Shear - :Y(Twist) -D r~:~ + (2 - v) 8~~~21 o


The physical reasoning behind the combined boundary condition for free edges is
that the twist and the shear both contribute vertical forces. There must be no net
vertical force at any point on the free edge. An equilibrium argument involves the
idea that a growing twist provides pairs (couples) of vertical force (Fig. 17.7). One
can cancel adjacent opposing pairs of forces (Fig. 17.8) and see that there is a net
resultant thrust downward at each intermediate point. The combined law is basically
a statement that
L:;VERTICAL FORCES= SHEAR- a T~IST = 0 (17.17)

You can see why equilibrium requires a great amount of intuition. Energy methods are
better at giving consistent boundary conditions.
CHAPTER 17. THIN PLATES 203

I I
r r

Figure 17.7: An increasing twist along an edge, and how it causes negative shear
distribution.

remainder {

cancelled
shear force

Figure 17.8: Cancellation of vertical components of shear due to twist.

17.3.4 Stress Recovery


The discussion to this point has centered on finding the midplane stresses first,
and then the lateral deflection w(x,y). This may end the problem for some cases, but
usually one wishes to find the bending stresses in the plate. This is done as a separate
step.

The moments in the plate are:

fJ2w fJ2w
Mx = D [ fJx2 + v fJ x2 J

fJ2w fJ2w
My = D [ 8y2 + v 8x2l
The maximum direct bending stresses at the top and bottom of the plate are:
CHAPTER 17. THIN PLATES 204

6My
(C!y )max= v
where the proper sign can be decided by the sense of the moment and whether the
surface is the top or bottom. (Positive Mx and My put compression in the top surface,
similar to the Timoshenko sign convention for beams .)

The maximum shearing stresses due to bending are found under the assumption of
a parabolic shear deformation through the plate thickness:

(Txz)max = ~ ';

17.4 FINITE ELEMENT FORMULATION


Plate elements are very important in vehicle industries : automotive, aerospace and
naval. Many structures such as bridges and buildings are built from plate components .
One can use quadrilaterals and triangles (Figs . 17.9) to model most situations. The

Figure 17.9: Quadrilateral and triangular plate elements.

plate elements can have midside and internal nodes in addition to corner nodes .

In principle, thin plates can have 6 degrees of freedom per node. Many theories
have been developed using only 5, however. The missing d .o.f. is the "drilling" degree
of freedom (rotation about an axis perpendicular to the plate middle surface) . The lack
of a discrete degree of freedom in this direction is because it is the one rotation where
the dimensionality of the surface has not been reduced through integration. For the
other two rotations, a discrete angle is needed to carry forward the stiffness property
of the surface that has been "lost" through integration.
CHAPTER 17. THIN PLATES 205

17.5 MELOSH, ZIENKIEWICZ AND CHEUNG


RECTANGULAR PLATE ELEMENT
This plate element [3),[2] is important for historical reasons and as an educational
tool. Consider a rectangular element with four nodes (Fig. 17.10). There are a total of
12 degrees of freedom . The plate is linearly elastic and with orthotropic material prop-
erties . The stiffness will be developed with force and moment resultants (generalized
stresses) rather than stresses . The main idea is to be able to calculate strain energy
created during flexure . This integration will be done over the plate area, rather than
the volume, because the force resultants already contain the thickness distribution of
stress.

Figure 17.10: Degrees of freedom in MZC rectangle. Pure flexure .

The relevant vector variables for the MZC rectangle are:

{Z: } ~:
Mxy
<== {
Xxy
} <== { w} <== (17.18)
CHAPTER 17. THIN PLATES 206

where the generalized strain-displacement relation defines the curvatures:

{
Xxy }
X = [ ::f,z {w ( x, y)}
]
(17.19)
Xx y axa y
~
[D]

The moment-curvature relation is:

~: D:J { ~J (17.20)

[G]

We will use a displacement-function approach, where a polynomial with 12 gener-


alized coordinates is needed to represent the 12 nodal degrees of freedom:
2 2 3 2 2 3 3 3
w(
x ,)
y = q1 +q2x+q3y+q4x +qsxy+q6y +q7x +qsx y+qgxy +q10y +qnx y+q12xy
(17.21)
This polynomial is geomet!ically isotropic (emphasizes x and y coordinates equally).
It is in the form:
w(x,y) = []{q} (17.22)
where

(17.23)

Rotations at nodes are related to the midplane deflection:


aw
e Xl By(x1,yi) (17.24)

aw
()Yl -ax (x1, Yl) (17.25)
aw
() X2 By (x 2, Y2) (17.26)

etc. This yields the 12 x 12 matrix H relating physical to generalized coordinates:

= [H] (17.27)
CHAPTER 17. THIN PLATES 207

Inverting this matrix analytically was once a formidable task, but now can be done
with symbolic languages such as Mathematica and Maple.

We now have all the mappings needed to calculate the stiffness of the element.

(17.28)

where dA = dx dy.

It is good for the student to see the symbolic form of the plate stiffness for this
case, because is it probably the most complicated element that can be written out
conveniently. (Isoparametric elements exist only in the computer's memory for a short
while, in numerical form.)

(17.29)

where [L] is a 4 x 4 matrix with submatrices [l] on the diagonal:

0 0

and where the matrix [l J is:


[L] =
[!
l
0
0
0
l
0 l (17.30)

~ 2~J
0
[I J = [ 2b (17.31)
0

60
0 0
30 0 20
30 0 15 60
0 0 0 0 0
b2 15 0 10 30 0 20
[I<1J a2 -60 0 -30 -30 -15
0 60
0 0 0 0 0 0 0 0
30 0 10 15 0 5 -30 0 20
-30 0 -15 -60 0 -30 30 0 -15 60
0 0 0 0 0 0 0 0 0 0 0
15 0 5 30 0 10 -15 0 10 -30 0 20
CHAPTER 17. THIN PLATES 208

60
-30 20
0 0 0
-60 30 0 60
-30 10 0 30 20
2
a 0 0 0 0 0 0
b2 30 -15 0 -30 -15 0 60
-15 10 0 15 5 0 -30 20
0 0 0 0 0 0 0 0 0
-30 15 0 30 15 0 -60 30 0 60
-15 5 0 15 10 0 -30 10 0 30 20
0 0 0 0 0 0 0 0 0 0 0 0
30
-15 0
15 -15 0
-30 0 -15 30
0 0 0 15 0
-15 0 0 15 15 0
-30 15 0 30 0 0 30
15 0 0 0 0 0 -15 0
0 0 0 0 0 0 -15 15 0
30 0 0 -30 -15 0 -30 0 15 30
0 0 0 -15 0 0 0 0 0 15 0
0 0 0 0 0 0 15 0 0 -15 -15 0
84
-6 8
6 0 8
-84 6 -6 84
-6 -2 0 6 8
-6 0 -8 6 0 8
-84 6 -6 84 6 6 84
6 -8 0 -6 2 0 -6 8
6 0 . -2 -6 0 2 -6 0 8
84 -6 6 -84 -6 -6 -84 6 6 84
6 2 0 -6 -8 0 -6 -2 0 6 8
-6 0 2 6 0 -2 6 0 -8 -6 0 8

This element has been found to be noncompatible in the sense that the rotation normal
to interelement boundaries is not continuous between elements. The element , although
intended to be a C 1 element , does not meet that requirement. This casts some doubt
on the convergence of the element , but it performs reasonably well in spite of the
noncompatibility.
CHAPTER 17. THIN PLATES 209

A serious drawback of such a rectangular element is the inability to mesh shapes of


arbitrary plan form . At present, most commercial codes contain an excellent quadrilat-
eral element, and a companion triangular element . These can mesh any shape, and are
widely used in practice. Examples are the QU AD4 and TRIA4 in MSC /N ASTRAN.

17.6 COMBINED BENDING AND STRETCH-


ING
A common approach to solving combined bending (flexure) and stretching problems
in plates is to superpose the plane stress problem with the pure bending problem. This
is a linear approach which does not directly couple the in-plane forces to the bending
moments . The two problems are solved separately and results are superposed.

The plane stress problem uses the degrees of freedom shown in Fig. 17.11. These
are the translations u(x,y) and v(x,y) . The flexure problem uses the degrees of freedom

plane
stress

plate
flexure

Figure 17.11 : Degrees of freedom in bending and stretching.

w(x,y), Bx(x,y), By(x,y) . Together, the degrees of freedom are shown in Fig. 17.12. One
can see that the drilling degree of freedom is missing.

There is a lot of confusion about the relation between bending and stretching.
There is no coupling between the two in the purely linear theory. One must use the
small-deflection equation in order to include the coupling, and it is obtained in only
one direction. To get two-way coupling, one must use a strongly nonlinear theory.
CHAPTER 17. THIN PLATES 210

Figure 17.12: Degrees of freedom for combined bending and stretching.

17.7 THIN SHELLS


Thin shells are structural components that have thickness small in comparison to
their plan dimensions, and have initial curvature in one or or both directions. The addi-
tion of curvature is physically important, because it couples the bending and stretching
effects. In linear thin plate theory, this coupling is missing entirely. In small-deflection
thin-plate theory, the coupling is of second order and in one direction only (in-plane
stresses affect out-of-plane deformations). Eggshells show the practical advantage of
curvature, where lateral loads can be carried by midplane stresses rather than by bend-
ing moments.

In spite of the fundamental differences in the mechanics of plates and shells, it is


common practice to model curved shells with many, small flat plates . This results in
a faceted surface, such as for an automobile fender sketched in Fig. 17.13. It has been

Figure 17.13: Faceted surface made from flat plate elements.

found that a mesh of moderate refinement will capture the essence of the interaction
between bending and stretching, even using linear plate elements. This is because at
each intersection, there is a balance of shear and in-plane forces (Fig. 17.14) which
CHAPTER 17. THIN PLATES 211

assembled

'~ ~'
.--------:ploded vi~
Figure 17.14: Force balance at intersection of plates.

effectively allows laterally applied loads to be converted to midplane stresses.

True shell elements account internally within each element for the interaction of
bending and stretching, rather than depending on the intersections between faceted
plates. These shell elements are typically 8-noded quadrilaterals or 6-noded triangles.
The intermediate nodes allow description of the initial curvature, as well as subsequent
deflection.

Current engineering practice is to use many flat plate elements to model a shell
structure, rather than a smaller number of shell elements. The switch to shell elements
will happen when shell elements are more robust and can be extended to nonlinear
problems more accurately.

17.8 THICK SHELLS


When shell thickness becomes more than 1/10 of the plan dimensions of a body,
the shell is "thick." Solid elements are often used to model such thick shells . This
is perfectly appropriate, because the solid elements inherently consider the shearing
deformations and other higher-order effects in thick shells. Figure 17.15 shows a side
view of deformation in a solid element. The 8-noded solid is capable of representing the
shearing deformation found in thick shells. The 20-noded solid element can represent
parabolic deformation through the thickness of the shell, in addition to the shearing
deformation . When the reader encounters a thick shell, he/she can use solids with
technical confidence, although sometimes the cost of analysis can be prohibitive.

