LS-DYNA Theory Manual (Updated 13dec2016) PDF

Download as pdf or txt
Download as pdf or txt
You are on page 1of 872

LS-DYNA®

Theory Manual

10/27/16 (r:8004)
LS-DYNA Dev

LIVERMORE SOFTWARE TECHNOLOGY CORPORATION (LSTC)


Corporate Address
Livermore Software Technology Corporation
P. O. Box 712
Livermore, California 94551-0712
Support Addresses
Livermore Software Technology Corporation Livermore Software Technology Corporation
7374 Las Positas Road 1740 West Big Beaver Road
Livermore, California 94551 Suite 100
Tel: 925-449-2500 Fax: 925-449-2507 Troy, Michigan 48084
Email: [email protected] Tel: 248-649-4728 Fax: 248-649-6328
Website: www.lstc.com
Disclaimer
Copyright © 1992-2016 Livermore Software Technology Corporation. All Rights Reserved.
LS-DYNA®, LS-OPT® and LS-PrePost® are registered trademarks of Livermore Software Technology
Corporation in the United States. All other trademarks, product names and brand names belong to their
respective owners.
LSTC products are protected under the following US patents: 7953578, 7945432, 7702494, 7702490, 7664623, 7660480,
7657394, 7640146, 7613585, 7590514, 7533577, 7516053, 7499050, 7472602, 7428713, 7415400, 7395128, 7392163, 7386428,
7386425, 7382367, 7308387, 7286972, 7167816, 8050897, 8069017, 7987143, 7996344, 8126684, 8150668, 8165856, 8180605,
8190408, 8200464, 8296109, 8209157, 8271237, 8200458, 8306793, Japan patent 5090426 and pending patents applications.

LSTC reserves the right to modify the material contained within this manual without prior notice.
The information and examples included herein are for illustrative purposes only and are not intended to be
exhaustive or all-inclusive. LSTC assumes no liability or responsibility whatsoever for any direct of indirect
damages or inaccuracies of any type or nature that could be deemed to have resulted from the use of this
manual.

Any reproduction, in whole or in part, of this manual is prohibited without the prior written approval of
LSTC. All requests to reproduce the contents hereof should be sent to [email protected].

AES. Copyright © 2001, Dr Brian Gladman < [email protected]>, Worcester, UK. All rights reserved.
LICENSE TERMS
The free distribution and use of this software in both source and binary form is allowed (with or without
changes) provided that:

1. distributions of this source code include the above copyright notice, this list of conditions and the
following disclaimer;

2. distributions in binary form include the above copyright notice, this list of conditions and the
following disclaimer in the documentation and/or other associated materials;

3. the copyright holder's name is not used to endorse products built using this software without
specific written permission.
DISCLAIMER
This software is provided 'as is' with no explicit or implied warranties in respect of any properties, including,
but not limited to, correctness and fitness for purpose.
Issue Date: 21/01/2002
This file contains the code for implementing the key schedule for AES (Rijndael) for block and key sizes of 16,
24, and 32 bytes.
LS-DYNA Theory Manual Table of Contents

Table of Contents

Table of Contents....................................................................................................................................... 0-1

1 Abstract ................................................................................................................................................... 1-1

2 History of LS-DYNA ................................................................................................................................ 2-1


2.1 Features add in 1989-1990 ..................................................................................................... 2-3
2.2 Options added in 1991-1992 .................................................................................................. 2-4
2.3 Options added in 1993-1994 .................................................................................................. 2-5
2.4 Version 936 ............................................................................................................................. 2-6
2.5 Version 940 ............................................................................................................................. 2-8
2.6 Version 950 ........................................................................................................................... 2-10
2.7 Version 960 ........................................................................................................................... 2-13
2.8 Version 970 ........................................................................................................................... 2-18

3 Preliminaries ............................................................................................................................................ 3-1


3.1 Governing Equations .............................................................................................................. 3-1

4 Solid Elements ........................................................................................................................................ 4-1


4.1 Volume Integration .................................................................................................................. 4-3
4.2 Solid Element 2 ....................................................................................................................... 4-6
4.2.1 Brief summary of solid element 2 ........................................................................... 4-6
4.2.2 Transverse shear locking example ......................................................................... 4-7
4.2.3 Solid element -2 ...................................................................................................... 4-8
4.2.4 Transverse shear locking example revisited........................................................... 4-9
4.2.5 Solid element -1 ...................................................................................................... 4-9
4.2.6 Example ................................................................................................................ 4-10
4.3 Hourglass Control ................................................................................................................. 4-10
4.4 Puso Hourglass Control ........................................................................................................ 4-16
4.5 Fully Integrated Brick Elements and Mid-Step Strain Evaluation......................................... 4-21
4.6 Four Node Tetrahedron Element .......................................................................................... 4-23
4.7 Nodal Pressure Tetrahedron ................................................................................................. 4-24
4.8 Six Node Pentahedron Element............................................................................................ 4-25
4.9 Fully Integrated Brick Element With 48 Degrees-of-Freedom ............................................. 4-26
4.10 Fully Integrated Tetrahedron Element With 24 Degrees-of-Freedom ................................ 4-29

LS-DYNA DEV 10/27/16 (r:8004) 0-1 (Table of Contents)


Table of Contents LS-DYNA Theory Manual

4.11 The Cosserat Point Elements in LS-DYNA ......................................................................... 4-32


4.11.1 Introduction ......................................................................................................... 4-32
4.11.2 Geometry ............................................................................................................ 4-33
4.11.3 Deformation and strain ....................................................................................... 4-33
4.11.4 Stress and Force................................................................................................. 4-34
4.11.5 CPE3D10 modification........................................................................................ 4-35
4.11.6 Hourglass ............................................................................................................ 4-36
4.11.7 Comparison to Jabareen & Rubin ...................................................................... 4-37
4.11.8 Nodal formulation ............................................................................................... 4-37
4.11.9 Cantilever Beam Example................................................................................... 4-38
4.11.10 Compression Test for Rubber Example ........................................................... 4-39
4.11.11 References ........................................................................................................ 4-39
4.12 Integral Difference Scheme as Basis For 2D Solids ........................................................... 4-40
4.12.1 Rezoning With 2D Solid Elements ...................................................................... 4-45

5 Cohesive elements ................................................................................................................................ 4-50


5.1 Kinematics ............................................................................................................................ 4-51
5.2 Constitutive law..................................................................................................................... 4-53
5.3 Nodal forces .......................................................................................................................... 4-53
5.4 Drilling constraint in shell 29 ................................................................................................. 4-55
5.5 Rotational masses in shell 29 ............................................................................................... 4-56
5.6 Stiffness matrix in shell 29 .................................................................................................... 4-57

6 Belytschko Beam .................................................................................................................................... 5-1


6.1 Co-rotational Technique ......................................................................................................... 5-1
6.2 Belytschko Beam Element Formulation.................................................................................. 5-5
6.2.1 Calculation of Deformations ................................................................................... 5-5
6.2.2 Calculation of Internal Forces ................................................................................. 5-6
6.2.3 Updating the Body Coordinate Unit Vectors .......................................................... 5-8

7 Hughes-Liu Beam ................................................................................................................................... 6-1


7.1 Geometry ................................................................................................................................ 6-1
7.2 Fiber Coordinate System ........................................................................................................ 6-5
7.2.1 Local Coordinate System ....................................................................................... 6-6
7.3 Strains and Stress Update ...................................................................................................... 6-7
7.3.1 Incremental Strain and Spin Tensors ..................................................................... 6-7
7.3.2 Stress Update ......................................................................................................... 6-7
7.3.3 Incremental Strain-Displacement Relations ........................................................... 6-8
7.3.4 Spatial Integration ................................................................................................... 6-9
0-2 (Table of Contents) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Table of Contents

8 Warped Beam Elements ......................................................................................................................... 7-1


8.1 Resultant Warped Beam ......................................................................................................... 7-1
8.1.1 Green-Lagrange Strains in Terms of Deformational Displacements ...................... 7-1
8.1.2 Deformational Displacements After Large Rotations ............................................. 7-1
8.1.3 Green-Lagrange Strains in terms of Centroidal Displacements and Angular
Rotations .......................................................................................................................... 7-2
8.1.4 Strain Energy........................................................................................................... 7-3

9 Belytschko-Lin-Tsay Shell ...................................................................................................................... 8-1


9.1 Co-rotational Coordinates ...................................................................................................... 8-1
9.2 Velocity-Strain Displacement Relations.................................................................................. 8-3
9.3 Stress Resultants and Nodal Forces ...................................................................................... 8-5
9.4 Hourglass Control (Belytschko-Lin-Tsay) ............................................................................... 8-6
9.5 Hourglass Control (Englemann and Whirley) .......................................................................... 8-7
9.6 Belytschko-Wong-Chiang Improvements .............................................................................. 8-9

10 Triangular Shells.................................................................................................................................... 9-1


10.1 C0 Triangular Shell ................................................................................................................ 9-1
10.1.1 Co-rotational Coordinates .................................................................................... 9-1
10.1.2 Velocity-Strain Relations....................................................................................... 9-2
10.1.3 Stress Resultants and Nodal Forces .................................................................... 9-6
10.2 Marchertas-Belytschko Triangular Shell............................................................................... 9-7
10.2.1 Element Coordinates ............................................................................................ 9-7
10.2.2 Displacement Interpolation ................................................................................. 9-10
10.2.3 Strain-Displacement Relations ........................................................................... 9-11
10.2.4 Nodal Force Calculations ................................................................................... 9-12

11 Fully Integrated Shell (Type 16)........................................................................................................... 10-1


11.1 Introduction ......................................................................................................................... 10-1
11.2 Hu-Washizu Three Field Principle....................................................................................... 10-1
11.3 In-plane Assumed Strain Field ............................................................................................ 10-2
11.4 Transverse Shear Assumed Strain Field ............................................................................. 10-3
11.5 Rigid Body Motion .............................................................................................................. 10-4
11.6 Belytschko-Leviathan Projection ........................................................................................ 10-5

12 Shells with Thickness Stretch ............................................................................................................. 10-6


12.1 Introduction ......................................................................................................................... 10-6
12.2 Shell Type 25 ...................................................................................................................... 10-7
12.2.1 Formulation ......................................................................................................... 10-7

LS-DYNA DEV 10/27/16 (r:8004) 0-3 (Table of Contents)


Table of Contents LS-DYNA Theory Manual

12.2.2 Hourglass modes ................................................................................................ 10-8


12.3 Shell Type 26 ...................................................................................................................... 10-9
12.3.1 Modifying shell type 16 ....................................................................................... 10-9
12.4 Shell Type 27 .................................................................................................................... 10-10
12.4.1 Modifying shell type 4 ....................................................................................... 10-10
12.5 Related Features ............................................................................................................... 10-10
12.5.1 Continuous vs decoupled thickness field ......................................................... 10-10
12.5.2 Nodal masses ................................................................................................... 10-11
12.5.3 Transfer of contact forces................................................................................. 10-11
12.6 Contact Pressure Treatment in Shells 2, 4 and 16 ........................................................... 10-12

13 Hughes-Liu Shell ................................................................................................................................. 11-1


13.1 Geometry ............................................................................................................................ 11-2
13.2 Kinematics .......................................................................................................................... 11-4
13.2.1 Fiber Coordinate System .................................................................................... 11-5
13.2.2 Lamina Coordinate System ................................................................................ 11-7
13.3 Strains and Stress Update .................................................................................................. 11-9
13.3.1 Incremental Strain and Spin Tensors ................................................................. 11-9
13.3.2 Stress Update ..................................................................................................... 11-9
13.3.3 Incremental Strain-Displacement Relations ..................................................... 11-10
13.4 Element Mass Matrix ........................................................................................................ 11-11
13.5 Accounting for Thickness Changes .................................................................................. 11-12
13.6 Fully Integrated Hughes-Liu Shells ................................................................................... 11-12

14 Transverse Shear Treatment For Layered Shell ................................................................................. 12-1

15 Eight-Node Solid Shell Element .......................................................................................................... 13-1

16 Eight-Node Solid Element for Thick Shell Simulations ....................................................................... 14-1


16.1 Abstract ............................................................................................................................... 14-1
16.2 Introduction ......................................................................................................................... 14-1
16.3 Element Formulations ......................................................................................................... 14-2
16.3.1 Strain Operator ................................................................................................... 14-2
16.3.2 Co-rotational Coordinate System ....................................................................... 14-7
16.3.3 Stress and Strain Measures................................................................................ 14-8
16.3.4 Co-rotational Stress and Strain Updates ........................................................... 14-9
Tangent Stiffness Matrix and Nodal Force Vectors..................................................... 14-12
16.4 Numerical Examples ......................................................................................................... 14-15
16.4.1 Timoshenko Cantilever Beam ........................................................................... 14-15

0-4 (Table of Contents) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Table of Contents

16.4.2 Pinched Cylinder ............................................................................................... 14-18


16.4.3 Scordelis-Lo Roof ............................................................................................. 14-18
16.4.4 Circular Sheet Stretched with a Tight Die ........................................................ 14-20
16.5 Conclusions ...................................................................................................................... 14-23

17 Truss Element ..................................................................................................................................... 15-1

18 Membrane Element ............................................................................................................................. 16-1


18.1 Co-rotational Coordinates .................................................................................................. 16-1
18.2 Velocity-Strain Displacement Relations.............................................................................. 16-2
18.3 Stress Resultants and Nodal Forces .................................................................................. 16-2
18.4 Membrane Hourglass Control............................................................................................. 16-3

19 Discrete Elements and Masses .......................................................................................................... 17-1


19.1 Orientation Vectors ............................................................................................................. 17-2
19.2 Dynamic Magnification “Strain Rate” Effects ..................................................................... 17-4
19.3 Deflection Limits in Tension and Compression .................................................................. 17-4
19.4 Linear Elastic or Linear Viscous .......................................................................................... 17-5
19.5 Nonlinear Elastic or Nonlinear Viscous ............................................................................... 17-5
19.6 Elasto-Plastic with Isotropic Hardening ............................................................................. 17-6
19.7 General Nonlinear ............................................................................................................... 17-6
19.8 Linear Visco-Elastic ............................................................................................................ 17-7
19.9 Muscle Model...................................................................................................................... 17-9
19.10 Seat Belt Material............................................................................................................ 17-12
19.11 Seat Belt Elements.......................................................................................................... 17-13
19.12 Sliprings .......................................................................................................................... 17-14
19.13 Retractors ....................................................................................................................... 17-15
19.14 Sensors ........................................................................................................................... 17-18
19.15 Pretensioners .................................................................................................................. 17-19
19.16 Accelerometers ............................................................................................................... 17-19

20 Simplified Arbitrary Lagrangian-Eulerian ............................................................................................ 18-1


20.1 Mesh Smoothing Algorithms .............................................................................................. 18-3
20.1.1 Equipotential Smoothing of Interior Nodes ........................................................ 18-3
20.1.2 Simple Averaging .............................................................................................. 18-11
20.1.3 Kikuchi’s Algorithm ........................................................................................... 18-11
20.1.4 Surface Smoothing ........................................................................................... 18-11
20.1.5 Combining Smoothing Algorithms.................................................................... 18-11
20.2 Advection Algorithms ........................................................................................................ 18-12

LS-DYNA DEV 10/27/16 (r:8004) 0-5 (Table of Contents)


Table of Contents LS-DYNA Theory Manual

20.2.1 Advection Methods in One Dimension ............................................................. 18-12


20.2.2 Advection Methods in Three Dimensions ......................................................... 18-16
20.3 The Manual Rezone .......................................................................................................... 18-23

21 Stress Update Overview ..................................................................................................................... 19-1


21.1 Jaumann Stress Rate.......................................................................................................... 19-1
21.2 Jaumann Stress Rate Used With Equations of State ......................................................... 19-2
21.3 Green-Naghdi Stress Rate .................................................................................................. 19-3
21.4 Elastoplastic Materials ........................................................................................................ 19-5
21.5 Hyperelastic Materials ........................................................................................................ 19-8
21.6 Layered Composites ......................................................................................................... 19-10
21.7 Constraints on Orthotropic Elastic Constants .................................................................. 19-12
21.8 Local Material Coordinate Systems in Solid Elements..................................................... 19-13
21.9 General Erosion Criteria for Solid Elements ..................................................................... 19-14
21.10 Strain Output to the LS-DYNA Database ....................................................................... 19-16
21.12 Strain Rate Effects in Material Models ........................................................................... 19-18
21.12.1 Running N-average option .............................................................................. 19-18
21.12.2 Last N-average option .................................................................................... 19-18
21.12.3 Averaging over a fixed amount of time 𝑻 ....................................................... 19-19

22 Material Models ................................................................................................................................ 20-20


22.1 Material Model 1: Elastic.................................................................................................. 20-34
22.2 Material Model 2: Orthotropic Elastic .............................................................................. 20-35
22.3 Material Model 3: Elastic Plastic with Kinematic Hardening ........................................... 20-37
22.4 Material Model 4: Thermo-Elastic-Plastic ....................................................................... 20-41
22.5 Material Model 5: Soil and Crushable Foam ................................................................... 20-43
22.6 Material Model 6: Viscoelastic ......................................................................................... 20-45
22.7 Material Model 7: Continuum Rubber ............................................................................. 20-46
22.8 Material Model 8: Explosive Burn .................................................................................... 20-47
22.9 Material Model 9: Null Material ........................................................................................ 20-50
22.10 Material Model 10: Elastic-Plastic-Hydrodynamic ........................................................ 20-51
22.11 Material Model 11: Elastic-Plastic With Thermal Softening .......................................... 20-54
22.12 Material Model 12: Isotropic Elastic-Plastic .................................................................. 20-56
22.13 Material Model 13: Isotropic Elastic-Plastic with Failure............................................... 20-57
22.14 Material Model 14: Soil and Crushable Foam With Failure ........................................... 20-58
22.15 Material Model 15: Johnson and Cook Plasticity Model ............................................... 20-59
22.16 Material Model 16: Pseudo Tensor ................................................................................ 20-61
22.17 Material Model 17: Isotropic Elastic-Plastic With Oriented Cracks .............................. 20-65

0-6 (Table of Contents) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Table of Contents

22.18 Material Model 18: Power Law Isotropic Plasticity ....................................................... 20-67
22.19 Material Model 19: Strain Rate Dependent Isotropic Plasticity..................................... 20-68
22.20 Material Model 20: Rigid ................................................................................................ 20-70
22.21 Material Model 21: Thermal Orthotropic Elastic ............................................................ 20-71
22.22 Material Model 22: Chang-Chang Composite Failure Model ........................................ 20-73
22.23 Material Model 23: Thermal Orthotropic Elastic with 12 Curves ................................... 20-75
22.24 Material Model 24: Piecewise Linear Isotropic Plasticity .............................................. 20-77
22.25 Material Model 25: Kinematic Hardening Cap Model ................................................... 20-79
22.26 Material Model 26: Crushable Foam ............................................................................. 20-84
22.27 Material Model 27: Incompressible Mooney-Rivlin Rubber .......................................... 20-88
22.27.1 Stress Update for Shell Elements ................................................................... 20-90
22.27.2 Derivation of the Continuum Tangent Stiffness.............................................. 20-92
22.27.3 The Algorithmic Tangent Stiffness for Shell Elements ................................... 20-93
22.28 Material Model 28: Resultant Plasticity ......................................................................... 20-95
22.29 Material Model 29: FORCE LIMITED Resultant Formulation......................................... 20-97
22.29.1 Internal Forces ................................................................................................ 20-99
22.29.2 Tangent Stiffness .......................................................................................... 20-102
22.30 Material Model 30: Closed-Form Update Shell Plasticity ........................................... 20-104
22.30.1 Mathematical Description of the Material Model.......................................... 20-104
22.30.2 Algorithmic Stress Update ............................................................................ 20-105
22.30.3 Tangent Stiffness Matrix ............................................................................... 20-106
22.31 Material Model 31: Slightly Compressible Rubber Model ........................................... 20-109
22.32 Material Model 32: Laminated Glass Model ................................................................ 20-110
22.33 Material Model 33: Barlat’s Anisotropic Plasticity Model ............................................ 20-111
22.34 Material Model 34: Fabric ............................................................................................ 20-112
22.35 Material Model 35: Kinematic/Isotropic Plastic Green-Naghdi Rate .......................... 20-114
22.36 Material Model 36: Barlat’s 3-Parameter Plasticity Model .......................................... 20-115
22.36.1 Material Tangent Stiffness ............................................................................ 20-116
22.36.2 Load curves in different directions ............................................................... 20-118
22.36.2.1 An introductory remark ................................................................. 20-118
22.36.2.2 The model ..................................................................................... 20-119
22.36.3 Variable Lankford parameters ...................................................................... 20-120
22.37 Material Model 37: Transversely Anisotropic Elastic-Plastic ...................................... 20-122
22.38 Material Model 38: Blatz-Ko Compressible Foam....................................................... 20-124
22.39 Material Model 39: Transversely Anisotropic Elastic-Plastic With FLD....................... 20-125
22.42 Material Model 42: Planar Anisotropic Plasticity Model .............................................. 20-126
22.51 Material Model 51: Temperature and Rate Dependent Plasticity ............................... 20-127
LS-DYNA DEV 10/27/16 (r:8004) 0-7 (Table of Contents)
Table of Contents LS-DYNA Theory Manual

22.52 Material Model 52: Sandia’s Damage Model .............................................................. 20-129


22.53 Material Model 53: Low Density Closed Cell Polyurethane Foam .............................. 20-130
22.54 Material Models 54 and 55: Enhanced Composite Damage Model............................ 20-132
22.57 Material Model 57: Low Density Urethane Foam ........................................................ 20-135
22.58 Material Model 58: Laminated Composite Fabric ....................................................... 20-137
22.60 Material Model 60: Elastic With Viscosity .................................................................... 20-139
22.61 Material Model 61: Maxwell/Kelvin Viscoelastic with Maximum Strain ...................... 20-140
22.62 Material Model 62: Viscous Foam ............................................................................... 20-141
22.63 Material Model 63: Crushable Foam ........................................................................... 20-142
22.64 Material Model 64: Strain Rate Sensitive Power-Law Plasticity.................................. 20-144
22.65 Material Model 65: Modified Zerilli/Armstrong ............................................................ 20-145
22.66 Material Model 66: Linear Stiffness/Linear Viscous 3D Discrete Beam ...................... 20-146
22.67 Material Model 67: Nonlinear Stiffness/Viscous 3D Discrete Beam ........................... 20-147
22.68 Material Model 68: Nonlinear Plastic/Linear Viscous 3D Discrete Beam .................... 20-148
22.69 Material Model 69: Side Impact Dummy Damper (SID Damper)................................. 20-149
22.70 Material Model 70: Hydraulic/Gas Damper ................................................................. 20-151
22.71 Material Model 71: Cable ............................................................................................. 20-152
22.73 Material Model 73: Low Density Viscoelastic Foam .................................................... 20-153
22.74 Material Model 74: Elastic Spring for the Discrete Beam............................................ 20-154
22.75 Material Model 75: Bilkhu/Dubois Foam Model .......................................................... 20-155
22.76 Material Model 76: General Viscoelastic ..................................................................... 20-156
22.76.1 Introduction ................................................................................................... 20-156
22.76.2 Constitutive Model ........................................................................................ 20-156
22.76.3 Tangent Modulus .......................................................................................... 20-156
22.76.4 Using the Relaxation Curve .......................................................................... 20-157
22.77 Material Model 77: Hyperviscoelastic Rubber............................................................. 20-159
22.77.1 General Hyperelastic Rubber Model ............................................................ 20-160
22.77.2 Ogden Rubber Model ................................................................................... 20-161
22.77.3 The Viscoelastic Contribution ....................................................................... 20-162
22.77.4 Stress Update for Shell Elements ................................................................. 20-162
22.78 Material Model 78: Soil/Concrete ................................................................................ 20-165
22.79 Material Model 79: Hysteretic Soil ............................................................................... 20-167
22.80 Material Model 80: Ramberg-Osgood Plasticity ......................................................... 20-168
22.81 Material Models 81 and 82: Plasticity with Damage and Orthotropic Option ............. 20-169
22.81.1 Material Model 82: Isotropic Elastic-Plastic with Anisotropic Damage ...... 20-171
22.83 Material Model 83: Fu-Chang’s Foam With Rate Effects ............................................ 20-174
22.84 Material Model 84 and 85: Winfrith Concrete .............................................................. 20-177
0-8 (Table of Contents) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Table of Contents

22.87 Material Model 87: Cellular Rubber ............................................................................. 20-178


22.88 Material Model 88: MTS Model ................................................................................... 20-180
22.89 Material Model 89: Plasticity Polymer ......................................................................... 20-182
22.90 Material Model 90: Acoustic ........................................................................................ 20-183
22.91 Material Model 91: Soft Tissue .................................................................................... 20-184
22.94 Material Model 94: Inelastic Spring Discrete Beam .................................................... 20-185
22.96 Material Model 96: Brittle Damage Model ................................................................... 20-186
22.97 Material Model 97: General Joint Discrete Beam ........................................................ 20-188
22.98 Material Model 98: Simplified Johnson Cook .............................................................. 20-189
22.100 Material Model 100: Spot Weld ................................................................................. 20-190
22.101 Material Model 101: GE Thermoplastics ................................................................... 20-193
22.102 Material Model 102: Hyperbolic Sine......................................................................... 20-194
22.103 Material Model 103: Anisotropic Viscoplastic ........................................................... 20-195
22.104 Material Model 104: Continuum Damage Mechanics Model .................................... 20-196
22.106 Material Model 106: Elastic Viscoplastic Thermal ..................................................... 20-199
22.110 Material Model 110: Johnson-Holmquist Ceramic Model......................................... 20-200
22.111 Material Model 111: Johnson-Holmquist Concrete Model ....................................... 20-202
22.115 Material Model 115: Elastic Creep Model ................................................................. 20-203
22.116 Material Model 116: Composite Layup ..................................................................... 20-204
22.117 Material Model 117-118: Composite Matrix .............................................................. 20-205
22.119 Material Model 119: General Nonlinear 6DOF Discrete Beam .................................. 20-206
22.120 Material Model 120: Gurson ...................................................................................... 20-207
22.120 Material Model 120: Gurson RCDC ........................................................................... 20-208
22.124 Material Model 124: Tension-Compression Plasticity ............................................... 20-209
22.126 Material Model 126: Metallic Honeycomb ................................................................. 20-210
22.126.1 Stress Update ............................................................................................. 20-212
22.126.2 Support for Independent Shear and Hydrostatic Yield Stress Limit .......... 20-213
22.127 Material Model 127: Arruda-Boyce Hyperviscoelastic Rubber ................................. 20-216
22.128 Material Model 128: Heart Tissue .............................................................................. 20-218
22.129 Material Model 129: Isotropic Lung Tissue ............................................................... 20-219
22.130 Material Model 130: Special Orthotropic ................................................................... 20-221
22.131 Material Model 131: Isotropic Smeared Crack .......................................................... 20-222
22.133 Material Model 133: Barlat_YLD2000 ........................................................................ 20-224
22.133.1 Closest point projection algorithm ............................................................. 20-227
22.134 Material Model 134: Viscoelastic Fabric .................................................................... 20-229
22.139 Material Model 139: Modified Force Limited ............................................................. 20-230
22.141 Material Model 141: Rate Sensitive Polymer ............................................................. 20-233
LS-DYNA DEV 10/27/16 (r:8004) 0-9 (Table of Contents)
Table of Contents LS-DYNA Theory Manual

22.142 Material Model 142: Transversely Anisotropic Crushable Foam ................................ 20-234
22.143 Material Model 143: Wood Model ............................................................................. 20-236
22.144 Material Model 144: Pitzer Crushable Foam ............................................................. 20-243
22.147 Material Model 147: FHWA Soil Model...................................................................... 20-244
22.154 Material Model 154: Deshpande-Fleck Foam ........................................................... 20-251
22.156 Material Model 156: Muscle ...................................................................................... 20-252
22.158 Material Model 158: Rate Sensitive Composite Fabric ............................................. 20-253
22.159 Material Model 159: Continuous Surface Cap Model ............................................... 20-254
22.161 Material Models 161 and 162: Composite MSC ....................................................... 20-260
22.163 Material Model 163: Modified Crushable Foam ........................................................ 20-266
22.164 Material Model 164: Brain Linear Viscoelastic........................................................... 20-267
22.166 Material Model 166: Moment Curvature Beam ......................................................... 20-268
22.169 Material Model 169: Arup Adhesive........................................................................... 20-270
22.170 Material Model 170: Resultant Anisotropic ................................................................ 20-271
22.175 Material Model 175: Viscoelastic Maxwell................................................................. 20-272
22.176 Material Model 176: Quasilinear Viscoelastic ............................................................ 20-273
22.177 Material Models 177 and 178: Hill Foam and Viscoelastic Hill Foam ....................... 20-274
22.177.1 Hyperelasticity Using the Principal Stretch Ratios ..................................... 20-274
22.177.2 Hill’s Strain Energy Function....................................................................... 20-275
22.177.3 Viscous Stress ............................................................................................ 20-277
22.177.4 Viscoelastic Stress Contribution................................................................. 20-277
22.177.5 Material Tangent Modulus for the Fully Integrated Brick ........................... 20-278
22.179 Material Models 179 and 180: Low Density Synthetic Foam .................................... 20-280
22.179.1 Hyperelasticity Using the Principal Stretch Ratios ..................................... 20-280
22.179.2 Strain Energy Function ............................................................................... 20-281
22.179.3 Modeling of the Hysteresis ......................................................................... 20-282
22.179.4 Viscous Stress ............................................................................................ 20-282
22.179.5 Viscoelastic Stress Contribution................................................................. 20-283
22.179.6 Stress Corresponding to First Load Cycle ................................................. 20-283
22.181 Material Model 181: Simplified Rubber/Foam ........................................................... 20-285
22.181.1 Hyperelasticity Using the Principal Stretch Ratios ..................................... 20-285
22.181.1.1 Determination of f, rubber option ............................................... 20-286
22.181.1.2 Determination of f, foam option .................................................. 20-287
22.181.2 Some Remarks ........................................................................................... 20-288
22.181.2.1 Strain rates .................................................................................. 20-288
22.181.2.2 Modeling of the Frequency Independent Damping .................... 20-288
22.187 Material Model 187: Semi-Analytical Model for the Simulation of Polymers ............ 20-289
0-10 (Table of Contents) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Table of Contents

22.187.1 Material law formulation.............................................................................. 20-289


22.187.2 Hardening formulation ................................................................................ 20-293
22.187.3 Rate effects ................................................................................................. 20-294
22.187.4 Damage and failure ..................................................................................... 20-294
22.196 Material Model 196: General Spring Discrete Beam ................................................. 20-296

23 Equation of State Models ................................................................................................................... 21-1


23.1 Equation of State Form 1: Linear Polynomial .................................................................... 21-1
23.2 Equation of State Form 2: JWL High Explosive................................................................. 21-2
23.3 Equation of State Form 3: Sack “Tuesday” High Explosives ............................................ 21-2
23.4 Equation of State Form 4: Gruneisen ................................................................................ 21-3
23.5 Equation of State Form 5: Ratio of Polynomials................................................................ 21-3
23.6 Equation of State Form 6: Linear With Energy Deposition ................................................ 21-4
23.7 Equation of State Form 7: Ignition and Growth Model ...................................................... 21-4
23.8 Equation of State Form 8: Tabulated Compaction ............................................................ 21-5
23.9 Equation of State Form 9: Tabulated................................................................................. 21-5
23.10 Equation of State Form 10: Propellant-Deflagration ....................................................... 21-6

24 Artificial Bulk Viscosity ........................................................................................................................ 22-1


24.1 Shock Waves ...................................................................................................................... 22-1
24.2 Bulk Viscosity ...................................................................................................................... 22-3

25 Time Step Control ............................................................................................................................... 23-1


25.1 Time Step Calculations for Solid Elements ........................................................................ 23-1
25.2 Time Step Calculations for Beam and Truss Elements ...................................................... 23-2
25.3 Time Step Calculations for Shell Elements......................................................................... 23-3
25.4 Time Step Calculations for Solid Shell Elements ............................................................... 23-4
25.5 Time Step Calculations for Discrete Elements ................................................................... 23-4

26 Boundary and Loading Conditions ..................................................................................................... 24-1


26.1 Pressure Boundary Conditions ........................................................................................... 24-1
26.2 Transmitting Boundaries ..................................................................................................... 24-3
26.3 Kinematic Boundary Conditions ......................................................................................... 24-3
26.4 Displacement Constraints................................................................................................... 24-3
26.5 Prescribed Displacements, Velocities, and Accelerations ................................................. 24-4
26.6 Body Force Loads............................................................................................................... 24-5

27 Time Integration .................................................................................................................................. 25-1


27.1 Background......................................................................................................................... 25-1
27.2 The Central Difference Method........................................................................................... 25-3

LS-DYNA DEV 10/27/16 (r:8004) 0-11 (Table of Contents)


Table of Contents LS-DYNA Theory Manual

27.3 Stability of Central Difference Method................................................................................ 25-3


27.4 Subcycling (Mixed Time Integration) .................................................................................. 25-5

28 Rigid Body Dynamics.......................................................................................................................... 26-1


28.1 Rigid Body Joints ................................................................................................................ 26-6
28.2 Deformable to Rigid Material Switching ............................................................................. 26-8
28.3 Rigid Body Welds ............................................................................................................... 26-9

29 Contact-Impact Algorithm .................................................................................................................. 27-1


29.1 Introduction ......................................................................................................................... 27-1
29.2 Kinematic Constraint Method ............................................................................................. 27-1
29.3 Penalty Method ................................................................................................................... 27-2
29.4 Distributed Parameter Method ........................................................................................... 27-3
29.5 Preliminaries........................................................................................................................ 27-4
29.6 Slave Search ....................................................................................................................... 27-5
29.7 Sliding With Closure and Separation .................................................................................. 27-9
29.7.1 Standard Penalty Formulation ............................................................................ 27-9
29.7.2 Soft Constraint Penalty Formulation................................................................. 27-10
29.7.3 Segment-based Penalty Formulation ............................................................... 27-11
29.8 Recent Improvements in Surface-to-Surface Contact ..................................................... 27-11
29.8.1 Improvements to the Contact Searching.......................................................... 27-11
29.8.2 Accounting For the Shell Thickness ................................................................. 27-13
29.8.3 Initial Contact Interpenetrations ....................................................................... 27-15
29.8.4 Contact Energy Calculation .............................................................................. 27-16
29.8.5 Contact Damping .............................................................................................. 27-17
Friction ......................................................................................................................... 27-20
29.9 Tied Interfaces .................................................................................................................. 27-21
29.10 Strongly Objective Tied Contacts ................................................................................... 27-24
29.10.1 Kinematics ...................................................................................................... 27-26
29.10.2 Translational constraint................................................................................... 27-27
29.10.3 Bending and torsion constraint ...................................................................... 27-28
29.11 Sliding-Only Interfaces.................................................................................................... 27-28
29.12 Bucket Sorting ................................................................................................................ 27-29
29.12.1 Bucket Sorting in TYPE 4 Single Surface Contact ......................................... 27-31
29.12.2 Bucket Sorting in Surface to Surface and TYPE 13 Single Surface Contact. 27-32
29.13 Single Surface Contact Algorithms in LS-DYNA ............................................................ 27-34
29.14 Surface to Surface Constraint Algorithm ........................................................................ 27-37
29.15 Planar Rigid Boundaries ................................................................................................. 27-40

0-12 (Table of Contents) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Table of Contents

29.16 Geometric Rigid Boundaries........................................................................................... 27-41


29.17 VDA/IGES Contact .......................................................................................................... 27-42
29.18 Simulated Draw Beads ................................................................................................... 27-45
29.19 Edge to Edge Contact .................................................................................................... 27-48
29.20 Beam to Beam Contact .................................................................................................. 27-50
29.21 Mortar contact ................................................................................................................ 27-52
29.21.1 Kinematics ...................................................................................................... 27-52
29.21.2 Constitutive relation ........................................................................................ 27-53
29.21.3 Contact nodal forces ...................................................................................... 27-54
29.21.4 Treatment of beams, sharp solid and shell edges ......................................... 27-54
29.21.5 Characteristic length and contact release ...................................................... 27-57
29.21.6 Outputs for debugging implicit models .......................................................... 27-58
29.21.7 Initial penetrations........................................................................................... 27-59
29.21.8 2D Mortar contact ........................................................................................... 27-61
29.21.9 Tied and tiebreak option ................................................................................. 27-66

30 Geometric Contact Entities ................................................................................................................. 28-1

31 Nodal Constraints ............................................................................................................................... 29-1


31.1 Nodal Constraint Sets ......................................................................................................... 29-1
31.2 Linear Constraint Equations ............................................................................................... 29-1

32 Vectorization and Parallelization ......................................................................................................... 30-1


32.1 Vectorization ....................................................................................................................... 30-1
32.2 Parallelization ...................................................................................................................... 30-4

33 Airbags ................................................................................................................................................ 31-1


33.1 Control Volume Modeling ................................................................................................... 31-1
33.2 Equation of State Model ..................................................................................................... 31-3
33.3 Airbag Inflation Model ......................................................................................................... 31-5
33.4 Wang's Hybrid Inflation Model ........................................................................................... 31-7
33.5 Constant Volume Tank Test ............................................................................................. 31-10

34 Dynamic Relaxation and System Damping ........................................................................................ 32-1


34.1 Dynamic Relaxation For Initialization .................................................................................. 32-1
34.2 Mass Weighted Damping.................................................................................................... 32-4
34.3 Dynamic Relaxation—How Fast Does it Converge? .......................................................... 32-4

35 Heat Transfer ...................................................................................................................................... 33-1


35.1 Conduction of Heat in an Orthotropic Solid ....................................................................... 33-1
35.2 Thermal Boundary Conditions ............................................................................................ 33-2
LS-DYNA DEV 10/27/16 (r:8004) 0-13 (Table of Contents)
Table of Contents LS-DYNA Theory Manual

35.3 Thermal Energy Balances ................................................................................................... 33-4


35.4 Heat Generation .................................................................................................................. 33-4
35.5 Initial Conditions ................................................................................................................. 33-4
35.6 Material Properties .............................................................................................................. 33-4
35.7 Nonlinear Analysis............................................................................................................... 33-5
35.8 Units .................................................................................................................................... 33-5

36 Adaptivity ............................................................................................................................................ 34-1

37 Implicit ................................................................................................................................................. 35-1


37.1 Introduction ......................................................................................................................... 35-1
37.2 Equations ............................................................................................................................ 35-2
37.2.1 Discretization ...................................................................................................... 35-2
37.2.2 Constraints .......................................................................................................... 35-3
37.3 Implicit Statics..................................................................................................................... 35-5
37.3.1 Linearization ........................................................................................................ 35-5
37.3.2 Linear theory ....................................................................................................... 35-6
37.3.3 Quasi-Newton iterations ..................................................................................... 35-8
37.3.4 Tangent stiffness reformations ......................................................................... 35-13
37.3.5 Line search........................................................................................................ 35-13
37.3.6 Convergence check .......................................................................................... 35-15
37.3.7 Automatic time stepping ................................................................................... 35-16
37.4 Implicit Dynamics .............................................................................................................. 35-17
37.4.1 Newmark time integration ................................................................................. 35-17
37.4.2 Practical considerations ................................................................................... 35-17
37.4.3 Linear theory ..................................................................................................... 35-18

38 Arc-length ......................................................................................................................................... 35-23


38.1 Overview ........................................................................................................................... 35-23
38.2 Nonlinear equations .......................................................................................................... 35-26
38.3 Newton iterations .............................................................................................................. 35-27
38.4 Arc-length constraint – predictor step .............................................................................. 35-29
38.5 Arc-length constraint – corrector steps ............................................................................ 35-30

39 Sparse Direct Linear Equation Solvers ............................................................................................... 36-1


39.1 Sparsity Preserving Orderings ............................................................................................ 36-1
39.2 Multifrontal Algorithm.......................................................................................................... 36-2
39.3 The Five Solver Options ...................................................................................................... 36-3
39.4 Treating Matrix Singularities ............................................................................................... 36-4

0-14 (Table of Contents) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Table of Contents

40 Sparse Eigensolver ............................................................................................................................. 37-1


40.1 The Eigenvalue Problem for Rotating Systems .................................................................. 37-2

41 Boundary Element Method ................................................................................................................. 38-1


41.1 Governing Equations .......................................................................................................... 38-1
41.2 Surface Representation ...................................................................................................... 38-2
41.3 The Neighbor Array ............................................................................................................. 38-3
41.4 Wakes ................................................................................................................................. 38-5
41.5 Execution Time Control ...................................................................................................... 38-6
41.6 Free-Stream Flow ............................................................................................................... 38-7

42 SPH ..................................................................................................................................................... 39-1


42.1 SPH Formulation ................................................................................................................. 39-1
42.1.1 Definitions ........................................................................................................... 39-1
42.1.2 Discrete Form of Conservation Equations .......................................................... 39-2
42.1.3 Applications to Conservation Equations ............................................................ 39-4
42.1.4 Formulation Available in LS-DYNA ..................................................................... 39-4
42.2 Sorting ................................................................................................................................. 39-5
42.3 Artificial Viscosity ................................................................................................................ 39-5
42.4 Time Integration .................................................................................................................. 39-6
42.5 Initial Setup ......................................................................................................................... 39-6

43 Element-Free Galerkin ........................................................................................................................ 40-1


43.1 Moving least-squares ......................................................................................................... 40-1
43.2 Integration constraint and strain smoothing ....................................................................... 40-3
43.3 Lagrangian strain smoothing for path-dependent problems ............................................. 40-5
43.4 Galerkin approximation for explicit dynamic computation ................................................. 40-5
43.5 Imposition of essential boundary condition........................................................................ 40-6
43.6 Mesh-free Shell ................................................................................................................... 40-8
43.6.1 Mesh-free Shell Surface Representation ............................................................ 40-8
43.6.2 Updated Lagrangian Formulation and Co-rotational Procedure...................... 40-10

44 Linear shells ........................................................................................................................................ 41-1


44.1 Shells for Linear Analysis .................................................................................................... 41-1
44.2 Wilson’s Shell (element #20) ............................................................................................... 41-1
44.2.1 Plate Element ...................................................................................................... 41-1
44.2.2 Membrane Element ............................................................................................. 41-4
44.3 Assumed Strain/Membrane with Drilling Degree-of-freedom (element #18) ..................... 41-7
44.3.1 Membrane Element ............................................................................................. 41-7

LS-DYNA DEV 10/27/16 (r:8004) 0-15 (Table of Contents)


Table of Contents LS-DYNA Theory Manual

44.3.2 Plate Element ...................................................................................................... 41-7


44.4 Differences between Element Types 18 and 20. ................................................................ 41-8

45 Random Geometrical Imperfections ................................................................................................... 42-1


45.1 Introduction to Random Geometrical Imperfections Using Karhunen-Loève Expansions 42-1
45.2 Methodology ....................................................................................................................... 42-1
45.2.1 Generation of random fields using Karhunen-Loève expansion ........................ 42-1
45.2.2 Solution of Fredholm integral of the second kind for analytical covariance
functions ........................................................................................................................ 42-2
45.2.3 Generating eigenfunctions from experimentally measured fields ...................... 42-4

46 Frequency Domain .............................................................................................................................. 42-6


46.1 Frequency Response Functions ......................................................................................... 42-6
46.1.1 FRF Computations .............................................................................................. 42-6
46.1.2 About the damping ............................................................................................. 42-8
46.2 ACOUSTIC FEM.................................................................................................................. 42-9
46.2.1 Theory basis........................................................................................................ 42-9

47 Rotor Dynamics ................................................................................................................................ 42-10


47.1 Introduction ....................................................................................................................... 42-10
47.2 Two Coordinate systems .................................................................................................. 42-10
47.3 Forces in the Rotating Coordinate System ...................................................................... 42-11
47.4 Transformation between Coordinate Systems ................................................................. 42-12
47.5 Equation of Motion in Rotating Coordinate System ......................................................... 42-13
47.6 Equation of Motion in Fixed Coordinate System .............................................................. 42-15
47.7 Arbitrary Rotation Axis ...................................................................................................... 42-15

48 References .......................................................................................................................................... 43-1

0-16 (Table of Contents) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Abstract

1
Abstract

LS-DYNA is a general purpose finite element code for analyzing the large
deformation static and dynamic response of structures including structures coupled to
fluids. The main solution methodology is based on explicit time integration. An implicit
solver is currently available with somewhat limited capabilities including structural
analysis and heat transfer. A contact-impact algorithm allows difficult contact problems to
be easily treated with heat transfer included across the contact interfaces. By a
specialization of this algorithm, such interfaces can be rigidly tied to admit variable zoning
without the need of mesh transition regions. Other specializations, allow draw beads in
metal stamping applications to be easily modeled simply by defining a line of nodes along
the draw bead. Spatial discretization is achieved by the use of four node tetrahedron and
eight node solid elements, two node beam elements, three and four node shell elements,
eight node solid shell elements, truss elements, membrane elements, discrete elements, and
rigid bodies. A variety of element formulations are available for each element type.
Specialized capabilities for airbags, sensors, and seatbelts have tailored LS-DYNA for
applications in the automotive industry. Adaptive remeshing is available for shell
elements and is widely used in sheet metal stamping applications. LS-DYNA currently
contains approximately one-hundred constitutive models and ten equations-of-state to
cover a wide range of material behavior. This theoretical manual has been written to
provide users and potential users with insight into the mathematical and physical basis of
the code.

LS-DYNA DEV 10/27/16 (r:8004) 1-1 (Abstract)


LS-DYNA Theory Manual History of LS-DYNA

2
History of LS-DYNA

The origin of LS-DYNA dates back to the public domain software, DYNA3D, which
was developed in the mid-seventies at the Lawrence Livermore National Laboratory. The
first version of DYNA3D [Hallquist 1976a] was released in 1976 with constant stress 4- or 8-
node solid elements, 16- and 20-node solid elements with 2 × 2 × 2 Gaussian quadrature, 3,
4, and 8-node membrane elements, and a 2-node cable element. A nodal constraint contact-
impact interface algorithm [Hallquist 1977] was available. On the Control Data CDC-7600,
a supercomputer in 1976, the speed of the code varied from 36 minutes per 106 mesh cycles
with 4-8 node solids to 180 minutes per 106 mesh cycles with 16 and 20 node solids.
Without hourglass control to prevent formation of non-physical zero energy deformation
modes, constant stress solids were processed at 12 minutes per 106 mesh cycles. A
moderate number of very costly solutions were obtained with this version of DYNA3D
using 16- and 20-node solids. Hourglass modes combined with the procedure for
computing the time step size prevented us from obtaining solutions with constant stress
elements.

In this early development, several things became apparent. Hourglass deformation


modes of the constant stress elements were invariably excited by the contact-impact
algorithm, showing that a new sliding interface algorithm was needed. Higher order
elements seemed to be impractical for shock wave propagation because of numerical noise
resulting from the ad hoc mass lumping necessary to generate a diagonal mass matrix.
Although the lower frequency structural response was accurately computed with these
elements, their high computer cost made analysis so expensive as to be impractical. It was
obvious that realistic three-dimensional structural calculations were possible, if and only if
the under-integrated eight node constant stress solid element could be made to function.
This implied a need for a much better sliding interface algorithm, a more cost-effective
hourglass control, more optimal programming, and a machine much faster than the CDC-
7600. This latter need was fulfilled several years later when LLNL took deliver of its first
CRAY-1. At this time, DYNA3D was completely rewritten.

The next version, released in 1979, achieved the aforementioned goals. On the
CRAY the vectorized speed was 50 times faster, 0.67 minutes per million mesh cycles. A

LS-DYNA DEV 10/27/16 (r:8004) 2-1 (History of LS-DYNA)


History of LS-DYNA LS-DYNA Theory Manual

symmetric, penalty-based, contact-impact algorithm was considerably faster in execution


speed and exceedingly reliable. Due to lack of use, the membrane and cable elements were
stripped and all higher order elements were eliminated as well. Wilkins’ finite difference
equations [Wilkins et al. 1974] were implemented in unvectorized form in an overlay to
compare their performance with the finite element method. The finite difference algorithm
proved to be nearly two times more expensive than the finite element approach (apart from
vectorization) with no compensating increase in accuracy, and was removed in the next
code update.

The 1981 version [Hallquist 1981a] evolved from the 1979 version. Nine additional
material models were added to allow a much broader range of problems to be modeled
including explosive-structure and soil-structure interactions. Body force loads were
implemented for angular velocities and base accelerations. A link was also established
from the 3D Eulerian code JOY [Couch, et. al., 1983] for studying the structural response to
impacts by penetrating projectiles. An option was provided for storing element data on
disk thereby doubling the capacity of DYNA3D.

The 1982 version of DYNA3D [Hallquist 1982] accepted DYNA2D [Hallquist 1980]
material input directly. The new organization was such that equations of state and
constitutive models of any complexity could be easily added. Complete vectorization of
the material models had been nearly achieved with about a 10 percent increase in execution
speed over the 1981 version.

In the 1986 version of DYNA3D [Hallquist and Benson 1986], many new features
were added, including beams, shells, rigid bodies, single surface contact, interface friction,
discrete springs and dampers, optional hourglass treatments, optional exact volume
integration, and VAX/VMS, IBM, UNIX, COS operating systems compatibility, that greatly
expanded its range of applications. DYNA3D thus became the first code to have a general
single surface contact algorithm.

In the 1987 version of DYNA3D [Hallquist and Benson 1987] metal forming
simulations and composite analysis became a reality. This version included shell thickness
changes, the Belytschko-Tsay shell element [Belytschko and Tsay, 1981], and dynamic
relaxation. Also included were non-reflecting boundaries, user specified integration rules
for shell and beam elements, a layered composite damage model, and single point
constraints.

New capabilities added in the 1988 DYNA3D [Hallquist 1988] version included a
cost effective resultant beam element, a truss element, a C0 triangular shell, the BCIZ
triangular shell [Bazeley et al., 1965], mixing of element formulations in calculations,
composite failure modeling for solids, noniterative plane stress plasticity, contact surfaces
with spot welds, tiebreak sliding surfaces, beam surface contact, finite stonewalls,
stonewall reaction forces, energy calculations for all elements, a crushable foam
constitutive model, comment cards in the input, and one-dimensional slidelines.

2-2 (History of LS-DYNA) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual History of LS-DYNA

In 1988 the Hallquist began working half-time at LLNL to devote more time to the
development and support of LS-DYNA for automotive applications. By the end of 1988 it
was obvious that a much more concentrated effort would be required in the development
of LS-DYNA if problems in crashworthiness were to be properly solved; therefore, at the
start of 1989 the Hallquist resigned from LLNL to continue code development full time at
Livermore Software Technology Corporation. The 1989 version introduced many
enhanced capabilities including a one-way treatment of slide surfaces with voids and
friction; cross-sectional forces for structural elements; an optional user specified minimum
time step size for shell elements using elastic and elastoplastic material models; nodal
accelerations in the time history database; a compressible Mooney-Rivlin material model; a
closed-form update shell plasticity model; a general rubber material model; unique penalty
specifications for each slide surface; external work tracking; optional time step criterion for
4-node shell elements; and internal element sorting to allow full vectorization of right-
hand-side force assembly.

2.1 Features add in 1989-1990


Throughout the past decade, considerable progress has been made as may be seen in
the chronology of the developments which follows. During 1989 many extensions and
developments were completed, and in 1990 the following capabilities were delivered to
users:
• arbitrary node and element numbers,
• fabric model for seat belts and airbags,
• composite glass model,
• vectorized type 3 contact and single surface contact,
• many more I/O options,
• all shell materials available for 8 node brick shell,
• strain rate dependent plasticity for beams,
• fully vectorized iterative plasticity,
• interactive graphics on some computers,
• nodal damping,
• shell thickness taken into account in shell type 3 contact,
• shell thinning accounted for in type 3 and type 4 contact,
• soft stonewalls,
• print suppression option for node and element data,
• massless truss elements, rivets – based on equations of rigid body dynamics,
• massless beam elements, spot welds – based on equations of rigid body dynamics,
LS-DYNA DEV 10/27/16 (r:8004) 2-3 (History of LS-DYNA)
History of LS-DYNA LS-DYNA Theory Manual

• expanded databases with more history variables and integration points,


• force limited resultant beam,
• rotational spring and dampers, local coordinate systems for discrete elements,
• resultant plasticity for C0 triangular element,
• energy dissipation calculations for stonewalls,
• hourglass energy calculations for solid and shell elements,
• viscous and Coulomb friction with arbitrary variation over surface,
• distributed loads on beam elements,
• Cowper and Symonds strain rate model,
• segmented stonewalls,
• stonewall Coulomb friction,
• stonewall energy dissipation,
• airbags (1990),
• nodal rigid bodies,
• automatic sorting of triangular shells into C0 groups,
• mass scaling for quasi static analyses,
• user defined subroutines,
• warpage checks on shell elements,
• thickness consideration in all contact types,
• automatic orientation of contact segments,
• sliding interface energy dissipation calculations,
• nodal force and energy database for applied boundary conditions,
• defined stonewall velocity with input energy calculations,

2.2 Options added in 1991-1992


• rigid/deformable material switching,
• rigid bodies impacting rigid walls,
• strain-rate effects in metallic honeycomb model 26,
• shells and beams interfaces included for subsequent component analyses,
• external work computed for prescribed displacement/velocity/accelerations,
• linear constraint equations,

2-4 (History of LS-DYNA) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual History of LS-DYNA

• MPGS database,
• MOVIE database,
• Slideline interface file,
• automated contact input for all input types,
• automatic single surface contact without element orientation,
• constraint technique for contact,
• cut planes for resultant forces,
• crushable cellular foams,
• urethane foam model with hysteresis,
• subcycling,
• friction in the contact entities,
• strains computed and written for the 8 node thick shells,
• “good” 4 node tetrahedron solid element with nodal rotations,
• 8 node solid element with nodal rotations,
• 2  2 integration for the membrane element,
• Belytschko-Schwer integrated beam,
• thin-walled Belytschko-Schwer integrated beam,
• improved LS-DYNA database control,
• null material for beams to display springs and seatbelts in TAURUS,
• parallel implementation on Cray and SGI computers,
• coupling to rigid body codes,
• seat belt capability.

2.3 Options added in 1993-1994


• Arbitrary Lagrangian Eulerian brick elements,
• Belytschko-Wong-Chiang quadrilateral shell element,
• Warping stiffness in the Belytschko-Tsay shell element,
• Fast Hughes-Liu shell element,
• Fully integrated brick shell element,
• Discrete 3D beam element,
• Generalized dampers,

LS-DYNA DEV 10/27/16 (r:8004) 2-5 (History of LS-DYNA)


History of LS-DYNA LS-DYNA Theory Manual

• Cable modeling,
• Airbag reference geometry,
• Multiple jet model,
• Generalized joint stiffnesses,
• Enhanced rigid body to rigid body contact,
• Orthotropic rigid walls,
• Time zero mass scaling,
• Coupling with USA (Underwater Shock Analysis),
• Layered spot welds with failure based on resultants or plastic strain,
• Fillet welds with failure,
• Butt welds with failure,
• Automatic eroding contact,
• Edge-to-edge contact,
• Automatic mesh generation with contact entities,
• Drawbead modeling,
• Shells constrained inside brick elements,
• NIKE3D coupling for springback,
• Barlat’s anisotropic plasticity,
• Superplastic forming option,
• Rigid body stoppers,
• Keyword input,
• Adaptivity,
• First MPP (Massively Parallel) version with limited capabilities.
• Built in least squares fit for rubber model constitutive constants,
• Large hystersis in hyperelastic foam,
• Bilhku/Dubois foam model,
• Generalized rubber model,

2.4 Version 936


New options added to version 936 in 1995 include:
• Belytschko - Leviathan Shell

2-6 (History of LS-DYNA) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual History of LS-DYNA

• Automatic switching between rigid and deformable bodies.


• Accuracy on SMP machines to give identical answers on one, two or more
processors.
• Local coordinate systems for cross-section output can now be specified.
• Null material for shell elements.
• Global body force loads now may be applied to a subset of materials.
• User defined loading subroutine.
• Improved interactive graphics.
• New initial velocity options for specifying rotational velocities.
• Geometry changes after dynamic relaxation can be considered for initial velocities.
• Velocities may also be specified by using material or part ID’s.
• Improved speed of brick element hourglass force and energy calculations.
• Pressure outflow boundary conditions have been added for the ALE options.
• More user control for hourglass control constants for shell elements.
• Full vectorization in constitutive models for foam, models 57 and 63.
• Damage mechanics plasticity model, material 81,
• General linear viscoelasticity with 6 term prony series.
• Least squares fit for viscoelastic material constants.
• Table definitions for strain rate effects in material type 24.
• Improved treatment of free flying nodes after element failure.
• Automatic projection of nodes in CONTACT_TIED to eliminate gaps in the surface.
• More user control over contact defaults.
• Improved interpenetration warnings printed in automatic contact.
• Flag for using actual shell thickness in single surface contact logic rather than the
default.
• Definition by exempted part ID’s.
• Airbag to Airbag venting/segmented airbags are now supported.
• Airbag reference geometry speed improvements by using the reference geometry
for the time step size calculation.
• Isotropic airbag material may now be directly for cost efficiency.
• Airbag fabric material damping is now specified as the ratio of critical damping.
• Ability to attach jets to the structure so the airbag, jets, and structure to move
together.

LS-DYNA DEV 10/27/16 (r:8004) 2-7 (History of LS-DYNA)


History of LS-DYNA LS-DYNA Theory Manual

• PVM 5.1 Madymo coupling is available.


• Meshes are generated within LS-DYNA3D for all standard contact entities.
• Joint damping for translational motion.
• Angular displacements, rates of displacements, damping forces, etc. in JNTFORC
file.
• Link between LS-NIKE3D to LS-DYNA3D via *INITIAL_STRESS keywords.
• Trim curves for metal forming springback.
• Sparse equation solver for springback.
• Improved mesh generation for IGES and VDA provides a mesh that can directly be
used to model tooling in metal stamping analyses.

2.5 Version 940


New options added to Version 940 in 1996 and 1997:
• Part/Material ID’s may be specified with 8 digits.
• Rigid body motion can be prescribed in a local system fixed to the rigid body.
• Nonlinear least squares fit available for the Ogden rubber model.
• Lease squares fit to the relaxation curves for the viscoelasticity in rubber.
• Fu-Chang rate sensitive foam.
• 6 term Prony series expansion for rate effects in model 57-now 73
• Viscoelastic material model 76 implemented for shell elements.
• Mechanical threshold stress (MTS) plasticity model for rate effects.
• Thermoelastic-plastic material model for Hughes-Liu beam element.
• Ramberg-Osgood soil model
• Invariant local coordinate systems for shell elements are optional.
• Second order accurate stress updates.
• Four-noded, linear, tetrahedron element.
• Co-rotational solid element for foam that can invert without stability problems.
• Improved speed in rigid body to rigid body contacts.
• Improved searching for the a_3, a_5 and a10 contact types.
• Invariant results on shared memory parallel machines with the a_n contact types.
• Thickness offsets in type 8 and 9 tie break contact algorithms.

2-8 (History of LS-DYNA) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual History of LS-DYNA

• Bucket sort frequency can be controlled by a load curve for airbag applications.
• In automatic contact each part ID in the definition may have unique:
◦ Static coefficient of friction
◦ Dynamic coefficient of friction
◦ Exponential decay coefficient
◦ Viscous friction coefficient
◦ Optional contact thickness
◦ Optional thickness scale factor
◦ Local penalty scale factor
• Automatic beam-to-beam, shell edge-to-beam, shell edge-to-shell edge and single
surface contact algorithm.
• Release criteria may be a multiple of the shell thickness in types a_3, a_5, a10, 13,
and 26 contact.
• Force transducers to obtain reaction forces in automatic contact definitions. Defined
manually via segments, or automatically via part ID’s.
• Searching depth can be defined as a function of time.
• Bucket sort frequency can be defined as a function of time.
• Interior contact for solid (foam) elements to prevent "negative volumes."
• Locking joint
• Temperature dependent heat capacity added to Wang-Nefske inflator models.
• Wang Hybrid inflator model [Wang, 1996] with jetting options and bag-to-bag
venting.
• Aspiration included in Wang’s hybrid model [Nucholtz, Wang, Wylie, 1996].
• Extended Wang’s hybrid inflator with a quadratic temperature variation for heat
capacities [Nusholtz, 1996].
• Fabric porosity added as part of the airbag constitutive model.
• Blockage of vent holes and fabric in contact with structure or itself considered in
venting with leakage of gas.
• Option to delay airbag liner with using the reference geometry until the reference
area is reached.
• Birth time for the reference geometry.
• Multi-material Euler/ALE fluids,
◦ 2nd order accurate formulations.
◦ Automatic coupling to shell, brick, or beam elements

LS-DYNA DEV 10/27/16 (r:8004) 2-9 (History of LS-DYNA)


History of LS-DYNA LS-DYNA Theory Manual

◦ Coupling using LS-DYNA contact options.


◦ Element with fluid + void and void material
◦ Element with multi-materials and pressure equilibrium
• Nodal inertia tensors.
• 2D plane stress, plane strain, rigid, and axisymmetric elements
• 2D plane strain shell element
• 2D axisymmetric shell element.
• Full contact support in 2D, tied, sliding only, penalty and constraint techniques.
• Most material types supported for 2D elements.
• Interactive remeshing and graphics options available for 2D.
• Subsystem definitions for energy and momentum output. and many more
enhancements not mentioned above.

2.6 Version 950


Capabilities added during 1997-1998 in Version 950 include:
• Adaptive refinement can be based on tooling curvature with FORMING contact.
• The display of draw beads is now possible since the draw bead data is output into
the d3plot database.
• An adaptive box option, *DEFINE_BOX_ADAPTIVE, allows control over the
refinement level and location of elements to be adapted.
• A root identification file, adapt.rid, gives the parent element ID for adapted
elements.
• Draw bead box option, *DEFINE_BOX_DRAWBEAD, simplifies draw bead input.
• The new control option, CONTROL_IMPLICIT, activates an implicit solution
scheme.
• 2D Arbitrary-Lagrangian-Eulerian elements.
• 2D automatic contact is defined by listing part ID's.
• 2D r-adaptivity for plane strain and axisymmetric forging simulations is available.
• 2D automatic non-interactive rezoning as in LS-DYNA2D.
• 2D plane strain and axisymmetric element with 2 × 2 selective-reduced integration
are implemented.
• Implicit 2D solid and plane strain elements are available.

2-10 (History of LS-DYNA) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual History of LS-DYNA

• Implicit 2D contact is available.


• The new keyword, *DELETE_CONTACT_2DAUTO, allows the deletion of 2D
automatic contact definitions.
• The keyword, *LOAD_BEAM is added for pressure boundary conditions on 2D
elements.
• A viscoplastic strain rate option is available for materials:
◦ *MAT_PLASTIC_KINEMATIC
◦ *MAT_JOHNSON_COOK
◦ *MAT_POWER_LAW_PLASTICITY
◦ *MAT_STRAIN_RATE_DEPENDENT_PLASTICITY
◦ *MAT_PIECEWISE_LINEAR_PLASTICITY
◦ *MAT_RATE_SENSITIVE_POWERLAW_PLASTICITY
◦ *MAT_ZERILLI-ARMSTRONG
◦ *MAT_PLASTICITY_WITH_DAMAGE
◦ *MAT_PLASTICITY_COMPRESSION_TENSION
• Material model, *MAT_PLASTICITY_WITH_DAMAGE, has a piecewise linear
damage curve given by a load curve ID.
• The Arruda-Boyce hyper-viscoelastic rubber model is available, see *MAT_ARRU-
DA_BOYCE.
• Transverse-anisotropic-viscoelastic material for heart tissue, see *MAT_HEART_-
TISSUE.
• Lung hyper-viscoelastic material, see *MAT_LUNG_TISSUE.
• Compression/tension plasticity model, see *MAT_PLASTICITY_COMPRESSION_-
TENSION.
• The Lund strain rate model, *MAT_STEINBERG_LUND, is added to Steinberg-
Guinan plasticity model.
• Rate sensitive foam model, *MAT_FU_CHANG_FOAM, has been extended to
include engineering strain rates, etc.
• Model, *MAT_MODIFIED_PIECEWISE_LINEAR_PLASTICITY, is added for
modeling the failure of aluminum.
• Material model, *MAT_SPECIAL_ORTHOTROPIC, added for television shadow
mask problems.
• Erosion strain is implemented for material type, *MAT_BAMMAN_DAMAGE.
• The equation of state, *EOS_JWLB, is available for modeling the expansion of
explosive gases.

LS-DYNA DEV 10/27/16 (r:8004) 2-11 (History of LS-DYNA)


History of LS-DYNA LS-DYNA Theory Manual

• The reference geometry option is extended for foam and rubber materials and can
be used for stress initialization, see *INITIAL_FOAM_REFERENCE_GEOMETRY.
• A vehicle positioning option is available for setting the initial orientation and
velocities, see *INITIAL_VEHICLE_KINEMATICS.
• A boundary element method is available for incompressible fluid dynamics
problems.
• The thermal materials work with instantaneous coefficients of thermal expansion:
◦ *MAT_ELASTIC_PLASTIC_THERMAL
◦ *MAT_ORTHOTROPIC_THERMAL
◦ *MAT_TEMPERATURE_DEPENDENT_ORTHOTROPIC
◦ *MAT_ELASTIC_WITH_VISCOSITY.
• Airbag interaction flow rate versus pressure differences.
• Contact segment search option, [bricks first optional]
• A through thickness Gauss integration rule with 1-10 points is available for shell
elements. Previously, 5 were available.
• Shell element formulations can be changed in a full deck restart.
• The tied interface which is based on constraint equations, TIED_SURFACE_TO_-
SURFACE, can now fail with FAILURE option.
• A general failure criteria for solid elements is independent of the material type, see
*MAT_ADD_EROSION
• Load curve control can be based on thinning and a flow limit diagram, see *DE-
FINE_CURVE_FEEDBACK.
• An option to filter the spotweld resultant forces prior to checking for failure has
been added the option, *CONSTRAINED_SPOTWELD, by appending,_FILTERED_-
FORCE, to the keyword.
• Bulk viscosity is available for shell types 1, 2, 10, and 16.
• When defining the local coordinate system for the rigid body inertia tensor a local
coordinate system ID can be used. This simplifies dummy positioning.
• Prescribing displacements, velocities, and accelerations is now possible for rigid
body nodes.
• One-way flow is optional for segmented airbag interactions.
• Pressure time history input for airbag type, LINEAR_FLUID, can be used.
• An option is available to independently scale system damping by part ID in each of
the global directions.
• An option is available to independently scale global system damping in each of the
global directions.
2-12 (History of LS-DYNA) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual History of LS-DYNA

• Added option to constrain global DOF along lines parallel with the global axes. The
keyword is *CONSTRAINED_GLOBAL. This option is useful for adaptive remesh-
ing.
• Beam end code releases are available, see *ELEMENT_BEAM.
• An initial force can be directly defined for the cable material, *MAT_CABLE_DIS-
CRETE_BEAM. The specification of slack is not required if this option is used.
• Airbag pop pressure can be activated by accelerometers.
• Termination may now be controlled by contact, via *TERMINATION_CONTACT.
• Modified shell elements types 8, 10 and the warping stiffness option in the
Belytschko-Tsay shell to ensure orthogonality with rigid body motions in the event
that the shell is badly warped. This is optional in the Belytschko-Tsay shell and the
type 10 shell.
• A one point quadrature brick element with an exact hourglass stiffness matrix has
been implemented for implicit and explicit calculations.
• Automatic file length determination for d3plot binary database is now implement-
ed. This insures that at least a single state is contained in each d3plot file and elimi-
nates the problem with the states being split between files.
• The dump files, which can be very large, can be placed in another directory by
specifying d=/home/user /test/d3dump on the execution line.
• A print flag controls the output of data into the MATSUM and RBDOUT files by
part ID's. The option, PRINT, has been added as an option to the *PART keyword.
• Flag has been added to delete material data from the d3thdt file. See *DATABASE_-
EXTENT_BINARY and column 25 of the 19th control card in the structured input.
• After dynamic relaxation completes, a file is written giving the displaced state
which can be used for stress initialization in later runs.

2.7 Version 960


Capabilities added during 1998-2000 in Version 960. Most new capabilities work on both
the MPP and SMP versions; however, the capabilities that are implemented for the SMP
version only, which were not considered critical for this release, are flagged below. These
SMP unique capabilities are being extended for MPP calculations and will be available in
the near future. The implicit capabilities for MPP require the development of a scalable
eigenvalue solver, which is under development for a later release of LS-DYNA.
• Incompressible flow solver is available. Structural coupling is not yet implemented.
• Adaptive mesh coarsening can be done before the implicit spring back calculation in
metal forming applications.

LS-DYNA DEV 10/27/16 (r:8004) 2-13 (History of LS-DYNA)


History of LS-DYNA LS-DYNA Theory Manual

• Two-dimensional adaptivity can be activated in both implicit and explicit calcula-


tions. (SMP version only)
• An internally generated smooth load curve for metal forming tool motion can be
activated with the keyword: *DEFINE_CURVE_SMOOTH.
• Torsional forces can be carried through the deformable spot welds by using the
contact type: *CONTACT_SPOTWELD_WITH_TORSION (SMP version only with a
high priority for the MPP version if this option proves to be stable.)
• Tie break automatic contact is now available via the *CONTACT_AUTOMATIC_..._
TIEBREAK options. This option can be used for glued panels. (SMP only)
• *CONTACT_RIGID_SURFACE option is now available for modeling road surfaces
(SMP version only).
• Fixed rigid walls PLANAR and PLANAR_FINITE are represented in the binary
output file by a single shell element.
• Interference fits can be modeled with the INTERFERENCE option in contact.
• A layered shell theory is implemented for several constitutive models including the
composite models to more accurately represent the shear stiffness of laminated
shells.
• Damage mechanics is available to smooth the post-failure reduction of the resultant
forces in the constitutive model *MAT_SPOTWELD_DAMAGE.
• Finite elastic strain isotropic plasticity model is available for solid elements. *MAT_-
FINITE_ELASTIC_STRAIN_PLASTICITY.
• A shape memory alloy material is available: *MAT_SHAPE_MEMORY.
• Reference geometry for material, *MAT_MODIFIED_HONEYCOMB, can be set at
arbitrary relative volumes or when the time step size reaches a limiting value. This
option is now available for all element types including the fully integrated solid
element.
• Non orthogonal material axes are available in the airbag fabric model. See *MAT_-
FABRIC.
• Other new constitutive models include for the beam elements:
◦ *MAT_MODIFIED_FORCE_LIMITED
◦ *MAT_SEISMIC_BEAM
◦ *MAT_CONCRETE_BEAM
• for shell and solid elements:
◦ *MAT_ELASTIC_VISCOPLASTIC_THERMAL
• for the shell elements:
◦ *MAT_GURSON

2-14 (History of LS-DYNA) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual History of LS-DYNA

◦ *MAT_GEPLASTIC_SRATE2000
◦ *MAT_ELASTIC_VISCOPLASTIC_THERMAL
◦ *MAT_COMPOSITE_LAYUP
◦ *MAT_COMPOSITE_LAYUP
◦ *MAT_COMPOSITE_direct
• for the solid elements:
◦ *MAT_JOHNSON_HOLMQUIST_CERAMICS
◦ *MAT_JOHNSON_HOLMQUIST_CONCRETE
◦ *MAT_INV_HYPERBOLIC_SIN
◦ *MAT_UNIFIED_CREEP
◦ *MAT_SOIL_BRICK
◦ *MAT_DRUCKER_PRAGER
◦ *MAT_RC_SHEAR_WALL
• and for all element options a very fast and efficient version of the Johnson-Cook
plasticity model is available:
◦ *MAT_SIMPLIFIED_JOHNSON_COOK
• A fully integrated version of the type 16 shell element is available for the resultant
constitutive models.
• A nonlocal failure theory is implemented for predicting failure in metallic materials.
The keyword *MAT_NONLOCAL activates this option for a subset of elastoplastic
constitutive models.
• A discrete Kirchhoff triangular shell element (DKT) for explicit analysis with three
in plane integration points is flagged as a type 17 shell element. This element has
much better bending behavior than the C0 triangular element.
• A discrete Kirchhoff linear triangular and quadrilaterial shell element is available as
a type 18 shell. This shell is for extracting normal modes and static analysis.
• A C0 linear 4-node quadrilaterial shell element is implemented as element type 20
with drilling stiffness for normal modes and static analysis.
• An assumed strain linear brick element is available for normal modes and statics.
• The fully integrated thick shell element has been extended for use in implicit
calculations.
• A fully integrated thick shell element based on an assumed strain formulation is
now available. This element uses a full 3D constitutive model which includes the
normal stress component and, therefore, does not use the plane stress assumption.

LS-DYNA DEV 10/27/16 (r:8004) 2-15 (History of LS-DYNA)


History of LS-DYNA LS-DYNA Theory Manual

• The 4-node constant strain tetrahedron element has been extended for use in
implicit calculations.
• Relative damping between parts is available, see *DAMPING_RELATIVE (SMP
only).
• Preload forces are can be input for the discrete beam elements.
• Objective stress updates are implemented for the fully integrated brick shell
element.
• Acceleration time histories can be prescribed for rigid bodies.
• Prescribed motion for nodal rigid bodies is now possible.
• Generalized set definitions, i.e., SET_SHELL_GENERAL etc. provide much
flexibility in the set definitions.
• The command "sw4." will write a state into the dynamic relaxation file, D3DRLF,
during the dynamic relaxation phase if the d3drlf file is requested in the input.
• Added mass by PART ID is written into the matsum file when mass scaling is used
to maintain the time step size, (SMP version only).
• Upon termination due to a large mass increase during a mass scaled calculation a
print summary of 20 nodes with the maximum added mass is printed.
• Eigenvalue analysis of models containing rigid bodies is now available using
BCSLIB-EXT solvers from Boeing. (SMP version only).
• Second order stress updates can be activated by part ID instead of globally on the
*CONTROL_ACCURACY input.
• Interface frictional energy is optionally computed for heat generation and is output
into the interface force file (SMP version only).
• The interface force binary database now includes the distance from the contact
surface for the FORMING contact options. This distance is given after the nodes are
detected as possible contact candidates. (SMP version only).
• Type 14 acoustic brick element is implemented. This element is a fully integrated
version of type 8, the acoustic element (SMP version only).
• A flooded surface option for acoustic applications is available (SMP version only).
• Attachment nodes can be defined for rigid bodies. This option is useful for NVH
applications.
• CONSTRAINED_POINTS tie any two points together. These points must lie on a
shell element.
• Soft constraint is available for edge-to-edge contact in type 26 contact.
• CONSTAINED_INTERPOLATION option for beam to solid interfaces and for
spreading the mass and loads. (SMP version only).

2-16 (History of LS-DYNA) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual History of LS-DYNA

• A database option has been added that allows the output of added mass for shell
elements instead of the time step size.
• A new contact option allows the inclusion of all internal shell edges in contact type
*CONTACT_GENERAL, type 26. This option is activated by adding INTERIOR
option.
• A new option allows the use deviatoric strain rates rather than total rates in material
model 24 for the Cowper-Symonds rate model.
• The CADFEM option for ASCII databases is now the default. Their option includes
more significant figures in the output files.
• When using deformable spot welds, the added mass for spot welds is now printed
for the case where global mass scaling is activated. This output is in the log file,
d3hsp file, and the messag file.
• Initial penetration warnings for edge-to-edge contact are now written into the
MESSAG file and the D3HSP file.
• Each compilation of LS-DYNA is given a unique version number.
• Finite length discrete beams with various local axes options are now available for
material types 66, 67, 68, 93, and 95. In this implementation the absolute value of
SCOOR must be set to 2 or 3 in the *SECTION_BEAM input.
• New discrete element constitutive models are available:
◦ *MAT_ELASTIC_SPRING_DISCRETE_BEAM
◦ *MAT_INELASTIC_SPRING_DISCRETE_BEAM
◦ *MAT_ELASTIC_6DOF_SPRING_DISCRETE_BEAM
◦ *MAT_INELASTIC_6DOF_SPRING_DISCRETE_BEAM
The latter two can be used as finite length beams with local coordinate systems.
• Moving SPC's are optional in that the constraints are applied in a local system that
rotates with the 3 defining nodes.
• A moving local coordinate system, CID, can be used to determine orientation of
discrete beam elements.
• Modal superposition analysis can be performed after an eigenvalue analysis. Stress
recovery is based on type 18 shell and brick (SMP only).
• Rayleigh damping input factor is now input as a fraction of critical damping, i.e.
0.10. The old method required the frequency of interest and could be highly unsta-
ble for large input values.
• Airbag option "SIMPLE_PRESSURE_VOLUME" allows for the constant CN to be
replaced by a load curve for initialization. Also, another load curve can be defined
which allows CN to vary as a function of time during dynamic relaxation. After
dynamic relaxation CN can be used as a fixed constant or load curve.

LS-DYNA DEV 10/27/16 (r:8004) 2-17 (History of LS-DYNA)


History of LS-DYNA LS-DYNA Theory Manual

• Hybrid inflator model utilizing CHEMKIN and NIST databases is now available.
Up to ten gases can be mixed.
• Option to track initial penetrations has been added in the automatic SMP contact
types rather than moving the nodes back to the surface. This option has been avail-
able in the MPP contact for some time. This input can be defined on the fourth card
of the *CONTROL_CONTACT input and on each contact definition on the third
optional card in the *CONTACT definitions.
• If the average acceleration flag is active, the average acceleration for rigid body
nodes is now written into the d3thdt and nodout files. In previous versions of LS-
DYNA, the accelerations on rigid nodes were not averaged.
• A capability to initialize the thickness and plastic strain in the crash model is
available through the option *INCLUDE_STAMPED_PART, which takes the results
from the LS-DYNA stamping simulation and maps the thickness and strain distribu-
tion onto the same part with a different mesh pattern.
• A capability to include finite element data from other models is available through
the option, *INCLUDE_TRANSFORM. This option will take the model defined in
an INCLUDE file: offset all ID's; translate, rotate, and scale the coordinates; and
transform the constitutive constants to another set of units.

2.8 Version 970


Many new capabilities were added during 2001-2002 to create version 970 of LS-DYNA.
Some of the new features, which are also listed below, were also added to later releases of
version 960. Most new explicit capabilities work for both the MPP and SMP versions;
however, the implicit capabilities for MPP require the development of a scalable eigenvalue
solver and a parallel implementation of the constraint equations into the global matrices.
This work is underway. A later release of version 970 is planned that will be scalable for
implicit solutions.
• MPP decomposition can be controlled using *CONTROL_MPP_DECOMPOSITION
commands in the input deck.
• The MPP arbitrary Lagrangian-Eulerian fluid capability now works for airbag
deployment in both SMP and MPP calculations.
• Euler-to-Euler coupling is now available through the keyword *CONSTRAINED_-
EULER_TO_EULER.
• Up to ten ALE multi-material groups may now be defined. The previous limit was
three groups.
• Volume fractions can be automatically assigned during initialization of multi-
material cells. See the GEOMETRY option of *INITIAL_VOLUME_FRACTION.
• A new ALE smoothing option is available to accurately predict shock fronts.