17.9 HOMEWORK
P roblem 1
A square plate is clamped on all four boundaries. It is loaded with 1000 N applied
CHAPTER 17. THIN PLATES 212

8-noded
hexagonal
element

20-noded
hexagonal
element

Figure 17.15: Deformation of thick shells as modeled by 8 and 20 noded solid elements.

normal to the center of the plate. It is desired to get an engineering estimate of the
deflection under the load, say within 20%. The plate is one meter square and has the
bending stiffness terms:

Dx 10 4 N m
Dy 10 4 N m
D1 5000 N m
Dxy 2500 N m

It is easiest to work out this problem in units of N, m, sec.

T
lm
P= lOOON

l . . ._______.
~lm~
Top View Oblique View

Figure 17.16: Plate under central load

Problem 2
A classical problem once encountered was how to create a triangular plate bending
element that uses a complete polynomial of degree 5. Imagine you are asked to do this
CHAPTER 17. THIN PLATES 213

task. How many grid points would you need to describe the triangle? What degrees of
freedom would you choose at each node?

Use similar logic as was used in the development of the MZC triangle to proceed.
You may not need all the degrees of freedom at each node to suffice. Don't include
stretching degrees of freedom- they can be left for a separate membrane element.

Problem 3.
The plate structure shown in Fig. 17.17 is to be modeled through the use of a
combined plate bending and stretching element. The element is made up of an MZC
flexural element plus an in-plane (plane stress) element.

It is decided to use a reflective plane and model half of the plate as shown. You are
to state the constraints on the degrees of freedom in the problem, as outlined in the
table following.

A) Use "S" for d.o.f. to be constrained because of symmetry arguments.


B) Use "R" for d.o.f. to be constrained to remove rigid body modes.
C) Use "M" for d.o.f. to be constrained to remove mechanisms.
D) Use "G" for d.o.f. to be constrained to remove nodal (grid point) singularities.

Node degree of freedom


1 2 3 4 5 6
Tx Ty Tz Rx Ry Rz
Constraint list for plate problem. 1
2
3
4

Problem 4.
Consider the plate in Fig. 17.18. It is 10 mm thick and 500 mm by 500 mm in
planform. It is clamped firmly on all four boundaries. A vertical load P of 100 N is
applied at its center. Using four MZC plate elements, estimate the lateral (vertical)
deflection under the load. Do you think the estimate will be high or low? (Remember
that the MZC element is nonconforming, i. e. violates Melosh's rules for convergence.)

Material properties are:


E 680,000 MPa
v 0.3
CHAPTER 17. THIN PLATES 214

physical problem finite element


model

Figure 17.17: Plate problem.

500mm
7
/
10 mm thick
500mm----Y

Figure 17.18: Square plate with central load.


Bibliography

[1] Hughes, Thomas J. R., "The Finite Element Method," Prentice-Hall, Englewood
Cliffs, NJ 1987.

[2] Strutt, J. W. (Lord Rayleigh), "Theory of Sound," (in an introduction by Robert


Lindsay) Dover Publications, NY, 1945 (p. xvi) .

[3] Melosh, R . J ., "Basis of derivation of matrices for the direct stiffness method,"
Journal of the American Institute for Aeronautics and Astronautics, Vol. 1, 1963
(pp . 1631-1637).

[4] Zienkiewicz, 0. C., and Cheung, Y. K., "The finite element method for the analysis
of elastic isotropic and orthotropic slabs," Proceedings of the Institution of Civil
Engineers, Vol. 28,1964 (pp . 471-488) .

215
Chapter 18

COORDINATE SYSTEMS

18.1 DEFINITIONS
The location of a point in space (Fig. 18.1) can be determined through the use of
a rectangular (Cartesian) coordinate system and a set of numbers (coordinates) which
measure distance along the base vector direction . The base vectors are a set of 3 cen-
tered vectors, with their center defining the origin of the system. The coordinates are
usually expressed in a vector form . These coordinates are meaningless when standing
alone, and only make sense in reference to the given set of base vectors . The concept

'I
'
lz
I
I / y ............ r 1
I /
- - - _y . . __ e ~ '.J

X
Rectangular Cylindrical
(Cartesian)

Figure 18.1: Rectangular and cylindrical coordinate systems

generalizes to include angular measure in the case of cylindrical and spherical systems .

A coordinate system defines the underlying space in which a physical problem is


posed, and allows location of geometric grid points (nodes). It also is the system by
which components of forces , displacements and stiffnesses are defined.

In a typical finite element solution, several coordinate systems are simultaneously


used . One, the basic coordinate system, is taken as the reference and the others are

216
CHAPTER 18. COORDINATE SYSTEMS 217

defined in relation to it . In general purpose finite element codes the basic coordinate
system is rectangular and right-handed. The user creates other "local" systems as
needed to more easily define holes, curved surfaces, and loads (Figure 18.2).

We will classify coordinate systems by their type and by their use. Four types of
coordinate systems are: (Figures 18.1, 18.2):

1) Rectangular (Cartesian)
2) Cylindrical
3) Spherical
4) Curvilinear

X
Spherical Curvilinear

Figure 18.2: Spherical and curvilinear coordinate systems

These coordinate systems are used in four different ways :

1) Grid point coordinate system; used to locate the grid points in space.

2) Element coordinate system; allow convenient construction of individual finite


elements.

3) Material coordinate system; for anisotropic materials.

4) Global coordinate system; for assembly of the stiffness matrix. Taken as the
union of all the local coordinate systems.

Much of the problem formulation for the finite element analyst deals with the grid
point coordinate systems . Postprocessing often involves element coordinates. Each of
the four coordinate uses will be discussed in turn .

18.1.1 Grid Point Coordinate System


Most finite element programs assume an underlying, Cartesian, right-handed coor-
dinate system in which to lay out geometry. User created local coordinates are also
CHAPTER 18. COORDINATE SYSTEMS 218

allowed.
a) Basic coordinate system- underlying coordinate system, implicit.
b) Other (local) coordinate systems - created by the user.

18.1.2 Element Coordinate System


These are typically created by the computer program at the stage of element stiff-
ness generation. Stress results are often written out in element coordinates.

Yelem

hexahedron
quadrilateral
beam

Figure 18.3: Assortment of element coordinate systems .

18.1.3 Material Coordinate System


Used for membrane, plate, shell and solid elements, but not needed for line elements
such as trusses and beams. Figure 18.4 shows one possible definition for material coor-
dinates in a quadrilateral plate, where the material coordinate system is a rectangular
system centered at one node and oriented w. r. t. one side.

For the case of anisotropic material, one must define the properties of the material
with respect to a set of "material" coordinates. These are often referred to element
coordinates, or by referring to some geometrical feature of the element. An example is
the case of the QU AD4 quadrilateral bending element in MSC /N ASTRAN . In Fig. 18.4,
the rectangular material axes are rotated an angle B measured from the side connecting
grids G 1 and G2 .

18.1.4 Global Coordinate System


The N ASTRAN series of finite element element codes do not use the basic set of
coordinates for equation solving. Rather, the most convenient coordinate system for
each grid is used . This means that the collection of local coordinates is important, and
is called the global set of coordinates. For instance, a rectangular plate with circular
CHAPTER 18. COORDINATE SYSTEMS 219

Figure 18.4: Definition of material coordinates in typical element : QUAD4.

hole (Fig. 18.5) might have pressure loads on the hole that favor a local cylindrical

Lx
basic coordinate
system

Figure 18.5: Example of "global" coordinates in NASTRAN.

system. The mathematical solution would be easier if one used the basic coordinate
system for nodes away from the hole and a local cylindrical system for nodes on the
hole. This allows boundary conditions to be applied directly in components of force
perpendicular and tangent to the hole's surface.

18.2 COORDINATE TRANSFORMATIONS


Coordinate transformations are necessary for several standard finite element oper-
ations . If nodal coordinates are given in a local system, then it is often necessary to
convert those coordinates to a global system. Stresses, for instance, are often calculated
in a local system but must be expressed in another system.

Translations of coordinate systems do not affect Cartesian components of displace-


ment, stress, etc. (Translations will affect cylindrical and spherical components, how-
CHAPTER 18. COORDINATE SYSTEMS 220

ever.) Rotations do affect components of field vectors in any of the coordinate systems .
As a result , it is important to understand the changes in components under a rotation
of the coordinate system.

18.2.1 Aborted Assembly in Element Coordinates


First, let us consider a false start at assembling a system of truss elements (Fig. 18.6).
One might try to directly add the X etem and Y elem components of displacement at the

Yelem

Figure 18.6: Attempted assembly of two trusses .

common joint. This would lead to failure , of course; a general consideration of compo-
nents relative to a commo n coordinate system is needed. The engineer has a choice
of an underlying (basic) coordinate system or a local system at the joint.

18.2.2 Cartesian Displacement Transformation in 2-D


The components of translation are expressed in matrix form:

(18 .1)

Unfortunately, this notation suppresses the base vectors. The full concept is that :

(18 .2)
ba s e vector bas e v ect or

The components of displacement will change if the coordinate system is rotated.

Consider a coordinate transformation from an element system to a global system


(Fig. 18.7). The displacement components in the element coordinates are:
CHAPTER 18. COORDINATE SYSTEMS 221

Yelem

u 2gJobal / elem

~... I.
lllglobal

Figure 18.7: Relation between element coordinates and global coordinates .

ulglobalcosa + U2globalsina (18.3)


-ulglobalsina + U2globalcosa (18.4)
This is better seen in matrix form :

u1 } [ cos a szna] { u1 } (18.5)


{
u2 elem -
-------
-sina
[t]
cosa

If this is done, node by node, one obtains for the total element :
U2 global

lilA lilA
li2A li2A
0 0
U1B [ [t(r)] [t(aB)] 0
lilB
~ li2B
0 [t(ac)] :] li2B (18 .6)

[T]
elem global

The operator [T] represents the element transformation:

{ U} elem = [T]{ U} global (18.7)


The matrix [T] can be shown to be orthogonal, i. e.

(18.8)
A necessary condition for this orthogonality is that

Det [T] = 1 (18 .9)


CHAPTER 18. COORDINATE SYSTEMS 222

18.2.3 Force Transformation in 2-D


Forces are physical vectors that transform the same way as displacements :

{J} elem = [T]{J}global (18.10)

18.2.4 Stiffness Transformation in 2-D


Consider the equilibrium law:

{J} elem = [k] elem{u}elem (18 .11)

We wish to transform this relation to a global system of coordinates .

[T] {j} global [k] elem[T] {U} global


{j} global [T]-l [k] elem[T] {U }global
[T]T[k] elem[T]{ U }global (18.12)
[k]gl obal

We have hence found the way to convert the stiffness matrix, based on the way that
the vectors transform .

(18 .13)

18.2.5 Example: Rod Element in 2-D


Consider the planar truss assembly shown in Fig. 18.8. We know how to find the
stiffness matrix of each truss in element coordinates . Find the stiffness of the outboard
truss inclined at 30 in terms of the basic coordinate system. The basic and global
coordinates are identical in this case.

The truss element in question (Fig. 18.9) has a stiffness matrix in 2- D:

(18.14)

The transformation matrix is:


sin30 0
cos30 0
(18.15)
0 cos30
0 -sin30
CHAPTER 18. COORDINATE SYSTEMS 223

Figure 18.8: Two-dimensional truss assembly, with basic coordinates shown.

u
2elem

GA

Figure 18.9: Inclined truss element.

-
[
0.866
-0.5
0
0
0.5
0.866
0
0
0
0
0.866
-0.5
0
0
0.5
0.866
l (18.16)

The global stiffness is:

(18.17)

After multiplication, one has:

[k ]global
[
2
cos a
EA cosa sina
L -cos a2
cosa szna
sin 2 a
-cosa szna
2
-cos 2 a
-cosa szna
cos 2 a
-cos a sma
-sin~a
cosa szna
l
(18.18)

-cosa szna -sin a cosa szna sin 2 a


0.433 -0.75
[ 0.75 -0.433]
EA 0.433 0.25 -0.433 -0.25
(18.19)
L -0.75 -0.433 0.75 0.433
-0.433 -0.25 0.433 0.25
CHAPTER 18. COORDINATE SYSTEMS 224

We can check the values in the matrix. The inclined truss ought to be stiffer in the
horizontal than in the vertical direction, i. e. , the kn 91 obal value should be larger than
the k22 91 obal value (Fig. 18.10). The values found from the matrix are:

unit displacement unit displacement


horizontally vertically

Figure 18.10: Comparison of horizontal and vertical stiffness.

EA
y(0.75)
EA
y(0.25)

The truss is indeed three times as stiff in the horizontal direction.

Displacement transformation
After the solution is carried out in global coordinates, the user often requires stresses
and displacements in local coordinates. One can recover the displacements by

{u}elem = [T(30)]{u}global (18.20)

The internal forces and stresses can all be calculated from the nodal displacements.

18.3 HOMEWORK
Problem 1.

A truss element is embedded in 2-D space as shown. If the stiffness matrix in basic
coordinates has a term k41 =- 4.5x10 6 N/mm, what is the stiffness EA/1 of the truss
member?
CHAPTER 18. COORDINATE SYSTEMS 225

___. u3

L X
__.u
1

Figure 18.11: Truss element


Chapter 19

THERMAL STRESS

19.1 EQUIVALENT NODAL LOAD


We have shown in previous lectures that a prestrain {Eo} causes an equivalent nodal
load

{f} e.n.l. = { [B]T[G]{ Eo}dV. (19.1)


lvol
This came as a result of the stress-strain law

{a} = [G] ( { E} - {Eo}) + {a o}. (19.2)

Let us concentrate on thermal strain. We will set {a 0 } = 0 and use the law

{a} = [G] ( { E} - {Eo}) . (19.3)

To develop feeling for the pres train {Eo} , use a scalar example of a line element with
left node fixed. Suppose the element is to be both loaded deflection and the stress-
strain curve for zero temperature (reference level) would pass through the origin, as
shown for T = 0 C in each figure. Once heated, however, a new load-deflection curve
shown as T = +100 C is used. Once could interpret the new stress-strain curve as a
prestress effect , namely as the stress needed to hold the heated line element in place

Figure 19.1: Line element

with zero strain, but this is not done. (Under that interpretation, a negative stress is
needed to hold the element in undeformed shape). Instead, one identifies Eo as the

226
CHAPTER 19. THERMAL STRESS 227

a) b) c)

Figure 19.2: Heated line element under load

amount of strain caused by the free thermal expansion with no stress applied.
This approach keeps signs straight - the thermal strain is positive when associated with
a positive temperature.

19.2 THERMAL STRAIN IN VARIOUS SITUA-


TIONS
In a 3-D solid, if one has local coordinates (x 0 , y 0 , z0 ) aligned with principal material
directions
Exo a1.6.T
Eyo a2.6.T
tzo a3.6.T
{co}= (19.4)
/xyo 0
/yzo 0
/zxo 0
where a 1 , a 2 and a 3 are coefficients of thermal expansion. These ai may be functions
of temperature and should account for all strain between the current temperature and
the reference (here taken as zero for convenience).

In a 2- D, plane stress case,

{Eo} = { ::: } = { (19.5)


/xyo

In a 2- D, plane strain case

{Eo} = { ::: (19.6)


/xyo
CHAPTER 19. THERMAL STRESS 228

If an element has orthotropic material properties, but the coordinate system used
(say x, y) does not lie in principal directions, one has for plane stress:

(19 .7)

If a material has a temperature dependent coefficient of expansion, then, in each of


the expressions above, one replaces the simple a!1T with

{T a 1 (T)dT,
lro
where T0 is a reference temperature.

19.3 EXAMPLE: TWO-NODE LINE ELEMENT


Consider an aluminum link 10" long. Model the link as a two-node line element.
Properties are

Figure 19.3: Al umin urn link

E 10 7 psi
0.3
'
a 1.23 X 10
-5 In
:----op
In
A 1 in 2

The link is to be heated at 100 F and then subjected to various boundary conditions
and loading situations .

The stress-strain law becomes

[E] ( { E} - {Eo}) (19.8)


7
[10 psi] ( { E} - {1.23 X 10- 3 }) (19.9)
CHAPTER 19. THERMAL STRESS 229

(J
/
unheated//
/
/
/
/ heated
~ T = 100F
/...,. ~
I

Figure 19.4: Stress-strain curves for heated and unheated case

The equations of equilibrium are

{ /
/2
1
}
+
{ /
/2
1
}
e.n.l., e:o = EA [ 1
-1
-~ l{:~ } (19.10)

where

{ j: } e.n .l. , 'O


1 01
[B]T[G]{ Eo}dV (19.11)

L [-~ji] [E]{allT}Adx

-EAaflT}
{ EAaflT
-12, 300 lb }
{ (19.12)
12, 300 lb
(19.13)
We have, prior to specifying loads and boundary conditions ,

{ !l
/2
}+{ -12,300} =
12, 300
106~b
zn
[ 1 -1] {
-1 1 u2
ul} (19.14)

A sketch of load vs. deflection is given. The initial loads (equivalent nodal loads
due to initial strain) are absorbed into {!}TOTAL

19.3.1 Constrained Element


If the nodes are constrained, the element is held in place by the external concen-
trated forces. Solve Eq. 19.14

~}
-12 , 300
12,300 } = {
CHAPTER 19. THERMAL STRESS 230

{f}
TOTAL

{u}

Figure 19.5: Combined loading

Hence
!!12 } = { 12, 300 }
{ -12,300 lb.

--+ c_.__________..) +----


12,300 lb 12,300 lb

Figure 19.6: Loads on constrained line element

The stress-strain diagram and the load-deflection diagram are given in Fig. 19.7.

f
TOTAL

Figure 19.7: Stress-strain and load-displacement for constrained line element

19.3.2 Freely Expanding Element


If the same line element as above is heated 100 F and allowed to expand freely,
find the resulting displacements . The external nodal forces are zero. Set u 1 = 0 to
remove the rigid body mode. Equation 19.14 gives

{ ~ }+{ -~~:i~~ }=(10 psi)[ -~ 6


-~ ]{ ~2 }
CHAPTER 19. THERMAL STRESS 231

From the second equation,


u 2 = 0.0123 in

.
) )
u 1=0

Figure 19.8: Free expansion of aluminum link

f
TOTAL

Figure 19.9: Stress-strain and load-displacement for freely expanding line element

19.4 HOMEWORK
Problem 1.
A continuous beam is clamped at both ends as shown in Fig. 19.10. A rod (truss) is
pinned at its ends, at the ceiling and the center of the beam. The loads on the system
are both thermal and mechanical. The rod is first cooled 200 C. It is then loaded with
a vertical load P of 400 N. What is the final deflection of the center point of the beam?

Truss properties:

EA/1 200 N/mm


EA 10 5 N
1 X 10- 5 1/C
CHAPTER 19. THERMAL STRESS 232

Beam properties:

EI = 10 10 N mm 2
Total length = 2000 mm

-'-/LL

rod
' beam /
/
'
,,
.... /

/
'
400N
...
2000mm~
Figure 19.10: Beam with thermal and mechanical loads

Problem 2.
A three-noded triangle for plane stress is given as shown. It has width "a" of 150
mm and height "b" of 100 mm, and is 10 mm thick. The element is pinned at the left
node and on a roller at the right node. What would be the equivalent nodal load in
the vertical direction at the top node due to a thermal load of + 100 C?
The element is made of aluminum, with E = 68,900 MPa, v = 0.3 and thermal
expansion coefficient a = 2.21 x 10- 5 l/C. The element stiffness matrix and other
features are given in Chapter 4.
The shape function matrix for this triangle is:
~-
~- {/;
1 JL 0 JL
[N(x,y)J = [ - a 2b
~-
b (19.15)
0 a
JL
2b 0

Problem 3.
Consider the sheet of composite material in the shape of a triangle as shown in
Fig. 19.12. The triangle is constrained at its 3 vertices. The material is a hypothetical
material that has isotropic elastic properties except for thermal expansion. For thermal
expansion, the terms ax and axy are zero. Find the vertical force exerted by the triangle
on the support at the top node, due to heating the triangle by 50 C. You may use a
single constant strain triangle to estimate this force.
CHAPTER 19. THERMAL STRESS 233

1
1'\
------~-4-----~,~
X
1
2
\
//// /~~/~/~

Figure 19.11: Plane stress triangle under thermal load.

HINT: DO NOT CARRY OUT ALL CALCULATIONS NEEDED T O DO THE EN-


TIRE SET OF THERMAL LOADS- THIS WILL TAKE TWO HOURS! Remember
that a matrix relation
{A}= [B][C]{D}
can be written in summation form:

A i = I: I: B i jcjknk
j k

and that i is a free index.


The solution must be first carried out in symbolic form . A final numerical answer
is desired, however , for sheet thickness of 10 mm, with properties:

E 5.0 x 10 5 N/mm 2
v 0.3
a x 0
ay 2.0 X 10- 4 1/C
a xy 0
~T 50 c
a 100 mm

19.5 REFERENCES
1) Zienkiewicz, 0 . C., and Taylor, R. L., "The Finite Element Method," 4th edition
1989 (pp. 47-50).
CHAPTER 19. THERMAL STRESS 234

Figure 19.12: Constant strain triangle under thermal load.


Chapter 20

GAUSS INTEGRATION

20.1 NUMERICAL INTEGRATION


We often need to integrate a scalar function (such as strain energy) over a volume.

fv f(x, y, z)dV = 0 (20 .1)

In finite elements, the integrand is often very complicated, and can involve large-size
determinants. This makes it important to approximate the integral numerically, by
sampling the integrand at relatively few points. Fortunately, we can develop this
numerical integration theory in terms of a scalar function of one variable, and the
concept generalizes to 2 and 3 dimensions.

The question of integration of an area under a curve (Fig. 20.1) is primarily a


problem of interpolating the curve. The reason is that the other three bounding lines
of the area are no problem. As a result, textbooks on numerical integration make little

f(x)

Figure 20.1: Area under a curve.

distinction between the topics of integration and interpolation.