2-18 (History of LS-DYNA) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual History of LS-DYNA

• DATABASE_FSI activates output of fluid-structure interaction data to ASCII file


DBFSI.
• Point sources for airbag inflators are available. The origin and mass flow vector of
these inflators are permitted to vary with time.
• A majority of the material models for solid materials are available for calculations
using the SPH (Smooth Particle Hydrodynamics) option.
• The Element Free Galerkin method (EFG or meshfree) is available for two-
dimensional and three-dimensional solids. This new capability is not yet imple-
mented for MPP applications.
• A binary option for the ASCII files is now available. This option applies to all ASCII
files and results in one binary file that contains all the information normally spread
between a large number of separate ASCII files.
• Material models can now be defined by numbers rather than long names in the
keyword input. For example the keyword *MAT_PIECEWISE_LINEAR_PLASTICI-
TY can be replaced by the keyword: *MAT_024.
• An embedded NASTRAN reader for direct reading of NASTRAN input files is
available. This option allows a typical input file for NASTRAN to be read directly
and used without additional input. See the *INCLUDE_NASTRAN keyword.
• Names in the keyword input can represent numbers if the *PARAMETER option is
used to relate the names and the corresponding numbers.
• Model documentation for the major ASCII output files is now optional. This option
allows descriptors to be included within the ASCII files that document the contents
of the file.
• ID’s have been added to the following keywords:
◦ *BOUNDARY_PRESCRIBED_MOTION
◦ *BOUNDARY_PRESCRIBED_SPC
◦ *CONSTRAINED_GENERALIZED_WELD
◦ *CONSTRAINED_JOINT
◦ *CONSTRAINED_NODE_SET
◦ *CONSTRAINED_RIVET
◦ *CONSTRAINED_SPOTWELD
◦ *DATABASE_CROSS_SECTION
◦ *ELEMENT_MASS
• The *DATABASE_ADAMS keyword is available to output a modal neutral file
d3mnf. This is available upon customer request since it requires linking to an
ADAMS library file.

LS-DYNA DEV 10/27/16 (r:8004) 2-19 (History of LS-DYNA)


History of LS-DYNA LS-DYNA Theory Manual

• Penetration warnings for the contact option, “ignore initial penetration,” are added
as an option. Previously, no penetration warnings were written when this contact
option was activated.
• Penetration warnings for nodes in-plane with shell mid-surface are printed for the
AUTOMATIC contact options. Previously, these nodes were ignored since it was
assumed that they belonged to a tied interface where an offset was not used; conse-
quently, they should not be treated in contact.
• For the arbitrary spot weld option, the spot welded nodes and their contact
segments are optionally written into the d3hsp file. See *CONTROL_CONTACT.
• For the arbitrary spot weld option, if a segment cannot be found for the spot welded
node, an option now exists to error terminate. See *CONTROL_CONTACT.
• Spot weld resultant forces are written into the swforc file for solid elements used as
spot welds.
• Solid materials have now been added to the failed element report and additional
information is written for the “node is deleted” messages.
• A new option for terminating a calculation is available, *TERMINATION_CURVE.
• A 10-noded tetrahedron solid element is available with either a 4 or 5 point
integration rule. This element can also be used for implicit solutions.
• A new 4 node linear shell element is available that is based on Wilson’s plate
element combined with a Pian-Sumihara membrane element. This is shell type 21.
• A shear panel element has been added for linear applications. This is shell type 22.
This element can also be used for implicit solutions.
• A null beam element for visualization is available. The keyword to define this null
beam is *ELEMENT_PLOTEL. This element is necessary for compatibility with
NASTRAN.
• A scalar node can be defined for spring-mass systems. The keyword to define this
node is *NODE_SCALAR. This node can have from 1 to 6 scalar degrees-of-
freedom.
• A thermal shell has been added for through-thickness heat conduction. Internally, 8
additional nodes are created, four above and four below the mid-surface of the shell
element. A quadratic temperature field is modeled through the shell thickness.
Internally, the thermal shell is a 12 node solid element.
• A beam OFFSET option is available for the *ELEMENT_BEAM definition to permit
the beam to be offset from its defining nodal points. This has the advantage that all
beam formulations can now be used as shell stiffeners.
• A beam ORIENTATION option for orienting the beams by a vector instead of the
third node is available in the *ELEMENT_BEAM definition for NASTRAN compati-
bility.

2-20 (History of LS-DYNA) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual History of LS-DYNA

• Non-structural mass has been added to beam elements for modeling trim mass and
for NASTRAN compatibility.
• An optional checking of shell elements to avoid abnormal terminations is available.
See *CONTROL_SHELL. If this option is active, every shell is checked each time
step to see if the distortion is so large that the element will invert, which will result
in an abnormal termination. If a bad shell is detected, either the shell will be deleted
or the calculation will terminate. The latter is controlled by the input.
• An offset option is added to the inertia definition. See *ELEMENT_INERTIA_OFF-
SET keyword. This allows the inertia tensor to be offset from the nodal point.
• Plastic strain and thickness initialization is added to the draw bead contact option.
See *CONTACT_DRAWBEAD_INITIALIZE.
• Tied contact with offsets based on both constraint equations and beam elements for
solid elements and shell elements that have 3 and 6 degrees-of-freedom per node,
respectively. See BEAM_OFFSET and CONSTRAINED_OFFSET contact options.
These options will not cause problems for rigid body motions.
• The segment-based (SOFT = 2) contact is implemented for MPP calculations. This
enables airbags to be easily deployed on the MPP version.
• Improvements are made to segment-based contact for edge-to-edge and sliding
conditions, and for contact conditions involving warped segments.
• An improved interior contact has been implemented to handle large shear
deformations in the solid elements. A special interior contact algorithm is available
for tetrahedron elements.
• Coupling with MADYMO 6.0 uses an extended coupling that allows users to link
most MADYMO geometric entities with LS-DYNA FEM simulations. In this cou-
pling MADYMO contact algorithms are used to calculate interface forces between
the two models.
• Release flags for degrees-of-freedom for nodal points within nodal rigid bodies are
available. This makes the nodal rigid body option nearly compatible with the RBE2
option in NASTRAN.
• Fast updates of rigid bodies for metal forming applications can now be accom-
plished by ignoring the rotational degrees-of-freedom in the rigid bodies that are
typically inactive during sheet metal stamping simulations. See the keyword:
*CONTROL_RIGID.
• Center of mass constraints can be imposed on nodal rigid bodies with the SPC
option in either a local or a global coordinate system.
• Joint failure based on resultant forces and moments can now be used to simulate the
failure of joints.
• CONSTRAINED_JOINT_STIFFNESS now has a TRANSLATIONAL option for the
translational and cylindrical joints.

LS-DYNA DEV 10/27/16 (r:8004) 2-21 (History of LS-DYNA)


History of LS-DYNA LS-DYNA Theory Manual

• Joint friction has been added using table look-up so that the frictional moment can
now be a function of the resultant translational force.
• The nodal constraint options *CONSTRAINED_INTERPOLATION and *CON-
STRAINED_LINEAR now have a local option to allow these constraints to be ap-
plied in a local coordinate system.
• Mesh coarsening can now be applied to automotive crash models at the beginning
of an analysis to reduce computation times. See the new keyword: *CONTROL_-
COARSEN.
• Force versus time seatbelt pretensioner option has been added.
• Both static and dynamic coefficients of friction are available for seat belt slip rings.
Previously, only one friction constant could be defined.
• *MAT_SPOTWELD now includes a new failure model with rate effects as well as
additional failure options.
• Constitutive models added for the discrete beam elements:
◦ *MAT_1DOF_GENERALIZED_SPRING
◦ *MAT_GENERAL_NONLINEAR_6dof_DISCRETE_BEAM
◦ *MAT_GENERAL_NONLINEAR_1dof_DISCRETE_BEAM
◦ *MAT_GENERAL_SPRING_DISCRETE_BEAM
◦ *MAT_GENERAL_JOINT_DISCRETE_BEAM
◦ *MAT_SEISMIC_ISOLATOR
• for shell and solid elements:
◦ *MAT_PLASTICITY_WITH_DAMAGE_ORTHO
◦ *MAT_SIMPLIFIED_JOHNSON_COOK_ORTHOTROPIC_DAMAGE
◦ *MAT_HILL_3R
◦ *MAT_GURSON_RCDC
• for the solid elements:
◦ *MAT_SPOTWELD
◦ *MAT_HILL_FOAM
◦ *MAT_WOOD
◦ *MAT_VISCOELASTIC_HILL_FOAM
◦ *MAT_LOW_DENSITY_SYNTHETIC_FOAM
◦ *MAT_RATE_SENSITIVE_POLYMER
◦ *MAT_QUASILINEAR VISCOELASTIC
◦ *MAT_TRANSVERSELY_ANISOTROPIC_CRUSHABLE_FOAM

2-22 (History of LS-DYNA) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual History of LS-DYNA

◦ *MAT_VACUUM
◦ *MAT_MODIFIED_CRUSHABLE_FOAM
◦ *MAT_PITZER_CRUSHABLE FOAM
◦ *MAT_JOINTED_ROCK
◦ *MAT_SIMPLIFIED_RUBBER
◦ *MAT_FHWA_SOIL
◦ *MAT_SCHWER_MURRAY_CAP_MODEL
• Failure time added to MAT_EROSION for solid elements.
• Damping in the material models *MAT_LOW_DENSITY_FOAM and *MAT_LOW_-
DENSITY_VISCOUS_FOAM can now be a tabulated function of the smallest stretch
ratio.
• The material model *MAT_PLASTICITY_WITH_DAMAGE allows the table
definitions for strain rate.
• Improvements in the option *INCLUDE_STAMPED_PART now allow all history
data to be mapped to the crash part from the stamped part. Also, symmetry planes
can be used to allow the use of a single stamping to initialize symmetric parts.
• Extensive improvements in trimming result in much better elements after the
trimming is completed. Also, trimming can be defined in either a local or global
coordinate system. This is a new option in *DEFINE_CURVE_TRIM.
• An option to move parts close before solving the contact problem is available, see
*CONTACT_AUTO_MOVE.
• An option to add or remove discrete beams during a calculation is available with the
new keyword: *PART_SENSOR.
• Multiple jetting is now available for the Hybrid and Chemkin airbag inflator
models.
• Nearly all constraint types are now handled for implicit solutions.
• Calculation of constraint and attachment modes can be easily done by using the
option: *CONTROL_IMPLICIT_MODES.
• Penalty option, see *CONTROL_CONTACT, now applies to all *RIGIDWALL
options and is always used when solving implicit problems.
• Solid elements types 3 and 4, the 4 and 8 node elements with 6 degrees-of-freedom
per node, are available for implicit solutions.
• The warping stiffness option for the Belytschko-Tsay shell is implemented for
implicit solutions. The Belytschko-Wong-Chang shell element is now available for
implicit applications. The full projection method is implemented due to it accuracy
over the drill projection.

LS-DYNA DEV 10/27/16 (r:8004) 2-23 (History of LS-DYNA)


History of LS-DYNA LS-DYNA Theory Manual

• Rigid to deformable switching is implemented for implicit solutions.


• Automatic switching can be used to switch between implicit and explicit calcula-
tions. See the keyword: *CONTROL_IMPLICIT_GENERAL.
• Implicit dynamics rigid bodies are now implemented. See the keyword *CON-
TROL_IMPLICIT_DYNAMIC.
• Eigenvalue solutions can be intermittently calculated during a transient analysis.
• A linear buckling option is implemented. See the new control input: *CONTROL_-
IMPLICIT_BUCKLE
• Implicit initialization can be used instead of dynamic relaxation. See the keyword
*CONTROL_DYNAMIC_RELAXATION where the parameter, IDFLG, is set to 5.
• Superelements, i.e., *ELEMENT_DIRECT_MATRIX_INPUT, are now available for
implicit applications.
• There is an extension of the option, *BOUNDARY_CYCLIC, to symmetry planes in
the global Cartesian system. Also, automatic sorting of nodes on symmetry planes
is now done by LS-DYNA.
• Modeling of wheel-rail contact for railway applications is now available, see
*RAIL_TRACK and *RAIL_TRAIN.
• A new, reduced CPU, element formulation is available for vibration studies when
elements are aligned with the global coordinate system. See *SECTION_SOLID and
*SECTION_SHELL formulation 98.
• An option to provide approximately constant damping over a range of frequen-
cies is implemented, see *DAMPING_FREQUENCY_RANGE.

2-24 (History of LS-DYNA) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Preliminaries

3
Preliminaries

NOTE: Einstein summation convention is used. For each re-


peated index there is an implied summation.

Consider the body shown in Figure 3.1. We are interested in time-dependent


deformation for which a point in b initially at 𝑋𝛼 (𝛼 = 1, 2, 3) in a fixed rectangular
Cartesian coordinate system moves to a point 𝑥𝑖 (𝑖 = 1, 2, 3) in the same coordinate system.
Since a Lagrangian formulation is considered, the deformation can be expressed in terms of
the convected coordinates 𝑋𝛼 , and time 𝑡
𝑥𝑖 = 𝑥𝑖 (𝑋𝛼 , 𝑡). (3.1)
At time 𝑡 = 0, we have the initial conditions
𝑥𝑖 (𝐗, 0) = 𝑋i
(3.2)
𝑥̇𝑖 (𝐗, 0) = 𝑉𝑖 (𝐗)
where 𝐕 is the initial velocity.

3.1 Governing Equations


We seek a solution to the momentum equation
𝜎𝑖𝑗,𝑗 + 𝜌𝑓𝑖 = 𝜌𝑥̈𝑖 (3.3)
satisfying the traction boundary conditions,
𝜎𝑖𝑗 𝑛𝑗 = 𝑡𝑖 (𝑡), (3.4)
on boundary 𝜕𝑏1 , the displacement boundary conditions,
𝑥𝑖 (𝑋𝛼 , 𝑡) = 𝐷𝑖 (𝑡), (3.5)
on boundary 𝜕𝑏2 , and the contact discontinuity condition,
(𝜎𝑖𝑗+ − 𝜎𝑖𝑗− )𝑛𝑖 = 0, (3.6)

LS-DYNA DEV 10/27/16 (r:8004) 3-1 (Preliminaries)


Preliminaries LS-DYNA Theory Manual

along an interior boundary 𝜕b3 when 𝑥𝑖+ = 𝑥𝑖− . Here 𝛔 is the Cauchy stress, 𝜌 is the current
density, 𝐟 is the body force density, and 𝐱̈ is acceleration. The comma on 𝜎𝑖𝑗,𝑗 denotes
covariant differentiation, and 𝑛𝑗 is a unit outward normal to a boundary element on ∂b.

Mass conservation is trivially stated as


𝜌𝑉 = 𝜌0 (3.7)
where 𝑉 is the relative volume, i.e., the determinant of the deformation gradient matrix, 𝐹𝑖𝑗 ,
∂𝑥𝑖
𝐹𝑖𝑗 = (3.8)
∂𝑋𝑗
and 𝜌0 is the reference density. The energy equation
𝐸̇ = 𝑉𝑠𝑖𝑗 𝜀̇𝑖𝑗 − (𝑝 + 𝑞)𝑉̇ (3.9)

is integrated in time and is used for evaluating equations of state and to track the global
energy balance. In Equation (3.9), 𝑠𝑖𝑗 and 𝑝 represent the deviatoric stresses and pressure,
𝑠𝑖𝑗 = 𝜎𝑖𝑗 + (𝑝 + 𝑞)𝛿𝑖𝑗 (3.10)

1
𝑝 = − 𝜎𝑖𝑗 𝛿𝑖𝑗 − 𝑞
3
(3.11)
1
= − 𝜎𝑘𝑘 − 𝑞
3
respectively, where 𝑞 is the bulk viscosity, 𝛿𝑖𝑗 is the Kronecker delta (𝛿𝑖𝑗 = 1 if 𝑖 = 𝑗;
otherwise 𝛿𝑖𝑗 = 0), and 𝜀̇𝑖𝑗 is the strain rate tensor. The strain rates and bulk viscosity are
discussed later.

3-2 (Preliminaries) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Preliminaries

X3
x3

x2 X2

∂b
t=0 b

∂B

B0

X1
x1
Figure 3.1. Notation.

We can write:

∫ (𝜌𝑥̈𝑖 − 𝜎𝑖𝑗,𝑗 − 𝜌𝑓 )𝛿𝑥𝑖 𝑑𝜐 + ∫ (𝜎𝑖𝑗 𝑛𝑗 − 𝑡𝑖 )𝛿𝑥𝑖 𝑑𝑠 + ∫ (𝜎𝑖𝑗+ − 𝜎𝑖𝑗− )𝑛𝑗 𝛿𝑥𝑖 𝑑𝑠 = 0 (3.12)
𝑣 𝜕𝑏1 𝜕𝑏3

where 𝛿𝑥𝑖 satisfies all boundary conditions on 𝜕𝑏2 , and the integrations are over the current
geometry. Application of the divergence theorem gives

∫ (𝜎𝑖𝑗 𝛿𝑥𝑖 ) 𝑑𝜐 = ∫ 𝜎𝑖𝑗 𝑛𝑗 𝛿𝑥𝑖 𝑑𝑠 + ∫ (𝜎𝑖𝑗+ − 𝜎𝑖𝑗− )𝑛𝑗 𝛿𝑥𝑖 𝑑𝑠 (3.13)


𝜐 ,𝑗 ∂𝑏1 ∂𝑏3

and noting that


(𝜎𝑖𝑗 𝛿𝑥𝑖 ),𝑗 − 𝜎𝑖𝑗,𝑗 𝛿𝑥𝑖 = 𝜎𝑖𝑗 𝛿𝑥𝑖,𝑗 (3.14)
leads to the weak form of the equilibrium equation,

𝛿𝜋 = ∫ 𝜌𝑥̈𝑖 𝛿𝑥𝑖 𝑑𝜐 + ∫ 𝜎𝑖𝑗 𝛿𝑥𝑖,𝑗 𝑑𝜐 − ∫ 𝜌𝑓𝑖 𝛿𝑥𝑖 𝑑𝜐 − ∫ 𝑡𝑖 𝛿𝑥𝑖 𝑑𝑠 = 0, (3.15)


𝜐 𝜐 𝜐 ∂𝑏1

which is a statement of the principle of virtual work.

We superimpose a mesh of finite elements interconnected at nodal points on the


reference configuration and track particles through time, i.e.,
𝑘
 𝑗
𝑥𝑖 (𝑋𝛼 , 𝑡) = 𝑥𝑖 (𝑋𝛼 (𝜉 , 𝜂, 𝜁 ), 𝑡) = ∑ 𝑁𝑗 (𝜉 , 𝜂, 𝜁 )𝑥𝑖 (𝑡) (3.16)
𝑗=1

LS-DYNA DEV 10/27/16 (r:8004) 3-3 (Preliminaries)


Preliminaries LS-DYNA Theory Manual

where 𝑁𝑗 are shape (interpolation) functions in the parametric coordinates (𝜉 , 𝜂, 𝜁 ), 𝑘 is the


 𝑗
number of nodal points defining the element, and 𝑥𝑖 is the nodal coordinate of the jth node
in the ith direction. Each shape function has a finite support that is limited to the elements
for which its associated node is a member (hence the name finite element method).
Consequently, within each element the interpolation only depends on the nodal values for
the nodes in that element, and hence expressions like Equation (3.16)are meaningful.

The condition 𝛿𝜋 = 0 holds for all variations, 𝛿𝑥𝑖 , and, in particular, it holds for
variations along the shape functions. In each of the 3 Cartesian directions upon setting the
variation to one of the shape functions the weak form reduces to a necessary (but not
sufficient) condition that must be satisfied by any solution so that the
number of equations = 3 × number of nodes.
At this stage it is useful to introduce a vector space having dimension ℝ(number of nodes) with
a corresponding cartesian basis {𝐞′𝑖 }number
𝑖=1
of nodes
. Since the body is discretized into 𝑛
disjoint elements, the integral in (3.15) may be separated using the spatial additively of
integration into 𝑛 terms, one for each element
𝑛
𝛿𝜋 = ∑ 𝛿𝜋𝑚 = 0. (3.17)
𝑚=1

The contribution from each element is

𝛿𝜋𝑚 = ∫ 𝜌𝑥̈𝑖 𝛿𝑥𝑖 𝑑𝜐 + ∫ 𝜎𝑖𝑗 𝛿𝑥𝑖,𝑗 𝑑𝜐 − ∫ 𝜌𝑓𝑖 𝛿𝑥𝑖 𝑑𝜐 − ∫ 𝑡𝑖 𝛿𝑥𝑖 𝑑𝑠. (3.18)
𝜐𝑚 𝜐𝑚 𝜐𝑚 ∂𝑏1 ∩𝜕𝑣𝑚

Assembling the element contributions back into a system of equations leads to


𝑛
∑ {∫ 𝜌𝑥̈𝑖 (𝐞i ⨂𝛎𝑚 )𝑑𝜐 + ∫ 𝜎𝑖𝑗𝑚 (𝐞i ⨂𝛎𝑚 𝑚
,𝑗 )𝑑𝜐 − ∫ 𝜌𝑓𝑖 (𝐞i ⨂𝛎 )𝑑𝜐
𝜐𝑚 𝜐𝑚 𝜐𝑚
𝑚=1
(3.19)
−∫ 𝑡𝑖 (𝐞i ⨂𝛎𝑚 )𝑑𝑠} = 0.
∂𝑏1 ∩𝜕𝑣𝑚

In which
𝑘
𝛎 = ∑ 𝑁𝑖 𝐞𝑛′ 𝑚 (𝑖)
𝑚
(3.20)
𝑖=1

where 𝑛𝑚 (𝑖) is the global node number.

Applying the approximation scheme of Equation (3.16) to the dependent variables


and substituting into Equation (3.19) yields
𝑛
∑ {∫ 𝜌𝐍T𝑚 𝐍𝑚 𝐚𝑑𝜐 + ∫ 𝐁T𝑚 𝛔𝑑𝜐 − ∫ 𝜌𝐍T𝑚 𝐛𝑑𝜐 − ∫ 𝐍T𝑚 𝐭𝑑𝑠} = 0 (3.21)
𝜐𝑚 𝜐𝑚 𝜐𝑚 ∂𝑏1
𝑚=1

where 𝐍 is an interpolation matrix; 𝜎 is the stress vector


𝛔T = (𝜎𝑥𝑥 , 𝜎𝑦𝑦 , 𝜎𝑧𝑧 , 𝜎𝑥𝑦 , 𝜎𝑦𝑧 , 𝜎zx ); (3.22)

3-4 (Preliminaries) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Preliminaries

B is the strain-displacement matrix; a is the nodal acceleration vector


𝑎1
⎡ 𝑥2 ⎤
𝑥̈1 ⎢ 𝑎𝑥 ⎥
⎡𝑥̈ ⎤ = 𝐍 ⎢ ⋮ ⎥ = 𝐍𝐚; (3.23)
⎢ 2⎥ ⎢ ⎥
⎣𝑥̈3 ⎦ ⎢𝑎𝑘 ⎥
⎢ 𝑦⎥
⎣𝑎𝑘𝑧 ⎦
b is the body force load vector; and 𝒕 is the applied traction load.
𝑓 𝑡𝑥
⎡ 𝑥⎤
𝐭 = ⎢ 𝑡𝑦 ⎤

⎢𝑓𝑦 ⎥
𝐛=⎢ ⎥, ⎥ (3.24)
⎣𝑓𝑧 ⎦ ⎣ 𝑡𝑧 ⎦

LS-DYNA DEV 10/27/16 (r:8004) 3-5 (Preliminaries)


LS-DYNA Theory Manual Solid Elements

4
Solid Elements

For a mesh of 8-node hexahedron solid elements, Equation ((3.16)) becomes:


8
 𝑗
𝑥𝑖 (𝑋𝛼 , 𝑡) = 𝑥𝑖 (𝑋𝛼 (𝜉 , 𝜂, 𝜁 ), 𝑡) = ∑ 𝜙𝑗 (𝜉 , 𝜂, 𝜁 )𝑥𝑖 (𝑡). (4.1)
𝑗=1

The shape function 𝜙𝑗 is defined for the 8-node hexahedron as


1
𝜙𝑗 = (1 + 𝜉 𝜉𝑗 )(1 + 𝜂𝜂𝑗 )(1 + 𝜁 𝜁𝑗 ), (4.2)
8
𝑗
where 𝜉𝑗 , 𝜂𝑗 , 𝜁𝑗 take on their nodal values of (±1, ±1, ±1) and 𝑥𝑖 is the nodal coordinate of the
jth node in the ith direction (see Figure 4.1).

For a solid element, N is the 3 × 24 rectangular interpolation matrix given by


𝜙1 0 0 𝜙2 0 ⋯ 0 0
⎡0 𝜙1 0 0 𝜙2 ⋯ 𝜙8 0 ⎤
N(𝜉 , 𝜂, 𝜁 ) = ⎢ ⎥, (4.3)
⎣0 0 𝜙1 0 0 ⋯ 0 𝜙8 ⎦
𝝈 is the stress vector
σ T = (𝜎𝑥𝑥 𝜎𝑦𝑦 𝜎𝑧𝑧 𝜎𝑥𝑦 𝜎𝑦𝑧 𝜎𝑧𝑥 ). (4.4)

LS-DYNA DEV 10/27/16 (r:8004) 4-1 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

5
8

6
Node ξ η ζ
7 1 -1 -1 -1
2 1 -1 -1
3 1 1 -1
4 -1 1 -1
5 -1 -1 1
1 4
6 1 -1 1
7 1 1 1
8 -1 1 1

2
3

Figure 4.1. Eight node solid hexahedron element.

𝐁 is the 6 × 24 strain-displacement matrix



⎡ 0 0 ⎤
⎢∂𝑥 ⎥
⎢ ∂ ⎥
⎢0 0 ⎥
⎢ ∂𝑦 ⎥
⎢ ⎥
⎢ ∂ ⎥
⎢0 0
∂𝑧 ⎥
𝐁=⎢∂ ∂
⎥ 𝐍. (4.5)
⎢ ⎥
⎢ 0 ⎥
⎢∂𝑦 ∂𝑥 ⎥
⎢ ∂ ∂⎥
⎢0 ⎥
⎢ ∂𝑧 ∂𝑦⎥
⎢ ⎥
⎢∂ ∂⎥
⎣∂𝑧 0 ∂𝑥⎦
In order to achieve a diagonal mass matrix the rows are summed giving the kth diagonal
term as
8
𝑚𝑘𝑘 = ∫ 𝜌𝜙𝑘 ∑ 𝜙𝑖 𝑑𝜐 = ∫ 𝜌𝜙𝑘 𝑑𝜐, (4.6)
𝜐 𝜐
𝑖=1

since the basis functions sum to unity.

Terms in the strain-displacement matrix are readily calculated. Note that

4-2 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

∂𝜙𝑖 ∂𝜙𝑖 ∂𝑥 ∂𝜙𝑖 ∂𝑦 ∂𝜙𝑖 ∂𝑧


= + + ,
∂𝜉 ∂𝑥 ∂𝜉 ∂𝑦 ∂𝜉 ∂𝑧 ∂𝜉
∂𝜙𝑖 ∂𝜙𝑖 ∂𝑥 ∂𝜙𝑖 ∂𝑦 ∂𝜙𝑖 ∂𝑧
= + + , (4.7)
∂𝜂 ∂𝑥 ∂𝜂 ∂𝑦 ∂𝜂 ∂𝑧 ∂𝜂
∂𝜙𝑖 ∂𝜙𝑖 ∂𝑥 ∂𝜙𝑖 ∂𝑦 ∂𝜙𝑖 ∂𝑧
= + + ,
∂𝜁 ∂𝑥 ∂𝜁 ∂𝑦 ∂𝜁 ∂𝑧 ∂𝜁
which can be rewritten as
∂𝜙 ∂𝑥 ∂𝑦 ∂𝑧 ∂𝜙𝑖 ∂𝜙𝑖
⎡ 𝑖⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
⎢ ∂𝜉 ⎥ ⎢∂𝜉 ∂𝜉 ∂𝜉 ⎥ ⎢ ∂𝑥 ⎥ ⎢ ∂𝑥 ⎥
⎢∂𝜙 ⎥ ⎢∂𝑥 ∂𝑧 ⎥
⎢ 𝑖⎥ ⎢ ∂𝑦 ⎥⎢ ∂𝜙𝑖 ⎥ ⎢∂𝜙 ⎥
⎥ = 𝐉 ⎢ 𝑖 ⎥.
⎢ ⎥=⎢ ⎥⎢ (4.8)
⎢ ∂𝜂 ⎥ ⎢∂𝜂 ∂𝜂 ∂𝜂⎥ ⎢ ∂𝑦 ⎥ ⎢ ∂𝑦 ⎥
⎢∂𝜙 ⎥ ⎢∂𝑥 ⎥ ⎢ ⎥ ⎢ ⎥
⎢ 𝑖⎥ ⎢ ∂𝑦 ∂𝑧 ⎥ ⎢∂𝜙𝑖 ⎥ ⎢∂𝜙𝑖 ⎥
⎣ ∂𝜁 ⎦ ⎣∂𝜁 ∂𝜁 ⎣
∂𝜁 ⎦ ∂𝑧 ⎦ ⎣ ∂𝑧 ⎦
Inverting the Jacobian matrix, J, we can solve for the desired terms
∂𝜙 ∂𝜙
⎡ 𝑖⎤ ⎡ 𝑖⎤
⎢ ∂𝑥 ⎥ ⎢ ∂𝜉 ⎥
⎢∂𝜙𝑖 ⎥ ⎢∂𝜙 ⎥
⎢ ⎥ = 𝐉−1 ⎢
⎢ ⎥.
𝑖⎥
(4.9)
⎢ ∂𝑦 ⎥ ⎢ ∂𝜂 ⎥
⎢ ⎥ ⎢∂𝜙 ⎥
⎢∂𝜙𝑖 ⎥ ⎢ 𝑖⎥
⎣ ∂𝑧 ⎦ ⎣ ∂𝜁 ⎦

4.1 Volume Integration


Volume integration is carried out with Gaussian quadrature. If 𝑔 is some function
defined over the volume, and 𝑛 is the number of integration points, then
1 1 1
∫ 𝑔𝑑𝜐 = ∫ ∫ ∫ 𝑔|𝐉|𝑑𝜉𝑑𝜂𝑑𝜁 , (4.10)
𝜐 −1 −1 −1

is approximated by
𝑛 𝑛 𝑛
∑ ∑ ∑ 𝑔𝑗𝑘𝑙 ∣𝐽𝑗𝑘𝑙 ∣𝑤𝑗 𝑤𝑘 𝑤𝑙 , (4.11)
𝑗=1 𝑘=1 𝑙=1

where 𝑤𝑗 , 𝑤𝑘 , 𝑤𝑙 are the weighting factors,


𝑔𝑗𝑘𝑙 = g(𝜉𝑗 , 𝜂𝑘 , 𝜁𝑙 ), (4.12)
and 𝐽 is the determinant of the Jacobian matrix. For one-point quadrature
𝑛 = 1,
(4.13)
𝑤𝑖 = 𝑤𝑗 = 𝑤𝑘 = 2,

LS-DYNA DEV 10/27/16 (r:8004) 4-3 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

Flanagan Wilkins
DYNA3D Belytschko FDM
Strain displacement matrix 94 357 843
Strain rates 87 156
Force 117 195 270
Subtotal 298 708 1113
Hourglass control 130 620 680
Total 428 1328 1793

Table 4.1. Operation counts for a constant stress hexahedron (includes adds,
subtracts, multiplies, and divides in major subroutines, and is independent of
vectorization). Material subroutines will add as little as 60 operations for the
bilinear elastic-plastic routine to ten times as much for multi-surface plasticity and
reactive flow models. Unvectorized material models will increase that share of
the cost a factor of four or more.

𝜉1 = 𝜂1 = 𝜁1 = 0,
and we can write

∫ 𝑔𝑑𝑣 = 8𝑔(0,0,0)|𝐉(0,0,0)|. (4.14)

Note that 8|𝐽(0,0,0)| approximates the element volume.

Perhaps the biggest advantage to one-point integration is a substantial savings in


computer time. An anti-symmetry property of the strain matrix
∂𝜙1 ∂𝜙 ∂𝜙3 ∂𝜙
=− 7, = − 5,
∂𝑥𝑖 ∂𝑥𝑖 ∂𝑥𝑖 ∂𝑥𝑖
(4.15)
∂𝜙2 ∂𝜙 ∂𝜙4 ∂𝜙
= − 8, = − 6,
∂𝑥𝑖 ∂𝑥𝑖 ∂𝑥𝑖 ∂𝑥𝑖
at 𝜉 = 𝜂 = 𝜁 = 0 reduces the amount of effort required to compute this matrix by more than
25 times over an 8-point integration. This cost savings extends to strain and element nodal
force calculations where the number of multiplies is reduced by a factor of 16. Because
only one constitutive evaluation is needed, the time spent determining stresses is reduced
by a factor of 8. Operation counts for the constant stress hexahedron are given in Table 4.1.
Included are counts for the Flanagan and Belytschko [1981] hexahedron and the
hexahedron used by Wilkins [1974] in his integral finite difference method, which was also
implemented [Hallquist 1979].

It may be noted that 8-point integration has another disadvantage in addition to


cost. Fully integrated elements used in the solution of plasticity problems and other
4-4 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Solid Elements

problems where Poisson’s ratio approaches 0.5 lock up in the constant volume bending
modes. To preclude locking, an average pressure must be used over the elements;
consequently, the zero energy modes are resisted by the deviatoric stresses. if the
deviatoric stresses are insignificant relative to the pressure or, even worse, if material
failure cause loss of this stress state component, hourglassing will still occur, but with no
means of resisting it. Sometimes, however, the cost of the fully integrated element may be
justified by increased reliability and if used sparingly may actually increase the overall
speed.

LS-DYNA DEV 10/27/16 (r:8004) 4-5 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

4.2 Solid Element 2


Solid element 2 is a selective reduced (S/R) integrated element that in general is regarded
as too stiff. In particular this is the case when the elements exhibit poor aspect ratio, i.e.,
when one element dimension is significantly smaller than the other(s). This occurs for
instance when modelling thin walled structures and the time for solving the problem
prevents using a sufficient number of elements for maintaining close to cubic elements
throughout the structure. The reason for the locking phenomenon is that the element is not
able to represent pure bending modes without introducing transverse shear strains, and
this may be bad enough to lock the element to a great extent. In an attempt to solve this
transverse shear locking problem, two new fully integrated solid elements are introduced
and documented herein that may become of practical use for these types of applications.

4.2.1 Brief summary of solid element 2


Let 𝑥𝐼𝑖 represent the nodal coordinate of dimension 𝑖 and node 𝐼, and likewise 𝑣𝐼𝑖 its
velocity. Furthermore denote
1
𝑁𝐼 (𝜉1 , 𝜉2 , 𝜉3 ) = (1 + 𝜉1𝐼 𝜉1 + 𝜉2𝐼 𝜉2 + 𝜉3𝐼 𝜉3 + 𝜉12
𝐼 𝐼
𝜉1 𝜉2 + 𝜉13 𝐼
𝜉1 𝜉3 + 𝜉23 𝐼
𝜉2 𝜉3 + 𝜉123 𝜉1 𝜉2 𝜉3 ),(4.2.16)
8
the shape functions for the standard isoparametric domain where
𝜉1∗ = [−1 1 1 −1 −1 1 1 −1],
𝜉2∗ = [−1 −1 1 1 −1 −1 1 1],
𝜉3∗ = [−1 −1 −1 −1 1 1 1 1],

𝜉12 =[ 1 −1 1 −1 1 −1 1 −1], (4.2.17)

𝜉13 =[ 1 −1 −1 1 1 −1 1 −1],

𝜉23 =[ 1 1 −1 −1 −1 −1 1 1],

𝜉123 = [−1 1 −1 1 1 −1 1 −1],
and let furthermore
𝐼 𝐼
𝜉21 = 𝜉12 ,
𝐼 𝐼
𝜉32 = 𝜉23 , (4.2.18)
𝐼 𝐼
𝜉31 = 𝜉13 .

The isoparametric representation of the coordinates of a material point in the


element is then given as (where the dependence on 𝜉1 , 𝜉2 , 𝜉3 is suppressed for brevity)
𝑥𝑖 = 𝑥𝐼𝑖 𝑁𝐼 , (4.2.19)
and its associated jacobian matrix is

4-6 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

p
Figure 4.2.2. Bending mode for a fully integrated brick.

𝜕𝑥𝑖 1
𝐽𝑖𝑗 = = 𝑥𝐼𝑖 (𝜉𝑗𝐼 + 𝜉𝑗𝑘𝐼 𝜉𝑘 + 𝜉𝑗𝑙𝐼 𝜉𝑙 + 𝜉123
𝐼
𝜉𝑘 𝜉𝑙 ), (4.2.20)
𝜕𝜉𝑗 8
where 𝑘 = 1 + mod(𝑗, 3) and 𝑙 = 1 + mod(𝑗 + 1,3). For future reference let
1
𝐽𝑖𝑗0 = 𝑥𝐼𝑖 (0) 𝜉𝑗𝐼 , (4.2.21)
8
be the jacobian evaluated in the element center and in the beginning of the simulation (i.e.,
at time zero). The velocity gradient computed directly from the shape functions and
velocity components is
𝜕𝑣𝑖 −1
𝐿𝑖𝑗 = = 𝐽𝑖𝑘̇ 𝐽𝑘𝑗 = 𝐵𝑖𝑗𝐼𝑘 𝑣𝐼𝑘 , (4.2.22)
𝜕𝑥𝑗
where
𝜕𝑁𝐼 −1
𝐵𝑖𝑗𝐼𝑘 = 𝐽 𝛿 , (4.2.23)
𝜕𝜉𝑙 𝑙𝑗 𝑖𝑘
is the gradient-displacement matrix and represents the element except for the alleviation of
volumetric locking. To do just that, let 𝐵0𝑖𝑗𝐼𝑘 be defined by

𝐽𝑖𝑘̅̇ 𝐽𝑘𝑗
̅ = 𝐵0𝑖𝑗𝐼𝑘 𝑣𝐼𝑘 ,
−1
(4.2.24)

with 𝐽𝑖𝑗̅ being the element averaged jacobian matrix, and construct the gradient-
displacement matrix used for the element as
1
̅̅̅̅ 𝑖𝑗𝐼𝑘 = 𝐵𝑖𝑗𝐼𝑘 + (𝐵0𝑙𝑙𝐼𝑘 − 𝐵𝑙𝑙𝐼𝑘 )𝛿𝑖𝑗 .
𝐵 (4.2.25)
3
This is what is often called the B-bar method.

4.2.2 Transverse shear locking example


To get the idea of the modifications needed to alleviate transverse shear locking let’s look at
the parallelepiped of dimensions 𝑙1 × 𝑙2 × 𝑙3 in the Figure above. For this simple geometry
the jacobian matrix is diagonal and the velocity gradient is expressed as

LS-DYNA DEV 10/27/16 (r:8004) 4-7 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

2 1
𝐿𝑖𝑗 = 𝐽𝑖𝑗̇ = 𝑣𝐼𝑖 (𝜉𝑗𝐼 + 𝜉𝑗𝑘𝐼 𝜉𝑘 + 𝜉𝑗𝑙𝐼 𝜉𝑙 + 𝜉123
𝐼
𝜉𝑘 𝜉𝑙 ), (4.2.26)
𝑙𝑗 4𝑙𝑗
where, again, 𝑘 = 1 + mod(𝑗, 3) and 𝑙 = 1 + mod(𝑗 + 1,3). Now let 𝑖 ≠ 𝑝 ≠ 𝑞 ≠ 𝑖, then a
pure bending mode in the plane with normal in direction 𝑞 and about axis 𝑝 is represented
by
𝑣𝐼𝑖 = 𝜉𝑖𝑞𝐼 ,
𝑣𝐼𝑝 = 0, (4.2.27)
𝑣𝐼𝑞 = 0,
and thus the velocity gradient is given as
1
𝐿𝑖𝑗 = (𝜉 𝐼 𝜉 𝐼 𝜉 + 𝜉𝑖𝑞𝐼 𝜉𝑗𝑙𝐼 𝜉𝑙 ),
4𝑙𝑗 𝑖𝑞 𝑗𝑘 𝑘
(4.2.28)
𝐿𝑝𝑗 = 0,
𝐿𝑞𝑗 = 0,
for 𝑗 = 1, 2, 3. The nonzero expression above amounts to
1
𝐿𝑖𝑖 = 𝜉 ,
4𝑙𝑖 𝑞
𝐿𝑖𝑝 = 0, (4.2.29)
1
𝐿𝑖𝑞 = 𝜉.
4𝑙𝑞 𝑖
Notable here is that a pure bending mode gives arise to a transverse shear strain
represented by the last expression in the above. Assuming that 𝑙𝑞 is small compared to 𝑙𝑖
this may actually lock the element.

4.2.3 Solid element -2


Given this insight the modifications in the expression of the jacobian matrix are as follows.
Let
0 0
√𝐽1𝑚 0 0 0 0
⎛ 𝐽1𝑚 + 𝐽2𝑚 𝐽2𝑚 + 𝐽3𝑚 𝐽3𝑚 ⎞
𝜅𝑚𝑛 ⎜
⎜1,
= min ⎜ ⎟
⎟,
⎟ (4.2.30)
√ 0 0 0 0 0 0
⎝ 𝐽 𝐽
1𝑛 1𝑛 + 𝐽 𝐽
2𝑛 2𝑛 + 𝐽 𝐽
3𝑛 3𝑛 ⎠

be the aspect ratio between dimensions 𝑚 and 𝑛 at time zero. The modified jacobian is
written
1
𝐽𝑖𝑗̃ = 𝑥𝐼𝑖 (𝜉𝑗𝐼 + 𝜉𝑗𝑘𝐼 𝜉𝑘𝑖 + 𝜉𝑗𝑙𝐼 𝜉𝑙𝑖 + 𝜉123
𝐼
𝜉𝑘𝑖 𝜉𝑙𝑖 ), (4.2.31)
8
where

4-8 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

𝜉𝑘 𝜅𝑗𝑘 𝑖≠𝑗
𝜉𝑘𝑖 = { , (4.2.32)
𝜉𝑘 otherwise
and
𝜉𝑙 𝜅𝑗𝑙 𝑖≠𝑗
𝜉𝑙𝑖 = { . (4.2.33)
𝜉𝑙 otherwise

The velocity gradient is now given as


𝐿𝑖𝑗 = 𝐽𝑖𝑘̃̇ 𝐽𝑘𝑗
−1
̃ = 𝐵̃𝑖𝑗𝐼𝑘 𝑣𝐼𝑘 , (4.2.34)

where 𝐵̃𝑖𝑗𝐼𝑘 is the gradient-displacement matrix used for solid element type -2 in LS-DYNA.
The B-bar method is used to prevent volumetric locking.

4.2.4 Transverse shear locking example revisited


Once again let’s look at the parallelepiped of dimensions 𝑙1 × 𝑙2 × 𝑙3 . The jacobian matrix is
still diagonal and the velocity gradient is with the new element formulation expressed as
2 ̇ 1
𝐿𝑖𝑗 = 𝐽𝑖𝑗̃ = 𝑣 (𝜉 𝐼 + 𝜉𝑗𝑘𝐼 𝜉𝑘𝑖 + 𝜉𝑗𝑙𝐼 𝜉𝑙𝑖 + 𝜉123
𝐼
𝜉𝑘𝑖 𝜉𝑙𝑖 ), (4.2.35)
𝑙𝑗 4𝑙𝑗 𝐼𝑖 𝑗
where, again, 𝑘 = 1 + mod(𝑗, 3) and 𝑙 = 1 + mod(𝑗 + 1,3). The velocity gradient for a pure
bending mode is now given as
1
𝐿𝑖𝑗 = (𝜉𝑖𝑞𝐼 𝜉𝑗𝑘𝐼 𝜉𝑘𝑖 + 𝜉𝑖𝑞𝐼 𝜉𝑗𝑙𝐼 𝜉𝑙𝑖 ), (4.2.36)
4𝑙𝑗
which amounts to (for the potential nonzero elements)
1
𝐿𝑖𝑖 = 𝜉𝑞 ,
4𝑙𝑖
𝐿𝑖𝑝 = 0, (4.2.37)
1
𝐿𝑖𝑞 = 𝜉𝜅 .
4𝑙𝑞 𝑖 𝑞𝑖

If we assume that this is the geometry in the beginning of the simulation and that 𝑙𝑞
is smaller than 𝑙𝑖 the transverse shear strain can be expressed as
1
𝐿𝑖𝑞 = 𝜉𝑖 , (4.2.38)
4𝑙𝑖
meaning that the transverse shear energy is not affected by poor aspect ratios, i.e., the
transverse shear strain does not grow with decreasing 𝑙𝑞 .

4.2.5 Solid element -1


Working out the details in the expression of the gradient-displacement matrix for solid
element type -2 reveals that this matrix is dense, i.e., there are 216 nonzero elements in this
matrix that needs to be processed compared to 72 for the standard solid element type 2. A

LS-DYNA DEV 10/27/16 (r:8004) 4-9 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

slight modification of the jacobian matrix will substantially reduce the computational
expense for this element. Simply substitute the expressions for 𝜉𝑘𝑖 and 𝜉𝑙𝑖 by
𝜉𝑘𝑖 = 𝜉𝑘 𝜅𝑗𝑘 , (4.2.39)
and
𝜉𝑙𝑖 = 𝜉𝑙 𝜅𝑗𝑙 . (4.2.40)

This will lead to a stiffness reduction for certain modes, in particular the out-of-
plane hourglass mode as can be seen by once again looking at the transverse shear locking
example. The velocity gradient for pure bending is now
1
𝐿𝑖𝑖 = 𝜉𝑞 𝜅𝑖𝑞 ,
4𝑙𝑖
𝐿𝑖𝑝 = 0, (4.2.41)
1
𝐿𝑖𝑞 = 𝜉𝜅 ,
4𝑙𝑞 𝑖 𝑞𝑖
and if it turns out that 𝑙𝑖 is smaller than 𝑙𝑞 , then this results in
1
𝐿𝑖𝑖 = 𝜉 . (4.2.42)
4𝑙𝑞 𝑞
That is, if 𝑖 represents the direction of the thinnest dimension, its corresponding bending
strain is inadequately reduced.

4.2.6 Example
A plate of dimensions 10 × 5 × 1 mm3 is clamped at one end and subjected to a 1 Nm
torque at the other end. The Young’s modulus is 210 GPa and the analytical solution for
the end tip deflection is 0.57143 mm. In order to study the mesh convergence for the three
fully integrated bricks the plate is discretized into 2 × 1 × 1, 4 × 2 × 2, 8 × 4 × 4, 16 × 8 × 8and
finally 32 × 16 × 16 elements, all elements having the same aspect ratio of 5 × 1. The table
below shows the results for the different fully integrated elements, and indicates an
accuracy improvement for solid elements −1 and −2.
Discretization Solid element type 2 Solid element type -2 Solid element type -1
2x1x1 0.0564 (90.1%) 0.6711 (17.4%) 0.6751 (18.1%)
4x2x2 0.1699 (70.3%) 0.5466 (4.3%) 0.5522 (3.4%)
8x4x4 0.3469 (39.3%) 0.5472 (4.2%) 0.5500 (3.8%)
16x8x8 0.4820 (15.7%) 0.5516 (3.5%) 0.5527 (3.3%)
32x16x16 0.5340 (6.6%) 0.5535 (3.1%) 0.5540 (3.1%)

4.3 Hourglass Control


The biggest disadvantage to one-point integration is the need to control the zero
energy modes, which arise, called hourglassing modes. Undesirable hourglass modes tend
to have periods that are typically much shorter than the periods of the structural response,
and they are often observed to be oscillatory. However, hourglass modes that have periods
that are comparable to the structural response periods may be a stable kinematic
4-10 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Solid Elements

component of the global deformation modes and must be admissible. One way of resisting
undesirable hourglassing is with a viscous damping or small elastic stiffness capable of
stopping the formation of the anomalous modes but having a negligible affect on the stable
global modes. Two of the early three-dimensional algorithms for controlling the hourglass
modes were developed by Kosloff and Frazier [1974] and Wilkins et al. [1974].

Since the hourglass deformation modes are orthogonal to the strain calculations,
work done by the hourglass resistance is neglected in the energy equation. This may lead
to a slight loss of energy; however, hourglass control is always recommended for the under
integrated solid elements. The energy dissipated by the hourglass forces reacting against
the formations of the hourglass modes is tracked and reported in the output files matsum
and glstat.

It is easy to understand the reasons for the formation of the hourglass modes.
Consider the following strain rate calculations for the 8-node solid element
1 8 ∂𝜙𝑘 𝑘 ∂𝜙𝑘 𝑘
𝜀̇𝑖𝑗 = (∑ 𝑥̇ + 𝑥̇ ). (4.43)
2 𝑘=1 ∂𝑥𝑖 𝑗 ∂𝑥𝑗 𝑖
Whenever diagonally opposite nodes have identical velocities, i.e.,
𝑥̇𝑖1 = 𝑥̇𝑖7 , 𝑥̇𝑖2 = 𝑥̇𝑖8 , 𝑥̇𝑖3 = 𝑥̇𝑖5 , 𝑥̇𝑖4 = 𝑥̇𝑖6 , (4.44)
the strain rates are identically zero:
𝜀̇𝑖𝑗 = 0, (4.45)

1k 2k

3k 4k
Figure 4.3. The hourglass modes of an eight-node element with one integration
point are shown [Flanagan and Belytschko 1981]. A total of twelve modes exist.

LS-DYNA DEV 10/27/16 (r:8004) 4-11 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

𝛼=1 𝛼=2 𝛼=3 𝛼=4


𝛤𝑗1 1 1 1 1
𝛤𝑗2 1 -1 -1 -1
𝛤𝑗3 -1 -1 1 1
𝛤𝑗4 -1 1 -1 -1
𝛤𝑗5 -1 -1 1 -1
𝛤𝑗6 -1 1 -1 1
𝛤𝑗7 1 1 1 -1
𝛤𝑗8 1 -1 -1 1

Table 4. Hourglass base vectors.

due to the asymmetries in Equations (4.15). It is easy to prove the orthogonality of the
hourglass shape vectors, which are listed in Table 4 and shown in Figure 4.3 with the
derivatives of the shape functions:
8
∂𝜙𝑘
∑ 𝛤 = 0, 𝑖 = 1, 2, 3, 𝛼 = 1, 2, 3, 4. (4.46)
𝑘=1
∂𝑥𝑖 𝛼𝑘
The product of the base vectors with the nodal velocities is zero when the element velocity
field has no hourglass component,
8
ℎ𝑖𝛼 = ∑ 𝑥̇𝑖𝑘 𝛤𝛼𝑘 = 0. (4.47)
𝑘=1
are nonzero if hourglass modes are present. The 12 hourglass-resisting force vectors, 𝑓𝑖𝛼𝑘 are
𝑓𝑖𝛼𝑘 = 𝑎ℎ ℎ𝑖𝛼 𝛤𝛼𝑘 , (4.48)
where
2⁄ 𝑐
𝑎ℎ = 𝑄HG 𝜌𝑣e 3 , (4.49)
4
in which 𝑣e is the element volume, 𝑐 is the material sound speed, and 𝑄HG is a user-defined
constant usually set to a value between .05 and .15. Equation (1.21) is hourglass control
type 1 in the LS-DYNA User’s Manual.

A shortcoming of hourglass control type 1 is that the hourglass resisting forces of Equation
(1.21) are not orthogonal to linear velocity field when elements are not in the shape of
parallelpipeds. As a consequence, such elements can generate hourglass energy with a
constant strain field or rigid body rotation. Flanagan and Belytschko [1981] developed an
hourglass control that is orthogonal to all modes except the zero energy hourglass modes.
Instead of resisting components of the bilinear velocity field that are orthogonal to the
strain calculation, Flanagan and Belytschko resist components of the velocity field that are
not part of a fully linear field. They call this field, defined below, the hourglass velocity
field
HG LIN
𝑥̇𝑖𝑘 = 𝑥̇𝑖𝑘 − 𝑥̇𝑖𝑘 , (4.50)

4-12 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

where
LIN
𝑥̇𝑖𝑘 ̇ (𝑥𝑗𝑘 − 𝑥̅𝑗 ),
= 𝑥̅i̇ + 𝑥̅𝑖,𝑗 (4.51)
and
1 8
𝑥̅𝑖 = ∑ 𝑥𝑖𝑘 ,
8 𝑘=1
(4.52)
1 8 𝑘
̇
𝑥̅𝑖 = ∑ 𝑥̇𝑖 .
8 𝑘=1
Flanagan and Belytschko construct geometry-dependent hourglass shape vectors that are
orthogonal to the fully linear velocity field and the rigid body field. With these vectors
they resist the hourglass velocity deformations. Defining hourglass shape vectors in terms
of the base vectors as
8
𝛾𝛼𝑘 = 𝛤𝛼𝑘 − 𝜙𝑘,𝑖 ∑ 𝑥𝑖𝑛 𝛤𝛼𝑛 , (4.53)
𝑛=1

the analogue for (4.47) is,


8
𝑔𝑖𝛼 = ∑ 𝑥̇𝑖𝑘 𝛾𝛼𝑘 = 0, (4.54)
𝑘=1

with the 12 resisting force vectors being


𝑓𝑖𝛼𝑘 = 𝑎ℎ 𝑔𝑖𝛼 𝛾𝛼𝑘 , (4.55)
where 𝑎ℎ is a constant given in Equation (4.48). Equation (1.28) corresponds to hourglass
control type 2 in the LS-DYNA User’s Manual. The 𝛾 terms used of equation of Equation
(1.26) are used not only type hourglass control type 2, but are the basis for all solid element
hourglass control except for form 1.

A cost comparison in Table 4.1 shows that the default type 1 hourglass viscosity
requires approximately 130 adds or multiplies per hexahedron, compared to 620 and 680
for the algorithms of Flanagan-Belytschko and Wilkins. Therefore, for a very regular mesh,
type 1 hourglass control may provide a faster, sufficiently accurate solution, but in general,
any of the other hourglass options which are all based on the 𝛾𝛼𝑘 terms of Equation (1.26)
will be a better choice.

Type 3 hourglass control is identical to type 2, except that the shape function
derivatives in Eq. (1.26) are evaluated at the centroid of the element rather than at the
origin of the referential coordinate system. With this method, Equation (1.14) produces the
exact element volume. However, the anti-symmetry property of Equation (1.15) is not true,
so there is some increased number of computations.

The remaining hourglass control types calculated hourglass forces proportional to


total hourglass deformation rather than hourglass viscosity. A stiffness form of hourglass
control allows elements to spring back and will absorb less energy than the viscous forms.

LS-DYNA DEV 10/27/16 (r:8004) 4-13 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

Types 4 and 5 hourglass control are similar to types 2 and 3, except that they
evaluate hourglass stiffness rather than viscosity. The hourglass rates of equation (1.27) are
multiplied by the solution time step to produce increments of hourglass deformation. The
hourglass stiffness is scaled by the element’s maximum frequency so that stability can is
maintained as long as the hourglass scale factor, 𝑎ℎ , is sufficiently small.

Type 6 hourglass control improves on type 5 by scaling the stiffness such that the
hourglass forces match those generated by a fully integrated element control by doing
closed form integration over the element volume scaling the hourglass stiffness by
matching the stabilization for the 3D hexahedral element is available for both implicit and
explicit solutions. Based on material properties and element geometry, this stiffness type
stabilization is developed by an assumed strain method [Belytschko and Bindeman 1993]
such that the element does not lock with nearly incompressible material. When the user
defined hourglass constant 𝑎h is set to 1.0, accurate coarse mesh bending stiffness is
obtained for elastic material. For nonlinear material, a smaller value of 𝑎h is suggested and
the default value is set to 0.1. In the implicit form, the assumed strain stabilization matrix
is:
𝐤 𝐤12 𝐤13
stab ⎡ 11 ⎤
𝐊 = 2𝜇𝑎h ⎢𝐤21 𝐤22 𝐤23 ⎥, (4.56)
⎣𝐤31 𝐤32 𝐤33 ⎦
where the 8 × 8 submatricies are calculated by:
1 1+𝜐 1
𝐤𝑖𝑖 ≡ 𝐻𝑖𝑖 [( ) (𝛄𝑗 𝛄𝑗T + 𝛄𝑘 𝛄𝑘T ) + ( ) 𝛄4 𝛄4T ] + (𝐻𝑗𝑗 + 𝐻𝑘𝑘 )𝛄𝑖 𝛄1T ,
1−𝜐 3 2 (4.57)
𝜐 1
𝐤𝑖𝑗 ≡ ,𝐻𝑖𝑗 [( ) 𝛄 𝛄T + 𝛄𝑖 𝛄𝑗T ]
1−𝜐 𝑗 𝑖 2
with,
2
𝐻𝑖𝑖 ≡ ∫(ℎ𝑗,𝑖 ) 𝑑𝑣 = ∫(ℎ𝑘,𝑖 )2 𝑑𝑣 = 3 ∫(ℎ4,𝑖 )2 𝑑𝑣,
𝑣 𝑣 𝑣 (4.58)
𝐻𝑖𝑗 ≡ ∫ ℎ𝑖,𝑗 ℎ𝑗,𝑖 𝑑𝑣,
𝑣
where,
ℎ1 = 𝜉𝜂 ℎ2 = 𝜂𝜁 ℎ3 = 𝜁𝜉 ℎ4 = 𝜉𝜂𝜁 , (4.59)

Subscripts 𝑖, 𝑗, and 𝑘 are permuted as in Table 44.2. A comma indicates a derivative


with respect to the spatial variable that follows. The hourglass vectors, 𝛾𝛼 are defined by
equation (4.53). The stiffness matrix is evaluated in a corotational coordinate system that is
aligned with the referential axis of the element. The use of a corotational system allows
direct evaluation of integrals in equations (4.58) by simplified equations that produce a
more accurate element than full integration.
T T
1 (𝚲𝑗 𝐱𝑗 )(𝚲𝑘 𝐱𝑘 ) (4.60)
𝐻𝑖𝑖 = 𝑖 ≠ 𝑗 ≠ 𝑘,
3 (𝚲𝑖T 𝐱𝑖 )
4-14 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Solid Elements

𝑖 𝑗 𝑘
1 2 3
1 3 2
2 3 1
2 1 3
3 1 2
3 2 1

Table 44.2. Permutations of i, j, and k.

1
𝐻𝑖𝑗 = (𝚲𝑘T 𝐱𝑘 ) 𝑖 ≠ 𝑗 ≠ 𝑘, (4.61)
3
Λ𝑖 are 8 × 1 matrices of the referential coordinates of the nodes as given in Figure 4.1, and
x𝑖 are 8 × 1 matrices of the nodal coordinates in the corotational system. For each material
type, a Poisson's ratio, 𝑣, and an effective shear modulus, 𝜇, is needed.

In the explicit form, the 12 hourglass force stabilization vectors are


4
𝐟𝑖stab = ∑ 𝑎h 𝑔𝑖𝛼 𝛄𝛼T , (4.62)
𝛼=1

where the 12 generalized stresses are calculated incrementally by


𝑛−1
𝑛
𝑔𝑖𝛼 𝑛−1
= 𝑔𝑖𝛼 + Δ𝑡𝑔̇𝑖𝛼 2 , (4.63)

and
𝑔̇𝑖𝑖 = 𝜇[(𝐻𝑗𝑗 + 𝐻𝑘𝑘 )𝑞𝑖𝑖̇ + 𝐻𝑖𝑗 𝑞𝑗𝑗̇ + 𝐻𝑖𝑘 𝑞𝑘𝑘
̇ ],
1
𝑔̇𝑖𝑗 = ,2𝜇 [ 𝐻 𝑞 ̇ + 𝜐𝐻𝑘𝑘 𝑞𝑖𝑖̇ ] (4.64)
1 − 𝜐 𝑖𝑖 𝑖𝑗
1+𝜈
𝑔̇𝑖4 = ,2𝜇 ( ) 𝐻𝑖𝑖 𝑞𝑖4
̇
3
where,
̇ = (𝛄𝛼T 𝐱̇𝑖 ).
𝑞𝑖𝛼 (4.65)
Subscripts 𝑖, 𝑗, and 𝑘 are permuted as per Table 44.2. As with the implicit form, calculations
are done in a corotational coordinate system in order to use the simplified equations (4.60)
and (4.61).

Type 7 hourglass control is very similar to type 6 hourglass control but with one significant
difference. As seen in Equation (1.36), type 6 obtains the current value of the generalized
stress from the previous value and the current increment. The incremental method is
nearly always sufficiently accurate, but it is possible for hourglass modes to fail to spring
back to the initial element geometry since the hourglass stiffness varies as the H terms
given by Equations (1.33) and (1.34) are recalculated in the deformed configuration each

LS-DYNA DEV 10/27/16 (r:8004) 4-15 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

cycle. Type 7 hourglass control eliminates this possible error by calculating the total
hourglass deformation in each cycle. For type 7 hourglass control, Equations (1.37) are
rewritten using 𝑔 and 𝑞 in place of 𝑔̇ and 𝑞 ̇, and Equation (1.38) is replaced by (1.39).
𝑞𝑖𝛼 = (𝛄𝛼T 𝐱𝑖 ) − (𝛄0𝛼
T
𝐱0𝑖 ). (4.39)
In Equation (1.39), 𝐱𝟎𝒊 and 𝛄𝟎𝜶 are evaluated using the initial, undeformed nodal coordinate
values. Type 7 hourglass control is considerably slower than type 6, so it is not generally
recommended, but may be useful when the solution involves at least several cycles of
loading and unloading that involve large element deformation of elastic or hyperelastic
material.

Both type 6 and 7 hourglass control are stiffness type methods, but may have viscosity
added through the VDC parameter on the *HOURGLASS card. The VDC parameter scales
the added viscosity, and VDC = 1.0 corresponds approximately to critical damping. The
primary motivation for damping is to reduce high frequency oscillations. A small
percentage of critical damping should be sufficient for this, but it is also possible to add
supercritical damping along with a small value of QM to simulate a very viscous material
that springs back slowly.

4.4 Puso Hourglass Control


Regarding the solid elements in LS-DYNA, the fully integrated brick uses selective-
reduced integration, which is known to alleviate volumetric locking but not shear locking
for elements with poor aspect ratio. The enhanced assumed strain methods have been the
most successful at providing coarse mesh accuracy for general non-linear material models.
In short, these elements tend to sacrifice computational efficiency for accuracy and are
hence of little interest in explicit analysis. Puso [2000] developed an enhanced assumed
strain element that combines coarse mesh accuracy with computational efficiency. It is
formulated as a single point integrated brick with an enhanced assumed strain physical
stabilization. In this project, we have implemented this element in LS-DYNA and made
comparisons with the assumed strain element developed by Belytschko and Bindeman
[1993] to see whether it brings anything new to the existing LS-DYNA element library.
This is hourglass control type 9.

The element formulation is that of Puso [2000], and is essentially the mean strain
hexahedral element by Flanagan and Belytschko [1981] in which the perturbation hourglass
control is substituted for an enhanced assumed strain stabilization force.
Given the matrices

4-16 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

1 −1 −1 −1 1 1 1 −1
⎡1⎤ ⎡ 1 −1 −1⎤ ⎡ 1 −1 −1 1⎤
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢1⎥ ⎢ 1 1 −1⎥ ⎢−1 −1 1 −1⎥
⎢1⎥ ⎢−1 1 −1 ⎥ ⎢−1 1 −1 1⎥
𝐒=⎢ ⎥
⎢1⎥ , 𝚵=⎢ ⎢−1 −1
⎥, 𝐇 = ⎢ ⎥, (4.66)
1⎥ ⎢−1 −1 1 1⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢1⎥ ⎢ 1 −1 1⎥ ⎢−1 1 −1 −1⎥
⎢1⎥ ⎢ 1 1 1⎥ ⎢ 1 1 1 1⎥
⎣1⎦ ⎣−1 1 1⎦ ⎣ 1 −1 −1 −1⎦
we can define the vector of shape functions as
1
𝐍(𝛏) = [𝐬 + 𝚵𝛏 + 𝐇𝐡(𝛏)], (4.67)
8
where
𝜂𝜍
𝜉 ⎡ 𝜉𝜍 ⎤
𝛏 = ⎢𝜂 ⎤

⎥, 𝐡(𝛏) = ⎢
⎢ ⎥. (4.68)
⎢ 𝜉𝜂 ⎥

⎣𝜍⎦
⎣𝜉𝜂𝜍⎦
The position vector
𝑥(𝛏)
⎡𝑦(𝛏)⎤
𝐱(𝛏) = ⎢ ⎥, (4.69)
⎣ 𝑧(𝛏)⎦
is for isoparametric finite elements given as
𝐱(𝛏) = 𝐗T 𝐍(𝛏), (4.70)
where
𝑥1 𝑦1 𝑧1
⎡𝑥2 𝑦2 𝑧2 ⎤
⎢𝑥 𝑦3 𝑧3 ⎥
⎢ 3 ⎥
⎢𝑥4 𝑦4 𝑧4 ⎥
𝐗=⎢
⎢𝑥5 𝑦5

𝑧5 ⎥, (4.71)
⎢𝑥 𝑦6 𝑧6 ⎥
⎢ 6 ⎥
⎢𝑥7 𝑦7 𝑧7 ⎥
⎣𝑥8 𝑦8 𝑧8 ⎦
is the matrix of nodal coordinates. The Jacobian matrix maps the isoparametric domain to
the physical domain as
∂𝐱(𝛏)
𝐉(𝛏) = , (4.72)
∂𝛏
and we find the Jabobian matrix at the element centroid to be
1
𝐉0 = 𝐉(0) = 𝐗T 𝚵. (4.73)
8
We may use this to rewrite the vector of shape functions partially in terms of the position
vector as

LS-DYNA DEV 10/27/16 (r:8004) 4-17 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

𝐍(𝛏) = 𝐛0 + 𝐁0 𝐱 + 𝚪𝐡(𝛏), (4.74)


Where
1
𝐛0 = {𝐈 − 𝐁0 𝐗T }𝐬 , (4.75a)
8
1
𝚪 = {𝐈 − 𝐁0 𝐗T }𝐇, (4.75b)
8
1
𝐁0 = 𝚵𝐉−1 . (4.75c)
8 0
The gradient-displacement matrix from this expression is given as
𝐁(𝛏) = 𝐁0 + 𝐁𝑠 (𝛏), (4.76)
where
∂𝐡(𝛏)
𝐁𝑠 (𝛏) = 𝚪 𝐉(𝛏)−1 . (4.77)
∂𝛏
We have
0 𝜍 𝜂
∂𝐡(𝛏) ⎡ 𝜍 0 𝜉⎥

=⎢
⎢𝜂 . (4.78)
∂𝛏 ⎢ 𝜉 0⎥⎥
⎣𝜂𝜍 𝜉𝜍 𝜉𝜂⎦
At this point we substitute the gradient-displacement matrix at the centroid of the element
𝐁0 with the mean gradient-displacement matrix 𝐁 defined as
1
𝐁= ∫ 𝐁(𝛏)𝑑𝑉𝑒 , (4.79)
𝑉𝑒 𝑒
where 𝑒 refers to the element domain and 𝑉𝑒 is the volume of the element, in all of the
expressions above. That is
1
𝚪 = {𝐈 − 𝐁𝐗T }𝐇, (4.80)
8
and
𝐁(𝛏) = 𝐁 + 𝐁𝑠 (𝛏), (4.81)
where
∂𝐡(𝛏)
𝐁𝑠 (𝛏) = 𝚪 𝐉(𝛏). (4.82)
∂𝛏
Proceeding, we write the expression for the rate-of-deformation as
1
𝛆̇ = [𝐗̇T 𝐁(𝛏) + 𝐁(𝛏)T 𝐗̇]
2
(4.83)
1 T T 1
= (𝐗̇ 𝐁 + 𝐁 𝐗̇) + [𝐗̇T 𝐁𝑠 (𝛏) + 𝐁𝑠 (𝛏)T 𝐗̇]
2 2

4-18 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

T
1 T T 1⎧
{ T ∂𝐡(𝛏) −1 −T ∂𝐡(𝛏) T ⎫}
= (𝐗̇ 𝐁 + 𝐁 𝐗̇) + ⎨𝐗̇ 𝚪 𝐉(𝛏) + 𝐉(𝛏) [ ] 𝚪 𝐗̇⎬
2 2{
⎩ ∂𝛏 ∂𝛏 }

⎧ T ⎫
1 T T 1 { ∂𝐡(𝛏) ∂𝐡(𝛏) T }
= (𝐗̇ 𝐁 + 𝐁 𝐗̇) + 𝐉(𝛏)−T ⎨𝐉(𝛏)T 𝐗̇T 𝚪 +[ ] 𝚪 𝐗̇𝐉(𝛏)⎬ 𝐉(𝛏)−1 ,
2 2 {
⎩ ∂𝛏 ∂𝛏 }

where we substitute the occurrences of the jacobian matrix 𝐉(ξ)with the following
expressions
⎧ T ⎫
1 T 1 ∂𝐡(𝛏) ∂𝐡(𝛏)
T
̂0 {𝐉T0 𝐗̇T 𝚪
𝛆̇ ≈ (𝐗̇ 𝐁 + 𝐁 𝐗̇) + 𝐉−T + [ ] 𝚪 T ̇ } −1
𝐗 𝐉 ̂ (4.84)
2 2 ⎨
{ ∂𝛏 ∂𝛏 0 ⎬ 𝐉0 ,
}
⎩ ⎭
where
∥𝐣1 ∥
⎡ ⎤
𝐉0̂ = ⎢ ∥𝐣2 ∥ ⎥, (4.85)
⎣ ∥𝐣3 ∥⎦
and j𝑖 is the i:th column in the matrix 𝐉0 . This last approximation is the key to the mesh
distortion insensitivity that characterizes the element.

Changing to Voigt notation, we define the stabilization portion of the strain rate as
𝛆̇𝑠 = 𝐉−1
̂ ̃ ̃
0 𝐁𝑠 (𝛏)𝐮̇, (4.86)
where now
∥𝐣1 ∥−2
⎡ ⎤
⎢ ⎥
⎢ ∥𝐣2 ∥−2 ⎥
⎢ ⎥
⎢ ∥𝐣3 ∥−2 ⎥
𝐉−1
0 =⎢ ⎥, (4.87)
⎢ −1
∥𝐣1 ∥ ∥𝐣2 ∥ −1 ⎥
⎢ ⎥
⎢ ∥𝐣3 ∥−1 ∥𝐣2 ∥−1 ⎥
⎢ ⎥
−1 −1
⎣ ∥𝐣1 ∥ ∥𝐣3 ∥ ⎦

γ2 𝜍 + γ3 𝜂 + γ4 𝜍𝜂 0 0
⎡ ⎤
⎢ 0 γ1 𝜍 + γ3 𝜉 + γ4 𝜍𝜉 0 ⎥
⎢ 0 0 γ1 𝜂 + γ2 𝜉 + γ4 𝜉𝜂⎥
̃ 𝑠 (𝛏) = ⎢
𝐁 ⎢

⎥, (4.88)
⎢ γ1 𝜍 γ2 𝜍 0 ⎥
⎢ 0 γ2 𝜉 γ3 𝜉 ⎥
⎢ ⎥
⎣ γ1 𝜂 0 γ3 𝜂 ⎦
and 𝐮̇̃ is the vector of nodal velocities transformed to the isoparametric system according to
𝐮̇̃ = 𝕵𝐮̇, (4.89)
where 𝕵 is the 24 × 24 matrix that transforms the 8 nodal velocity vectors to the
isoparametric domain given by

LS-DYNA DEV 10/27/16 (r:8004) 4-19 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

𝐉T0
⎡ ⎤
𝕵 = perm ⎢
⎢ ⋱ ⎥.
⎥ (4.90)
T
⎣ 𝐉0 ⎦
T
Moreover, 𝛄𝑖 is the ith row of 𝚪 . We have deliberately neglected terms that cause
parallelepiped finite elements to lock in shear.

To eliminate Poisson type locking in bending and volumetric locking, an enhanced


isoparametric rate-of-strain field is introduced as
𝛆̇𝑒 = 𝐉−1
̂ ̃
0 𝐆(𝛏)𝛂̇ , (4.91)
with
𝜉 0 0 𝜉𝜂 𝜂𝜍 𝜍𝜉
⎡ ⎤
⎢0 𝜂 0 𝜉𝜂 𝜂𝜍 𝜍𝜉 ⎥
̃ (𝛏) = ⎢
𝐆 ⎢
0 0 𝜍 𝜉𝜂 𝜂𝜍 𝜍𝜉 ⎥.
⎥ (4.92)
⎢0 0 0 0 0 0 ⎥
⎢0 0 0 0 0 0 ⎥
⎣0 0 0 0 0 0 ⎦
Hence, the stabilized strain field becomes
𝛆̇𝑠 = 𝐉−1
̂ ̃ ̃ ̃ −1
̂ ̃
0 [𝐁𝑠 (𝛏)𝐮̇ + 𝐆(𝛏)𝛂̇ ] = 𝐉0 𝛆̇𝑠 , (4.93)
where 𝜶̇ is the enhanced strain vector that must be determined from an equilibrium
condition.

The virtual work equation can be written


𝛿𝑊int = ∫ 𝛿𝛆T 𝛔𝐝𝑉𝑒 , (4.94a)
𝑒
= ∫ 𝛿𝛆T 𝐽 −1 𝐉T 𝐒𝑑𝑉𝑒 , (4.94b)
𝑒
𝑡
= ∫ 𝛿𝛆T 𝐽 −1 𝐉T ⎛
⎜∫ 𝐂𝑆𝐸 𝐄̇ 𝑑𝜏 ⎞
⎟ 𝑑𝑉𝑒 , (4.94c)
𝑒 ⎝ 0 ⎠
𝑡
T −1 T ⎛
= ∫ 𝛿𝛆 𝐽 𝐉 ⎜∫ 𝐽𝐉−T 𝐂𝜎 𝛆̇𝑑𝜏 ⎞
⎟ 𝑑𝑉𝑒 , (4.94d)
𝑒 ⎝0 ⎠
𝑡
T⎛ ⎞
= ∫ 𝛿𝛆 𝑗0 𝐉0 ⎜∫ 𝐽𝐉−T
T 𝜎
0 𝐂 𝛆̇𝑑𝜏 ⎟ 𝑑𝑉𝑝 , (4.94e)
𝑝 ⎝0 ⎠
𝑡
𝑉
≈ ∫ 𝛿𝛆̃T ⎛
⎜∫ 𝑒 𝐉−T̂ ̂ ̃ ⎞
𝜎 −1
0 𝐂 𝐉0 𝛆̇𝑑𝜏 ⎟ 𝑑𝑉𝑝 , (4.94f)
𝑝 ⎝0 8 ⎠
where 𝐽 is the determinant of the deformation gradient, 𝐉 is the push-forward operator of a
symmetric 2nd order tensor, 𝑗0 is the determinant of the jacobian matrix, 𝑗0 is the
determinant of the jacobian matrix at time 0, σ is the true stress tensor, 𝐒 is the 2nd Piola-
Kirchhoff stress tensor, 𝐂𝑆𝐸 is the material tangent modulus, 𝐂σ is the spatial tangent
modulus and 𝑉𝑒 is the volume of the element. In the above, we have used various

4-20 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

transformation formulae between different stress and constitutive tensors. At this point we
are only interested in how to handle the stabilization portion of the strain rate field, the
constant part is only used to update the midpoint stress as usual. Because of orthogonality
properties of the involved matrices, it turns out that we may just insert the expression for
the stabilization strain rate field to get
𝑡
𝑉
𝑠
𝛿𝑊int ≈ ∫ 𝛿𝛆̃𝑠T ⎛
⎜∫ 𝑒 𝐉−T
̂ ̂ ̃ ⎞
𝜎 −1
0 𝐂 𝐉0 𝛆̇𝑠 𝑑𝜏 ⎟ 𝑑𝑉𝑝 , (4.95a)
𝑝 ⎝0 8 ⎠

̃ (𝛏)𝑇 ⎛ 𝑡 𝑉𝑒 −T 𝜎 −1
𝚩
= [𝛿𝐮̃ 𝑇
𝛿𝜶 T] ∫ [ 𝑠
𝑇
] ⎜∫ 𝐉0̂ 𝐂 𝐉0̂ [𝚩
̃ 𝑠 (𝛏)
𝐆̃ (𝛏)] [𝐮̇̃] 𝑑𝜏 ⎞
⎟ 𝑑𝑉𝑝 . (4.95b)
𝑝
̃
𝐆 (𝛏) ⎝0 8 𝛂̇ ⎠
The stabilization contribution to the internal force vector is given by
𝑡
𝐟 T ̃ 𝑠 (𝛏)T ⎛ 𝑉𝑒 −T 𝜎 −1
0 ] ∫ [𝚩
[ 𝑢 ] = [𝕵 ] ⎜∫ 𝐉 ̂ 𝐂 𝐉0̂ [𝚩 ̃ (𝛏)][𝕵 0][𝐮̇]𝑑𝜏 ⎞
̃ 𝑠 (𝛏) 𝐆 ⎟ 𝑑𝑉𝑝 . (4.96)
𝐟𝛼 0 𝐈 𝑝 𝐆 ̃ (𝛏)T ⎝ 8 0 0 𝐈 𝛂̇ ⎠
0

In a discretization, the condition 𝐟𝛼 = 0 is used to determine Δ𝛂, the increment of the


enhanced strain variables, from Δu, the increment in displacements. This is inserted back
into the expression for the internal force vectors to determine 𝐟𝑢 , the stabilization
contribution to the internal force vector.

The implementation of the element is very similar to the implementation of the one
point integrated mean strain hexahedral by Flanagan and Belytschko [1981]. The hourglass
forces are calculated in a different manner.

From the midpoint stress update we get a bulk and shear modulus characterizing
the material at this specific point in time. From this we form the isotropic spatial tangent
modulus 𝐂𝛔 to be used for computing the stabilization force from Equation (4.96).

4.5 Fully Integrated Brick Elements and Mid-Step Strain


Evaluation
To avoid locking in the fully integrated brick elements strain increments at a point in
a constant pressure, solid element are defined by [see Nagtegaal, Parks, anmmmmmd Rice
1974]
∂Δ𝑣 + ∂Δ𝑢
𝑛+1⁄2 𝑛+1⁄2
∂Δ𝑢 ∂𝑥 ∂𝑦
Δ𝜀𝑥𝑥 = + 𝜙, Δ𝜀𝑥𝑦 = ,
𝑛+1⁄2 2
∂𝑥
(4.97)
∂Δ𝑤 + ∂Δ𝑣
𝑛+1⁄2 𝑛+1⁄2
∂Δ𝑣 ∂𝑦 ∂𝑧
Δ𝜀𝑦𝑦 = + 𝜙, Δ𝜀𝑦𝑧 = ,
𝑛+1⁄2 2
∂𝑦

LS-DYNA DEV 10/27/16 (r:8004) 4-21 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

∂Δ𝑢 + ∂Δ𝑤
∂Δ𝑤 𝑛+1⁄2 𝑛+1⁄2
Δ𝜀𝑧𝑧 = + 𝜙, Δ𝜀𝑧𝑥 = ∂𝑧 ∂𝑥 ,
𝑛+1⁄2 2
∂𝑧
where 𝜙 modifies the normal strains to ensure that the total volumetric strain increment at
each integration point is identical
∂Δ𝑢 + ∂Δ𝑣 + ∂Δ𝑤
𝑛+1⁄2 𝑛+1⁄2 𝑛+1⁄2 (4.98)
𝑛+1⁄2 ∂𝑥 ∂𝑦 ∂𝑧
𝜙 = Δ𝜀𝑣 − ,
3
𝑛+1⁄2
and Δ𝜀𝑣 is the average volumetric strain increment in the midpoint geometry

1∫ ⎛ ∂Δ𝑢 + ∂Δ𝑣 + ∂Δ𝑤 ⎞ 𝑛+1⁄2


𝑛+1⁄2 ⎜ ⎟ 𝑑𝑣
3 𝑣 1
𝑛+ ⁄2 1
𝑛+ ⁄2 1
𝑛+ ⁄2
⎝∂𝑥 ∂𝑦 ∂𝑧 ⎠ (4.99)
,
𝑛+1⁄2
∫ 𝑛+1⁄2 𝑑𝑣
𝑣

Δ𝑢,. Δ𝑣, and Δ𝑤 are displacement increments in the x, y, and z directions, respectively, and

𝑛+1⁄2 (𝑥𝑛 + 𝑥𝑛+1 )


𝑥 = , (4.100a)
2

𝑛+1⁄2 (𝑦𝑛 + 𝑦𝑛+1 )


𝑦 = , (4.100b)
2

(𝑧𝑛 + 𝑧𝑛+1 )
𝑛+1⁄2
𝑧 = . (4.100c)
2
To satisfy the condition that rigid body rotations cause zero straining, it is necessary to use
the geometry at the mid-step in the evaluation of the strain increments. As the default, LS-
DYNA currently uses the geometry at step 𝑛 + 1 to save operations; however, for implicit
calculations the mid-step strain calculation is always recommended, and, for explicit
calculations, which involve rotating parts, the mid-step geometry should be used especially
if the number of revolutions is large. The mid-step geometry can be activated either
globally or for a subset of parts in the model by using the options on the control card,
*CONTROL_ACCURACY.