235
CHAPTER 20. GAUSS INTEGRATION 236

20.2 INTERPOLATION
Hamming[1] gives four basic ways of finding a functionr~.l value f(x) at a general
point x (Fig. 20. 2) when the function is not known every\i\ re analytically.

f(x)

A X B
Figure 20.2: Interpolation

1. Methods which use tabulated functional values from points near f(x), say f(A)
and f(B), to estimate a value between.

2. Methods using differences of functional values, e.g., the straight line approxima-
tion:
f(x) = f(A) + ~-=-~(!(B)- f(A)) (20.2)

3. Methods involving derivatives e.g., Newton's method,


df
f (x) = f (A) + dx I A ( x - A) (20 .3)

4. Gaussian quadrature using arbitrary sample points on the x axis.

The fourth method is the one of interest here. The optimum position of the sample
points is found during the process.

20.3 TRAPEZOIDAL INTEGRATION RULE


The simplest and most stable numerical method of integration is to choose sample
points as in Fig. 20 .3 and to form trapezoidal areas under the curve. The formula for
the sum of the areas under the curve is:
1 1 1
[" f(x)dx 2[f(xi) + J(x2)]~x + 2[J(x2) + f(x3)]~x + + 2[f(xn-l) + f(xn)]~x
~X ~X
2 f(xl) + ~xj(x2) + + ~xf(xn-l) + 2 f(xn) (20.4)
CHAPTER 20. GAUSS INTEGRATION 237

f(x)

Figure 20 .3: Trapezoidal integration.

The "weighting matrix" is :


1 1
~X l -2 1 1 1 1 1 1 -
2
J (20.5)

The end values are weighted only one-half as much as the interior values .

20.4 SIMPSON'S INTEGRATION RULE


If a parabola is fit through each group of three points (Fig. 20.4), the integrating
formula becomes :

(20.6)

where the weighting matrix is


1 4 2 2
~ J
4 4
~xl 3 3 3 3 3 3 3
(20 .7)

The end points are weighted less heavily than the interior points, which themselves
have alternating weights.

20.5 GAUSS INTEGRATION


20.5.1 Symmetric Form
We will discuss the symmetric Gaussian quadrature. The goal is to integrate the
e
area under a curve for -1 ~ ~ 1, where the function can be sampled at any point .
One proposes a series solution:

(20.8)
CHAPTER 20. GAUSS INTEGRATION 238

f(x)

Figure 20.4: Simpson's rule integration.

f(x)
f( ~)

- .......
' --
a x. b -1 ~1 1
1

Figure 20.5: Unsymmetric and symmetric domains.


CHAPTER 20. GAUSS INTEGRATION 239

In other words, one replaces the integral with a finite sum of terms, each consisting
of the product of a "weight" Wk and the function evaluated at an "abscissa" ~k The
weight factors and the abscissae must be found. If done in a general way, using a
symmetric form, this determination of weights and abscissae need only be done once
and for all. Published tables of these constants are available. [2)

20.5.2 One-Term Gauss Integration Formula


Suppose one wishes to integrate a general function on the symmetric domain (Fig. 20.6)
with only one sample point:

(20.9)

f(~)

/
~ -- - f--- --

~
I I

-1 1

Figure 20.6: An arbitrary function to be integrated with one Gauss term.

Using symmetry arguments, one could state that the sample point will be at the
origin of the ~ axis, but we will instead let that fact appear from the solution. For a
trial function f(~) = 1, The first defining equation is:

{ (l)d~ = wd(~,) (20.10)

Therefore
2 = w1 1 t-- 1st defining equation (20.11)
The second defining equation is

(20.12)

Therefore:
0 = w 1~1 t-- 2nd defining equation (20.13)
The solution w 1 = 2 ~1 = 0 leads to the integrating formula:

(20.14)
CHAPTER 20. GAUSS INTEGRATION 240

This will exactly integrate linear functions, e.g. (Fig. 20 .7) :

/(~) = 3+~ (20 .15)

The integrating formula yields:

{
1
(3 + e)ae 2/(o)
2(3)
6 ~ Gauss, exact (20 .16)

f( ~ )

Figure 20 .7: One term integration of linear function.

The same one-point formula can be used to integrate a higher degree polynomial
as sketched in Fig. 20.8. In this case, the formula would underestimate the integral,

f( ~ )