Since the bulk modulus is constant in the plastic and viscoelastic material models,
constant pressure solid elements result. In the thermoelastoplastic material, a constant
temperature is assumed over the element. In the soil and crushable foam material, an
average relative volume is computed for the element at time step 𝑛 + 1, and the pressure
and bulk modulus associated with this relative volume is used at each integration point.
For equations of state, one pressure evaluation is done per element and is added to the
deviatoric stress tensor at each integration point.

4-22 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

2 r
s
Figure 4.4. Four node tetrahedron.

The foregoing procedure requires that the strain-displacement matrix corresponding


to Equations (4.66) and consistent with a constant volumetric strain, ̅𝐁
̅̅̅̅, be used in the nodal
force calculations [Hughes 1980]. It is easy to show that:
̅̅̅̅̅𝑛+1 )T 𝛔𝑛+1 𝑑𝑣 = ∫ (𝐁𝑛+1 )T 𝛔𝑛+1 𝑑𝑣,
𝐅 = ∫ 𝑛+1(𝐁 (4.101)
𝑣 𝑛+1 𝑣

and avoid the needless complexities of computing ̅𝐁


̅̅̅̅.

4.6 Four Node Tetrahedron Element


The four node tetrahedron element with one point integration, shown in Figure 4.4,
is a simple, fast, solid element that has proven to be very useful in modeling low density
foams that have high compressibility. For most applications, however, this element is too
stiff to give reliable results and is primarily used for transitions in meshes. The formulation
follows the formulation for the one point solid element with the difference that there are no
kinematic modes, so hourglass control is not needed. The basis functions are given by:
𝑁1 (𝑟, 𝑠, 𝑡) = 𝑟, (4.102a)
𝑁2 (𝑟, 𝑠, 𝑡) = 𝑠, (4.102b)
𝑁3 (𝑟, 𝑠, 𝑡) = 1 − 𝑟 − 𝑠 − 𝑡, (4.102c)
𝑁4 (𝑟, 𝑠, 𝑡) = 𝑡. (4.102d)

If a tetrahedron element is needed, this element should be used instead of the


collapsed solid element since it is, in general, considerably more stable in addition to being
much faster. Automatic sorting can be used, see *CONTROL_SOLID keyword, to
segregate these elements in a mesh of 8 node solids for treatment as tetrahedrons.

LS-DYNA DEV 10/27/16 (r:8004) 4-23 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

4.7 Nodal Pressure Tetrahedron


For applications requiring a tetrahedron mesh, the volume averaged tetrahedron type 13 is
an interesting alternative to the standard single point tetrahedron, also known as the 1
point nodal pressure tetrahedron. The idea is to average the volumetric strain over
adjacent elements to smooth the pressure response. To this end, we assume that 𝐸 is the set
of all such elements in a model, and likewise 𝑁 is the set of all nodes belonging to these
elements. We introduce the indicator
1/4 if node 𝑛 is connecting to element 𝑒
𝜒𝑒𝑛 = { (4.103)
0 otherwise
The volume of a node 𝑛 is defined as
𝑉𝑛 = ∑ 𝜒𝑒𝑛 𝑉𝑒 (4.104)
𝑒∈𝐸
where 𝑉𝑒 is the exact volume of element 𝑒. This allows us to define an average jacobian 𝐽𝑒̅
for element 𝑒 as
𝑉
𝐽𝑒̅ = ∑ 𝜒𝑒𝑛 { 𝑛0 } (4.105)
𝑛∈𝑁 𝑉𝑛
which is essentially the relative volume. Here 𝑉𝑛0 is the nodal volume in the reference
configuration. The element is completely defined by the assumed deformation gradient
given as
1/3
𝐽𝑒̅
̅
𝑭𝑒 = ( ) 𝑭𝑒 (4.106)
𝐽𝑒
where 𝑭𝑒 is the deformation gradient from the isoparametric shape functions and 𝐽𝑒 =
det𝑭𝑒 , also given as
𝑉
𝐽𝑒 = 0𝑒 . (4.107)
𝑉𝑒
The assumed rate of deformation, derived from 𝑭𝑒̅ , is given as
1 𝛿𝐽 ̅ 𝛿𝐽
𝛿𝜺̅𝑒 = ( 𝑒 − 𝑒 ) 𝑰 + 𝛿𝑭𝑒 𝑭𝑒−1 (4.108)
3 𝐽𝑒̅ 𝐽𝑒
meaning that the volumetric strain is replaced by that of the averaged one. Continuing, the
virtual work equation is given as
∑ 𝒇𝑛𝑇 𝛿𝒙𝑛 = ∑ 𝝈𝑒 : 𝛿𝜺̅𝑒 𝐽𝑒 𝑉𝑒0 (4.109)
𝑛∈𝑁 𝑒∈𝐸
which provides the equation for the nodal forces, 𝒇𝑛 . An expression for the strain-
displacement matrix, 𝑩𝑛𝑒 , is deduced from combining (4.108) and the relation 𝛿𝜺̅𝑒 =
∑𝑛∈𝑁 𝑩𝑛𝑒 𝛿𝒙𝑛 . It turns out to be given as
𝑉𝑓 𝜒𝑒𝑚 𝑛
𝑩𝑛𝑒 = ∑ ∑ 𝜒𝑚 𝑩 ̃ 𝑛𝑒
̅̅̅̅ 𝑓 + 𝑩 (4.110)
𝑓 ∈𝐸 𝑚∈𝑁 𝑓 𝐽𝑒̅ 𝑉𝑚
0

̅̅̅ 𝑛𝑒 and
where ̅𝑩

4-24 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

4
6

r
2
s

1
5
Figure 4.5. Six node Pentahedron.

4.8 Six Node Pentahedron Element


The pentahedron element with two point Gauss integration along its length, shown
in Figure 4.5, is a solid element that has proven to be very useful in modeling axisymmetric
structures where wedge shaped elements are used along the axis-of-revolution. The
formulation follows the formulation for the one point solid element with the difference
that, like the tetrahedron element, there are no kinematic modes, so hourglass control is not
needed. The basis functions are given by:
1
𝑁1 (𝑟, 𝑠, 𝑡) = (1 − 𝑡)𝑟, (4.111a)
2
1
𝑁2 (𝑟, 𝑠, 𝑡) = (1 − 𝑡)(1 − 𝑟 − 𝑠), (4.111b)
2
1
𝑁3 (𝑟, 𝑠, 𝑡) = (1 + 𝑡)(1 − 𝑟 − 𝑠), (4.111c)
2
1
𝑁4 (𝑟, 𝑠, 𝑡) = (1 + 𝑡)𝑟, (4.111d)
2
1
𝑁5 (𝑟, 𝑠, 𝑡) = (1 − 𝑡)𝑠, (4.111e)
2
1
𝑁6 (𝑟, 𝑠, 𝑡) = (1 + 𝑡)𝑠. (4.111f)
2

LS-DYNA DEV 10/27/16 (r:8004) 4-25 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

8
8
15
7 7
16
14
5 20 5
13
19 6
6
4 4
17 11
3 3
12 18

10
1 1
9 2 2

DOF ui, vi, wi DOF ui, vi, wi, θxi, θyi, θzi
Figure 4.6. The 20-node solid element is transformed to an 8-node solid with 6
degrees-of-freedom per node.

If a pentahedron element is needed, this element should be used instead of the


collapsed solid element since it is, in general, more stable and significantly faster.
Automatic sorting can be used, see *CONTROL_SOLID keyword, to segregate these
elements in a mesh of 8 node solids for treatment as pentahedrons. Selective-reduced
integration is used to prevent volumetric locking, i.e., a constant pressure over the domain
of the element is assumed.

4.9 Fully Integrated Brick Element With 48 Degrees-of-


Freedom
The forty-eight degree of freedom brick element is derived from the twenty node
solid element; see Figure 4.6, through a transformation of the nodal displacements and
rotations of the mid-side nodes [Yunus, Pawlak, and Cook, 1989]. This element has the
advantage that shell nodes can be shared with brick nodes and that the faces have just four
nodes — a real advantage for the contact-impact logic. The accuracy of this element is
relatively good for problems in linear elasticity but degrades as Poisson’s ratio approaches
the incompressible limit. This can be remedied by using incompatible modes in the
element formulation, but such an approach seems impractical for explicit computations.

4-26 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

θz

w
z

k
v θy
x
u
θx i
Figure 4.7. A typical element edge is shown from [Yunus, Pawlak, and Cook,
1989].
The instantaneous velocity for a midside node 𝑘 is given as a function of the corner
node velocities as (See Figure 4.7),
1 𝑦𝑗 − 𝑦𝑖 𝑧𝑗 − 𝑧𝑖
𝑢̇𝑘 = (𝑢̇𝑖 + 𝑢̇𝑗 ) + (𝜃𝑧𝑗̇ − 𝜃𝑧𝑖̇ ) + ̇ − 𝜃𝑦𝑗
(𝜃𝑦𝑖 ̇ ),
2 8 8
1 𝑧𝑗 − 𝑧𝑖 𝑥 − 𝑥𝑖
𝑣̇𝑘 = (𝑣̇𝑖 + 𝑣̇𝑗 ) + ̇ − 𝜃𝑥𝑖
(𝜃𝑥𝑗 ̇ )+ 𝑗 (𝜃𝑧𝑖̇ − 𝜃𝑧𝑗̇ ), (4.112)
2 8 8
1 𝑥𝑗 − 𝑥𝑖 𝑦 − 𝑦𝑖
𝑤̇ 𝑘 = (𝑤̇ 𝑖 + 𝑤̇ 𝑗 ) + (𝜃𝑦𝑗̇ − 𝜃𝑦𝑖 ̇ )+ 𝑗 (𝜃𝑥𝑖 ̇ − 𝜃𝑥𝑗̇ ),
2 8 8
where 𝑢, 𝑣, 𝑤, 𝜃𝑥 , 𝜃𝑦 , and 𝜃z are the translational and rotational displacements in the global
𝑥, 𝑦, and 𝑧 directions. The velocity field for the twenty-node hexahedron element in terms
of the nodal velocities is:
⎧ 𝑢̇1 ⎫
{
{ ⋮ }}
{
{ 𝑢̇ }
{ 20 }
}
𝜙 𝜙2 … 𝜙20 0 0 … 0 0 0 … 0 { { 𝑣̇1 }
{𝑢̇ ⎫
⎧ } ⎡ 1 ⎤ }
⎨ 𝑣̇ ⎬ = ⎢0 0 … 0 𝜙1 𝜙2 … 𝜙20 0 0 … 0 ⎥ ⎨ ⋮ ⎬, (4.113)
{
⎩𝑤̇ }⎭ ⎣0 0 … 0 0 0 … 0 𝜙1 𝜙2 … 𝜙20 ⎦ {
{ 𝑣̇20 }
}
{
{ 𝑤̇ 1 }
}
{
{ ⋮ }}
{ }
⎩𝑤̇ 20 ⎭
where 𝜙𝑖 are given by [Bathe and Wilson 1976] as,

LS-DYNA DEV 10/27/16 (r:8004) 4-27 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

DOF ui, vi, wi DOF ui, vi, wi, θxi, θyi, θzi

Figure 4.8. Twenty-four degrees of freedom tetrahedron element [Yunus,


Pawlak, and Cook, 1989].

(𝑔9 + 𝑔12 + 𝑔17 ) (𝑔13 + 𝑔16 + 𝑔17 )


𝜙1 = 𝑔1 − , 𝜙5 = 𝑔5 − ,
2 2
(𝑔 + 𝑔10 + 𝑔18 ) (𝑔 + 𝑔14 + 𝑔18 )
𝜙2 = 𝑔2 − 9 , 𝜙6 = 𝑔6 − 13 ,
2 2
(𝑔 + 𝑔11 + 𝑔19 ) (𝑔 + 𝑔15 + 𝑔19 )
𝜙3 = 𝑔3 − 10 , 𝜙7 = 𝑔7 − 14 ,
2 2
(𝑔 + 𝑔12 + 𝑔20 ) (𝑔 + 𝑔16 + 𝑔20 ) (4.114)
𝜙4 = 𝑔4 − 11 , 𝜙8 = 𝑔8 − 15 ,
2 2
𝜙𝑖 = 𝑔𝑖 for 𝑗 = 9, … , 20
𝑔𝑖 = 𝐺(𝜉 , 𝜉𝑖 )𝐺(𝜂, 𝜂𝑖 )𝐺(𝜁 , 𝜁𝑖 ),
⎧1
{ (1 + 𝛽𝛽𝑖 ) for, 𝛽𝑖 = ±1; 𝛽 = 𝜉 , 𝜂, 𝜁
𝐺(𝛽, 𝛽𝑖 ) = ⎨2 .
{
⎩ 1 − 𝛽2 for, 𝛽𝑖 = 0

The standard formulation for the twenty node solid element is used with the above trans-
formations. The element is integrated with a fourteen point integration rule [Cook 1974]:
1 1 1
∫ ∫ ∫ 𝑓 (𝜉 , 𝜂 , 𝜁 )𝑑𝜉𝑑𝜂𝑑𝜁 =
−1 −1 −1

𝐵6 [𝑓 (−𝑏, 0,0) + 𝑓 (𝑏, 0,0) + 𝑓 (0, −𝑏, 0) + ⋯ + (6 terms)] + (4.115)

𝐶8 [𝑓 (−𝑐, −𝑐, −𝑐) + 𝑓 (𝑐, −𝑐, −𝑐) + 𝑓 (𝑐, 𝑐, −𝑐) + ⋯ + (8 terms)],


where
𝐵6 = 0.8864265927977938, 𝑏 = 0.7958224257542215,
(4.116)
𝐶8 = 0.3351800554016621, 𝑐 = 0.7587869106393281.

4-28 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

Cook reports that this rule has nearly the same accuracy as the twenty-seven point Gauss
rule, which is very costly. The difference in cost between eight point and fourteen point
integration, though significant, is necessary to eliminate the zero energy modes.

4.10 Fully Integrated Tetrahedron Element With 24 Degrees-


of-Freedom
The twenty-four degree of freedom tetrahedron element is derived from the ten-
node tetrahedron element; see Figure 4.8, following the same procedure used above for the
forty-eight degree of freedom brick element [Yunus, Pawlak, and Cook, 1989]. This
element has the advantage that shell nodes can be shared with its nodes and it is
compatible with the brick element discussed above. The accuracy of this element is
relatively good-at least when compared to the constant strain tetrahedron element. This is
illustrated by the bar impact example in Figure 4.9 which compares the 12 and 24 degree of
freedom tetrahedron elements. The 12 degree-of-freedom tetrahedron displays severe
volumetric locking.

In our implementation we have not strictly followed the reference. In order to


prevent locking in applications that involve incompressible behavior, selective reduced
integration is used with a total of 5 integration points. Although this is rather expensive,
no zero energy modes exist. We use the same approach in determining the rotary mass
that is used in the implementation of the shell elements.

LS-DYNA DEV 10/27/16 (r:8004) 4-29 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

Figure 4.9. A comparison of the 12 and 24 degree-of-freedom tetrahedron


elements is shown. The 12 degree-of-freedom tetrahedron element on the top
displays severe volumetric locking.

4-30 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

8 7 8 7 8 7

5 6 5 5
6 6
3 3 4 3
4
4
1 2 1 1
2 2
8 8
7 7

5 6 5 6

4
3 4 3

1 2 1 2
Figure 4.10a. Construction of a hexahedron element with five tetrahedrons.

8 7 8 7 8 7

5 5 5 6
6 6
4 4
3 4
3 3

1 2 1 2 1 2
8 7 8 7 8 7

5 6 5 5 6
6

3 4 4
4 3 3

1 2 1 2 1 2
Figure 4. 10b. Construction of a hexahedron element with six tetrahedrons

Figures 4.10a and 4. 10b show the construction of a hexahedron element from five
and six tetrahedron elements, respectively. When two sides of the adjacent bricks made
from five tetrahedrons are together, it is likely that four unique triangular segments exist.
This creates a problem in LS-PREPOST, which uses the numbering as a basis for
eliminating interior polygons prior to display. Consequently, the graphics in the post-
processing phase can be considerably slower with the degeneration in Figure 4.10a.
However, marginally better results may be obtained with five tetrahedrons per hexahedron
due to a better constraint count.

LS-DYNA DEV 10/27/16 (r:8004) 4-31 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

4.11 The Cosserat Point Elements in LS-DYNA


4.11.1 Introduction
The Cosserat Point Elements (CPE) are based on the works by Jabareen et.al.[1,2]. In
contrast with a conventional approach, the CPE is treated as a structure rather than a
continuum. The kinematic variables of the CPE are characterized by a volume averaged
deformation gradient and other measures of inhomogeneous deformations. The CPE
models the response of a simple continuum (not a generalized Cosserat media) and the
additional kinematic degrees of freedom model physical modes of deformation of the
structure. Moreover, the CPE uses a hyperelastic constitutive equation for elastic response
with the strain energy of the CPE being separated additively into a part dependent on the
strain energy of the three-dimensional material and another strain energy associated with
inhomogeneous deformations. Also, for the tetrahedral CPE use is made of a new measure
of dilatation that stabilizes hourglass type modes in large deformations. This formulation
is valid for large deformations and the coefficients in the inhomogeneous strain energy are
ingeniously determined by comparison with exact linear solutions. This ensures that CPE
yields accurate results for elementary deformation modes in linear elasticity. Moreover,
using the average deformation gradient for the response to homogeneous deformations the
CPE formulation can be coupled with arbitrary material models in LS-DYNA. Still the
element is, due to its complexity and slight loss of generality, first and foremost
recommended for hyperelasticy in implicit analysis.

Section 4.11.2 through 4.11.8 describes the theory for the hexahedral CPE element,
the tetrahedron is based on the same concepts except for the volumetric correction
presented in Section 4.11.5. For more details we refer to [1] and [2]. We end with two
examples in Sections 4.11.9 and 4.11.10.

d3
D3

L
d1
D2
d2

D1
H
W

Figure 4.11. The CPE hexahedron

4-32 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

4.11.2 Geometry
The geometry of the hexahedral CPE element is characterized by the three-dimensional
directors 𝐃𝑖 and 𝐝𝑖 , 𝑖 = 0,1, … ,7, where the formers are associated with the reference
configuration and the latters with the current configuration. The reciprocal vectors 𝐃𝑖 and
𝐝𝑖 , 𝑖 = 1,2,3, are such that
𝑗
𝐝𝑖 ⋅ 𝐝 𝑗 = 𝐃 𝑖 ⋅ 𝐃 𝑗 = 𝛿 𝑖 , 𝑖, 𝑗 = 1, 2, 3, (4.11.117)
and we also have that
|𝐃𝑖 | = 1, 𝑖 = 1, 2, 3, (4.11.118)
The coordinates 𝜃𝑖 , 𝑖 = 1,2,3, ranges between
−𝐻/2 ≤ 𝜃1 ≤ 𝐻/2, − 𝑊/2 ≤ 𝜃2 ≤ 𝑊/2, − 𝐿/2 ≤ 𝜃3 ≤ 𝐿/2, (4.11.119)
and the 𝐀 matrix is given such that
7 7
𝑗 𝑗
𝐃𝑖 = ∑ 𝐴𝑖 𝐗𝑗 , 𝐝𝑖 = ∑ 𝐴𝑖 𝐱𝑗 , 𝑖 = 0, 1, . . . , 7, (4.11.120)
𝑗=0 𝑗=0

where 𝐗𝑖 and 𝐱𝑖 are the nodal coordinates in the reference and current configuration,
respectively. Hence the 𝐀 matrix represents the mapping between the nodal coordinates
and the Cosserat point directors. The reference position vector 𝐗 is expressed as
7
𝐗 = 𝐗(𝜃1 , 𝜃2 , 𝜃3 ) = ∑ 𝑁 𝑗 (𝜃1 , 𝜃2 , 𝜃3 )𝐃𝑗 , (4.11.121)
𝑗=0

and likewise the current position 𝐱 as


7
𝐱 = 𝐱(𝜃1 , 𝜃2 , 𝜃3 ) = ∑ 𝑁 𝑗 (𝜃1 , 𝜃2 , 𝜃3 )𝐝𝑗 , (4.11.122)
𝑗=0

The shape functions are given as


𝑁 0 = 1, 𝑁 1 = 𝜃1 , 𝑁 2 = 𝜃2 , 𝑁 3 = 𝜃3 ,
(4.11.123)
𝑁 4 = 𝜃1 𝜃2 , 𝑁 5 = 𝜃1 𝜃3 , 𝑁 6 = 𝜃2 𝜃3 , 𝑁 7 = 𝜃1 𝜃2 𝜃3 .

4.11.3 Deformation and strain


The deformation measures used are
3 4
𝑖 −1 ̅̅̅̅ = 𝐅 (𝐈 + ∑ 𝛃𝑖 ⊗ 𝐕𝑖 ), (4.11.124)
𝐅 = ∑ 𝐝𝑖 ⊗ 𝐃 , 𝛃𝑖 = 𝐅 𝐝𝑖+3 − 𝐃𝑖+3 (𝑖 = 1,2,3,4), 𝐅
𝑖=1 𝑖=1

where 𝐅 ̅̅̅̅ is the volume averaged deformation gradient and thus represents the
homogeneous deformations whereas 𝛃𝑖 are measures of the inhomogeneous deformations.
As for the 𝐕𝑖 we have

LS-DYNA DEV 10/27/16 (r:8004) 4-33 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

𝐻2 𝑊2
𝑉 = 𝐻𝑊𝐿 (𝐃1 × 𝐃2 ⋅ 𝐃3 + 𝐃 × 𝐃5 ⋅ 𝐃1 + 𝐃
12 4 12 6
𝐿2
× 𝐃4 ⋅ 𝐃2 + 𝐃5 × 𝐃6 ⋅ 𝐃3 ) ,
12
𝐻2 𝑊2
𝐕1 = 𝑉 −1 𝐻𝑊𝐿 ( 𝐃5 × 𝐃1 + 𝐃 × 𝐃6 ) ,
12 12 2 (4.11.125)
2 2
𝐻 𝐿
𝐕2 = 𝑉 −1 𝐻𝑊𝐿 ( 𝐃1 × 𝐃4 + 𝐃6 × 𝐃3 ) ,
12 12
𝑊2 𝐿2
𝐕3 = 𝑉 −1 𝐻𝑊𝐿 ( 𝐃4 × 𝐃2 + 𝐃3 × 𝐃5 ) ,
12 12
𝐕4 = 𝟎,

̅̅̅̅ is given as
The velocity gradient consistent with 𝐅
4
̅̅̅̅̇𝐅
𝐋̅ = 𝐅 ̅̅̅̅−1 = 𝐋 + ∑(𝐝̇ 𝑗+3 − 𝐋𝐝𝑗+3 ) ⊗ 𝐕𝑗 𝐅
̅̅̅̅−1 , (4.11.126)
𝑗=1

where 𝐋 = 𝐅̇ 𝐅−1 , which in turn gives the rate –of-deformation and spin tensors as
1 1
𝛆̅̇ = (𝐋̅ + 𝐋̅ 𝑇 ), ̅̅ ̅̅̅ = (𝐋̅ − 𝐋̅ 𝑇 ).
𝛚 (4.11.127)
2 2

4.11.4 Stress and Force


On the other hand,
3
𝐋 = ∑ 𝐝̇ 𝑖 ⊗ 𝐝𝑖 , (4.11.128)
𝑖=1

so we can rewrite 𝐋̅ as
3 4 4
𝐋̅ = ∑ 𝐝̇ 𝑖 ⊗ 𝐝𝑖 ⎛
⎜𝐅̅̅̅̅ − ∑ 𝐝𝑗+3 ⊗ 𝐕𝑗 ⎞
⎟𝐅̅̅̅̅−1 + ∑ 𝐝̇ 𝑖+3 ⊗ 𝐕𝑖 𝐅
̅̅̅̅−1 . (4.11.129)
𝑖=1 ⎝ 𝑗=1 ⎠ 𝑖=1

The Cauchy stress is given by the constitutive law


𝛔∇ = 𝐟hypo (𝛆̅,̇ . . . ), (4.11.130)
or for the hyperelastic case
̅̅̅̅, . . . ),
𝛔 = 𝐟hyper (𝐅 (4.11.131)
so the nonzero internal force vectors are given as

4-34 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

4
𝐭𝜎𝑖 = 𝐽 𝑉 ̅̅̅̅−𝑇 ⎛
̅ 𝛔𝐅 ⎜𝐅̅̅̅̅𝑇 − ∑ 𝐕𝑗 ⊗ 𝐝𝑗+3 ⎞
⎟ 𝐝𝑖 , 𝐭𝜎𝑖+3 = 𝑉𝛔𝐅
̅̅̅̅−𝑇 𝐕𝑖 , 𝑖 = 1, 2, 3. (4.11.132)
⎝ 𝑗=1 ⎠
The first expression above can in turn be rewritten as
4
𝐭𝜎𝑖 = ⎛ ̅ 𝛔 − ∑ 𝐭𝜎𝑗+3 ⊗ 𝐝𝑗+3 ⎞
⎜𝐽 𝑉 ⎟ 𝐝𝑖 , 𝑖 = 1, 2, 3. (4.11.133)
⎝ 𝑗=1 ⎠

4.11.5 CPE3D10 modification


For the 10-noded tetrahedron, a modified deformation gradient is used in the constitutive
law, given by
̃ 1/3 ̅̅̅̅
𝐅̃ = (𝐽⁄ ) 𝐅, (4.11.134)
𝐽̅
meaning it has been modified for the volumetric response. Here
𝐽 ̃ = 𝐽 ̅ + 𝜂𝐽, (4.11.135)
This gives a consistent velocity gradient as
1𝐽⎛ ⎧ }̅̇
{𝐽 ̇ 𝐽 ⎫
𝐋̃ = 𝐋̅ + ⎜𝜂̇ + 𝜂 ⎨ − ⎬⎞ ⎟ 𝐈. (4.11.136)
3 𝐽 ̃⎝ ⎩𝐽 𝐽 }
{ ⎭̅ ⎠
The Cauchy stress is now given by the constitutive law
1
𝛔∇ = 𝐟hypo (𝛆̃,̇ … ), 𝛆̃̇ = (𝐋̃ + 𝐋̃ 𝑇 ), (4.11.137)
2
or for the hyperelastic case
𝛔 = 𝐟hyper (𝐅̃, . . . ), (4.11.138)
and the corresponding internal force can be identified through a principle of virtual work
9
∑ 𝐭𝜎𝑖 ⋅ 𝐝̇ 𝑖 = 𝐽 𝑉
̃ 𝐋̃ ⋅ 𝛔. (4.11.139)
𝑖=0

For the hyperelastic special case we have


∂Σ ∂Σ 𝑇 1 ∂Σ
𝛔= 𝐈 + 2𝐽 −1
̃ [𝐅
̅̅̅̅′ ̅̅̅̅′ − (
𝐅 ⋅ 𝐂′̅ ) 𝐈], (4.11.140)
∂𝐽 ̃ ∂𝐂′̅ 3 ∂𝐂′̅
and since

LS-DYNA DEV 10/27/16 (r:8004) 4-35 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

3 6 6
𝐋̃ = ∑ 𝐝̇ 𝑖 ⊗ 𝐝𝑖 ⎛
⎜𝐅̅̅̅̅ − ∑ 𝐝𝑗+3 ⊗ 𝐕𝑗 ⎞
⎟𝐅̅̅̅̅−1 + ∑ 𝐝̇ 𝑖+3 ⊗ 𝐕𝑖 𝐅
̅̅̅̅−1
𝑖=1 ⎝ 𝑗=1 ⎠ 𝑖=1

1 𝐽 6 3 ⎡ ∂𝜂 𝑘 𝑇 ̇
3
+ ∑ ∑ ⎢ 𝑘 ((𝐝 ) 𝐝𝑗+3 − ∑ 𝐝𝑖 ⋅ 𝐝𝑗+3 𝐝𝑘 ⋅ 𝐝̇ 𝑖 )⎤
⎥𝐈
3 𝐽 ̃ 𝑗=1 𝑘=1 ⎣∂𝑏𝑗 𝑖=1 ⎦ (4.11.141)

1𝐽 ⎡ ⎛ 6 3 𝑖 6
+ ⎢𝜂 ⎜ ∑ ∑ 𝐝 ⋅ 𝐝 𝐝̇
𝑗+3 𝑖 ⋅ ̅
𝐅̅̅̅ −𝑇 𝑗
𝐕 − ∑ ̅̅̅̅−𝑇 𝐕𝑖 ⎞
𝐝̇ 𝑖+3 ⋅ 𝐅 ⎟⎤⎥ 𝐈,
3 𝐽 ̃ ⎣ ⎝ 𝑗=1 𝑖=1 𝑖=1 ⎠⎦

putting back these expressions into the virtual work expression above and adding the
hourglass internal forces from below we get the same as in [2].

4.11.6 Hourglass
The hourglass resistance is based on a strain energy potential given as
4 3 4 3
𝑉𝜇 𝑗
𝜓= ∑ ∑ ∑ ∑ 𝑏𝑖 𝐵𝑖𝑘 𝑙
𝑗𝑙 𝑏𝑘 , (4.11.142)
12(1 − 𝜈) 𝑖=1 𝑗=1 𝑘=1 𝑙=1

where the inhomogeneous (hourglass) strain quantities are defined as


𝑗
𝑏𝑖 = 𝛃𝑖 ⋅ 𝐃𝑗 , 𝑖 = 1, 2, 3, 4, 𝑗 = 1, 2, 3, (4.11.143)
and the constant symmetric matrix 𝐁 contains geometry and constitutive information for
obtaining accurate results for small deformations. Furthermore, 𝜇 represents the shear
modulus of the material and 𝜈 is the Poisson’s ratio. The hourglass force is then given as
T 𝑘 T
4 3
𝑖 ∂𝜓 ∂𝜓 ⎛∂𝑏𝑗 ⎞
𝐭ℎ = ( ) = ∑ ∑ 𝑘 ⎜ ⎟ , 𝑖 = 0, 1, … ,7. (4.11.144)
∂𝐝𝑖 𝑗=1 𝑘=1 ∂𝑏𝑗 ⎝
∂𝐝𝑖 ⎠

Differentiating the strain quantities results in

∂𝑏𝑘𝑗 T
= −𝐝𝑖 ⋅ 𝐝𝑗+3 (𝐝𝑘 ) , 𝑖 = 1, 2, 3
∂𝐝𝑖
(4.11.145)
∂𝑏𝑘𝑗 𝑗 T
= 𝛿𝑖 (𝐝𝑘 ) , 𝑖 = 1,2,3,4
∂𝐝𝑖+3

which inserted into the expression for the force yields


4
𝐭ℎ𝑖 = −⎛ ⎜∑ 𝐭ℎ𝑗+3 ⊗ 𝐝𝑗+3 ⎞
⎟ 𝐝𝑖 , 𝑖 = 1,2,3
⎝ 𝑗=1 ⎠
3 (4.11.146)
∂𝜓
𝐭ℎ𝑖+3 = ∑ 𝑗 𝐝𝑗 , 𝑖 = 1,2,3,4
𝑗=1 ∂𝑏𝑖

4-36 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

4.11.7 Comparison to Jabareen & Rubin


Putting the material and hourglass force together yields
𝐭𝑖 = 𝐭𝜎𝑖 + 𝐭ℎ𝑖 , 𝑖 = 0,1, . . . ,7, (4.11.147)
hence
4
𝐭𝑖 = ⎛ ̅ 𝛔 − ∑ 𝐭𝑗+3 ⊗ 𝐝𝑗+3 ⎞
⎜𝐽 𝑉 ⎟ 𝐝𝑖 𝑖 = 1,2,3
⎝ 𝑗=1 ⎠
3 (4.11.148)
∂𝜓 𝑗
𝐭𝑖+3 = 𝐽 𝑉
̅ 𝛔𝐅̅̅̅̅−T 𝐕𝑖 + ∑
𝑗
𝐝 𝑖 = 1,2,3,4
𝑗=1 ∂𝑏𝑖
𝑗
In [1], the hourglass strains 𝑏𝑖 are augmented with the geometry parameters 𝐻,𝑊 and 𝐿,
whereas here we have merged this information into the constitutive matrix 𝐁 for the
purpose of simplifying the presentation. In [1] a hyperelastic constitutive law is assumed
with a strain energy potential Σ = Σ(𝐂)̅ and the Cauchy stress is then given as
∂Σ T
𝛔 = 2𝐽 −1
̅ 𝐅̅̅̅̅ ̅̅̅̅ .
𝐅 (4.11.149)
∂𝐂̅
Taking all this into account, and consulting (2.13-2.14) in [1], the implemented CPE element
in LS-DYNA should be consistent with the theory.

4.11.8 Nodal formulation


To be of use in LS-DYNA, the CPE element has to be formulated in terms of the nodal
variables, meaning that the internal forces that are conjugate to 𝐱̇𝑖 , 𝑖 = 0,1, . . . ,7, with
respect to internal energy rate are given as
7
𝐟𝑖 = ∑ 𝐴𝑖𝑗 𝐭𝑗 , 𝑖 = 0,1, … ,7. (4.11.150)
𝑗=0

LS-DYNA DEV 10/27/16 (r:8004) 4-37 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

Figure 4.11. Stress profile for tip loading in two directions for various mesh
densities and distortions

4.11.9 Cantilever Beam Example


First a mesh distortion test for small deformations where a cantilever beam with various
mesh densities and distortions were simulated. The stress profiles for loading in two
directions are shown in Figure 4.11. The tip displacements were monitored and compared
to the analytical solution, and the results for the CPE element is very promising as is shown
in figure below and Table 4.11 when compared to the Belytschko-Bindeman and Puso
element [3,4].
0.7
CPE
0.6 B-B
Puso
0.5

0.4
Error

0.3

0.2

0.1

0
0.2 0.4 0.6 0.8 1
Mesh size

Figure 4.11. Mesh convergence rate for different element formulations

4-38 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

Cosserat Belytschko-Bindeman Puso


1.7% 61.8% 24.8%
0.8% 46.8% 14.7%
0.6% 40.0% 14.5%
0.3% 39.8% 9.2%
0.2% 33.9% 8.5%
0.2% 27.0% 6.2%
0.1% 24.6% 5.3%
0.1% 22.3% 3.6%
0.1% 19.0% 0.9%
0.1% 15.4% 0.3%

Table 4.11. Comparison to analytical solution for tip loading

4.11.10 Compression Test for Rubber Example


The second example is a plane strain deformation of an incompressible rubber block on a
frictionless surface when partially loaded by a rigid plate. The block is modeled with
quadratic tetrahedrons and a large deformation formulation applies. Five different mesh
topologies are investigated for two different element formulations (CPE and full
integration) where the orientation of the tetrahedral elements is different in each mesh
whereas the mesh density is kept constant. The outer edges of the block for the different
meshes are depicted in the figure above and indicate once again that the CPE element
formulation is relatively insensitive to the mesh.

Figure 4.11. Insensitivity illustration of the 10-noded CPE tetrahedron

4.11.11 References
[1]M. Jabareen and M.B. Rubin, A Generalized Cosserat Point Element (CPE) for Isotropic
Nonlinear Elastic Materials including Irregular 3-D Brick and Thin Structures,
J. Mech. Mat. Struct. 3-8, pp. 1465-1498, 2008.
[2]M. Jabareen, E. Hanukah and M.B. Rubin, A Ten Node Tetrahedral Cosserat Point
Element (CPE) for Nonlinear Isotropic Elastic Materials. Comput Mech 52, pp
257-285, 2013.

LS-DYNA DEV 10/27/16 (r:8004) 4-39 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

y
S

n
x i
Figure 4.11. The contour S encloses an area A.

[3]M.A. Puso, A Highly Efficient Enhanced Assumed Strain Physically Stabilized


Hexahedral Element, In. J. Num. Methods. Eng 49-8, pp. 1029-1064, 2000.
[4]T. Belytschko and L.P. Bindeman , Assumed Strain Stabilization of the Eight Node
Hexahedral Element, Comp. Methods Appl. Mech. Eng. 105-2, pp. 225-260,
1993.

4.12 Integral Difference Scheme as Basis For 2D Solids


Two dimensional solid element in LS-DYNA include:
• Plane stress 2D element
• Plane strain 2D shell element
• Axisymmetric 2D Petrov-Galerkin (area weighted) element
• Axisymmetric 2D Galerkin (volume weighted) element
These elements have their origins in the integral difference method of Noh [1964] which is
also used the HEMP code developed by Wilkins [1964, 1969]. In LS-DYNA, both two
dimensional planar and axisymmetric geometries are defined in the 𝑥𝑦 plane. In
axisymmetric geometry, however, the 𝑥 axis corresponds to the radial direction and the 𝑦
axis becomes the axis of symmetry. The integral difference method defines the components
of the gradient of a function 𝐹 in terms of the line integral about the contour 𝑆 which
encloses the area 𝐴:
∂𝐹 ∫ 𝐹(𝐧̂ ⋅ 𝐱̂)𝑑𝑆
= lim 𝑐 ,
∂𝑥 A→0 |A|
(4.151)
∂𝐹 ∫ 𝐹(𝐧̂ ⋅ 𝐲̂ )𝑑𝑆
= lim 𝑐 .
∂𝑦 A→0 |A|

4-40 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

strain rates nodal forces


Figure 4.12. Strain rates are element centered and nodal forces are node centered.

Here, 𝐧̂ is the normal vector to 𝑆 and 𝐱̂ and 𝐲̂ are unit vectors in the x and y directions,
respectively. See Figure 4.11.

In this approach the velocity gradients which define the strain rates are element
centered, and the velocities and nodal forces are node centered. See Figure 4.12. Noting
that the normal vector 𝐧̂ is defined as:
∂𝑦 ∂𝑥
𝐧̂ = 𝐱̂ + 𝐲̂ , (4.152)
∂𝑆 ∂𝑆
and referring to Figure 4.13, we can expand the numerator in equation (4.64):
∂𝑦
∫ 𝐹(n̂ ⋅ x̂ )𝑑𝑆 = ∫ 𝐹 𝑑𝑆
𝑐
∂𝑆 (4.153)
= 𝐹23 (𝑦3 − 𝑦2 ) + 𝐹34 (𝑦4 − 𝑦3 ) + 𝐹41 (𝑦1 − 𝑦4 ) + 𝐹12 (𝑦2 − 𝑦1 ),
where 𝐹𝑖𝑗 = (𝐹𝑖 + 𝐹𝑗 )/2.

Therefore, letting 𝐴 again be the enclosed area, the following expressions are
obtained:
∂𝐹 𝐹23 (𝑦3 − 𝑦2 ) + 𝐹34 (𝑦4 − 𝑦3 ) + 𝐹41 (𝑦1 − 𝑦4 ) + 𝐹12 (𝑦2 − 𝑦1 )
=
∂𝑥 𝐴
(4.154)
(𝐹 − 𝐹4 )(𝑦3 − 𝑦1 ) + (𝑦2 − 𝑦4 )(𝐹3 − 𝐹1 )
= 2 .
2𝐴
Hence, the strain rates in the x and y directions become:
∂𝐹 𝐹23 (𝑦3 − 𝑦2 ) + 𝐹34 (𝑦4 − 𝑦3 ) + 𝐹41 (𝑦1 − 𝑦4 ) + 𝐹12 (𝑦2 − 𝑦1 ) (4.155)
=
∂𝑥 𝐴
LS-DYNA DEV 10/27/16 (r:8004) 4-41 (Solid Elements)
Solid Elements LS-DYNA Theory Manual

y 4 3

1 2

Figure 4.13. Element numbering.

(𝐹2 − 𝐹4 )(𝑦3 − 𝑦1 ) + (𝑦2 − 𝑦4 )(𝐹3 − 𝐹1 )


= ,
2𝐴
and
∂𝑦̇ (𝑦̇2 − 𝑦̇4 )(𝑥3 − 𝑥1 ) + (𝑥2 − 𝑥4 )(𝑦̇3 − 𝑦̇1 )
𝜀𝑦𝑦 = = . (4.156)
∂𝑦 2𝐴
The shear strain rate is given by:
1 ∂𝑦̇ ∂𝑥̇
𝜀𝑥𝑦 = ( + ), (4.157)
2 ∂𝑥 ∂𝑦
where
∂𝑦̇ (𝑦̇2 − 𝑦̇4 )(𝑦3 − 𝑦1 ) + (𝑦2 − 𝑦4 )(𝑦̇3 − 𝑦̇1 )
= ,
∂𝑥 2𝐴
(4.158)
∂𝑥̇ (𝑥̇2 − 𝑥̇4 )(𝑥3 − 𝑥1 ) + (𝑥2 − 𝑥4 )(𝑥̇3 − 𝑥̇1 )
= .
∂𝑦 2𝐴

The zero energy modes, called hourglass modes, as in the three dimensional solid
elements, can be a significant problem. Consider the velocity field given by: 𝑥̇3 = 𝑥̇1 , 𝑥̇2 =
𝑥̇4 , 𝑦̇3 = 𝑦̇1 , and 𝑦̇2 = 𝑦̇4 . As can be observed from Equations (4.97) and (4.98), 𝜀𝑥𝑥 = 𝜀𝑦𝑦 =
𝜀𝑥𝑦 = 0 and the element "hourglasses" irrespective of the element geometry. In the two-
dimensional case, two modes exist versus twelve in three dimensions. The hourglass
treatment for these modes is identical to the approach used for the shell elements, which
are discussed later.

In two-dimensional planar geometries for plane stress and plane strain, the finite
element method and the integral finite difference method are identical. The velocity strains
are computed for the finite element method from the equation:
𝛆̇ = 𝐁𝐯, (4.159)

4-42 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

III

4 3
IV II
1 2

I
Figure 4.14. The finite difference stencil for computing nodal forces is shown.

where 𝛆̇ is the velocity strain vector, B is the strain displacement matrix, and 𝐯 is the nodal
velocity vector. Equation (4.100a) exactly computes the same velocity strains as the integral
difference method if
𝐁 = 𝐁(𝑠, 𝑡)|𝑠=𝑡=0 . (4.160)

The update of the nodal forces also turns out to be identical. The momentum
equations in two-dimensional planar problems are given by
1 ∂𝜎𝑥𝑥 ∂𝜎𝑥𝑦
( + ) = 𝑥̈,
𝜌 ∂𝑥 ∂𝑦
(4.161)
1 ∂𝜎𝑥𝑦 ∂𝜎𝑦𝑦
( + ) = 𝑦̈.
𝜌 ∂𝑥 ∂𝑦
Referring to Figure 4.14, the integral difference method gives Equation (4.113):

1 ∂𝜎𝑥𝑥 𝜎𝑥𝑥1 (𝑦𝐼 − 𝑦𝐼𝑉 ) + 𝜎𝑥𝑥2 (𝑦𝐼𝐼 − 𝑦𝐼 ) + 𝜎𝑥𝑥3 (𝑦𝐼𝐼𝐼 − 𝑦𝐼𝐼 ) + 𝜎𝑥𝑥4 (𝑦𝐼𝑉 − 𝑦𝐼𝐼𝐼 )
= . (4.162)
𝜌 ∂𝑥 1 (𝜌 𝐴 + 𝜌 𝐴 + 𝜌 𝐴 + 𝜌 𝐴 )
1 1 2 2 3 3 4 4
2
An element wise assembly of the discretized finite difference equations is possible
leading to a finite element like finite difference program. This approach is used in the
DYNA2D program by Hallquist [1980].

In axisymmetric geometries additional terms arise that do not appear in planar


problems:

LS-DYNA DEV 10/27/16 (r:8004) 4-43 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

1 ∂𝜎𝑥𝑥 ∂𝜎𝑥𝑦 𝜎𝑥𝑥 − 𝜎𝜃𝜃


( + + ) = 𝑥̈,
𝜌 ∂𝑥 ∂𝑦 𝑥
(4.163)
1 ∂𝜎𝑥𝑦 ∂𝜎𝑦𝑦 ∂𝜎𝑥𝜃
( + + ) = 𝑦̈,
𝜌 ∂𝑥 ∂𝑦 𝑥

where again note that 𝑦 is the axis of symmetry and 𝑥 is the radial direction. The only
difference between finite element approach and the finite difference method is in the
treatment of the terms, which arise from the assumption of axisymmetry. In the finite
difference method the radial acceleration is found from the calculation:
1 𝜎𝑥𝑥 (𝑦𝐼 − 𝑦𝐼𝑉 ) + 𝜎𝑥𝑥2 (𝑦𝐼𝐼 − 𝑦𝐼 ) + 𝜎𝑥𝑥3 (𝑦𝐼𝐼𝐼 − 𝑦𝐼𝐼 ) + 𝜎𝑥𝑥4 (𝑦𝐼𝑉 − 𝑦𝐼𝐼𝐼 )
𝑥̈ = [ 1 −
2 (𝜌1 𝐴1 + 𝜌2 𝐴2 + 𝜌3 𝐴3 + 𝜌4 𝐴4 )
(4.164)
𝜎𝑥𝑦1 (𝑥𝐼 − 𝑥𝐼𝑉 ) + 𝜎𝑥𝑦2 (𝑥𝐼𝐼 − 𝑥𝐼 ) + 𝜎𝑥𝑦3 (𝑥𝐼𝐼𝐼 − 𝑥𝐼𝐼 ) + 𝜎𝑥𝑦4 (𝑥𝐼𝑉 − 𝑥𝐼𝐼𝐼 )
] + 𝛽,
(𝜌1 𝐴1 + 𝜌2 𝐴2 + 𝜌3 𝐴3 + 𝜌4 𝐴4 )
where 𝛽𝑓𝑒 is found by a summation over the four surrounding elements:
1 4 𝜎𝑥𝑥 − 𝜎𝜃𝜃𝑖
𝛽 = ∑[ 𝑖 ]. (4.165)
4 𝑖=1 (𝜌𝑥)𝑖
𝑥𝑖 is the centroid of the ith element defined as the ratio of its volume 𝑉𝑖 and area 𝐴𝑖 :
𝑉𝑖
𝑥𝑖 = , (4.166)
𝐴𝑖
𝜎𝜃𝜃𝑖 is the hoop stress, and 𝜌𝑖 is the current density.

When applying the Petrov-Galerkin finite element approach, the weighting


functions are divided by the radius, 𝑟:
1
∫ 𝛟T (∇ ⋅ 𝛔 + 𝐛 − 𝜌𝐮̈)𝑑𝑉 = ∫ 𝛟T (∇ ⋅ 𝛔 + 𝐛 − 𝜌𝐮̈)𝑑𝐴 = 0, (4.167)
𝑟
where the integration is over the current geometry. This is sometimes referred to as the
"Area Galerkin" method. This approach leads to a time dependent mass vector. LS-DYNA
also has an optional Galerkin axisymmetric element, which leads to a time independent
mass vector. For structural analysis problems where pressures are low the Galerkin
approach works best, but in problems of hydrodynamics where pressures are a large
fraction of the elastic modulus, the Petrov-Galerkin approach is superior since the behavior
along the axis of symmetry is correct.

The Petrov-Galerkin approach leads to equations similar to finite differences. The


radial acceleration is given by.

4-44 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

1 𝜎𝑥𝑥 (𝑦𝐼 − 𝑦𝐼𝑉 ) + 𝜎𝑥𝑥2 (𝑦𝐼𝐼 − 𝑦𝐼 ) + 𝜎𝑥𝑥3 (𝑦𝐼𝐼𝐼 − 𝑦𝐼𝐼 ) + 𝜎𝑥𝑥4 (𝑦𝐼𝑉 − 𝑦𝐼𝐼𝐼 )
𝑥̈ = [ 1 −
2 (𝜌1 𝐴1 + 𝜌2 𝐴2 + 𝜌3 𝐴3 + 𝜌4 𝐴4 )
(4.168)
𝜎𝑥𝑦1 (𝑥𝐼 − 𝑥𝐼𝑉 ) + 𝜎𝑥𝑦2 (𝑥𝐼𝐼 − 𝑥𝐼 ) + 𝜎𝑥𝑦3 (𝑥𝐼𝐼𝐼 − 𝑥𝐼𝐼 ) + 𝜎𝑥𝑦4 (𝑥𝐼𝑉 − 𝑥𝐼𝐼𝐼 )
] + 𝛽𝑓𝑒 ,
(𝜌1 𝐴1 + 𝜌2 𝐴2 + 𝜌3 𝐴3 + 𝜌4 𝐴4 )
where 𝛽𝑓𝑒 is now area weighted.
4 (𝜎𝑥𝑥𝑖 − 𝜎𝜃𝜃𝑖 )𝐴𝑖
1
𝛽𝑓𝑒 = ∑⎡ ⎢ ⎤
⎥. (4.169)
4(𝜌1 𝐴1 + 𝜌2 𝐴2 + 𝜌3 𝐴3 + 𝜌4 𝐴4 ) 𝑖=1 ⎣ 𝑥𝑖 ⎦

In LS-DYNA, the two-dimensional solid elements share the same constitutive


subroutines with the three-dimensional elements. The plane stress element calls the plane
stress constitutive models for shells. Similarly, the plane strain and axisymmetric elements
call the full three-dimensional constitutive models for solid elements. Slight overheads
exists since the strain rate components 𝜀̇𝑦𝑧 and 𝜀̇𝑧𝑥 are set to zero in the two-dimensional
case prior to updating the six stress component; consequently, the additional work is
related to having six stresses whereas only four are needed. A slowdown of LS-DYNA
compared with DYNA2D of fifteen percent has been observed; however, some of the added
cost is due to the internal and hourglass energy calculations, which were not done in
DYNA2D.

4.12.1 Rezoning With 2D Solid Elements


Lagrangian solution techniques generally function well for problems when element
distortions are moderate. When distortions become excessive or when material breaks up,
i.e., simply connected regions become multi-connected, these codes break down, and an
Eulerian approach is a necessity. Between these two extremes, applications exist for which
either approach may be appropriate but Lagrangian techniques are usually preferred for
speed and accuracy. Rezoning may be used to extend the domain of application for
Lagrangian codes.

Rezoning capability was added to DYNA2D in 1980 and to LS-DYNA in version 940.
In the current implementation the rezoning can be done interactively and used to relocate
the nodal locations within and on the boundary of parts. This method is sometimes
referred to as r-adaptive.

The rezoning is accomplished in three steps listed below:

1. Generate nodal values for all variables to be remapped

2. Rezone one or more materials either interactively or automatically with com-


mand file.

3. Initialize remeshed regions by interpolating from nodal point values of old mesh.

LS-DYNA DEV 10/27/16 (r:8004) 4-45 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

2 1 8
Ajusted
node
3 7

4 5 6

Figure 4.15. The stencil used to relax an interior nodal point.

In the first step each variable is approximated globally by a summation over the
number of nodal points 𝑛:
𝑛
𝑔(𝑟, 𝑧) = ∑ 𝑔𝑖 Φ𝑖 (𝑟, 𝑧), (4.170)
𝑖=1

where
Φ𝑖 = set of piecewise continuous global basis functions
𝑔𝑖 = nodal point values

Given a variable to be remapped h(𝑟, 𝑧), a least squares best fit is found by
minimizing the functional

Π = ∫(𝑔 − ℎ)2 𝑑𝐴, (4.171)

i.e.,
𝑑Π
= 0, 𝑖 = 1, 2, … , 𝑛. (4.172)
𝑑𝑔𝑖
This yields the set of matrix equations
𝐌𝐠 = 𝐟, (4.173)
where

𝐌 = ∑ 𝐌𝑒 = ∑ ∫ 𝚽𝚽T 𝑑𝐴 ,
𝑒
(4.174)
𝑒
𝐟 = ∑ 𝐟 = ∑ ∫ ℎ𝚽𝑑𝐴.
𝑒

Lumping the mass makes the calculation of 𝑔 trivial


𝑛
𝑀𝑖 = ∑ 𝑀𝑖𝑗 , (4.175)
𝑗

4-46 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

𝑓𝑖
𝑔𝑖 = .
𝑀𝑖
In step 2, the interactive rezoning phase permits:
• Plotting of solution at current time
• Deletion of elements and slidelines
• Boundary modifications via dekinks, respacing nodes, etc.
• Mesh smoothing

A large number of interactive commands are available and are described in the Help
package. Current results can be displayed by
• Color fringes
• Contour lines
• Vectors plots
• Principal stress lines
• Deformed meshes and material outlines
• Profile plots
• Reaction forces
• Interface pressures along 2D contact interfaces

Three methods are available for smoothing:


• Equipotential
• Isoparametric
• Combination of equipotential and isoparametric.

LS-DYNA DEV 10/27/16 (r:8004) 4-47 (Solid Elements)


Solid Elements LS-DYNA Theory Manual

old mesh
new mesh

Figure 4.16. A four point Gauss quadrature rule over the new element is used to
determine the new element centered value.

In applying the relaxation, the new nodal positions are found and given by Equation
(1.176)
8
∑𝑖=1 𝜉𝑖 𝑥𝑖
𝑥= 8
,
∑𝑖=1 𝜉𝑖
(4.176)
8
∑𝑖=1 𝜉𝑖 𝑦𝑖
𝑦= 8
,
∑𝑖=1 𝜉𝑖
where the nodal positions relative to the node being moved are shown in the sketch in
Figure 4.15.

The weights, 𝜉𝑖 , for equipotential smoothing are


1
𝜉1 = 𝜉5 = [(𝑥7 − 𝑥3 )2 + (𝑦7 − 𝑦3 )2 ], (4.177a)
4
𝜉4 = 𝜉8 = −𝜉2 , (4.177b)
1
𝜉2 = 𝜉6 = [(𝑥1 − 𝑥5 )(𝑥7 − 𝑥3 ) + (𝑦1 − 𝑦5 )(𝑦7 − 𝑦3 )], (4.177c)
2
1
𝜉3 = 𝜉7 = [(𝑥1 − 𝑥5 )2 + (𝑦1 − 𝑦5 )2 ], (4.177d)
4

4-48 (Solid Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Solid Elements

g4 g3

a
s

g1 g2

Figure 4.17. A four point Gauss quadrature rule over the new element is used to
determine the new element centered value.

and are given by


𝜉1 = 𝜉3 = 𝜉5 = 𝜉7 = .50,
(4.178)
𝜉2 = 𝜉4 = 𝜉6 = 𝜉8 = −.25,
for isoparametric smoothing. Since logical regularity is not assumed in the mesh, we
construct the nodal stencil for each interior node and then relax it. The nodes are
iteratively moved until convergence is obtained. In Chapter 14 of this manual, the
smoothing procedures are discussed for three-dimensional applications.

The new element centered values, ℎ∗ , computed in Equation (1.179) are found by a 4
point Gauss Quadrature as illustrated in Figure 4.16.
∫ 𝑔𝑑𝐴
ℎ∗ = . (4.179)
∫ 𝑑𝐴
The Gauss point values are interpolated from the nodal values according to Equation
(1.180). This is also illustrated by Figure 4.17.
𝑔𝑎 = ∑ 𝜙𝑖 (𝑠𝑎 , 𝑡𝑎 )𝑔𝑖 . (4.180)

LS-DYNA DEV 10/27/16 (r:8004) 4-49 (Solid Elements)


Type 16: Fully Integrated Shell LS-DYNA Theory Manual

5
Cohesive elements

The cohesive elements are used for modelling cohesive interfaces between faces of solid
elements (types 19 and 21), faces of shell elements (types 20 and 22) and edges of shell
elements (type 29), typically for treating delamination.
Element 19 𝒒3 Element 20 𝒒3

𝑛4
𝑛1 𝑛4 𝑛1
𝒒1 𝒒1 𝑛3
𝑛3 𝑡
𝑛2 𝑚4 𝑛2 𝑚4
𝑚1
𝑚1
𝑚3 𝑚3
𝑚2 𝒒2 𝑚2 𝒒2

Element 29 𝒒3
𝑡

𝑛4 𝐴
Cohesive layer 𝒒3
𝑛3 𝒙t
𝑛1 𝒒1
𝒒1
𝑛2 𝒒2
𝑚4
𝒒2
𝒙b
𝑚1 𝑚3

𝑚2

Figure 5.18. Illustration of cohesive elements 19, 20 and 29

As a comparison to other elements, the cohesive “strain” is the separation distance (length)
between the two surfaces, and the cohesive stress (force per area) is its conjugate with
respect to the energy surface density (energy per area). Cohesive elements 21 and 22 are
9.50
LS-DYNA Theory Manual Cohesive elements

pentahedral versions of elements 19 and 20, respectively, where the top and bottom surface
are triangles. These elements are not treated here per se, but the only difference is that the
iso-parametric interpolation functions change.

5.1 Kinematics
For this presentation we refer to Figure 5.18 for an illustration. Let
𝒅 = 𝑸𝑻 (𝒙𝑡 − 𝒙𝑏 ) − 𝒅0 , (5.181)
be the separation of the cohesive layer in the local system, where
(5.182)
𝑸 = [𝒒1 𝒒2 𝒒3 ],

is the local coordinate system and 𝒙𝑡 and 𝒙𝑏 are global coordinates on the top and bottom
surfaces for a given iso-parametric coordinate (𝜉 , 𝜂). The distance vector 𝒅0 represents the
initial gap for cases where the cohesive interface has a nonzero thickness, so 𝒅 = 𝟎 initially.
For cohesive element 19 the separation is given directly from the solid element geometries
(sum over i)
𝒙𝑡 = 𝒙𝑖𝑛 𝑁 𝑖 (𝜉 , 𝜂),
(5.183)
𝒙𝑏 = 𝒙𝑖𝑚 𝑁 𝑖 (𝜉 , 𝜂),
where
1
𝑁 𝑖 (𝜉 , 𝜂) = (1 + 𝜉 𝑖 𝜉 )(1 + 𝜂𝑖 𝜂), (5.184)
4
and
𝜉 ∗ = [−1, 1, 1, −1],
𝜂∗ = [−1, −1, 1, 1], (5.185)

are the in-plane shape functions. From here and onwards, superscripts n and m denote the
top and bottom surfaces, respectively, and thus 𝒙𝑖𝑛 and 𝒙𝑖𝑚 are the nodal coordinates
associated with the two surfaces.
For cohesive elements 20 and 29, the separation 𝒅 is updated using an incremental
formulation and we have
𝒅 ̇ = 𝑸𝑇 (𝒙̇𝑡 − 𝒙̇𝑏 ) + 𝑸̇ 𝑇 𝑸𝒅. (5.186)
Furthermore, for cohesive element 20 we have (sum over i)
𝑡
𝒙̇𝑡 = {𝒙̇𝑖𝑛 − 𝝎𝑖𝑛 × 𝒏𝑡 } 𝑁 𝑖 (𝜉 , 𝜂),
2
(5.187)
𝑡
𝒙̇𝑏 = {𝒙̇𝑖𝑚 + 𝝎𝑖𝑚 × 𝒏𝑏 } 𝑁 𝑖 (𝜉 , 𝜂),
2

LS-DYNA DEV 10/27/16 (r:8004) 4-51 (Cohesive elements)


Type 16: Fully Integrated Shell LS-DYNA Theory Manual

where the thicknesses of the two shells adjacent to the cohesive layer are denoted t,
currently assumed to be the same for the both. In these equations, 𝒏𝑡 and 𝒏𝑏 are the top and
bottom shell normal, initially equal to 𝒒3 but they may evolve independently with time.
For cohesive element 29 we instead have
𝑡 1−𝜂 𝑡 1+𝜂
𝒙̇𝑡 = {𝒙̇1𝑛 + 𝜉 𝝎1𝑛 × 𝒏𝑡 } + {𝒙̇2𝑛 + 𝜉 𝝎2𝑛 × 𝒏𝑡 } ,
2 2 2 2
(5.188)
𝑡 1−𝜂 𝑡 1+𝜂
𝐱̇𝑏 = {𝐱̇4𝑚 + 𝜉 𝝎4𝑚 × 𝐧𝑏 } + {𝐱̇3𝑚 + 𝜉 𝝎3𝑚 × 𝐧𝑏 } .
2 2 2 2
In these expressions 𝝎𝑖𝑚 and 𝝎𝑖𝑛 denote nodal rotational velocities, and also note that for
evaluation the velocities of 𝒙𝑡 and 𝒙𝑏 we assume that the fiber pointing from assumed mid
layer coincides with that of the coordinate axes. This is in analogy to how the Belytschko-
Tsay element is treating the fiber vectors and presumably enhances robustness of the
elements. Note also that the shell normal are in this case initially equal to 𝒒1 .
For the local coordinate system, cohesive elements 19 and 20 evaluate this according to the
invariant node numbering approach for shells using the mid layer node coordinates
𝒙𝑖𝑛 + 𝒙𝑖𝑚
𝒙̅𝑖 = , 𝑖 = 1,2,3,4, (5.189)
2
as follows. First let
𝒙̅3 − 𝒙̅1
𝒆1 = ,
∣𝒙̅3 − 𝒙̅1 ∣
(5.190)
𝒙̅ − 𝒙̅2
𝒆2 = 4 ,
|𝒙̅4 − 𝒙̅2 |
and then
𝒆1 + 𝒆2
𝒒1 = − ,
|𝒆1 + 𝒆2 |
(5.191)
𝒆 − 𝒆2
𝒒2 = 1 ,
|𝒆1 − 𝒆2 |

followed by
𝒒3 = 𝒒1 × 𝒒2 . (5.192)
Cohesive element 29 starts by computing
𝒙2𝑛 + 𝒙3𝑚 − 𝒙1𝑛 − 𝒙4𝑚
𝒒2 = 𝑛 , (5.193)
∣𝒙2 + 𝒙3𝑚 − 𝒙1𝑛 − 𝒙4𝑚 ∣
followed by
𝒒 = 𝒙4𝑛 + 𝒙3𝑛 − 𝒙1𝑚 − 𝒙2𝑚 , (5.194)

𝒒 − 𝒒2 𝒒𝑇 𝒒2
𝒒3 = , (5.195)
|𝒒 − 𝒒2 𝒒𝑇 𝒒2 |

9.52
LS-DYNA Theory Manual Cohesive elements

and
𝒒1 = 𝒒2 × 𝒒3 . (5.196)

5.2 Constitutive law


The cohesive constitutive law amounts to determine the normal and shear stress, expressed
here as the stress vector 𝝈, as function of the separation vector 𝒅, 𝝈 = 𝝈(𝒅, … ). The typical
appearance of each component of this vector is illustrated in Figure 5.19., the interface
behaves elastically up to a critical separation distance 𝑑𝑒 and peak stress 𝜎𝑒 after which
damage commences. The interface is damaged and failure occur at a certain critical
separation distance 𝑑𝑐 , the unloading is typically elastic as indicated by the dashed arrow.

For detailed information on individual cohesive constitutive laws we refer to the materials
section.
𝜎
𝜎𝑒

𝑑
𝑑𝑒 𝑑𝑐

Figure 5.19.Common stress versus separation for a cohesive interface

5.3 Nodal forces


The principle of virtual work states that (sum over i)

∫ 𝒅𝑇̇ 𝝈𝑑𝐴 = {𝒙̇𝑖𝑚 }𝑇 𝒇𝑖𝑚 + {𝒙̇𝑖𝑛 }𝑇 𝒇𝑖𝑛 + {𝝎𝑖𝑚 }𝑇 𝒓𝑖𝑚 + {𝝎𝑖𝑛 }𝑇 𝒓𝑖𝑛 , (5.197)
𝐴
𝑗 𝑗
where 𝒇𝑖 and 𝒓𝑖 is the nodal force and moment for node i on element j, respectively. The
area A represents the cohesive mid layer spanned by the iso-parametric representation and
this is used to identify the nodal forces and moments. In the following we work out the
details for cohesive element 29. Cohesive elements 19 and 20 are treated analogous.

LS-DYNA DEV 10/27/16 (r:8004) 4-53 (Cohesive elements)


Type 16: Fully Integrated Shell LS-DYNA Theory Manual

Using Equation (5.186) we also have

∫ 𝐝̇ 𝑇 𝛔𝑑𝐴 = ∫ 𝐱̇𝑡𝑇 𝐐𝛔𝑑𝐴 − ∫ 𝐱̇𝑏𝑇 𝐐𝛔𝑑𝐴 − ∫ 𝐝𝑇 𝐐̇ 𝑇 𝐐𝛔𝑑𝐴, (5.198)


𝐴 𝐴 𝐴 𝐴

and before continuing we rewrite (5.188) as


1−𝜂 1−𝜂 𝑡 1+𝜂 1+𝜂 𝑡
𝒙̇𝑡𝑇 = {𝒙̇1𝑛 }𝑇 − {𝝎1𝑛 }𝑇 𝑹𝑡𝑇 𝜉 + {𝒙̇2𝑛 }𝑇 − {𝝎2𝑛 }𝑇 𝑹𝑡𝑇 𝜉 ,
2 2 2 2 2 2
(5.199)
1−𝜂 1−𝜂 𝑡 1+𝜂 1+𝜂 𝑡
𝒙̇𝑏𝑇 = {𝒙̇4𝑚 }𝑇 − {𝝎4𝑚 }𝑇 𝑹𝑏𝑇 𝜉 + {𝒙̇3𝑚 }𝑇 − {𝝎3𝑚 }𝑇 𝑹𝑏𝑇 𝜉 ,
2 2 2 2 2 2
where 𝑹∗ is the linear operator defined by
𝑹∗ 𝝎 = 𝒏∗ × 𝝎 (5.200)
and insert this into the right of (5.198) which after some simplifications gives

∫ 𝒅𝑇̇ 𝝈𝑑𝐴 = − ∫ 𝒅𝑇 𝑸̇ 𝑇 𝑸𝝈𝑑𝐴


𝐴 𝐴

1−𝜂 1−𝜂 𝑡
+{𝐱̇1𝑛 }𝑇 ∫ ( ) 𝐐𝛔𝑑𝐴 + {𝝎1𝑛 }𝑇 ∫ (− 𝜉 ) 𝐑𝑡𝑇 𝐐𝛔𝑑𝐴
2 2 2
𝐴 𝐴

1+𝜂 1+𝜂 𝑡 (5.201)


+{𝒙̇2𝑛 }𝑇 ∫ ( ) 𝑸𝝈𝑑𝐴 + {𝝎2𝑛 }𝑇 ∫ (− 𝜉 ) 𝑹𝑡𝑇 𝑸𝝈𝑑𝐴
2 2 2
𝐴 𝐴

1−𝜂 1−𝜂 𝑡
+{𝒙̇4𝑚 }𝑇 ∫ (− ) 𝑸𝝈𝑑𝐴 + {𝝎4𝑚 }𝑇 ∫ ( 𝜉 ) 𝑹𝑏𝑇 𝑸𝝈𝑑𝐴
2 2 2
𝐴 𝐴

1+𝜂 1+𝜂 𝑡
+{𝒙̇3𝑚 }𝑇 ∫ (− ) 𝑸𝝈𝑑𝐴 + {𝝎3𝑚 }𝑇 ∫ ( 𝜉 ) 𝑹𝑏𝑇 𝑸𝝈𝑑𝐴.
2 2 2
𝐴 𝐴
If the first term on the right hand side of (5.201) is neglected we can, using (5.197) and
(5.201), identify the nonzero nodal forces and moments

1−𝜂 𝑡 1−𝜂
𝒇1𝑛 = ∫ ( ) 𝑸𝝈𝑑𝐴, 𝒓1𝑛 = − ∫ ( 𝜉 ) 𝑹𝑡𝑇 𝑸𝝈𝑑𝐴,
2 2 2
𝐴 𝐴

1+𝜂 𝑡 1+𝜂
𝒇2𝑛 = ∫ ( ) 𝑸𝝈𝑑𝐴, 𝒓2𝑛 = − ∫ ( 𝜉 ) 𝑹𝑡𝑇 𝑸𝝈𝑑𝐴,
2 2 2
𝐴 𝐴
(5.202)
𝑡 1+𝜂
𝒇3𝑚 = −𝒇2𝑛 , 𝒓3𝑚 = ∫ ( 𝜉 ) 𝑹𝑏𝑇 𝑸𝝈𝑑𝐴,
2 2
𝐴

𝑡 1−𝜂
𝒇4𝑚 = −𝒇1𝑛 , 𝒓4𝑚 = ∫ ( 𝜉 ) 𝑹𝑏𝑇 𝑸𝝈𝑑𝐴.
2 2
𝐴

9.54
LS-DYNA Theory Manual Cohesive elements

In the implementation these integrals are evaluated using 4-point Gaussian quadrature,
where the integration point locations are given by
1 1 1 1
𝜉∗ = ⎡
⎢− , , ,− ⎤⎥,
⎣ √3 √3 √3 √3⎦
(5.203)
1 11 1
𝜂∗ = ⎡
⎢− ,− , , ⎤⎥.
⎣ √3 √3 √3 √3⎦

Thus an integral is evaluated as


4
∫ 𝜙(𝜉 , 𝜂)𝛔𝑑𝐴 ≈ ∑ 𝜙(𝜉𝑖 , 𝜂𝑖 )𝛔𝑖 𝐴𝑖 , (5.204)
𝐴 𝑖=1

where 𝜙 is an arbitrary function of the iso-parametric coordinates, 𝐴𝑖 in the right hand side
stands for the area of the cohesive layer and 𝛔𝑖 is the cohesive interface stress, both
evaluated at and with respect to integration point i.

5.4 Drilling constraint in shell 29


From (5.188) and the picture of shell type 29 in Figure 5.18, we see that rotational velocities
with respect to the adjacent shell normals will not induce translational velocities in the
integration points, so a stabilization scheme is applied. To this end we introduce the
generalized drilling strains δ𝑛𝑖 and δ𝑚𝑗 , 𝑖 = 1,2, 𝑗 = 3,4, distances that are incremented by
their respective velocities

𝛿𝑖𝑛̇ = 𝒏𝑇𝑡 {𝝎𝑖𝑛 𝑑21 − 𝑹21 (𝒙̇2𝑛 − 𝒙̇1𝑛 )}, 𝑖 = 1,2,


(5.205)
̇ = 𝒏𝑇𝑏 {𝝎𝑗𝑚 𝑑34 − 𝑹34 (𝒙̇3𝑚 − 𝒙̇4𝑚 )}, 𝑗 = 3,4,
𝛿𝑗𝑚

where
𝑑21 = ∣𝒙2𝑛 − 𝒙1𝑛 ∣,
𝑑34 = ∣𝒙3𝑚 − 𝒙4𝑚 ∣, (5.206)

LS-DYNA DEV 10/27/16 (r:8004) 4-55 (Cohesive elements)


Type 16: Fully Integrated Shell LS-DYNA Theory Manual

and we make use of the following definitions for arbitrary vector 𝐯,


1
𝑹21 𝒗 = (𝒙 𝑛 − 𝒙1𝑛 ) × 𝒗,
𝑑21 2
(5.207)
1
𝑹34 𝒗 = (𝒙3𝑚 − 𝒙4𝑚 ) × 𝒗.
𝑑34
A characteristic material stiffness 𝐸, typically a fraction of the elastic stiffness of the
underlying cohesive material, is used to set up the drilling stress
𝜍𝑖𝑛 = 𝐸𝛿𝑖𝑛 , 𝑖 = 1,2,
(5.208)
𝜍𝑗𝑚
= 𝐸𝛿𝑗𝑚 ,
𝑗 = 3,4,
and the stabilization nodal forces are evaluated as
𝑇
𝐟1𝑛 = 𝐴(𝜍1𝑛 +𝜍2𝑛 )𝐑21 𝐧𝑡 , 𝐫1𝑛 = 𝐴𝜍1𝑛 𝑑21 𝐧𝑡 ,
𝐟2𝑛 = −𝐟1𝑛 , 𝐫2𝑛 = 𝐴𝜍2𝑛 𝑑21 𝐧𝑡 ,
(5.209)
𝐟3𝑚 = −𝐟4𝑚 , 𝐫3𝑚 = 𝐴𝜍3𝑚 𝑑34 𝐧𝑏 ,
𝑇
𝐟4𝑚 = 𝐴(𝜍3𝑚 +𝜍4𝑚 )𝐑34 𝐧𝑏 , 𝐫4𝑚 = 𝐴𝜍4𝑚 𝑑34 𝐧𝑏 ,
and these forces and moments are added to the structural ones in the previous section.

5.5 Rotational masses in shell 29


The rotational mass in shell 29 is determined from a simple energy criterion. Assuming
that the cohesive layer is thin and the layer spins with a rotational velocity 𝜔 around axis
𝒒2 , the kinetic energy from the nodal rotational masses 𝑚𝑟 is
1
𝑊 = (𝑚𝑟 + 𝑚𝑟 + 𝑚𝑟 + 𝑚𝑟 )𝜔2 = 2𝑚𝑟 𝜔2 . (5.210)
2

This is compared to the corresponding kinetic energy for an equivalent solid type 19
cohesive layer, using that the shell type 29 translational nodal mass 𝑚𝑡 is twice that of the
solid type 19 nodal masses,
2
1 𝑚𝑡 𝑚𝑡 𝑚𝑡 𝑚𝑡 𝑚𝑡 𝑚𝑡 𝑚𝑡 𝑚𝑡 𝑡 1
𝑊= ( + + + + + + + ) ( 𝜔) = 𝑚𝑡 𝑡2 𝜔2 , (5.211)
2 2 2 2 2 2 2 2 2 2 2

which results in
1
𝑚𝑟 = 𝑚𝑡 𝑡 2 . (5.212)
4

9.56
LS-DYNA Theory Manual Cohesive elements

5.6 Stiffness matrix in shell 29


The stiffness matrix is the sum of contributions from the constitutive law and the
drilling force, and is for simplicity implemented for a force free configuration, i.e.,
𝝈 = 𝟎,
(5.213)
𝝇 = 𝟎,
where we have used the notation
𝜍𝑛
⎡ 1𝑛 ⎤
𝜍2 ⎥
𝝇=⎢
⎢𝜍𝑚 ⎥. (5.214)
⎢ 3⎥
⎣𝜍4𝑚 ⎦

For a given integration point and referring to Equation (5.202) and (5.209), we collect these
nodal force triplets into complete nodal vectors 𝒇 and 𝒈, associated with the constitutive
and drilling part, respectively. That is,
𝒇1𝑛
⎡ 𝑛⎤
⎢ 𝒓1 ⎥
⎢𝒇 𝑛 ⎥
⎢ 2⎥
⎢ 𝒓𝑛 ⎥
𝒇 =⎢ 𝑚 2⎥
, (5.215)
⎢𝒇3 ⎥
⎢ 𝑚⎥
⎢ 𝒓3 ⎥
⎢ 𝑚⎥
⎢𝒇4 ⎥
⎣𝒓4𝑚 ⎦
and the same expression holds for 𝒈. Likewise we let 𝒗 be the collection of nodal velocities,
i.e.,
𝒙̇𝑛
⎡ 1𝑛 ⎤
⎢ 𝝎1 ⎥
⎢ 𝒙̇𝑛 ⎥
⎢ 2𝑛 ⎥
⎢𝝎 ⎥
𝒗 = ⎢ 𝑚2 ⎥, (5.216)
⎢ 𝒙̇3 ⎥
⎢𝝎3𝑚 ⎥
⎢ 𝑚⎥
⎢ 𝒙̇4 ⎥
⎣𝝎4𝑚 ⎦
and note that we can identify generalized strain-displacement matrices 𝑩𝑓 (3 by 24 matrix)
and 𝑩𝑔 (4 by 24 matrix) from (5.186) and (5.188) as well as (5.205) so that
𝒅 ̇ = 𝑩𝑓 𝒗,
(5.217)
𝜹 ̇ = 𝑩𝑔 𝒗,
where we have collected the drilling kinematic velocities in a vector

LS-DYNA DEV 10/27/16 (r:8004) 4-57 (Cohesive elements)


Type 16: Fully Integrated Shell LS-DYNA Theory Manual

𝛿𝑛̇
⎡ 1𝑛 ⎤
⎢ 𝛿2̇ ⎥
𝜹̇ = ⎢
⎢𝛿𝑚

⎥ (5.218)
⎢ 3̇ ⎥
⎣𝛿4𝑚̇ ⎦
With this notation we can rewrite (5.202) and (5.209) into a more compact form
𝒇 = 𝑩𝑇𝑓 𝝈𝐴,
(5.219)
𝒈 = 𝑩𝑇𝑔 𝝇𝐴,
with 𝐴 here being the area of the integration point of interest. The stiffness matrix is then
simply the differentiation of these force vectors with respect to the nodal coordinates, and
by using (5.208), (5.213) and the chain rule of differentiation we get
𝜕𝝈
𝑲𝑓 = 𝑩𝑇𝑓 𝑩 𝐴,
𝜕𝒅 𝑓 (5.220)
𝑲𝑔 = 𝐸𝑩𝑇𝑔 𝑩𝑔 𝐴,
where 𝜕𝝈/𝜕𝒅 is the constitutive tangent from the cohesive material used.

9.58
LS-DYNA Theory Manual Belytschko Beam

6
Belytschko Beam

The Belytschko beam element formulation [Belytschko et al. 1977] is part of a family
of structural finite elements, by Belytschko and other researchers that employ a ‘co-
rotational technique’ in the element formulation for treating large rotation. This section
discusses the co-rotational formulation, since the formulation is most easily described for a
beam element, and then describes the beam theory used to formulate the co-rotational
beam element.

6.1 Co-rotational Technique


In any large displacement formulation, the goal is to separate the deformation
displacements from the rigid body displacements, as only the deformation displacements
give rise to strains and the associated generation of strain energy. This separation is
usually accomplished by comparing the current configuration with a reference
configuration.

The current configuration is a complete description of the deformed body in its


current spatial location and orientation, giving locations of all points (nodes) comprising
the body. The reference configuration can be either the initial configuration of the body,
i.e., nodal locations at time zero, or the configuration of the body at some other state (time).
Often the reference configuration is chosen to be the previous configuration, say at time
𝑡𝑛 = 𝑡𝑛+1 − Δ𝑡.

The choice of the reference configuration determines the type of deformations that
will be computed: total deformations result from comparing the current configuration with
the initial configuration, while incremental deformations result from comparing with the
previous configuration. In most time stepping (numerical) Lagrangian formulations,
incremental deformations are used because they result in significant simplifications of
other algorithms, chiefly constitutive models.

LS-DYNA DEV 10/27/16 (r:8004) 5-1 (Belytschko Beam)


Belytschko Beam LS-DYNA Theory Manual

^
Ȳ Y
b2
e02
e01
^
b1 X
I J

(a) Initial Configuration


Figure 6.1. Co-rotational coordinate system: (a) initial configuration, (b) rigid
rotational configuration and (c) deformed configuration.

A direct comparison of the current configuration with the reference configuration


does not result in a determination of the deformation, but rather provides the total (or
incremental) displacements. We will use the unqualified term displacements to mean
either the total displacements or the incremental displacements, depending on the choice of
the reference configuration as the initial or the last state. This is perhaps most obvious if
the reference configuration is the initial configuration. The direct comparison of the
current configuration with the reference configuration yields displacements, which contain
components due to deformations and rigid body motions. The task remains of separating
the deformation and rigid body displacements. The deformations are usually found by
subtracting from the displacements an estimate of the rigid body displacements. Exact
rigid body displacements are usually only known for trivial cases where they are
prescribed a priori as part of a displacement field. The co-rotational formulations provide
one such estimate of the rigid body displacements.

The co-rotational formulation uses two types of coordinate systems: one system
associated with each element, i.e., element coordinates which deform with the element, and
another associated with each node, i.e., body coordinates embedded in the nodes. (The
term ‘body’ is used to avoid possible confusion from referring to these coordinates as
‘nodal’ coordinates. Also, in the more general formulation presented in [Belytschko et al.,
1977], the nodes could optionally be attached to rigid bodies. Thus the term ‘body
coordinates’ refers to a system of coordinates in a rigid body, of which a node is a special
case.) These two coordinate systems are shown in the upper portion of Figure 6.1.

The element coordinate system is defined to have the local x-axis 𝐱̂ originating at
node 𝐼 and terminating at node 𝐽; the local y-axis 𝐲̂ and, in three dimension, the local z-axis
𝐳̂, are constructed normal to 𝐱̂. The element coordinate system (𝐱̂, 𝐲̂ , 𝐳̂) and associated unit
vector triad (𝐞1 , 𝐞2 , 𝐞3 ) are updated at every time step by the same technique used to
construct the initial system; thus the unit vector e1 deforms with the element since it always
points from node 𝐼 to node 𝐽.

The embedded body coordinate system is initially oriented along the principal
inertial axes; either the assembled nodal mass or associated rigid body inertial tensor is
5-2 (Belytschko Beam) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Belytschko Beam

^
X
J

^
Y b2
e1
e2 X̄
I b1

(b) Rigid Rotation Configuration

Figure 6.2. Co-rotational coordinate system: (a) initial configuration, (b) rigid
rotational configuration and (c) deformed configuration.

used in determining the inertial principal values and directions. Although the initial
orientation of the body axes is arbitrary, the selection of a principal inertia coordinate
system simplifies the rotational equations of motion, i.e., no inertial cross product terms are
present in the rotational equations of motion. Because the body coordinates are fixed in the
node, or rigid body, they rotate and translate with the node and are updated by integrating
the rotational equations of motion, as will be described subsequently.

The unit vectors of the two coordinate systems define rotational transformations
between the global coordinate system and each respective coordinate system. These
transformations operate on vectors with global components 𝐀 = (𝐴𝑥 , 𝐴𝑦 , 𝐴𝑧 ), body
coordinates components 𝐀 ̅̅̅̅̅̅ = (𝐴𝑥̅ , 𝐴𝑦̅ , 𝐴𝑧̅ ), and element coordinate components A
̂=
(𝐴𝑥̂ , 𝐴𝑦̂ , 𝐴𝑧̂ ) which are defined as:

⎧𝐴 𝑏 𝑏2𝑥 𝑏3𝑥 ⎧𝐴 𝑥⎫
̅
{ 𝑥⎫} ⎡ 1𝑥 ⎤{ ̅ }
𝐀 = ⎨𝐴𝑦 ⎬ = ⎢
⎢𝑏1𝑦 𝑏2𝑦 𝑏3𝑦 ⎥ ̅̅̅̅̅̅},
⎥ ⎨𝐴𝑦 ⎬ = [𝛌]{𝐀 (6.1)
{ }
⎩𝐴𝑧 ⎭ ⎣𝑏1𝑧 { }
𝑏2𝑧 𝑏3𝑧 ⎦ ⎩𝐴𝑧̅ ⎭
where 𝑏𝑖𝑥 , 𝑏𝑖𝑦 , 𝑏𝑖𝑧 are the global components of the body coordinate unit vectors. Similarly
for the element coordinate system:

𝑒3𝑥 ⎧ ̂ ⎫
⎧𝐴
{ 𝑥⎫
𝑒
} ⎡ 1𝑥
𝑒2𝑥 {𝐴𝑥 }
{ }
𝐀 = ⎨ 𝑦 ⎬ = ⎢𝑒1𝑦
𝐴 𝑒2𝑦 𝑒3𝑦 ⎤
⎥ ⎨𝐴𝑦̂ ⎬ = [𝛍]{𝐀
̂ }, (6.2)
{
⎩𝐴𝑧 }
⎭ ⎣𝑒1𝑧 𝑒2𝑧 {
𝑒3𝑧 ⎦ { ̂ } }
⎩𝐴𝑧 ⎭
where 𝑒𝑖𝑥 , 𝑒𝑖𝑦 , 𝑒𝑖𝑧 are the global components of the element coordinate unit vectors. The
inverse transformations are defined by the matrix transpose, i.e.,
̅̅̅̅̅̅} = [𝛌]T {𝐀}
{𝐀
(6.3)
̂ } = [𝛍]T {𝐀},
{𝐀
since these are proper rotational transformations.

LS-DYNA DEV 10/27/16 (r:8004) 5-3 (Belytschko Beam)


Belytschko Beam LS-DYNA Theory Manual

^ Ȳ
Y b2
e2 ^
e1 X
J
I e01
b1

(c) Deformed Configuration

Figure 6.3. Co-rotational coordinate system: (a) initial configuration, (b) rigid
rotational configuration and (c) deformed configuration.

The following two examples illustrate how the element and body coordinate system
are used to separate the deformations and rigid body displacements from the displace-
ments:

Rigid Rotation. First, consider a rigid body rotation of the beam element about node 𝐼, as
shown in the center of Figure 6.2, i.e., consider node 𝐼 to be a pinned connection. Because
the beam does not deform during the rigid rotation, the orientation of the unit vector 𝐞1 in
the initial and rotated configuration will be the same with respect to the body coordinates.
If the body coordinate components of the initial element unit vector 𝐞01 were stored, they
would be identical to the body coordinate components of the current element unit vector
e1 .

Deformation Rotation. Next, consider node 𝐼 to be constrained against rotation, i.e., a


clamped connection. Now node 𝐽 is moved, as shown in the lower portion of Figure 6.3,
causing the beam element to deform. The updated element unit vector e1 is constructed
and its body coordinate components are compared to the body coordinate components of
the original element unit vector 𝐞01 . Because the body coordinate system did not rotate, as
node I was constrained, the original element unit vector and the current element unit
vector are not colinear. Indeed, the angle between these two unit vectors is the amount of
rotational deformation at node I, i.e.,
𝐞1 × 𝐞01 = 𝜃ℓ 𝐞3 . (6.4)

Thus the co-rotational formulation separates the deformation and rigid body
deformations by using:
• a coordinate system that deforms with the element, i.e., the element coordinates;
• or a coordinate system that rigidly rotates with the nodes, i.e., the body coordinates;
Then it compares the current orientation of the element coordinate system with the initial
element coordinate system, using the rigidly rotated body coordinate system, to determine
the deformations.

5-4 (Belytschko Beam) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Belytschko Beam

6.2 Belytschko Beam Element Formulation


The deformation displacements used in the Belytschko beam element formulation
are:
𝐝̂ T = {𝛿𝐼𝐽 , 𝜃𝑥𝐽𝐼
̂ , 𝜃𝑦𝐼
̂ , 𝜃𝑦𝐽
̂ , 𝜃𝑧𝐼
̂ , 𝜃𝑧𝐽
̂ }, (6.5)
where,
𝛿𝐼𝐽 = length change
̂ = torsional deformation
𝜃𝑥𝐽𝐼
̂ , 𝜃𝑦𝐽
𝜃𝑦𝐼 ̂ , 𝜃𝑧𝐼
̂ , 𝜃𝑧𝐽
̂ = bending rotation deformations

The superscript ^ emphasizes that these quantities are defined in the local element
coordinate system, and 𝐼 and 𝐽 are the nodes at the ends of the beam.

The beam deformations, defined above in Equation (6.5), are the usual small
displacement beam deformations (see, for example, [Przemieniecki 1986]). Indeed, one
advantage of the co-rotational formulation is the ease with which existing small
displacement element formulations can be adapted to a large displacement formulation
having small deformations in the element system. Small deformation theories can be easily
accommodated because the definition of the local element coordinate system is
independent of rigid body rotations and hence deformation displacement can be defined
directly.

6.2.1 Calculation of Deformations


The elongation of the beam is calculated directly from the original nodal coordinates
(𝑋𝐼 , 𝑌𝐼 , 𝑍𝐼 ) and the total displacements (𝑢𝑥𝐼 , 𝑢𝑦𝐼 , 𝑢𝑧𝐼 ):
1
𝛿𝐼𝐽 = [2(𝑋𝐽𝐼 𝑢𝑥𝐽𝐼 + 𝑌𝐽𝐼 𝑢𝑦𝐽𝐼 + 𝑍𝐽𝐼 𝑢𝑧𝐽𝐼 ) + 𝑢2𝑥𝐽𝐼 + 𝑢2𝑦𝐽𝐼 + 𝑢2𝑧𝐽𝐼 ], (6.6)
𝑙 + 𝑙𝑜
where
𝑋𝐽𝐼 = 𝑋𝐽 − 𝑋𝐼
(6.7)
𝑢𝑥𝐽𝐼 = 𝑢𝑥𝐽 − 𝑢𝑥𝐼 , etc.
The deformation rotations are calculated using the body coordinate components of
the original element coordinate unit vector along the beam axis, i.e., 𝐞01 , as outlined in the
previous section. Because the body coordinate components of initial unit vector 𝐞01 rotate
with the node, in the deformed configuration it indicates the direction of the beam’s axis if
no deformations had occurred. Thus comparing the initial unit vector 𝐞01 with its current

LS-DYNA DEV 10/27/16 (r:8004) 5-5 (Belytschko Beam)


Belytschko Beam LS-DYNA Theory Manual

orientation 𝐞1 indicates the magnitude of deformation rotations. Forming the vector cross
product between 𝐞01 and 𝐞1 :
𝐞1 × 𝐞01 = 𝜃𝑦̂ 𝐞2 + 𝜃𝑧̂ 𝐞3 , (6.8)
where
𝜃𝑦̂ = is the incremental deformation about the local 𝑦̂ axis

𝜃𝑧̂ = is the incremental deformation about the local 𝑧̂ axis


The calculation is most conveniently performed by transforming the body components of
the initial element vector into the current element coordinate system:
0 ⎫
⎧𝑒1𝑥 0 ⎫
⎧𝑒1𝑥
{ ̂ ̅ }
{  0 } } T
{
{  0 }
⎨𝑒1𝑦 ̂ ⎬ = [𝛍] [𝛌] ⎨𝑒1𝑦
̅ ⎬. (6.9)
{ 0}
{ } { 0}
{ }
⎩𝑒1𝑧 ̂ ⎭ ⎩𝑒1𝑧 ̅ ⎭
Substituting the above into Equation (4.10)
𝐞1 𝐞2 𝐞3
0 ⎡
𝐞1 × 𝐞1 = det ⎢ 1 0 0⎤  0  0
̂ 𝐞3 = 𝜃𝑦̂ 𝐞2 + 𝜃𝑧̂ 𝐞3 .
 0  0  0
⎥ = −𝑒1𝑧
̂ 𝐞2 + 𝑒1𝑦 (6.10)
⎣𝑒1𝑥̂ 𝑒1𝑦
̂ 𝑒1𝑧
̂ ⎦
Thus,
0
𝜃𝑦̂ = −𝑒1𝑧
̂
(6.11)
 0
𝜃𝑧̂ = 𝑒1𝑦
̂ .
The torsional deformation rotation is calculated from the vector cross product of
initial unit vectors, from each node of the beam, that were normal to the axis of the beam,
̂ 0 and 𝑒2𝐽
i.e., 𝑒2𝐼 ̂ 0 ; note that 𝑒3𝐼
̂ 0 and 𝑒3𝐽
̂ 0 could also be used. The result from this vector cross
product is then projected onto the current axis of the beam, i.e.,
𝐞1 𝐞2 𝐞3
⎡  0  0  0 ⎤
̂ = 𝐞1 ⋅ (𝐞̂02𝐼 × 𝐞̂02𝐽 ) = 𝐞1 ⋅ det ⎢𝑒𝑥2𝐼
𝜃𝑥𝐽𝐼 ̂ 𝑒𝑦2𝐼
̂ 𝑒𝑧2𝐼
̂ ⎥ = 𝑒 0̂ 𝑒 0̂ − 𝑒 0̂ 𝑒 0̂ . (6.12)
⎢ ⎥ 𝑦2𝐼 𝑧2𝐽 𝑦2𝐽 𝑧2𝐼
 0  0  0
⎣𝑒𝑥2𝐽̂ 𝑒𝑦2𝐽̂ 𝑒𝑧2𝐽
̂ ⎦

̅ 0 and 𝑒2𝐽
Note that the body components of 𝑒2𝐼 ̅ 0 are transformed into the current element
coordinate system before performing the indicated vector products.

6.2.2 Calculation of Internal Forces


There are two methods for computing the internal forces for the Belytschko beam
element formulation:

1. functional forms relating the overall response of the beam, e.g., moment-curvature
relations,

2. direct through-the-thickness integration of the stress.


5-6 (Belytschko Beam) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Belytschko Beam

Currently only the former method, as explained subsequently, is implemented; the direct
integration method is detailed in [Belytschko et al., 1977].

Axial Force. The internal axial force is calculated from the elongation of the beam 𝛿 as
given by Equation (6.6), and an axial stiffness:
𝑓𝑥𝐽̂ = 𝐾 𝑎 𝛿, (6.13)
where
𝐴𝐸
𝐾𝑎 = = is the axial stiffness
𝑙0
𝐴 = cross sectional area of the beam
𝐸 = Young's Modulus
𝑙0 = original length of the beam

Bending Moments. The bending moments are related to the deformation rotations by
𝑚
̂ 𝑦𝐼 𝐾𝑦𝑏 4 + 𝜙𝑦 2 − 𝜙𝑦 ⎧ ̂ ⎫
{𝜃𝑦𝐼 }
{ }= [ ] ⎨ ⎬, (6.14a)
𝑚
̂ 𝑦𝐽 1 + 𝜙𝑦 2 − 𝜙𝑦 4 + 𝜙𝑦 { ̂
⎩𝜃𝑦𝐽 }

𝑚
̂ 𝑧𝐼 𝐾𝑧𝑏 4 + 𝜙𝑧 2 − 𝜙𝑧 𝜃𝑧𝐼̂
{ }= [ ] { }, (6.14b)
𝑚
̂ 𝑧𝐽 1 + 𝜙𝑧 2 − 𝜙𝑧 4 + 𝜙𝑧 𝜃𝑧𝐽 ̂
where Equation (6.14a) is for bending in the 𝐱̂ − 𝐳̂ plane and Equation (6.14b) is for bending
in the 𝐱̂ − 𝐲̂ plane. The bending constants are given by
𝐸𝐼𝑦𝑦
𝐾𝑦𝑏 = (6.15a)
𝑙0
𝐸𝐼𝑧𝑧
𝐾𝑧𝑏 = (6.15b)
𝑙0

𝐼𝑦𝑦 = ∫ ∫ 𝑧̂2 𝑑𝑦̂𝑑𝑧̂ (6.15c)

𝐼𝑧𝑧 = ∫ ∫ 𝑦̂2 𝑑𝑦̂𝑑𝑧̂ (6.15d)

12𝐸𝐼𝑦𝑦
𝜙𝑦 = (6.15e)
𝐺𝐴𝑠 𝑙2

12𝐸𝐼𝑧𝑧
𝜙𝑧 = . (6.15f)
𝐺𝐴𝑠 𝑙2
Hence 𝜙 is the shear factor, 𝐺 the shear modulus, and 𝐴𝑠 is the effective area in shear.

LS-DYNA DEV 10/27/16 (r:8004) 5-7 (Belytschko Beam)


Belytschko Beam LS-DYNA Theory Manual

Torsional Moment. The torsional moment is calculated from the torsional deformation
rotation as
𝑚 ̂ ,
̂ 𝑥𝐽 = 𝐾 𝑡 𝜃𝑥𝐽𝐼 (6.16)
where
𝐺𝐽
𝐾𝑡 = , (6.17)
𝑙0
and,
𝐽 = ∫ ∫ 𝑦̂𝑧𝑑̂ 𝑦̂𝑑𝑧̂. (6.18)

The above forces are conjugate to the deformation displacements given previously in
Equation (6.5), i.e.,
𝐝̂ T = {𝛿𝐼𝐽 , 𝜃𝑥𝐽𝐼
̂ , 𝜃𝑦𝐼
̂ , 𝜃𝑦𝐽
̂ , 𝜃𝑧𝐼
̂ , 𝜃𝑧𝐽
̂ }, (6.19)
where
𝐝̂ T 𝐟 ̂ = 𝑊 int . (6.20)
And with
𝐟 ̂T = {𝑓𝑥𝐽̂ , 𝑚
̂ 𝑥𝐽 , 𝑚
̂ 𝑦𝐼 , 𝑚
̂ 𝑦𝐽 , 𝑚
̂ 𝑧𝐼 , 𝑚
̂ 𝑧𝐽 }. (6.21)
The remaining internal force components are found from equilibrium:

𝑓𝑥𝐼̂ = −𝑓𝑥𝐽̂ 𝑚
̂ 𝑥𝐼 = −𝑚
̂ 𝑥𝐽
𝑚
̂ 𝑦𝐼 + 𝑚
̂ 𝑦𝐽
𝑓𝑧𝐼̂ = − 𝑓𝑧𝐼̂ = −𝑓𝑧𝐽̂ (6.22)
𝑙0
̂ 𝑧𝐼 + 𝑚
𝑚 ̂ 𝑧𝐽
𝑓𝑦𝐽̂ = − 𝑓𝑦𝐼̂ = −𝑓𝑦𝐽̂
𝑙0

6.2.3 Updating the Body Coordinate Unit Vectors


The body coordinate unit vectors are updated using the Newmark 𝛽-Method
[Newmark 1959] with 𝛽 = 0, which is almost identical to the central difference method
[Belytschko 1974]. In particular, the body component unit vectors are updated using the
formula
𝑗 𝑗
 𝑗+1  𝑗 d𝐛𝑖 Δ𝑡2 d2 𝐛𝑖
𝐛𝑖 = 𝐛𝑖 + Δ𝑡 + , (6.23)
d𝑡 2 d𝑡2
where the superscripts refer to the time step and the subscripts refer to the three unit
vectors comprising the body coordinate triad. The time derivatives in the above equation
are replaced by their equivalent forms from vector analysis:
 𝑗
d𝐛𝑖 (6.24)
= 𝛚 × 𝐛𝑖
d𝑡

5-8 (Belytschko Beam) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Belytschko Beam

 𝑗
d2 𝐛𝑖
= 𝛚 × (𝛚 × 𝐛𝑖 ) + (𝛂𝑖 × 𝐛𝑖 ),
d𝑡2
where 𝜔 and 𝛼 are vectors of angular velocity and acceleration, respectively, obtained from
the rotational equations of motion. With the above relations substituted into Equation
(6.23), the update formula for the unit vectors becomes

 𝑗+1  𝑗 Δ𝑡2
𝐛𝑖 = 𝐛𝑖 + Δ𝑡(𝛚 × 𝐛𝑖 ) + {[𝛚 × (𝛚 × 𝐛𝑖 ) + (𝛂𝑖 × 𝐛𝑖 )]}. (6.25)
2

To obtain the formulation for the updated components of the unit vectors, the body
coordinate system is temporarily considered to be fixed and then the dot product of
Equation (6.25) is formed with the unit vector to be updated. For example, to update the 𝑥̅
component of 𝐛3 , the dot product of Equation (6.25), with 𝑖 = 3, is formed with b1 , which
can be simplified to the relation

 𝑗+1  𝑗  𝑗+1  𝑗 Δ𝑡2  𝑗  𝑗  𝑗


𝑏̅𝑥3 = 𝐛1 ⋅ 𝐛3 = Δ𝑡𝜔𝑦 + (𝜔𝑥 𝜔𝑧 + 𝛼𝑦 ), (6.26)
2
Similarly,

 𝑗+1  𝑗  𝑗+1  𝑗 Δ𝑡2  𝑗  𝑗  𝑗


𝑏̅𝑦3 = 𝐛2 ⋅ 𝐛3 = Δ𝑡𝜔𝑥 + (𝜔𝑦 𝜔𝑧 + 𝛼𝑥 )
2
(6.27)
 𝑗+1  𝑗  𝑗+1 𝑗 Δ𝑡2  𝑗  𝑗  𝑗
𝑏̅𝑧3 = 𝐛1 ⋅ 𝐛2 = Δ𝑡𝜔𝑧 + (𝜔𝑥 𝜔𝑦 + 𝛼𝑧 ).
2
 𝑗+1  𝑗+1
The remaining components 𝐛3 and 𝐛1 are found by using normality and orthogonality,
where it is assumed that the angular velocities w are small during a time step so that the
 𝑗+1
quadratic terms in the update relations can be ignored. Since 𝐛3 is a unit vector,
normality provides the relation

 𝑗+1  𝑗+1  𝑗+1 2 2


𝑏̅𝑧3 = √1 − (𝑏̅𝑥3 ) − (𝑏̅𝑦3 ) . (6.28)

 𝑗+1
Next, if it is assumed that 𝑏̅𝑥1 ≈ 1, orthogonality yields
𝑗+1 𝑗+1 𝑗+1
 𝑗+1
𝑏̅𝑥3 + 𝑏̅𝑦1 𝑏̅𝑦3
𝑏̅𝑧1 =−  𝑗+1
. (6.29)
𝑏̅𝑧3
𝑗+1
The component 𝑏̅𝑥1 is then found by enforcing normality:

 𝑗+1  𝑗+1 2  𝑗+1 2


𝑏̅𝑥1 = √1 − (𝑏̅𝑦1 ) − (𝑏̅𝑧1 ) . (6.30)

The updated components of 𝐛1 and 𝐛3 are defined relative to the body coordinates at time
step 𝑗. To complete the update and define the transformation matrix, Equation (6.1), at time
step 𝑗 + 1, the updated unit vectors 𝐛1 and 𝐛3 are transformed to the global coordinate
system, using Equation (6.1) with [𝛌] defined at step 𝑗, and their vector cross product is
used to form 𝐛2 .
LS-DYNA DEV 10/27/16 (r:8004) 5-9 (Belytschko Beam)
LS-DYNA Theory Manual Hughes-Liu Beam

7
Hughes-Liu Beam

The Hughes-Liu beam element formulation, based on the shell [Hughes and Liu
1981a, 1981b] discussed later, was the first beam element we implemented. It has several
desirable qualities:
• It is incrementally objective (rigid body rotations do not generate strains), allowing
for the treatment of finite strains that occur in many practical applications;
• It is simple, which usually translates into computational efficiency and robustness
• It is compatible with the brick elements, because the element is based on a
degenerated brick element formulation;
• It includes finite transverse shear strains. The added computations needed to retain
this strain component, compare to those for the assumption of no transverse shear
strain, are insignificant.

7.1 Geometry
The Hughes-Liu beam element is based on a degeneration of the isoparametric 8-
node solid element, an approach originated by Ahmad et al., [1970]. Recall the solid
element isoparametric mapping of the biunit cube
8
𝐱(𝜉 , 𝜂, 𝜁 ) = ∑ 𝑁𝑎 (𝜉 , 𝜂, 𝜁 )𝑥𝑎 , (7.1)
𝑎=1

with,
(1 + 𝜉𝑎 𝜉 )(1 + 𝜂𝑎 𝜂)(1 + 𝜁𝑎 𝜁 )
𝑁𝑎 (𝜉 , 𝜂, 𝜁 ) = , (7.2)
8
where 𝐱 is an arbitrary point in the element, (𝜉 , 𝜂, 𝜁 ) are the parametric coordinates, 𝐱𝑎 are
the global nodal coordinates of node 𝑎, and 𝑁𝑎 are the element shape functions evaluated at
node 𝑎, i.e., (𝜉𝑎 , 𝜂𝑎 , 𝜁𝑎 ) are (𝜉 , 𝜂, 𝜁 ) evaluated at node 𝑎.

LS-DYNA DEV 10/27/16 (r:8004) 6-1 (Hughes-Liu Beam)


Hughes-Liu Beam LS-DYNA Theory Manual

In the beam geometry, 𝜉 determines the location along the axis of the beam and the
coordinate pair (𝜂, 𝜁 ) defines a point on the cross section. To degenerate the 8-node brick
geometry into the 2-node beam geometry, the four nodes at 𝜉 = −1 and at 𝜉 = 1 are
combined into a single node with three translational and three rotational degrees of
freedom. Orthogonal, inextensible nodal fibers are defined at each node for treating the
rotational degrees of freedom. Figure 7.1 shows a schematic of the biunit cube and the
beam element. The mapping of the biunit cube into the beam element is separated into
three parts:
𝐱(𝜉 , 𝜂, 𝜁 ) = 𝐱̅(𝜉 ) + 𝐗(𝜉 , 𝜂, 𝜁 ),
(7.3)
= 𝐱̅(𝜉 ) + 𝐗𝜁 (𝜉 , 𝜁 ) + 𝐗𝜂 (𝜉 , 𝜂),
where 𝐱̅ denotes a position vector to a point on the reference axis of the beam, and 𝐗𝜁 and
𝐗𝜂 are position vectors at point 𝐱̅ on the axis that define the fiber directions through that
point. In particular,

ζ
Biunit Cube
ζ
X
ξ
η

Beam Element ξ

Nodal fibers

Top Surface z+ζ


+1
x+ζ

ζ̄ x̄ζ
x^ζ

0 x^
η
xζ-
-1 -
Bottom Surface zζ
Figure 7.1. Hughes-Liu beam element.

6-2 (Hughes-Liu Beam) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Hughes-Liu Beam

2
𝐱̅(𝜉 ) = ∑ 𝑁𝑎 (𝜉 )𝐱̅𝑎 ,
𝑎=1
2
𝐗𝜂 (𝜉 , 𝜂) = ∑ 𝑁𝑎 (𝜉 )𝐗𝜂𝑎 (𝜂) , (7.4)
𝑎=1
2
𝐗𝜁 (𝜉 , 𝜁 ) = ∑ 𝑁𝑎 (𝜉 )𝐗𝜁𝑎 (𝜁 ).
𝑎=1

With this description, arbitrary points on the reference line 𝐱̅ are interpolated by the one-
dimensional shape function 𝑁(𝜉 ) operating on the global position of the two beam nodes
that define the reference axis, i.e., 𝐱̅a . Points off the reference axis are further interpolated
by using a one-dimensional shape function along the fiber directions, i.e., 𝐗𝜂𝑎 (𝜂) and𝐗𝜁𝑎 (𝜁 )
where
̂𝜂𝑎
𝐗𝜂𝑎 (𝜂) = 𝑧𝜂𝑎 (𝜂)𝐗 ̂𝜁𝑎
𝐗𝜁𝑎 (𝜁 ) = 𝑧𝜁𝑎 (𝜁 )𝐗
𝑧𝜂𝑎 (𝜂) = 𝑁+ (𝜂)𝑧+ −
𝜂𝑎 + 𝑁− (𝜂)𝑧𝜂𝑎 𝑧𝜁𝑎 (𝜁 ) = 𝑁+ (𝜁 )𝑧+ −
𝜁𝑎 + 𝑁− (𝜁 )𝑧𝜁𝑎

(1 + 𝜂) (1 + 𝜁 ) (7.5)
𝑁+ (𝜂) = 𝑁+ (𝜁 ) =
2 2
(1 − 𝜂) (1 − 𝜁 )
𝑁− (𝜂) = 𝑁− (𝜁 ) =
2 2
where 𝑧𝜁 (𝜁 ) and 𝑧𝜂 (𝜂) are “thickness functions”.

The Hughes-Liu beam formulation uses four position vectors, in addition to 𝜉 , to


locate the reference axis and define the initial fiber directions. Consider the two position
+ −
vectors 𝐱𝜁𝑎 and 𝐱𝜁𝑎 located on the top and bottom surfaces, respectively, at node 𝑎. Then
1 − + 1
̅ 𝜁𝑎
𝐱̅𝜁𝑎 = (1 − 𝜁 )𝐱 ̅ 𝜁𝑎
+ (1 + 𝜁 )𝐱 , ̅ 𝜂𝑎
𝐱̅𝜂𝑎 = (1 − 𝜁 )𝐱 − ̅ 𝜂𝑎
+ (1 + 𝜁 )𝐱 +
,
2 2
+ − + −
(𝐱𝜁𝑎 − 𝐱𝜁𝑎 ) (𝐱𝜂𝑎 − 𝐱𝜂𝑎 )
̂𝜁𝑎 =
𝐗 , ̂𝜂𝑎 =
𝐗 ,
+ − + −∥
∥𝐱𝜁𝑎 − 𝐱𝜁𝑎 ∥ ∥𝐱𝜂𝑎 − 𝐱𝜂𝑎
(7.6)
1 1
𝑧+ ̅ + −
𝜁𝑎 = (1 − 𝜁 )∥𝐱𝜁𝑎 − 𝐱𝜁𝑎 ∥, 𝑧+ + −
𝜂𝑎 = (1 − 𝜂̅)∥𝐱𝜂𝑎 − 𝐱𝜂𝑎 ∥,
2 2
1 1
𝑧− ̅ + −
𝜁𝑎 = − (1 + 𝜁 )∥𝐱𝜁𝑎 − 𝐱𝜁𝑎 ∥, 𝑧− + −
𝜂𝑎 = − (1 + 𝜂̅)∥𝐱𝜂𝑎 − 𝐱𝜂𝑎 ∥,
2 2
where ‖ ⋅ ‖ is the Euclidean norm. The reference surface may be located at the midsurface
of the beam or offset at the outer surfaces. This capability is useful in several practical
situations involving contact surfaces, connection of beam elements to solid elements, and
offsetting elements such as for beam stiffeners in stiffened shells. The reference surfaces
are located within the beam element by specifying the value of the parameters 𝜂̅ and 𝜁 ,̅ (see
lower portion of Figure 7.1). When these parameters take on the values –1 or +1, the

LS-DYNA DEV 10/27/16 (r:8004) 6-3 (Hughes-Liu Beam)


Hughes-Liu Beam LS-DYNA Theory Manual

reference axis is located on the outer surfaces of the beam. If they are set to zero, the
reference axis is at the center.

The same parametric representation used to describe the geometry of the beam
elements is used to interpolate the beam element displacements, i.e., an isoparametric
representation. Again the displacements are separated into the reference axis displace-
ments and rotations associated with the fiber directions:
𝐮(𝜉 , 𝜂, 𝜁 ) = 𝐮
̅̅̅̅(𝜉 ) + 𝐔(𝜉 , 𝜂, 𝜁 ),
(7.7)
=𝐮
̅̅̅̅(𝜉 ) + 𝐔𝜁 (𝜉 , 𝜁 ) + 𝐔𝜂 (𝜉 , 𝜂).
The reference axis is interpolated as usual
2
̅̅̅̅(𝜉 ) = ∑ 𝑁𝑎 (𝜉 )𝐮
𝐮 ̅̅̅̅𝑎 . (7.8)
𝑎=1

The displacements are also interpolated along the reference axis


2
𝐔𝜂 (𝜉 , 𝜂) = ∑ 𝑁𝑎 (𝜉 )𝐔𝜂𝑎 (𝜂),
𝑎=1
(7.9)
2
𝐔𝜁 (𝜉 , 𝜁 ) = ∑ 𝑁𝑎 (𝜉 )𝐔𝜁𝑎 (𝜁 ).
𝑎=1

The fiber displacement is interpolated consistently with the thickness,


̂𝜂𝑎 ,
𝐔𝜂𝑎 (𝜂) = 𝑧𝜂𝑎 (𝜂)𝐔
(7.10)
̂𝜁𝑎 ,
𝐔𝜁𝑎 (𝜁 ) = 𝑧𝜁𝑎 (𝜁 )𝐔
where 𝐮 is the displacement of a generic point, 𝐮 ̅̅̅̅ is the displacement of a point on the
reference surface, and 𝑈 is the “fiber displacement” rotations. The motion of the fibers can
be interpreted as either displacements or rotations as will be discussed.

Hughes and Liu introduced the notation that follows, and the associated schematic
shown in Figure 7.2, to describe the current deformed configuration with respect to the
reference configuration.
𝐲
̅̅̅̅ = 𝐱̅ + 𝐮
̅̅̅̅, 𝐘 = 𝐗 + 𝐔, ̂𝜂𝑎 = 𝐗
𝐘 ̂𝜂𝑎 + 𝐔
̂𝜂𝑎 ,
̅̅̅̅ + 𝐘,
𝐲=𝐲 (7.11)
𝐲
̅̅̅̅𝑎 = 𝐱̅𝑎 + 𝐮
̅̅̅̅𝑎 , 𝐘𝑎 = 𝐗𝑎 + 𝐔𝑎 , ̂𝜁𝑎 = 𝐗
𝐘 ̂𝜁𝑎 + 𝐔
̂𝜁𝑎 ,

In the above relations, and in Figure 7.2, the 𝐱 quantities refer to the reference configura-
tion, the 𝐲 quantities refer to the updated (deformed) configuration and the 𝐮 quantities are
the displacements. The notation consistently uses a superscript bar (⋅ ̅) to indicate reference
surface quantities, a superscript caret (⋅ ̂) to indicate unit vector quantities, lower case letter
for translational displacements, and upper case letters for fiber displacements. Thus to
update to the deformed configuration, two vector quantities are needed: the reference
surface displacement 𝐮 ̅̅̅̅ and the associated nodal fiber displacement 𝐔. The nodal fiber
displacements are defined in the fiber coordinate system, described in the next subsection.

6-4 (Hughes-Liu Beam) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Hughes-Liu Beam

u U

X (parallel construction)

x
Y

reference axis in
undeformed
x̄ geometry
Deformed Configuration
Reference Surface
Figure 7.2. Schematic of deformed configuration displacements and position
vectors.

7.2 Fiber Coordinate System


For a beam element, the known quantities will be the displacements of the reference
surface 𝑢̅ obtained from the translational equations of motion and the rotational quantities
at each node obtained from the rotational equations of motion. What remains to complete
the kinematics is a relation between nodal rotations and fiber displacements 𝐔. The
linearized relationships between the incremental components Δ𝐔 ̂ the incremental rotations
are given by
̂𝜂1 ⎫
⎧Δ𝑈
{ } 0 𝑌̂𝜂3 −𝑌̂𝜂2 ⎧Δ𝜃 ⎫
{ } ⎡ ⎤ { 1}
̂ ⎢−𝑌̂ 𝑌̂𝜂1 ⎥ Δ𝜃2 ⎬ = 𝐡𝜂 Δ𝛉,
⎨Δ𝑈𝜂2 ⎬ = ⎢
⎢ 𝜂3
0 ⎥
⎥⎨{ }
{ }
{Δ𝑈 }
⎩ 𝜂3 ⎭ ⎣ 𝑌̂𝜂2
̂ −𝑌̂𝜂1 0 ⎦ ⎩Δ𝜃3 ⎭
(7.12)
̂𝜁1 ⎫
⎧Δ𝑈
{ } 0 𝑌̂𝜁3 −𝑌̂𝜁2 ⎧Δ𝜃 ⎫
{ } ⎡ ⎤ { 1}
̂ ⎢−𝑌̂𝜁3 𝑌̂𝜁1 ⎥ Δ𝜃2 ⎬ = 𝐡𝜁 Δ𝛉.
⎨Δ𝑈𝜁2 ⎬ = ⎢

0 ⎥
⎥⎨{ }
{
{Δ𝑈 }
}
⎩ 𝜁3 ⎭ ⎣ 𝑌̂𝜁2
̂ −𝑌̂𝜁1 0 ⎦ Δ𝜃3 ⎭

Equations (7.12) are used to transform the incremental fiber tip displacements to
rotational increments in the equations of motion. The second-order accurate rotational
update formulation due to Hughes and Winget [1980] is used to update the fiber vectors:
𝑌̂𝜂𝑖𝑛+1 = 𝑅𝑖𝑗 (Δ𝜃)𝑌̂𝜂𝑖𝑛 ,
(7.13)
𝑌̂𝜁𝑖𝑛+1 = 𝑅𝑖𝑗 (Δ𝜃)𝑌̂𝜁𝑖𝑛 ,
then

LS-DYNA DEV 10/27/16 (r:8004) 6-5 (Hughes-Liu Beam)


Hughes-Liu Beam LS-DYNA Theory Manual

𝑛+1
̂𝜂𝑎
̂𝜂𝑎 = 𝐘 𝑛
̂𝜂𝑎
Δ𝐔 −𝐘 ,
(7.14)
𝑛+1
̂𝜁𝑎
̂𝜁𝑎 = 𝐘 𝑛
̂𝜁𝑎
Δ𝐔 −𝐘 ,
where
(2𝛿𝑖𝑗 + Δ𝑆𝑖𝑘 )Δ𝑆𝑖𝑘
𝑅𝑖𝑗 (Δ𝜃) = 𝛿𝑖𝑗 + ,
2𝐷
Δ𝑆𝑖𝑗 = 𝑒𝑖𝑘𝑗 Δ𝜃𝑘 , (7.15)
1
2𝐷 = 2 + (Δ𝜃12 + Δ𝜃22 + Δ𝜃32 ).
2
Here 𝛿𝑖𝑗 is the Kronecker delta and 𝑒𝑖𝑘𝑗 is the permutation tensor.

7.2.1 Local Coordinate System


In addition to the above described fiber coordinate system, a local coordinate system
is needed to enforce the zero normal stress conditions transverse to the axis. The
orthonormal basis with two directions 𝐞̂2 and 𝐞̂3 normal to the axis of the beam is
constructed as follows:
̅̅̅̅2 − 𝐲
𝐲 ̅̅̅̅1
𝐞̂1 = ,
∥𝐲
̅̅̅̅2 − 𝐲
̅̅̅̅1 ∥
̂𝜂1 + 𝐘
̂𝜂2 (7.16)
𝐘
𝐞′2 = .
̂𝜂1 + 𝐘
∥𝐘 ̂𝜂2 ∥

From the vector cross product of these local tangents.


𝐞̂3 = 𝐞̂1 × 𝐞′2 , (7.17)
and to complete this orthonormal basis, the vector
𝐞̂2 = 𝐞̂3 × 𝐞̂1 , (7.18)
is defined. This coordinate system rigidly rotates with the deformations of the element.

The transformation of vectors from the global to the local coordinate system can
now be defined in terms of the basis vectors as
⎧𝐴𝑥̂ ⎫ 𝑒3𝑥 T ⎧𝐴𝑥 ⎫
{
{ } } ⎡𝑒1𝑥 𝑒2𝑥
{ }
̂ ̂
𝐀 = ⎨𝐴𝑦 ⎬ = ⎢𝑒1𝑦 𝑒2𝑦 𝑒3𝑦 ⎤
⎥ ⎨𝐴𝑦 ⎬ = [𝐪]{𝐀}, (7.19)
{
{𝐴 ̂ }} ⎣𝑒1𝑧 𝑒2𝑧 𝑒3𝑧 ⎦ {
⎩𝐴𝑧 }

⎩ 𝑧⎭
̂ is a
where 𝑒𝑖𝑥 , 𝑒𝑖𝑦 , 𝑒𝑖𝑧 are the global components of the local coordinate unit vectors, 𝐀
vector in the local coordinates, and 𝐀 is the same vector in the global coordinate system.

6-6 (Hughes-Liu Beam) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Hughes-Liu Beam

7.3 Strains and Stress Update


7.3.1 Incremental Strain and Spin Tensors
The strain and spin increments are calculated from the incremental displacement
gradient
∂Δ𝑢𝑖
𝐺𝑖𝑗 = , (7.20)
∂𝑦𝑗
where Δ𝑢𝑖 are the incremental displacements and 𝑦𝑗 are the deformed coordinates. The
incremental strain and spin tensors are defined as the symmetric and skew-symmetric
parts, respectively, of 𝐺𝑖𝑗 :
1
Δ𝜀𝑖𝑗 = (𝐺𝑖𝑗 + 𝐺𝑗𝑖 ),
2
(7.21)
1
Δ𝜔𝑖𝑗 = (𝐺𝑖𝑗 − 𝐺𝑗𝑖 ).
2

The incremental spin tensor Δ𝜔𝑖𝑗 is used as an approximation to the rotational


contribution of the Jaumann rate of the stress tensor; in an implicit implementation
[Hallquist 1981b] the more accurate Hughes-Winget [1980] transformation matrix is used,
Equation (7.15), with the incremental spin tensor for the rotational update. Here the
Jaumann rate update is approximated as
𝜎 𝑖𝑗 = 𝜎𝑖𝑗𝑛 + 𝜎𝑖𝑝
𝑛 𝑛
Δ𝜔𝑝𝑗 + 𝜎𝑗𝑝 Δ𝜔𝑝𝑖 , (7.22)
where the superscripts on the stress tensor refer to the updated (𝑛 + 1) and reference (𝑛)
configurations. This update of the stress tensor is applied before the constitutive
evaluation, and the stress and strain are stored in the global coordinate system.

7.3.2 Stress Update


To evaluate the constitutive relation, the stresses and strain increments are rotated
from the global to the local coordinate system using the transformation defined previously
in Equation (7.19), viz.
𝑛
𝜎𝑖𝑗𝑙 = 𝑞𝑖𝑘 𝜎 𝑘𝑛 𝑞𝑗𝑛 ,
(7.23)
Δ𝜀𝑖𝑗𝑙 = 𝑞𝑖𝑘 Δ𝜀𝑘𝑛 𝑞𝑗𝑛 ,
where the superscript 𝑙 indicates components in the local coordinate system. The stress is
updated incrementally:
𝑛+1 𝑛 𝑛+1
𝜎𝑖𝑗𝑙 = 𝜎𝑖𝑗𝑙 + Δ𝜎𝑖𝑗𝑙
2
, (7.24)

and rotated back to the global system:

LS-DYNA DEV 10/27/16 (r:8004) 6-7 (Hughes-Liu Beam)


Hughes-Liu Beam LS-DYNA Theory Manual

𝑛+1
𝑙
𝜎𝑖𝑗𝑛+1 = 𝑞𝑘𝑖 𝜎𝑘𝑛 𝑞𝑛𝑗 , (7.25)
before computing the internal force vector.

7.3.3 Incremental Strain-Displacement Relations


After the constitutive evaluation is completed, the fully updated stresses are rotated
back to the global coordinate system. These global stresses are then used to update the
internal force vector

𝐟𝑎int = ∫ 𝐁T𝑎 𝛔𝑑𝜐, (7.26)

where 𝐟𝑎int are the internal forces at node 𝑎 and 𝐁𝑎 is the strain-displacement matrix in the
global coordinate system associated with the displacements at node 𝑎. The 𝐁 matrix relates
six global strain components to eighteen incremental displacements [three translational
displacements per node and the six incremental fiber tip displacements of Equation (7.14)].
It is convenient to partition the 𝐁 matrix:
𝐁 = [𝐁1 , 𝐁2 ]. (7.27)
Each 𝐵𝑎 sub matrix is further partitioned into a portion due to strain and spin with the
following sub matrix definitions:
𝐵 0 0 𝐵4 0 0 𝐵7 0 0
⎡0 1 𝐵2 0 0 𝐵5 0 0 𝐵8 0 ⎤
⎢ ⎥
⎢0 0 𝐵3 0 0 𝐵6 0 0 𝐵9 ⎥
𝐁𝑎 = ⎢
⎢𝐵2
⎥, (7.28)
𝐵1 0 𝐵5 𝐵4 0 𝐵8 𝐵7 0 ⎥
⎢ ⎥
⎢0 𝐵3 𝐵2 0 𝐵6 𝐵5 0 𝐵9 𝐵8 ⎥
⎣𝐵3 0 𝐵1 𝐵6 0 𝐵4 𝐵9 0 𝐵7 ⎦
where,
⎧ 𝜕𝑁𝑎
{ 𝑁 𝑎,𝑖 = 𝑖 = 1,2,3
{ 𝜕𝑦𝑖
{
{
{ 𝜕(𝑁𝑎 𝑧𝜂𝑎 )
𝐵𝑖 = ⎨(𝑁𝑎 𝑧𝜂𝑎 ) = 𝑖 = 4,5,6. (7.29)
{ ,𝑖−3 𝜕𝑦𝑖−3
{
{ 𝜕(𝑁𝑎 𝑧𝜁𝑎 )
{ (𝑁 𝑧 )
{ 𝑎 𝜁𝑎 ,𝑖−6 = 𝑖 = 7,8,9
⎩ 𝜕𝑦𝑖−6
With respect to the strain-displacement relations, note that:
• The derivative of the shape functions are taken with respect to the global coordi-
nates;
• The 𝐁 matrix is computed on the cross-section located at the mid-point of the axis;
• The resulting 𝐁 matrix is a 6  18 matrix.
The internal force, 𝑓 , given by
𝐟′ = 𝐓T 𝐟𝑎int (7.30)
6-8 (Hughes-Liu Beam) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Hughes-Liu Beam

4 3

1 2
4 7 3

8 9 6

1 5 2
4 7 3

8 9 6

1 5 2 4 10 9 3

11 16 15 8

12 13 14 7

1 5 6 2

Figure 7.3. Integration possibilities for rectangular cross sections in the Hughes-
Liu beam element.

is assembled into the global right hand side internal force vector. 𝐓 is defined as (also see
Equation (7.12):
𝐈 𝟎
⎡𝟎 𝐡𝜂 ⎤
𝐓=⎢ ⎥, (7.31)
⎣𝟎 𝐡𝜁 ⎦
where 𝐈 the 3  3 identity matrix.

7.3.4 Spatial Integration


The integration of Equation (7.26) for the beam element is performed with one- point
integration along the axis and multiple points in the cross section. For rectangular cross
sections, a variety of choices are available as is shown in Figure 7.3. The beam has no zero
energy or locking modes.

LS-DYNA DEV 10/27/16 (r:8004) 6-9 (Hughes-Liu Beam)


Hughes-Liu Beam LS-DYNA Theory Manual

tt

A
st

Figure 7.4. Specification of the nodal thickness, 𝑠𝑡 and 𝑡𝑡 , for a beam with an
arbitrary cross-section.

For the user defined rule, it is necessary to specify the number of integration points
and the relative area for the total cross section:
𝐴
𝐴𝑟 =
𝐬𝑡 ⋅ 𝐭 𝑡
where 𝑠𝑡 and 𝑡𝑡 are the beam thickness specified on either the cross section or beam element
cards. The rectangular cross-section which contains 𝑠𝑡 and 𝑡𝑡 should completely contain the
cross-sectional geometry. Figure 7.4 illustrates this for a typical cross-section. In Figure
5.5, the area is broken into twelve integration points. For each integration point, it is
necessary to define the 𝑠 and 𝑡 parametric coordinates, (𝑠𝑖 ,𝑡𝑖 ), of the centroid of the ith
integration point and the relative area associated with the point
𝐴𝑖
𝐴𝑟𝑖 =
𝐴

6-10 (Hughes-Liu Beam) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Hughes-Liu Beam

A1 A2 A3 A4 A5

A6

s6 t6

A7 s

A8

A12 A11 A10 A9

Figure 7.5. A breakdown of the cross section geometry in Figure 7.4 into twelve
integration points.

where 𝐴𝑖 is the ratio of the area of the integration point and the actual area of the cross-
section, 𝐴.

LS-DYNA DEV 10/27/16 (r:8004) 6-11 (Hughes-Liu Beam)


LS-DYNA Theory Manual Warped Beam Elements

8
Warped Beam Elements

8.1 Resultant Warped Beam


8.1.1 Green-Lagrange Strains in Terms of Deformational Displacements
All quantities in this section are referred to the local element coordinate system 𝐞𝑖 ,
𝑖 = 1, 2, 3. The origin of the local system is taken at node 1, with 𝐞1 directed along the line
of centroids, while 𝐞2 , and 𝐞3 are directed along the principal axes of the cross-section.

With respect to the local system, the Green-Lagrange strain tensor can be written as:
𝜀𝑖𝑗 = 𝑒𝑖𝑗 + 𝜂𝑖𝑗 , (8.1)
where,
𝑒𝑖𝑗 = 0.5(𝑢𝑖,𝑗 + 𝑢𝑗,𝑖 ),
(8.2)
𝜂𝑖𝑗 = 0.5𝑢𝑘,𝑖 𝑢𝑘,𝑗 .

The geometric assumption of infinite in-plane rigidity implies 𝜀22 = 𝜀33 = 𝛾23 = 0.
Then the non-zero strain components which contribute to the strain energy are:
1
𝜀11 = 𝑢1,1 + (𝑢21,1 + 𝑢22,1 + 𝑢23,1 ),
2
2𝜀12 = 𝑢1,2 + 𝑢2,1 + 𝑢1,1, 𝑢1,2 + 𝑢2,1 𝑢2,2 + 𝑢3,1 𝑢3,2 , (8.3)

2𝜀13 = 𝑢1,3 + 𝑢3,1 + 𝑢1,1 𝑢1,3 + 𝑢2,1 𝑢2,3 + 𝑢3,1 𝑢3,3 .

8.1.2 Deformational Displacements After Large Rotations


The position vectors of an arbitrary point P in the initial and current local
configurations are:

LS-DYNA DEV 10/27/16 (r:8004) 7-1 (Warped Beam Elements)


Warped Beam Elements LS-DYNA Theory Manual

0
0
𝐱𝑃0 = 𝐱𝐶 + [𝐞 1 𝐞2 𝐞3 ] ⎡ ⎤
⎢𝑥2 ⎥, (8.4)
⎣𝑥3 ⎦

𝜛𝜙
𝐱𝑃 = 𝐱𝐶 + [𝐞′1 𝐞′2 𝐞′3 ] ⎢𝑥2 ⎤

⎥, (8.5)
⎣𝑥 3 ⎦
respectively, with
1
[𝐞′1 𝐞′2 𝐞′3 ] = [𝐈 + 𝛉 + 𝛉2 ] [𝐞1 𝐞2 𝐞3 ], (8.6)
2
where
0 −𝜃3 𝜃2
⎡𝜃 0 −𝜃1 ⎤
𝛉=⎢ 3 ⎥, (8.7)
⎣−𝜃2 𝜃1 0 ⎦
and 𝜛 is the Saint-Venant warping function about the centroid C. By the transfer theorem,
the following relation holds:
𝜛 = 𝜔 + 𝑐2 𝑥3 − 𝑐3 𝑥2 , (8.8)
where 𝜔 refers to the shear center S, and 𝑐2 and 𝑐3 are the coordinates of S.

Subtracting Equation (8.4) from Equation (8.5) and neglecting third-order terms, the
displacements vector of point P can be computed:
1 1
𝑢1 = 𝑢̅1 − 𝑥2 𝜃3 + 𝑥3 𝜃2 + 𝑥2 𝜃1 𝜃2 + 𝑥3 𝜃1 𝜃3 + 𝜛𝜙,
2 2
1 1
𝑢2 = 𝑢̅2 − 𝑥3 𝜃1 − 𝑥2 (𝜃12 + 𝜃32 ) + 𝑥3 𝜃2 θ3 + 𝜛θ3 𝜙, (8.9)
2 2
1 1
u3 = u ̅̅̅̅3 + x2 θ1 − x3 (θ21 + θ22 ) + x2 θ2 θ3 − 𝜛θ2 𝜙,
2 2
where 𝑢̅1 , 𝑢̅2 , and 𝑢̅3 are the displacements of the centroid C.

8.1.3 Green-Lagrange Strains in terms of Centroidal Displacements and Angular


Rotations
From Equations (8.3) and (8.9), a second-order approximation of the Green-Lagrange
strains can be evaluated. Neglecting term 12 𝑢21,1 and the nonlinear strain components
generated by warping, the strain components are simplified as
1
𝜀11 = 𝜀0 + 𝑥2 𝜅2 + 𝑥3 𝜅3 + (𝑥22 + 𝑥32 )𝜃1,1
2
+ 𝜔𝜙,1 ,
2
2𝜀12 = 𝛾12 + 𝜛,2 𝜙 − 𝑥3 𝜅1 , (8.10)

2𝜀13 = 𝛾13 + 𝜛,3 𝜙 + 𝑥2 𝜅1 ,


with

7-2 (Warped Beam Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Warped Beam Elements

1
𝜀0 = 𝑢̅1,1 + (𝑢̅22,1 + 𝑢̅23,1 ),
2
1
𝜅1 = 𝜃1,1 + (𝜃2,1 𝜃3 − 𝜃3,1 𝜃2 ),
2
1
𝜅2 = −𝜃3,1 + (𝜃1 𝜃2,1 + 𝜃1,1 𝜃2 ) + 𝑢̅3,1 𝜃1,1 − 𝑐3 𝜙,1 ,
2
(8.11)
1
𝜅3 = 𝜃2,1 + (𝜃1 𝜃3,1 + 𝜃1,1 𝜃3 ) − 𝑢̅2,1 𝜃1,1 + 𝑐2 𝜙,1 ,
2
1
𝛾12 = 𝑢̅2,1 − 𝜃3 + 𝜃1 𝜃2 + 𝑢̅3,1 𝜃1 − 𝑢̅1,1 𝜃3 ,
2
1
𝛾13 = 𝑢̅3,1 + 𝜃2 + 𝜃1 𝜃3 − 𝑢̅2,1 𝜃1 + 𝑢̅1,1 𝜃2 .
2

Numerical testing has shown that neglecting the nonlinear terms in the curvatures 𝜅1 , 𝜅2 ,𝜅3
and bending shear strains 𝛾12 , 𝛾13 has little effect on the accuracy of the results. Therefore,
Equation (8.11) can be simplified to
𝜅1 = 𝜃1,1 , 𝜅2 = −𝜃3,1 − 𝑐3 𝜙,1 , 𝜅3 = 𝜃2,1 + 𝑐2 𝜙,1 ,
1 (8.12)
𝜀0 = 𝑢̅1,1 + (𝑢̅22,1 + 𝑢̅23,1 ), 𝛾12 = 𝑢̅2,1 − 𝜃3 , 𝛾13 = 𝑢̅3,1 + 𝜃2 .
2
Adopting Bernoulli’s assumption (𝛾12 = 𝛾13 = 0) and Vlasov’s assumption (𝜙 = θ1,1 ),
Equation (8.10) can be rewritten as:
1
𝜀11 = 𝜀0 + 𝑥2 𝜅2 + 𝑥3 𝜅3 + 𝑟2 𝜅12 + 𝜔𝜃1,11 ,
2
2𝜀12 = (𝜛,2 − 𝑥3 )𝜅1 , (8.13)

2𝜀13 = (𝜛,3 − 𝑥2 )𝜅1 ,


where
𝑟2 = 𝑥22 + 𝑥32 , 𝜅1 = 𝜃1,1 , 𝜅2 = −𝑢̅2,11 − 𝑐3 𝜃1,11 , 𝜅3 = −𝑢̅3,11 + 𝑐2 𝜃1,11 . (8.14)

To avoid membrane locking, ε11 in Equation (8.13) is reformulated as


1 𝐼
𝜀11 = 𝜀𝑎 + 𝑥2 𝜅2 + 𝑥3 𝜅3 + (𝑟2 − 𝑜 ) 𝜅12 + 𝜔𝜃1,11 , (8.15)
2 𝐴
where
1 1 1 𝐼
𝜀𝑎 = ∫ [𝑢̅1,1 + (𝑢̅22,1 + 𝑢̅23,1 + 𝑜 𝜅12 )] 𝑑𝑥1 . (8.16)
𝑙 0 2 𝐴

8.1.4 Strain Energy


Assuming material is linear elastic, the strain energy can be evaluated from:

LS-DYNA DEV 10/27/16 (r:8004) 7-3 (Warped Beam Elements)


Warped Beam Elements LS-DYNA Theory Manual

𝐿 1 1
𝑈 = ∫ ( 𝐸 ∫ 𝜀211 𝑑𝐴 + 𝐺 ∫ [(2𝜀12 )2 + (2𝜀13 )2 ]𝑑𝐴) 𝑑𝑥1 . (8.17)
0 2 𝐴 2 𝐴

The following relations are used in integrating the previous equations:


(1) Since the reference frame is located at centroid C with e2 and e3 directed along the
principal axes,

∫ 𝑥2 𝑑𝐴 = 0 , ∫ 𝑥3 𝑑𝐴 = 0 , ∫ 𝑥2 𝑥3 𝑑𝐴 = 0. (8.18)
𝐴 𝐴 𝐴

(2) Since sectorial area ω refers to shear center S,

∫ 𝜔𝑑𝐴 = 0 , ∫ 𝑥2 𝜔𝑑𝐴 = 0 , ∫ 𝑥3 𝜔𝑑𝐴 = 0. (8.19)


𝐴 𝐴 𝐴

Integration through the cross-section gives:

∫ 𝜀211 𝑑𝐴 + 𝐴𝜀2𝑎 + 𝐼22 𝜅22 + 𝐼33 𝜅32 + 𝐼𝜔 𝜃1,11


2
+ 𝐼2𝑟 𝜅2 𝜅12 + 𝐼3𝑟 𝜅3 𝜅12 + 𝐼𝜔𝑟 𝜃1,11 𝜅12
𝐴

1 𝐼𝑜2 4 (8.20)
+ (𝐼𝑟𝑟 − ) 𝜅1
4 𝐴

∫ [(2ε12 )2 + (2ε13 )2 ] dA = 𝐽κ12 (8.21)


A

with

𝐼22 = ∫ 𝑥22 𝑑𝐴 , 𝐼33 = ∫ 𝑥32 𝑑𝐴 , 𝐼𝑂 = 𝐼22 + 𝐼33


𝐴 𝐴
𝐼2𝑟 = ∫ 𝑥2 𝑟2 𝑑𝐴, 𝐼3𝑟 = ∫ 𝑥3 𝑟2 𝑑𝐴, 𝐼𝑟𝑟 = ∫ 𝑟4 𝑑𝐴 (8.22)
𝐴 𝐴 𝐴
𝐼𝜔 = ∫ 𝜔2 𝑑𝐴 , 𝐼𝜔𝑟 = ∫ 𝜔𝑟2 𝑑𝐴 , ̅̅̅̅,3 + 𝑥2 )2 + (𝜔
𝐽 = ∫ [(𝜔 ̅̅̅̅,2 − 𝑥3 )2 ] 𝑑𝐴
𝐴 𝐴 𝐴

6.1.5 Displacement Field


Linear interpolation is used for axial displacement 𝐮 ̅̅̅̅, whereas Hermitian
interpolations are used for 𝑢2 , 𝑢3 , and 𝜃1 , considering the following relations used in
deriving the final expression of strain energy:

𝜃2 = −𝑢3,1 𝜃3 = 𝑢2,1 𝜙 = 𝜃1,1 . (8.23)

The nodal displacement field is constructed by


𝑢̅1 = 𝐍1 𝐝, (8.24)

7-4 (Warped Beam Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Warped Beam Elements

⎧𝑢̅
{ 2⎫ }
⎨𝑢̅3 ⎬ = 𝐍2 𝐝, (8.25)
{
⎩𝜃1 }⎭
where
𝐝T = [0 0 0 0 𝜃2𝐼 𝜃3𝐼 𝜙𝐼 𝑢̅1𝐽𝐼 0 0 𝜃1𝐽𝐼 𝜃2𝐽 𝜃3𝐽 𝜙𝐽 ]T , (8.26)

𝐍1 = [1 − 𝜉 ⋅⋅⋅⋅⋅⋅| 𝜉⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ],
⋅𝑓 ⋅ ⋅ ⋅ 𝑔 ⋅ ∣ ⋅ 1−𝑓 ⋅ ⋅ ⋅ ℎ ⋅ (8.27)
⎡ ⎤
𝐍2 = ⎢⋅⋅ 𝑓 ⋅ −𝑔 ⋅ ⋅∣∣ ⋅ ⋅ 1−𝑓 ⋅ −ℎ ⋅ ⋅ ⎥,

⎣⋅⋅ ⋅ 𝑓 ⋅ ⋅ 𝑔∣ ⋅ ⋅ ⋅ 1−𝑓 ⋅ ⋅ ℎ⎦
with
𝑓 = 1 − 3𝜉 2 + 2𝜉 3 𝑔 = 𝑙(𝜉 − 2𝜉 2 + 𝜉 3 )ℎ = 𝑙(𝜉 3 − 𝜉 2 ). (8.28)

Equations (8.25) and (8.27) also imply


𝜃1,1 = 𝐍3 𝐝, (8.29)
where
𝐍3 = [⋅⋅ 𝑓,1 ⋅⋅ 𝑔,1 ∣ ⋅⋅⋅ −𝑓,1 ⋅⋅ ℎ,1 ]. (8.30)

6.1.6 Strain Energy in Matrix Form


The strain energy due to the average strain εa defined in Equation (8.16) can be
expressed in matrix form as
2
1 1 1 T 1 T
𝑈1 = 𝐸𝐴𝑙 [∫ (𝐍1,1 𝐝)𝑑𝜉 + 𝐝 [∫ (𝐍2,1 𝐃𝐍2,1 )𝑑𝜉 ] 𝐝] , (8.31)
2 0 2 0

where
𝐼
𝐃 = diag (1,1, 𝑜 ). (8.32)
𝐴

The strain energy due to the second through fourth terms is


1 1
𝑈2 = 𝐸𝑙𝐝T [∫ 𝐍T2,11 𝐇𝐍2,11 𝑑𝜉 ] 𝐝, (8.33)
2 0

where
𝐼22 𝐼22 𝑐3
⎡ −𝐼33 𝑐2 ⎤
𝐇=⎢ 𝐼33 ⎥, 𝐼′𝜔 = 𝐼𝜔 + 𝐼22 𝑐32 + 𝐼33 𝑐22 . (8.34)
⎣𝐼22 𝑐3 −𝐼33 𝑐2 𝐼′𝜔 ⎦

The strain energy due to the fifth through seventh terms is

LS-DYNA DEV 10/27/16 (r:8004) 7-5 (Warped Beam Elements)


Warped Beam Elements LS-DYNA Theory Manual

1 1
𝑈3 = 𝐸𝑙 [∫ (𝐍3 𝐝)2 𝛖𝐍2,11 𝑑𝜉 ] 𝐝, (8.35)
2 0

where
𝛖 = (−𝐼2𝑟 − 𝐼3𝑟 𝐼′𝜔𝑟 ), 𝐼′𝜔𝑟 = 𝐼𝜔𝑟 − 𝑐3 𝐼2𝑟 + 𝑐2 𝐼3𝑟 . (8.36)

The strain energy due to the eighth and ninth terms is


1 𝐼2 1 1 1
𝑈4 = 𝐸𝑙 (𝐼𝑟𝑟 − 𝑜 ) ∫ (𝐍3 𝐝)4 𝑑𝜉 + 𝐺𝐽𝑙 ∫ (𝐍3 𝐝)2 𝑑𝜉 (8.37)
8 𝐴 0 2 0

6.1.7 Internal Nodal Force Vector


The internal force can be evaluated from
̅̅̅̅1𝐽𝐿 1 T
𝐮 1 1
𝐟e = 𝐸𝐴𝑙 ( + 𝐝 𝐐) (𝐏 + 𝐐) + 𝐸 (𝐑 + 𝐒 + 𝐓 + 𝐕) + G𝐖, (8.38)
l 2 2 2
where
1 1 1
𝐏 = ∫ 𝐍T1,1 𝑑𝜉 , 𝐐 = [∫ 𝐍T2,1 𝐃𝐍2,1 𝑑𝜉 ] 𝐝, 𝐑 = l [∫ 𝐍T2,11 𝐇𝐍2,11 𝑑𝜉 ] 𝐝,
0 0 0
1 1
𝐒 = l [∫ (𝐍3 𝐝)2 𝐍T2,11 𝑑𝜉 ] 𝛖T , 𝐓 = 𝑙 ∫ (𝐍3 𝐝)(𝛖𝐍2,11 𝐝) 𝐍T3 𝑑𝜉 , (8.39)
0 0
𝐼𝑜2 1 1 𝑢̅1𝐽𝐼 1 T
𝐕 = (𝐼𝑟𝑟 − ) l ∫ (𝐍3 𝐝)3 𝐍T3 𝑑𝜉 , 𝐖 = 𝐽𝑙 ∫ (𝐍3 𝐝)𝐍T3 𝑑𝜉 , 𝜆= + 𝐝 𝐐.
𝐴 0 0 l 2

With respect to the local coordinate system, there are totally eight independent
components in the nodal force vector, in correspondence to the eight nodal displacement
components.

Other forces can be calculated by:


𝐅6 + 𝐅13 𝐅 + 𝐅12
𝐅1 = −𝐅8 , 𝐅2 = , 𝐅3 = − 5 ,
l l (8.40)
𝐅4 = −𝐅11 , 𝐅9 = −𝐅2 , 𝐅10 = −𝐅3 .

6.2 Integrated Warped Beam

6.2.1 Kinematics
We introduce three coordinate systems that are mutually interrelated. The first
coordinate system is the orthogonal Cartesian coordinate system (𝑥, 𝑦, 𝑧), for which the y
and zaxes lie in the plane of the cross-section and the 𝑥-axis parallel to the longitudinal axis
of the beam. The second coordinate system is the local plate coordinate system (𝑥, 𝑠, 𝑛) as
shown in Figure 6.1, wherein the n-axis is normal to the middle surface of a plate element,
the s-axis is tangent to the middle surface and is directed along the contour line of the

7-6 (Warped Beam Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Warped Beam Elements

cross-section. The (𝑥, 𝑠, 𝑛) and (𝑥, 𝑦, 𝑧) coordinate systems are related through an angle of
orientation 𝜃 as defined in Figure 40.1. The third coordinate set is the contour coordinate s
along the profile of the section with its origin at some point O on the profile section. Point
P is called the pole through which the axis parallel to the x-axis is called the pole axis. To
derive the analytical model for a thin-walled beam, the following two assumptions are
made:
1. The contour of the thin wall does not deform in its own plane.
2. The shear strain γsx of the middle surface is zero.

According to assumption 1, the midsurface displacement components 𝑣 and 𝑤 with


respect to the (𝑥, 𝑠, 𝑛) coordinate system at a point A can be expressed in terms of
displacements 𝐕 and 𝐖 of the pole P in the (𝑥, 𝑦, 𝑧) coordinate system and the rotation
angle 𝜙𝑥 about the pole axis
𝐯(𝑥, 𝑠) = 𝐕(𝑥)cos𝜃(𝑠) + 𝐖(𝑥)sin𝜃(𝑠) − 𝐫(𝑠)ϕ𝑥 (𝑥),
(8.41)
𝐰(𝑥, 𝑠) = −𝐕(𝑥)sin𝜃(𝑠) + 𝐖(𝑥)cos𝜃(𝑠) − 𝐪(𝑠)ϕ𝑥 (𝑥).

These equations apply to the whole contour. The out-of-plane displacement u can
now be found from assumption 2. On the middle surface
∂𝐮 ∂𝐯
+ = 𝟎, (8.42)
∂𝑠 ∂𝑥
which can be written
∂𝐮 ∂𝐯
=− = −𝐕′ (𝑥)cos𝜃(𝑠) − 𝐖′ (𝑥)sin𝜃(𝑠) + 𝐫(𝑠)ϕ′𝑥 (𝑥). (8.43)
∂s ∂𝑥

Integrating this relation from point O to an arbitrary point on the contour yields
(using t as a dummy for s)
P

s
q(s)
n
r(s) θ(s)
A
z

y
Figure 8.1. Definition of coordinates in thin-walled open section

LS-DYNA DEV 10/27/16 (r:8004) 7-7 (Warped Beam Elements)


Warped Beam Elements LS-DYNA Theory Manual

s s s s
∂𝐮
∫ 𝑑𝑡 = −𝐕′ (𝑥) ∫ cos𝜃(𝑡)𝑑𝑡 − 𝐖′ (𝑥) ∫ sin𝜃(𝑡)𝑑𝑡 + ϕ′𝑥 (𝑥) ∫ 𝐫(𝑡)𝑑𝑡. (8.44)
∂𝑡
0 0 0 0

Noting that
𝑑𝑦 = cos𝜃(𝑡)𝑑𝑡,
(8.45)
𝑑𝑧 = sin𝜃(𝑡)𝑑𝑡.
we end up with
𝑢(𝑥, 𝑠) = ⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟⏟
𝑢(𝑥, 0) + V′(𝑥)𝑦(0) + W′(𝑥)𝑧(0) + ϕ′𝑥 (𝑥)ϖ
=:𝐔(𝑥)
− 𝑉′(𝑥)
⏟ 𝑦(𝑠) − W′(𝑥)
⏟ z(𝑠) + ϕ′𝑥 (𝑥)(ω(𝑠) − ϖ) (8.46)
=:𝜙𝑧 (𝑥) =:−𝜙𝑦 (𝑥)
= 𝑈(𝑥) − 𝜙𝑧 (𝑥)𝑦 + 𝜙𝑦 (𝑥)𝑧 + 𝜙′𝑥 (𝑥)(𝜔(𝑠) − ϖ).
where 𝑈 denotes the average out-of-plane displacement over the section, 𝜙𝑦 and 𝜙𝑧 denote
the rotation angle about the y and z axis1, respectively, ω is the sectorial area defined as
𝑠
𝜔(𝑠) = ∫ 𝑟(𝑡)𝑑𝑡, (8.47)
0

and ϖ is the average of the sectorial area over the section.

The expression for the displacements in the (x, y, z) coordinate system is


𝑢(𝑥, 𝑦, 𝑧) = 𝑈(𝑥) − 𝜙𝑧 (𝑥)𝑦 + 𝜙𝑦 (𝑥)𝑧 + 𝜗(𝑥)𝜔(𝑦, 𝑧),
𝑣(𝑥, 𝑦, 𝑧) = 𝑉(𝑥) − 𝜙𝑥 (𝑥)𝑧, (8.48)
𝑤(𝑥, 𝑦, 𝑧) = 𝑊(𝑥) + 𝜙𝑥 (𝑥)𝑦,
where we have introduced ϑ to represent the twist constrained by the condition
𝜗(𝑥) = 𝜙𝑥 (𝑥). (8.49)
and 𝜔 denotes the sectorial coordinate that is adjusted for zero average over the section.

6.2.2 Kinetics
The kinetic energy of the beam can be written
1
𝑇= ∫ 𝜌{𝑢̇2 + 𝑣̇2 + 𝑤̇ 2 } 𝑑𝑉. (8.50)
2
𝑉

Taking the variation of this expression leads to

1The substitution of ′ for and ′ for can be seen as a conversion from an Euler-
Bernoulli kinematic assumption to that of Timoschenko.
7-8 (Warped Beam Elements) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Warped Beam Elements

δT = ∫ ρ{u̇ δu̇ + v̇ δv̇ + ẇ δẇ } dV


V
= ∫ ρ{U̇ − 𝜙ż y + 𝜙ẏ z + 𝜗̇ ω}{δU̇ − δ𝜙ż y + δ𝜙ẏ z + δ𝜗̇ ω}dV +
V
∫ ρ{V̇ − 𝜙ẋ z}{δV̇ − δ𝜙ẋ z}dV + ∫ ρ{Ẇ + 𝜙ẋ y}{δẆ + δ𝜙ẋ y}dV
V V
= ∫ ρ{U̇ δU̇ + y2 𝜙ż δ𝜙ż − yω𝜙ż δ𝜗̇ + z2 𝜙ẏ δ𝜙ẏ }dV +
V
∫ ρ{zω𝜙ẏ δ𝜗̇ − yω𝜗̇ δ𝜙ż + zω𝜗̇ δ𝜙ẏ + ω2 𝜗̇ δ𝜗̇ }dV + (8.51)
V
∫ ρ{V̇ δV̇ + z2 𝜙ẋ δ𝜙ẋ + Ẇ δẆ + y2 𝜙ẋ δ𝜙ẋ }dV
V
= ρA ∫{U̇ δU̇ + V̇ δV̇ + Ẇ δẆ }dV + ρIzz ∫{𝜙ż δ𝜙ż + 𝜙ẋ δ𝜙ẋ }dV +
l l
ρIýy ∫{𝜙ẏ δ𝜙ẏ + 𝜙ẋ δ𝜙ẋ }dV + ρIyω ∫{𝜙ẏ δ𝜗̇ + 𝜗̇ δ𝜙ẏ }dV −
l l
ρIzω ∫{𝜗̇ δ𝜙ż + 𝜙ż δ𝜗̇ }dV + ρIωω ∫ 𝜗̇ δ𝜗̇ dV ,
l l

from which the consistent mass matrix can be read out. Here A is the cross sectional area,
Izz and Iyy are the second moments of area with respect to the z and y axes, respectively,
Iωω is the sectorial second moment and Izω and Iyω are the sectorial product moments. An
approximation of this mass matrix can be made by neglecting the off diagonal components.
The diagonal components are
𝜌𝐴𝑙
𝑚TRNS = ,
3
𝜌(𝐼𝑦𝑦 + 𝐼𝑧𝑧 )𝑙
𝑚RT𝑥 = ,
3
𝜌𝐼𝑦𝑦 𝑙
𝑚RT𝑦 = , (8.52)
3
𝜌𝐼 𝑙
𝑚RT𝑧 = 𝑧𝑧 ,
3
𝜌𝐼 𝑙
𝑚TWST = 𝜔𝜔 .
3

With 𝐸 as Young’s modulus and 𝐺 as the shear modulus, the strain energy can be
written
LS-DYNA DEV 10/27/16 (r:8004) 7-9 (Warped Beam Elements)
Warped Beam Elements LS-DYNA Theory Manual

1
𝛱 = (𝐸𝜀2𝑥𝑥 + 𝐺𝛾𝑥𝑦2 2
+ 𝐺𝛾𝑥𝑧 ), (8.53)
2
where the infinitesimal strain components are (neglecting the derivatives of sectorial area)
𝜀𝑥𝑥 = 𝑈 ′ + 𝜙′𝑦 𝑧 − 𝜙′𝑧 𝑦 + 𝜗 ′ 𝜔,
𝛾𝑥𝑦 = 𝑉 ′ − 𝜙′𝑥 𝑧 − 𝜙𝑧 , (8.54)
𝛾𝑥𝑧 = 𝑊 ′ + 𝜙′𝑥 𝑦 + 𝜙𝑦 .
and the variation of the same can be written
𝛿𝜀𝑥𝑥 = 𝛿𝑈 ′ + 𝛿𝜙′𝑦 𝑧 − 𝛿𝜙′𝑧 𝑦 + 𝛿𝜗 ′ 𝜔,
𝛿𝛾𝑥𝑦 = 𝛿𝑉 ′ − 𝛿𝜙′𝑥 𝑧 − 𝛿𝜙𝑧 , (8.55)
𝛿𝛾𝑥𝑧 = 𝛿𝑊 ′ + 𝛿𝜙′𝑥 𝑦 + 𝛿𝜙𝑦 .

The variation of the strain energy is

𝛿𝛱 = ∫{𝐸𝜀𝑥𝑥 𝛿𝜀𝑥𝑥 + 𝐺𝛾𝑥𝑦 𝛿𝛾𝑥𝑦 + 𝐺𝛾𝑥𝑧 𝛿𝛾𝑥𝑧 }𝑑𝑉


𝑉
= 𝐸𝐴 ∫ 𝑈 ′ 𝛿𝑈 ′ 𝑑𝑙 + 𝐺𝐴 ∫ 𝑉 ′ 𝛿𝑉 ′ 𝑑𝑙 + 𝐺𝐴 ∫ 𝑊 ′ 𝛿𝑊 ′ 𝑑𝑙 + 𝐺(𝐼𝑦𝑦 + 𝐼𝑧𝑧 ) ∫ 𝜙′𝑥 𝛿𝜙′𝑥 𝑑𝑙 +
𝑙 𝑙 𝑙 𝑙
′ ′ ′ ′
𝐸𝐼𝑦𝑦 ∫ 𝜙𝑦 𝛿𝜙𝑦 𝑑𝑙 + 𝐺𝐴 ∫ 𝜙𝑦 𝛿𝜙𝑦 𝑑𝑙 + 𝐸𝐼𝑧𝑧 ∫ 𝜙𝑧 𝛿𝜙𝑧 𝑑𝑙 + 𝐺𝐴 ∫ 𝜙𝑧 𝛿𝜙𝑧 𝑑𝑙 + (8.56)
𝑙 𝑙 𝑙 𝑙
𝐸𝐼𝜔𝜔 ∫ 𝜗 𝛿𝜗 𝑑𝑙 − 𝐺𝐴 ∫(𝑉 𝛿𝜙𝑧 + 𝜙𝑧 𝛿𝑉 )𝑑𝑙 + 𝐺𝐴 ∫(𝑊 𝛿𝜙𝑦 + 𝜙𝑦 𝛿𝑊 ′ )𝑑𝑙
′ ′ ′ ′ ′
+
𝑙 𝑙 𝑙

𝐸𝐼𝑦𝜔 ∫(𝜙′𝑦 𝛿𝜗 ′ + 𝜗 ′ 𝛿𝜙′𝑦 )𝑑𝑙 − 𝐸𝐼𝑧𝜔 ∫(𝜙′𝑧 𝛿𝜗 ′ + 𝜗 ′ 𝛿𝜙′𝑧 )𝑑𝑙 ,


𝑙 𝑙

where the stiffness matrix can be read. Again the diagonal components are

7-10 (Warped Beam Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Warped Beam Elements

𝐸𝐴
𝑘TRNS = ,
𝑙
𝐺𝐴
𝑘SHR = ,
𝑙
𝐺(𝐼𝑦𝑦 + 𝐼𝑧𝑧 )
𝑘RTx = ,
𝑙
(8.57)
𝐸𝐼𝑦𝑦
𝐺𝐴𝑙
𝑘RTy = + ,
𝑙 3
𝐸𝐼 𝐺𝐴𝑙
𝑘RTz = 𝑧𝑧 + ,
𝑙 3
𝐸𝐼
𝑘TWST = 𝜔𝜔 .
𝑙

From the expressions of the mass and stiffness matrix, the frequencies of the most
common modes can be estimated. These are
√3 𝐸
1. The tensile and twisting modes with frequency 𝜔 = 𝑙
√𝜌.
√3 𝐺
2. The transverse shear and torsional mode with frequency 𝜔 = 𝑙
√𝜌.
3𝐸 𝐺𝐴 3𝐸 𝐺𝐴
3. The bending modes with frequencies 𝜔 = √𝜌𝑙 2
+ 𝜌𝐼 and 𝜔 = √𝜌𝑙 2
+ 𝜌𝐼 .
𝑦𝑦 𝑧𝑧

Which one of these four that is the highest depends on the geometry of the beam
element. In LS-DYNA the first of these frequencies is used for calculating a stable time
step. We have found no reason for changing approach regarding this element.