-1
~~~ 1

Figure 20.8: One term integration of higher degree polynomial.

because more area would be neglected on the positive ~ axis than included on the
negative ~ axis.

20.5.3 Two-Term Gauss Integration Formula


Suppose we wish to use a two-term approximation to the symmetric integration:

(20 .17)
CHAPTER 20. GAUSS INTEGRATION 241

This means there are four unknowns, w 1 , w 2 , ~ 1 and ~ 2 . We assume f(~) is available
for numerical sampling, that is, we can insert a value of~ into f(~) and get a number,
even though we can not write out f(~) analytically.

We can impose four conditions on the problem, and have four defining equations
for the unknowns. Assume that f(~) is a polynomial (it often is). Iff(~) were cubic,

(20.18)

The integral can be expanded

~~ (ao+ ale+ a2e 2+ a3e 3 )d~


ao l1 (l)de + a1 ~~ ede + a2l1 ede + a3l1 ede (20.19)

The integration process would be exact if it is separately exact for constant, linear,
parabolic and cubic terms.

Defining Equations
For f(~) = 1, the integration formula becomes

therefore:
2 = w1 + w2 f-- 1st defining equation (20.20)

(20.21)

therefore:
0= w1~1 + w2~2 f-- 2nd defining equation (20.22)

(20.23)

then:
2
3= w1~i + w2~~ f-- 3rd defining equation (20.24)

For f(~) = ~ 3

then:
0 = w 1 ~{ + w 2 ~~ f-- 4th defining equation (20.25)
CHAPTER 20. GAUSS INTEGRATION 242

Solution of Defining Equations


We have created 4 defining equations and can solve for the 4 unknowns w1, w 2 , ~ 1
and ~ 2 . The equations are nonlinear and need a little intuitive trick for ease of solution.
It is known that the abscissa should be symmetrically placed along the ~ axis, since
there is right/left symmetry in the problem. Hence, one can use
~2 = -~1 (20.26)
in place of Eq. 20.25. Equation 20.20 is solved for w 2 :
(20.27)
This and Eq. 20.26 are put in Eq. 20.22:
0 = w16 + (2- wl)( -~1)
= 2~1(w1- 1) (20.28)
To satisfy this equation, we cannot take ~ 1 to be zero, or both sample points would lie
on the same (center) point. Rather, we must choose
w1 = 1

which leads to

Equation 20.24 now gives


2/3 = ~i + (-~1) 2
(20.29)

~1 j1j3
0.577350269 (20.30)
Due to symmetry:
~2 = -0.577350269 (20.31)
We conclude that the expression

tl f( X) = (l.O)f( -0.577350269) + (l.O)f(0.577350269) (20.32)

can be used to exactly integrate polynomials f(~) up to cubics, or can integrate ap-
proximately polynomials of higher degree, or transcendental functions.

20.5.4 Gaussian Formulas with Many Sample Points


The process described above can be continued to cases with many Gauss points.
There are multiple solutions to each number of Gauss points, however, some of them
require sampling the function outside of the double-unit-line domain. These are dis-
carded. Solution of the nonlinear set of equations involves the zeroes of the Legendre
polynomials in general, which is above the scope of this text. Results can be tabulated,
however, and used with confidence for the varied applications (Table 20.1).
CHAPTER 20. GAUSS INTEGRATION 243

Table 20.1 : Abscissae and Weights for Gauss Integration

Gauss points Abscissa, ~i Weights, Wi


1 0.0 2.00000 00000
2 0.55735 02691 1.00000 00000
-0 .55735 02691 1.00000 00000
3 0. 77 459 66692 0.55555 55556
-0.77 459 66692 0.55555 55556
0.0 0.88888 88889
4 0.86113 63116 0.34 785 48451
-0 .86113 63116 0.34 785 48451
0.33998 10436 0.65214 51548
-0.33998 10436 0.65214 51548
5 0.90617 98459 0.23692 68851
-0 .90617 98459 0.23692 68851
0.53846 93101 0.4 7862 86705
-0 .53846 93101 0.4 7862 86705
0.0 0.56888 88889

20.6 EXAMPLE OF GAUSSIAN INTEGRATION


IN ONE DIMENSION
Consider the polynomial sketched in Fig. 20.9:

!(~) = ~3 + ~~2 - ~~ (20.33)


4 8
This curve is sufficiently intricate to be interesting. One would intuitively expect an
exact interpolation to involve many sample points.

Use Gauss integration to find the exact value of

1 (~3 + ~~2
1
-1 4
- ~~)d~
8
(20 .34)

The exact solution only requires a two-term Gauss integration formula. It is amazing
that only two functional samples can account for all the positive and negative areas!
The form of solution is :

t j(e)de = 1.00000f(0.577350) + l.OOOOOj( -0.577350)


= (0.57735o? + ~(0.577350) 2 - ~(0.577350)
CHAPTER 20. GAUSS INTEGRATION 244

1.0

f(~ )

Figure 20.9: Cubic polynomial on symmetric domain .

( -o.57735W + ~( -o.57735W - ~( -o.57735o)


0.166667 {= Gauss

The analytical answer, for comparison, is :

1 (~3 + ~~2
1
-1 4
- ~~)d~
8
1

1
4 1 3 3
4~ + 12 ~ - 16 ~ - 1
211

Exact
6
There is no algorithm error in this two-point formula as applied to a cubic, only round-
off error.

20.6.1 Nonsyrnmetric Gaussian integration


The nonsymmetric form of Gaussian quadrature is:

(20.35)

where
b-a b+a
Xi=
2
(--)~i + (-2- ) (20. 36)

This allows one to apply the symmetric, normalized data pairs ( wi, ~i) given in math
tables to a given physical problem which is not symmetric. One determines the Gauss
points in physical coordinates Xi and evaluates the integral directly.
CHAPTER 20. GAUSS INTEGRATION 245

20.7 GAUSS INTEGRATION IN 2-D


20 .7.1 Best Theoretical Procedure
The logical extension of the 1-D Gauss procedure would be to integrate over a
square region (Fig. 20 .10) with the formula:

llll J(e, TJ)ded'T] = ~ wd(ei, TJi) (20.37)

This would yield 3 defining equations per sample point. Since the polynomial in 2-D

~i o
:;l
ill i ...

Figure 20 .10: Two-dimensional domain .

is of the form:
!(~, ry) =a+ b~ +cry+ d~2 + e~ry + /7]2 + 9"73 + h~27] + P~"72 + qry3 + ... '
one needs 2 sample points to define a quadratic, 4 sample points for a cubic, 5 for a
. quartic and 7 for a quintic. But-this is not the way it is done in practice!

20.7.2 Practical Procedure


It has been found expedient to use the 1-D Gauss integration repeatedly in the 2nd
(and 3rd) dimensions, so that the pattern of Gauss sample points is a regular pattern
(Fig. 20.11). The approach is approximate in general, but is exact for certain functions,
such as those that are separable in the variables. An example would be:

llll (e + TJ2)d~dTJ ll [~
= w;W + TJ2)JdTJ
2 2
L Wj[L Wi(~; + ry})]
j=l i=l

L L WjWi(~; + ryJ) (20.38)


j
CHAPTER 20. GAUSS INTEGRATION 246

-
I
~-
_._
I
-

I I
I I
~- -
- -
I
I I
Figure 20.11: Regular pattern of Gauss points .

For most practical problems, the number and location of Gauss points will be
automatically chosen by the general purpose finite element program. In research ap-
plications, however, the engineer will want to control the number of Gauss points .

20.7.3 Comments
1. If you know the degree of a polynomial to be integrated, it does not pay to use
more integration points than necessary for an exact answer . This arises in plate
and shell elements .

2. Abramowitz and Stegun [2] give tables for abscissae and weight factors for Gaus-
sian integration for up to 96 sample points and carry 15 decimal places!

3. The solution for the defining equations in general involves finding zeroes of the
Legendre polynomial.

4. There are multiple solutions for the abscissa and weight factors, which is due to
the nonlinearity of the defining equations . There are 3 sets for cases with less
than 13 samples points (n < 13) and 4 sets for 16 ::; n ::; 96 . There may be
additional solutions, but we discard them because we desire real numbers and
desire that the abscissae lie in the integration region.

20.8 RADAU AND LOBATTO INTEGRATION


A variation in Gaussian quadrature is to take one or two sample points to be
fixed, typically at the endpoints of the region (vertices of an element). If you choose
one sample point a priori to be at the end of the region, you have Radau integration
(Fig. 20.12) whereas choosing two such points gives Lobatto integration (Fig. 20.13) .
Specification of a sample point provides more accurate results at that point; however,
CHAPTER 20. GAUSS INTEGRATION 247

f(~)
_....- --
.---- - -
I

-1
-- /

1
~2

Figure 20 .12: Radau integration, with sample point at left end.

f(~)

...---- -
1 I
-- I

-1 ~ 1- 1

Figure 20.13: Lobatto integration with sample points at both ends.

it also eliminates a defining equation and reduces the overall accuracy of the method.
There is a trade-off involved in the choice of methods. Radau integration has been
used in triangular plate bending elements (Fig. 20 .14).

20.9 Finite Element Applications


There are two major integrations performed in FEA . The first is for the stiffness
matrix:
[k] = /.)B]r[G][B]jJ(e , 7] , ()ided7Jd(
The second is the calculation of equivalent nodal loads:

f [N]T{P}dSPACE
} s PACE

The number of Gauss points needed depends on the complexity of the shape func-
tions , the Jacobian and distributed loads.

20.10 HOMEWORK
Problem 1.

A polynomial f( x ) = 1 + x 3 - 3x 4 is defined over the region 2 ~ x ~ 10. You wish


to get an approximate value of the integral
CHAPTER 20. GAUSS INTEGRATION 248

0 0

0 0

4-node rectangle 4-node quadrilateral

3-node triangle. Radau. 8-node hexahedron

Figure 20.14: Examples of Gauss point locations.

Use a two-point Gaussian integration to provide the estimate of the integral. Is this
actually an exact answer, or not? (No Riemann integrals accepted for the answer!)

Problem 2.
An engineer wants to develop a special-purpose Gaussian integration formula for the
symmetric domain from -1. to +1. She wants to integrate exactly functions f(x) which
are known to be 5th degree polynomials, but do not contain any constant or linear
terms. Write down the relevant defining equations that will be needed to develop an
exact Gaussian integration scheme for this problem, and which uses the fewest sample
points. Carry out a few steps of the solution for the abscissa and weights; you do not
need to carry it to completion.

Problem 3.
Integrate the function

over the interval [-1 ,1] using Gaussian integration and 3 sample points. Is your answer
exact or not? Why?
Bibliography

[1] Hamming, Numerical Methods for Scientists and Engineers, McGraw- Hill Book
Co., New York, 2nd Edition, 1973, Chapters 14-19.

[2] Abramowitz, M. and Stegun, I. A., Handbook of Mathematical Functions, National


Bureau of Standards, 2nd printing, Nov. 1964, pp. 887, 917-919 .

249
Chapter 21

ISOPARAMETRIC ELEMENTS

21.1 INTERPOLATING INDEPENDENT VARI-


ABLES
In previous chapters, we have often interpolated dependent variables (functions).
We now want to broaden this process to interpolation of independent variables . This
is rather unmotivated, so we will review the old concept first .

21.1.1 Review of Interpolation of Functions


To interpolate the function f(x) (Fig. 21.1), we depend on sample values at the

f(x)

- -- -- --
f(x)

N 1 (x) N 2 (x)

>< x
Figure 21.1: Shape functions for line element.

f(x) f(xt) + [f(x2) - f(xt)] (x - Xt)


X2- Xl

[1- x-xl]f(xt)+ [x-xl]f(x2) (21.1)


x2- xl x2- xl
N1(x)j(x1) + N2(x)j(x2) (21.2)

250
CHAPTER 21. ISOPARAMETRIC ELEMENTS 251

It seems logical, by now , to scale the dependent variable f( x) according to the value of
the independent variable x.

21.1.2 Interpolation of Independent Variable


If we propose to generalize the interpolation to the x variable, perhaps using the
same shape functions as above, we get:

(21.3)

This is a curious approach, interpolating a variable by referring to its own endpoint


values . It is not as crazy as it seems, however, and is often used when a transformation
between two sets of independent coordinates is used. The approach illustrated, map-
ping both the dependent and the independent variables with identical shape functions,
is called an isoparametric approach, and leads to isoparametric elements.

Equation 21.3 can be shown to be an identity:

X2X1- xi- XX1 +xi+ XX2- X1X2


X = x2- xl

The utility of this type of interpolation lies in the case of transformation of coordinates
between two sets of independent variables.

21.2 ISOPARAMETRIC LINE ELEMENT


Suppose one wished to integrate to find the stiffness of the line element, using
isoparametric mapping. The element would be mapped to the double unit line (Fig. 21.2).

We define ( such that

(21.4)

We will interpolate the field variable u( x):

1 1
u(x) = (1- ()u 1 + (1 + ()u 2 (21.5)
2 2
CHAPTER 21. ISOPARAMETRIC ELEMENTS 252

I ::::9 I
-1

Figure 21.2: Mapping the line element onto the parent element.

The line element formulation can be converted from physical to ~ coordinate:

[k] = fvlB(x)f[G(x)][B(x)]dV(x)

= fvlB(x(e))f[G(x(e))][B(x(O)]IJ(e)ldV(e)