6.2.3 Penalty on Twist


The twist is constrained using a penalty that is introduced in the strain energy as
𝑃𝐸A
𝛱P = ∫(𝜙′𝑥 − 𝜗)2 𝑑𝑙, (8.58)
2
𝑙

and the corresponding variation is

𝛿𝛱 = 𝑃𝐸𝐴 ∫(𝜙′𝑥 − 𝜗)(𝛿𝜙′𝑥 − 𝛿𝜗)𝑑𝑙. (8.59)


𝑙

The diagonal of the stiffness matrix is modified as follows

LS-DYNA DEV 10/27/16 (r:8004) 7-11 (Warped Beam Elements)


Warped Beam Elements LS-DYNA Theory Manual

𝐺(𝐼𝑦𝑦 + 𝐼𝑧𝑧 ) 𝑃𝐸𝐴


𝑘RTx = + ,
𝑙 𝑙
(8.60)
𝐸𝐼 𝑃𝐸𝐴𝑙
𝑘TWST = 𝜔𝜔 + .
𝑙 3

3𝐸 𝑃𝐸𝐴
This increases the twist mode frequency to √𝜌𝑙 2
+ 𝜌𝐼 and the torsional mode to
𝜔𝜔

√3 𝐺(𝐼𝑦𝑦 + 𝐼𝑧𝑧 ) + 𝑃𝐸𝐴


√ . (8.61)
𝑙 𝜌

Even though this gives an indication of a frequency increase we have made no


modifications on the computation of the critical time step in an explicit analysis. We have
used 𝑃 = 1 in the implementation. This decision may have to be reconsidered depending
on the choice of the parameter 𝜇, in the end it will come down to trial and error from
numerical simulations.

6.3 Generalization to Large Displacements


A generalization of the small displacement theory to nonlinear theory is quite
straightforward. We have used a corotational formulation where the small strains in the
linear theory are used directly as strain rates in the element system. We emphasize that the
nonlinear beam formulation is obtained by simply replacing displacements for velocities
and strains with strain rates in the previous section.

The nodal velocities for a beam element in the local system is written
T
𝐯 = (𝑣𝑥1 𝑣𝑦1 𝑣𝑧1 𝜔𝑥1 𝜔𝑦1 𝜔𝑧1 𝜗̇ 1 𝑣𝑥2 𝑣𝑦2 𝑣𝑧2 𝜔𝑥2 𝜔𝑦2 𝜔𝑧2 𝜗̇ 2 ) , (8.62)

where the superscript refers to the local node number. These are obtained by transforming
the translational velocities and rotational velocities using the local to global transformation
matrix qij . The strain rate – velocity matrix in the local system can be written
𝐁0 =
−𝑙−1 0 0 0 −𝑙−1 −1
0 𝑧 𝑙0 𝑦 −𝑙−1
0 𝜔 𝑙−1 0 0 0 𝑙−1 −1
0 𝑧 −𝑙0 𝑦 𝑙−1
0 𝜔
⎡ 0 0 ⎤
⎢ 1 ⎥ 1
⎢0 −𝑙−1
0 0 𝑙−1
0 𝑧 0 − 0 0 0 𝑙−1
0 0−𝑙−1
0 𝑧 0
⎥ −
⎢ 2 ⎥,(8.63) 2
⎢ 1 ⎥ 1
⎢0 0 −𝑙−1
0 −𝑙−1
0 𝑦 0 0 0 0 𝑙−1
0 0𝑙−1
0 𝑦⎥ 0
⎢ 2 ⎥ 2
⎢ −1 1 −1 1 ⎥
⎣0 0 0 −𝑙0 0 0 −
2
0 0 0 𝑙0 0 0 − ⎦
2
where 𝑙0 is the beam length in the reference configuration, i.e., beginning of the time step.
A corresponding matrix w.r.t. the current configuration is

7-12 (Warped Beam Elements) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Warped Beam Elements

𝐁=
−𝑙−1 0 0 0 −𝑙−1 𝑧 −𝑙−1 𝑦 −𝑙−1 𝜔 𝑙−1 0 0 0 𝑙−1 𝑧 −𝑙−1 𝑦 𝑙−1 𝜔
⎡ ⎤
⎢ −1 −1 1 −1 1 ⎥
⎢0 −𝑙 0 𝑙 𝑧 0 − 0 0 𝑙 0 −𝑙−1 𝑧 0 − 0 ⎥
⎢ 2 2 ⎥ (8.64)
⎢ 1 1 ⎥,
⎢0 0 −𝑙−1 −𝑙−1 𝑦 0 0 0 0 𝑙−1 𝑙−1 𝑦 0 0 ⎥
⎢ 2 2 ⎥
⎢ 1 1 ⎥
⎣0 0 0 −𝑙−1 0 0 − 0 0 0 𝑙−1 0 0 − ⎦
2 2
where we use the current length of the beam. These matrices are evaluated in each
integration point (𝑥, 𝑦) of the cross section. To compute the strain rate in the local system
we simply apply
𝛆̇ = 𝐁0 𝐯, (8.65)
which is then used to update the local stresses 𝛔. The internal force vector is then
assembled as
𝐟 = 𝐁T 𝛔. (8.66)

Finally the internal force is transformed to the global system using the transfor-
mation matrix.

To compute the stiffness matrix for implicit we neglect the geometric contribution
and just apply
𝐊 = 𝐁T 𝐂𝐁, (8.67)
where 𝐂 is the material tangent modulus. Again the matrix must be transformed to the
global system before used in the implicit solver.

LS-DYNA DEV 10/27/16 (r:8004) 7-13 (Warped Beam Elements)


LS-DYNA Theory Manual Belytschko-Lin-Tsay Shell

9
Belytschko-Lin-Tsay Shell

The Belytschko-Lin-Tsay shell element ([Belytschko and Tsay 1981], [Belytschko et


al., 1984a]) was implemented in LS-DYNA as a computationally efficient alternative to the
Hughes-Liu shell element. For a shell element with five through thickness integration
points, the Belytschko-Lin-Tsay shell elements requires 725 mathematical operations
compared to 4050 operations for the under integrated Hughes-Liu element. The selectively
reduced integration formulation of the explicit Hughes-Liu element requires 35,350
mathematical operations. Because of its computational efficiency, the Belytschko-Lin-Tsay
shell element is usually the shell element formulation of choice. For this reason, it has
become the default shell element formulation for explicit calculations.

The Belytschko-Lin-Tsay shell element is based on a combined co-rotational and


velocity-strain formulation. The efficiency of the element is obtained from the
mathematical simplifications that result from these two kinematical assumptions. The co-
rotational portion of the formulation avoids the complexities of nonlinear mechanics by
embedding a coordinate system in the element. The choice of velocity-strain or rate-of-
deformation in the formulation facilitates the constitutive evaluation, since the conjugate
stress is the physical Cauchy stress. We closely follow the notation of Belytschko, Lin, and
Tsay in the following development.

9.1 Co-rotational Coordinates


The midsurface of the quadrilateral shell element, or reference surface, is defined by
the location of the element’s four corner nodes. An embedded element coordinate system
(see Figure 7.1) that deforms with the element is defined in terms of these nodal
coordinates. Then the procedure for constructing the co-rotational coordinate system
begins by calculating a unit vector normal to the main diagonal of the element:
𝐬
𝐞̂3 = 3 , (9.1)
∥𝐬3 ∥
LS-DYNA DEV 10/27/16 (r:8004) 8-1 (Belytschko-Lin-Tsay Shell)
Belytschko-Lin-Tsay Shell LS-DYNA Theory Manual

^
y s3
4 3

e^3 r31
r42

e^2

s1 ^
x
1
e^1
r21 2

Figure 9.1. Construction of element coordinate system is shown.

∥𝐬3 ∥ = √[𝐬3 ]21 + [𝐬3 ]22 + [𝐬3 ]23 , (9.2)

𝐬3 = 𝐫31 × 𝐫42 , (9.3)


where the superscript caret (⋅ ̂) is used to indicate the local (element) coordinate system.

It is desired to establish the local 𝑥 axis 𝑥̂ approximately along the element edge
between nodes 1 and 2. This definition is convenient for interpreting the element stresses,
which are defined in the local 𝑥̂ − 𝑦̂ coordinate system. The procedure for constructing this
unit vector is to define a vector 𝐬1 that is nearly parallel to the vector 𝐫21 , viz.
𝐬1 = 𝐫21 − (𝐫21 ⋅ 𝐞̂3 )𝐞̂3 , (9.4)
𝐬1
𝐞̂1 = . (9.5)
‖𝐬1 ‖

The remaining unit vector is obtained from the vector cross product
𝐞̂2 = 𝐞̂3 × 𝐞̂1 . (9.6)

If the four nodes of the element are coplanar, then the unit vectors 𝐞̂1 and 𝐞̂2 are
tangent to the midplane of the shell and 𝐞̂3 is in the fiber direction. As the element
deforms, an angle may develop between the actual fiber direction and the unit normal 𝐞̂3 .
The magnitude of this angle may be characterized as
∣𝐞̂3 ⋅ 𝐟 − 1∣ < 𝛿, (9.7)
where 𝐟 is the unit vector in the fiber direction and the magnitude of 𝛿 depends on the
magnitude of the strains. According to Belytschko et al., for most engineering applications,
acceptable values of 𝛿 are on the order of 10-2 and if the condition presented in Equation
8-2 (Belytschko-Lin-Tsay Shell) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Belytschko-Lin-Tsay Shell

(9.7) is met, then the difference between the rotation of the co-rotational coordinates 𝑒 ̂ and
the material rotation should be small.

The global components of this co-rotational triad define a transformation matrix


between the global and local element coordinate systems. This transformation operates on
vectors with global components 𝐀 = (𝐴𝑥 𝐴𝑦 𝐴𝑧 ) and element coordinate components
̂ = (𝐴𝑥̂ 𝐴𝑦̂ 𝐴𝑧̂ ), and is defined as:
𝐀

⎧ 𝐴 𝑒 𝑒2𝑥 𝑒3𝑥 ⎧ 𝐴𝑥̂ ⎫


{ 𝑥⎫} ⎡ 1𝑥 {
{ }
}
𝐀 = ⎨𝐴𝑦 ⎬ = ⎢𝑒1𝑦 𝑒2𝑦 𝑒3𝑦 ⎤ 𝐴 ̂ ̂ = 𝐪T 𝐀
⎥ ⎨ 𝑦 ⎬ = 𝛍𝐀 ̂, (9.8)
{
⎩𝐴𝑧 }
⎭ ⎣𝑒1𝑧 𝑒2𝑧 { }
𝑒3𝑧 ⎦ { }
⎩𝐴𝑧̂ ⎭
where 𝑒𝑖𝑥 , 𝑒𝑖𝑦 , 𝑒𝑖𝑧 are the global components of the element coordinate unit vectors. The
inverse transformation is defined by the matrix transpose, i.e.,
̂ = 𝛍T 𝐀.
𝐀 (9.9)

9.2 Velocity-Strain Displacement Relations


The above small rotation condition, Equation (9.7), does not restrict the magnitude
of the element’s rigid body rotations. Rather, the restriction is placed on the out-of-plane
deformations, and, thus, on the element strain. Consistent with this restriction on the
magnitude of the strains, the velocity-strain displacement relations used in the Belytschko-
Lin-Tsay shell are also restricted to small strains.

As in the Hughes-Liu shell element, the displacement of any point in the shell is
partitioned into a midsurface displacement (nodal translations) and a displacement
associated with rotations of the element’s fibers (nodal rotations). The Belytschko-Lin-Tsay
shell element uses the Mindlin [1951] theory of plates and shells to partition the velocity of
any point in the shell as:
𝐯 = 𝐯 𝑚 − 𝑧̂ 𝐞3 × 𝛉, (9.10)
where 𝐯 𝑚 is the velocity of the mid-surface, 𝛉 is the angular velocity vector, and 𝑧̂ is the
distance along the fiber direction (thickness) of the shell element. The corresponding co-
rotational components of the velocity strain (rate of deformation) are given by
1 ∂𝜐̂𝑖 ∂𝜐̂𝑗
𝑑𝑖𝑗̂ = ( + ). (9.11)
2 ∂𝑥̂𝑗 ∂𝑥̂𝑖
Substitution of Equation (9.10) into the above yields the following velocity-strain relations:

∂𝑣̂𝑥𝑚 ∂ 𝜃𝑦̂
̂
𝑑𝑥 = + 𝑧̂ , (9.12)
∂𝑥̂ ∂𝑥̂

LS-DYNA DEV 10/27/16 (r:8004) 8-3 (Belytschko-Lin-Tsay Shell)


Belytschko-Lin-Tsay Shell LS-DYNA Theory Manual

∂𝜐̂𝑦𝑚 ∂ 𝜃𝑥̂
𝑑𝑦̂ = − 𝑧̂ , (9.13)
∂𝑦̂ ∂𝑦̂
𝑚
∂𝜐̂𝑥𝑚 ∂𝜐̂𝑦 ∂ 𝜃𝑦̂ ∂ 𝜃𝑥̂ ⎞
̂
2𝑑𝑥𝑦 = + + 𝑧̂ ⎛
⎜ − ⎟, (9.14)
∂𝑦̂ ∂𝑥̂ ⎝ ∂𝑦̂ ∂𝑥̂ ⎠

̂ = ∂𝜐̂𝑧𝑚
2𝑑𝑦𝑧 − 𝜃𝑥̂ , (9.15)
∂𝑦̂

̂ = ∂𝜐̂𝑧𝑚
2𝑑𝑥𝑧 + 𝜃𝑦̂ . (9.16)
∂𝑥̂

The above velocity-strain relations need to be evaluated at the quadrature points


within the shell. Standard bilinear nodal interpolation is used to define the mid-surface
velocity, angular velocity, and the element’s coordinates (isoparametric representation).
These interpolations relations are given by
𝐯𝑚 = 𝑁𝐼 (𝜉 , 𝜂)𝐯𝐼 ,
𝛉𝑚 = 𝑁𝐼 (𝜉 , 𝜂)𝛉𝐼 , (9.17)
𝐱𝑚 = 𝑁𝐼 (𝜉 , 𝜂)𝐱𝐼 .
where the subscript 𝐼 is summed over all the nodes of the element and the nodal velocities
are obtained by differentiating the nodal coordinates with respect to time, i.e., 𝜐𝐼 = 𝑥̇𝐼 . The
bilinear shape functions are
1
𝑁1 = (1 − 𝜉 )(1 − 𝜂), (9.18)
4
1
𝑁2 = (1 + 𝜉 )(1 − 𝜂), (9.19)
4
1
𝑁3 = (1 + 𝜉 )(1 + 𝜂), (9.20)
4
1
𝑁4 = (1 − 𝜉 )(1 + 𝜂). (9.21)
4

The velocity-strains at the center of the element, i.e., at 𝜉 = 0, and 𝜂 = 0, are obtained
by substitution of the above relations into the previously defined velocity-strain
displacement relations, Equations (9.12) and (9.16). After some algebra, this yields
𝑑𝑥̂ = 𝐵1𝐼 𝜐̂𝑥𝐼 + 𝑧̂𝐵1𝐼 𝜃𝑦𝐼
̂ , (9.22a)

𝑑𝑦̂ = 𝐵2𝐼 𝜐̂𝑦𝐼 − 𝑧̂𝐵2𝐼 𝜃𝑥𝐼


̂ , (9.22b)

8-4 (Belytschko-Lin-Tsay Shell) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Belytschko-Lin-Tsay Shell

̂ = 𝐵2𝐼 𝜐̂𝑥𝐼 + 𝐵1𝐼 𝜐̂𝑦𝐼 + 𝑧̂ (𝐵2𝐼 𝜃𝑦𝐼


2𝑑𝑥𝑦 ̂ − 𝐵1𝐼 𝜃𝑥𝐼
̂ ), (9.22c)

̂ = 𝐵1𝐼 𝜐̂𝑧𝐼 + 𝑁𝐼 𝜃𝑦𝐼


2𝑑𝑥𝑧 ̂ , (9.22d)

̂ = 𝐵2𝐼 𝜐̂𝑧𝐼 − 𝑁𝐼 𝜃𝑥𝐼


2 𝑑𝑦𝑧 ̂ , (9.22e)

∂𝑁𝐼
𝐵1𝐼 = , (9.22f)
∂𝑥̂
∂𝑁𝐼
𝐵2𝐼 = . (9.22g)
∂𝑦̂
The shape function derivatives 𝐵𝑎𝐼 are also evaluated at the center of the element, i.e., at
𝜉 = 0, and 𝜂 = 0.

9.3 Stress Resultants and Nodal Forces


After suitable constitutive evaluations using the above velocity-strains, the resulting
stresses are integrated through the thickness of the shell to obtain local resultant forces and
moments. The integration formula for the resultants are
𝑅̂ = ∫ 𝜎̂ 𝛼𝛽 𝑑𝑧̂,
𝑓𝛼𝛽 (9.23)

̂𝑅
𝑚 𝛼𝛽 = − ∫ 𝑧̂𝜎̂𝛼𝛽 𝑑𝑧̂, (9.24)

where the superscript, 𝑅, indicates a resultant force or moment, and the Greek subscripts
emphasize the limited range of the indices for plane stress plasticity.

The above element-centered force and moment resultants are related to the local
nodal forces and moments by invoking the principle of virtual power and integrating with
a one-point quadrature. The relations obtained in this manner are
𝑓𝑥𝐼̂ = 𝐴(𝐵1𝐼 𝑓𝑥𝑥
𝑅̂ + 𝐵2𝐼 𝑓𝑥𝑦
𝑅̂ ), (9.25)

𝑓𝑦𝐼̂ = 𝐴(𝐵2𝐼 𝑓𝑦𝑦


𝑅̂ + 𝐵1𝐼 𝑓𝑥𝑦
𝑅̂ ), (9.26)

𝑓𝑧𝐼̂ = 𝐴𝜅(𝐵1𝐼 𝑓𝑥𝑧


𝑅̂ + 𝐵2𝐼 𝑓𝑦𝑧
𝑅̂ ), (9.27)

𝜅 𝑅̂
̂ 𝑥𝐼 = 𝐴 (𝐵2𝐼 𝑚
𝑚 ̂𝑅 ̂𝑅
𝑦𝑦 + 𝐵1𝐼 𝑚 𝑥𝑦 − 𝑓𝑦𝑧 ), (9.28)
4
𝜅 𝑅̂
𝑚 ̂𝑅
̂ 𝑦𝐼 = −𝐴 (𝐵1𝐼 𝑚 ̂𝑅
𝑥𝑥 + 𝐵2𝐼 𝑚 𝑥𝑦 − 𝑓𝑥𝑧 ), (9.29)
4
𝑚
̂ 𝑧𝐼 = 0, (9.30)

LS-DYNA DEV 10/27/16 (r:8004) 8-5 (Belytschko-Lin-Tsay Shell)


Belytschko-Lin-Tsay Shell LS-DYNA Theory Manual

where 𝐴 is the area of the element, and 𝜅 is the shear factor from the Mindlin theory. In the
Belytschko-Lin-Tsay formulation, 𝜅 is used as a penalty parameter to enforce the Kirchhoff
normality condition as the shell becomes thin.

The above local nodal forces and moments are then transformed to the global
coordinate system using the transformation relations given previously as Equation (9.8).
The global nodal forces and moments are then appropriately summed over all the nodes
and the global equations of motion are solved for the next increment in nodal accelerations.

9.4 Hourglass Control (Belytschko-Lin-Tsay)


In part, the computational efficiency of the Belytschko-Lin-Tsay and the under
integrated Hughes-Liu shell elements are derived from their use of one-point quadrature in
the plane of the element. To suppress the hourglass deformation modes that accompany
one-point quadrature, hourglass viscosity stresses are added to the physical stresses at the
local element level. The discussion of the hourglass control that follows pertains to the
Hughes-Liu and the membrane elements as well.

The hourglass control used by Belytschko et al., extends an earlier derivation by


Flanagan and Belytschko [1981], (see also Kosloff and Frazier [1978], Belytschko and Tsay
[1983]). The hourglass shape vector, 𝛕𝐼 , is defined as
𝛕𝐼 = 𝐡𝐼 − (𝐡𝐽 𝐱̂𝑎𝐽 )𝐁𝑎𝐼 , (9.31)
where
+1
⎡−1⎤
𝐡=⎢
⎢+1⎥⎥, (9.32)
⎣−1⎦
is the basis vector that generates the deformation mode that is neglected by one-point
quadrature. In Equation (9.31) and the reminder of this subsection, the Greek subscripts
have a range of 2, e.g., 𝐱̂𝑎𝐼 = (𝑥̂1𝐼 , 𝑥̂2𝐼 ) = (𝑥̂𝐼 , 𝑦̂𝐼 ).

The hourglass shape vector then operates on the generalized displacements, in a


manner similar to Equations (7.11a - e), to produce the generalized hourglass strain rates
̂ ,
𝐪̇𝛼𝐵 = 𝛕𝐼 𝜃𝛼𝐼 (9.33)

𝐪̇3𝐵 = 𝝉𝐼 𝜐̂𝑧𝐼 , (9.34)

𝐪̇𝛼𝑀 = 𝛕𝐼 𝜐̂𝛼𝐼 , (9.35)


where the superscripts 𝐁 and 𝐌 denote bending and membrane modes, respectively. The
corresponding hourglass stress rates are then given by

8-6 (Belytschko-Lin-Tsay Shell) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Belytschko-Lin-Tsay Shell

𝑟𝜃 𝐸𝑡3 𝐴
𝑄̇𝐵𝛼 = 𝐵𝛽𝐼 𝐵𝛽𝐼 𝑞𝐵𝛼̇ , (9.36)
192

𝑟𝑤 𝜅𝐺𝑡3 𝐴
𝑄̇𝐵3 = 𝐵𝛽𝐼 𝐵𝛽𝐼 𝑞𝐵3̇ , (9.37)
12
𝑟𝑚 𝐸𝑡𝐴
𝑄̇𝑀
𝛼 = 𝐵𝛽𝐼 𝐵𝛽𝐼 𝑞𝑀
𝛼̇ , (9.38)
8
where 𝑡 is the shell thickness and the parameters, 𝑟𝜃 , 𝑟𝑤 , and 𝑟𝑚 are generally assigned
values between 0.01 and 0.05.

Finally, the hourglass stresses, which are updated from the stress rates in the usual
way, i.e.,
𝐐𝑛+1 = 𝐐𝑛 + Δ𝑡𝐐̇ , (9.39)
and the hourglass resultant forces are then
̂𝐻
𝑚 𝐵
𝛼𝐼 = 𝜏𝐼 𝑄𝛼 , (9.40)

𝐻̂ = 𝜏𝐼 𝑄𝐵,
𝑓3𝐼 (9.41)
3

𝐻̂ = 𝜏𝐼 𝑄𝑀,
𝑓𝛼𝐼 𝛼 (9.42)
where the superscript 𝐻 emphasizes that these are internal force contributions from the
hourglass deformations. These hourglass forces are added directly to the previously
determined local internal forces due to deformations Equations (7.14a - f). These force
vectors are orthogonalized with respect to rigid body motion.

9.5 Hourglass Control (Englemann and Whirley)


Englemann and Whirley [1991] developed an alternative hourglass control, which
they implemented in the framework of the Belytschko, Lin, and Tsay shell element. We
will briefly highlight their procedure here that has proven to be cost effective-only twenty
percent more expensive than the default control.

In the hourglass procedure, the in-plane strain field (subscript p) is decomposed into
the one point strain field plus the stabilization strain field:
𝛆̅ṗ = 𝛆̅p0̇ + 𝛆̅p𝑠̇ , (9.43)

𝛆̅p𝑠̇ = 𝐖𝑚 𝐪̇𝑚 + 𝑧𝐖𝑏 𝐪̇𝑏 . (9.44)


Here, 𝐖𝑚 and 𝐖𝑏 play the role of stabilization strain velocity operators for membrane and
bending:

LS-DYNA DEV 10/27/16 (r:8004) 8-7 (Belytschko-Lin-Tsay Shell)


Belytschko-Lin-Tsay Shell LS-DYNA Theory Manual

𝑝 𝑝
𝑓 (𝜉 , 𝜂) 𝑓4 (𝜉 , 𝜂)
⎡ 1𝑝 𝑝

𝐖𝑚 = ⎢
⎢ 𝑓 2 (𝜉 , 𝜂) 𝑓 5 (𝜉 , 𝜂)⎥,
⎥ (9.45)
𝑝 𝑝
⎣𝑓3 (𝜉 , 𝜂) 𝑓6 (𝜉 , 𝜂)⎦
𝑝 𝑝
−𝑓 (𝜉 , 𝜂) 𝑓1 (𝜉 , 𝜂)
⎡ 4𝑝 𝑝

𝐖𝑏 = ⎢
⎢−𝑓 5 (𝜉 , 𝜂) 𝑓 2 (𝜉 , 𝜂)⎥,
⎥ (9.46)
𝑝 𝑝
⎣−𝑓6 (𝜉 , 𝜂) 𝑓3 (𝜉 , 𝜂)⎦
𝑝
where the terms 𝑓𝑖 (𝜉 , 𝜂) 𝑖 = 1, 2, . . . , 6, are rather complicated and the reader is referred to
the reference [Englemann and Whirley, 1991].

To obtain the transverse shear assumed strain field, the procedure given in [Bathe
and Dvorkin, 1984] is used. The transverse shear strain field can again be decomposed into
the one point strain field plus the stabilization field:
𝛆̅ṡ = 𝛆̅s0̇ + 𝛆̅s𝑠̇ , (9.47)
that is related to the hourglass velocities by
𝛆̅s𝑠̇ = 𝐖s 𝐪̇𝑠 , (9.48)
where the transverse shear stabilization strain-velocity operator 𝐖𝑠 is given by
𝑓 𝑠 (𝜉 , 𝜂) −𝑔1𝑠 𝜉 𝑔2𝑠 𝜂 𝑔3𝑠 𝜉 𝑔3𝑠 𝜂
𝐖𝑠 = [ 1𝑠 ]. (9.49)
𝑓2 (𝜉 , 𝜂) 𝑔4𝑠 𝜉 𝑔4𝑠 𝜂 −𝑔2𝑠 𝜉 𝑔1𝑠 𝜂
Again, the coefficients 𝑓1𝑠 (𝜉 , 𝜂) and 𝑔1𝑠 are defined in the reference.

In their formulation, the hourglass forces are related to the hourglass velocity field
through an incremental hourglass constitutive equation derived from an additive
decomposition of the stress into a “one-point stress,” plus a “stabilization stress.” The
integration of the stabilization stress gives a resultant constitutive equation relating
hourglass forces to hourglass velocities. The in-plane and transverse stabilization stresses
are updated according to:
𝛕s𝑠,𝑛+1 = 𝛕s𝑠,𝑛 + Δ𝑡𝑐𝑠 𝐂𝑠 𝛆̅𝑠𝑠̇ , (9.50)

𝛕s𝑠,𝑛+1 = 𝛕s𝑠,𝑛 + Δ𝑡𝑐s 𝐂s 𝛆̅s𝑠̇ , (9.51)


where the tangent matrix is the product of a matrix 𝐂, which is constant within the shell
domain, and a scalar 𝑐 that is constant in the plane but may vary through the thickness.

The stabilization stresses can now be used to obtain the hourglass forces:

𝐐𝑚 = ∫2ℎ ∫ 𝐖𝑚
T 𝑠
𝛕𝑝 𝑑𝐴 𝑑𝑧, (9.52)
− 𝐴
2

8-8 (Belytschko-Lin-Tsay Shell) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Belytschko-Lin-Tsay Shell

Figure 9.2. The twisted beam problem fails with the Belytschko-Tsay shell
element.


𝐐𝑏 = ∫2ℎ ∫ 𝐖𝑏T 𝛕𝑝𝑠 𝑑𝐴 𝑑𝑧, (9.53)
− 𝐴
2


𝐐𝑠 = ∫2ℎ ∫ 𝐖𝑠T 𝛕𝑠𝑠 𝑑𝐴 𝑑𝑧. (9.54)
− 𝐴
2

9.6 Belytschko-Wong-Chiang Improvements


Since the Belytschko-Tsay element is based on a perfectly flat geometry, warpage is
not considered. Although this generally poses no major difficulties and provides for an
efficient element, incorrect results in the twisted beam problem, See Figure 7.2, are obtained
where the nodal points of the elements used in the discretization are not coplanar. The
Hughes-Liu shell element considers non-planar geometry and gives good results on the
twisted beam, but is relatively expensive. The effect of neglecting warpage in typical a
application cannot be predicted beforehand and may lead to less than accurate results, but
the latter is only speculation and is difficult to verify in practice. Obviously, it would be
better to use shells that consider warpage if the added costs are reasonable and if this
unknown effect is eliminated. In this section we briefly describe the simple and
computationally inexpensive modifications necessary in the Belytschko-Tsay shell to
include the warping stiffness. The improved transverse shear treatment is also described
which is necessary for the element to pass the Kirchhoff patch test. Readers are directed to
the references [Belytschko, Wong, and Chang 1989, 1992] for an in depth theoretical
background.

LS-DYNA DEV 10/27/16 (r:8004) 8-9 (Belytschko-Lin-Tsay Shell)


Belytschko-Lin-Tsay Shell LS-DYNA Theory Manual

P3

P1
P2
h

Figure 9.3. Nodal fiber vectors 𝐩1 , 𝐩2 , and 𝐩3 , where ℎ is the thickness.

In order to include warpage in the formulation it is convenient to define nodal fiber


vectors as shown in Figure 7.3. The geometry is interpolated over the surface of the shell
from:
𝑥 = 𝑥𝑚 + 𝜁 𝑝̅ = (𝑥𝐼 + 𝜁 𝑝̅ 𝐼 )𝑁𝐼 (𝜉 , 𝜂), (9.55)
where
𝜁ℎ
𝜁̅ = . (9.56)
2

The in plane strain components are given by:


̅ 𝑐𝑥𝐼 𝑣̂𝑥𝐼 + 𝑏𝑥𝐼 𝑝̇𝑥𝐼 ),
𝑑𝑥𝑥 = 𝑏𝑥𝐼 𝑣̂𝑥𝐼 + 𝜁 (𝑏 (9.57)

̅ 𝑐𝑦𝐼 𝑣̂𝑦𝐼 + 𝑏𝑦𝐼 𝑝̇𝑦𝐼 ),


𝑑𝑦𝑦 = 𝑏𝑦𝐼 𝑣̂𝑦𝐼 + 𝜁 (𝑏 (9.58)

1
̅ 𝑐𝑥𝐼 𝑣̂𝑦𝐼 + 𝑏𝑥𝐼 𝑝̇𝑦𝐼 + 𝑏𝑐𝑦𝐼 𝑣̂𝑥𝐼 + 𝑏𝑦𝐼 𝑝̇𝑥𝐼 ).
𝑑𝑥𝑦 = 𝑏𝑥𝐼 𝑣̂𝑦𝐼 + 𝑏𝑦𝐼 𝑣̂𝑥𝐼 + 𝜁 (𝑏 (9.59)
2
The coupling terms are come in through 𝑏𝑐𝑖𝐼 : which is defined in terms of the components of
the fiber vectors as:
𝑏𝑐𝑥𝐼 𝑝𝑦̂2 − 𝑝𝑦̂4 𝑝𝑦̂3 − 𝑝𝑦̂1 𝑝𝑦̂4 − 𝑝𝑦̂2 𝑝𝑦̂1 − 𝑝𝑦̂3
{ 𝑐 } = [𝑝 − 𝑝 𝑝𝑥̂3 − 𝑝𝑥̂1 𝑝𝑥̂4 − 𝑝𝑥̂2 𝑝𝑥̂1 − 𝑝𝑥̂3 ], (9.60)
𝑏𝑦𝐼 𝑥̂2 𝑥̂4

For a flat geometry the normal vectors are identical and no coupling can occur. Two
methods are used by Belytschko for computing 𝑏𝑐𝑖𝐼 and the reader is referred to his papers
for the details. Both methods have been tested in LS-DYNA and comparable results were
obtained.
The transverse shear strain components are given as
𝛾̂ 𝑥𝑧 = −𝑁𝐼 (𝜉 , 𝜂)𝜃𝑦̂̅ 𝐼 , (9.61)

𝛾̂ 𝑦𝑧 = −𝑁𝐼 (𝜉 , 𝜂)𝜃𝑥̂̅ 𝐼 , (9.62)

8-10 (Belytschko-Lin-Tsay Shell) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Belytschko-Lin-Tsay Shell

^
y

K
r
e^y
Lk

enk eni

J
K
I e^x
Figure 9.4. Vector and edge definitions for computing the transverse shear strain
components.

where the nodal rotational components are defined as:


𝜃𝑥̂̅ 𝐼 = (𝐞𝐼𝑛 ⋅ 𝐞𝑥̂ )𝜃𝑛𝐼̅ + (𝐞𝐾 ̅
𝐾
𝑛 ⋅ 𝐞𝑥̂ )𝜃𝑛 , (9.63)

𝜃𝑦̂̅ 𝐼 = (𝐞𝐼𝑛 ⋅ 𝐞𝑦̂ )𝜃𝑛𝐼̅ + (𝐞𝐾 ̅


𝐾
𝑛 ⋅ 𝐞𝑦̂ )𝜃𝑛 , (9.64)

1 𝐼 1
𝜃𝑛𝐼̅ = (𝜃𝑛𝐼 𝐼
+ 𝜃𝑛𝐽 ) + 𝐼𝐽 (𝜐̂𝑧𝐽 − 𝜐̂𝑧𝐽 ), (9.65)
2 𝐿
where the subscript n refers to the normal component of side 𝐼 as seen in Figure 7.3 and 𝐿𝐼𝐽
is the length of side 𝐼𝐽.

LS-DYNA DEV 10/27/16 (r:8004) 8-11 (Belytschko-Lin-Tsay Shell)


LS-DYNA Theory Manual Triangular Shells

10
Triangular Shells

10.1 C0 Triangular Shell

The 𝐶0 shell element due to Kennedy, Belytschko, and Lin [1986] has been
implemented as a computationally efficient triangular element complement to the
Belytschko-Lin-Tsay quadrilateral shell element ([Belytschko and Tsay 1981], [Belytschko et
al., 1984a]). For a shell element with five through-the-thickness integration points, the
element requires 649 mathematical operations (the Belytschko-Lin-Tsay quadrilateral shell
element requires 725 mathematical operations) compared to 1417 operations for the
Marchertas-Belytschko triangular shell [Marchertas and Belytschko 1974] (referred to as the
BCIZ [Bazeley, Cheung, Irons, and Zienkiewicz 1965] triangular shell element in the
DYNA3D user’s manual).

Triangular shell elements are offered as optional elements primarily for


compatibility with local user grid generation and refinement software. Many computer
aided design (CAD) and computer aided manufacturing (CAM) packages include finite
element mesh generators, and most of these mesh generators use triangular elements in the
discretization. Similarly, automatic mesh refinement algorithms are typically based on
triangular element discretization. Also, triangular shell element formulations are not
subject to zero energy modes inherent in quadrilateral element formulations.

The triangular shell element’s origins are based on the work of Belytschko et al.,
[Belytschko, Stolarski, and Carpenter 1984b] where the linear performance of the shell was
demonstrated. Because the triangular shell element formulations parallels closely the
formulation of the Belytschko-Lin-Tsay quadrilateral shell element presented in the
previous section (Section 7), the following discussion is limited to items related specifically
to the triangular shell element.

10.1.1 Co-rotational Coordinates


The mid-surface of the triangular shell element, or reference surface, is defined by
the location of the element’s three nodes. An embedded element coordinate system (see
LS-DYNA DEV 10/27/16 (r:8004) 9-1 (Triangular Shells)
Triangular Shells LS-DYNA Theory Manual

^
z

^
y
3

e^2
e^3
^
x
1
e^1 2
Figure 10.1. Local element coordinate system for C0 shell element.

Figure 10.1) that deforms with the element is defined in terms of these nodal coordinates.
The procedure for constructing the co-rotational coordinate system is simpler than the
corresponding procedure for the quadrilateral, because the three nodes of the triangular
element are guaranteed coplanar.

The local x-axis 𝑥̂ is directed from node 1 to 2. The element’s normal axis 𝑧̂ is
defined by the vector cross product of a vector along 𝑥̂ with a vector constructed from node
1 to node 3. The local y-axis 𝑦̂ is defined by a unit vector cross product of 𝐞̂3 with 𝐞̂1 ,
which are the unit vectors in the 𝑧̂ directions, respectively. As in the case of the
quadrilateral element, this triad of co-rotational unit vectors defines a transformation
between the global and local element coordinate systems. See Equations (7.5 a, b).

10.1.2 Velocity-Strain Relations


As in the Belytschko-Lin-Tsay quadrilateral shell element, the displacement of any
point in the shell is partitioned into a mid-surface displacement (nodal translations) and a
displacement associated with rotations of the element’s fibers (nodal rotations). The
Kennedy-Belytschko-Lin triangular shell element also uses the Mindlin [Mindlin 1951]
theory of plates and shells to partition the velocity of any point in the shell (recall Equation
(7.6)):
𝐯 = 𝐯m − 𝑧̂ 𝐞3 × 𝛉, (10.1)
where 𝐯m is the velocity of the mid-surface, 𝛉 is the angular velocity vector, and 𝑧̂ is the
distance along the fiber direction (thickness) of the shell element. The corresponding co-
rotational components of the velocity strain (rate of deformation) were given previously in
Equation (7.11 a - e).

Standard linear nodal interpolation is used to define the midsurface velocity,


angular velocity, and the element’s coordinates (isoparametric representation). These
interpolation functions are the area coordinates used in triangular element formulations.
Substitution of the nodally interpolated velocity fields into the velocity-strain relations (see
Belytschko et al., for details), leads to strain rate-velocity relations of the form
𝐝̂ = 𝐁 𝐯̂ . (10.2)

9-2 (Triangular Shells) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Triangular Shells

It is convenient to partition the velocity strains and the 𝐁 matrix into membrane and
bending contributions. The membrane relations are given by
⎧𝜐̂𝑥1 ⎫
{
⎧ 𝑑𝑥̂ ⎫
M {
{ 𝜐̂𝑦1 }
}
}
{
{ ̂ } 𝑦̂3 0 𝑦̂3 0 0 0 { { }
} 1 ⎡ 𝑥2 }
𝜐̂
⎨ 𝑑 𝑦 ⎬ = ⎢ 0 𝑥̂3 − 𝑥̂2 0 −𝑥̂3 0 𝑥̂2 ⎤
⎥ ⎨𝜐̂ ⎬, (10.3)
{ 𝑥̂2 𝑦̂3
{2𝑑 ̂ }
} ⎣𝑥̂3 − 𝑥̂2 −𝑦̂3 −𝑥̂3 𝑦̂3 𝑥̂2 0⎦ { {
𝑦2 }
}
⎩ 𝑥𝑦 ⎭ { 𝑥3 }
𝜐̂
{
{𝜐̂ } }
⎩ 𝑦3 ⎭

𝐝̂ M = 𝐁M 𝐯̂ , (10.4)

⎧𝜃𝑥1 ̂ ⎫
{
{ }
{ ̂ }
𝜃𝑦1
{ } }
⎧ 𝜅̂𝑥 ⎫ 0 −𝑦̂3 0 𝑦̂3 0 0 { { ̂ }
{ } −1 ⎡ 𝜃𝑥2 }

𝜅̂𝑦 =
⎬ 𝑥̂
⎢ 3 𝑥̂2
− 0 𝑥̂3 0 −𝑥̂2 0⎤ ⎥ ⎨ ̂ ⎬, (10.5)
{2𝜅̂ } 𝑥̂2 𝑦̂3 ⎣ 𝑦̂ 𝑥̂3 − 𝑥̂2 −𝑦̂3 −𝑥̂3 0 𝑥̂2 ⎦ { 𝜃𝑦2 }
⎩ 𝑥𝑦 ⎭ 3 {
{ }
{ 𝜃𝑥3 ̂ }}
{
{𝜃 ̂ } }
⎩ 𝑦3 ⎭
or
𝛋̂ M = 𝐁M 𝛉̂ def . (10.6)

The local element velocity strains are then obtained by combining the above two
relations:
M
⎧ 𝑑𝑥̂ ⎫ ⎧ 𝑑𝑥̂ ⎫ ⎧ 𝜅̂𝑥 ⎫
{ } {
{ ̂ } { ̂ } } { }
𝜅̂𝑦 = 𝐝̂ M − 𝑧̂𝛋̂ . (10.7)
⎨ 𝑑𝑦 ⎬ = ⎨ 𝑑𝑦 ⎬ − 𝑧̂ ⎨ ⎬
{
{2𝑑 ̂ }
} {{ ̂ }} {2𝜅̂ }
⎩ 𝑥𝑦 ⎭ ⎩2𝑑𝑥𝑦 ⎭ ⎩ 𝑥𝑦 ⎭

The remaining two transverse shear strain rates are given by


̂
2𝑑𝑥𝑧 1
{ ̂ } =
2𝑑𝑦𝑧 6 𝑥̂2 𝑦̂3
(10.8)
−𝑦̂32 𝑦̂3 (2 𝑥̂2 + 𝑥̂3 ) 𝑦̂32 𝑦̂3 (3 𝑥̂2 − 𝑥̂3 ) 0 𝑥̂2 𝑦̂3
[ ]
𝑦̂3 ( 𝑥̂2 − 2𝑥̂2 ) 𝑥̂22 − 𝑥̂32 −𝑦̂32 ( 𝑥̂2 + 𝑥̂3 ) 𝑥̂3 ( 𝑥̂3 − 2𝑥̂2 ) −3𝑥̂2 𝑦̂3 𝑥̂2 ( 2𝑥̂3 − 𝑥̂2 )

LS-DYNA DEV 10/27/16 (r:8004) 9-3 (Triangular Shells)


Triangular Shells LS-DYNA Theory Manual

⎧𝜃𝑥1 ̂ ⎫def
{
{ }
{ ̂ }
𝜃𝑦1 }
{
{ }
{𝜃𝑥2 ̂ } }
⎨𝜃 ̂ ⎬ ,
{
{ 𝑦2 }
}
{
{ 𝜃𝑥3 ̂ }}
{
{𝜃 ̂ } }
⎩ 𝑦3 ⎭

𝐝̂ S = 𝐁S 𝛉̂ def
. (10.9)
All of the above velocity-strain relations have been simplified by using one-point
quadrature.

In the above relations, the angular velocities 𝛉̂def are the deformation component of
the angular velocity 𝛉̂ obtained by subtracting the portion of the angular velocity due to
rigid body rotation, i.e.,
𝛉̂def = 𝛉̂ − 𝛉̂rig , (10.10)
The two components of the rigid body angular velocity are given by
rig 𝜐̂𝑧1 − 𝜐̂𝑧2
𝜃𝑦̂ = , (10.11)
𝑥̂2

rig (𝜐̂𝑧3 − 𝜐̂𝑧1 )𝑥̂2 − (𝜐̂𝑧2 − 𝜐̂𝑧1 )𝑥̂3


𝜃𝑥̂ = . (10.12)
𝑥̂2 𝑦̂3
The first of the above two relations is obtained by considering the angular velocity of the
local x-axis about the local y-axis. Referring to Figure 10.1, by construction nodes 1 and 2
lie on the local x-axis and the distance between the nodes is 𝑥̂2 i.e., the 𝑥̂ distance from node
2 to the local coordinate origin at node 1. Thus the difference in the nodal 𝑧̂ velocities
divided by the distance between the nodes is an average measure of the rigid body rotation
rate about the local y-axis.

The second relation is conceptually identical, but is implemented in a slightly


different manner due to the arbitrary location of node 3 in the local coordinate system.
Consider the two local element configurations shown in Figure 10.2. For the leftmost
configuration, where node 3 is the local y-axis, the rigid body rotation rate about the local
x-axis is given by
rig 𝜐̂𝑧3 − 𝜐̂𝑧1
̂
𝜃𝑥−left = , (10.13)
𝑦̂3
and for the rightmost configuration the same rotation rate is given by
rig 𝜐̂𝑧3 − 𝜐̂𝑧2
̂
𝜃𝑥−right = . (10.14)
𝑦̂3

9-4 (Triangular Shells) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Triangular Shells

^
z

3 ^
y

^
x
1 2

^
z

^
y
3

^
x
1 2
Figure 10.2. Element configurations with node 3 aligned with node 1 (left) and
node 3 aligned with node 2 (right).

Although both of these relations yield the average rigid body rotation rate, the selection of
the correct relation depends on the configuration of the element, i.e., on the location of
node 3. Since every element in the mesh could have a configuration that is different in
general from either of the two configurations shown in Figure 10.2, a more robust relation
is needed to determine the average rigid body rotation rate about the local x-axis. In most
typical grids, node 3 will be located somewhere between the two configurations shown in
Figure 10.2. Thus a linear interpolation between these two rigid body rotation rates was
devised using the distance 𝑥̂3 as the interpolant:
rig rig 𝑥̂3 rig 𝑥̂
𝜃𝑥̂ = 𝜃𝑥−left
̂ (1 − ̂
) + 𝜃𝑥−right ( 3 ). (10.15)
𝑥̂2 𝑥̂2
Substitution of Equations (10.13) and (10.14) into (10.15) and simplifying produces the
relations given previously as Equation (10.12).

LS-DYNA DEV 10/27/16 (r:8004) 9-5 (Triangular Shells)


Triangular Shells LS-DYNA Theory Manual

10.1.3 Stress Resultants and Nodal Forces


After suitable constitutive evaluation using the above velocity strains, the resulting
local stresses are integrated through the thickness of the shell to obtain local resultant
forces and moments. The integration formulae for the resultants are
𝑅̂ = ∫ 𝜎̂ 𝛼𝛽 𝑑𝑧̂,
𝑓𝛼𝛽 (10.16)

̂𝑅
𝑚 𝛼𝛽 = − ∫ 𝑧̂ 𝜎̂ 𝛼𝛽 𝑑𝑧̂, (10.17)

where the superscript 𝑅 indicates a resultant force or moment and the Greek subscripts
emphasize the limited range of the indices for plane stress plasticity.

The above element-centered force and moment resultant are related to the local
nodal forces and moments by invoking the principle of virtual power and performing a
one-point quadrature. The relations obtained in this manner are
⎧𝑓𝑥1̂ ⎫
{ }
{𝑓 ̂ }
{ 𝑦1 }
{ } 𝑅
⎧𝑓𝑥𝑥 ̂ ⎫
{
{𝑓𝑥2̂ }
} { }
{
T 𝑓 ̂
𝑅
}
⎨𝑓 ̂ ⎬ = 𝐴𝐁M⎨ 𝑦𝑦 ⎬, (10.18)
{ 𝑦2 } { 𝑅̂ }
{ }
{ } ⎩𝑓𝑥𝑦 ⎭
{𝑓 ̂ }
{ 𝑥3 }
{ }
{𝑓 ̂ }
⎩ 𝑦3 ⎭

⎧𝑚̂ 𝑥1 ⎫
{
{ }
{ ̂ 𝑦1 }
𝑚 }
{ } ⎧𝑚
{ ̂𝑅𝑥𝑥 ⎫
}
{𝑚̂ 𝑥2 } { 𝑅 }
̂𝑅
T 𝑚 ̂ T 𝑓𝑥𝑧
⎨𝑚̂ 𝑦2 ⎬ = 𝐴𝐁M ⎨ 𝑦𝑦 ⎬ + 𝐴𝐁S { ̂𝑅 }, (10.19)
{ } {
{𝑚 𝑅 }
} 𝑓 𝑦𝑧
{
{ } ⎩ ̂ 𝑥𝑦 ⎭
{ ̂ 𝑥3 }
𝑚 }
{𝑚 }
⎩ ̂ 𝑦3 ⎭
where 𝐴 is the area of the element (2𝐴 = 𝑥̂2 𝑦̂3 ).

The remaining nodal forces, the 𝑧̂ component of the force (𝑓𝑧3̂ , 𝑓𝑧2̂ , 𝑓𝑧1̂ ), are
determined by successively solving the following equilibration equations
𝑚
̂ 𝑥1 + 𝑚 ̂ 𝑥3 + 𝑦̂3 𝑓𝑧3̂ = 0,
̂ 𝑥2 + 𝑚 (10.20)

𝑚
̂ 𝑦1 + 𝑚 ̂ 𝑦3 − 𝑥̂3 𝑓𝑧3̂ − 𝑥̂2 𝑓𝑧2̂ = 0,
̂ 𝑦2 + 𝑚 (10.21)

𝑓𝑧1̂ + 𝑓𝑧2̂ + 𝑓𝑧3̂ = 0, (10.22)


which represent moment equilibrium about the local x-axis, moment equilibrium about the
local y-axis, and force equilibrium in the local z-direction, respectively.

9-6 (Triangular Shells) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Triangular Shells

10.2 Marchertas-Belytschko Triangular Shell


The Marchertas-Belytschko [1974] triangular shell element, or the BCIZ triangular
shell element as it is referred to in the LS-DYNA user’s manual, was developed in the same
time period as the Belytschko beam element [Belytschko, Schwer, and Klein, 1977], see
Section 4, forming the first generation of co-rotational structural elements developed by
Belytschko and co-workers. This triangular shell element became the first triangular shell
implemented in DYNA3D. Although the Marchertas-Belytschko shell element is relatively
expensive, i.e., the 𝐶0 triangular shell element with five through-the-thickness integration
points requires 649 mathematical operations compared to 1,417 operations for the
Marchertas-Belytschko triangular shell, it is maintained in LS-DYNA for compatibility with
earlier user models. However, as the LS-DYNA user community moves to application of
the more efficient shell element formulations, the use of the Marchertas-Belytschko
triangular shell element will decrease.

As mentioned above, the Marchertas-Belytschko triangular shell has a common co-


rotational formulation origin with the Belytschko beam element. The interested reader is
referred to the beam element description, see Section 4, for details on the co-rotational
formulation. In the next subsection a discussion of how the local element coordinate
system is identical for the triangular shell and beam elements. The remaining subsections
discuss the triangular element’s displacement interpolants, the strain displacement
relations, and calculations of the element nodal forces. In the report [1974], much greater
detail is provided.

10.2.1 Element Coordinates


Figure 10.3(a) shows the element coordinate system, (𝐱̂, 𝐲̂ , 𝐳̂) originating at Node 1,
for the Marchertas-Belytschko triangular shell. The element coordinate system is
associated with a triad of unit vectors (𝐞1 , 𝐞2 , 𝐞3 ) the components of which form a
transformation matrix between the global and local coordinate systems for vector
quantities. The nodal or body coordinate system unit vectors (𝐛1 , 𝐛2 , 𝐛3 ) are defined at
each node and are used to define the rotational deformations in the element, see Section
8.4.4.

The unit normal to the shell element 𝐞3 is formed from the vector cross product
𝐞3 = 𝐥21 × 𝐥31 , (10.23)
where 𝐥21 and 𝐥31 are unit vectors originating at Node 1 and pointing towards Nodes 2 and
3 respectively, see Figure 10.3(b).

Next a unit vector g, see Figure 10.3(b), is assumed to be in the plane of the
triangular element with its origin at Node 1 and forming an angle 𝛽 with the element side
between Nodes 1 and 2, i.e., the vector 𝑙21 . The direction cosines of this unit vector are
represented by the symbols (𝑔𝑥 , 𝑔𝑦 , 𝑔𝑧 ). Since g is the unit vector, its direction cosines will
satisfy the equation

LS-DYNA DEV 10/27/16 (r:8004) 9-7 (Triangular Shells)


Triangular Shells LS-DYNA Theory Manual

𝑔𝑥2 + 𝑔𝑦2 + 𝑔𝑧2 = 1. (10.24)

Also, since 𝐠 and 𝐞3 are orthogonal unit vectors, their vector dot product must
satisfy the equation
𝑒3𝑥 𝑔𝑥 + 𝑒3𝑦 𝑔𝑦 + 𝑒3𝑧 𝑔𝑧 = 0. (10.25)
In addition, the vector dot product of the co-planar unit vectors 𝐠 and 𝐥21 satisfies the
equation
𝐼21𝑥 𝑔𝑥 + 𝐼21𝑦𝑔𝑦 𝑔𝑦 + 𝐼21𝑧 𝑔𝑧 = cos𝛽, (10.26)
where (𝑙21𝑥 , 𝑙21𝑦 , 𝑙21𝑧 )are the direction cosines of 𝐥21 .

9-8 (Triangular Shells) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Triangular Shells

3
y^

x^
^
z e2

e3 e1
1 b2

b3 b1
2
(a) Element and body coordinates

3
y^

x^
^ I31
z e2
α
e3
α/2 β
1 b2
I21

b3 b1
2

(b) Construction of element coordinates


Figure 10.3. Construction of local element coordinate system.

Solving this system of three simultaneous equation, i.e., Equation (10.24), (10.25),
and (10.26), for the direction cosines of the unit vector g yields
𝑔𝑥 = 𝑙21𝑥 cos𝛽 + (𝑒3𝑦 𝑙21𝑧 − 𝑒3𝑧 𝑙21𝑦 )sin𝛽, (10.27)

𝑔𝑦 = 𝑙21𝑦 cos𝛽 + (𝑒3𝑧 𝑙21𝑥 − 𝑒3𝑥 𝑙21𝑧 )sin𝛽, (10.28)

LS-DYNA DEV 10/27/16 (r:8004) 9-9 (Triangular Shells)


Triangular Shells LS-DYNA Theory Manual

𝑔𝑧 = 𝑙21𝑧 cos𝛽 + (𝑒3𝑥 𝑙21𝑦 − 𝑒3𝑦 𝑙21𝑥 )sin𝛽. (10.29)


These equations provide the direction cosines for any vector in the plane of the triangular
element that is oriented at an angle 𝛽 from the element side between Nodes 1 and 2. Thus
the unit vector components of 𝐞1 and 𝐞2 are obtained by setting 𝛽 = 𝛼/2 and 𝛽 = (𝜋 + 𝛼)/2
in Equation (8.22), respectively. The angle 𝛼 is obtained from the vector dot product of the
unit vectors 𝐥21 and 𝐥31 ,
cos𝛼 = 𝐥21 ⋅ 𝐥31 . (10.30)

10.2.2 Displacement Interpolation


As with the other large displacement and small deformation co-rotational element
formulations, the nodal displacements are separated into rigid body and deformation
displacements,
𝐮 = 𝐮rigid + 𝐮def , (10.31)
where the rigid body displacements are defined by the motion of the local element
coordinate system, i.e., the co-rotational coordinates, and the deformation displacement are
defined with respect to the co-rotational coordinates. The deformation displacement are
defined by

⎧ 𝑢̂𝑥 ⎫
def 𝜙𝑚
𝑥
{
{ 𝑢̂ } } ⎡ 𝑚 ⎤ ⎧ 𝛿 ⎫
𝑦 = ⎢ 𝜙𝑦 ⎥ { }
(10.32)
⎨ − − −⎬ ⎢− − −⎥ ⎨− − −⎬,
{ } ⎢ {
⎥ ⎩ }
{
⎩ 𝑢̂𝑧 ⎭
} 𝑓 𝜃̂ ⎭
⎣ 𝜙𝑧 ⎦

𝛅T = {𝛿12 𝛿23 𝛿31 }, (10.33)


are the edge elongations and
̂
𝛉̂ = {𝜃1𝑥 ̂
𝜃1𝑦 ̂
𝜃2𝑥 ̂
𝜃2𝑦 ̂
𝜃3𝑥 ̂ },
𝜃3𝑦 (10.34)
are the local nodal rotation with respect to the co-rotational coordinates.

𝑓
The matrices 𝜙𝑚 𝑚
𝑥 , 𝜙𝑦 and 𝜙𝑧 are the membrane and flexural interpolation functions,
respectively. The element’s membrane deformation is defined in terms of the edge
elongations. Marchertas and Belytschko adapted this idea from Argyris et al., [1964],
where incremental displacements are used, by modifying the relations for total
displacements,
2(𝑥𝑗𝑖 𝑢𝑗𝑖𝑥 + 𝑦𝑗𝑖 𝑢𝑗𝑖𝑦 + 𝑧𝑗𝑖 𝑢𝑗𝑖𝑧 ) + 𝑢2𝑗𝑖𝑥 + 𝑢2𝑗𝑖𝑦 + 𝑢2𝑗𝑖𝑧
𝛿𝑖𝑗 = , (10.35)
𝑙 0𝑖𝑗 + 𝑙𝑖𝑗
where 𝑥𝑗𝑖 = 𝑥𝑗 − 𝑥𝑖 , etc.

The non-conforming shape functions used for interpolating the flexural


𝑓
deformations, 𝜙𝑧 were originally derived by Bazeley, Cheung, Irons, and Zienkiewicz
9-10 (Triangular Shells) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Triangular Shells

𝑓
[1965]; hence the LS-DYNA reference to the BCIZ element. Explicit expressions for 𝜙𝑧 are
quite tedious and are not given here. The interested reader is referred to Appendix G in
the original work of Marchertas and Belytschko [1974].

The local nodal rotations, which are interpolated by these flexural shape functions,
are defined in a manner similar to those used in the Belytschko beam element. The current
components of the original element normal are obtained from the relation
𝐞03 = 𝛍T 𝛌𝐞̅ 03 , (10.36)
where 𝛍 and 𝛌 are the current transformations between the global coordinate system and
the element (local) and body coordinate system, respectively. The vector 𝐞̅ 03 is the original
element unit normal expressed in the body coordinate system. The vector cross product
between this current-original unit normal and the current unit normal,
𝐞3 × 𝐞03 = 𝜃𝑥̂ 𝐞1 + 𝜃𝑦̂ 𝐞2 , (10.37)
define the local nodal rotations as
𝛉̂𝑥 = − 𝐞̂03𝑦 , (10.38)

𝛉̂𝑦 = 𝐞̂03𝑥 . (10.39)


Note that at each node the corresponding 𝛌 transformation matrix is used in Equation
(10.36).

10.2.3 Strain-Displacement Relations


Marchertas-Belytschko impose the usual Kirchhoff assumptions that normals to the
midplane of the element remain straight and normal, to obtain
∂𝑢𝑥 ∂2 𝑢
𝑒𝑥𝑥 = − 𝑧 2𝑧 , (10.40)
∂𝑥 ∂𝑥
∂𝑢𝑦 ∂ 2 𝑢𝑧
𝑒𝑦𝑦 = −𝑧 , (10.41)
∂𝑦 ∂𝑦2

∂𝑢𝑥 ∂𝑢𝑦 ∂ 2 𝑢𝑧
2𝑒𝑥𝑦 = + − 2𝑧 , (10.42)
∂𝑦 ∂𝑥 ∂𝑥 ∂𝑦
where it is understood that all quantities refer to the local element coordinate system.

Substitution of Equations (10.32) into the above strain-displacement relations yields


𝛆 = 𝐄m 𝜹 − 𝑧𝐄f 𝛉̂, (10.43)
where
𝛆 T = {𝜀𝑥𝑥 𝜀𝑦𝑦 2𝜀𝑥𝑦 }, (10.44)
with

LS-DYNA DEV 10/27/16 (r:8004) 9-11 (Triangular Shells)


Triangular Shells LS-DYNA Theory Manual

∂𝜙𝑚
𝑥𝑖
⎡ ⎤
⎢ ∂𝑥⎥
⎢ ∂𝜙𝑚⎥
⎢ ⎥
𝑦𝑖
𝐄m =⎢ ⎥, (10.45)
⎢ ∂𝑦 ⎥
⎢∂𝜙𝑚 ∂𝜙𝑚 ⎥
⎢ 𝑥𝑖 + 𝑦𝑖 ⎥
⎣ ∂𝑦 ∂𝑥 ⎦
and
2 𝑓
⎡ ∂ 𝜙𝑧𝑖 ⎤
⎢ ∂𝑥2 ⎥
⎢ 2 𝑓 ⎥
⎢∂ 𝜙 ⎥
𝐄f = ⎢ 𝑧𝑖 ⎥
. (10.46)
⎢ ∂𝑦2 ⎥
⎢ ⎥
⎢ ∂2 𝜙𝑓 ⎥
⎢2 𝑧𝑖 ⎥
⎣ ∂𝑥 ∂𝑦⎦
Again, the interested reader is referred to Appendices F and G in the original work of
Marchertas and Belytschko [1974] for explicit expressions of the above two matrices.

10.2.4 Nodal Force Calculations


The local element forces and moments are found by integrating the local element
stresses through the thickness of the element. The local nodal forces are given by
T
𝐟 ̂ = ∫ 𝐄m 𝛔
̂ 𝑑𝑉, (10.47)

𝐟T̂ = {𝑓12 , 𝑓23 , 𝑓31 }, (10.48)

̂T = {𝜎𝑥𝑥 , 𝜎𝑦𝑦 , 𝜎𝑥𝑦 },


𝛔 (10.49)
where the side forces and stresses are understood to all be in the local convected coordinate
system.

Similarly, the local moments are given by


T
̂ = − ∫ 𝑧 𝐄f 𝛔
𝐦 ̂ 𝑑𝑉, (10.50)

̂ T = {𝑚
𝐦 ̂ 1𝑥 𝑚
̂ 1𝑦 𝑚
̂ 2𝑥 𝑚
̂ 2𝑦 𝑚
̂ 3𝑥 𝑚
̂ 3𝑦 }. (10.51)
The through-the-thickness integration portions of the above local force and moment
integrals are usually performed with a 3- or 5-point trapezoidal integration. A three-point
inplane integration is also used; it is inpart this three-point inplane integration that
increases the operation count for this element over the 𝐶0 shell, which used one-point
inplane integration with hourglass stabilization.

9-12 (Triangular Shells) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Triangular Shells

The remaining transverse nodal forces are obtained from element equilibrium
considerations. Moment equilibrium requires
𝑓̂ 1 −𝑥̂3 𝑦̂3 𝑚
̂ 1𝑥 + 𝑚
̂ 2𝑥 + 𝑚
̂ 3𝑥
{ 2𝑧 } = [ ] {𝑚
̂ 1𝑦 + 𝑚 ̂ 3𝑦 },
̂ 2𝑦 + 𝑚 (10.52)
𝑓3𝑧̂ 2𝐴 𝑥̂2 − 𝑦̂2
where 𝐴 is the area of the element. Next transverse force equilibrium provides
𝑓1𝑧̂ = −𝑓2𝑧̂ − 𝑓3𝑧̂ . (10.53)

The corresponding global components of the nodal forces are obtained from the
following transformation
⎧𝑓𝑖𝑥 ⎫ 𝑓 ⎧ 𝑥 + 𝑢𝑖𝑗𝑥 ⎫ ⎧𝑥 +𝑢 𝑒
{ } 𝑖𝑗 { 𝑖𝑗 } 𝑓
𝑦𝑖𝑗 + 𝑢𝑖𝑗𝑦 + 𝑖𝑘 {𝑦𝑖𝑘 + 𝑢𝑖𝑘𝑥 ⎫
} ̂

{𝑒3𝑥 ⎫
}
⎨𝑓𝑖𝑦 ⎬ = 𝑙 ⎨ ⎬ 𝑙 ⎨ 𝑖𝑘 𝑖𝑘𝑦 ⎬ + 𝑓𝑖𝑧 ⎨ 3𝑦 ⎬.
{
(10.54)
{𝑓 }
⎩ 𝑖𝑧 ⎭ 𝑖𝑗 {
⎩𝑧𝑖𝑗 + 𝑢𝑖𝑗𝑧 }
⎭ 𝑖𝑘
{
⎩𝑧𝑖𝑘 + 𝑢𝑖𝑘𝑧 }
⎭ ⎩𝑒3𝑧 }

Finally, the local moments are transformed to the body coordinates using the relation
⎧ 𝑚
̅̅̅̅̅ ⎧𝑚̂ 𝑖𝑥 ⎫
{ 𝑖𝑥 ⎫ } T {𝑚 }
⎨ 𝑚
̅̅̅̅̅𝑖𝑦 ⎬ = 𝛌 𝛍 ⎨ ̂ 𝑖𝑦 ⎬. (10.55)
{ ̅̅̅̅̅𝑖𝑧 } {
⎩𝑚 ⎭ ⎩𝑚̂ 𝑖𝑧 }

LS-DYNA DEV 10/27/16 (r:8004) 9-13 (Triangular Shells)


LS-DYNA Theory Manual Fully Integrated Shell (Type 16)

11
Fully Integrated Shell (Type 16)

11.1 Introduction
Shell type 16 in LS-DYNA is a fully integrated shell with assumed strain interpolants
used to alleviate locking and enhance in-plane bending behavior, see Engelmann, Whirley,
and Goudreau [1989]; Simo and Hughes [1986]; Pian and Sumihara [1985]. It uses a local
element coordinate system that rotates with the material to account for rigid body motion
and automatically satisfies frame invariance of the constitutive relations. The local element
coordinate system is similar to the one used for the Belytschko-Tsay element, where the the
first two basis vectors are tangent to the shell midsurface at the center of the element, and
the third basis vector is in the normal direction of this surface and initially coincident with
the fiber vectors.

11.2 Hu-Washizu Three Field Principle


The element is derived starting from the Hu-Washizu three-field principle stated as
̅̅̅̅ , ̅𝛔
0 = 𝛿Π(𝐯, 𝐃 ̅̅̅̅ : 𝛔(𝐃
̅̅̅̅) = ∫ 𝛿𝐃 ̅̅̅̅ )𝑑Ω + ∫ 𝛿[𝛔 ̅̅̅̅ )]𝑑Ω − 𝛿𝑃ext + 𝛿𝑃kin ,
̅̅̅̅̅: (𝐃(𝐯) − 𝐃 (11.1)
Ω Ω
̅̅̅̅ is the assumed strain rate, ̅𝛔
where 𝐯 is the velocity, 𝐃 ̅̅̅̅ is the assumed stress, 𝛔 denotes the
constitutive update as a function of the assumed strain rate, and 𝐃 is the strain rate
computed from the velocity field. 𝛿𝑃kin and 𝛿𝑃ext are the virtual power contributions from
the inertial and external forces, respectively, and Ω denotes the domain of the shell
element. The contribution from the internal forces can be decomposed in the in-plane and
transverse shear parts as
𝑝 𝑠
0 = 𝛿𝑃int + 𝛿𝑃int + 𝐻 𝑝 + 𝐻 𝑠 − 𝛿𝑃ext + 𝛿𝑃kin , (11.2)
where

LS-DYNA DEV 10/27/16 (r:8004) 10-1 (Fully Integrated Shell (Type 16))
Fully Integrated Shell (Type 16) LS-DYNA Theory Manual

p ̅̅̅̅ p : 𝛔p (𝐃
̅̅̅̅ )𝑑Ω,
𝛿𝑃int = ∫ 𝛿𝐃 (11.3)
Ω

s ̅̅̅̅ s : 𝛔s (𝐃
̅̅̅̅ )𝑑Ω,
𝛿𝑃int = 𝜅 ∫ 𝛿𝐃 (11.4)
Ω

𝐻 p = ∫ 𝛿[𝛔
̅̅̅̅̅p : (𝐃p (𝐯) − 𝐃
̅̅̅̅ p )]𝑑Ω,
(11.5)
Ω

𝐻 s = 𝜅 ∫ 𝛿[𝛔
̅̅̅̅̅s : (𝐃s (𝐯) − 𝐃
̅̅̅̅ s )]𝑑Ω.
(11.6)
Ω

Here κ is the shear correction factor and the superscripts mean that only the in-plane
components (p) or transverse shear (s) components are treated.

11.3 In-plane Assumed Strain Field


Using the standard isoparametric interpolation for the four-node quadrilateral
element, the in-plane strain rate can be written
p
𝐃p = [𝐁m 𝑧𝐁b ] [𝐯p ], (11.7)
𝛉̇
where 𝐁m and 𝐁b are strain-displacement matrices for membrane and bending modes,
respectively, 𝑧 is the through thickness coordinate and 𝐯p and 𝛉̇p are the nodal (in-plane)
translational and rotational velocities, respectively.

To derive the in-plane assumed strain field, the interpolants for the assumed stress
and strain rates are chosen as
𝐬
̅̅̅̅̅p = [𝐒p 𝐒p ][𝐬m ],
𝛔 (11.8)
b

𝐞m
̅̅̅̅ p = 𝐂−1 [𝐒p
𝐃 𝐒p ][𝐞 ], (11.9)
b

where
1 0 0 𝑎21 𝜂̂ 𝑏21 𝜉 ̂
p
⎡ ⎤
𝐒 =⎢
⎢0 1 0 𝑎22 𝜂̂ 𝑏22 𝜉 ̂ ⎥
⎥, (11.10)
⎣0 0 1 𝑎1 𝑎2 𝜂̂ 𝑏1 𝑏2 𝜉 ⎦̂
and
1
𝜉̂ = 𝜉 − ∫ 𝜉𝑑Ω, (11.11)
|Ω|
Ω

10-2 (Fully Integrated Shell (Type 16)) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Fully Integrated Shell (Type 16)

1
𝜂̂ = 𝜂 − ∫ 𝜂𝑑Ω. (11.12)
|Ω|
Ω

Furthermore, 𝐂 is the plane stress constitutive matrix and 𝜉 and 𝜂 are the isoparametric
coordinates. The coefficients 𝑎𝑖 and 𝑏𝑖 are defined through
1 𝑎 𝑏1
𝐉0 = [ 1 ], (11.13)
4 𝑎2 𝑏2
where 𝐉0 is the area jacobian matrix from the isoparametric to physical domain computed
at the element center.

Inserting the expressions for the strain rate and assumed stress and strain rate into
the expression for 𝐻 p and requiring 𝐻 p = 0 for arbitrary 𝐬m , 𝐬b , 𝐞m and 𝐞b , yields the
following expression for the assumed strain rate in terms of the nodal velocities
p
̅̅̅̅ p = [̅𝐁
𝐃 ̅̅̅̅m ̅̅̅̅̅b ] [𝐯p ],
𝑧𝐁 (11.14)
𝛉̇
where
̅̅̅̅m = 𝐂−1 𝐒p 𝐄̂ 𝐁
̅𝐁 ̂m, (11.15)

̅̅̅̅b = 𝐂−1 𝐒p 𝐄̂ 𝐁
̅𝐁 ̂b, (11.16)
and
T
𝐄̂ = ∫ 𝐒p 𝐂−1 𝐒p 𝑑Ω, (11.17)
𝛺

̂ m = ∫ 𝐒p T 𝐁m 𝑑Ω,
𝐁 (11.18)
𝛺

̂ b = ∫ 𝐒p T 𝐁b 𝑑Ω.
𝐁 (11.19)
𝛺

11.4 Transverse Shear Assumed Strain Field


The transverse shear strain is the Bathe-Dvorkin [1984] assumed natural strain field
and is derived as follows. Using the standard isoparametric interpolation for the four-node
quadrilateral element, the transverse shear strain rate can be written
𝐯
𝐃s = 𝐁𝑡 [ ̇𝑝𝑧 ], (11.20)
𝛉
where 𝐁𝑡 is the corresponding strain-displacement matrix and 𝐯𝑧 and 𝛉̇𝑝 are the nodal out-
of-plane translational and in-plane rotational velocities, respectively.

The assumed strain rate is defined as


LS-DYNA DEV 10/27/16 (r:8004) 10-3 (Fully Integrated Shell (Type 16))
Fully Integrated Shell (Type 16) LS-DYNA Theory Manual

𝐯
𝐃 ̅̅̅̅𝑡 [ p𝑧 ],
̅̅̅̅ s = ̅𝐁 (11.21)
𝛉̇
where
̅̅̅̅̅𝑡 = 𝐉−T 𝐄 ∫ 𝐒𝑠 T 𝐉T 𝐁𝑡 𝑑Ω,
𝐁 (11.22)
Ω

Here 𝐉 is the area jacobian matrix from the isoparametric domain to the physical domain,
1 1−𝜉 1+𝜉 0 0
𝐄= [ ], (11.23)
2 0 0 1−𝜂 1+𝜂

δ(η)δ(1 + ξ) δ(η)δ(1 − ξ) 0 0
𝐒s = [ ], (11.24)
0 0 δ(ξ)δ(1 + η) δ(ξ)δ(1 − η)
and 𝛿 is the Dirac delta function. Defining the assumed stress as
̅̅̅̅̅p = 𝐉𝐒s 𝐬,
𝛔 (11.25)
yields 𝐻 s = 0 regardless of the choice of 𝐬 and thus a B-bar expression for the assumed
transverse strain rates is obtained as given above. The result is equivalent to defining the
isoparametric assumed shear strain rates by interpolating the corresponding strain rates
from the mid-side points A, B, C and D shown in Figure 11.1.

η
ξ B

Figure 11.1. Midside locations of isoparametric strain rates

11.5 Rigid Body Motion


For the in-plane assumed strain field, a rigid body motion may induce a nonzero
strain rate. The expression for the in-plane strain rate for a rigid body motion is

10-4 (Fully Integrated Shell (Type 16)) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Fully Integrated Shell (Type 16)

̅̅̅̅ r = ̅𝐁
𝐃 ̅̅̅̅r 𝛉̇, (11.26)
where
̅̅̅̅̅r = 𝑤𝐁
𝐁 ̅̅̅̅̅m 𝐑. (11.27)

11.6 Belytschko-Leviathan Projection


For warped configurations and since the geometry of the current shell element is
flat, extremely flexible behavior can be expected for some modes of deformation.
Following [Belytschko and Leviathan 1994], a 7-mode projection matrix 𝐏 (3 rigid body
rotation modes and 4 nodal drill rotation modes) is constructed used for projecting out
these zero energy modes. The explicit formula for the projection matrix is given by
𝐏 = 𝐈 − 𝐑(𝐑T 𝐑)−1 𝐑T , (11.28)
where 𝐑 is a matrix where each column corresponds to the nodal velocity of a zero energy
mode. This projection matrix operates on the nodal velocities prior to computing the strain
rates, and also on the resulting internal force vector to maintain invariance of the internal
power.

LS-DYNA DEV 10/27/16 (r:8004) 10-5 (Fully Integrated Shell (Type 16))
Shells with Thickness Stretch LS-DYNA Theory Manual

12
Shells with Thickness Stretch

12.1 Introduction
Thickness stretch is of considerable importance in problems involving finite thickness
strains, contact and surface loads in nonlinear shell applications. As an example, in sheet
metal forming applications, the presence of normal stresses in thickness direction improves
the accuracy of the solution and also its response on the double sided contact zone between
dies and sheet. It has also been shown that a kinematical representation of a continuous
thickness field improves the instability characteristics when compared to experimental
results, see Figure Figure 1210-2 and Björklund [2014].
There have been several attempts to account for the through-thickness deformation in the
literature, this implementation is inspired by the 7P-CYSE shell introduced in Cardoso and
Yoon [2005]. However, the formulation in this paper is rather complicated and involves for
instance assumed shear strains (ANS) to alleviate shear locking and complicated setup of
internal forces and stiffness matrices by combined analytical and numerical integration to
restore rank deficiencies. A direct implementation of this shell following the theory was
ruled out for efficiency reasons, and another approach was taken in order to presumably
get a useful shell.
The Belytschko-Tsay shell element is one of the fastest elements for thin shell simulations.
This, together with its robustness, is the reason why it is popular in finite element codes.
The implementation of shell type 25, the reduced integrated shell with thickness stretch, is
based on the formulation of the Belytschko-Tsay shell with a relaxation of the thickness
variable. This ensures that it will be efficient and hopefully also possess properties useful
for applications where through thickness deformation is important. As a fully integrated
alternative, shell type 26 is available as an analogue extension of shell type 16 (fully
integrated Belytschko-Tsay), and a triangular shell with thickness stretch is available as
type 27 mainly to allow for hybrid meshes (quadrilaterals combined with triangles) in this
context.
The theory that follows is very similar to that of the Belytschko-Tsay (type 2), Fully
Integrated Shell (type 16) and C0-shell (type 4), and here we emphasize on the parts
involving the amendments to those shell formulations.

9.6
LS-DYNA Theory Manual Shells with Thickness Stretch

25

20

Shell 16, no transverse shear

Force, F [kN]
15
stress and discontinuous
thickness field
10

5
Experiment
Shell 26, transverse shear stress Plane Stress
With Normal Stress
and continuous thickness field 0
0 0.5 1 1.5 2 2.5 3
Displacement, δ [mm]

Figure 1210-2 Example of premature localization with shell type 16


while shell type 26 matches experimental result (from Björklund

12.2 Shell Type 25


12.2.1 Formulation
The kinematics, i.e., position 𝒙 and velocity 𝒗, of a material point in the shell with through
thickness stretch can be written in the shell element local coordinate system as (sum over
nodal indices 𝐼)
𝒙 = (𝒙𝐼 + 𝑠𝐼 𝒏)𝑁𝐼 (𝜉 , 𝜂) (12.29)
𝒗 = (𝒗𝐼 + 𝑠𝐼 𝝎𝐼 × 𝒏 + 𝑠𝐼̇ 𝒏)𝑁𝐼 (𝜉 , 𝜂)
where we have set
𝜍 (12.30)
𝑠𝐼 = 𝑡𝐼 + (1 − 𝜍2 )𝑞𝐼 .
2
The kinematics is based on the Belytschko-Tsay shell with the additional feature that the
thickness is variable, whence the last term in the second of (12.29). The thickness variable is
represented by 𝑡𝐼 and an additional strain variable 𝑞𝐼 to allow for a linear strain through the
thickness. The latter represents the location of the mid-surface and is important to avoid
“Poisson locking” in bending modes of deformation, so the shell has two additional
degrees of freedom compared to the Belytschko-Tsay shell. The other variables and
parameters are the nodal coordinates 𝒙𝐼 , nodal velocities 𝒗𝐼 , nodal rotational velocities 𝝎𝐼 ,
shell normal 𝒏 = (0 0 1)𝑇 and bilinear isoparametric shape functions 𝑁𝐼 . From this we
can determine the local velocity gradient as
𝜕𝒗 𝜕𝑁 𝜕𝑠 𝜕𝑠 ̇ (12.31)
= (𝒗𝐼 + 𝑠𝐼 𝝎𝐼 × 𝒏 + 𝑠𝐼̇ 𝒏) 𝐼 + (𝝎𝐼 × 𝒏 𝐼 + 𝒏 𝐼 ) 𝑁𝐼 .
𝜕𝒙 𝜕𝒙 𝜕𝒙 𝜕𝒙

9.7
Shells with Thickness Stretch LS-DYNA Theory Manual

𝜕𝑁𝐼
In the local system one can assume a vanishing third component of both 𝝎𝐼 × 𝒏 and 𝜕𝒙
,
and the thickness strain rate is given by
𝜕𝑠 ̇ 𝑡 ̇ − 4𝜍𝑞 ̇ (12.32)
𝜀̇𝑡 = =
𝜕𝑥3 𝑡 − 4𝜍𝑞
where we used the notation 𝑠 = 𝑠𝐼 𝑁𝐼 , 𝑡 = 𝑡𝐼 𝑁𝐼 and 𝑞 = 𝑞𝐼 𝑁𝐼 , sum over 𝐼. For small strains
𝑞 = 0, and this shows that the thickness strain rate is at most linear.
For evaluating internal forces we define the strain-displacement tensor through (assuming
Voigt notation and sum over 𝐼)
𝜕𝒗 (12.33)
= 𝑩𝐼 𝒖𝐼 ,
𝜕𝒙
and the nodal vector 𝒖𝐼 is given by
(12.34)
𝒖𝐼 = (𝒗𝑇𝐼 𝝎𝐼𝑇 𝑡𝐼̇ 𝑞𝐼̇ )𝑇 ,
indicating the 8 degrees of freedom per node in these elements. The principle of virtual
work results in an internal force vector
(12.35)
𝒇𝐼 = ∫ 𝑩𝑇𝐼 𝝈 ,

where 𝝈 is the Cauchy stress and the integral is over the current element configuration.

12.2.2 Hourglass modes


Shell type 25 is a reduced integration element and in addition to the six hourglass modes
present in the original Belytschko-Tsay shell, two more are added by the introduction of
the thickness variables. Fortunately these are orthogonal to any rigid body motion and/or
any other hourglass mode, and are given by
𝑡 =ℎ (12.36)
𝐼 𝐼

𝑞𝐼 = ℎ𝐼
where
(12.37)
𝐼
ℎ𝐼 = (−1) /4

and all other displacement components are zero. To restrain these modes we have
included generalized strains and stresses according to the following (sum over 𝐼)
ℎ𝐼 𝑡𝐼̇ (12.38)
𝜀̇ℎ𝑡 =
𝑡
4ℎ 𝑞 ̇
𝜀̇ℎ𝑞 = − 𝐼 𝐼
𝑡
The corresponding generalized stresses are obtained through
(12.39)
𝜎̇ 𝑡 = 𝐸𝐻 𝜀̇ℎ𝑡

9.8
LS-DYNA Theory Manual Shells with Thickness Stretch

𝐸𝐻 ℎ
𝜎̇ 𝑞 = 𝜀̇
3 𝑞
and the forces are then given by
(12.40)
𝑓𝐼𝑡 = 𝐴ℎ𝐼 𝜎𝑡
𝑞
𝑓𝐼 = −4𝐴ℎ𝐼 𝜎𝑞

where 𝐴 is the area of the element and the value of 𝐸𝐻 is taken as 𝐸𝐻 = 0.05𝐸, i.e., 5% of the
Young’s modulus.

12.3 Shell Type 26


12.3.1 Modifying shell type 16
When using full integration, care must be taken in order to avoid the well-known shear
locking phenomenon. Common techniques developed for this are the ANS (Assumed
Natural Strain) and various types of EAS (Enhanced Assumed Strain) techniques. In
Cardoso and Yoon [2004] the ANS technique is used in which the transverse shear strains
are interpolated from the mid points of the shell element edges. It is reported that this is a
successful technique, but when consulting Bischoff and Ramm [1997] it is reported that it is
not well suited to avoid membrane locking and reduce mesh distortion sensitivity. They
are proposing a combination of ANS and EAS to get a decent shell element formulation.
However, elements that use an EAS in general require a nonlinear solution for the assumed
strain variables which can be computationally rather expensive. The approach taken here
is to suitably modifying an existing LS-DYNA standard assumed strain element, i.e.,
element type 16, which is a fully integrated extension of the Belytschko-Tsay element. Due
to the rather complex setup of assumed strain components, we use the following in order to
extend the element to variable thickness stretch without having to bother about all the
details of the kinematics.
In the beginning of a time step, the thickness components are reset to the corresponding
value in the center of the shell element. That is
𝑠 + 𝑠2 + 𝑠3 + 𝑠4 (12.41)
𝑠̃ = 1
4
is used for evaluating the strain increments. However, the rates of the thickness variables
are unchanged, so this can be seen as an assumed strain approach in the thickness
direction. Using this, the velocity gradient expression is an augmentation of that of
element 16
𝜕𝒗 𝜕𝒗 𝜕𝑁 𝜕𝑠 ̇ (12.42)
= ∣ + 𝑠𝐼̇ 𝒏 𝐼 + 𝒏 𝐼 𝑁𝐼 .
𝜕𝒙 𝜕𝒙 16 𝜕𝒙 𝜕𝒙
This should be interpreted that the velocity gradient used in shell element type 16 as a
function of the current thickness coordinate 𝑠 ̃ (and its isoparametric derivative) is
augmented by thickness component rates. Furthermore, to avoid locking phenomena and
9.9
Shells with Thickness Stretch LS-DYNA Theory Manual

still maintaining a non-singular element we use single point integration of the thickness
strain component and an ANS approach for the transverse shear strain components
emanating from the rates of thickness components. This approach is inspired by the
methodology used for the Belytschko-Tsay shell in which the nodal fiber vectors are reset
in the beginning of each time step, and in an analogous way we reset the thickness
components in the beginning of each time step in this shell element formulation.

12.4 Shell Type 27


12.4.1 Modifying shell type 4
A triangular element with thickness stretch was added to allow consistent sorting of
triangular shell elements, and not necessarily intended to be used for other purposes. The
approach taken is very similar to that of shell type 26, the existing LS-DYNA standard C0
triangular element was modified according to the same principles as described in the
previous section. In this case the velocity gradient expression is modified according to
𝜕𝒗 𝜕𝒗 𝜕𝑠 ̇̃ (12.43)
= ∣ +𝒏
𝜕𝒙 𝜕𝒙 4 𝜕𝒙
where again the subindex 4 indicates that it’s the velocity gradient of element type 4
evaluated with respect to the average thickness value 𝑠 ̃ and its isoparametric derivative.
For the triangular element type 27 the thickness is always constant in the element, a
decision that was taken for the sake of simplicity. More on thickness distribution in the
next section.

12.5 Related Features


12.5.1 Continuous vs decoupled thickness field
A drawback with having a continuous thickness field is that complicated geometries tend
to lock the structure. This approach assumes that the geometry is relatively flat and may be
suitable in metal forming but not in other situations. To remedy this we added the option
to decouple the thickness field so that the thickness is discontinuous between elements,
which makes the element suitable for crash analysis. To activate this option the user
should put the variable IDOF on *SECTION_SHELL equal to 2, which actually is the
default. For shell element type 25, this option will make the thickness variable constant in
the element, and the implementation follows the one described above with the restriction
𝑡𝐼 = 𝑡 (constant) and 𝑞𝐼 = 𝑞 (constant). No additional zero energy modes are present
with this approach.
Using the same approach for element type 26 turns out to be a poor choice since it leads to
locking phenomenon due to full integration. Instead we use the approach where the
thickness is bilinear within but discontinuous between elements, which turns out to be

9.10
LS-DYNA Theory Manual Shells with Thickness Stretch

successful. That means that the same code can be executed regardless of the choice of
IDOF for this element.

12.5.2 Nodal masses


For dynamic analysis some mass quantity must be associated with the extra degrees of
freedom, and although a consistent finite element approach is applicable the mass is here
empirically estimated from elaborating with the kinetic energy for compressive in-plane
and out-of-plane modes. That is, the kinetic energy of an in-plane uniaxial strain mode
should have the same kinetic energy as the equivalent out-of-plane ditto. This assumption
leads to a mass of the scalar nodes given by 𝑚 = 𝑚𝑇 /4 where 𝑚𝑇 is the translational mass.
A problem with this approach is that it leads to instabilities due to high eigenfrequencies of
the shell. For this reason we have set
3𝐴 (12.44)
𝑚 = max(1, )𝑚
2𝑡2 𝑇
where 𝐴 is the element area and 𝑡 is the thickness.
𝒇 𝒏

−𝒇

Figure 1210-3 Contact influence on thickness stretch shells.

12.5.3 Transfer of contact forces


The new degrees of freedom allows for a different treatment of how contact forces are
transferred from master to slave, for this discussion we refer to Figure Figure 1210-3. The
nodal forces in local system acting on the slave from contact on the upper and lower
surface can roughly be written
𝑇
𝒇𝐼𝑢/𝑙 = (± 𝒇
𝑇
𝑡 𝒇 𝑇 𝒏∙𝒇 (12.45)
(𝒏 × ) 0)
4 2 4 8
where we used the same order of the degrees of freedom as in (12.34). Summing these two
force contributions give nodal contact forces acting on the slave given by

9.11
Shells with Thickness Stretch LS-DYNA Theory Manual

𝑇
𝒇𝐼𝑐 = (𝟎 𝒇 𝑇 𝒏∙𝒇 (12.46)
𝑇
𝑡 (𝒏 × ) 0)
4 4

We can see that the total force include nodal moments caused by the frictional forces acting
at an offset from the midsurface of the shell, but also a pressure acting on the thickness
degree of freedom. In conclusion, the relaxation of the thickness in the Belytschko-Tsay
shell allows for double sided contact zones, i.e., the shell is affected by contact pressure
from both sides even though they are of equal magnitude. This is not possible in
traditional shell with zero normal stress, unless a modified option is used.

12.6 Contact Pressure Treatment in Shells 2, 4 and 16


By specifying IDOF = 3 on *SECTION_SHELL for the Belytschko-Tsay (type 2), C0-element
(type 4) and Fully Integrated (type 16) shell, the contact pressure influences the stress and
can induce thickness changes. This is a short explanation of the theory behind.
The z-stress in a shell element is usually restricted to be zero, but in this case we intend to
solve the constitutive update using the constraint
(12.47)
𝜎𝑧𝑧 = 𝛼𝜎𝑐 (𝑧)
where
𝜎𝑐𝑏 − 𝜎𝑐𝑡 3 𝜎 𝑏 + 𝜎𝑐𝑡 (12.48)
𝜎𝑐 (𝑧) = − (𝑧 − 3𝑧) − 𝑐
4 2
and
𝜎𝑐𝑏 = contact pressure at bottom surface of shell
𝜎𝑐𝑡 = contact pressure at top surface of shell
𝑧= isoparametric coordinate through the thickness between -1 and 1
𝛼= scaling parameter
The scaling parameter can be set as the 8th parameter on card 3 on *CONTROL_CONTACT.
The constitutive update for en elastic-plastic material can be written
(12.49)
𝛔𝑛+1 = 𝛔𝑛 + 𝐾∆𝜀𝑣𝑜𝑙 𝐈 + ∆𝐬(𝐬𝑛 , ∆𝛆𝑑𝑒𝑣 )

𝑛+1 (12.50)
𝜎𝑧𝑧 = 𝜎𝑐𝑛+1
where
𝛔𝑛+1 = stress in step n+1
𝛔𝑛 = stress in step n
𝐾= bulk modulus
∆𝛆 = strain increment
∆𝜀𝑣𝑜𝑙 = ∆𝛆: 𝐈 = volumetric strain increment
𝐈 = unit tensor
∆𝐬= deviatoric stress increment

9.12
LS-DYNA Theory Manual Shells with Thickness Stretch

𝑛
𝐬𝑛 = 𝛔𝑛 − 𝛔 3:𝐈 𝐈 = deviatoric stress in step n
∆𝛆𝑑𝑒𝑣 = ∆𝛆 − ∆𝛆:
3
𝐈
𝐈 = deviatoric strain increment
𝑛+1 𝑛+1 𝑛+1 𝑛+1 𝑛+1
The independent variables in (12.49) and (12.50) are 𝜎𝑥𝑥 , 𝜎𝑦𝑦 , 𝜎𝑥𝑦 , 𝜎𝑦𝑧 , 𝜎𝑥𝑧 and ∆𝜀𝑧𝑧 .
Here we assume that the stress response can be decoupled into a volumetric and deviatoric
part and the deviatoric stress increment depends only on the deviatoric part of the stress
and strain increment as indicated in the formula. We can rewrite (12.49) and (12.50) as
̃ 𝑛+1 = 𝛔
𝛔 ̃ 𝑛 + 𝐾∆𝜀̃𝑣𝑜𝑙 𝐈 + ∆𝐬(𝐬̃𝑛 , ∆𝛆̃𝑑𝑒𝑣 ) (12.51)
𝑛+1 (12.52)
𝜎̃ 𝑧𝑧 =0
by substituting
̃ 𝑛 = 𝛔𝑛 − 𝜎𝑐𝑛 𝐈
𝛔 (12.53)
(12.54)
̃ 𝑛+1 = 𝛔𝑛+1 − 𝜎𝑐𝑛+1 𝐈
𝛔
and
𝜎𝑐𝑛+1 − 𝜎𝑐𝑛 (12.55)
∆𝛆̃ = ∆𝛆 − 𝐈.
3𝐾
Since the deviatoric stress and strain increment is not changed, i.e.,
𝐬𝑛 = 𝐬̃𝑛 (12.56)
(12.57)
∆𝛆𝑑𝑒𝑣 = ∆𝛆̃𝑑𝑒𝑣
with this substitution it follows that the existing material routines can be used for solving
𝑛+1 𝑛+1 𝑛+1 𝑛+1 𝑛+1
(12.51) and (12.52) in terms of 𝜎̃ 𝑥𝑥 , 𝜎̃ 𝑦𝑦 , 𝜎̃ 𝑥𝑦 , 𝜎̃ 𝑦𝑧 , 𝜎̃ 𝑥𝑧 and ∆𝜀̃𝑧𝑧 and then use the inverse
of (12.54) and (12.55) to establish the stress and through thickness strain increment. Thus,
the algorithm is as follows
1.Given 𝛔𝑛 , ∆𝛆, 𝜎𝑐𝑛 and 𝜎𝑐𝑛+1
2.Use (12.53) and (12.55) to compute 𝛔 ̃ 𝑛 and ∆𝛆̃.
3.Do a constitutive update, (12.51) and (12.52), to get 𝛔 ̃ 𝑛+1 and ∆𝜀̃𝑧𝑧 .
4.Use (12.54) and (12.55) to compute 𝛔𝑛+1 and ∆𝜀𝑧𝑧 .

9.13
Shells with Thickness Stretch LS-DYNA Theory Manual

Figure 1210-4 Compression of a shell sheet between two rigid


plates using IDOF = 3 on shell 16.

9.14
LS-DYNA Theory Manual Hughes-Liu Shell

13
Hughes-Liu Shell

The Hughes-Liu shell element formulation ([Hughes and Liu 1981a, b], [Hughes et
al., 1981], [Hallquist et al., 1985]) was the first shell element implemented in LS-DYNA. It
was selected from among a substantial body of shell element literature because the element
formulation has several desirable qualities:
• it is incrementally objective (rigid body rotations do not generate strains), allowing
for the treatment of finite strains that occur in many practical applications;
• it is simple, which usually translates into computational efficiency and robustness;
• it is compatible with brick elements, because the element is based on a degenerated
brick element formulation. This compatibility allows many of the efficient and
effective techniques developed for the DYNA3D brick elements to be used with this
shell element;
• it includes finite transverse shear strains;
• a through-the-thickness thinning option (see [Hughes and Carnoy 1981]) is also
available when needed in some shell element applications.

The remainder of this section reviews the Hughes-Liu shell element (referred to by
Hughes and Liu as the U1 element) which is a four-node shell with uniformly reduced
integration, and summarizes the modifications to their theory as it is implemented in LS-
DYNA. A detailed discussion of these modifications, as well as those associated with the
implementation of the Hughes-Liu shell element in NIKE3D, are presented in an article by
Hallquist and Benson [1986].

LS-DYNA DEV 10/27/16 (r:8004) 11-1 (Hughes-Liu Shell)


Hughes-Liu Shell LS-DYNA Theory Manual

13.1 Geometry
The Hughes-Liu shell element is based on a degeneration of the standard 8-node
brick element formulation, an approach originated by Ahmad et al. [1970]. Recall from the
discussion of the solid elements the isoparametric mapping of the biunit cube:
𝐱(𝜉 , 𝜂, 𝜁 ) = 𝑁𝑎 (𝜉 , 𝜂, 𝜁 )𝐱𝑎 , (13.1)

(1 + 𝜉𝑎 𝜉 )(1 + 𝜂𝑎 𝜂)(1 + 𝜁𝑎 𝜁 )
𝑁𝑎 (𝜉 , 𝜂, 𝜁 ) = , (13.2)
8
where 𝐱 is an arbitrary point in the element, (𝜉 , 𝜂, 𝜁 ) are the parametric coordinates, 𝐱𝑎 are
the global nodal coordinates of node 𝑎, and 𝑁𝑎 are the element shape functions evaluated at
node 𝑎, i.e., (𝜉𝑎 , 𝜂𝑎 , 𝜁𝑎 ) are (𝜉 , 𝜂, 𝜁 ) evaluated at node 𝑎.

In the shell geometry, planes of constant 𝜁 will define the lamina or layers of the
shell and fibers are defined by through-the-thickness lines when both 𝜉 and 𝜂 are constant
(usually only defined at the nodes and thus referred to as ‘nodal fibers’). To degenerate the
8-node brick geometry into the 4-node shell geometry, the nodal pairs in the 𝜁 direction
(through the shell thickness) are combined into a single node, for the translation degrees of
freedom, and an inextensible nodal fiber for the rotational degrees of freedom. Figure 13.1
shows a schematic of the bi-unit cube and the shell element.

The mapping of the bi-unit cube into the shell element is separated into two parts
𝐱(𝜉 , 𝜂, 𝜁 ) = 𝐱̅(𝜉 , 𝜂) + 𝐗(𝜉 , 𝜂, 𝜁 ), (13.3)
where 𝐱̅ denotes a position vector to a point on the reference surface of the shell and X is a
position vector, based at point 𝐱̅ on the reference, that defines the fiber direction through
that point. In particular, if we consider one of the four nodes which define the reference
surface, then
𝐱̅ (𝜉 , 𝜂) = 𝑁𝑎 (𝜉 , 𝜂) 𝐱̅𝑎 , (13.4)

𝐗(𝜉 , 𝜂, 𝜁 ) = 𝑁𝑎 (𝜉 , 𝜂)𝐗𝑎 (𝜁 ). (13.5)

With this description, arbitrary points on the reference surface 𝐱̅ are interpolated by
the two-dimensional shape function 𝑁(𝜉 , 𝜂) operating on the global position of the four
shell nodes that define the reference surfaces, i.e., 𝐱̅𝑎 . Points off the reference surface are
further interpolated by using a one-dimensional shape function along the fiber direction,
i.e., 𝐗𝑎 (𝜁 ), where
̂𝑎 ,
𝐗𝑎 (𝜁 ) = 𝑧𝑎 (𝜁 ) 𝐗 (13.6)

𝑧𝑎 (𝜁 ) = 𝑁+ (𝜁 )𝑧+ −
𝑎 + 𝑁− (𝜁 )𝑧𝑎 , (13.7)

(1 + 𝜁 )
𝑁+ (𝜁 ) = , (13.8)
2

11-2 (Hughes-Liu Shell) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Hughes-Liu Shell

ζ
Biunit Cube
ζ
X
ξ
η

Beam Element ξ

Nodal fibers

Top Surface z+ζ


+1
x+ζ

ζ̄ x̄ζ
x^ζ

0 x^
η
xζ-
-1 -
Bottom Surface zζ
Figure 13.1. Mapping of the biunit cube into the Hughes-Liu shell element and
nodal fiber nomenclature.

(1 − 𝜁 )
𝑁− (𝜁 ) = (13.9)
2
As shown in the lower portion of Figure 13.1, 𝐗 ̂𝑎 is a unit vector in the fiber direction and
𝑧(𝜁 ) is a thickness function. (Thickness changes (see [Hughes and Carnoy 1981]) are
accounted for by explicitly adjusting the fiber lengths at the completion of a time step based
on the amount of straining in the fiber direction. Updates of the fiber lengths always lag
one time step behind other kinematical quantities.)

The reference surface may be located at the mid-surface of the shell or at either of the
shell’s outer surfaces. This capability is useful in several practical situations involving
contact surfaces, connection of shell elements to solid elements, and offsetting elements
such as stiffeners in stiffened shells. The reference surface is located within the shell
element by specifying the value of the parameter 𝜁 ̅ (see lower portion of Figure 13.1).

LS-DYNA DEV 10/27/16 (r:8004) 11-3 (Hughes-Liu Shell)


Hughes-Liu Shell LS-DYNA Theory Manual

When 𝜁 ̅ = – 1, 0, +1, the reference surface is located at the bottom, middle, and top surface

of the shell, respectively.

The Hughes-Liu formulation uses two position vectors, in addition to 𝜁 ,̅ to locate the
reference surface and define the initial fiber direction. The two position vectors 𝑥𝑎+ and 𝑥𝑎−
are located on the top and bottom surfaces, respectively, at node 𝑎. From these data the
following are obtained:
1
̅ 𝑎− + (1 + 𝜁 )𝑥
𝑥̅𝑎 = (1 − 𝜁 )𝑥 ̅ 𝑎+ , (13.10)
2

(𝑥𝑎+ − 𝑥𝑎− )
𝑋̂𝑎 = , (13.11)
ℎ𝑎

1
𝑧+
𝑎 =
̅ 𝑎,
(1 − 𝜁 )ℎ (13.12)
2
1
𝑧− ̅
𝑎 = − (1 + 𝜁 )ℎ𝑎 , (13.13)
2

ℎ𝑎 = ∥𝑥𝑎+ − 𝑥𝑎− ∥, (13.14)


where ‖ ⋅ ‖ is the Euclidean norm.

13.2 Kinematics
The same parametric representation used to describe the geometry of the shell
element, i.e., reference surface and fiber vector interpolation, are used to interpolate the
shell element displacement, i.e., an isoparametric representation. Again, the displacements
are separated into the reference surface displacements and rotations associated with the
fiber direction:
𝐮(𝜉 , 𝜂, 𝜁 ) = 𝐮
̅̅̅̅(𝜉 , 𝜂) + 𝐔(𝜉 , 𝜂, 𝜁 ), (13.15)

̅̅̅̅(𝜉 , 𝜂) = 𝑁𝑎 (𝜉 , 𝜂)𝐮
𝐮 ̅̅̅̅𝑎 , (13.16)

𝐔(𝜉 , 𝜂, 𝜁 ) = 𝑁𝑎 (𝜉 , 𝜂)𝐔𝑎 (𝜁 ), (13.17)

̂𝑎 ,
𝐔𝑎 (𝜁 ) = 𝑧𝑎 (𝜁 )𝐔 (13.18)
where 𝐮 is the displacement of a generic point; 𝐮 ̅̅̅̅ is the displacement of a point on the
reference surface, and 𝐔 is the ‘fiber displacement’ rotations; the motion of the fibers can be
interpreted as either displacements or rotations as will be discussed.

11-4 (Hughes-Liu Shell) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Hughes-Liu Shell

Hughes and Liu introduce the notation that follows, and the associated schematic
shown in Figure 13.2, to describe the current deformed configuration with respect to the
reference configuration:
𝐲=𝐲
̅̅̅̅ + 𝐘, (13.19)

𝐲
̅̅̅̅ = 𝐱̅ + 𝐮
̅̅̅̅, (13.20)

𝐲
̅̅̅̅𝑎 = 𝐱̅𝑎 + 𝐮
̅̅̅̅𝑎 , (13.21)

𝐘 = 𝐗 + 𝐔, (13.22)

𝐘𝑎 = 𝐗𝑎 + 𝐔𝑎 , (13.23)

̂𝑎 = 𝐗
𝐘 ̂𝑎 + 𝐔
̂𝑎 . (13.24)
In the above relations, and in Figure 13.2, the 𝐱 quantities refer to the reference
configuration, the 𝐲 quantities refer to the updated (deformed) configuration and the 𝐮
quantities are the displacements. The notation consistently uses a superscript bar (⋅ ̅) to
indicate reference surface quantities, a superscript caret (⋅ ̂) to indicate unit vector
quantities, lower case letters for translational displacements, and upper case letters
indicating fiber displacements. To update to the deformed configuration, two vector
quantities are needed: the reference surface displacement 𝐮
̅̅̅̅ and the associated nodal fiber
displacement 𝐔. The nodal fiber displacements are defined in the fiber coordinate system,
described in the next subsection.

13.2.1 Fiber Coordinate System


For a shell element with four nodes, the known quantities will be the displacements
of the reference surface 𝐮̅̅̅̅ obtained from the translational equations of motion and some
rotational quantities at each node obtained from the rotational equations of motion. To
complete the kinematics, we now need a relation between nodal rotations and fiber
displacements 𝐔.

LS-DYNA DEV 10/27/16 (r:8004) 11-5 (Hughes-Liu Shell)


Hughes-Liu Shell LS-DYNA Theory Manual

u U

X (parallel construction)

x
Y

reference axis in
undeformed
x̄ geometry
Deformed Configuration
Reference Surface
Figure 13.2. Schematic of deformed configuration displacements and position
vectors.

At each node a unique local Cartesian coordinate system is constructed that is used
as the reference frame for the rotation increments. The relation presented by Hughes and
Liu for the nodal fiber displacements (rotations) is an incremental relation, i.e., it relates the
current configuration to the last state, not to the initial configuration. Figure 13.3 shows
𝑓 𝑓 𝑓
two triads of unit vectors: (𝐛1 , 𝐛2 , 𝐛3 ) comprising the orthonormal fiber basis in the
reference configuration (where the fiber unit vector is now 𝐘 ̂ = 𝐛𝑓 ) and (𝐛1 , 𝐛2 , 𝐛3 )
3
indicating the incrementally updated current configuration of the fiber vectors. The
reference triad is updated by applying the incremental rotations, Δ𝜃1 and Δ𝜃2 , obtained
𝑓 𝑓
from the rotational equations of motion, to the fiber vectors (𝐛 1 and 𝐛 2 ) as shown in
Figure 13.3. The linearized relationship between the components of Δ𝑈 ̂ in the fiber system
viz, Δ𝑈̂ 𝑓 , Δ𝑈
̂ 𝑓 , Δ𝑈
̂ 𝑓 , and the incremental rotations is given by
1 2 3

̂𝑓 ⎫
⎧Δ𝑈
{
{ 1𝑓 }
} −1 0
⎡ Δ𝜃1
̂
Δ𝑈
⎨ 2⎬ = ⎢ 0 −1⎤⎥ {Δ𝜃 }. (13.25)
{
{Δ𝑈 }
𝑓} ⎣ 0 0⎦ 2
⎩ ̂ ⎭3

Although the above Hughes-Liu relation for updating the fiber vector enables a
reduction in the number of nodal degrees of freedom from six to five, it is not implemented
in LS-DYNA because it is not applicable to beam elements.

11-6 (Hughes-Liu Shell) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Hughes-Liu Shell

fiber

^
b3f =Y
b3

b1f b2
Δθ2
Δθ1

b1 b2f

Figure 13.3. Incremental update of fiber vectors using Hughes-Liu incremental


rotations.

In LS-DYNA, three rotational increments are used, defined with reference to the
global coordinate axes:
⎧ ̂
{Δ𝑈1 ⎫
} 0 𝑌̂3 −𝑌̂2 ⎧Δ𝜃1 ⎫
⎡ ⎤ { }
⎨ Δ𝑈
{ ̂ ⎬} ⎢−𝑌̂3
̂2 = ⎢ 0 𝑌̂1 ⎥
⎥ ⎨
{
Δ𝜃2 ⎬.
}
(13.26)
⎩Δ𝑈3 ⎭ ⎣ 𝑌̂2 −𝑌̂1 0 ⎦ ⎩Δ𝜃3 ⎭
Equation (13.26) is adequate for updating the stiffness matrix, but for finite rotations
the error is significant. A more accurate second-order technique is used in LS-DYNA for
updating the unit fiber vectors:
𝑌̂𝑖𝑛+1 = 𝑅𝑖𝑗 (Δ𝜃)𝑌̂𝑖𝑛 , (13.27)

1 (2𝛿𝑖𝑗 + Δ𝑆𝑖𝑘 )Δ𝑆𝑖𝑘 (13.28)


𝑅𝑖𝑗 (Δ𝜃) = 𝛿𝑖𝑗 + ,
2 𝐷
ΔS𝑖𝑗 = e𝑖𝑘𝑗 Δ𝜃𝑘 , (13.29)

1
2𝐷 = 2 + (Δ𝜃12 + Δ𝜃22 + Δ𝜃32 ). (13.30)
2
Here, 𝛿𝑖𝑗 is the Kronecker delta and 𝑒𝑖𝑗𝑘 is the permutation tensor. This rotational update is
often referred to as the Hughes-Winget formula [Hughes and Winget 1980]. An exact rota-
tional update using Euler angles or Euler parameters could easily be substituted in
Equation (13.27), but it is doubtful that the extra effort would be justified.

13.2.2 Lamina Coordinate System


In addition to the above described fiber coordinate system, a local lamina coordinate
system is needed to enforce the zero normal stress condition, i.e., plane stress. Lamina are
LS-DYNA DEV 10/27/16 (r:8004) 11-7 (Hughes-Liu Shell)
Hughes-Liu Shell LS-DYNA Theory Manual

η = con η
stant

nt
sta
e^3

con
e^2

ξ=
ξ
e^1

Figure 13.4. Schematic of lamina coordinate unit vectors.

layers through the thickness of the shell that correspond to the locations and associated
thicknesses of the through-the-thickness shell integration points; the analogy is that of
lamina in a fibrous composite material. The orthonormal lamina basis (Figure 13.4), with
one direction 𝑒3̂ normal to the lamina of the shell, is constructed at every integration point
in the shell.

The lamina basis is constructed by forming two unit vectors locally tangent to the
lamina:
𝐲,𝜉
𝐞̂1 = , (13.31)
∥𝐲,𝜉 ∥

𝐲,𝜂
𝐞′2 = , (13.32)
∥𝐲,𝜂 ∥
where, as before, 𝐲 is the position vector in the current configuration. The normal to the
lamina at the integration point is constructed from the vector cross product of these local
tangents:
𝐞̂3 = 𝐞̂1 × 𝐞′2 , (13.33)

𝐞̂2 = 𝐞̂3 × 𝐞̂1 , (13.34)


is defined, because 𝐞̂2 , although tangent to both the lamina and lines of constant 𝜉 , may not
be normal to 𝐞̂1 and 𝐞̂3 . The lamina coordinate system rotates rigidly with the element.

The transformation of vectors from the global to lamina coordinate system can now
be defined in terms of the lamina basis vectors as
⎧𝐴𝑥̂ ⎫ 𝑒1𝑥 𝑒2𝑥 𝑒3𝑥 T ⎧𝐴𝑥 ⎫
{ }
{ } { }
̂ = 𝐴𝑦̂ = ⎡
𝐀 𝑒 𝑒 𝑒 ⎤ 𝐴𝑦 = 𝐪𝐀, (13.35)
⎨ ⎬ ⎢ 1𝑦 2𝑦 3𝑦 ⎥ ⎨ ⎬
{
{𝐴̂ }} ⎣𝑒1𝑧 𝑒2𝑧 𝑒3𝑧 ⎦ {
⎩ 𝐴 }

⎩ 𝑧⎭ 𝑧

̂ is a
where 𝑒𝑖𝑥 , 𝑒𝑖𝑦 , 𝑒𝑖𝑧 are the global components of the lamina coordinate unit vectors; 𝐀
vector in the lamina coordinates, and 𝐴 is the same vector in the global coordinate system.

11-8 (Hughes-Liu Shell) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Hughes-Liu Shell

13.3 Strains and Stress Update


13.3.1 Incremental Strain and Spin Tensors
The strain and spin increments are calculated from the incremental displacement
gradient
∂Δ𝑢𝑖
𝐺𝑖𝑗 = , (13.36)
∂𝑦𝑗
where Δ𝑢𝑖 are the incremental displacements and 𝑦𝑗 are the deformed coordinates. The
incremental strain and spin tensors are defined as the symmetric and skew-symmetric
parts, respectively, of 𝐺𝑖𝑗 :
1
Δ𝜀𝑖𝑗 = (𝐺𝑖𝑗 + 𝐺𝑗𝑖 ), (13.37)
2
1
Δ𝜔𝑖𝑗 = (𝐺𝑖𝑗 − 𝐺𝑗𝑖 ). (13.38)
2

The incremental spin tensor Δ𝜔𝑖𝑗 is used as an approximation to the rotational


contribution of the Jaumann rate of the stress tensor; LS-DYNA implicit uses the more
accurate Hughes-Winget transformation matrix (Equation (13.27)) with the incremental
spin tensor for the rotational update. The Jaumann rate update is approximated as:
𝜎𝑖𝑗𝑛+1 = 𝜎𝑖𝑗𝑛 + 𝜎𝑖𝑝
𝑛 𝑛
Δ𝜔𝑝𝑗 + 𝜎𝑗𝑝 Δ𝜔𝑝𝑖 , (13.39)
where the superscripts on the stress refer to the updated (𝑛 + 1) and reference (𝑛)
configurations. The Jaumann rate update of the stress tensor is applied in the global
configuration before the constitutive evaluation is performed. In the Hughes-Liu shell the
stresses and history variables are stored in the global coordinate system.

13.3.2 Stress Update


To evaluate the constitutive relation, the stresses and strain increments are rotated
from the global to the lamina coordinate system using the transformation defined
previously in Equation (13.35), viz.
𝑛+1
𝑛+1
𝜎𝑖𝑗𝑙 = 𝑞𝑖𝑘 𝜎𝑘𝑛 𝑞𝑗𝑛 , (13.40)

𝑛+1⁄2 𝑛+1⁄2
Δ𝜀𝑖𝑗𝑙 = 𝑞𝑖𝑘 Δ𝜀𝑘𝑛 𝑞𝑗𝑛 , (13.41)

where the superscript 𝑙 indicates components in the lamina (local) coordinate system.

The stress is updated incrementally:


LS-DYNA DEV 10/27/16 (r:8004) 11-9 (Hughes-Liu Shell)
Hughes-Liu Shell LS-DYNA Theory Manual

𝑛+1 𝑛+1 𝑛+1⁄2


𝜎𝑖𝑗𝑙 = 𝜎𝑖𝑗𝑙 + Δ𝜎𝑖𝑗𝑙 , (13.42)

and rotated back to the global system:


𝑛+1
𝑙
𝜎𝑖𝑗𝑛+1 = 𝑞𝑘𝑖 𝜎𝑘𝑛 𝑞𝑛𝑗 , (13.43)
before computing the internal force vector.

13.3.3 Incremental Strain-Displacement Relations


The global stresses are now used to update the internal force vector

𝐟𝑎int = ∫ 𝐓𝑎T 𝐁T𝑎 𝛔𝑑𝜐, (13.44)

where 𝐟𝑎int are the internal forces at node 𝑎, 𝐁𝑎 is the strain-displacement matrix in the
lamina coordinate system associated with the displacements at node 𝑎, and 𝐓𝑎 is the
transformation matrix relating the global and lamina components of the strain-
displacement matrix. Because the B matrix relates six strain components to twenty-four
displacements (six degrees of freedom at four nodes), it is convenient to partition the B
matrix into four groups of six:
𝐁 = [𝐁1 𝐁2 𝐁3 𝐁4 ], (13.45)
Each 𝐁𝑎 submatrix is further partitioned into a portion due to strain and spin:
𝐁𝜀𝑎
𝐁𝑎 = [ ], (13.46)
𝐁𝜔𝑎

𝐵 0 0 𝐵4 0 0
⎡0 1 𝐵2 0 0 𝐵5 0 ⎤
⎢ ⎥
𝐁𝑎 = ⎢
𝜀
⎢𝐵̅̅̅̅ 2 ̅̅̅̅ 1
𝐵 0 ̅̅̅̅ 5
𝐵 ̅̅̅̅ 4
𝐵 0 ⎥ ⎥, (13.47)
⎢0 ̅̅̅̅ 3
𝐵 ̅̅̅̅ 2
𝐵 0 ̅̅̅̅ 6
𝐵 ̅̅̅̅ 5 ⎥
𝐵
⎢ ⎥
⎣𝐵̅̅̅̅ 3 0 ̅̅̅̅ 1
𝐵 ̅̅̅̅ 6
𝐵 0 ̅̅̅̅ 4 ⎦
𝐵

̅̅̅̅
𝐵 −𝐵 ̅̅̅̅ 1 0 ̅̅̅̅ 5
𝐵 −𝐵 ̅̅̅̅ 4 0
⎡ 2 ⎤
𝐁𝜔
𝑎 =⎢
⎢ 0 𝐵̅̅̅̅ 3 −𝐵 ̅̅̅̅ 2 0 ̅̅̅̅ 6
𝐵 −𝐵 ̅̅̅̅ 5 ⎥
⎥, (13.48)
⎣−𝐵 ̅̅̅̅ 3 0 ̅̅̅̅ 1
𝐵 −𝐵 ̅̅̅̅ 6 0 ̅̅̅̅ 4 ⎦
𝐵
where
⎧ ∂𝑁𝑎
{
{ 𝑁𝑎,𝑖 = for 𝑖 = 1, 2, 3
{ ∂𝑦𝑖𝑙
𝐵𝑖 = ⎨ . (13.49)
{ ∂(𝑁𝑎 𝑧𝑎 )
{
{(𝑁𝑎 𝑧𝑎 ),𝑖−3 = ∂𝑦𝑙 for 𝑖 = 4, 5, 6
⎩ 𝑖−3

Notes on strain-displacement relations:


• The derivatives of the shape functions are taken with respect to the lamina
coordinate system, e.g.,𝑦 = 𝑞𝑦.

11-10 (Hughes-Liu Shell) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Hughes-Liu Shell

• The superscript bar indicates the 𝐵’s are evaluated at the center of the lamina (0, 0,
𝜁 ). The strain-displacement matrix uses the ‘B-Bar’ (𝐵 ̅̅̅̅ )approach advocated by
Hughes [1980]. In the NIKE3D and DYNA3D implementations, this entails replac-
ing certain rows of the B matrix and the strain increments with their counterparts
evaluated at the center of the element. In particular, the strain-displacement matrix
is modified to produce constant shear and spin increments throughout the lamina.
• The resulting B-matrix is a 8 × 24 matrix. Although there are six strain and three
rotations increments, the B matrix has been modified to account for the fact that 𝜎33
will be zero in the integration of Equation (13.44).

13.4 Element Mass Matrix


Hughes, Liu, and Levit [Hughes et al., 1981] describe the procedure used to form the
shell element mass matrix in problems involving explicit transient dynamics. Their
procedure, which scales the rotary mass terms, is used for all shell elements in LS-DYNA
including those formulated by Belytschko and his co-workers. This scaling permits large
critical time step sizes without loss of stability.

The consistent mass matrix is defined by

𝐌 = ∫ 𝜌𝐍T 𝐍 𝑑𝜐𝑚 , (13.50)


𝜐𝑚

but cannot be used effectively in explicit calculations where matrix inversions are not
feasible. In LS-DYNA only three and four-node shell elements are used with linear
interpolation functions; consequently, we compute the translational masses from the
consistent mass matrix by row summing, leading to the following mass at element node a:

𝑀disp𝑎 = ∫ 𝜌𝜙𝑎 𝑑𝜐. (13.51)


𝑣

The rotational masses are computed by scaling the translational mass at the node by the
factor 𝛼:
𝑀rot𝑎 = ∝ 𝑀disp𝑎 , (13.52)

∝ = max{∝1 , ∝2 }, (13.53)

1
∝1 = ⟨𝑧𝑎 ⟩2 + [𝑧𝑎 ]2 , (13.54)
12
𝑉
∝2 = , (13.55)
8ℎ

(𝑧+ −
𝑎 + 𝑧𝑎 )
⟨𝑧𝑎 ⟩ = , (13.56)
2

LS-DYNA DEV 10/27/16 (r:8004) 11-11 (Hughes-Liu Shell)


Hughes-Liu Shell LS-DYNA Theory Manual

[𝑧𝑎 ] = 𝑧+ −
𝑎 − 𝑧𝑎 . (13.57)

13.5 Accounting for Thickness Changes


Hughes and Carnoy [1981] describe the procedure used to update the shell thickness
due to large membrane stretching. Their procedure with any necessary modifications is
used across all shell element types in LS-DYNA. One key to updating the thickness is an
accurate calculation of the normal strain component Δ𝜀33 . This strain component is easily
obtained for elastic materials but can require an iterative algorithm for nonlinear material
behavior. In LS-DYNA we therefore default to an iterative plasticity update to accurately
determine Δ𝜀33 .

Hughes and Carnoy integrate the strain tensor through the thickness of the shell in
order to determine a mean value Δ𝜀̅𝑖𝑗 :
1 1
Δ𝜀̅𝑖𝑗 = ∫ Δ𝜀𝑖𝑗 𝑑𝜁 , (13.58)
2 −1
and then project it to determine the straining in the fiber direction:
𝛆̅ 𝑓 = 𝐘
̂ T Δ𝛆̅𝐘
̂. (13.59)
Using the interpolation functions through the integration points the strains in the fiber
directions are extrapolated to the nodal points if 2 × 2 selectively reduced integration is
employed. The nodal fiber lengths can now be updated:
𝑓
ℎ𝑛+1
𝑎 = ℎ𝑛𝑎 (1 + 𝜀̅𝑎 ). (13.60)

13.6 Fully Integrated Hughes-Liu Shells


It is well known that one-point integration results in zero energy modes that must be
resisted. The four-node under integrated shell with six degrees of freedom per node has
nine zero energy modes, six rigid body modes, and four unconstrained drilling degrees of
freedom. Deformations in the zero energy modes are always troublesome but usually not a
serious problem except in regions where boundary conditions such as point loads are
active. In areas where the zero energy modes are a problem, it is highly desirable to
provide the option of using the original formulation of Hughes-Liu with selectively
reduced integration.

The major disadvantages of full integration are two-fold:

1. nearly four times as much data must be stored;

11-12 (Hughes-Liu Shell) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Hughes-Liu Shell

Figure 13.5. Selectively reduced integration rule results in four inplane points
being used.

2. the operation count increases three- to fourfold. The level 3 loop is added as
shown in Figure 13.6
However, these disadvantages can be more than offset by the increased reliability and
accuracy.

We have implemented two version of the Hughes-Liu shell with selectively reduced
integration. The first closely follows the intent of the original paper, and therefore no
assumptions are made to reduce costs, which are outlined in operation counts in Table 10.1.
These operation counts can be compared with those in Table 10.2 for the Hughes-Liu shell
with uniformly reduced integration. The second formulation, which reduces the number of
operation by more than a factor of two, is referred to as the co-rotational Hughes-Liu shell
in the LS-DYNA user’s manual. This shell is considerably cheaper due to the following
simplifications:
• Strains rates are not centered. The strain displacement matrix is only computed at
time 𝑛 + 1 and not at time 𝑛 + 1 ⁄ 2.
• The stresses are stored in the local shell system following the Belytschko-Tsay shell.
The transformations of the stresses between the local and global coordinate systems
are thus avoided.
• The Jaumann rate rotation is not performed, thereby avoiding even more computa-
tions. This does not necessarily preclude the use of the shell in large deformations.
• To study the effects of these simplifying assumptions, we can compare results with
those obtained with the full Hughes-Liu shell. Thus far, we have been able to get
comparable results.

LS-DYNA DEV 10/27/16 (r:8004) 11-13 (Hughes-Liu Shell)


Hughes-Liu Shell LS-DYNA Theory Manual

LEVEL L1 - Do over each element group


gather data, midstep geometry calculation

LEVEL 2 - For each thickness integration point


center of element calculations for selective
reduced integration

LEVEL 3 - Do over 4 Gauss points


stress update and force
contributions

LEVEL 2 - Completion

LEVEL L1 - Completion
Figure 13.6. An inner loop, LEVEL 3, is added for the Hughes-Liu shell with
selectively reduced integration.

LEVEL L1 - Once per element


Midstep translation geometry, etc. 204
Midstep calculation of 𝑌̂ 318
LEVEL L2 - For each integration point through thickness (NT points)
Strain increment at (0, 0, 𝜁 ) 316
Hughes-Winget rotation matrix 33
Square root of Hughes-Winget matrix 47
Rotate strain increments into lamina coordinates 66
Calculate rows 3-8 of B matrix 919
LEVEL L3 - For each integration point in lamina
Rotate stress to n+1/2 configuration 75
Incremental displacement gradient matrix 370
Rotate stress to lamina system 75
Rotate strain increments to lamina system 55
Constitutive model model dependent
Rotate stress back to global system 69
Rotate stress to n+1 configuration 75
Calculate rows 1 and 2 of B matrix 358
Stresses in n+1 lamina system 75
Stress divergence 245

11-14 (Hughes-Liu Shell) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Hughes-Liu Shell

TOTAL 522 +NT {1381 +4 * 1397}

Table 10.1. Operation counts for the Hughes-Liu shell with selectively reduced
integration.

LEVEL L1 - Once per element


Calculate displacement increments 24
Element areas for time step 53
̂
Calculate 𝑌 238
LEVEL L2 and L3 - Integration point through thickness (NT points)
Incremental displacement gradient matrix 284
Jaumann rotation for stress33
Rotate stress into lamina coordinates 75
Rotate stain increments into lamina coordinates 81
Constitutive model model dependent
Rotate stress to n+1 global coordinates 69
Stress divergence 125
LEVEL L1 - Cleanup
Finish stress divergence 60
Hourglass control 356

TOTAL 731 +NT * 667

Table 10.2. Operation counts for the LS-DYNA implementation of the uniformly
reduced
Hughes-Liu shell.

LS-DYNA DEV 10/27/16 (r:8004) 11-15 (Hughes-Liu Shell)


LS-DYNA Theory Manual Transverse Shear Treatment For Layered Shell

14
Transverse Shear Treatment For
Layered Shell

The shell element formulations that include the transverse shear strain components
are based on the first order shear deformation theory, which yield constant through
thickness transverse shear strains. This violates the condition of zero traction on the top
and bottom surfaces of the shell. Normally, this is corrected by the use of a shear correction
factor. The shear correction factor is 5/6 for isotropic materials; however, this value is
incorrect for sandwich and laminated shells. Not accounting for the correct transverse
shear strain and stress could yield a very stiff behavior in sandwich and laminated shells.
This problem is addressed here by the use of the equilibrium equations without gradient in
the y-direction as described by what follows. Consider the stresses in a layered shell:
(𝑖) ∘ (𝑖) ∘ (𝑖) ∘ (𝑖) ∘ (𝑖) (𝑖)
𝜎𝑥 (𝑖) = 𝐶11 (𝜀𝑥 + 𝑧𝜒𝑥 ) + 𝐶12 (𝜀𝑦 + 𝑧𝜒𝑦 ) = 𝐶11 𝜀𝑥 + 𝐶12 𝜀𝑦 + 𝑧(𝐶11 𝜒𝑥 + 𝐶12 𝜒𝑦 ),
(𝑖) ∘ (𝑖) ∘ (𝑖) (𝑖)
𝜎𝑦(𝑖) = 𝐶12 𝜀𝑥 + 𝐶22 𝜀𝑦 + 𝑧(𝐶12 𝜒𝑥 + 𝐶22 𝜒𝑦 , (14.1)
(𝑖) (𝑖) ∘
𝜏𝑥𝑦 = 𝐶44 (𝜀𝑥𝑦 + 𝑧𝜒𝑥𝑦 ).
Assume that the bending center 𝑧̅𝑥 is known. Then
(𝑖) (𝑖) (𝑖) (𝑖)
𝜎𝑥(𝑖) = (𝑧 − 𝑧̅𝑥 )(𝐶11 𝜒𝜒 + 𝐶12 𝜒𝑦 ) + 𝐶11 𝜀𝑥 (𝑧̅𝑥 ) + 𝐶12 𝜀𝑦 (𝑧̅𝑥 ). (14.2)
The bending moment is given by the following equation:

𝑁𝐿 𝑧𝑖 𝑁𝐿 𝑧𝑖
𝑀𝑥𝑥 = 𝜒𝑥 ⎛
⎜∑ 𝐶11(𝑖)
∫ (𝑧 − 𝑧̅𝑥 )2 𝑑𝑧⎞
⎟ + 𝜒𝑦 ⎛
⎜∑ 𝐶12(𝑖)
∫ (𝑧 − 𝑧̅𝑥 )2 𝑑𝑧⎞
⎟ (14.3)
⎝ 𝑖=1 𝑧𝑖−1 ⎠ ⎝ 𝑖=1 𝑧𝑖−1 ⎠

or
1 𝑁𝐿 (𝑖)
𝑀𝑥𝑥 = [𝜒𝑥 ∑ 𝐶11 [(𝑧3𝑖 − 𝑧3𝑖−1 ) − (𝑧𝑖 − 𝑧𝑖−1 )𝑧̅2𝑥 ] (14.4)
3 𝑖=1
𝑁𝐿 (𝑖)
+ 𝜒𝑦 ∑ 𝐶12 [(𝑧3𝑖 − 𝑧3𝑖−1 ) − (𝑧𝑖 − 𝑧𝑖−1 )𝑧̅2𝑥 ]]
𝑖=1

LS-DYNA DEV 10/27/16 (r:8004) 12-1 (Transverse Shear Treatment For Layered Shell)
Transverse Shear Treatment For Layered Shell LS-DYNA Theory Manual

where “𝑁𝐿” is the number of layers in the material.

Assume 𝜀𝑦 = 0 and 𝜎𝑥 = 𝐸𝜒 𝜀𝜒 , let


𝑁𝐿
(𝐸𝐼)𝑥 = ∑ 𝐸𝑥(𝑖) [(𝑧3𝑖 − 𝑧3𝑖−1 ) − (𝑧𝑖 − 𝑧𝑖−1 )𝑧̅2𝑥 ], (14.5)
𝑖=1

then
𝑧 − 𝑧̅𝑥
𝜀𝑥 = = (𝑧 − 𝑧̅𝑥 )𝜒𝑥 , (14.6)
𝜌
and
1
𝑀𝑥𝑥 = (𝜒𝑥 (𝐸𝐼)𝑥 ), (14.7)
3
3𝑀𝑥𝑥
𝜒𝑥 = . (14.8)
(𝐸𝐼)𝑥
Therefore, the stress becomes
3𝑀𝑥𝑥 𝐸(𝑖)
𝑥 (𝑧 − 𝑧̅𝑥 )
𝜎𝑥(𝑖) = . (14.9)
(𝐸𝐼)𝑥
Now considering the first equilibrium equation, one can write the following:
(𝑗)
∂𝜏𝑥𝑧 ∂𝜎 3𝑄 𝐸 (𝑧 − 𝑧̅𝑥 )
= − 𝑥 = − 𝑥𝑧 𝑥 , (14.10)
∂𝑧 ∂𝑥 (𝐸𝐼)𝑥

2
3𝑄𝑥𝑧 𝐸𝑥 (𝑧 − 𝑧𝑧̅𝑥 )
(𝑗)
(𝑗) 2 (14.11)
𝜏𝑥𝑧 =− + 𝐶𝑗 ,
(𝐸𝐼)𝑥
where 𝑄𝑥𝑧 is the shear force and 𝐶𝑗 is the constant of integration. This constant is obtained
from the transverse shear stress continuity requirement at the interface of each layer. Let
then
𝑧2𝑖−1
𝑄𝑥𝑧 𝐸(𝑖)
𝑥 ( − 𝑧𝑖−1 𝑧̅𝑥 )
2 𝑖−1 (14.12)
𝐶𝑗 = + 𝜏𝑥𝑧 ,
(𝐸𝐼)𝑥
and
𝑄𝑥𝑧 𝐸(𝑖)
𝑥 𝑧2 𝑧2
(𝑖)
𝜏𝑥𝑧 (𝑖−1)
= 𝜏𝑥𝑧 + [ 𝑖−1 − 𝑧𝑖−1 𝑧̅𝑥 − + 𝑧𝑧̅𝑥 ]. (14.13)
(𝐸𝐼)𝑥 2 2
For the first layer
(1)
3𝑄𝑥𝑧 𝐶11 𝑧2 − 𝑧2𝑜
𝜏𝑥𝑧 = − [ − 𝑧̅𝑥 (𝑧 − 𝑧𝑜 )], (14.14)
(𝐸𝐼)𝑥 2
12-2 (Transverse Shear Treatment For Layered Shell) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Transverse Shear Treatment For Layered Shell

for subsequent layers


(𝑖)
(𝑖−1) 3𝑄𝑥𝑧 𝐶11 𝑧2 − 𝑧2𝑖−1
𝜏𝑥𝑧 = 𝜏𝑥𝑧 − [ − 𝑧̅𝑥 (𝑧 − 𝑧𝑖−1 )] , 𝑧𝑖−1 ≤ 𝑧 ≤ 𝑧𝑖 . (14.15)
(𝐸𝐼)𝑥 2
(𝑖−1)
Here 𝜏𝑥𝑧 is the stress in previous layer at the interface with the current layer. The shear
stress can also be expressed as follows:
(𝑖)
3𝑄𝑥𝑧 𝐶11 𝑧2 − 𝑧2𝑖−1
𝜏𝑥𝑧 = − [𝑓𝑥(𝑖) + − 𝑧̅𝑥 (𝑧 − 𝑧𝑖−1 )], (14.16)
(𝐸𝐼)𝑥 2
where
𝑖−1 𝑧𝑗 + 𝑧𝑗+1
1 (𝑗)
𝑓𝑥(𝑖) = (𝑖)
∑ 𝐶11 ℎ𝑗 [ − 𝑧̅𝑥 ], (14.17)
𝐶11 𝑗=1 2

and
ℎ𝑗 = 𝑧𝑗 − 𝑧𝑗−1 . (14.18)

To find 𝑄𝑥𝑧 , the shear force, assume that the strain energy expressed through
̅ , is equal to the strain energy expressed through the derived
average shear modules, 𝐶66
expressions as follows:
1 𝑄𝑥𝑧 1 𝜏 2
𝑈= = ∫ 𝑥𝑧 𝑑𝑧, (14.19)
̅ ℎ 2 𝐶66
2 𝐶66

2 2
1 9ℎ 𝐶11 (𝑖) (𝑧2 − 𝑧2𝑖−1 )
= ∫ [𝑓 + − 𝑧̅𝑥 (𝑧 − 𝑧𝑖−1 )] 𝑑𝑧
̅
𝐶66 (𝐸𝐼)𝑥
2 𝐶66 𝑥 2
𝑧𝑖 2
9ℎ 𝑁𝐿
(𝐶11 (𝑖) )2 𝑧2 − 𝑧𝑖−1 2
= 2
∑ (𝑖)
∫ [𝑓𝑥(𝑖) + − 𝑧̅𝑥 (𝑧 − 𝑧𝑖−1 )] 𝑑𝑧
(𝐸𝐼)𝑥 𝑖=1 𝐶66 𝑧𝑖−1
2
(14.20)
𝑁𝐿 (𝑖) 2
1 9ℎ (𝐶11 ) ℎ 𝑖
= 2
∑ 𝑖
{𝑓𝑥 [60𝑓𝑥𝑖 + 20ℎ𝑖 (𝑧𝑖 + 2𝑧𝑖−1 − 3𝑧̅𝑥 )]
60 (𝐸𝐼) 𝑖 𝐶66
𝑥

+ 𝑧̅𝑥 ℎ𝑖 [20𝑧̅𝑥 ℎ𝑖 + 35𝑧2𝑖−1 − 10𝑧𝑖−1 (𝑧𝑖 + 𝑧𝑖−1 ) − 15𝑧2𝑖 ] + 𝑧𝑖 (𝑧𝑖 + 𝑧𝑖−1 )(3𝑧2𝑖
− 7𝑧2𝑖−1 ) + 8𝑧4𝑖−1 },
then
̅ 𝛾
𝑄𝑥𝑧 = 𝜏̅𝑥𝑧 ℎ = 𝐶66 ̅̅̅̅𝑥𝑧 ℎ, (14.21)
to calculate 𝑧̅𝑥 use 𝜏𝑥𝑧 for last layer at surface 𝑧 = 0,
𝑁𝐿
(𝑖) 𝑧2𝑖 − 𝑧2𝑖−1
∑ 𝐶11 [( ) − 𝑧̅𝑥 (𝑧𝑖 − 𝑧𝑖−1 )] = 0, (14.22)
𝑖=1
2

LS-DYNA DEV 10/27/16 (r:8004) 12-3 (Transverse Shear Treatment For Layered Shell)
Transverse Shear Treatment For Layered Shell LS-DYNA Theory Manual

where
𝑁𝐿 (𝑖)
∑𝑖=1 𝐶11 ℎ𝑖 (𝑧𝑖 + 𝑧𝑖+1 )
𝑧̅𝑥 = 𝑁𝐿
. (14.23)
(𝑖)
2 ∑𝑖=1 𝐶11 ℎ𝑖

Algorithm:

The following algorithm is used in the implementation of the transverse shear treatment.

1. Calculate 𝑧̅𝑥 according to equation (14.23)

2. Calculate 𝑓𝑥𝑖 according to equation (14.17)


𝑁𝐿 (𝑖)
3. Calculate 13 ∑𝑖=1 𝐶11 (𝑧3𝑖 − 𝑧3𝑖−1 )

𝑁𝐿 (𝑖) 2
4. Calculate ℎ[13 ∑𝑖=1 𝐶11 (𝑧3𝑖 − 𝑧3𝑖−1 )]

5. ̅ according to equation (14.20)


Calculate 𝐶66

6. ̅ 𝛾
Calculate 𝑄𝑥𝑧 = 𝐶66 ̅̅̅̅𝑥𝑧 ℎ

7. Calculate 𝜏𝑥𝑧 according to equation (14.16)

Steps 1-5 are performed at the initialization stage. Step 6 is performed in the shell
formulation subroutine, and step 7 is performed in the stress calculation inside the
constitutive subroutine.

12-4 (Transverse Shear Treatment For Layered Shell) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Eight-Node Solid Shell Element

15
Eight-Node Solid Shell Element

The isoparametric eight-node brick element discussed in Section 3 forms the basis
for tshell formulation 1, a solid shell element with enhancements based on the Hughes-Liu
and the Belytschko-Lin-Tsay shells. Like the eight-node brick, the geometry is interpolated
from the nodal point coordinates as:
8
𝑗
𝑥𝑖 (𝑋𝛼 , 𝑡) = 𝑥𝑖 (𝑋𝛼 (𝜉 , 𝜂, 𝜁 ), 𝑡) = ∑ 𝜙𝑗 (𝜉 , 𝜂, 𝜁 )𝑥𝑖 (𝑡), (15.1)
𝑗=1

1
𝜙𝑗 = (1 + 𝜉 𝜉𝑗 )(1 + 𝜂𝜂𝑗 )(1 + 𝜁 𝜁𝑗 ). (15.2)
8
As with solid elements, 𝐍 is the 3 × 24 rectangular interpolation matrix:
𝜑1 0 0 𝜑2 0 … 0 0
⎡0 𝜑1 0 0 𝜑2 … 𝜑8 0 ⎤
𝐍(𝜉 , 𝜂, 𝜁 ) = ⎢ ⎥, (15.3)
⎣0 0 𝜑1 0 0 … 0 𝜑8 ⎦
𝛔 is the stress vector:
𝛔T = (𝜎𝑥𝑥 , 𝜎𝑦𝑦 , 𝜎𝑧𝑧 , 𝜎𝑥𝑦 , 𝜎𝑦𝑧 , 𝜎𝑧𝑥 ), (15.4)
and 𝐁 is the 6 × 24 strain-displacement matrix:

LS-DYNA DEV 10/27/16 (r:8004) 13-1 (Eight-Node Solid Shell Element)


Eight-Node Solid Shell Element LS-DYNA Theory Manual

ζ Upper shell surface. The numbering of


the solid shell determines its orientation

5 Node ξ η ζ
8
1 -1 -1 -1
6 2 1 -1 -1
3 1 1 -1
4
η 4 -1 1 -1
7 5 -1 -1 1
6 1 -1 1
7 1 1 1
2 8 -1 1 1
3
ξ

Figure 15.1. Eight node solid shell element


⎡ 0 0 ⎤
⎢∂𝑥 ⎥
⎢ ∂ ⎥
⎢0 0 ⎥
⎢ ∂𝑦 ⎥
⎢ ⎥
⎢ ∂⎥
⎢0 0
∂𝑧 ⎥
𝐁=⎢∂ ∂
⎥ 𝐍, (15.5)
⎢ ⎥
⎢ 0 ⎥
⎢∂𝑦 ∂𝑥 ⎥
⎢ ∂ ∂⎥
⎢0 ⎥
⎢ ∂𝑧 ∂𝑦⎥
⎢ ⎥
⎢∂ ∂⎥
⎣∂𝑧 0
∂𝑥⎦

Terms in the strain-displacement matrix are readily calculated. Note that


∂𝜑𝑖 ∂𝜑𝑖 ∂𝑥 ∂𝜑𝑖 ∂𝑦 ∂𝜑𝑖 ∂𝑧
= + + ,
∂𝜉 ∂𝑥 ∂𝜉 ∂𝑦 ∂𝜉 ∂𝑧 ∂𝜉
∂𝜑𝑖 ∂𝜑𝑖 ∂𝑥 ∂𝜑𝑖 ∂𝑦 ∂𝜑𝑖 ∂𝑧
= + + , (15.6)
∂𝜂 ∂𝑥 ∂𝜂 ∂𝑦 ∂𝜂 ∂𝑧 ∂𝜂
∂𝜑𝑖 ∂𝜑𝑖 ∂𝑥 ∂𝜑𝑖 ∂𝑦 ∂𝜑𝑖 ∂𝑧
= + + ,
∂𝜁 ∂𝑥 ∂𝜁 ∂𝑦 ∂𝜁 ∂𝑧 ∂𝜁
which can be rewritten as

13-2 (Eight-Node Solid Shell Element) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Eight-Node Solid Shell Element

∂𝜑 ∂𝑥 ∂𝑦 ∂𝑧 ∂𝜑𝑖 ∂𝜑𝑖
⎡ 𝑖⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
⎢ ∂𝜉 ⎥ ⎢∂𝜉 ∂𝜉 ∂𝜉 ⎥ ⎢ ∂𝑥 ⎥ ⎢ ∂𝑥 ⎥
⎢∂𝜑 ⎥ ⎢∂𝑥 ∂𝑧 ⎥
⎢ 𝑖⎥ ⎢ ∂𝑦 ⎥⎢ ∂𝜑𝑖 ⎥ ⎢∂𝜑 ⎥
⎥ = 𝐉 ⎢ 𝑖 ⎥.
⎢ ⎥=⎢ ⎥⎢ (15.7)
⎢ ∂𝜂 ⎥ ⎢∂𝜂 ∂𝜂 ∂𝜂⎥ ⎢ ∂𝑦 ⎥ ⎢ ∂𝑦 ⎥
⎢∂𝜑 ⎥ ⎢∂𝑥 ⎥ ⎢ ⎥ ⎢ ⎥
⎢ 𝑖⎥ ⎢ ∂𝑦 ∂𝑧 ⎥ ⎢∂𝜑𝑖 ⎥ ⎢∂𝜑𝑖 ⎥
⎣ ∂𝜁 ⎦ ⎣∂𝜁 ∂𝜁 ∂𝜁 ⎦ ⎣ ∂𝑧 ⎦ ⎣ ∂𝑧 ⎦
Inverting the Jacobian matrix, 𝐉, we can solve for the desired terms
∂𝜑 ∂𝜑
⎡ 𝑖⎤ ⎡ 𝑖⎤
⎢ ∂𝑥 ⎥ ⎢ ∂𝜉 ⎥
⎢∂𝜑𝑖 ⎥ ⎢∂𝜑 ⎥
⎢ ⎥ = 𝐉−1 ⎢
⎢ ⎥.
𝑖⎥
(15.8)
⎢ ∂𝑦 ⎥ ⎢ ∂𝜂 ⎥
⎢ ⎥ ⎢∂𝜑 ⎥
⎢∂𝜑𝑖 ⎥ ⎢ 𝑖⎥
⎣ ∂𝑧 ⎦ ⎣ ∂𝜁 ⎦

To obtain shell-like behavior from the solid element, it is necessary to use multiple
integration points through the shell thickness along the 𝜁 axis while employing a plane
stress constitutive subroutine. Consequently, it is necessary to construct a reference surface
within the brick shell. We locate the reference surface midway between the upper and
lower surfaces and construct a local coordinate system exactly as was done for the
Belytschko-Lin-Tsay shell element. Following the procedure outlined in Section 7,
Equations (7.1) – (7.3), the local coordinate system can be constructed as depicted in Figure
15.2. Equation (7.5a) gives the transformation matrix in terms of the local basis:

⎧ 𝐴 𝑒 𝑒 𝑒 ⎧𝐴𝑥̂ ⎫
{ 𝑥⎫} ⎡ 1𝑥 2𝑥 3𝑥 ⎤ { { } }
{𝐀} = ⎨𝐴𝑦 ⎬ = ⎢𝑒1𝑦 𝑒2𝑦 𝑒3𝑦 ⎥ ⎨𝐴𝑦̂ ⎬ = [𝛍]{𝐀
̂ } = [𝐪]T {𝐀
̂ }. (15.9)
{ }
⎩𝐴𝑧 ⎭ ⎣ 𝑒 𝑒 𝑒 ⎦ {
{ }
}
1𝑧 2𝑧 3𝑧
⎩𝐴𝑧̂ ⎭

As with the Hughes-Liu shell, the next step is to perform the Jaumann rate update:
𝜎𝑖𝑗𝑛+1 = 𝜎𝑖𝑗𝑛 + 𝜎𝑖𝑝
𝑛 𝑛
Δ𝜔𝑝𝑗 + 𝜎𝑗𝑝 Δ𝜔𝑝𝑖 , (15.10)
to account for the material rotation between time steps 𝑛 and 𝑛 + 1. The Jaumann rate
update of the stress tensor is applied in the global configuration before the constitutive
evaluation is performed. In the solid shell, as in the Hughes-Liu shell, the stresses and
history variables are stored in the global coordinate system. To evaluate the constitutive
relation, the stresses and the strain increments are rotated from the global to the lamina
coordinate system using the transformation defined previously:
𝑛+1
𝑛+1
𝜎𝑖𝑗𝑙 = 𝑞𝑖𝑘 𝜎𝑘𝑛 𝑞𝑗𝑛 , (15.11)

𝑛+1⁄2 𝑛+1⁄2
Δ𝜀𝑖𝑗𝑙 = 𝑞𝑖𝑘 Δ𝜀𝑘𝑛 𝑞𝑗𝑛 , (15.12)

LS-DYNA DEV 10/27/16 (r:8004) 13-3 (Eight-Node Solid Shell Element)


Eight-Node Solid Shell Element LS-DYNA Theory Manual

A reference surface is constructed within


the solid shell element and the local reference
system is defined.

^
y s3 3
4

e^3 r31
r42

e^2

s1 ^
x
1 e^1
r21 2

Figure 15.2. Construction of the reference surface in the solid shell element.

where the superscript l indicates components in the lamina (local) coordinate system. The
stress is updated incrementally:
𝑛+1 𝑛+1 𝑛+1⁄2
𝜎𝑖𝑗𝑙 = 𝜎𝑖𝑗𝑙 + Δ𝜎𝑖𝑗𝑙 . (15.13)

Independently from the constitutive evaluation


𝑙 (15.14)
𝜎33 = 0,
which ensures that the plane stress condition is satisfied, we update the normal stress
which is used as a penalty to maintain the thickness of the shell:
penalty n+1 penalty n
(𝜎33 ) = (𝜎33 ) + 𝐸Δ𝜀𝑙33 , (15.15)

where 𝐸 is the elastic Young’s modulus for the material. The stress tensor of Equation
(15.13) is rotated back to the global system:
𝑙 𝑛+1
𝜎𝑖𝑗𝑛+1 = 𝑞𝑘𝑖 (𝜎𝑘𝑛 ) 𝑞𝑛𝑗 . (15.16)

A penalty stress tensor is then formed by transforming the normal penalty stress tensor (a
null tensor except for the 33 term) back to the global system:

13-4 (Eight-Node Solid Shell Element) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Eight-Node Solid Shell Element

𝑛+1
penalty n+1 penalty 𝑙
(𝜎𝑖𝑗 ) = 𝑞𝑘𝑖 [(𝜎𝑖𝑗 )] 𝑞𝑛𝑗 , (15.17)

before computing the internal force vector. The internal force vector can now be computed:
T 𝑛+1
𝐟int = ∫(𝐁𝑛+1 ) [𝝈 𝑛+1 + (𝝈 penalty ) ] 𝑑𝜐. (15.18)

The brick shell exhibits no discernible locking problems with this approach.

The treatment of the hourglass modes is identical to that described for the solid
elements in Section 3.

LS-DYNA DEV 10/27/16 (r:8004) 13-5 (Eight-Node Solid Shell Element)


LS-DYNA Theory Manual Eight-Node Solid Element for Thick Shell Simulations

16
Eight-Node Solid Element for Thick
Shell Simulations

16.1 Abstract
Tshell formulation 3 is an eight-node hexahedral element incorporated into LS-
DYNA to simulate thick shell structures. The element formulation is derived in a co-
rotational coordinate system and the strain operator is calculated with a Taylor series
expansion about the center of the element. Special treatments are made on the dilatational
strain component and shear strain components to eliminate the volumetric and shear
locking. The use of consistent tangential stiffness and geometric stiffness greatly improves
the convergence rate in implicit analysis.

16.2 Introduction
Large-scale finite element analyses are extensively used in engineering designs and
process controls. For example, in automobile crashworthiness, hundreds of thousands of
unknowns are involved in the computer simulation models, and in metal forming
processing, tests in the design of new dies or new products are done by numerical
computations instead of costly experiments. The efficiency of the elements is of crucial
importance to speed up the design processes and reduce the computational costs for these
problems. Over the past ten years, considerable progress has been achieved in developing
fast and reliable elements.

In the simulation of shell structures, Belytschko-Lin-Tsay [Belytschko, 1984a] and


Hughes-Liu [Hughes, 1981a and 1981b] shell elements are widely used. However, in some
cases thick shell elements are more suitable. For example, in the sheet metal forming with
large curvature, traditional thin shell elements cannot give satisfactory results. Also thin
shell elements cannot give us detailed strain information though the thickness. In LS-
LS-DYNA DEV 10/27/16 (r:8004) 14-1 (Eight-Node Solid Element for Thick Shell Simulations)
Eight-Node Solid Element for Thick Shell Simulations LS-DYNA Theory Manual

DYNA, the eight-node solid thick shell element is still based on the Hughes-Liu and
Belytschko-Lin-Tsay shells [Hallquist, 1998]. A new eight-node solid element based on Liu,
1985, 1994 and 1998 is incorporated into LS-DYNA, intended for thick shell simulation.
The strain operator of this element is derived from a Taylor series expansion and special
treatments on strain components are utilized to avoid volumetric and shear locking.

The organization of this paper is as follows. The element formulations are described
in the next section. Several numerical problems are studied in the third section, followed
by the conclusions.

16.3 Element Formulations


16.3.1 Strain Operator
The new element is based on the eight-node hexahedral element proposed and
enhanced by Liu, 1985, 1994, 1998. For an eight-node hexahedral element, the spatial
coordinates, 𝑥𝑖 , and the velocity components, 𝑣𝑖 , in the element are approximated in terms
of nodal values, x𝑖a and v𝑖a , by
8
𝑥𝑖 = ∑ 𝑁𝑎 (𝜉 , 𝜂, 𝜁 )𝑥𝑖𝑎 , (16.1)
𝑎=1

8
𝑣𝑖 = ∑ 𝑁𝑎 (𝜉 , 𝜂, 𝜁 )𝑣𝑖𝑎 , 𝑖 = 1, 2, 3, (16.2)
𝑎=1

1
𝑁𝑎 (𝜉 , 𝜂, 𝜁 ) = (1 + 𝜉𝑎 𝜉 )(1 + 𝜂𝑎 𝜂)(1 + 𝜁𝑎 𝜁 ), (16.3)
8
and the subscripts 𝑖 and a denote coordinate components ranging from one to three and the
element nodal numbers ranging from one to eight, respectively. The referential coordinates
𝜉 , 𝜂, and 𝜁 of node a are denoted by 𝜉𝑎 , 𝜂𝑎 , and 𝜁𝑎 , respectively.

The strain rate (or rate of deformation), 𝛆̇, is composed of six components,
𝛆̇T = [𝜀𝑥𝑥 𝜀𝑦𝑦 𝜀𝑧𝑧 𝜀𝑥𝑦 𝜀𝑦𝑧 𝜀𝑧𝑥 ], (16.4)
and is related to the nodal velocities by a strain operator, ̅𝐁
̅̅̅̅,
̅̅̅̅̅(𝜉 , 𝜂, 𝜁 )𝐯,
𝛆̇ = 𝐁 (16.5)
where
𝐯T = [vx1 vy1 vz1 ⋯ vx8 vy8 vz8 ], (16.6)

14-2 (Eight-Node Solid Element for Thick Shell Simulations) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Eight-Node Solid Element for Thick Shell Simulations

̅B̅̅̅
⎡ 𝑥𝑥 ⎤ 𝐵1 (1) 0 0 ⋯ 𝐵1 (8) 0 0
⎢̅B̅̅̅𝑦𝑦 ⎥ ⎡ 0 𝐵2 (1) 0 ⋯ 0 𝐵2 (8) 0 ⎤
⎢ ⎥ ⎢ ⎥
⎢̅B̅̅̅ ⎥ ⎢ 𝐵3 (8)⎥
̅̅̅̅ = ⎢ 𝑧𝑧 ⎥ = ⎢ 0
̅𝐁 0 𝐵3 (1) ⋯ 0 0 ⎥, (16.7)
⎢B ̅̅̅̅ ⎥ ⎢𝐵 (1) 𝐵1 (1) 0 ⋯ 𝐵2 (8) 𝐵1 (8) 0 ⎥
⎢ 𝑥𝑦 ⎥ ⎢ 2 ⎥
⎢̅B̅̅̅ ⎥ ⎢ 0 𝐵3 (1) 𝐵2 (1) ⋯ 0 𝐵3 (8) 𝐵2 (8)⎥
⎢ 𝑦𝑧 ⎥ ⎣𝐵 (1) 0 𝐵1 (1) ⋯ 𝐵3 (8) 0 𝐵1 (8)⎦
3
⎣̅B̅̅̅𝑧𝑥 ⎦

B1 𝑁,𝑥 (𝜉 , 𝜂, 𝜁 )
⎡B ⎤ = ⎡ ⎤
⎢ 2⎥ ⎢ ⎢𝑁,𝑦 (𝜉 , 𝜂, 𝜁 )⎥⎥. (16.8)
⎣B3 ⎦ ⎣𝑁,𝑧 (𝜉 , 𝜂, 𝜁 ) ⎦

Unlike standard solid element where the strain operator is computed by


differentiating the shape functions, the strain operator for this new element is expanded in
a Taylor series about the element center up to bilinear terms as follows [Liu, 1994, 1998],
̅̅̅̅̅(𝜉 , 𝜂, 𝜁 ) = 𝐁
𝐁 ̅̅̅̅̅(0) + ̅𝐁 ̅̅̅̅,𝜉 (0)𝛏 + 𝐁
̅̅̅̅̅,𝜂 (0)𝛈 + ̅𝐁 ̅̅̅̅,𝜁 (0)𝛇
(16.9)
+ 2[𝐁 ̅̅̅̅̅,𝜉𝜂 (0)𝛏𝛈 + ̅𝐁 ̅̅̅̅,𝜂𝜁 (0)𝛈𝛇 + 𝐁 ̅̅̅̅̅,𝜁𝜉 (0)𝛇𝛏] .

Let
𝐱1T = 𝐱T = [𝑥1 𝑥2 𝑥3 𝑥4 𝑥5 𝑥6 𝑥7 𝑥8 ], (16.10)

𝐱2T = 𝐲T = [𝑦1 𝑦2 𝑦3 𝑦4 𝑦5 𝑦6 𝑦7 𝑦8 ], (16.11)

𝐱3T = 𝐳T = [𝑧1 𝑧2 𝑧3 𝑧4 𝑧5 𝑧6 𝑧7 𝑧8 ], (16.12)

𝛏T = [−1 1 1 −1 −1 1 1 −1], (16.13)

𝛈T = [−1 −1 1 1 −1 −1 1 1], (16.14)

𝛇T = [−1 −1 −1 −1 1 1 1 1], (16.15)


the Jacobian matrix at the center of the element can be evaluated as
𝛏T 𝐱 𝛏T 𝐲 𝛏T 𝐳

1 T ⎤
𝐉(0) = [J𝑖𝑗 ] = ⎢ 𝛈 𝐱 𝛈T 𝐲 𝛈 T 𝐳⎥
⎥; (16.16)
8⎢ T
⎣𝛇 𝐱 𝛇T 𝐲 T
𝛇 𝐳⎦
the determinant of the Jacobian matrix is denoted by j0 and the inverse matrix of 𝐉(0) is
denoted by 𝐃
𝐃 = [D𝑖𝑗 ] = 𝐉−1 (0). (16.17)

The gradient vectors and their derivatives with respect to the natural coordinates at
the center of the element are given as follows,

LS-DYNA DEV 10/27/16 (r:8004) 14-3 (Eight-Node Solid Element for Thick Shell Simulations)
Eight-Node Solid Element for Thick Shell Simulations LS-DYNA Theory Manual

1
𝐛1 = 𝐍,𝑥 (0) = [𝐷11 𝛏 + 𝐷12 𝛈 + 𝐷13 𝛇], (16.18)
8
1
𝐛2 = 𝐍,𝑦 (0) = [𝐷21 𝛏 + 𝐷22 𝛈 + 𝐷23 𝛇], (16.19)
8
1
𝐛3 = 𝐍,𝑧 (0) = [𝐷31 𝛏 + 𝐷32 𝛈 + 𝐷33 𝛇], (16.20)
8
1
𝐛1,𝜉 = 𝐍,𝑥𝜉 (0) = [𝐷12 𝜸1 + 𝐷13 𝛄2 ], (16.21)
8
1
𝐛2,𝜉 = 𝐍,𝑦𝜉 (0) = [𝐷22 𝛄1 + 𝐷23 𝛄2 ], (16.22)
8
1
𝐛3,𝜉 = 𝐍,𝑧𝜉 (0) = [𝐷32 𝛄1 + 𝐷33 𝛄2 ], (16.23)
8
1
𝐛1,𝜂 = 𝐍,𝑥𝜂 (0) = [𝐷11 𝛄1 + 𝐷13 𝛄3 ], (16.24)
8
1
𝐛2,𝜂 = 𝐍,𝑦𝜂 (0) = [𝐷21 𝛄1 + 𝐷23 𝛄3 ], (16.25)
8
1
𝐛3,𝜂 = 𝐍,𝑧𝜂 (0) = [𝐷31 𝛄1 + 𝐷33 𝛄3 ], (16.26)
8
1
𝐛1,𝜁 = 𝐍,𝑥𝜁 (0) = [𝐷11 𝛄2 + 𝐷12 𝛄3 ], (16.27)
8
1
𝐛2,𝜁 = 𝐍,𝑦𝜁 (0) = [𝐷21 𝛄2 + 𝐷22 𝛄3 ], (16.28)
8
1
𝐛3,𝜁 = 𝐍,𝑧𝜁 (0) = [𝐷31 𝛄2 + 𝐷32 𝛄3 ], (16.29)
8
1
𝐛1,𝜉𝜂 = 𝐍,𝑥𝜉𝜂 (0) = [𝐷13 𝛄4 − (𝐩T1 𝐱𝑖 )𝐛𝑖,𝜉 − (𝐫1T 𝐱𝑖 )𝐛𝑖,𝜂 ], (16.30)
8
1
𝐛2,𝜉𝜂 = 𝐍,𝑦𝜉𝜂 (0) = [𝐷23 𝛄4 − (𝐛T2 𝐱𝑖 )𝐛𝑖,𝜉 − (𝐫2T 𝐱𝑖 )𝐛𝑖,𝜂 ], (16.31)
8
1
𝐛3,𝜉𝜂 = 𝐍,𝑧𝜉𝜂 (0) = [𝐷33 𝛄4 − (𝐩T3 𝐱𝑖 )𝐛𝑖,𝜉 − (𝐫3T 𝐱𝑖 )𝐛𝑖,𝜂 ], (16.32)
8
1
𝐛1,𝜂𝜁 = 𝐍,𝑥𝜂𝜁 (0) = [𝐷11 𝛄4 − (𝐪1T 𝐱𝑖 )𝐛𝑖,𝜂 − (𝐩T1 𝐱𝑖 )𝐛𝑖,𝜁 ], (16.33)
8

14-4 (Eight-Node Solid Element for Thick Shell Simulations) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Eight-Node Solid Element for Thick Shell Simulations

1
𝐛2,𝜂𝜁 = 𝐍,𝑦𝜂𝜁 (0) = [𝐷21 𝛄4 − (𝐪2T 𝐱𝑖 )𝐛𝑖,𝜂 − (𝐩T2 𝐱𝑖 )𝐛𝑖,𝜁 ], (16.34)
8
1
𝐛3,𝜂𝜁 = 𝐍,𝑧𝜂𝜁 (0) = [𝐷31 𝛄4 − (𝐪3T 𝐱𝑖 )𝐛𝑖,𝜂 − (𝐩T3 𝐱𝑖 )𝐛𝑖,𝜁 ], (16.35)
8
1
𝐛1,𝜁𝜉 = 𝐍,𝑥𝜁𝜉 (0) = [𝐷12 𝛄4 − (𝐫1T 𝐱𝑖 )𝐛𝑖,𝜁 − (𝐪1T 𝐱𝑖 )𝐛𝑖,𝜉 ], (16.36)
8
1
𝐛2,𝜁𝜉 = 𝐍,𝑦𝜁𝜉 (0) = [𝐷22 𝛄4 − (𝐫2T 𝐱𝑖 )𝐛𝑖,𝜁 − (𝐪2T 𝐱𝑖 )𝐛𝑖,𝜉 ], (16.37)
8
1
𝐛3,𝜁𝜉 = 𝐍,𝑧𝜁𝜉 (0) = [𝐷32 𝛄4 − (𝐫3T 𝐱𝑖 )𝐛𝑖,𝜁 − (𝐪3T 𝐱𝑖 )𝐛𝑖,𝜉 ], (16.38)
8
where
𝐩𝑖 = 𝐷𝑖1 𝐡1 + 𝐷𝑖3 𝐡3 , (16.39)

𝐪𝑖 = 𝐷𝑖1 𝐡2 + 𝐷𝑖2 𝐡3 , (16.40)

𝐫𝑖 = 𝐷𝑖2 𝐡1 + 𝐷𝑖3 𝐡2 , (16.41)

𝜸𝛼 = 𝐡𝛼 − (𝐡𝛼T 𝐱𝑖 )𝐛𝑖 , (16.42)


and
𝐡1T = [1 −1 1 −1 1 −1 1 −1], (16.43)

𝐡2T = [1 −1 −1 1 −1 1 1 −1], (16.44)

𝐡3T = [1 1 −1 −1 −1 −1 1 1], (16.45)

𝐡4T = [−1 1 −1 1 1 −1 1 −1]. (16.46)

In the above equations 𝐡1 is the 𝜉𝜂-hourglass vector, 𝐡2 the 𝜂𝜁 -hourglass vector, 𝐡3


the 𝜁𝜉 -hourglass vector and 𝐡4 the 𝜉𝜂𝜁 -hourglass vector. They are the zero energy-
deformation modes associated with the one-point-quadrature element which result in a
non-constant strain field in the element [Flanagan, 1981, Belytschko, 1984 and Liu, 1984].
The 𝛾𝛼 in equations (16.21)–(16.38) are the stabilization vectors. They are orthogonal to the
linear displacement field and provide a consistent stabilization for the element.