where
dx
J -
d~
-xl x2
= -2+ -
2
X2- X1
= 2
L
-
2
[G] - [E]
[D] - [~]
dx
= [~ d~]
d~ dx
2d
= [L d~]

A two-noded line element lying on [-1,1] on the~ axis has shape functions:

(21.6)

Hence the strain matrix [B] is:

[B] = [D][N]
2d 1 1
= [- - ] [ -(1- ~)
L d~ 2
-(1
2
+ ~)]
1
= [-- ~]
L L
CHAPTER 21 . ISOPARAMETRIC ELEMENTS 253

Finally,

IL/21 Ad~
1
1
[k]
! -1
[ - / Ll [E)[-1/ L
1/ L 1/ L]

EA [
L
1 -1]1
-1
This is the same stiffness matrix as found earlier using physical coordinates for the
derivation.

21.3 ISOPARAMETRIC QUADRILATERAL


The creation of isoparametric elements in two dimensions is much more interesting.
The distinction between mapping the curved sides of an element into straight sides
on the "parent" element in the~' r; coordinates becomes apparent . Unfortunately, to
illustrate the details of such mapping becomes difficult, so that the example chosen
here will be a "straight-sided" element.

Consider the four-noded quadrilateral shown in Fig. 21.3 . We wish to find the

Figure 21.3: Two-dimensional quadrilateral.

stiffness matrix:
[k] = j~)B]T[G][B]dV
which, for plane stress, reduces to:

[k] = h DBf[G][B]dxdy

where

[G)
(1}!v2)
Ev
2
[ (1-v )
(t~') G~]
(1-v 2 )
0 0
CHAPTER 21. ISOPARAMETRIC ELEMENTS 254

[D] [J oy
~]
ax
We need to invent a~, TJ coordinate system which maps the quadrilateral (Fig. 21.4)
onto the double-unit square (the parent element). We interpolate the independent

Figure 21.4: Mapping to the parent element.

variables (underlying coordinates x and y) using ~ and TJ as parameters:

(21. 7)

The coordinate y is handled similarly.


4

X = LNi (~,TJ)Xi (21.8)


i =l
4
y = L Ni (~ , TJ )Yi (21.9)
i =l

The shape functions satisfy the conditions that they have unit value at the "home"
node and zero value at the other nodes. For instance, the first shape function (Fig. 21.5)
satisfies:

N1( -1 , -1) 1
N1(1 , -1) 0
N1 ( 1, 1) 0
N1( -1 , 1) 0

We can use a product form for the shape functions. Choose:

N 1 ( ~ , TJ) = Constant ( 1 - ~) ( 1 - TJ) (21.10)


CHAPTER 21. ISOPARAMETRIC ELEMENTS 255

N 1 (x,y)

Figure 21.5: The shape function N1 (x,y).

Evaluate the constant:


N1( -1, -1) 1 = Constant(2)(2)
Constant 1/4
1
~)(1- TJ)
-(1-
4
By similar reasoning, the entire shape function matrix is found.

N2(~,ry)

N3(~,ry)

N4(~,ry)

xl
Y1
X2
{ ~} = [ ~1 N1
0
~J
N2 0 N3 0 N4 Y2
0 N2 0 N3 0 X3
Y3
x4
Y4
This is the "serendipity" family of shape functions.

The field variables are interpolated the same way as the independent variables:
7
Ux = L Ni(~, TJ )ui
i=odd
8
uy L Ni(C TJ )ui
t = even
CHAPTER 21. ISOPARAMETRIC ELEMENTS 256

Our goal is to convert

[k] = h j .L[B(x,y)]T[G(x,y)][D(x,y)][N(x,y)]dx dy (21.11)

into

(21.12)

where the determinant of the Jacobian matrix is defined:

ax
IJI =- ax ae I
ae u
ay (21.13)
ary ary
The Jacobian appears naturally in the chain rule:

a a ax
- - +ay
- a~
-
a ay
a~ ax a~
a a ax
- -+--
a ay
ar; ax ar; ay ar;
which is written in matrix form:

=
ax
~;
[ ary g;ay
ary
l{gx
a }
ay
The derivatives found in the Jacobian can be found from the basic interpolation for-
mulas:

We require the derivatives with respect to x andy in the original integration formula.
If [J]- 1 exists, we can invert the relation:

The stiffness matrix can now be written out:

1 1
+ J-1
J-1 a
11
a
12 a:;;
8[
[kJ=hL/)Bf[GJ [ _ o _
18 18
]21 8[ + ]22 a:;;
CHAPTER 21. ISOPARAMETRIC ELEMENTS 257

(~)[(1-~)(1-ry) 0 (1+~)(1-ry) 0
4 0 (1-~)(1-ry) 0 (1 +~)(1- ry)
11(1+~)(1+ry)
II o
0
(1+~)(1+ 77 )
(1-~)(1+ry)
0 (1-~)(1+ry)
0 ll Jn
]21
]121 d~ d17
]22
(21.14)
where the terms J~J refer to the i, j component of the inverted matrix [J)- , not the 1

inverted component of [J). This expression is numerically integrated on the parent


element, as shown in Fig. 21.6 . Each of the terms in the expression is a scalar function

4 3
0 0

---

0 0

1 2

Figure 21.6 : Gauss integration in two dimensions.

on the double-unit square (Fig. 21. 7) .

Figure 21.7: Integration of strain energy term in two dimensions .

21.4 HOMEWORK
Problem 1.

A rectangular finite element in two dimensions is given in Fig. 21.8 . It is to be


numerically integrated using a mapping to the double unit square, also shown.

A) To what point (x,y) does the point(~, ry) = (0.5, 0.2) correspond?
CHAPTER 21. ISOPARAMETRIC ELEMENTS 258

B) What is the value of IJI? You should show your calculations or your logic for your
value.

(-1 ,3) (1,3)

y 11

L ~X

~~
(0.5,0.2)
(-1,-3) (1,-3)

Figure 21.8: Two-dimensional finite element

Problem 2.
Consider the mapping between the quadrilateral element shown in Fig. 21.9 and
the double unit square (Fig. 21.10). Suppose a Gauss integration point lies at a point
(0.5,0.5) in the(~, r;) plane. To what point in the (x,y) plane does this correspond?

y (5,5)
(2,4) 3
4

2
1 (4,2)
(1,1)
X

Figure 21.9: Isoparametric element in 2-D

Problem 3.

An eight-noded plane stress element is shown in Fig. 21.11. The element is mapped
onto a parent element in the ~, r; plane.

A) A displacement polynomial has been proposed for the displacement u( ~, r;) as:
CHAPTER 21. ISOPARAMETRIC ELEMENTS 259

(-1,1) (1,1)
4 3

(0.5,0.5)

1 2
(-1,-1) (1,-1)

Figure 21.10: Double unit square

Does this have the proper number of terms? If not, propose a different polynomial.

B) Discuss serendipity-type shape functions for the quadrilateral. These shape func-
tions need only be cubic (no higher). Describe the functions in general, and then give
two qualitatively different specific cases for N i ( C1]).

-
1 2
X

Figure 21.11: Eight-noded quadrilateral plate element

Problem 4.
Consider the three finite elements with the specific load cases in Fig. 21.12. In each
case, fill in the charts below to tell which of the components of equivalent nodal load
will be zero. Use an x for a nonzero term and a 0 for a zero term. Explain the type of
shape functions you are using for the quadrilateral problem-try to use only cubic (not
higher degree) polynomials.

Nonzero terms in beam problem.


CHAPTER 21. ISOPARAMETRIC ELEMENTS 260

1 3

2(! ++++++!)4 2

3
Figure 21.12 : Euler- Bernoulli beam, Turner triangle and plane stress quadrilateral.

Grid fy Mz
1
2

Nonzero terms in constant strain triangle.


Grid fx fy
1
2
3
CHAPTER 21. ISOPARAMETRIC ELEMENTS 261

Nonzero terms in eight-noded quadrilateral.

Grid fx fy
1
2
3
4
5
6
7
8

P robletn 5.

A special parabolic line element is to be developed using isoparametric concepts.


The element lies on the x axis as shown in Fig. 21.13.

A) Find the mapping that maps the element onto the "parent" element on thee axis.
You will use a quadratic relation for this.

B) Two-point Gauss integration will be used. What is the physical coordinate' x 1


corresponding to the Gauss point on the right?

Gauss points

X ~ .-o
_)\._
a Os ~
0.0 1.0 2.4 3.0 -1.0 0.0 1.0

Figure 21.13: Parabolic (quadratic) line element.


Chapter 22

SOLID ELEMENTS

22.1 TYPICAL TYPES OF SOLID ELEMENTS


Solid bodies are often modeled with tetrahedral, wedge and hexahedral elements.
There is a current debate about the order of the element that should be used. General
consensus is that the linear tetrahedron is not good. Automated meshing software must
use quadratic or higher order tetrahedrals. On the other hand, there is evidence that

tetrahedron wedge hexahedron

Figure 22 .1: Solid elements.

the linear hexahedron is more effective, when used in larger numbers than quadratic
hexahedra. This may change in the future, as more robust higher-order hexahedra are
developed.

262
CHAPTER 22. SOLID ELEMENTS 263

22.2 ELASTICITY THEORY FOR SOLIDS


The theory for solid elements is based on the linear theory of elasticity. There are
no moments defined at points, as there are in the structural theories . We will use the
general mapping shown in Fig. 22 .2. The generalized Hooke's law [G] is the important

-1
[H]
~ u

Figure 22.2: Vector relations for elastic finite elements.

material relationship. For the most general material (anisotropic) :

G1 ,1
G1,2 G2 ,2 (symm)
G1,3 G2,3 G33
[G] = ' (22.1)
G1,4 G2,4 G3,4 G4,4
G1,s G2,s G3,s G4,s Gs,s
G16 G2,6 G36 G1,6 Gs 6 G6,6
' ' '
For orthotropic materials, such as laminated composites (Fig. 22.3):

Figure 22.3: Laminated material.


CHAPTER 22. S OLID ELEMENTS 264

(symm)
G3,3
[G] = 0
(22.2)
0 Gs s
'
0 0

The number of elastic constants is reduced from 21 to 9.

Many engineering materials are isotropic, i.e., have the same properties in all di-
rections at any point. In this case, there are only two material constants , typically
represented by Young's modulus E and Poisson's ratio v.

1-v
v 1-v SYM
v v 1-v
[G] = E 1-2v (22.3)
(1+v)(1-2v) 0 0 0 2
1-2v
0 0 0 0 2
1-2v
0 0 0 0 0 2

The strain-displacement law is purely geometric and does not depend on the ma-
terial:
ax a0 0
0 a;; 0
0 0 fua
[D]= a a (22.4)
a;; ax 0
0 a a
az a;;
!L_
0 a
az ax

22.3 CONSTANT-STRAIN TETRAHEDRON


This element is the generalization to 3-D of the constant strain (Turner) triangle.
It has four nodes and four faces. The displacement field can be expressed as a complete

Figure 22.4: Constant strain tetrahedron.


CHAPTER 22. S OLID ELEMENTS 265

linear function of x,y,z :


u( x ,y, z ) q1 + q2 x + q3y + q4 z
v( x, y ,z) qs + q6x + q7y + qs z
w( x, y,z ) qg + q10x + quy + q12 z

The standard finite element theory applies :


[k] f [B]T[G][B]dV
lvoL
{f} enz f [N]T {P}dSPACE
}sPACE
There exist a set of "natural" volume coordinates which assist in the integration of
these quantities. The volume coordinates are shown in Fig. 22.5, where the quantities
vl, v2, v3 and v4 are the fractional volumes contained in the subtetrahedron facing
away from the node in question. These add to unity, therefore there are only three

4 4
1
2 2

Figure 22.5: Volume coordinates defined on tetrahedron.

independent coordinates to define the point. The coordinates themselves are shape
functions, because they take unit value (the whole volume) when evaluated at the
home node and zero volume when the point of interest is at the other nodes .
Linear tetrahedrons are not practical elements to use in stress problems because of
their simple stress field . On the other hand, they are widely used in heat conduction,
often to generate a hexahedron element by filling the internal space with tetrahedra.
The quadratic tetrahedron is becoming important because of automated meshing
software that fills arbitrary volumes with tetrahedra. We will not study tetrahedra
further, however .

22.4 HEXAHEDRON
The hexahedron is currently the favorite solid element. Typical commercial FE
programs allow use of 8 to 20 nodes in each element. The user must provide appropriate
transitioning from coarse to fine mesh.
CHAPTER 22. SOLID ELEMENTS 266

The solid hexahedron is usually an isoparametric element, and is numerically inte-


grated. It is difficult to develop intuition about such an element, because the analytical
form of the stiffness matrix cannot be shown!

Solid elements often suffer from being either too stiff or too flexible. If there are not
enough Gauss integration points, they can have zero-energy deformation modes and
be too flexible. [1 J Figure 22.6, for example, shows a linear "brick" element with only
one integration point. When exposed to the particular reversed loading through the
thickness, there would be no strain energy density at the mid-elevation line. A series

\ 7 I I 7
\ 0 v 0
v 0 v 0
I
v
~ I I I
\ I .....______,_l_.______.&...-l___.______./'-------'

Figure 22.6: Zero energy deformation mode for hypothetical element solid element

of such elements might be very flexible under such loading.

Figure 22.7 shows a brick element with a 2x2x2 Gauss point integration, where the
"hour glass" deformation shown would cause no strain-energy density at the integration
points. Elements must be formulated to avoid these modes of deformation.

--

Figure 22.7: Hour-glass deformation mode in solid element.

Another problem in solid elements is "locking," where under certain loading on


certain distorted element shapes, the element can be as much as 100 times stiffer
than normal. This even happens in commercial FE programs. One avoidance is to
use several solid elements through the thickness of a structural element, so as not to
subject an individual element to a complete stress reversal. Another avoidance is keep
solid elements as close to rectangular prisms as possible.