The strain operators, 𝐁̅̅̅̅̅(𝜉 , 𝜂, 𝜁 ), can be decomposed into two parts, the dilatational
dil
̅̅̅̅̅ (𝜉 , 𝜂, 𝜁 ), and the deviatoric part, ̅𝐁 ̅̅̅̅dev (𝜉 , 𝜂, 𝜁 ), both of which can be expanded
part, 𝐁
about the element center as in Equation (16.9)
̅̅̅̅̅dil (𝛏, 𝛈, 𝛇) = ̅𝐁
𝐁 ̅̅̅̅dil
̅̅̅̅dil (0) + ̅𝐁 ̅̅̅̅̅dil ̅̅̅̅̅dil
,𝜉 (0)𝛏 + 𝐁,𝜂 (0)𝛈 + 𝐁,𝜁 (0)𝛇
(16.47)
+2[𝐁 ̅̅̅̅̅dil ̅̅̅̅̅dil ̅̅̅̅̅dil
,𝜉𝜂 (0)𝛏𝛈 + 𝐁,𝜂𝜁 (0)𝛈𝛇 + 𝐁,𝜁𝜉 (0)𝛇𝛏],

LS-DYNA DEV 10/27/16 (r:8004) 14-5 (Eight-Node Solid Element for Thick Shell Simulations)
Eight-Node Solid Element for Thick Shell Simulations LS-DYNA Theory Manual

̅̅̅̅̅dev (𝜉 , 𝜂, 𝜁 ) = 𝐁
𝐁 ̅̅̅̅dev
̅̅̅̅̅dev (0) + ̅𝐁 ̅̅̅̅̅dev ̅̅̅̅̅dev
,𝜉 (0)𝜉 + 𝐁,𝜂 (0)𝛈 + 𝐁,𝜁 (0)𝛇
(16.48)
+2[𝐁 ̅̅̅̅̅dev ̅̅̅̅̅dev ̅̅̅̅̅dev
,𝜉𝜂 (0)𝛏𝛈 + 𝐁,𝜂𝜁 (0)𝛈𝛇 + 𝐁,𝜁𝜉 (0)𝛇𝛏],

To avoid volumetric locking, the dilatational part of the strain operators is evaluated
only at one quadrature point, the center of the element, i.e., they are constant terms
̅̅̅̅̅dil (𝝃 , 𝜼, 𝜻) = ̅𝐁
𝐁 ̅̅̅̅dil (0). (16.49)

To remove shear locking, the deviatoric strain submatrices can be written in an


orthogonal co-rotational coordinate system rotating with the element as
̅̅̅̅dev
B ̅̅̅̅dev ̅̅̅̅dev ̅̅̅̅dev ̅̅̅̅dev
𝑥𝑥 (𝜉 , 𝜂, 𝜁 ) = B𝑥𝑥 (0) + B𝑥𝑥,𝜉 (0)𝜉 + B𝑥𝑥,𝜂 (0)𝜂 + B𝑥𝑥,𝜁 (0)𝜁
(16.50)
+2[B ̅̅̅̅dev ̅̅̅̅dev ̅̅̅̅dev
𝑥𝑥,𝜉𝜂 (0)𝜉𝜂 + B𝑥𝑥,𝜂𝜁 (0)𝜂𝜁 + B𝑥𝑥,𝜁𝜉 (0)𝜁𝜉 ],

̅B̅̅̅dev ̅̅̅̅dev ̅̅̅̅dev ̅̅̅̅dev ̅̅̅̅dev


𝑦𝑦 (𝜉 , 𝜂, 𝜁 ) = B𝑦𝑦 (0) + B𝑦𝑦,𝜉 (0)𝜉 + B𝑦𝑦,𝜂 (0)𝜂 + B𝑦𝑦,𝜁 (0)𝜁
(16.51)
+2[B ̅̅̅̅dev ̅̅̅̅dev ̅̅̅̅dev
𝑦𝑦,𝜉𝜂 (0)𝜉𝜂 + B𝑦𝑦,𝜂𝜁 (0)𝜂𝜁 + B𝑦𝑦,𝜁𝜉 (0)𝜁𝜉 ],

̅̅̅̅dev
B ̅̅̅̅dev ̅̅̅̅dev ̅̅̅̅dev ̅̅̅̅dev
𝑧𝑧 (𝜉 , 𝜂, 𝜁 ) = B𝑧𝑧 (0) + B𝑧𝑧,𝜉 (0)𝜉 + B𝑧𝑧,𝜂 (0)𝜂 + B𝑧𝑧,𝜁 (0)𝜁
(16.52)
+2[B ̅̅̅̅dev ̅̅̅̅dev ̅̅̅̅dev
𝑧𝑧,𝜉𝜂 (0)𝜉𝜂 + B𝑧𝑧,𝜂𝜁 (0)𝜂𝜁 + B𝑧𝑧,𝜁𝜉 (0)𝜁𝜉 ],

̅̅̅̅dev
B ̅̅̅̅dev ̅̅̅̅dev
𝑥𝑦 (𝜉 , 𝜂, 𝜁 ) = B𝑥𝑦 (0) + B𝑥𝑦,𝜁 (0)𝜁 , (16.53)

̅B̅̅̅dev ̅̅̅̅dev ̅̅̅̅dev


𝑦𝑧 (𝜉 , 𝜂, 𝜁 ) = B𝑦𝑧 (0) + B𝑦𝑧,𝜉 (0)𝜉 , (16.54)

̅̅̅̅dev
B ̅̅̅̅dev ̅̅̅̅dev
𝑧𝑥 (𝜉 , 𝜂, 𝜁 ) = B𝑧𝑥 (0) + B𝑧𝑥,𝜂 (0)𝜂. (16.55)

Here, only one linear term is left for shear strain components such that the modes
causing shear locking are removed. The normal strain components keep all non-constant
terms given in equation (16.48).

Summation of equation (16.49) and equations (16.50)–(16.55) yields the following


strain submatrices which can eliminate the shear and volumetric locking:
̅B̅̅̅𝑥𝑥 (𝜉 , 𝜂, 𝜁 ) = ̅B̅̅̅𝑥𝑥 (0) + B ̅̅̅̅dev ̅̅̅̅dev ̅̅̅̅dev
𝑥𝑥,𝜉 (0)𝜉 + B𝑥𝑥,𝜂 (0)𝜂 + B𝑥𝑥,𝜁 (0)𝜁
(16.56)
+2[B ̅̅̅̅dev ̅̅̅̅dev ̅̅̅̅dev
𝑥𝑥,𝜉𝜂 (0)𝜉𝜂 + B𝑥𝑥,𝜂𝜁 (0)𝜂𝜁 + B𝑥𝑥,𝜁𝜉 (0)𝜁𝜉 ],

̅̅̅̅𝑦𝑦 (𝜉 , 𝜂, 𝜁 ) = B
B ̅̅̅̅𝑦𝑦 (0) + B ̅̅̅̅dev ̅̅̅̅dev ̅̅̅̅dev
𝑦𝑦,𝜉 (0)𝜉 + B𝑦𝑦,𝜂 (0)𝜂 + B𝑦𝑦,𝜁 (0)𝜁
(16.57)
+2[B ̅̅̅̅dev ̅̅̅̅dev ̅̅̅̅dev
𝑦𝑦,𝜉𝜂 (0)𝜉𝜂 + B𝑦𝑦,𝜂𝜁 (0)𝜂𝜁 + B𝑦𝑦,𝜁𝜉 (0)𝜁𝜉 ],

̅B̅̅̅𝑧𝑧 (𝜉 , 𝜂, 𝜁 ) = ̅B̅̅̅𝑧𝑧 (0) + ̅B̅̅̅dev ̅̅̅̅dev ̅̅̅̅dev


𝑧𝑧,𝜉 (0)𝜉 + B𝑧𝑧,𝜂 (0)𝜂 + B𝑧𝑧,𝜁 (0)𝜁
(16.58)
+2[B ̅̅̅̅dev ̅̅̅̅dev ̅̅̅̅dev
𝑧𝑧,𝜉𝜂 (0)𝜉𝜂 + B𝑧𝑧,𝜂𝜁 (0)𝜂𝜁 + B𝑧𝑧,𝜁𝜉 (0)𝜁𝜉 ],

14-6 (Eight-Node Solid Element for Thick Shell Simulations) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Eight-Node Solid Element for Thick Shell Simulations

̅̅̅̅𝑥𝑦 (𝜉 , 𝜂, 𝜁 ) = B
B ̅̅̅̅dev
̅̅̅̅𝑥𝑦 (0) + B 𝑥𝑦,𝜁 (0)𝜁 , (16.59)

̅̅̅̅𝑦𝑧 (𝜉 , 𝜂, 𝜁 ) = B
B ̅̅̅̅dev
̅̅̅̅𝑦𝑧 (0) + B 𝑦𝑧,𝜉 (0)𝜉 , (16.60)

̅B̅̅̅𝑧𝑥 (𝜉 , 𝜂, 𝜁 ) = ̅B̅̅̅𝑧𝑥 (0) + ̅B̅̅̅dev


𝑧𝑥,𝜂 (0)𝜂. (16.61)

It is noted that the elements developed above cannot pass the patch test if the
elements are skewed. To remedy this drawback, the gradient vectors defined in (16.18)–
(16.20) are replaced by the uniform gradient matrices, proposed by Flanagan [1981],
b̃ B1 (𝜉 , 𝜂, 𝜁 )
⎡ 1⎤ 1 ⎡ ⎤
⎢b̃2 ⎥
⎢ ⎥ = 𝑉 ∫Ω ⎢B2 (𝜉 , 𝜂, 𝜁 )⎥ 𝑑𝑉 . (16.62)
𝑒
⎣b̃3 ⎦
𝑒
⎣B3 (𝜉 , 𝜂, 𝜁 )⎦
Where 𝑉𝑒 is the element volume and the stabilization vector are redefined as
𝛄̃ 𝛼 = 𝐡𝛼 − (𝐡𝛼T 𝐱𝑖 )𝐛̃𝑖 . (16.63)

The element using the strain submatrices (16.56)-(16.61) and uniform gradient
matrices (16.62) with four-point quadrature scheme is called HEXDS element.

16.3.2 Co-rotational Coordinate System


In elements for shell/plate structure simulations, the elimination of the shear
locking depends on the proper treatment of the shear strain. It is necessary to attach a local

g2
e^3 e^1
g1
y e^2
ζ=0

x
Figure 16.1. Definition of co-rotational coordinate system

coordinate system to the element so that the strain tensor in this local system is relevant for
LS-DYNA DEV 10/27/16 (r:8004) 14-7 (Eight-Node Solid Element for Thick Shell Simulations)
Eight-Node Solid Element for Thick Shell Simulations LS-DYNA Theory Manual

the treatment. The co-rotational coordinate system determined here is one of the most
convenient ways to define such a local system.

A co-rotational coordinate system is defined as a Cartesian coordinate system which


rotates with the element. Let {𝑥𝑎 , 𝑦𝑎 , 𝑧𝑎 } denote the current nodal spatial coordinates in the
global system. For each quadrature point with natural coordinates(𝜉 , 𝜂, 𝜁 ), we can have
two tangent directions on the mid-surface (𝜁 = 0) within the element (see Fig. 16.1)
∂𝐱 ∂𝑥 ∂𝑦 ∂𝑧
𝐠1 = =[ ] = [𝑁𝑎,𝜉 𝑥𝑎 𝑁𝑎,𝜉 𝑦𝑎 𝑁𝑎,𝜉 𝑧𝑎 ](𝜉 ,𝜂,0) , (16.64)
∂𝜉 ∂𝜉 ∂𝜉 ∂𝜉

∂𝐱 ∂𝑥 ∂𝑦 ∂𝑧
𝐠2 = =[ ] = [𝑁𝑎,𝜂 𝑥𝑎 𝑁𝑎,𝜂 𝑦𝑎 𝑁𝑎,𝜂 𝑧𝑎 ](𝜉,𝜂,0) . (16.65)
∂𝜂 ∂𝜂 ∂𝜂 ∂𝜂

The unit vector 𝐞̂1 of the co-rotational coordinate system is defined as the bisector of
the angle intersected by these two tangent vectors 𝐠1 and 𝐠2 ; the unit vector 𝐞̂3 is
perpendicular to the mid-surface and the other unit vector is determined by 𝐞̂1 and 𝐞̂3 , i.e.,
𝐠 𝐠 𝐠 𝐠
𝐞̂1 = ( 1 + 2 )⁄(∣ 1 + 2 ∣), (16.66)
∣𝐠1 ∣ ∣𝐠2 ∣ ∣𝐠1 ∣ ∣𝐠2 ∣
𝐠1 × 𝐠2
𝐞̂3 = , (16.67)
∣𝐠1 × 𝐠2 ∣

𝐞̂2 = 𝐞̂3 × 𝐞̂1 , (16.68)


which lead to the transformation matrix
𝐞̂1
𝐑 = ⎢𝐞̂2 ⎤

⎥. (16.69)
⎣𝐞̂3 ⎦

16.3.3 Stress and Strain Measures


Since the co-rotational coordinate system rotates with the configuration, the stress
defined in this co-rotational system does not change with the rotation or translation of the
material body and is thus objective. Therefore, we use the Cauchy stress in the co-
rotational coordinate system, called the co-rotational Cauchy stress, as our stress measure.

The rate of deformation (or velocity strain tensor), also defined in the co-rotational
coordinate system, is used as the measure of the strain rate,
T
1 ∂𝐯̂ def ∂𝐯̂ def ⎤
𝛆̇ = 𝐝̂ = ⎡
⎢ + ( ) ⎥, (16.70)
2 ∂𝐱̂ ∂𝐱̂
⎣ ⎦
where 𝐯̂ def is the deformation part of the velocity in the co-rotational system 𝐱̂. If the initial
strain 𝛆̂ (𝐗, 0) is given, the strain tensor can be expressed as,

14-8 (Eight-Node Solid Element for Thick Shell Simulations) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Eight-Node Solid Element for Thick Shell Simulations

𝑡
𝛆̂(X, 𝑡) = 𝛆̂(𝐗, 0) + ∫ 𝐝̂ (𝐗, 𝜏) 𝑑𝜏. (16.71)
0

The strain increment is then given by the mid-point integration of the velocity strain
tensor,
T
𝑡𝑛+1 1 ⎡∂Δ𝐮̂def ⎛∂Δ𝐮̂

def

⎟ ⎤
Δ𝛆̂ = ∫ ̂
𝐝𝑑𝜏 =̇ ⎢ ⎜
+⎜ ⎟ ⎥ (16.72)
𝑡𝑛

2 ∂𝐱̂ 1 ∂𝐱̂ 1 ⎟ ⎥,
𝑛+ ⎝ 𝑛+
⎣ 2 2⎠ ⎦
where Δ𝐮̂def is the deformation part of the displacement increment in the co-rotational
system 𝐱̂𝑛 + 1 referred to the mid-point configuration.
2

16.3.4 Co-rotational Stress and Strain Updates


For stress and strain updates, we assume that all variables at the previous time step
𝑡𝑛 are known. Since the stress and strain measures defined in the earlier section are
objective in the co-rotational system, we only need to calculate the strain increment from
the displacement field within the time increment [𝑡𝑛 , 𝑡𝑛 + 1 ]. The stress is then updated by
using the radial return algorithm.

All the kinematical quantities must be computed from the last time step
configuration, Ω𝑛 , at 𝑡 = 𝑡𝑛 and the current configuration, Ω𝑛 + 1 at 𝑡 = 𝑡𝑛 + 1 since these
are the only available data. Denoting the spatial coordinates of these two configurations as
𝐱n and 𝐱n + 1 in the fixed global Cartesian coordinate system 𝑂x, as shown in Figure 16.2,
the coordinates in the corresponding co-rotational Cartesian coordinate systems, 𝑂𝐱̂𝑛 and
𝑂𝐱̂𝑛 + 1 , can be obtained by the following transformation rules:
𝐱̂𝑛 = 𝐑𝑛 𝐱𝑛 , (16.73)

𝐱̂𝑛+1 = 𝐑𝑛+1 𝐱𝑛+1 , (16.74)


where 𝐑𝑛 and 𝐑𝑛 + 1 are the orthogonal transformation matrices which rotate the global
coordinate system to the corresponding co-rotational coordinate systems, respectively.

LS-DYNA DEV 10/27/16 (r:8004) 14-9 (Eight-Node Solid Element for Thick Shell Simulations)
Eight-Node Solid Element for Thick Shell Simulations LS-DYNA Theory Manual

Ωn+1

^
Xn+1
Ωn+1/2

Ωn ^
Xn+1/2
^
Xn
O

Figure 16.2. Configurations at times 𝑡𝑛 , 𝑡𝑛 + 1 , and 𝑡𝑛+1 ,


2

Since the strain increment is referred to the configuration at 𝑡 = 𝑡𝑛 + 1 , by assuming


2
the velocities within the time increment [𝑡𝑛 , 𝑡𝑛 + 1 ] are constant, we have
1
𝐱 = (𝐱𝑛 + 𝐱𝑛+1 ), (16.75)
𝑛+1 2
2

and the transformation to the co-rotational system associated with this mid-point
configuration, Ω𝑛 + 1 , is given by
2

𝐱̂ =𝐑 𝐱 . (16.76)
𝑛+1 𝑛+1 𝑛+1
2 2 2

Similar to polar decomposition, an incremental deformation can be separated into


the summation of a pure deformation and a pure rotation [Belytschko, 1973]. Letting Δ𝐮
indicate the displacement increment within the time increment [𝑡𝑛 , 𝑡𝑛 + 1 ], we write
2

Δ𝐮 = Δ𝐮def + Δ𝐮rot , (16.77)


where Δ𝐮def and Δ𝐮rot are, respectively, the deformation part and the pure rotation part of
the displacement increment in the global coordinate system. The deformation part also
includes the translation displacements which cause no strains.

In order to obtain the deformation part of the displacement increment referred to the
configuration at 𝑡 = 𝑡𝑛 + 1 , we need to find the rigid rotation from Ω𝑛 to Ω𝑛 + 1 provided that
2

14-10 (Eight-Node Solid Element for Thick Shell Simulations) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Eight-Node Solid Element for Thick Shell Simulations

the mid-point configuration, Ω𝑛 + 1 , is held still. Defining two virtual configurations, Ω′𝑛
2
and Ω′𝑛 + 1 , by rotating the element bodies Ω𝑛 and Ω𝑛 + 1 into the co-rotational system
̂
𝑂x̂𝑛 + 1 (Fig. 13.3) and denoting and 𝐱′ 𝑛 + 1 as the coordinates of Ω′𝑛 and Ω′𝑛 + 1 in the co-
2
rotational system 𝑂𝐱̂𝑛 + 1 , we have
2

̂ = 𝐱̂ ,
𝐱′ ̂
𝐱′ (16.78)
𝑛 𝑛 𝑛 + 1 = 𝐱̂𝑛 + 1 .

We can see that from Ω𝑛 to Ω′𝑛 and from Ω′𝑛 + 1 to Ω𝑛 + 1 , the body experiences two
rigid rotations and the rotation displacements are given by
T ̂ T
Δ𝐮rot
1 = 𝐱′𝑛 − 𝐱𝑛 = 𝐑 1 𝐱′𝑛 − 𝐱𝑛 = 𝐑 1 𝐱̂𝑛 − 𝐱𝑛 , (16.79)
𝑛+ 𝑛+
2 2

T ̂ T
Δ𝐮rot
2 = 𝐱𝑛+1 − 𝐱′𝑛+1 = 𝐱𝑛+1 − 𝐑 1 𝐱′𝑛+1 = 𝐱𝑛+1 − 𝐑 1 𝐱̂𝑛+1 . (16.80)
𝑛+ 𝑛+
2 2

Thus the total rotation displacement increment can be expressed as


T
Δ𝐮rot = Δ𝐮rot rot
1 + Δ𝐮2 = 𝐱𝑛+1 − 𝐱𝑛 − 𝐑 1 (𝐱̂𝑛+1 − 𝐱̂𝑛 ) 𝑛+
2
(16.81)
= Δ𝐮 − 𝐑T 1 (𝐱̂𝑛+1 − 𝐱̂𝑛 ).
𝑛+
2

Then the deformation part of the displacement increment referred to the


configuration Ω𝑛 + 1 is
2

Δ𝐮def = Δ𝐮 − Δ𝐮rot = 𝐑T 1 (𝐱̂𝑛+1 − 𝐱̂𝑛 ). (16.82)


𝑛+
2

Therefore, the deformation displacement increment in the co-rotational coordinate


system 𝑂x̂𝑛 + 1 is obtained as
2

Δ𝐮̂def = 𝐑 Δ𝐮def = 𝐱̂𝑛+1 − 𝐱̂𝑛 . (16.83)


𝑛+1
2

Once the strain increment is obtained by equation (16.72), the stress increment, also
referred to the mid-point Configuration, can be calculated with the radial return algorithm.
The total strain and stress can then be updated as
𝛆̂𝑛+1 = 𝛆̂𝑛 + Δ𝛆̂, (16.84)

𝛔
̂𝑛+1 = 𝛔
̂𝑛 + Δ𝛔
̂. (16.85)

LS-DYNA DEV 10/27/16 (r:8004) 14-11 (Eight-Node Solid Element for Thick Shell Simulations)
Eight-Node Solid Element for Thick Shell Simulations LS-DYNA Theory Manual

^
Ωn+1 Xn+1
Pn+1
,
Δu2rot Ωn+1
,
Pn+1
^
Δu Xn+1/2

Ωn+1/2
Δudef

,
Pn
,
Ωn Δu1rot
^
Ωn Xn
Pn
O

Figure 16.3. Separation of the displacement increment

Note that the resultant stress and strain tensors are both referred to the current
configuration and defined in the current co-rotational coordinate system. By using the
tensor transformation rule we can have the strain and stress components in the global
coordinate system.

Tangent Stiffness Matrix and Nodal Force Vectors


From the Hu-Washizu variational principle, at both 𝑣th and (𝑣 + 1)th iteration, we
have

∫ 𝑣 𝛿𝜀̂𝑣𝑖𝑗 𝜎̂ 𝑖𝑗𝑣 𝑑𝑉 = 𝛿𝜋̂ext


𝑣
, (16.86)
̂
Ω

∫ 𝛿𝜀̂𝑣+1 𝑣+1
𝑖𝑗 𝜎̂ 𝑖𝑗
𝑣+1
𝑑𝑉 = 𝛿𝜋̂ext , (16.87)
̂ 𝑣+1
Ω

where 𝛿𝜋̂ext is the virtual work done by the external forces. Note that both equations are
written in the co-rotational coordinate system defined in the 𝑣th iterative configuration
𝑣
given by x𝑛+1 . The variables in this section are within the time step [𝑡𝑛 , 𝑡𝑛+1 ] and
2
superscripts indicate the number of iterations.

Assuming that all external forces are deformation-independent, linearization of


Equation (16.87) gives [Liu, 1992]

14-12 (Eight-Node Solid Element for Thick Shell Simulations) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Eight-Node Solid Element for Thick Shell Simulations

𝑣 𝑣
∫ 𝑣 𝛿𝑢̂𝑣𝑖,𝑗 𝐶𝑖𝑗𝑘𝑙
̂ Δ𝑢̂𝑘,𝑙 𝑑𝑉 + ∫ 𝛿𝑢̂𝑣𝑖,𝑗 𝑇̂𝑖𝑗𝑘𝑙 𝑣+1
Δ𝑢̂𝑘,𝑙 𝑑𝑉 = 𝛿𝜋̂ext 𝑣
− 𝛿𝜋̂ext , (16.88)
̂
Ω 𝑣 ̂
Ω

where the Green-Naghdi rate of Cauchy stress tensor is used, i.e.,


𝑣
𝑇̂𝑖𝑗𝑘𝑙 = 𝛿𝑖𝑘 𝜎̂ 𝑗𝑙𝑣 . (16.89)

The first term on the left hand side of (16.88) denotes the material response since it is
due to pure deformation or stretching; the second term is an initial stress part resulting
from finite deformation effect.

Taking account of the residual of the previous iteration, Equation (16.87) can be
approximated as
𝑣 𝑣
∫ 𝑣 𝛿𝑢̂𝑣𝑖,𝑗 (𝐶𝑖𝑗𝑘𝑙
̂ + 𝑇̂𝑖𝑗𝑘𝑙 𝑣+1
)Δ𝑢̂𝑘,𝑙 𝑑𝑉 = 𝛿𝜋̂ext − ∫ 𝑣 𝛿𝜀̂𝑣𝑖𝑗 𝜎̂ 𝑖𝑗𝑣 𝑑𝑉 . (16.90)
̂
Ω ̂
Ω

If the strain and stress vectors are defined as


𝛆 T = [ 𝜀𝑥 𝜀𝑦 𝜀𝑧 2𝜀𝑥𝑦 2𝜀𝑦𝑧 2𝜀𝑧𝑥 2𝜔𝑥𝑦 2𝜔𝑦𝑧 2𝜔𝑧𝑥 ], (16.91)

𝛔T = [𝜎𝑥 𝜎𝑦 𝜎𝑧 𝜎𝑥𝑦 𝜎𝑦𝑧 𝜎𝑧𝑥 ], (16.92)

We can rewrite equation (16.90) as

∫ 𝛿𝜀̂𝑣𝑖 (𝐶𝑖𝑗𝑣̂ + 𝑇̂𝑖𝑗𝑣 )𝛿𝜀̂𝑗 𝑑𝑉 = 𝛿𝜋̂ext


𝑣+1
− ∫ 𝑣 𝛿𝜀̂𝑣𝑖 𝜎̂ 𝑗𝑣 𝑑𝑉 , (16.93)
̂
𝑣
Ω ̂
Ω

where 𝐶𝑖𝑗𝑣̂ is the consistent tangent modulus tensor corresponding to pure deformation (see
Section 3.2.3) but expanded to a 9 by 9 matrix; 𝑇̂𝑖𝑗𝑣 is the geometric stiffness matrix which is
given as follows [Liu 1992]:

LS-DYNA DEV 10/27/16 (r:8004) 14-13 (Eight-Node Solid Element for Thick Shell Simulations)
Eight-Node Solid Element for Thick Shell Simulations LS-DYNA Theory Manual

T =
𝜎4 𝜎6 𝜎4 𝜎6
⎡𝜎1 0 0
2
0
2 2
0 −
2 ⎤
⎢ 𝜎4 𝜎5 𝜎 𝜎5 ⎥
⎢ 𝜎2 0 0 − 4 0 ⎥
⎢ 2 2 2 2 ⎥
⎢ 𝜎5 𝜎6 𝜎 𝜎6 ⎥
⎢ 𝜎3 0 0 − 5 ⎥
⎢ 2 2 2 2 ⎥
⎢ 𝜎1 + 𝜎2 𝜎6 𝜎5 𝜎2 − 𝜎1 𝜎6 𝜎5 ⎥
⎢ − ⎥
⎢ 4 4 4 4 4 4 ⎥
⎢ 𝜎2 + 𝜎3 𝜎4 𝜎 𝜎3 − 𝜎2 𝜎4 ⎥ (16.94)
⎢ − 6 ⎥.
⎢ 4 4 4 4 4 ⎥
⎢ 𝜎1 + 𝜎3 𝜎5 𝜎 𝜎1 − 𝜎3 ⎥
⎢ − 4 ⎥
⎢ 4 4 4 4 ⎥
⎢ 𝜎1 + 𝜎2 𝜎6 𝜎5 ⎥
⎢ symm. − − ⎥
⎢ 4 4 4 ⎥
⎢ 𝜎2 + 𝜎3 𝜎 ⎥
⎢ − 4 ⎥
⎢ 4 4 ⎥
⎢ 𝜎3 + 𝜎1 ⎥
⎣ 4 ⎦

By interpolation
Δ𝐮 = 𝐍Δ𝐝, 𝛿𝐮 = 𝐍𝛿𝐝; (16.95)

̅̅̅̅̅Δ𝐝,
Δ𝛆 = 𝐁 𝛿𝛆 = ̅𝐁
̅̅̅̅𝛿𝐝, (16.96)
where 𝐍 and ̅𝐁 ̅̅̅̅ are, respectively, the shape functions and strain operators defined in
Section 2. This leads to a set of equations
̂ 𝑣 Δ𝐝̂ = 𝐫 𝑣+1
𝐊 ̂ ̂ − 𝐟int
𝑣+1
= 𝐟ext 𝑣̂
, (16.97)
𝑣̂
̂ 𝑣 , and the internal nodal force vector, 𝐟int
where the tangent stiffness matrix, 𝐊 , are

𝐊 ̂̅̅̅̅T (𝐂𝑣̂ + 𝐓
̂ 𝑣 = ∫ ̅𝐁 ̅̂̅̅̅̅dV,
̂ 𝑣 )𝐁 (16.98)
𝑣 ̂
Ω

𝑣̂
𝐟int ̂̅̅̅̅T 𝛔
= ∫ 𝑣̅𝐁 ̂𝑣 dV. (16.99)
̂
Ω

The tangent stiffness and nodal force are transformed into the global coordinate
system tensorially as
𝐊𝑣 = 𝐑𝑣T 𝐊
̂ 𝑣 𝐑𝑣 , (16.100)
𝑣
𝐫 𝑣+1 = 𝐑𝑣T 𝐫int
̂ , (16.101)
𝑣
where 𝐑𝑣 is the transformation matrix of the co-rotational system defined by 𝐱𝑛+1 . Finally,
we get a set of linear algebraic equations

14-14 (Eight-Node Solid Element for Thick Shell Simulations) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Eight-Node Solid Element for Thick Shell Simulations

𝐊𝑣 Δ𝐝𝑣+1 = 𝐫 𝑣+1 . (16.102)

16.4 Numerical Examples


To investigate the performance of the element introduced in this paper, a variety of
problems including linear elastic and nonlinear elastic-plastic/large deformation problems
are studied. Since the element is developed to avoid locking, the applicability to problems
of thin structures is studied by solving the standard test problems including pinched
cylinder and Scordelis-Lo roof, which are proposed by MacNeal, 1985 and Belytschko,
1984b. Also a sheet metal forming problem is solved to test and demonstrate the
effectiveness and efficiency of this element.

16.4.1 Timoshenko Cantilever Beam


The first problem is a linear, elastic cantilever beam with a load at its end as shown
in Fig. 16.4, where 𝑀 and 𝑃 at the left end of the cantilever are reactions at the support. The
analytical solution from Timoshenko, 1970 is
−𝑃𝑦 1
𝑢𝑥 (𝑥, 𝑦) = [(6𝐿 − 3𝑥)𝑥 + (2 + 𝑣̅) (𝑦2 − 𝐷2 )], (16.103)
̅̅̅̅𝐼
6𝐸 4

𝑃 1
𝑢𝑦 (𝑥, 𝑦) = [3𝑣̅𝑦2 (𝐿 − 𝑥) + (4 + 5𝑣̅)𝐷2 𝑥 + (3𝐿 − 𝑥)𝑥2 ], (16.104)
̅̅̅̅𝐼
6𝐸 4
where
1 3
𝐼= 𝐷 , (16.105)
12

̅𝐸̅̅̅ = {𝐸, { 𝑣
⎧ for plane stress
𝑣̅ = ⎨ 𝑣 (16.106)
𝐸/(1 − 𝑣2 ) , {
⎩1 − 𝑣 for plane strain

The displacements at the support end, 𝑥 = 0, − 12 𝐷 ≤ 𝑦 ≤ 12 𝐷 are nonzero except at


the top, bottom and midline (as shown in Fig. 13.5). Reaction forces are applied at the
support based on the stresses corresponding to the displacement field at 𝑥 = 0, which are
𝑃𝑦 𝑃 1 2
𝜎𝑥𝑥 = − (𝐿 − 𝑥), 𝜎𝑦𝑦 = 0, 𝜎𝑥𝑦 = ( 𝐷 − 𝑦2 ). (16.107)
𝐼 2𝐼 4

The distribution of the applied load to the nodes at 𝑥 = 𝐿 is also obtained from the
closed-form stress fields.

LS-DYNA DEV 10/27/16 (r:8004) 14-15 (Eight-Node Solid Element for Thick Shell Simulations)
Eight-Node Solid Element for Thick Shell Simulations LS-DYNA Theory Manual

M P x
P D

L
Figure 16.4. Timoshenko cantilever beam.

p/2
D/2 x

L Pt. A

(a) Regular mesh

y
θ

p/2
D/2 x

L/4 Pt. A

(b) Skewed mesh


Figure 16.5. Top half of anti-symmetric beam mesh

The parameters for the cantilever beam are: 𝐿 = 1.0, 𝐷 = 0.02, 𝑃 = 2.0, 𝐸 = 1 × 107 ;
and two values of Poisson’s ratio: (1)𝑣 = 0.25, (2)𝑣 = 0.4999.

Since the problem is anti-symmetric, only the top half of the beam is modeled. Plane
strain conditions are assumed in the z-direction and only one layer of elements is used in
this direction. Both regular mesh and skewed mesh are tested for this problem.

Normalized vertical displacements at point A for each case are given in Table 13.1.
Tables 13.1a and 13.1b show the normalized displacement at point A for the regular mesh.

14-16 (Eight-Node Solid Element for Thick Shell Simulations) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Eight-Node Solid Element for Thick Shell Simulations

There is no shear or volumetric locking for this element. For the skewed mesh, with the
skewed angle increased, we need more elements to get more accurate solution (Table
13.1c).
(a) 𝑣 = 0.25, regular mesh
Analytical solution 𝑤A = 9.3777 × 10−2

Mesh 411 811 821

HEXDS 1.132 1.142 1.029

(b) 𝑣 = 0.4999, regular mesh


Analytical solution 𝑤A = 7.5044 × 10−2

Mesh 411 811 821

HEXDS 1.182 1.197 1.039

(c) 𝑣 = 0.25, skewed mesh

1 5 10

4  1 1 1.078 0.580 0.317

8  1 1 1.136 0.996 0.737

16  1 1 1.142 1.090 .955

Table 13.1. Normalized displacement at point A of cantilever beam.

LS-DYNA DEV 10/27/16 (r:8004) 14-17 (Eight-Node Solid Element for Thick Shell Simulations)
Eight-Node Solid Element for Thick Shell Simulations LS-DYNA Theory Manual

16.4.2 Pinched Cylinder


Figure 16.6 shows a pinched cylinder subjected to a pair of concentrated loads. Two
cases are studied in this example. In the first case, both ends of the cylinder are assumed to
be free. In the second case, both ends of the cylinder are covered with rigid diaphragms so
that only the displacement in the axial direction is allowed at the ends. The parameters for
the first case
(without diaphragms) are
𝐸 = 1.05 × 106 , 𝑣 = 0.3125, 𝐿 = 10.35, 𝑅 = 1.0, 𝑡 = 0.094, 𝑃 = 100.0; (16.108)
while for the second case (with diaphragms), the parameters are set to be
𝐸 = 3 × 106 , 𝑣 = 0.3, 𝐿 = 600.0, 𝑅 = 300.0, 𝑡 = 3.0, 𝑃 = 1.0. (16.109)

Due to symmetry only one octant of the cylinder is modeled. The computed
displacements at the loading point are compared to the analytic solutions in Table 13.2.
HEXDS element works well in both cases, indicating that this element can avoid not only
shear locking but also membrane locking; this is not unexpected since membrane locking
occurs primarily in curved elements [Stolarski, 1983].

16.4.3 Scordelis-Lo Roof


Scordelis-Lo roof subjected to its own weight is shown in Figure 16.7. Both ends of
the roof are assumed to be covered with rigid diaphragms. The parameters are selected to
be: 𝐸 = 4.32 × 108 , 𝑣 = 0.0, 𝐿 = 50.0, 𝑅 = 25.0, 𝑡 = 0.25, 𝜃 = 40∘ , and the gravity is
360.0 per volume.

free or with diaphragm


ric
et
m
m
sy

p
R L
ric
et
m

P
m

symmetric
sy

t
Figure 16.6. Pinched cylinder and the element model

14-18 (Eight-Node Solid Element for Thick Shell Simulations) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Eight-Node Solid Element for Thick Shell Simulations

t
2L

R
θ

Figure 16.7. Scordelis-Lo roof under self-weight

(a) First case without diaphragms


Analytical solution 𝑤max = 0.1137

Mesh 10  10  2 16  16  4 20  20  4

HEXDS 1.106 1.054 1.067

(b) Second case with diaphragms


Analytical solution wmax = 1.8248 × 10−5

Mesh 10  10  2 16  16  4 20  20  4

HEXDS 0.801 0.945 .978

Table 13.2. Normalized displacement at loading point of pinched cylinder

Due to symmetry only one quarter of the roof is modeled. The computed
displacement at the midpoint of the edge is compared to the analytic solution in Table 13.3.
In this example the HEXDS element can get good result with 100 × 2 elements.

LS-DYNA DEV 10/27/16 (r:8004) 14-19 (Eight-Node Solid Element for Thick Shell Simulations)
Eight-Node Solid Element for Thick Shell Simulations LS-DYNA Theory Manual

R0=50.8mm

tight die
t=2mm r=2mm
Rd=54mm

R=54mm

Figure 16.8. Circular sheet stretched with a tight die

Analytical solution 𝑤max = 0.3024

Mesh 881 16  16  1 32  32  1 10  10  2

HEXDS 1.157 1.137 1.132 1.045

16.4.4 Circular Sheet Stretched with a Tight Die


A circular sheet is stretched under a hemisphere punch and a tight die with a small
corner radius (Fig. 16.8). The material is elastoplastic with nonlinear hardening rule. The
elastic material constants are: 𝐸 = 206 GPa and 𝑣 = 0.3. In the plastic range, the uniaxial
stress-strain curve is given by
𝜎 = 𝐾𝜀𝑛 , (16.110)
where 𝐾 = 509.8MPa, 𝑛 = 0.21, 𝜎 is Cauchy stress and 𝜀 is natural strain (logarithmic
strain). The initial yield stress is obtained to be 𝜎0 = 103.405Mpa and the tangent modulus
at the initial yield point is 𝐸t = 0.4326 × 105 MPa.

Because of the small corner radius of the die, the same difficulties as in the problem
of sheet stretch under the rigid cylinders lead the shell elements to failure in this problem.
Three-dimensional solid elements are needed and fine meshes should be put in the areas
near the center and the edge of the sheet.

One quarter of the sheet is modeled with 1400  2 HEXDS elements due to the
double symmetries. The mesh is shown in Fig. 16.9. Two layers of elements are used in the
thickness. Around the center and near the circular edge of the sheet, fine mesh is used.
The nodes on the edge are fixed in x- and y-directions and the bottom nodes on the edge
are prescribed in three directions. No friction is considered in this simulation. For

14-20 (Eight-Node Solid Element for Thick Shell Simulations) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Eight-Node Solid Element for Thick Shell Simulations

Figure 16.9. Mesh for circular sheet stretching

comparison, the axisymmetric four-node element with reduced integration (CAX4R) is also
used and the mesh for this element is the same as shown in the top of Figure 13.9.

The results presented here are after the punch has traveled down 50 mm. The
profile of the circular sheet is shown in Figure 16.10 where we can see that the sheet under
the punch experiences most of the stretching and the thickness of the sheet above the die

Figure 16.10. Deformed shape of a circular sheet with punch travel 50 mm

changes a lot. The deformation between the punch and the die is small. However, the
LS-DYNA DEV 10/27/16 (r:8004) 14-21 (Eight-Node Solid Element for Thick Shell Simulations)
Eight-Node Solid Element for Thick Shell Simulations LS-DYNA Theory Manual

Figure 16.11. Reaction force vs. punch travel for the circular sheet

sheet thickness obtained by the CAX4R element is less than that by the HEXDS element
and there is slight difference above the die. These observations can be verified by the strain
distributions in the sheet along the radial direction (Figure 13.12). The direction of the
radial strain is the tangent of the mid-surface of the element in the rz plane and the
thickness strain is in the direction perpendicular to the mid-surface of the element. The
unit vector of the circumferential strain is defined as the cross-product of the directional
cosine vectors of the radial strain and the thickness strain. We can see that the CAX4R
element yields larger strain components in the area under the punch than the HEXDS
element. The main difference of the strain distributions in the region above the die is that
the CAX4R element gives zero circumferential strain in this area but the HEXDS element
yields non-zero strain. The value of the reaction force shown in the Figure 13.11 is only one
quarter of the total punch reaction force since only one quarter of the sheet is modeled.
From this figure we can see that the sheet begins softening after the punch travels down
about 45 mm, indicating that the sheet may have necking though this cannot be seen clearly
from Figure 16.10.

(a) Radial strain distribution


14-22 (Eight-Node Solid Element for Thick Shell Simulations) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Eight-Node Solid Element for Thick Shell Simulations

(b) Circumferential strain distribution

(c) Thickness strain distribution

Figure 13.12. Strain distributions in circular sheet with punch travel 50 mm

16.5 Conclusions
A new eight-node hexahedral element is implemented for the large deformation
elastic-plastic analysis. Formulated in the co-rotational coordinate system, this element is
shown to be effective and efficient and can achieve fast convergence in solving a wide
variety of nonlinear problems.

By using a co-rotational system which rotates with the element, the locking
phenomena can be suppressed by omitting certain terms in the generalized strain
operators. In addition, the integration of the constitutive equation in the co-rotational

LS-DYNA DEV 10/27/16 (r:8004) 14-23 (Eight-Node Solid Element for Thick Shell Simulations)
Eight-Node Solid Element for Thick Shell Simulations LS-DYNA Theory Manual

system takes the same simple form as small deformation theory since the stress and strain
tensors defined in this co-rotational system are objective.

Radial return algorithm is used to integrate the rate-independent elastoplastic


constitutive equation. The tangent stiffness matrix consistently derived from this
integration scheme is crucial to preserve the second order convergence rate of the Newton’s
iteration method for the nonlinear static analyses.

Test problems studied in this paper demonstrate that the element is suitable to
continuum and structural numerical simulations. In metal sheet forming analysis, this
element has advantages over shell elements for certain problems where through the
thickness deformation and strains are significant.

14-24 (Eight-Node Solid Element for Thick Shell Simulations) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Truss Element

17
Truss Element

One of the simplest elements is the pin-jointed truss element shown in Figure 17.1.
This element has three degrees of freedom at each node and carries an axial force. The
displacements and velocities measured in the local system are interpolated along the axis
according to
𝑥
𝑢 = 𝑢1 + (𝑢2 − 𝑢1 ), (17.1)
𝐿
𝑥
𝑢̇ = 𝑢̇1 + (𝑢̇2 − 𝑢̇1 ), (17.2)
𝐿
where at 𝑥 = 0, 𝑢 = 𝑢1 and at 𝑥 = 𝐿, 𝑢 = 𝑢2 . Incremental strains are found from
(𝑢̇2 − 𝑢̇1 )
Δ𝜀 = Δ𝑡 (17.3)
𝐿
and are computed in LS-DYNA using
𝑛+1⁄2 𝑛+1⁄2
2 (𝑢̇2 − 𝑢̇1 )
𝑛+1⁄2 𝑛+1⁄2 (17.4)
Δ𝜀 = Δ𝑡
𝐿𝑛 + 𝐿𝑛+1

The normal force 𝑁 is then incrementally updated using a tangent modulus 𝐸𝑡


according to
𝑁 𝑛+1 = N𝑛 𝐴𝐸𝑡 + Δ𝜀𝑛+1/2 (17.5)

LS-DYNA DEV 10/27/16 (r:8004) 15-1 (Truss Element)


Truss Element LS-DYNA Theory Manual

N1

u1

N2

u2
Figure 17.1. Truss element.

Two constitutive models are implemented for the truss element: elastic and elastic-
plastic with kinematic hardening.

15-2 (Truss Element) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Membrane Element

18
Membrane Element

The Belytschko-Lin-Tsay shell element {Belytschko and Tsay [1981], Belytschko et


al., [1984a]} is the basis for this very efficient membrane element. In this section we briefly
outline the theory employed which, like the shell on which it is based, uses a combined co-
rotational and velocity-strain formulation. The efficiency of the element is obtained from
the mathematical simplifications that result from these two kinematical assumptions. The
co-rotational portion of the formulation avoids the complexities of nonlinear mechanics by
embedding a coordinate system in the element. The choice of velocity strain or rate of
deformation in the formulation facilitates the constitutive evaluation, since the conjugate
stress is the more familiar Cauchy stress.

In membrane elements the rotational degrees of freedom at the nodal points may be
constrained, so that only the translational degrees-of-freedom contribute to the straining of
the membrane. A triangular membrane element may be obtained by collapsing adjacent
nodes of the quadrilateral.

18.1 Co-rotational Coordinates


The mid-surface of the quadrilateral membrane element is defined by the location of
the element’s four corner nodes. An embedded element coordinate system (Figure 7.1) that
deforms with the element is defined in terms of these nodal coordinates. The co-rotational
coordinate system follows the development in Section 7, Equations (7.1)—(7.3).

LS-DYNA DEV 10/27/16 (r:8004) 16-1 (Membrane Element)


Membrane Element LS-DYNA Theory Manual

18.2 Velocity-Strain Displacement Relations


The co-rotational components of the velocity strain (rate of deformation) are given
by:
1 ∂𝜐̂ ∂𝜐̂𝑗
𝑑𝑖𝑗̂ = ( 𝑖 + ), (18.1)
2 ∂𝑥̂𝑗 ∂𝑥̂𝑖

The above velocity-strain relations are evaluated only at the center of the shell.
Standard bilinear nodal interpolation is used to define the mid-surface velocity, angular
velocity, and the element’s coordinates (isoparametric representation). These interpolation
relations are given by
𝑣𝑚 = 𝑁𝐼 (𝜉 , 𝜂)𝑣𝐼 , (18.2)

𝑥𝑚 = 𝑁𝐼 (𝜉 , 𝜂)𝑥𝐼 , (18.3)
where the subscript 𝐼 is summed over all the element’s nodes and the nodal velocities are
obtained by differentiating the nodal coordinates with respect to time, i.e., 𝜐𝐼 = ẋ𝐼 . The
bilinear shape functions are defined in Equations (7.10).

The velocity strains at the center of the element, i.e., at 𝜉 = 0, and 𝜂 = 0, are obtained
as in Section 7 giving:
𝑑𝑥̂ = 𝐵1𝐼 𝜐̂𝑥𝐼 , (18.4)

𝑑𝑦̂ = 𝐵2𝐼 𝜐̂𝑦𝐼 , (18.5)

̂ = 𝐵2𝐼 𝑣̂𝑥𝐼 + 𝐵1𝐼 𝑣̂𝑦𝐼 ,


2𝑑𝑥𝑦 (18.6)
where
∂𝑁𝐼
𝐵1𝐼 = , (18.7)
∂𝑥̂
∂𝑁𝐼
𝐵2𝐼 = . (18.8)
∂𝑦̂

18.3 Stress Resultants and Nodal Forces


After suitable constitutive evaluations using the above velocity strains, the resulting
stresses are multiplied by the thickness of the membrane, h, to obtain local resultant forces.
Therefore,

16-2 (Membrane Element) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Membrane Element

𝑅̂ = ℎ𝜎̂ 𝛼𝛽 ,
𝑓𝛼𝛽 (18.9)
where the superscript R indicates a resultant force and the Greek subscripts emphasize the
limited range of the indices for plane stress plasticity.

The above element centered force resultants are related to the local nodal forces by
𝑓𝑥𝐼̂ = 𝐴(𝐵1𝐼 𝑓𝑥𝑥
𝑅̂ + 𝐵2𝐼 𝑓𝑥𝑦
𝑅̂ ), (18.10)

𝑓𝑦𝐼̂ = 𝐴(𝐵2𝐼 𝑓𝑦𝑦


𝑅̂ + 𝐵1𝐼 𝑓𝑥𝑦
𝑅̂ ), (18.11)
where 𝐴 is the area of the element.

The above local nodal forces are then transformed to the global coordinate system
using the transformation relations given in Equation (7.5a).

18.4 Membrane Hourglass Control


Hourglass deformations need to be resisted for the membrane element. The
hourglass control for this element is discussed in Section 7.4.

LS-DYNA DEV 10/27/16 (r:8004) 16-3 (Membrane Element)


LS-DYNA Theory Manual Discrete Elements and Masses

19
Discrete Elements and Masses

The discrete elements and masses in LS-DYNA provide a capability for modeling
simple spring-mass systems as well as the response of more complicated mechanisms.
Occasionally, the response of complicated mechanisms or materials needs to be included in
LS-DYNA models, e.g., energy absorbers used in passenger vehicle bumpers. These
mechanisms are often experimentally characterized in terms of force-displacement curves.
LS-DYNA provides a selection of discrete elements that can be used individually or in
combination to model complex force-displacement relations.

The discrete elements are assumed to be massless. However, to solve the equations
of motion at unconstrained discrete element nodes or nodes joining multiple discrete
elements, nodal masses must be specified at these nodes. LS-DYNA provides a direct
method for specifying these nodal masses in the model input.

All of the discrete elements are two-node elements, i.e., three-dimensional springs or
trusses. A discrete element may be attached to any of the other LS-DYNA continuum,
structural, or rigid body element. The force update for the discrete elements may be
written as
𝐟𝑖̂ +1 = 𝐟𝑖̂ + Δ𝐟,̂ (19.1)
where the superscript 𝑖 + 1 indicates the time increment and the superposed caret (⋅ ̂)
indicates the force in the local element coordinates, i.e., along the axis of the element. In the
default case, i.e., no orientation vector is used; the global components of the discrete
element force are obtained by using the element’s direction cosines:
⎧𝐹 ⎧Δ𝑙 ⎫ ⎧𝑛𝑥 ⎫
{ 𝑥⎫} 𝑓 ̂ { 𝑥} ̂{𝑛𝑦 } = 𝑓 𝐧̂ ,
⎨𝐹𝑦 ⎬ = ⎨Δ𝑙𝑦 ⎬ = 𝑓 ⎨ ⎬ (19.2)
{ 𝑙 { } {
⎩𝑛𝑧 }
~
⎩𝐹𝑧 }
⎭ ⎩Δ𝑙𝑧 ⎭ ⎭

where

16.1
Discrete Elements and Masses LS-DYNA Theory Manual

⎧ Δ𝑙
{ 𝑥⎫ } ⎧𝑥 −𝑥
{𝑦2 − 𝑦1 ⎫
}
Δ𝐥 = ⎨Δ𝑙𝑦 ⎬ = ⎨ 2 1 ⎬. (19.3)
{Δ𝑙 } {
⎩𝑧2 − 𝑧1 }

⎩ 𝑧⎭
𝑙 is the length

𝑙 = √Δ𝑙2𝑥 + Δ𝑙2𝑦 + Δ𝑙2𝑧 , (19.4)

and (𝑥𝑖 , 𝑦𝑖 , 𝑧𝑖 ) are the global coordinates of the nodes of the spring element. The forces in
Equation (19.2) are added to the first node and subtracted from the second node.

For a node tied to ground we use the same approach but for the (𝑥2 , 𝑦2 , 𝑧2 )
coordinates in Equation (19.2) the initial coordinates of node 1, i.e., (𝑥0 , 𝑦0 , 𝑧0 ) are used
instead; therefore,
⎧ 𝐹 𝑥 −𝑥 𝑛𝑥 ⎫
{ 𝑥⎫} 𝑓̂⎧ {𝑦0 − 𝑦1 ⎫
} ⎧
̂ 𝑛𝑦 }.
{
⎨ 𝐹𝑦⎬ = ⎨ 0 1⎬ = 𝑓 ⎨ ⎬ (19.5)
{ 𝑙 {
⎩𝑧0 − 𝑧1 } {
⎩𝑛𝑧 }
⎩𝐹𝑧 }
⎭ ⎭ ⎭

The increment in the element force is determined from the user specified force-
displacement relation. Currently, nine types of force-displacement/velocity relationships
may be specified:
1. linear elastic;
2. linear viscous;
3. nonlinear elastic;
4. nonlinear viscous;
5. elasto-plastic with isotropic hardening;
6. general nonlinear;
7. linear viscoelastic.
8. inelastic tension and compression only.
9. muscle model.

The force-displacement relations for these models are discussed in the following
later.

19.1 Orientation Vectors


An orientation vector,
⎧ 𝑚
{𝑚 1 ⎫
}
𝐦 = ⎨ 2 ⎬, (19.6)
{
⎩𝑚 3 }

can be defined to control the direction the spring acts. If orientation vectors are used, it is
strongly recommended that the nodes of the discrete element be coincident and remain
16.2
LS-DYNA Theory Manual Discrete Elements and Masses

approximately so throughout the calculation. If the spring or damper is of finite length,


rotational constraints will appear in the model that can substantially affect the results. If
finite length springs are needed with directional vectors, then the discrete beam elements,
the type 6 beam, should be used with the coordinate system flagged for the finite length
case.

We will first consider the portion of the displacement that lies in the direction of the
vector. The displacement of the spring is updated based on the change of length given by
Δ𝐼 = 𝐼 − 𝐼0 , (19.7)
where 𝐼0 is the initial length in the direction of the vector and lis the current length given
for a node to node spring by
𝐼 = 𝑚1 (𝑥2 − 𝑥1 ) + 𝑚2 (𝑦2 − 𝑦1 ) + 𝑚3 (𝑧2 − 𝑧1 ), (19.8)
and for a node to ground spring by
𝐼 = 𝑚1 (𝑥0 − 𝑥1 ) + 𝑚2 (𝑦0 − 𝑦1 ) + 𝑚3 (𝑧0 − 𝑧1 ), (19.9)
The latter case is not intuitively obvious and can affect the sign of the force in unexpected
ways if the user is not familiar with the relevant equations. The nodal forces are then given
by
⎧ 𝐹 ⎧𝑚1 ⎫
{ 𝑥⎫ } ̂{𝑚2 }.
⎨ 𝐹𝑦 ⎬ = 𝑓 ⎨ ⎬ (19.10)
{ {
⎩𝑚3 }
⎩𝐹𝑧 }⎭ ⎭

The orientation vector can be either permanently fixed in space as defined in the
input or acting in a direction determined by two moving nodes which must not be
coincident but may be independent of the nodes of the spring. In the latter case, we
recompute the direction every cycle according to:

⎧ 𝑚 ⎧ 𝑥𝑛 − 𝑥1𝑛 ⎫
{𝑚 1 ⎫
} 1 { 2𝑛 }
⎨ 2⎬ = 𝑛 ⎨ 𝑦2 − 𝑦1𝑛 ⎬. (19.11)
{
⎩𝑚 3 }
⎭ 𝑙 {
⎩𝑧𝑛2 − 𝑧𝑛1 }

In Equation (19.9) the superscript, 𝑛, refers to the orientation nodes.

For the case where we consider motion in the plane perpendicular to the orientation
vector we consider only the displacements in the plane, Δ𝑙𝑝 , given by,
Δ𝐥𝑝 = Δ𝐥 − 𝐦(𝐦 ⋅ Δ𝐥). (19.12)
We update the displacement of the spring based on the change of length in the plane given
by
𝑝
Δ𝑙𝑝 = 𝑙𝑝 − 𝑙0 , (19.13)
𝑝
where 𝑙0 is the initial length in the direction of the vector and 𝑙 is the current length given
for a node to node spring by

16.3
Discrete Elements and Masses LS-DYNA Theory Manual

𝑝 𝑝 𝑝
𝑙𝑝 = 𝑚1 (𝑥2 − 𝑥1 ) + 𝑚2 (𝑦2 − 𝑦1 ) + 𝑚3 (𝑧2 − 𝑧1 ), (19.14)
and for a node to ground spring by
𝑝 𝑝 𝑝
𝑙𝑝 = 𝑚1 (𝑥0 − 𝑥1 ) + 𝑚2 (𝑦0 − 𝑦1 ) + 𝑚3 (𝑧0 − 𝑧1 ), (19.15)
where
⎧𝑚𝑝1 ⎫ ⎧Δ𝑙 𝑝

{
{ 𝑝} } 1 {
{ 𝑥𝑝 }}
⎨𝑚 2 ⎬ = ⎨Δ𝑙𝑦 ⎬. (19.16)
{ 𝑝}
{ } 𝑝2 𝑝2 { 𝑝}
𝑝2 { }
⎩𝑚3 ⎭ √Δ𝑙𝑥 + Δ𝑙𝑦 + Δ𝑙𝑧 ⎩Δ𝑙𝑧 ⎭
After computing the displacements, the nodal forces are then given by

⎧ 𝐹 ⎧𝑚𝑝1 ⎫
{ 𝑥⎫ } {
{ }
̂ 𝑚𝑝 .
}
⎨ 𝐹𝑦 ⎬ = 𝑓 ⎨ 2⎬ (19.17)
{
⎩𝐹𝑧 }⎭ { 𝑝}
{ }
⎩𝑚3 ⎭

19.2 Dynamic Magnification “Strain Rate” Effects


To account for “strain rate” effects, we have a simple method of scaling the forces
based on the relative velocities that applies to all springs. The forces computed from the
spring elements are assumed to be the static values and are scaled by an amplification
factor to obtain the dynamic value:
𝑉
𝐹dynamic = (1. + 𝑘𝑑 )𝐹 , (19.18)
𝑉0 static
where
𝑘𝑑 = is a user defined input value
𝑉 = absolute relative velocity
𝑉0 = dynamic test velocity

For example, if it is known that a component shows a dynamic crush force at 15 m/s
equal to 2.5 times the static crush force, use 𝑘𝑑 = 1.5 and 𝑉0 = 15.

19.3 Deflection Limits in Tension and Compression


The deflection limit in compression and tension is restricted in its application to no
more than one spring per node subject to this limit, and to deformable bodies only. For
example in the former case, if three spring are in series either the center spring or the two
end springs may be subject to a limit but not all three. When the limiting deflection is

16.4
LS-DYNA Theory Manual Discrete Elements and Masses

reached momentum conservation calculations are performed and a common acceleration is


computed:
𝑓1̂ + 𝑓2̂
𝑎common
̂ = . (19.19)
𝑚1 + 𝑚2
An error termination will occur if a rigid body node is used in a spring definition where
compression is limited.

19.4 Linear Elastic or Linear Viscous


These discrete elements have the simplest force-displacement relations. The linear
elastic discrete element has a force-displacement relation of the form
𝑓 ̂ = 𝐾Δ𝑙, (19.20)
where 𝐾 is the element’s stiffness and Δ𝑙 is the change in length of the element. The linear
viscous element has a similar force-velocity (rate of displacement) relation:
Δ𝑙
𝑓 ̂= 𝐶 . (19.21)
Δ𝑡
where 𝐶 is a viscous damping parameter and Δ𝑡 is the time step increment.

19.5 Nonlinear Elastic or Nonlinear Viscous


These discrete elements use piecewise force-displacement or force-relative velocity
relations. The nonlinear elastic discrete element has a tabulated force-displacement
relation of the form
𝑓 ̂ = 𝐾Δ𝑙, (19.22)
where 𝐾(Δ𝑙) is the tabulated force that depends on the total change in the length of the
element (Figure 19.1) The nonlinear viscous element has a similar tabulated force-relative
velocity relation:
Δ𝑙
𝑓 ̂= 𝐶 , (19.23)
Δ𝑡
Δ𝑙
where 𝐶(Δ𝑡 ) is the viscous damping force that depends on the rate of change of the
element’s length. Nonlinear discrete elements unload along the loading paths.

If the spring element is initially of zero length and if no orientation vectors are used
then only the tensile part of the stress strain curve needs to be defined. However, if the

16.5
Discrete Elements and Masses LS-DYNA Theory Manual

Nolinear Elastic/Viscous

Force

Displacement/Velocity
Figure 19.1. Piecewise linear force-displacement curve for nonlinear elastic
discrete element.

spring element is initially of finite length then the curve must be defined in both the
positive and negative quadrants.

19.6 Elasto-Plastic with Isotropic Hardening


The elasto-plastic discrete element has a bilinear force-displacement relationship
that is specified by the elastic stiffness, a tangent stiffness and a yield force (Figure 19.2).
This discrete element uses the elastic stiffness model for unloading until the yield force is
exceeded during unloading. The yield force is updated to track its maximum value which
is equivalent to an isotropic hardening model. The force-displacement relation during
loading may be written as
𝐾
𝑓 ̂ = 𝐹𝑦 (1 − 𝑡 ) + 𝐾𝑡 Δ𝑙, (19.24)
𝐾
where 𝐹𝑦 is the yield force and 𝐾𝑡 is the tangent stiffness.

19.7 General Nonlinear


The general nonlinear discrete element allows the user to specify independent and
nonsymmetrical piecewise linear loading and unloading paths (Figure 19.3(a)).

16.6
LS-DYNA Theory Manual Discrete Elements and Masses

Elsto-Plastic with
Isotropic Hardening

ET
FY

Force

E
Elasto-Plastic Unloading

Displacement
Figure 19.2. Loading and unloading force-displacement curves for elasto-plastic
discrete element.

This element combines the features of the above-described nonlinear elastic and
elasto-plastic discrete elements by allowing the user to specify independent initial yield
forces in tension (FYT) and in compression (FYC). If the discrete element force remains
between these initial yield values, the element unloads along the loading path (Figure
19.3(b)). This corresponds to the nonlinear elastic discrete element.

However, if the discrete element force exceeds either of these initial yield values, the
specified unloading curve is used for subsequent unloading. Additionally, the initial
loading and unloading curves are allowed to move in the force-displacement space by
specifying a mixed hardening parameter 𝛽 where 𝛽 = 0 corresponds to kinematic
hardening (Figure 19.3(c)) and 𝛽 = 0 𝛽 = 1 corresponds to isotropic hardening (Figure
19.3(d)).

19.8 Linear Visco-Elastic


The linear viscoelastic discrete element [Schwer, Cheva, and Hallquist 1991] allows
the user to model discrete components that are characterized by force relaxation or
displacement creep behavior. The element’s variable stiffness is defined by three
parameters and has the form
𝐾(𝑡) = 𝐾∞ + (𝐾0 − 𝐾∞ )𝑒−𝛽𝑡 , (19.25)

16.7
Discrete Elements and Masses LS-DYNA Theory Manual

loading
β>0 β>0
curve
options
β=0 Fyt β=0
Fyt - Fyc
δ δ
Fyc

unloading curve

kinematic hardening β<1 isotropic hardening β=1

force force
F2
Fyt-Fyc F1

δ δ
F2

Figure 19.3. Loading and unloading force displacement curves for general
nonlinear discrete element.

where 𝐾∞ is the long duration stiffness, 𝐾0 is the short time stiffness, and 𝛽 is a decay
parameter that controls the rate at which the stiffness transitions between the short and
long duration stiffness (Figure 16.4).
This model corresponds to a three-parameter Maxwell model (see insert in
Figure 19.4) which consists of a spring and damper in series connected to another spring in
parallel. Although this discrete element behavior could be built up using the above-
described linear elastic and linear viscous discrete elements, such a model would also
require the user to specify the nodal mass at the connection of the series spring and
damper. This mass introduces a fourth parameter which would further complicate fitting
the model to experimental data.

16.8
LS-DYNA Theory Manual Discrete Elements and Masses

Log K0
K∞ Visco-Elastic

Log K(t)
K0-K∞
1/β
Log K∞

Log t
Figure 19.4. Typical stiffness relaxation curve used for the viscoelastic discrete
element.

19.9 Muscle Model


This is Material Type 15 for discrete springs and dampers. This material is a Hill-
type muscle model with activation. It is for use with discrete elements. The LS-DYNA
implementation is due to Dr. J.A. Weiss.
L0 Initial muscle length, Lo.
VMAX Maximum CE shortening velocity, Vmax.
SV Scale factor, Sv, for Vmax vs. active state.
LT.0: absolute value gives load curve ID
GE.0: constant value of 1.0 is used
A Activation level vs. time function.
LT.0: absolute value gives load curve ID
GE.0: constant value of A is used
FMAX Peak isometric force, Fmax.
TL Active tension vs. length function.
LT.0: absolute value gives load curve ID
GE.0: constant value of 1.0 is used

16.9
Discrete Elements and Masses LS-DYNA Theory Manual

a(t) LM
vM
FCE CE
M SEE
F FM
FPE
LM PE

Figure 19.5. Discrete model for muscle contraction dynamics, based on a Hill-
type representation. The total force is the sum of passive force FPE and active
force FCE. The passive element (PE) represents energy storage from muscle
elasticity, while the contractile element (CE) represents force generation by the
muscle. The series elastic element (SEE), shown in dashed lines, is often neglected
when a series tendon compliance is included. Here, a(t) is the activation level, LM
is the length of the muscle, and vM is the shortening velocity of the muscle.

TV Active tension vs. velocity function.


LT.0: absolute value gives load curve ID
GE.0: constant value of 1.0 is used
FPE Force vs. length function, Fpe, for parallel elastic element.
LT.0: absolute value gives load curve ID
EQ.0: exponential function is used (see below)
GT.0: constant value of 0.0 is used
LMAX Relative length when Fpe reaches Fmax. Required if Fpe = 0 above.
KSH Constant, Ksh, governing the exponential rise of Fpe. Required if Fpe = 0
above.

The material behavior of the muscle model is adapted from the original model
proposed by Hill (1938). Reviews of this model and extensions can be found in Winters
(1990) and Zajac (1989). The most basic Hill-type muscle model consists of a contractile
element (CE) and a parallel elastic element (PE) (Figure 19.5). An additional series elastic
element (SEE) can be added to represent tendon compliance. The main assumptions of the
Hill model are that the contractile element is entirely stress free and freely distensible in the
resting state, and is described exactly by Hill’s equation (or some variation). When the
muscle is activated, the series and parallel elements are elastic, and the whole muscle is a
simple combination of identical sarcomeres in series and parallel. The main criticism of
Hill’s model is that the division of forces between the parallel elements and the division of
extensions between the series elements is arbitrary, and cannot be made without
introducing auxiliary hypotheses. However, these criticisms apply to any discrete element
model. Despite these limitations, the Hill model has become extremely useful for modeling
musculoskeletal dynamics, as illustrated by its widespread use today.

When the contractile element (CE) of the Hill model is inactive, the entire resistance
to elongation is provided by the PE element and the tendon load-elongation behavior. As
activation is increased, force then passes through the CE side of the parallel Hill model,
providing the contractile dynamics. The original Hill model accommodated only full
16.10
LS-DYNA Theory Manual Discrete Elements and Masses

activation - this limitation is circumvented in the present implementation by using the


modification suggested by Winters (1990). The main features of his approach were to
realize that the CE force-velocity input force equals the CE tension-length output force.
This yields a three-dimensional curve to describe the force-velocity-length relationship of
the CE. If the force-velocity y-intercept scales with activation, then given the activation,
length and velocity, the CE force can be determined.

Without the SEE, the total force in the muscle FM is the sum of the force in the CE
and the PE because they are in parallel:
𝐹M = 𝐹PE + 𝐹CE . (19.26)
The relationships defining the force generated by the CE and PE as a function of LM, VM
and 𝑎(𝑡) are often scaled by 𝐹max , the peak isometric force (p. 80, Winters 1990), L0, the
initial length of the muscle (p. 81, Winters 1990), and 𝑉max , the maximum unloaded CE
shortening velocity (p. 80, Winters 1990). From these, dimensionless length and velocity
can be defined:
𝐿M
𝐿= ,
𝐿o
(19.27)
𝑉M
𝑉= .
𝑉max ∗ 𝑆V (𝑎(t))
Here, 𝑆V scales the maximum CE shortening velocity 𝑉max and changes with activation
level 𝑎(𝑡). This has been suggested by several researchers, i.e. Winters and Stark [1985].
The activation level specifies the level of muscle stimulation as a function of time. Both
have values between 0 and 1. The functions 𝑆V (𝑎(𝑡)) and 𝑎(𝑡) are specified via load curves
in LS-DYNA, or default values of 𝑆V = 1 and 𝑎(𝑡) = 0 are used. Note that L is always
positive and that 𝑉 is positive for lengthening and negative for shortening.

The relationship between FCE, V and L was proposed by Bahler et al. [1967]. A
three-dimensional relationship between these quantities is now considered standard for
computer implementations of Hill-type muscle models [i.e., eqn 5.16, p. 81, Winters 1990].
It can be written in dimensionless form as:
𝐹CE = 𝑎(𝑡) ∗ 𝐹max ∗ 𝑓TL (𝐿) ∗ 𝑓TV (𝑉), (19.28)
The force in the parallel elastic element FPE is determined directly from the current length
of the muscle using an exponential relationship [eqn 5.5, p. 73, Winters 1990]:
𝐹PE
𝑓PE = =0 𝐿≤1
𝐹MAX
(19.29)
𝐹PE 1 𝐾
𝑓PE = = [exp ⎛
⎜ sh (L − 1)⎞
⎟ − 1] 𝐿>1
𝐹MAX exp(𝐾sh ) − 1 ⎝𝐿max ⎠

For computation of the total force developed in the muscle FM, the functions for the
tension-length 𝑓TL and force-velocity 𝑓TV relationships used in the Hill element must be
16.11
Discrete Elements and Masses LS-DYNA Theory Manual

Figure 19.6. Typical normalized tension-length (TL) and tension-velocity (TV)


curves for skeletal muscle.

defined. These relationships have been available for over 50 years, but have been refined to
allow for behavior such as active lengthening. The active tension-length curve 𝑓TL describes
the fact that isometric muscle force development is a function of length, with the maximum
force occurring at an optimal length. According to Winters, this optimal length is typically
around 𝐿 = 1.05, and the force drops off for shorter or longer lengths, approaching zero
force for 𝐿 = 0.4 and 𝐿 = 1.5. Thus the curve has a bell-shape. Because of the variability in
this curve between muscles, the user must specify the function 𝑓TL via a load curve,
specifying pairs of points representing the normalized force (with values between 0 and 1)
and normalized length 𝐿 (Figure 19.6).

The active tension-velocity relationship 𝑓TV used in the muscle model is mainly due
to the original work of Hill. Note that the dimensionless velocity V is used. When V = 0,
the normalized tension is typically chosen to have a value of 1.0. When V is greater than or
equal to 0, muscle lengthening occurs. As V increases, the function is typically designed so
that the force increases from a value of 1.0 and asymptotes towards a value near 1.4. When
V is less than zero, muscle shortening occurs and the classic Hill equation hyperbola is
used to drop the normalized tension to 0 (Figure 16.6). The user must specify the function
𝑓TV via a load curve, specifying pairs of points representing the normalized tension (with
values between 0 and 1) and normalized velocity V.

19.10 Seat Belt Material


The seat belt capability reported here was developed by Walker and co-workers
[Walker and Dallard 1991, Strut, Walker, et al., 1991] and this section excerpted from their
documentation. Each belt material defines stretch characteristics and mass properties for a

16.12
LS-DYNA Theory Manual Discrete Elements and Masses

set of belt elements. The user enters a load curve for loading, the points of which are
(Strain, Force). Strain is defined as engineering strain, i.e.
current length
Strain = − 1. (19.30)
initial length

Another similar curve is entered to describe the unloading behavior. Both


loadcurves should start at the origin (0,0) and contain positive force and strain values only.
The belt material is tension only with zero forces being generated whenever the strain
becomes negative. The first non-zero point on the loading curve defines the initial yield
point of the material. On unloading, the unloading curve is shifted along the strain axis
until it crosses the loading curve at the ‘yield’ point from which unloading commences. If
the initial yield has not yet been exceeded or if the origin of the (shifted) unloading curve is
at negative strain, the original loading curves will be used for both loading and unloading.
If the strain is less than the strain at the origin of the unloading curve, the belt is slack and
no force is generated. Otherwise, forces will then be determined by the unloading curve
for unloading and reloading until the strain again exceeds yield after which the loading
curves will again be used.

A small amount of damping is automatically included. This reduces high frequency


oscillation, but, with realistic force-strain input characteristics and loading rates, does not
significantly alter the overall forces-strain performance. The damping forced opposes the
relative motion of the nodes and is limited by stability:
. 1 × mass × relative velocity
𝐷= . (19.31)
timestep size

In addition, the magnitude of the damping forces is limited to one tenth of the force
calculated from the forces-strain relationship and is zero when the belt is slack. Damping
forces are not applied to elements attached to sliprings and retractors.

The user inputs a mass per unit length that is used to calculate nodal masses on
initialization.

A ‘minimum length’ is also input. This controls the shortest length allowed in any
element and determines when an element passes through sliprings or are absorbed into the
retractors. One tenth of a typical initial element length is usually a good choice.

19.11 Seat Belt Elements


Belt elements are single degree of freedom elements connecting two nodes and are
treated in a manner similar to the spring elements. When the strain in an element is

16.13
Discrete Elements and Masses LS-DYNA Theory Manual

positive (i.e., the current length is greater then the unstretched length), a tension force is
calculated from the material characteristics and is applied along the current axis of the
element to oppose further stretching. The unstretched length of the belt is taken as the
initial distance between the two nodes defining the position of the element plus the initial
slack length. At the beginning of the calculation the seatbelt elements can be obtained
within a retractor.

19.12 Sliprings
Sliprings are defined in the LS-DYNA input by giving a slipring ID and element ID’s
for two elements who share a node which is coincident with the slipring node. The slipring
node may not be attached to any belt elements.

Sliprings allow continuous sliding of a belt through a sharp change of angle. Two
elements (1 and 2 in Figure 19.7) meet at the slipring. Node B in the belt material remains
attached to the slipring node, but belt material (in the form of unstretched length) is passed
from element 1 to element 2 to achieve slip. The amount of slip at each timestep is
calculated from the ratio of forces in elements 1 and 2. The ratio of forces is determined by
the relative angle between elements 1 and 2 and the coefficient of friction, 𝜇. The tension in
the belts is taken as T1 and T2, where T2 is on the high-tension side and T1 is the force on
the low-tension side. Thus if T2 is sufficiently close to T1 no slip occurs; otherwise, slip is
just sufficient to reduce the ratio T2⁄T1 to 𝑒𝜇𝜃 . No slip occurs if both elements are slack.
The out-of-balance force at node B is reacted on the slipring node; the motion of node B
follows that of slipring node.

If, due to slip through the slipring, the unstretched length of an element becomes
less than the minimum length (as entered on the belt material card), the belt is remeshed
locally: the short element passes through the slipring and reappears on the other side (see
Figure 19.7). The new unstretched length of e1 is 1.1 × minimum length. Force and strain
in e2 and e3 are unchanged; force and strain in e1 are now equal to those in e2. Subsequent
slip will pass material from e3 to e1. This process can continue with several elements
passing in turn through the slipring.

To define a slipring, the user identifies the two belt elements which meet at the
slipring, the friction coefficient, and the slipring node. The two elements must have a
common node coincident with the slipring node. No attempt should be made to restrain or
constrain the common node for its motion will automatically be constrained to follow the
slipring node. Typically, the slipring node is part of the vehicle body structure and,
therefore, belt elements should not be connected to this node directly, but any other feature
can be attached, including rigid bodies.

16.14
LS-DYNA Theory Manual Discrete Elements and Masses

Slip ring

B B
Element 2 Element 1

Element 1 Element 3
Element 2

Element 3

Before After
Figure 19.7. Elements passing through slipring.

19.13 Retractors
Retractors are defined by giving a node, the “retractor node” and an element ID of
an element outside the retractor but with one node that is coincident with the retractor
node. Also sensor ID’s must be defined for up to four sensors which can activate the
seatbelt.
Retractors allow belt material to be paid out into a belt element, and they operate in
one of two regimes: unlocked when the belt material is paid out or reeled in under constant
tension and locked when a user defined force-pullout relationship applies.

The retractor is initially unlocked, and the following sequence of events must occur
for it to become locked:
• Any one of up to four sensors must be triggered. (The sensors are described below).
• Then a user-defined time delay occurs.
• Then a user-defined length of belt must be payed out (optional).
• Then the retractor locks.
and once locked, it remains locked.

In the unlocked regime, the retractor attempts to apply a constant tension to the belt.
This feature allows an initial tightening of the belt, and takes up any slack whenever it
occurs. The tension value is taken from the first point on the force-pullout load curve. The

16.15
Discrete Elements and Masses LS-DYNA Theory Manual

maximum rate of pull out or pull in is given by 0.01 × fed length per time step. Because of
this, the constant tension value is not always achieved.

In the locked regime, a user-defined curve describes the relationship between the
force in the attached element and the amount of belt material paid out. If the tension in the
belt subsequently relaxes, a different user-defined curve applies for unloading. The
unloading curve is followed until the minimum tension is reached.

The curves are defined in terms of initial length of belt. For example, if a belt is
marked at 10mm intervals and then wound onto a retractor, and the force required to make
each mark emerge from the (locked) retractor is recorded, the curves used for input would
be as follows:

0 Minimum tension (should be > zero)


10 mm Force to emergence of first mark
20 mm Force to emergence of second mark
..
..
..

Pyrotechnic pretensions may be defined which cause the retractor to pull in the belt
at a predetermined rate. This overrides the retractor force-pullout relationship from the
moment when the pretensioner activates.