CHAPTER 22. S OLID ELEMENTS 267

Care must be taken in the theoretical development of solid elements so that when
the element is "flattened out," perhaps to model a plate, it properly represents the
shear behavior of a plate (or shell).

In overcoming problems of the sort mentioned above, many element developers use
different Gauss points for integrating direct strain energy than for shear strain energy.
Usually the shear strain integration is done with fewer Gauss points.

22.5 EXAMPLE- THE MSC/NASTRAN HEXA


ELEMENT
A well-known element for solid stress analysis is the MSC/NASTRAN HEXA ele-
ment, which can be used with 8 to 20 nodes . It was originally based on the work on
isoparametric elements by Irons, but has evolved over the years in a proprietary way.

A solid element described in physical coordinates is mapped to the double unit


cube (parent element) shown in Figure 22 .8 In this particular element, if the engineer

z
k!x
Figure 22.8: Mapping of solid hexahedron onto the parent element.

chooses to use only 8 nodes, the direct strain energy integration is carried out with
a 2x2x2 array of Gauss points . The shear strain energy uses a similar array, but at
different locations .

The 9 to 20 noded HEXA element uses a 3x3x3 array of Gauss points for evaluating
direct strain energy and a different 2x2x2 array of Gauss points for shear strain energy
evaluation. Richard MacNeal says "These procedures are necessary to relieve internal
constraints which destroy accuracy when the element is used to model thin shells." [2]
CHAPTER 22. S OLID ELEMENTS 268

o 3 x 3 x 3 Gauss points
2 x 2 x 2 shear points

Figure 22 .9: Gauss point arrays for 20 noded HEXA element

22.6 PENTAGONAL SOLID ELEMENT


If an engineer meshes a body primarily with hexahedral elements he/ she will find
corners where wedge elements are needed. Most commercial codes have a pentagonal
element which is a partner with the hexagonal element in the code. Some derivations
of the pentagonal element allow it to be formed from a hexagon, by collapsing one face
of the hexagon to a line.

The PENTA element , which is the wedge companion to the HEXA element in
the MSC /N ASTRAN program uses a separate formulation from the hexahedron. The
PENTA is mapped from its physical shape to that of a 90 triangular prism in the C
ry , ( space (Fig. 22.10) .

Figure 22 .10: Mapping the wedge element on the right angled prism.

The 6 noded PENTA element uses 6 Gauss points for direct' strain energy and some
shear components, and 3 special points for transverse shear strain. The 7-15 noded
PENTA uses 9 Gauss points for Ey, Ez, and ! xy It then uses 6 special points for those
components Ez, !yz and !zx, which control the stiffness behavior through the thickness
CHAPTER 22. SOLID ELEMENTS 269

when the PENTA gets thin in the z direction (The element is intended to approximate
a plate by thinning in that direction only.)

z y

r
X

Figure 22.11: PENTA element reduced to plate-like dimensions.

The author has used 6-noded wedge elements to model bolts, with the elements
arranged in a pie-shaped fan through the bolt cross section. The results were not useful
for bending behavior of the bolt because the stress distribution was too distorted. (The
stress through the triangular section of the 6 noded wedge element is constant, just
as it is in the constant strain triangle. This is too imprecise to approximate many
problems.)

22.7 HOMEWORK
Problem 1.
A cube of porous material behaves as a linear elastic solid. It has zero Poisson's
ratio. An experiment is carried out for a cube (Fig. 22.12) which is clamped on the
bottom surface and subjected to the set of loads in Fig 22.13. The only nonzero
displacement is u 15 = 0.001 mm. The other loads shown on the top surface are the
reactions needed to hold those vertices at zero displacement. There are no horizontal
forces at any of the nodes , including the reactions at the base. Find as many of the
stiffness coefficients as possible in the 24x24 stiffness matrix for the cube.
CHAPTER 22. SOLID ELEMENTS 270

~X
Figure 22.12: Cube of elastic material.

lOOON
200N lOON

t +
0.001 m f
'
z

~X
Figure 22.13: Forces required to cause displacement at one node only.
Bibliography

[1] Cook, R. D., "Concepts and Applications of Finite Element Analysis," John Wiley
& Sons, New York, Second Edition. 1981, pp. 134-136.

[2] MacNeal, Richard, MSC/NASTRAN Application Manual, The MacNeal-


Schwendler Corp., Los Angeles, 1991.

271
Chapter 23

CASE STUDY: HYDRAULIC


VALVE

23.1 PHYSICAL PROBLEM


A major hydraulic equipment manufacturer was interested in increasing the operat-
ing pressures of its valves from 3500 psi to 5000 psi. To do so required better knowledge
of the stress and displacement distribution in its existing line of valves. Fatigue tests
had shown some cracking in one design due to fatigue, during a 2,000,000 cycle test.
This was more cycles than normally seen in the field, indeed there had been no field
failures at the time.
The baseline design (Fig. 23.1) had 3 bores, which contained 6 plungers that reg-
ulated the flow of high-pressure hydraulic fluid . The typical application was heavy
off-road equipment such as road graders. The casting was a single piece, gray iron, and

/
/

./ ./ ./
./ ./ ./
/ / /

Figure 23.1: Sketch of three bore hydraulic pump.

made by sand casting. The bores and internal fluid paths were formed by sand/epoxy

272
CHAPTER 23. CASE STUDY: HYDRAULIC VALVE 273

core, stiffened with metal rods .

The valve in question had failed at the parting line of the mold. This may have
been due partly to some offset in the upper and lower mold halves, which introduced
a joggle at the mold line.

The goal of the project was to analyze the valve housing for stresses and displace-
ments . If a better design could be seen, it would be proposed, as well.

23.2 COST AND TIME ESTIMATES


An experimental program requiring the casting of a hydraulic valve prototype re-
quires 6 months time and $10,000 for the casting, alone (Table 1) . Any proposed
change in the casting requires another 6 months and $10,000. On the other hand, the
initial finite element study was estimated to cost $5,000, and take 1 month (the actual
costs were $6,200 and 5 weeks). Additional studies for design parameter changes cost
about $1,000 and took about a week to complete. Finite element modeling saves cost
and calendar time compared with making and testing prototype castings .

Table 1. Cost and time estimates for hydraulic valve study.


FEA EXPERIMENTAL
First Case Additional Cases First Case Additional Cases
Calendar Time 1 mo. 1 wk. 6 mo. 6 mo.
Total Cost $5,000 . $1,000. $10,000 . $1 0,000.

23.3 PHYSICAL MODELING


The valve is a solid body, as opposed to a beam, plate or shell structure. It is
expected that the material will remain in the linear elastic range. The displacements
are expected to be small so that linear geometric effects are retained. The pressure
loads are to be applied to certain chambers of the fluid-filled region, in a "worst-case"
scenano.

A small portion of the body (1/12) can be isolated (Fig. 23 .2) by claiming that
there are 5 planes of reflective symmetry. Three of these can be taken as fixed and 2
must be allowed to move. The idea that two of them must move was discovered by
trial and error. The body is internally pressurized and must be free to expand in all
directions. If all 5 planes are held fixed in space, the material between reflective planes
is "trapped" and cannot expand. The situation, properly posed, is something like the
generalized plane strain concept, where free expansion is allowed in certain directions.
In our case, the two reflective planes must remain planar, and parallel to their original
position.
CHAPTER 23. CASE STUDY: HYDRAULIC VALVE 274

Figure 23.2: Three fixed reflective planes.

It was not sufficient to create a moving plane which remained parallel to its original

I
I
_ _ _ _j

Figure 23.3: Use of rigid body technique for moving reflective plane.

position. It was also necessary to add an "unresolved" force, namely the force that
the mirror image body would have exerted on the real body. This total force was
determined by the projection of the pressure upon the image bore hole (sketched in
Fig. 23.3). The two forces so resolved are oriented as shown in Fig. 23.4.

Calculation of the force in the y direction is not so clear. Here, the pressure load is
eventually resolved at a far wall (a return oil duct), which is not modeled by the finite
element mesh. The need for this force is argued only by general logic. The modeling
ideas used on this body are considered to be advanced, in terms of general practice.

The material properties for the casting are:

Gray cast iron:

E 18,000, OOO.psi
v 0.3
CHAPTER 23. CASE STUDY: HYDRAULIC VALVE 275

Figure 23.4: Fixed (top, right side) and moving (left side) reflective planes.

Malleable iron (alternate):

E 24,000, OOO.psi
v 0.3

It is also desired to consider other geometrically scaled valves, from 1/2 as large to
2 times larger.

23.4 FINITE ELEMENT MODELING


A description of the MSC /N ASTRAN solution will be given. Eighty-eight HEXA
elements with 8-20 nodes were used (Fig. 23.5). The 20-noded elements were used in
the narrow sections at the top of the model where the fatigue failure had occurred.
The 8-noded elements were used at the thick bottom of the valve, where low stresses
were expected.

The 3 fixed reflective planes were constrained by using single point constraints on
the relevant nodes. This is easily accommodated in any finite element program, either
appearing on the node definition statement or as a separate constraint list.
The two moving reflective planes were modeled with 2 RBE2 rigid body elements.
The degrees of freedom on the reflected plane were constrained in 5 of the 6 components,
and the x translation was governed by the motion of one master node. The forces were
applied with FORCE statements, on the master node.

The FE model had 1364 d.o.f. and 54 7 nodes. The commercial cost for running the
problem was $175. With today's faster computers, the CPU expense would be half of
that.
CHAPTER 23. CASE STUDY: HYDRAULIC VALVE 276

Figure 23 .5: Mesh for hydraulic valve segment.

z
X

Figure 23 .6: Side view of FE mesh.

23.5 RESULTS
The program MSC /N ASTRAN defines a parameter epsilon, which is the ratio of
unbalanced force at a node to the nominal value of the force there. This is a measure
of loss of precision. It had a value 3x10- 13 , which means that there were effectively
13 decimal figures of precision in the solution. This is excellent. A second overall
measure of the "health" of the solution is the work done by the external loads, which
was 1.09 inlb. This is small, and shows that there are no mechanisms allowing large
load motion.

The results show that the round bores became elliptical (Fig. 23.8) . The maximum
deflections on the bore diameter were found to be 0.0005 in. The maximum principal
stress was found to be 22,600 psi (Fig. 23.9). Both the displacement and the stress
were larger than expected, but there was physical measurement on which to base this
CHAPTER 23. CASE STUDY: HYDRAULIC VALVE 277

Figure 23.7: Top view of FE mesh.

Figure 23.8: Side view, showing elliptical holes.

expectation. It was hoped that the stresses were less than 12,000 psi. The internal
stresses had never been measured, because of the hot, flowing hydraulic fluid, which
would strip away the strain gage wires.

After some thought, it was realized that although the displacements were higher
than expected, the "blow-by" of fluid would not be significant because of the long
distances the fluid would need to pass to get around the plunger. Likewise, although
the stress was higher than expected, no field failures were being experienced. The
worst-case pressurization of the bore holes was probably more severe than the field
operation.

A second area of moderate stress was found at the bottom of the main pressurized
fluid channel. Figure 23.10 shows the stresses there, as predicted from neighboring
elements. The mesh is not fine enough for accurate results. The stress at the critical
node averages 11,000 psi, which is not enough to cause trouble. One of the engineers
remembered a different model which had experienced a crack in this area. It was felt
that it was not worth remeshing the model for a more accurate study of this area.
CHAPTER 23. CASE STUDY: HYDRAULIC VALVE 278

13,000 psi
\ 22,400
~

z
Y,t,..x
Figure 23.9: High-stress regions at mold line.

4750, -11600

15100, 18000
\
5820,5470
~
' 10400,
9810, 9420

Figure 23.10: Stresses at bottom of fluid channel.

23.6 TECHNICAL CONCLUSIONS


Points of highest stress were found which coincided with the location of the fatigue
crack (through the mold parting line).
Stresses were higher than expected.
Displacements were higher than expected.
Certain column-like regions should be straightened as suggested in Fig. 23.11.

23.7 MANAGEMENT CONCLUSIONS


The actual cost of the project was $6200, as opposed to the estimated cost of $5000.
The actual calendar time spent was 5 weeks, as opposed to the estimate of one month.
The finite element technology is very productive for the stress and displacement analysis
of such valves. It is difficult, however, to provide rules-of-thumb for the valve designer.
No "design handbook" followed from the analysis.
CHAPTER 23. CASE STUDY: HYDRAULIC VALVE 279

bad better best


(original)

Figure 23.11: Column-like region to be straightened.

The data base was useful for studies that followed, some 14 runs in all . Additional
runs cost as little as $450 and could be run in one day.

23.8 ADDITIONAL, LATER RESULTS


Specific topics studied later showed:

Changing to malleable iron doesn't change stresses.

Geometric scaling doesn't change stresses .

Adding material to the post helps.

Adding material to the floor of the main channel helps (5% on stress).

Adding material to the top and bottom of the housing is not cost effective.

23.9 SCALING LAWS


The hydraulic valve had a special type of loading due only to internal pressure.
Another special loading is that due to gravity. One wonders how the stresses in a body
will scale with respect to the size of the body under such load cases. Some simple
examples will help us understand the scaling of such a system.

23.9.1 Pressure Loading


Bodies that are internally pressurized have internal forces that are proportional to
areas exposed to the pressure:
Forces ex L 2
CHAPTER 23. CASE STUDY: HYDRAULIC VALVE 280

~/----{7
~~
'------------/
I 2L-~I

Figure 23.12: Internally pressurized hydraulic valve

The resulting stresses can be categorized as direct, shear and bending stresses. Direct
and shear stresses vary as:
p
a = -
A
L2
ex: -
L2
ex: 1 (not dependent on length)

Bending stress varies as:

Me
a = I
(PL)c
ex: I
2
L LL
ex: --
L4
ex: 1 (not dependent on length)

This shows that two internally pressurized bodies which are geometrically scaled
will have the same stresses. One could design a whole range of hydraulic valves that
are geometrically scaled, of the same material, and have the same stresses in all sizes.

23.9.2 Gravitational and Inertial Loading


Many bodies such as large buildings and vehicles are self loaded. They have gravity
loads that are proportional to their mass. Moving bodies also experience inertial loads
due to their acceleration, again in proportion to their mass. Consider the two airplanes,
say, shown in Figure 23.13.

The forces on the airplane scale as:

Forces ex: L 3
CHAPTER 23. CASE STUDY: HYDRAULIC VALVE 281

~;yz/)
I --2L--~
~
Figure 23.13 : Geometrically scaled aircraft

Direct and shear stresses scale as:


p
a -
A
L3
ex -
L2
ex L

Bending stresses scale as:

MC
a
I
(PL)c
ex
I
L 3 LL
ex
L4
ex L

We see that inertially loaded bodies have stresses that are proportional to their length
scale. Therefore, as larger vehicles are made, they must be made of stronger materials,
or made in more effective geometries. In conclusion, we can't have bigger elephants
until nature provides stronger bones!
Chapter 24

MULTIPOINT CONSTRAINTS

24.1 OVERVIEW
The multipoint constraint (MPC) is a linear relation between two or more global
degrees of freedom u 9 . The relation is homogeneous, that is, there is no fixed displace-
ment imposed, but rather the degrees of freedom are scaled proportionately to each
other. The capability for MPC relations is present in the N ASTRAN series of commer-
cial programs, and the discussion here will follow the development in the N ASTRAN
Theoretical Manual. [2] The concept of the MPC is very important to industry. The
notation is a bit abstract in one way: there are subscripts that play two roles. One
role is the conventional index notation, but the
other is to denote the category of the vector. This causes a little trouble at first,
but the notation is widely used. The index g stands for global coordinates, a for
analysis (retained) coordinates, m for coordinates removed by MPC relations, and c
for constraint cases. Whether the symbol is to be summed over, or merely used as a
flag can be seen by the context.

The general expression for a constraint c is:


N
L R cg Ug =0 (c=1,2, ... ,m) (24.1)
g=l

An example would be a joint between a truss member and a plate (Fig. 24.1). The
component u 1 takes the average value of u 2 and u 3 .
1 1
u1 - -u2 - -u3 = 0
2 2

One can interpret the constraint equation as the equation of a hyperplane which
passes through the origin of an N-dimensional space. Figure 24.2 shows a general
case. Such a constraint is a linear, scleronomic constraint- it is holonomic and not a
function of time [1]. Constraints of this type are "workless"- they do no work on any
displacement consistent with the constraint.

282
CHAPTER 24. MULTIPOINT CONSTRAINTS 283

Figure 24.1: Joint between truss and plate

Figure 24.2: Hyperplane in displacement space

24.2 NONWORKING CONSTRAINT FORCES


The equation of equilibrium for a typical static problem is:

[K]{u} = {P} (24.2)

where the load vector {P} includes both the live loads and the external reactions . If
one now imposes constraints c on the system, there can be m (say) additional forces
of constraint:
m

[K]{ U} = {P} external + L {qc} (24.3)


c=l

Each such constraint force {qc} does no work:

(c= 1, 2, ... , m) (24.4)


CHAPTER 24. MULTIP OINT CONSTRAINTS 284

and therefore { qe} must act perpendicular to a hyperplane in the displacement space
(Fig. 24.3) . 1

Figure 24.3: Nonworking constraint force

24.3 MATHEMATICAL RELATIONS


Consider the physical example given above in Fig. 24.1, and follow each specific
calculation with the expression valid for a general constraint . We will denote the
coordinates which are to be removed as {urn} and the coordinates to be retained as
{un} In addition, the coefficient Reg which is attached to the degree of freedom to
be removed will be given a value unity- this causes no loss of generality in the linear
constraint equation.

The constraint equation

(24.5)

can be solved for u 1 (or the "home" coordinate ue) in terms of the other coordinates:

1 1
2u2 + 2u3
- L Re9 u 9 (24.6)
g:f=e
1
A constraint involves a limited number of displacement variables and the constraint force must
be normal to the hyperplane defined in that subspace.
CHAPTER 24. MULTIP OINT CONSTRAINTS 285

One imposes the condition of no work by the constraint' forces:

c(1 1 ) c c
q1 2u2 + 2u3 + q2u2 + q3u3 0

q~ (- L Rc u 9 9) + L q~u 9 0 (c=1,2, .. ,m) (24. 7)


g=f= c g=f= c

Gathering terms:
1 1
[q12 + q2]u2 + [q12 + q3]u3 = 0
[-q~Rcl + q~]u1 + .. + [-qcR c,c-1 + q~-1]u c-1 + [-q~Rc,c+1 + q~+1]u c+1
+ ... + [-qcR cN + qN]uN = 0 (c = 1, 2, ... ,m) (24.8)
Since the equation must be true for all un, each bracketed term must be separately zero.
This allows one to solve for all components of each force of constraint in terms of the
"force of constraint" q~ acting on the "home" degree of freedom Uc for the constraint :

1
q3 = --q1
2
(g = 1, 2, ... N) (c= 1,2, ... ,m)

This defines the components of each constraint in terms of a single parameter- a ho-
mogeneous relation among force components . For our example (Fig. 24.1), the sum of
the constraint forces is zero.

24.4 MATRIX FORM OF MPC RELATIONS


We need to cast the MPC relations into matrix form for total solution. The dis-
placements in the body can be partioned into Un that are to be retained and Urn that
are to be eliminated:

The matrix of constraint coefficients [R] is also partitioned:

The constraint equation becomes:

(24.9)

The matrix [Rm] must be nonsingular (since the constraints should be independent
conditions) , so that we can solve for { un}:

(24.1 0)
CHAPTER 24. MULTIPOINT CONSTRAINTS 286

where

In a like way, we need to relate the physical components of the constraint force
acting on the n degrees of freedom to those on the m degrees of freedom. The total
constraint force vector { q} can be broken into forces acting on the Um set, { qm} and
the forces on the Un set, {qn } The condition that no work be done (Eq. 24.4) is:

lun umJ { ::} 0 (24.11)

or lUnJ { qn} + lUmJ {Un} 0 (24.12)

We substitute the known displacements on the m degrees of freedom from Eq. 24.10
to get:

(24.13)
Collecting terms, one has:

(24.14)

Since the remaining degrees of freedom must be arbitrary (dependent on live loading)
the vector of constraint forces must vanish separately, yielding:

(24.15)

We use all of the relationships available to us to pose the larger problem:

(24.16)

where the barred quantities are physical stiffnesses and forces which will be eliminated.
One eliminates Um and qm to get

where

f{nn

Pn

The I<nn and P n are the generalized stiffness and force , respectively, acting on the
degrees of freedom Un which remain after the Um set has been removed.
CHAPTER 24. MULTIP OINT CONSTRAINTS 287

24.5 SAMPLE PROBLEM AND SOLUTION


24.5.1 Problem Statement
Consider the matrix equation:

A) Apply a multipoint constraint:

B) Apply a singlepoint constraint:

C) Apply a standard static condensation where u 1 is eliminated in terms of the


remaining coordinates. Express your answer as a solution for the analysis degree of
freedom Ua

D) Recover the solution for the global d . o. f. {u 9 }.

E) Recover the reaction forces and forces of constraint .

24.5.2 Solution
A) The MPC relation is :

[0 -1 0 1] g~ )
= { 0}

Hence ,

[0 -1 OJ
[1J

-[1J [0 -1 OJ
[0 1 0]
1
[1 0 ~]
2
CHAPTER 24. MULTIPOINT CONSTRAINTS 288

[f{nm] -
[~]
[Kmm] = [1]

~]+[~][o OJ+[~] [1 0 1J+[~] [1 ][ 0 1 0J


1
[Knn] = [l 2
0
1

[~ !]
2
= 3
1

{Pn} = {!}+[~]{2}
-
{0
The reduced set of equations is:

B) The reduction due to a SPC of u 3 = 0 is easy:

C) Do a standard static condensation (Gauss elimination in preferred order):

Solve for U0 from the 1st partitioned set of equations:

[J<oo]{ Uo} + [J<oa]{ Ua} = {Po}


{uo} = [I<oo]- 1 {Po}- [Koo]- 1 [1<oa]{ua}
Substitute in the 2nd partitioned set of equations to get:
CHAPTER 24. MULTIPOINT CONSTRAINTS 289

[Kaa] {Pa}
In our case

[Kaa] - [3] - [2][~][2]


= [2]
{Pa} - {5}- [2][~]{5}
= {~}
2
Hence:

and
{ u2} = 1.25
D) Recover all displacements { u 9 }:

The omitted coordinate { U0 } = {u1}.

{Uo} = [Go]{ Ua} + [Koo]- 1{Po}


- -[Koo]- 1 [Loa]{ua} + [Koo]- 1{Po}
1 1
= - ( 4) (2) ( 1. 25) + (4) (5)
{ u1} - 0.625

Hence:
{UJ } -= { u1
u2
} = { 0.625 }
1.250
Of course,

Hence:

{Un} = { ~~
u3
} = { ~:~~~
0.0
}

The multipoint constraint removed { u 4 }, which is now recovered:


CHAPTER 24. MULTIPOINT CONSTRAINTS 290

and finally the global displacement vector is:

24.6 HOMEWORK
Problem 1.
Two plates have been modeled in finite elements as shown. The system of equations
has been "condensed" to the remaining two degrees of freedom u 78 and u 99 . (This can
be done by a normal process of Gauss elimination, in which the other degrees of freedom
are eliminated.) The resulting set of equations in only those two variables is:

2
[ -1 -1] { U7g } = { 3 } (24.17)
2 Ugg 1

A weld is now added to force the two plates to move together in the vertical direction.
This is to be modeled by a multipoint constraint between the two remaining degrees
of freedom. Of course, physically, the constraint adds a pair of forces between the two
points. Solve for the final displacements and forces of constraint for the system.

Finite element model Simple model

Figure 24.4: Two plates to be welded, with reduced, spring model


Bibliography

[1] Greenwood, D. T., Principles of Dynamics, Prentice-Hall, Inc., 1965, pp. 232.

[2] MacNeal, R. H., (Editor), "The Nastran Theoretical Manual," The MacNeal-
Schwendler Corporation, Los Angeles, 1972 (pp . 3.5-1 to 3.5-4, and 5.4-1 to 5.4-4).

291

Anda mungkin juga menyukai