If desired, belt elements may be defined which are initially inside the retractor.
These will emerge as belt material is paid out, and may return into the retractor if sufficient
material is reeled in during unloading.

Elements e2, e3 and e4 are initially inside the retractor, which is paying out material
into element e1. When the retractor has fed Lcrit into e1, where:
Lcrit = fed length − 1.1 × minimum length (19.32)
Here, minimum length is defined on belt material input, and fed length is defined on
retractor input.

element e2 emerges with an unstretched length of 1.1 × minimum length; the unstretched
length of element e1 is reduced by the same amount. The force and strain in e1 are
unchanged; in e2, they are set equal to those in e1. The retractor now pays out material into
e2.

If no elements are inside the retractor, e2 can continue to extend as more material is
fed into it.

16.16
LS-DYNA Theory Manual Discrete Elements and Masses

Element 1
Before Element 1

Element 2
Element 3 Element 2
After
Element 4
Element 3

Element 4

All nodes within this area


are coincident

Figure 19.8. Elements in a retractor.

As the retractor pulls in the belt (for example, during initial tightening), if the
unstretched length of the mouth element becomes less than the minimum length, the
element is taken into the retractor.

To define a retractor, the user enters the retractor node, the ‘mouth’ element (into
which belt material will be fed, e1 in Figure 19.8, up to 4 sensors which can trigger
unlocking, a time delay, a payout delay (optional), load and unload curve numbers, and
the fed length. The retractor node is typically part of the vehicle stricture; belt elements
should not be connected to this node directly, but any other feature can be attached
including rigid bodies. The mouth element should have a node coincident with the
retractor but should not be inside the retractor. The fed length would typically be set either
to a typical element initial length, for the distance between painted marks on a real belt for
comparisons with high-speed film. The fed length should be at least three times the
minimum length.

If there are elements initially inside the retractor (e2, e3 and e4 in the Figure) they
should not be referred to on the retractor input, but the retractor should be identified on
the element input for these elements. Their nodes should all be coincident with the
retractor node and should not be restrained or constrained. Initial slack will automatically
be set to 1.1 × minimum length for these elements; this overrides any user-defined value.

Weblockers can be included within the retractor representation simply by entering a


‘locking up’ characteristic in the force pullout curve, see Figure 19.9. The final section can
be very steep (but must have a finite slope).
16.17
Discrete Elements and Masses LS-DYNA Theory Manual

19.14 Sensors
Sensors are used to trigger locking of retractors and activate pretensioners. Four
types of sensor are available which trigger according to the following criteria:

Type 1–When the magnitude of x-, y-, or z- acceleration of a given node has remained
above a given level continuously for a given time, the sensor triggers. This does not work
with nodes on rigid bodies.

Type 2–When the rate of belt payout from a given retractor has remained above a given
level continuously for a given time, the sensor triggers.

Type 3–The sensor triggers at a given time.

Type 4–The sensor triggers when the distance between two nodes exceeds a given
maximum or becomes less than a given minimum. This type of sensor is intended for use
with an explicit mas/spring representation of the sensor mechanism.

By default, the sensors are inactive during dynamic relaxation. This allows initial
tightening of the belt and positioning of the occupant on the seat without locking the
retractor or firing any pretensioners. However, a flag can be set in the sensor input to make
the sensors active during the dynamic relaxation phase.

with weblockers
without weblockers
Force

Pullout
Figure 19.9. Retractor force pull characteristics.

16.18
LS-DYNA Theory Manual Discrete Elements and Masses

19.15 Pretensioners
Pretensioners allow modeling of three types of active devices which tighten the belt
during the initial stages of a crash. The first type represents a pyrotechnic device which
spins the spool of a retractor, causing the belt to be reeled in. The user defines a pull-in
versus time curve which applies once the pretensioner activates. The remaining types
represents preloaded springs or torsion bars which move the buckle when released. The
pretensioner is associated with any type of spring element including rotational. Note that
the preloaded spring, locking spring and any restraints on the motion of the associated
nodes are defined in the normal way; the action of the pretensioner is merely to cancel the
force in one spring until (or after) it fires. With the second type, the force in the spring
element is cancelled out until the pretensioner is activated. In this case the spring in
question is normally a stiff, linear spring which acts as a locking mechanism, preventing
motion of the seat belt buckle relative to the vehicle. A preloaded spring is defined in
parallel with the locking spring. This type avoids the problem of the buckle being free to
‘drift’ before the pretensioner is activated.

To activate the pretensioner the following sequence of events must occur:


1. Any one of up to four sensors must be triggered.
2. Then a user-defined time delay occurs.
3. Then the pretensioner acts.

19.16 Accelerometers
The accelerometer is defined by three nodes in a rigid body which defines a triad to
measure the accelerations in a local system. The presence of the accelerometer means that
the accelerations and velocities of node 1 will be output to all output files in local instead of
global coordinates.

The local coordinate system is defined by the three nodes as follows:


• local 𝐱 from node 1 to node 2
• local 𝐳 perpendicular to the plane containing nodes, 1, 2, and 3 (𝐳 = 𝐱 × 𝐚), where 𝐚
is from node 1 to node 3).
• local 𝐲 = 𝐱 × 𝐳

The three nodes should all be part of the same rigid body. The local axis then rotates
with the body.

16.19
LS-DYNA Theory Manual Simplified Arbitrary Lagrangian-Eulerian

20
Simplified Arbitrary Lagrangian-
Eulerian

Arbitrary Lagrangian-Eulerian (ALE) formulations may be thought of as algorithms


that perform automatic rezoning. Users perform manual rezoning by

1. Stopping the calculation when the mesh is distorted,

2. Smoothing the mesh,

3. Remapping the solution from the distorted mesh to the smooth mesh.

An ALE formulation consists of a Lagrangian time step followed by a “remap” or


“advection” step. The advection step performs an incremental rezone, where “incremen-
tal” refers to the fact that the positions of the nodes are moved only a small fraction of the
characteristic lengths of the surrounding elements. Unlike a manual rezone, the topology
of the mesh is fixed in an ALE calculation. An ALE calculation can be interrupted like an
ordinary Lagrangian calculation and a manual rezone can be performed if an entirely new
mesh is necessary to continue the calculation.

The accuracy of an ALE calculation is often superior to the accuracy of a manually


rezoned calculation because the algorithm used to remap the solution from the distorted to
the undistorted mesh is second order accurate for the ALE formulation while the algorithm
for the manual rezone is only first order accurate.

In theory, an ALE formulation contains the Eulerian formulation as a subset.


Eulerian codes can have more than one material in each element, but most ALE implemen-
tations are simplified ALE formulations which permit only a single material in each
element. The primary advantage of a simplified formulation is its reduced cost per time
step. When elements with more than one material are permitted, the number and types of
materials present in an element can change dynamically. Additional data is necessary to

LS-DYNA DEV 10/27/16 (r:8004) 18-1 (Simplified Arbitrary Lagrangian-Eulerian)


Simplified Arbitrary Lagrangian-Eulerian LS-DYNA Theory Manual

specify the materials in each element and the data must be updated by the remap
algorithms.

The range of problems that can be solved with an ALE formulation is a direct
function of the sophistication of the algorithms for smoothing the mesh. Early ALE codes
were not very successful largely because of their primitive algorithms for smoothing the
mesh. In simplified ALE formulations, most of the difficulties with the mesh are associated
with the nodes on the material boundaries. If the material boundaries are purely
Lagrangian, i.e., the boundary nodes move with the material at all times, no smooth mesh
maybe possible and the calculation will terminate. The algorithms for maintaining a
smooth boundary mesh are therefore as important to the robustness of the calculations as
the algorithms for the mesh interior.

The cost of the advection step per element is usually much larger than the cost of the
Lagrangian step. Most of the time in the advection step is spent in calculating the material
transported between the adjacent elements, and only a small part of it is spent on
calculating how and where the mesh should be adjusted. Second order accurate monotonic
advection algorithms are used in LS-DYNA despite their high cost per element because
their superior coarse mesh accuracy which allows the calculation to be performed with far
fewer elements than would be possible with a cheaper first order accurate algorithm.

The second order transport accuracy is important since errors in the transport
calculations generally smooth out the solution and reduce the peak values in the history
variables. Monotonic advection algorithms are constructed to prevent the transport
calculations from creating new minimum or maximum values for the solution variables.
They were first developed for the solution of the Navier Stokes equations to eliminate the
spurious oscillations that appeared around the shock fronts. Although monotonic
algorithms are more diffusive than algorithms that are not monotonic, they must be used
for stability in general purpose codes. Many constitutive models have history variables
that have limited ranges, and if their values are allowed to fall outside of their allowable
ranges, the constitutive models are undefined. Examples include explosive models, which
require the burn fraction to be between zero and one, and many elastoplasticity models,
such as those with power law hardening, which require a non-negative plastic strain.

The overall flow of an ALE time step is:

1. Perform a Lagrangian time step.

2. Perform an advection step.

a) Decide which nodes to move.

b) Move the boundary nodes.

c) Move the interior nodes.

18-2 (Simplified Arbitrary Lagrangian-Eulerian) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Simplified Arbitrary Lagrangian-Eulerian

d) Calculate the transport of the element-centered variables.

e) Calculate the momentum transport and update the velocity.

Each element solution variable must be transported. The total number of solution
variables, including the velocity, is at least six and depends on the material models. For
elements that are modeled with an equation of state, only the density, the internal energy,
and the shock viscosity are transported. When the elements have strength, the six compo-
nents of the stress tensor and the plastic strain must also be advected, for a total of ten
solution variables. Kinematic hardening, if it is used, introduces another five solution
variables, for a total of fifteen.

The nodal velocities add an extra three solution variables that must be transported,
and they must be advected separately from the other solution variables because they are
centered at the nodes and not in the elements. In addition, the momentum must be
conserved, and it is a product of the node-centered velocity and the element-centered
density. This imposes a constraint on how the momentum transport is performed that is
unique to the velocity field. A detailed consideration of the difficulties associated with the
transport of momentum is deferred until later.

Perhaps the simplest strategy for minimizing the cost of the ALE calculations is to
perform them only every few time steps. The cost of an advection step is typically two to
five times the cost of the Lagrangian time step. By performing the advection step only
every ten steps, the cost of an ALE calculation can often be reduced by a factor of three
without adversely affecting the time step size. In general, it is not worthwhile to advect an
element unless at least twenty percent of its volume will be transported because the gain in
the time step size will not offset the cost of the advection calculations.

20.1 Mesh Smoothing Algorithms


The algorithms for moving the mesh relative to the material control the range of the
problems that can be solved by an ALE formulation. The equipotential method which is
used in LS-DYNA was developed by Winslow [1990] and is also used in the DYNA2D ALE
code [Winslow 1963]. It, and its extensions, have proven to be very successful in a wide
variety of problems. The following is extracted from reports prepared by Alan Winslow
for LSTC.

20.1.1 Equipotential Smoothing of Interior Nodes


“Equipotential” zoning [Winslow, 1963] is a method of making a structured mesh
for finite difference or finite element calculations by using the solutions of Laplace
equations (later extended to Poisson equations) as the mesh lines. The same method can be
used to smooth selected points in an unstructured three-dimensional mesh provided that it

LS-DYNA DEV 10/27/16 (r:8004) 18-3 (Simplified Arbitrary Lagrangian-Eulerian)


Simplified Arbitrary Lagrangian-Eulerian LS-DYNA Theory Manual

is at least locally structured. This chapter presents a derivation of the three-dimensional


equipotential zoning equations, taken from the references, and gives their finite difference
equivalents in a form ready to be used for smoothing interior points. We begin by
reviewing the well-known two-dimensional zoning equations, and then discuss their
extension to three dimensions.

In two dimensions we define curvilinear coordinates 𝜉 𝜂 which satisfy Laplace’s


equation:
∇2 𝜉 = 0, (20.1.1a)

∇2 𝜂 = 0. (1.1.1b)

We solve Equations (1.1.1b) for the coordinates 𝑥(𝜉 , 𝜂) and 𝑦(𝜉 , 𝜂) of the mesh lines:
that is, we invert them so that the geometric coordinates 𝑥, 𝑦 become the dependent
variables and the curvilinear coordinates 𝜉 , 𝜂 the independent variables. By the usual
methods of changing variables we obtain
𝛼𝑥𝜉𝜉 − 2𝛽𝑥𝜉𝜂 + 𝛾𝑥𝜂𝜂 = 0, (20.1.2a)

𝛼𝑦𝜉𝜉 − 2𝛽𝑦𝜉𝜂 + 𝛾𝑦𝜂𝜂 = 0, (20.1.2b)


where
𝛼 ≡ 𝑥𝜂2 + 𝑦𝜂2 , 𝛽 ≡ 𝑥𝜉 𝑥𝜂 + 𝑦𝜉 𝑦𝜂 , 𝛾 ≡ 𝑥𝜉2 + 𝑦𝜉2 . (20.1.3)

Equations (16.1.2) can be written in vector form:


𝛼𝐫𝜉𝜉 − 2𝛽𝐫𝜉𝜂 + 𝛾𝐫𝜂𝜂 = 𝟎, (20.1.4)
where
𝐫 ≡ 𝑥𝐢 + 𝑦𝐣. (20.1.5)

We differentiate Equations (20.1.4) and solve them numerically by an iterative


method, since they are nonlinear. In (𝜉 , 𝜂) space, we use a mesh whose curvilinear
coordinates are straight lines which take on integer values corresponding to the usual
numbering in a two-dimensional mesh. The numerical solution then gives us the location
of the “equipotential” mesh lines.

In three dimensions 𝑥, 𝑦, 𝑧, we add a third curvilinear coordinate 𝜁 and a third


Laplace equation
∇2 𝜁 = 0. (20.1.1c)

Inversion of the system of three equations (17.1.1) by change of variable is rather


complicated. It is easier, as well as more illuminating, to use the methods of tensor analysis

18-4 (Simplified Arbitrary Lagrangian-Eulerian) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Simplified Arbitrary Lagrangian-Eulerian

pioneered by Warsi [1982]. Let the curvilinear coordinates be represented by 𝜉 𝑖 (𝑖 = 1,2,3).


For a scalar function 𝐴(𝑥, 𝑦, 𝑧),Warsi shows that the transformation of its Laplacian from
rectangular Cartesian to curvilinear coordinates is given by
3 3
∇ 𝐴 = ∑ 𝑔 𝐴𝜉 𝑖 𝜉 𝑗 + ∑(∇2 𝜉 𝑘 )𝐴𝜉 𝑘 ,
2 𝑖𝑗
(20.1.6)
𝑖,𝑗=1 𝑘=1

where a variable subscript indicates differentiation with respect to that variable. Since the
curvilinear coordinates are each assumed to satisfy Laplace’s equation, the second
summation in Equation (20.1.6) vanishes and we have
3
∇2 𝐴 = ∑ 𝑔𝑖𝑗 𝐴𝜉 𝑖 𝜉 𝑗 . (20.1.7)
𝑖,𝑗=1

If now we let 𝐴 = 𝑥, 𝑦, and 𝑧 successively, the left-hand side of (20.1.7) vanishes in


each case and we get three equations which we can write in vector form
3
∑ 𝑔𝑖𝑗 𝐫𝜉 𝑖𝜉 𝑗 = 0. (20.1.8)
𝑖,𝑗=1

Equation (20.1.8) is the three-dimensional generalization of Equations (20.1.4), and it


only remains to determine the components of the contravariant metric tensor 𝑔𝑖𝑗 in three
dimensions. These are defined to be
𝑔𝑖𝑗 ≡ 𝐚𝑖 ⋅ 𝐚𝑗 , (20.1.9)
where the contravariant base vectors of the transformation from (𝑥, 𝑦, 𝑧) to (𝜉 1 , 𝜉 2 , 𝜉 3 ) are
given by
𝐚𝑗 × 𝐚𝑘
𝐚𝑖 ≡ ∇𝜉 𝑖 = , (20.1.10)
√𝑔
(𝑖, 𝑗, 𝑘 cyclic). Here the covariant base vectors, the coordinate derivatives, are given by
𝐚𝑖 ≡ 𝐫𝜉 𝑖 , (20.1.11)
where
𝐫 ≡ 𝑥ı̂ + 𝑦ĵ + 𝑧𝐤̂. (20.1.12)

Also,
𝑔 ≡ det(𝑔𝑖𝑗 ) = [𝐚1 ⋅ (𝐚2 × 𝐚3 )]2 = 𝐽 2 , (20.1.13)
where 𝑔𝑖𝑗 is the covariant metric tensor given by
𝑔𝑖𝑗 ≡ 𝐚𝑖 ⋅ 𝐚𝑗 , (20.1.14)
and 𝐽 is the Jacobian of the transformation.

LS-DYNA DEV 10/27/16 (r:8004) 18-5 (Simplified Arbitrary Lagrangian-Eulerian)


Simplified Arbitrary Lagrangian-Eulerian LS-DYNA Theory Manual

Substituting (20.1.10) into (20.1.9), and using the vector identity


(𝐚 × 𝐛) ⋅ (𝐜 × 𝐝) ≡ (𝐚 ⋅ 𝐜) ⋅ (𝐛 ⋅ 𝐝) − (𝐚 ⋅ 𝐝)(𝐛 ⋅ 𝐜), (20.1.15)
we get
2 2
𝑔𝑔𝑖𝑖 = 𝐚𝑗2 𝐚𝑘2 − (𝐚𝑗 ⋅ 𝐚𝑘 ) = 𝑔𝑗𝑗 𝑔𝑘𝑘 − (𝑔𝑗𝑘 ) , (20.1.16)

𝑔𝑔𝑖𝑗 = (𝐚𝑖 ⋅ 𝐚𝑘 )(𝐚𝑗 ⋅ 𝐚𝑘 ) − (𝐚𝑖 ⋅ 𝐚𝑗 )𝐚𝑘2 = 𝑔𝑖𝑘 𝑔𝑗𝑘 − 𝑔𝑖𝑗 𝑔𝑘𝑘 . (20.1.17)

Before substituting (20.1.11) into (17.1.13a, b), we return to our original notation:
𝜉 + 𝜉 1, 𝜂 + 𝜉 2, 𝜁 + 𝜉 3. (20.1.18)
Then, using (20.1.11), we get
2
𝑔𝑔11 = 𝐫𝜂2 𝐫𝜁2 − (𝐫𝜂 ⋅ 𝐫𝜁 ) , (20.1.19)

2
𝑔𝑔22 = 𝐫𝜁2 𝐫𝜉2 − (𝐫𝜁 ⋅ 𝐫𝜉 ) , (20.1.20)

2
𝑔𝑔33 = 𝐫𝜉2 𝐫𝜂2 − (𝐫𝜉 ⋅ 𝐫𝜂 ) , (20.1.21)
for the three diagonal components, and
𝑔𝑔12 = (𝐫𝜉 ⋅ 𝐫𝜁 )(𝐫𝜂 ⋅ 𝐫𝜁 ) − (𝐫𝜉 ⋅ 𝐫𝜂 )𝐫𝜁2 , (20.1.22)

𝑔𝑔23 = (𝐫𝜂 ⋅ 𝐫𝜉 )(𝐫𝜁 ⋅ 𝐫𝜉 ) − (𝐫𝜂 ⋅ 𝐫𝜁 )𝐫𝜉2 , (20.1.23)

𝑔𝑔31 = (𝐫𝜁 ⋅ 𝐫𝜂 )(𝐫𝜉 ⋅ 𝐫𝜂 ) − (𝐫𝜁 ⋅ 𝐫𝜉 )𝐫𝜂2 , (20.1.24)


for the three off-diagonal components of this symmetric tensor.

When we express Equations (17.1.15) in terms of the Cartesian coordinates, some


cancellation takes place and we can write them in the form
𝑔𝑔11 = (𝑥𝜂 𝑦𝜁 − 𝑥𝜁 𝑦𝜂 )2 + (𝑥𝜂 𝑧𝜁 − 𝑥𝜁 𝑧𝜂 )2 + (𝑦𝜂 𝑧𝜁 − 𝑦𝜁 𝑧𝜂 )2 , (20.1.25)

𝑔𝑔22 = (𝑥𝜁 𝑦𝜉 − 𝑥𝜉 𝑦𝜁 )2 + (𝑥𝜁 𝑧𝜉 − 𝑥𝜉 𝑧𝜁 )2 + (𝑦𝜁 𝑧𝜉 − 𝑦𝜉 𝑧𝜁 )2 , (20.1.26)

𝑔𝑔33 = (𝑥𝜉 𝑦𝜂 − 𝑥𝜂 𝑦𝜉 )2 + (𝑥𝜉 𝑧𝜂 − 𝑥𝜂 𝑧𝜉 )2 + (𝑦𝜉 𝑧𝜂 − 𝑦𝜂 𝑧𝜉 )2 , (20.1.27)


guaranteeing positivity as required by Equations (20.1.9). Writing out Equations (17.1.16)
we get
𝑔𝑔12 = (𝑥𝜉 𝑥𝜁 + 𝑦𝜉 𝑦𝜁 + 𝑧𝜉 𝑧𝜁 )(𝑥𝜂 𝑥𝜁 + 𝑦𝜂 𝑦𝜁 + 𝑧𝜂 𝑧𝜁 )
(20.1.28)
−(𝑥𝜉 𝑥𝜂 + 𝑦𝜉 𝑦𝜂 + 𝑧𝜉 𝑧𝜂 )(𝑥𝜁2 + 𝑦𝜁2 + 𝑧2𝜁 ),

18-6 (Simplified Arbitrary Lagrangian-Eulerian) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Simplified Arbitrary Lagrangian-Eulerian

𝑔𝑔23 = (𝑥𝜂 𝑥𝜉 + 𝑦𝜂 𝑦𝜉 + 𝑧𝜂 𝑧𝜉 )(𝑥𝜁 𝑥𝜉 + 𝑦𝜁 𝑦𝜉 + 𝑧𝜁 𝑧𝜉 )


(20.1.29)
−(𝑥𝜂 𝑥𝜁 + 𝑦𝜂 𝑦𝜁 + 𝑧𝜂 𝑧𝜁 )(𝑥𝜉2 + 𝑦𝜉2 + 𝑧2𝜉 ),

𝑔𝑔31 = (𝑥𝜁 𝑥𝜂 + 𝑦𝜁 𝑦𝜂 + 𝑧𝜁 𝑧𝜂 )(𝑥𝜉 𝑥𝜂 + 𝑦𝜉 𝑦𝜂 + 𝑧𝜉 𝑧𝜂 )


(20.1.30)
−(𝑥𝜁 𝑥𝜉 + 𝑦𝜁 𝑦𝜉 + 𝑧𝜁 𝑧𝜉 )(𝑥𝜂2 + 𝑦𝜂2 + 𝑧2𝜂 ).

Hence, we finally write Equations (20.1.8) in the form


𝑔(𝑔11 𝐫𝜉𝜉 + 𝑔22 𝐫𝜂𝜂 + 𝑔33 𝐫𝜁𝜁 + 2𝑔12 𝐫𝜉𝜂 + 2𝑔23 𝐫𝜂𝜁 + 2𝑔31 𝐫𝜁𝜉 ) = 0, (20.1.31)
where the 𝑔𝑔𝑖𝑗 are given by Equations (17.1.17) and (17.1.18). Because Equations (20.1.8) are
homogeneous, we can use 𝑔𝑔𝑖𝑗 in place of 𝑔𝑖𝑗 as long as 𝑔 is positive, as it must be for a
nonsingular transformation. We can test for positivity at each mesh point by using
Equation (1.1.32):
𝑥 𝑦𝜉 𝑧𝜉
∣ 𝜉 ∣
√𝑔 = 𝐽 = ∣∣𝑥𝜂 𝑦𝜂 𝑧𝜂 ∣∣, (20.1.32)
𝑥
∣ 𝜁 𝑦 𝜁 𝑧 𝜁∣

and requiring that 𝐽 > 0.

To check that these equations reproduce the two-dimensional equations when there
is no variation in one-dimension, we take 𝜁 as the invariant direction, thus reducing
(17.1.19) to
𝑔𝑔11 𝐫𝜉𝜉 + 2𝑔𝑔12 𝐫𝜉𝜂 + 𝑔𝑔22 𝐫𝜂𝜂 = 0. (20.1.33)
If we let 𝜁 = 𝑧, then the covariant base vectors become
𝐚1 = 𝑥𝜉 𝐢 + 𝑦𝜉 𝐣, (20.1.34)

𝐚2 = 𝑥𝜂 𝐢 + 𝑦𝜂 𝐣, (20.1.35)

𝐚3 = 𝐤. (20.1.36)
From (17.1.22), using (17.1.13), we get
𝑔𝑔11 = 𝑥𝜂2 + 𝑦𝜂2 , (20.1.37)

𝑔𝑔22 = 𝑥𝜉2 + 𝑦𝜉2 , (20.1.38)

𝑔𝑔12 = −(𝑥𝜉 𝑥𝜂 + 𝑦𝜉 𝑦𝜂 ). (20.1.39)


Substituting (17.1.23) into (20.1.33) yields the two-dimensional equipotential zoning
Equations (17.1.2).

Before differencing Equations (17.1.19) we simplify the notation and write them in
the form

LS-DYNA DEV 10/27/16 (r:8004) 18-7 (Simplified Arbitrary Lagrangian-Eulerian)


Simplified Arbitrary Lagrangian-Eulerian LS-DYNA Theory Manual

ζ η
ξ
Figure 20.1. Example Caption

𝛼1 𝐫𝜉𝜉 + 𝛼2 𝐫𝜂𝜂 + 𝛼3 𝐫𝜁𝜁 + 2𝛽1 𝐫𝜉𝜂 + 2𝛽2 𝐫𝜂𝜁 + 2𝛽3 𝐫𝜁𝜉 = 0, (20.1.40)
where
𝛼1 = (𝑥𝜂 𝑦𝜁 − 𝑥𝜁 𝑦𝜂 )2 + (𝑥𝜂 𝑧𝜁 − 𝑥𝜁 𝑧𝜂 )2 + (𝑦𝜂 𝑧𝜁 − 𝑦𝜁 𝑧𝜂 )2 , (20.1.41)

𝛼2 = (𝑥𝜁 𝑦𝜉 − 𝑥𝜉 𝑦𝜁 )2 + (𝑥𝜁 𝑧𝜉 − 𝑥𝜉 𝑧𝜁 )2 + (𝑦𝜁 𝑧𝜉 − 𝑦𝜉 𝑧𝜁 )2 , (20.1.42)

𝛼3 = (𝑥𝜉 𝑦𝜂 − 𝑥𝜂 𝑦𝜉 )2 + (𝑥𝜉 𝑧𝜂 − 𝑥𝜂 𝑧𝜉 )2 + (𝑦𝜉 𝑧𝜂 − 𝑦𝜂 𝑧𝜉 )2 . (20.1.43)

𝛽1 = (𝑥𝜉 𝑥𝜁 + 𝑦𝜉 𝑦𝜁 + 𝑧𝜉 𝑧𝜁 )(𝑥𝜂 𝑥𝜁 + 𝑦𝜂 𝑦𝜁 + 𝑧𝜂 𝑧𝜁 )
(20.1.44)
−(𝑥𝜉 𝑥𝜂 + 𝑦𝜉 𝑦𝜂 + 𝑧𝜉 𝑧𝜂 )(𝑥𝜁2 + 𝑦𝜁2 + 𝑧2𝜁 ),

𝛽2 = (𝑥𝜂 𝑥𝜉 + 𝑦𝜂 𝑦𝜉 + 𝑧𝜂 𝑧𝜉 )(𝑥𝜉 𝑥𝜁 + 𝑦𝜉 𝑦𝜁 + 𝑧𝜉 𝑧𝜁 )
(20.1.45)
−(𝑥𝜁 𝑥𝜂 + 𝑦𝜁 𝑦𝜂 + 𝑧𝜁 𝑧𝜂 )(𝑥𝜉2 + 𝑦𝜉2 + 𝑧2𝜉 ),

𝛽3 = (𝑥𝜂 𝑥𝜁 + 𝑦𝜂 𝑦𝜁 + 𝑧𝜂 𝑧𝜁 )(𝑥𝜉 𝑥𝜂 + 𝑦𝜉 𝑦𝜂 + 𝑧𝜉 𝑧𝜂 )
(20.1.46)
−(𝑥𝜉 𝑥𝜂 + 𝑦𝜉 𝑦𝜂 + 𝑧𝜉 𝑧𝜂 )(𝑥𝜂2 + 𝑦𝜂2 + 𝑧2𝜂 ),

We difference Equations (20.1.40) in a cube in the rectangular 𝜉 𝜂 𝜁 space with unit


spacing between the coordinate surfaces, using subscript 𝑖 to represent the 𝜉 direction, 𝑗 the
𝜂 direction, and 𝑘 the 𝜁 direction, as shown in Figure 20.1.

18-8 (Simplified Arbitrary Lagrangian-Eulerian) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Simplified Arbitrary Lagrangian-Eulerian

Using central differencing, we obtain the following finite difference approximations


for the coordinate derivatives:
𝐫𝜉 = (𝐫𝑖+1 − 𝐫𝑖−1 )⁄2, (20.1.47)

𝐫𝜂 = (𝐫𝑗+1 − 𝐫𝑗−1 )⁄2, (20.1.48)

𝐫𝜁 = (𝐫𝑘+1 − 𝐫𝑘−1 )⁄2, (20.1.49)

𝐫𝜉𝜉 = (𝐫𝑖+1 − 2𝐫 + 𝐫𝑖−1 ), (20.1.50)

𝐫𝜂𝜂 = (𝐫𝑗+1 − 2𝐫 + 𝐫𝑗−1 ), (20.1.51)

𝐫𝜁𝜁 = (𝐫𝑘+1 − 2𝐫 + 𝐫𝑘−1 ), (20.1.52)

1
𝐫𝜉𝜂 = [(𝐫𝑖+1,𝑗+1 + 𝐫𝑖−1,𝑗−1 ) − (𝐫𝑖+1,𝑗−1 + 𝐫𝑖−1,𝑗+1 )], (20.1.53)
4
1
𝐫𝜂𝜁 = [(𝐫𝑗+1,𝑘+1 + 𝐫𝑗−1,𝑘−1 ) − (𝐫𝑗+1,𝑘−1 + 𝐫𝑗−1,𝑘+1 )], (20.1.54)
4
1
𝐫𝜁𝜉 = [(𝐫𝐼+1,𝑘+1 + 𝐫𝐼−1,𝑘−1 ) − (𝐫𝐼+1,𝑘−1 + 𝐫𝐼−1,𝑘+1 )], (20.1.55)
4
where for brevity we have omitted subscripts 𝑖, 𝑗, or 𝑘 (e.g., 𝑘 + 1 stands for 𝑖, 𝑗, 𝑘 + 1).
Note that these difference expressions use only coordinate planes that pass through the
central point, and therefore do not include the eight corners of the cube.

Substituting Equations (17.1.26) into (17.1.24,17.1.25) and collecting terms, we get


18
∑ 𝜔𝑚 (𝐫𝑚 − 𝐫) = 0, (20.1.56)
𝑚=1

where the sum is over the 18 nearest (in the transform space) neighbors of the given point.
The coefficients 𝜔𝑚 are given in Table 17.1.

Equations (20.1.56) can be written


∑𝑚 𝜔𝑚 𝐫𝑚
𝐫𝑚 = , (20.1.57)
∑𝑚 ω𝑚

expressing the position of the central point as a weighted mean of its 18 nearest neighbors.
The denominator of (20.1.57) is equal to 2(𝛼1 + 𝛼2 + 𝛼3 ) which is guaranteed to be positive
by (17.1.25). This vector equation is equivalent to the three scalar equations
∑𝑚 𝜔𝑚 x𝑚
𝑥= , (20.1.58)
∑𝑚 ω𝑚

LS-DYNA DEV 10/27/16 (r:8004) 18-9 (Simplified Arbitrary Lagrangian-Eulerian)


Simplified Arbitrary Lagrangian-Eulerian LS-DYNA Theory Manual

∑ 𝑚 𝜔 𝑚 y𝑚
𝑦= , (20.1.59)
∑ 𝑚 𝜔𝑚

m Index 𝜔𝑚
1 𝑖+1 𝛼1
2 𝑖– 1 𝛼1
3 𝑗+1 𝛼2
4 𝑗– 1 𝛼2
5 𝑘+1 𝛼3
6 𝑘– 1 𝛼3
7 𝑖 + 1, 𝑗 + 1 𝛽1 /2
8 𝑖– 1, 𝑗– 1 𝛽1 /2
9 𝑖 + 1, 𝑗– 1 −𝛽1 /2
10 𝑖– 1, 𝑗 + 1 −𝛽1 /2
11 𝑗 + 1, 𝑘 + 1 𝛽2 /2
12 𝑗– 1, 𝑘– 1 𝛽2 /2
13 𝑗 + 1, 𝑘– 1 −𝛽2 /2
14 𝑗– 1, 𝑘 + 1 −𝛽2 /2
15 𝑖 + 1, 𝑘 + 1 𝛽3 /2
16 𝑖– 1, 𝑘– 1 𝛽3 /2
17 𝑖 + 1, 𝑘– 1 −𝛽3 /2
18 𝑖– 1, 𝑘 + 1 −𝛽3 /2

Table 17.1. 3D Zoning Weight Coefficients

∑ 𝑚 𝜔𝑚 𝑧𝑚
𝑧= . (20.1.60)
∑ 𝑚 𝜔𝑚
the same weights 𝜔𝑚 appearing in each equation.

These equations are nonlinear, since the coefficients are functions of the coordinates.
Therefore we solve them by an iterative scheme, such as SOR. When applied to Equation
(20.1.58), for example, this gives for the (n+1)st iteration
∑ 𝑚 𝜔 𝑚 𝑥𝑚
𝑥𝑛+1 = (1 − f)𝑥𝑛 + f ( ), (20.1.61)
∑ 𝑚 𝜔𝑚
where the over relaxation factor 𝑓 must satisfy 0 < 𝑓 < 2. In (20.1.61) the values of 𝑥𝑚 at the
neighboring points are the latest available values. The coefficients 𝜔𝑚 are recalculated
before each iteration using Table 17.1 and Equations (17.1.25).

To smooth one interior point in a three-dimensional mesh, let the point to be


smoothed be the interior point of Figure 17.1, assuming that its neighborhood has the
logical structure shown. Even though Equations (17.1.29) are nonlinear, the 𝜔𝑚 do not

18-10 (Simplified Arbitrary Lagrangian-Eulerian) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Simplified Arbitrary Lagrangian-Eulerian

involve the coordinates of the central point, since the ’s and’s do not. Hence we simply
solve Equations (17.1.29) for the new coordinates (𝑥, 𝑦, 𝑧), holding the 18 neighboring
points fixed, without needing to iterate.

If we wish to smooth a group of interior points, we solve iteratively for the


coordinates using equations of the form (20.1.61).

20.1.2 Simple Averaging


The coordinates of a node is the simple average of the coordinates of its surrounding
nodes.
tot
𝑛+1 1 𝑚 𝑛
𝑥SA = tot ∑ 𝑥𝑚 . (20.1.62)
𝑚 𝑚=1

20.1.3 Kikuchi’s Algorithm


Kikuchi proposed an algorithm that uses a volume-weighted average of the coordi-
nates of the centroids of the elements surrounding a node. Variables that are subscripted
with Greek letters refer to element variables, and subscripts with capital letters refer to the
local node numbering within an element.
1 𝑛
𝑥αn = ∑ 𝑥𝐴 , (20.1.63)
8 𝐴

αtot
𝑛+1
∑α=1 𝑉a 𝑥αn
𝑥𝐾 = . (20.1.64)
αtot
∑α=1 Va

20.1.4 Surface Smoothing


The surfaces are smoothed by extending the two-dimensional equipotential stencils
to three dimensions. Notice that the form of Equation (20.1.2a) and (20.1.2b) for the 𝑥 and𝑦
directions are identical. The third dimension, 𝑧, takes the same form. When Equation
(16.1.2) is applied to all three dimensions, it tends to flatten out the surface and alter the
total volume. To conserve the volume and retain the curvature of the surface, the point
given by the relaxation stencil is projected on to the tangent plane defined by the normal at
the node.

20.1.5 Combining Smoothing Algorithms


The user has the option of using a weighted average of all three algorithms to
generate a composite algorithm, where the subscripts E, SA, and K refer to the
equipotential, simple averaging, and Kikuchi’s smoothing algorithm respectively, and ww
is the weighting factor.

LS-DYNA DEV 10/27/16 (r:8004) 18-11 (Simplified Arbitrary Lagrangian-Eulerian)


Simplified Arbitrary Lagrangian-Eulerian LS-DYNA Theory Manual

𝑛+1
𝑥𝑛+1 = 𝑤E 𝑥E𝑛+1 + 𝑤SA 𝑥SA 𝑛+1
+ 𝑤K 𝑥K . (20.1.65)

20.2 Advection Algorithms


LS-DYNA follows the SALE3D strategy for calculating the transport of the element-
centered variables (i.e., density, internal energy, the stress tensor and the history variables).
The van Leer MUSCL scheme [van Leer 1977] is used instead of the donor cell algorithm to
calculate the values of the solution variables in the transport fluxes to achieve second order
accurate monotonic results. To calculate the momentum transport, two algorithms have
been implemented. The less expensive of the two is the one that is implemented in
SALE3D, but it has known dispersion problems and may violate monotonocity (i.e.,
introduce spurious oscillations) [Benson 1992]. As an alternative, a significantly more
expensive method [Benson 1992], which can be shown analytically to not have either
problem on a regular mesh, has also been implemented.
In this section the donor cell and van Leer MUSCL scheme are discussed. Both
methods are one-dimensional and their extensions to multidimensional problems are
discussed later.

20.2.1 Advection Methods in One Dimension


In this section the donor cell and van Leer MUSCL scheme are discussed. Both
methods are one-dimensional and their extensions to multidimensional problems are
discussed later.

The remap step maps the solution from a distorted Lagrangian mesh on to the new
mesh. The underlying assumptions of the remap step are 1) the topology of the mesh is
fixed (a complete rezone does not have this limitation), and 2) the mesh motion during a
step is less than the characteristic lengths of the surrounding elements. Within the fluids
community, the second condition is simply stated as saying the Courant number, 𝐶, is less
than one.
𝑢Δ𝑡 𝑓
𝐶= = ≤ 1, (20.2.66)
Δ𝑥 V

Since the mesh motion does not occur over any physical time scale, t is arbitrary,
and ut is the transport volume, 𝑓 , between adjacent elements. The transport volume
calculation is purely geometrical for ALE formulations and it is not associated with any of
the physics of the problem.

The algorithms for performing the remap step are taken from the computational
fluids dynamics community, and they are referred to as “advection” algorithms after the

18-12 (Simplified Arbitrary Lagrangian-Eulerian) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Simplified Arbitrary Lagrangian-Eulerian

first order, scalar conservation equation that is frequently used as a model hyperbolic
problem.
∂φ ∂φ
+ a(𝑥) = 0. (20.2.67)
∂t ∂𝑥

A good advection algorithm for the remap step is accurate, stable, conservative and
monotonic. Although many of the solution variables, such as the stress and plastic strain,
are not governed by conservation equations like momentum and energy, it is still highly
desirable that the volume integral of all the solution variables remain unchanged by the
remap step. Monotonicity requires that the range of the solution variables does not
increase during the remap. This is particularly important with mass and energy, where
negative values would lead to physically unrealistic solutions.

Much of the research on advection algorithms has focused on developing monotonic


algorithms with an accuracy that is at least second order. Not all recent algorithms are
monotonic. For example, within the finite element community, the streamline upwind
Petrov-Galerkin (SUPG) method developed by Hughes and coworkers [Brooks and Hughes
1982] is not monotonic. Johnson et al., [1984] have demonstrated that the oscillations in the
SUPG solution are localized, and its generalization to systems of conservation equations
works very well for the Euler equations. Mizukami and Hughes [1985] later developed a
monotonic SUPG formulation. The essentially non-oscillatory (ENO) [Harten 1989] finite
difference algorithms are also not strictly monotonic, and work well for the Euler
equations, but their application to hydrodynamics problems has resulted in negative
densities [McGlaun 1990]. Virtually all the higher order methods that are commonly used
were originally developed for solving the Euler equations, usually as higher order
extensions to Godunov’s method. Since the operator split approach is the dominant one in
Eulerian hydrocodes, these methods are implemented only to solve the scalar advection
equation.

The Donor Cell Algorithm. Aside from its first order accuracy, it is everything a good
𝜑
advection algorithm should be: stable, monotonic, and simple. The value of 𝑓𝑗 is
dependent on the sign of a at node 𝑗, which defines the upstream direction.
𝑛+1 𝑛 Δ𝑡 𝜑 𝜑
φ𝑗+ 1⁄ = φ𝑗+1⁄ + (𝑓𝑗 − 𝑓𝑗+1 ), (20.2.68)
2 2 Δ𝑥

𝜑 𝑎𝑗 𝑛 𝑛
|𝑎𝑗 | 𝑛 𝑛
𝑓𝑗 = (φ𝑗− 1⁄ + φ𝑗+1⁄ ) + (φ𝑗− 1⁄ − φ𝑗+1⁄ ). (20.2.69)
2 2 2 2 2 2

The donor cell algorithm is a first order Godunov method applied to the advection
equation. The initial values of 𝜙 to the left and the right of node 𝑗 are φ𝑛𝑗−1⁄ and φ𝑛𝑗+1⁄ , and
2 2
the velocity of the contact discontinuity at node 𝑗 is 𝑎𝑗 .

LS-DYNA DEV 10/27/16 (r:8004) 18-13 (Simplified Arbitrary Lagrangian-Eulerian)


Simplified Arbitrary Lagrangian-Eulerian LS-DYNA Theory Manual

The Van Leer MUSCL Algorithm. Van Leer [1977] introduced a family of higher order
Godunov methods by improving the estimates of the initial values of left and right states
for the Riemann problem at the nodes. The particular advection algorithm that is
presented in this section is referred to as the MUSCL (monotone upwind schemes for
conservation laws) algorithm for brevity, although MUSCL really refers to the family of
algorithms that can be applied to systems of equations.

The donor cell algorithm assumes that the distribution of 𝜙 is constant over an
element. Van Leer replaces the piecewise constant distribution with a higher order
interpolation function, φ𝑛𝑗+1⁄ (𝑥) that is subject to an element level conservation constraint.
2
The value of 𝜙 at the element centroid is regarded in this context as the average value of 𝜙
over the element instead of the spatial value at 𝑥𝑗+1⁄ .
2
𝑥𝑗+1
𝑛 𝑛
φ𝑗+ 1⁄ = ∫ φ𝑗+ 1⁄ (𝑥)d𝑥, (20.2.70)
2 𝑥𝑗 2

To determine the range of 𝜙, [φmin


𝑗+1⁄
, φmax
𝑗+1⁄
], for imposing the monotonicity
2 2
constraint, the maximum and minimum values of φ𝑛𝑗−1⁄ , φ𝑛𝑗+1⁄ , and φ𝑛𝑗+3⁄ are used.
2 2 2
Monotonicity can be imposed in either of two ways. The first is to require that the
𝑛
maximum and minimum values of φ𝑗+ 1⁄ (𝑥) fall within the range determined by the three
2
elements. The second is to restrict the average value of 𝜙 in the transport volumes
associated with element 𝑗 + 1/2. While the difference may appear subtle, the actual
difference between the two definitions is quite significant even at relatively low Courant
numbers. The second definition allows the magnitude of the 𝜙 transported to adjacent
elements to be larger than the first definition. As a consequence, the second definition is
better able to transport solutions with large discontinuities. The magnitude of 𝜙 an
algorithm is able to transport before its monotonicity algorithm restricts 𝜙 is a measure of
the algorithm’s “compressiveness.”

The first step up from a piecewise constant function is a piecewise linear function,
where 𝑥 is now the volume coordinate. The volume coordinate of a point is simply the
volume swept along the path between the element centroid and the point. Conservation is
guaranteed by expanding the linear function about the element centroid.
𝑛 𝑛 𝑛 𝑛
φ𝑗+ 1⁄ (𝑥) = 𝑆𝑗+1⁄ (𝑥 − 𝑥𝑗+1⁄ ) + φj+1⁄ . (20.2.71)
2 2 2 2

Letting 𝑠𝑛𝑗+1⁄ be a second order approximation of the slope, the monotonicity limited
2
𝑛
value of the slope, 𝑠𝑗+ 1⁄ , according to the first limiting approach, is determined by
2
assuming the maximum permissible values at the element boundaries.
𝑛 1 L R L 𝑛 R
𝑆𝑗+ 1⁄ = (sgn(s ) + sgn(s )) × min (∣s ∣, ∣s𝑗+1⁄ ∣ , ∣s ∣), (20.2.72)
2 2 2

18-14 (Simplified Arbitrary Lagrangian-Eulerian) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Simplified Arbitrary Lagrangian-Eulerian

φ𝑛𝑗+1⁄ − φ𝑛𝑗−1⁄
2 2
𝑠L = , (20.2.73)
1 Δ𝑥
2 𝑗+1⁄2

φ𝑛𝑗+3⁄ − φ𝑛𝑗+1⁄
2 2
𝑠R = . (20.2.74)
1 Δ𝑥
2 𝑗+1⁄2

The second limiter is similar to the first, but it assumes that the maximum
permissible values occur at the centroid of the transport volumes. Note that as stated in
Equation (17.2.6), this limiter still limits the slope at the element boundary even if the
element is the downstream element at that boundary. A more compressive limiter would
not limit the slope based on the values of 𝜙 at the downstream boundaries. For example, if
𝑎𝑗 is negative, only 𝑆𝑅 would limit the value of 𝑆𝑛 in Equation (17.2.6). If the element is the
downstream element at both boundaries, then the slope in the element has no effect on the
solution.
φ𝑛𝑗+1⁄ − φ𝑛𝑗−1⁄
2 2
𝑠L = . (20.2.75)
1 Δ𝑥 − 1 max(0, 𝑎𝑗 Δ𝑡)
2 𝑗+1⁄2 2

φ𝑛𝑗+3⁄ − φ𝑛𝑗+1⁄
2 2
𝑠R = . (20.2.76)
1𝑥 + 1 min(0, 𝑎𝑗+1 Δ𝑡)
2 𝑗+1⁄2 2

The flux at node 𝑗 is evaluated using the upstream approximation of 𝜙.


φ 𝑎𝑗 +
|𝑎𝑗 | +
𝑓𝑗 = (φ−
𝑗 + φ𝑗 ) + (φ−
𝑗 − φ𝑗 ),
(20.2.77)
2 2
𝑛 𝑛 𝑛
φ+ C
𝑗 = S𝑗+1⁄ (𝑥 − 𝑥𝑗+1⁄ ) + φ𝑗+1⁄ , (20.2.78)
2 2 2

𝑛 𝑛 𝑛
φ− C
𝑗 = S𝑗−1⁄ (𝑥 − 𝑥𝑗−1⁄ ) + φ𝑗−1⁄ , (20.2.79)
2 2 2

1
𝑥C = 𝑥𝑗𝑛 + 𝑎𝑗 Δ𝑡. (20.2.80)
2

The method for obtaining the higher order approximation of the slope is not unique.
Perhaps the simplest approach is to fit a parabola through the centroids of the three
𝑛
adjacent elements and evaluate its slope at 𝑥𝑗+ 1⁄ . When the value of 𝜙 at the element
2
centroids is assumed to be equal to the element average this algorithm defines a projection.

LS-DYNA DEV 10/27/16 (r:8004) 18-15 (Simplified Arbitrary Lagrangian-Eulerian)


Simplified Arbitrary Lagrangian-Eulerian LS-DYNA Theory Manual

(𝜑𝑛𝑗+3⁄ − φn𝑗+1⁄ ) Δ𝑥𝑗2 + (φ𝑛𝑗+1⁄ − φ𝑛𝑗−1⁄ ) Δ𝑥𝑗+1


2

𝑠𝑛𝑗+1⁄ = 2 2 2 2
, (20.2.81)
2 Δ𝑥𝑗 Δ𝑥𝑗+1 (Δ𝑥𝑗 + Δ𝑥𝑗+1 )
𝑛 𝑛
Δ𝑥𝑗 = 𝑥𝑗+ 1⁄ − 𝑥𝑗−1⁄ . (20.2.82)
2 2

20.2.2 Advection Methods in Three Dimensions


For programs that use a logically regular mesh, one-dimensional advection methods
are extended to two and three dimensions by using a sequence of one-dimensional sweeps
along the logically orthogonal mesh lines. This strategy is not possible for unstructured
meshes because they don’t have uniquely defined sweep directions through the mesh.
CAVEAT [Addessio, et al., 1986] uses one-dimensional sweeps in the spatial coordinate
system, but their approach is expensive relative to the other algorithms and it does not
always maintain spherical symmetry, which is an important consideration in underwater
explosion calculations.

The advection in LS-DYNA is performed isotropically. The fluxes through each face
of element A are calculated simultaneously, but the values of 𝜙 in the transport volumes
are calculated using the one-dimensional expressions developed in the previous sections.
6
1 ⎛ 𝑛 𝑛
f𝑗 ⎞
φ
φ𝑛+1
A = 𝑛+1
⎜ V φ
A A + ∑ ⎟. (20.2.83)
VA ⎝ 𝑗=1 ⎠

The disadvantage of isotropic advection is that there is no coupling between an


element and the elements that are joined to it only at its corners and edges (i.e., elements
that don’t share faces). The lack of coupling introduces a second order error that is
significant only when the transport is along the mesh diagonals.

The one-dimensional MUSCL scheme, which requires elements on either side of the
element whose transport is being calculated, cannot be used on the boundary elements in
the direction normal to the boundary. Therefore, in the boundary elements, the donor cell
algorithm is used to calculate the transport in the direction that is normal to the boundary,
while the MUSCL scheme is used in the two tangential directions.

It is implicitly assumed by the transport calculations that the solution variables are
defined per unit current volume. In LS-DYNA, some variables, such as the internal energy,
are stored in terms of the initial volume of the element. These variables must be rescaled
before transport, then the initial volume of the element is advected between the elements,
and then the variables are rescaled using the new “initial” volumes. Hyperelastic materials
are not currently advected in LS-DYNA because they require the deformation gradient,
which is calculated from the initial geometry of the mesh. If the deformation gradient is
integrated by using the midpoint rule, and it is advected with the other solution variables,
then hyperelastic materials can be advected without any difficulties.

18-16 (Simplified Arbitrary Lagrangian-Eulerian) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Simplified Arbitrary Lagrangian-Eulerian

Δ𝑡 Δ𝑡
F𝑛+1 = (I − )−1 (I + ) F𝑛 . (20.2.84)
𝑛+1⁄2 𝑛+1⁄2
2L 2L

Advection of the Nodal Velocities. Except for the Godunov schemes, the velocity is
centered at the nodes or the edges while the remaining variables are centered in the
elements. Momentum is advected instead of the velocity in most codes to guarantee that
momentum is conserved. The element-centered advection algorithms must be modified to
advect the node-centered momentum. Similar difficulties are encountered when node-
centered algorithms, such as the SUPG method [Brooks and Hughes 1982], are applied to
element-centered quantities [Liu, Chang, and Belytschko, to be published]. There are two
approaches: 1) construct a new mesh such that the nodes become the element centroids of
the new mesh and apply the element-centered advection algorithms, and 2) construct an
auxiliary set of element-centered variables from the momentum, advect them, and then
reconstruct the new velocities from the auxiliary variables. Both approaches can be made
to work well, but their efficiency is heavily dependent on the architecture of the codes. The
algorithms are presented in detail for one dimension first for clarity. Their extensions to
three dimensions, which are presented later, are straightforward even if the equations do
become lengthy. A detailed discussion of the algorithms in two dimensions is presented in
Reference [Benson 1992].

Notation. Finite difference notation is used in this section so that the relative locations of
the nodes and fluxes are clear. The algorithms are readily applied, however, to
unstructured meshes. To avoid limiting the discussion to a particular element-centered
advection algorithm, the transport volume through node 𝑖 is 𝑓 , the transported mass is 𝑓𝑖 ,̃
and the flux of 𝜙 is 𝜙𝑖 𝑓𝑖 . Most of the element-centered flux-limited advection algorithms
calculate the flux of 𝜙 directly, but the mean value of 𝜙 in the transport volumes is
calculated by dividing the 𝜙i 𝑓i , by the transport volume. A superscript “-” or “+” denotes
the value of a variable before or after the advection. Using this notation, the advection of 𝜙
in one dimension is represented by Equation (17.2.12), where the volume is V.
(φ− V−
𝑗+1⁄ 𝑗+1⁄
+ φ𝑖 𝑓𝑖 − φ𝑖+1 𝑓𝑖+1 )
+
φ𝑗+ 2 2 (20.2.85)
1⁄ = ,
2 V+
+ −
V𝑗+ 1⁄ = V𝑗+1⁄ + 𝑓𝑖 − 𝑓𝑖+1 . (20.2.86)
2 2

The Staggered Mesh Algorithm. YAQUI [Amsden and Hirt 1973] was the first code to
construct a new mesh that is staggered with respect to original mesh for momentum
advection. The new mesh is defined so that the original nodes become the centroids of the
new elements. The element-centered advection algorithms are applied to the new mesh to
advect the momentum. In theory, the momentum can be advected with the transport
volumes or the velocity can be advected with the mass.

LS-DYNA DEV 10/27/16 (r:8004) 18-17 (Simplified Arbitrary Lagrangian-Eulerian)


Simplified Arbitrary Lagrangian-Eulerian LS-DYNA Theory Manual

(M𝑗− v− ̃ ̃
𝑗 + v𝑗−1⁄ 𝑓𝑗−1⁄ − v𝑗+1⁄ 𝑓𝑗+1⁄ )
2 2 2 2
v+
𝑗 = , (20.2.87)
M𝑗+

(M𝑗− v−
𝑗 + {ρv}𝑗−1⁄ 𝑓𝑗−1⁄ − {ρv}𝑗+1⁄ 𝑓𝑗−1⁄ )
2 2 2 2
v+
𝑗 = , (20.2.88)
M𝑗+

M𝑗+ = M𝑗− + 𝑓𝑗−̃ 1⁄ − 𝑓𝑗+̃ 1⁄ . (20.2.89)


2 2

A consistency condition, first defined by DeBar [1974], imposes a constraint on the


formulation of the staggered mesh algorithm: if a body has a uniform velocity and a
spatially varying density before the advection, then the velocity should be uniform and
unchanged after the advection. The new mass of a node can be expressed in terms of the
quantities used to advect the element-centered mass.
1 + +
M𝑗+ = (M𝑗− 1⁄ + M𝑗+1⁄ ), (20.2.90)
2 2 2

1 − −
M𝑗+ = (M𝑗− 1⁄ + 𝜌𝑗−1 𝑓𝑗−1 − 𝜌𝑗 𝑓𝑗 + M𝑗+1⁄ + 𝜌𝑗 𝑓𝑗 − 𝜌𝑗+1 𝑓𝑗+1 ), (20.2.91)
2 2 2

1
M𝑗+ = M𝑗− + [(𝜌𝑗−1 𝑓𝑗−1 − 𝜌𝑗 𝑓𝑗 ) + (𝜌𝑗 𝑓𝑗 − 𝜌𝑗+1 𝑓𝑗+1 )]. (20.2.92)
2
The staggered mass fluxes and transport volumes are defined by equating Equation
(20.2.90) and Equation (17.2.15).
1
𝜌𝑗+1⁄ 𝑓𝑗+1⁄ = 𝑓𝑗+̃ 1⁄ = (𝜌𝑗 𝑓𝑗 + 𝜌𝑗+1 𝑓𝑗+1 ). (20.2.93)
2 2 2 2
The density 𝜌𝑗+1⁄ is generally a nonlinear function of the volume 𝑓𝑗+1⁄ , hence calculating
2 2
𝑓𝑗+1⁄ from Equation (20.2.93) requires the solution of a nonlinear equation for each
2
transport volume. In contrast, the mass flux is explicitly defined by Equation (20.2.93).
Most codes, including KRAKEN [Debar 1974], CSQ [Thompson 1975], CTH [McGlaun
1989], and DYNA2D [Hallquist 1980], use mass fluxes with the staggered mesh algorithm
because of their simplicity.

The dispersion characteristics of this algorithm are identical to the underlying


element-centered algorithm by construction. This is not true, however, for some of the
element-centered momentum advection algorithms. There are some difficulties in
implementing the staggered mesh method in multi-dimensions. First, the number of edges
defining a staggered element equals the number of elements surrounding the correspond-
ing node. On an unstructured mesh, the arbitrary connectivity results in an arbitrary
number of edges for each staggered element. Most of the higher order accurate advection
algorithms assume a logically regular mesh of quadrilateral elements, making it difficult to
use them with the staggered mesh. Vectorization also becomes difficult because of the
18-18 (Simplified Arbitrary Lagrangian-Eulerian) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Simplified Arbitrary Lagrangian-Eulerian

random number of edges that each staggered element might have. In the ALE calculations
of DYNA2D, only the nodes that have a locally logically regular mesh surrounding them
can be moved in order to avoid these difficulties [Benson 1992]. These difficulties do not
occur in finite difference codes which process logically regular blocks of zones. Another
criticism is the staggered mesh algorithm tends to smear out shocks because not all the
advected variables are element-centered [Margolin 1989]. This is the primary reason,
according to Margolin [1989], that the element-centered algorithm was adopted in SALE
[Amdsden, Ruppel, and Hirt 1980].

The SALE Algorithm. SALE advects an element-centered momentum and redistributes its
changes to the nodes [Amdsden, Ruppel, and Hirt 1980]. The mean element velocity,
𝐯
̅̅̅̅𝑗+1⁄ , specific momentum, 𝐩𝑗+1⁄ , element momentum, 𝐏𝑗+1⁄ , and nodal momentum are
2 2 2
defined by Equation (17.2.17).
1
̅̅̅̅𝑗+1⁄ = (𝐯𝑗 + 𝐯𝑗+1 ),
𝐯 (20.2.94)
2 2
𝐩𝑗+1⁄ = ρ𝑗+1⁄ 𝐯
̅̅̅̅𝑗+1⁄ , (20.2.95)
2 2 2

𝐏j+1⁄ = M𝑗+1⁄ 𝐯
̅̅̅̅𝑗+1⁄ . (20.2.96)
2 2 2

Denoting the change in the element momentum Δ𝐏𝑗+1⁄ , the change in the velocity at a
2
node is calculated by distributing half the momentum change from the two adjacent
elements.
Δ𝐏𝑗−1⁄ = p𝑗−1 𝑓𝑗−1 − p𝑗 𝑓𝑗 , (20.2.97)
2

1
P𝑗+ = P𝑗− + (Δ𝐏𝑗−1⁄ + Δ𝐏𝑗+1⁄ ), (20.2.98)
2 2 2

𝐏𝑗+
𝐯𝑗+ = . (20.2.99)
M𝑗+

̅̅̅̅𝑗+1⁄ with
This algorithm can also be implemented by advecting the mean velocity, 𝐯
2
̅̅̅̅𝑗 𝑓𝑗 .̃
the transported mass, and the transported momentum 𝐩𝑗 𝑓𝑗 is changed to 𝐯

The consistency condition is satisfied regardless of whether masses or volumes are


used. Note that the velocity is not updated from the updated values of the adjacent
element momenta. The reason for this is the original velocities are not recovered if 𝑓𝑖 = 0,
which indicates that there is an inversion error associated with the algorithm.

The HIS (Half Index Shift) Algorithm. Benson [1992] developed this algorithm based on
his analysis of other element-centered advection algorithms. It is designed to overcome the
dispersion errors of the SALE algorithm and to preserve the monotonicity of the velocity

LS-DYNA DEV 10/27/16 (r:8004) 18-19 (Simplified Arbitrary Lagrangian-Eulerian)


Simplified Arbitrary Lagrangian-Eulerian LS-DYNA Theory Manual

field. The SALE algorithm is a special case of a general class of algorithms. To sketch the
idea behind the HIS algorithm, the discussion is restricted to the scalar advection equation.
Two variables, Ψ1,𝑗+1⁄ and Ψ2,𝑗+1⁄ are defined in terms of a linear transformation of 𝜙𝑗 and
2 2
𝜙𝑗+1 . The linear transformation may be a function of the element 𝑗 + 1/2.

{Ψ1,𝑗+1⁄2 ⎫
⎧ −
}
= [a b ] {φ𝑗 }. (20.2.100)
⎨ − ⎬ −
{

Ψ2,𝑗+ 1⁄ ⎭} c d φ𝑗+1
2

This relation is readily inverted.

φ+ 1 ⎧Ψ+ ⎫
𝑗
{ + }= [d −b] { 1,𝑗+1⁄2 }. (20.2.101)
φ𝑗+1 ad − bc −c a ⎨ { Ψ+ ⎬
⎩ 2,𝑗+1⁄2 }

A function is monotonic over an interval if its derivative does not change sign. The
sum of two monotonic functions is monotonic, but their difference is not necessarily
− −
monotonic. As a consequence, Ψ1,𝑗+ 1⁄ and Ψ2,𝑗+1⁄ are monotonic over the same intervals
2 2
as φ−
𝑗 if all the coefficients in the linear transformation have the same sign. On the other
+ +
hand, φ+𝑗 is not necessarily monotonic even if Ψ1,𝑗+1⁄ and Ψ2,𝑗+1⁄ are monotonic because of
2 2
the appearance of the negative signs in the inverse matrix. Monotonicity can be maintained
by transforming in both directions provided that the transformation matrix is diagonal.
Symmetry in the overall algorithm is obtained by using a weighted average of the values of
𝜙𝑗 calculated in elements 𝑗 + 1/2 and 𝑗 − 1/2.

A monotonic element-centered momentum advection algorithm is obtained by


choosing the identity matrix for the transformation and by using mass weighting for the
inverse relationship.

{Ψ1,𝑗+1⁄2 ⎫
⎧ −
}
= [1 0] {v𝑗 } (20.2.102)
⎨ ⎬ −
{ Ψ
⎩ 2,𝑗+1⁄2 }
⎭ 0 1 v𝑗+1

To conserve momentum, Ψ is advected with the transport masses.

(M− 1 Ψ− −
+ Ψ𝑚,𝑗 𝑓𝑗 ̃ − Ψ𝑚,𝑗+1
− ̃ )
𝑓𝑗+1
𝑗+ 𝑚,𝑗+1
+ 2 2 (20.2.103)
Ψ𝑚,𝑗+1/2 = ,
M+ 1
𝑗+
2

1
v𝑗 = (M𝑗+1/2 Ψ1,𝑗+1/2 + M𝑗−1/2 Ψ2,𝑗−1/2 ). (20.2.104)
2M𝑗

Dispersion Errors. A von Neumann analysis [Trefethen 1982] characterizes the dispersion
errors of linear advection algorithms. Since the momentum advection algorithm modifies
the underlying element-centered advection algorithm, the momentum advection algorithm
does not necessarily have the same dispersion characteristics as the underlying algorithm.
18-20 (Simplified Arbitrary Lagrangian-Eulerian) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Simplified Arbitrary Lagrangian-Eulerian

The von Neumann analysis provides a tool to explore the changes in the dispersion
characteristics without considering a particular underlying advection algorithm.

The model problem is the linear advection equation with a constant value of c. A
class of solutions can be expressed as complex exponentials, where i is √−1 ,  is the
frequency, and χ is the wave number.
∂φ ∂φ
+c = 0, (20.2.105)
∂𝑡 ∂𝑥

φ(𝑥, 𝑡) = 𝑒𝑖(ω𝑡−χ𝑥) . (20.2.106)

For Equation (17.2.24), the dispersion equation is 𝜔 = 𝑐𝜒, but for discrete
approximations of the equation and for general hyperbolic equations, the relation is 𝜔 =
𝜔𝜒. The phase velocity, cp, and the group speed, cg, are defined by Equation (17.2.25).
ω
cp = , (20.2.107)
χ

∂ω
cg = . (20.2.108)
∂χ

The mesh spacing is assumed to have a constant value 𝐽, and the time step, ℎ, is also
constant. The + and - states in the previous discussions correspond to times n and n + 1 in
the dispersion analysis. An explicit linear advection method that has the form given by
Equation (1.2.109) results in a complex dispersion equation, Equation (17.2.27), where Π is
a complex polynomial.
𝜑𝑛+1
𝑗 = 𝜑𝑛𝑗 + F(c, ℎ, 𝐽, . . . , 𝜑𝑛𝑗−1 , 𝜑𝑛𝑗, 𝜑𝑛𝑗+1 , . . . ), (20.2.109)

𝑒𝑖ωℎ = 1 + P(𝑒𝑖χ𝐽 ), (20.2.110)

Π(𝑒𝑖χ𝐽 ) = ∑ 𝛽𝑗 𝑒𝑖χ𝑗𝐽 . (20.2.111)


𝑗

The dispersion equation has the general form given in Equation (1.2.112), where Πr
and Πi denote the real and imaginary parts of Π, respectively.
Π𝑖
ωℎ = tan−1 ( ). (20.2.112)
1 + Πr
Recognizing that the relations in the above equations are periodic in 𝜔ℎ and χ 𝐽 , the
normalized frequency and wave number are defined to simplify the notation.
𝜔
̅̅̅̅ = ωℎ, ̅̅̅̅ = χ 𝐽 .
χ (20.2.113)

LS-DYNA DEV 10/27/16 (r:8004) 18-21 (Simplified Arbitrary Lagrangian-Eulerian)


Simplified Arbitrary Lagrangian-Eulerian LS-DYNA Theory Manual

The von Neumann analysis of the SALE algorithm proceeds by first calculating the
increment in the cell momentum.
1
p𝑛𝑗+1/2 = (v𝑛𝑗 + v𝑗+1
𝑛
), (20.2.114)
2
1
p𝑛𝑗+1/2 = (1 + 𝑒−𝑖χ̅̅̅̅ )v𝑛𝑗 , (20.2.115)
2
𝑛+1 𝑛
Δp𝑛+1
𝑗+1/2 = P𝑗+1⁄ − P𝑗+1⁄ , (20.2.116)
2 2

1
Δp𝑛+1
𝑗+1/2 = (1 + e
−𝑖χ
̅̅ ̅̅
)Πv𝑛𝑗 . (20.2.117)
2

The velocity is updated from the changes in the cell momentum.


1
v𝑛+1
𝑗 = v𝑛𝑗 + (Δp𝑛+1 𝑛+1
𝑗+1/2 + Δp𝑗−1/2 ), (20.2.118)
2
1
v𝑛+1
𝑗 = (1 + 𝑒𝑖χ̅̅̅̅ )(1 + 𝑒−𝑖χ̅̅̅̅ )Πv𝑛𝑗 , (20.2.119)
4
1
v𝑛+1
𝑗 ̅̅̅̅))Πv𝑛𝑗 .
= (1 + cos(χ (20.2.120)
2

The dispersion relation for the SALE advection algorithm is given by Equation
(1.2.121).
1 (1 + cos(χ
̅̅̅̅))Π𝑖 ⎞
⎛ 2
̅̅̅̅ = tan−1 ⎜
𝜔 ⎜



⎟. (20.2.121)
1 + 1 (1 + cos(χ ̅̅̅̅))Π
⎝ 2 r⎠

By comparing Equation (20.2.112) and Equation (20.2.121), the effect of the SALE
momentum advection algorithm on the dispersion is to introduce a factor λ, equal to
1
2
(1 + cos(χ ̅̅̅̅))Π, into the spatial part of the advection stencil. For small values of χ ̅̅̅̅, λ is
close to one, and the dispersion characteristics are not changed, but when χ ̅̅̅̅ is π, the phase
and group velocity go to zero and the amplification factor is one independent of the
underlying advection algorithm. Not only is the wave not transported, it is not damped
out. The same effect is found in two dimensions, where λ, has the form 14 (1 + cos(χ ̅̅̅̅) +
cos(χ̅̅̅̅
̅̅̅̅ ) + cos(χ ̅̅̅̅
̅̅̅̅)cos(χ
̅̅̅̅ )).

In contrast, none of the other algorithms alter the dispersion characteristics of the
underlying algorithm. Benson has demonstrated for the element-centered algorithms that
the SALE inversion error and the dispersion problem are linked. Algorithms that fall into
the same general class as the SALE and HIS algorithms will, therefore, not have dispersion
problems [Benson 1992].
18-22 (Simplified Arbitrary Lagrangian-Eulerian) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Simplified Arbitrary Lagrangian-Eulerian

Three-Dimensional Momentum Advection Algorithms. The momentum advection


algorithms discussed in the previous sections are extended to three dimensions in a
straightforward manner. The staggered mesh algorithm requires the construction of a
staggered mesh and the appropriate transport masses. Based on the consistency
arguments, the appropriate transport masses are given by Equation (1.2.122).
𝑗+1 𝑘+1⁄2 𝑙+1⁄2
̃ 1
𝑓𝑗+1/2,𝑘,𝑙 = ∑ ∑ ∑ 𝑓𝐽,𝐾,𝐿 . (20.2.122)
8 𝐽=𝑗
𝐾=𝑘−1⁄2 𝐿=𝑙−1⁄2

The SALE advection algorithm calculates the average momentum of the element
from the four velocities at the nodes and distributes 1⁄8 of the change in momentum to each
node.
𝑘+1 𝑙+1𝑗+1
1
p𝑗+1⁄ ,𝑘+1⁄ ,𝑙+1⁄ = 𝜌𝑗+1⁄ ,𝑘+1⁄ ,𝑙+1⁄ ∑ ∑ ∑ v𝐽𝐾𝐿 , (20.2.123)
2 2 2 8 2 2 2
𝐽=𝑗 𝐾=𝑘 𝐿=𝑙

𝑘+1 𝑙+1 𝑗+1


1
p𝑗+1⁄ ,𝑘+1⁄ ,𝑙+1⁄ = 𝑀𝑗+1⁄ ,𝑘+1⁄ ,𝑙+1⁄ ∑ ∑ ∑ v𝐽𝐾𝐿 , (20.2.124)
2 2 2 8 2 2 2
𝐽=𝑗 𝐾=𝑘 𝐿=𝑙

2 𝑗+1⁄
2 𝑘+1⁄
2 𝑙+1⁄
1 ⎛ − − 1 ⎞
v+
𝑗,𝑘,𝑙 = + ⎜ ⎜
⎜ 𝑀 𝑣
𝑗,𝑘,𝑙 𝑗,𝑘,𝑙 + ∑ ∑ ∑ 𝛥𝑃𝐽,𝐾,𝐿 ⎟

⎟. (20.2.125)
𝑀𝑗,𝑘,𝑙 8 1 1 1
⎝ 𝐽=𝑗− ⁄2 𝐾=𝑘− ⁄2 𝐿=𝑙− ⁄2 ⎠

The HIS algorithm is also readily extended to three dimensions. The variable
definitions are given in Equation (1.2.126) and Equation (1.2.127), where the subscript A
refers to the local numbering of the nodes in the element. In an unstructured mesh, the
relative orientation of the nodal numbering within the elements may change. The subscript
̃ is the local
A is always with reference to the numbering in element 𝑗, 𝑘, 𝑙. The subscript A
node number in an adjacent element that refers to the same global node number as A.
ΨA,𝑗+1⁄ ,𝑘+1⁄ ,𝑙+1⁄ = vA,𝑗+1⁄ ,𝑘+1⁄ ,𝑙+1⁄ , (20.2.126)
2 2 2 2 2 2

2 𝑗+1⁄ 𝑘+1⁄2 𝑙+1⁄2


1 +
v+
𝑗,𝑘,𝑙 = + ∑ ∑ ∑ MJ,K,L ΨA
̃ ,J,K,L . (20.2.127)
M𝑗,𝑘,𝑙
𝐽=𝑗− ⁄2 𝐾=𝑘−1⁄2 𝐿=𝑙−1⁄2
1

20.3 The Manual Rezone


The central limitation to the simplified ALE formulation is that the topology of the
mesh is fixed. For a problem involving large deformations, a mesh that works well at early
LS-DYNA DEV 10/27/16 (r:8004) 18-23 (Simplified Arbitrary Lagrangian-Eulerian)
Simplified Arbitrary Lagrangian-Eulerian LS-DYNA Theory Manual

times may not work at late times regardless of how the mesh is distributed over the
material domain. To circumvent this difficulty, a manual rezoning capability has been
implemented in LS-DYNA. The general procedure is to 1) interrupt the calculation when
the mesh is no longer acceptable, 2) generate a new mesh with INGRID by using the
current material boundaries from LS-DYNA (the topologies of the new and old mesh are
unrelated), 3) remap the solution from the old mesh to the new mesh, and 4) restart the
calculation.

This chapter will concentrate on the remapping algorithm since the mesh generation
capability is documented in the INGRID manual [Stillman and Hallquist 1992]. The
remapping algorithm first constructs an interpolation function on the original mesh by
using a least squares procedure, and then interpolates for the solution values on the new
mesh.

The one point quadrature used in LS-DYNA implies a piecewise constant


distribution of the solution variables within the elements. A piecewise constant
distribution is not acceptable for a rezoner since it implies that for even moderately large
changes in the locations of the nodes (say, displacements on the order of fifty percent of the
elements characteristic lengths) that there will be no changes in the values of the element-
centered solution variables. A least squares algorithm is used to generate values for the
solution variables at the nodes from the element-centered values. The values of the
solution variables can then be interpolated from the nodal values, 𝜙A , using the standard
trilinear shape functions anywhere within the mesh.
𝜙(𝜉 , 𝜂, 𝜁 ) = 𝜙𝐴 NA (𝜉 , 𝜂, 𝜁 ). (20.3.128)

The objective function for minimization, 𝐽, is defined material by material, and each
material is remapped independently.
1
𝐽 = ∫(φA NA − φ)2 dV. (20.3.129)
2
V

The objective function is minimized by setting the derivatives of 𝐽 with respect to𝜙𝐴
equal to zero.
∂𝐽
= ∫(φB NB − φ)NA dV = 0. (20.3.130)
∂φA
V

The least square values of 𝜙A are calculated by solving the system of linear
equations, Equation (17.3.4).

MAB φB = ∫ NA φdV, (20.3.131)


V

18-24 (Simplified Arbitrary Lagrangian-Eulerian) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Simplified Arbitrary Lagrangian-Eulerian

MAB = ∫ NA NB dV. (20.3.132)


V

The “mass matrix”, MAB , is lumped to give a diagonal matrix. This eliminates the spurious
oscillations that occur in a least squares fit around the discontinuities in the solution (e.g.,
shock waves) and facilitates an explicit solution for 𝜙A . The integral on the right hand side
of Equation (20.3.131) is evaluated using one point integration. By introducing these
simplifications, Equation (17.3.4) is reduced to Equation (1.3.133), where the summation
over a is restricted to the elements containing node A.
∑α φα Vα
φA = . (20.3.133)
∑α Vα

The value of 𝜙α is the mean value of 𝜙 in element a. From this definition, the value
of 𝜙α is calculated using Equation (1.3.134).
1
φα = ∫ φA NA dV. (20.3.134)

The integrand in Equation (20.3.134) is defined on the old mesh, so that Equation
(20.3.134) is actually performed on the region of the old mesh that overlaps element  in the
new mesh, where the superscript “*” refers to elements on the old mesh.
1
φα = ∑ ∗ ∫ φA NA dV∗ . (20.3.135)
Vα β
Vα ∩V∗β

One point integration is currently used to evaluate Equation (20.3.135), although it would
be a trivial matter to add higher order integration. By introducing this simplification,
Equation (20.3.135) reduces to interpolating the value of 𝜙𝛼 from the least squares fit on the
old mesh.
𝜙a = 𝜙A NA (𝜉 ∗ , 𝜂∗ , 𝜁 ∗ ). (20.3.136)

The isoparametric coordinates in the old mesh that correspond to the spatial location
of the new element centroid must be calculated for Equation (20.3.136). The algorithm that
is described here is from Shapiro [1990], who references [Thompson and Maffeo 1985,
Maffeo 1984, Maffeo 1985] as the motivations for his current strategy, and we follow his
notation. The algorithm uses a “coarse filter” and a “fine filter” to make the search for the
correct element in the old mesh efficient.

The coarse filter calculates the minimum and maximum coordinates of each element
in the old mesh. If the new element centroid, (𝑥𝑠 , 𝑦𝑠 , 𝑧𝑠 ), lies outside of the box defined by
the maximum and minimum values of an old element, then the old element does not
contain the new element centroid.

LS-DYNA DEV 10/27/16 (r:8004) 18-25 (Simplified Arbitrary Lagrangian-Eulerian)


Simplified Arbitrary Lagrangian-Eulerian LS-DYNA Theory Manual

Several elements may pass the coarse filter but only one of them contains the new
centroid. The fine filter is used to determine if an element actually contains the new
centroid. The fine filter algorithm will be explained in detail for the two-dimensional case
since it easier to visualize than the three-dimensional case, but the final equations will be
given for the three-dimensional case.

The two edges adjacent to each node in Figure 20.2 (taken from [Shapiro 1990])
define four skew coordinate systems. If the coordinates for the new centroid are positive
for each coordinate system, then the new centroid is located within the old element.
Because of the overlap of the four positive quarter spaces defined by the skew coordinate
systems, only two coordinate systems actually have to be checked. Using the first and third
coordinate systems, the coordinates, α𝑖 , are the solution of Equation (17.3.9).
𝑉𝑠 = 𝑉1 + 𝑎1 𝑉12 + 𝑎2 𝑉14 , (20.3.137)

𝑉𝑠 = 𝑉3 + 𝑎3 𝑉32 + 𝑎4 𝑉34 . (20.3.138)

Two sets of linear equations are generated for the α𝑖 by expanding the vector
equations.
𝑥 −𝑥 𝑥 −𝑥 𝛼 𝑥 −𝑥
[𝑦2 − 𝑦1 𝑦4 − 𝑦1 ] {𝛼1 } = {𝑦𝑠 − 𝑦1 }, (20.3.139)
2 1 4 1 2 𝑠 1

𝑥 −𝑥 𝑥4 − 𝑥3 𝛼3 𝑥𝑠 − 𝑥3
[𝑦2 − 𝑦3 𝑦4 − 𝑦3 ] {𝛼4 } = {𝑦𝑠 − 𝑦3 }. (20.3.140)
2 3

The generalization of Equation (17.3.9) to three dimensions is given by Equation


(17.3.11), and it requires the solution of four sets of three equations. The numbering
convention for the nodes in Equation (17.3.11) follows the standard numbering scheme
used in LS-DYNA for eight node solid elements.
𝑉s = 𝑉1 + 𝑎1 𝑉12 + 𝑎2 𝑉14 + 𝑎3 𝑉15 , (20.3.141)

𝑉s = 𝑉3 + 𝑎4 𝑉37 + 𝑎5 𝑉34 + 𝑎6 𝑉32 , (20.3.142)

𝑉s = 𝑉6 + 𝑎7 𝑉62 + 𝑎8 𝑉65 + 𝑎9 𝑉67 , (20.3.143)

𝑉s = 𝑉8 + 𝑎10 𝑉85 + 𝑎11 𝑉84 + 𝑎12 𝑉87 . (20.3.144)

18-26 (Simplified Arbitrary Lagrangian-Eulerian) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Simplified Arbitrary Lagrangian-Eulerian

y
4

3
(xs, ys)
1 α4v34 α3v32
α1v12
v1 α2v14

vs 2

x
Figure 20.2. Skew Coordinate System

The fine filter sometimes fails to locate the correct element when the mesh is
distorted. When this occurs, the element that is closest to the new centroid is located by
finding the element for which the sum of the distances between the new centroid and the
nodes of the element is a minimum.

Once the correct element is found, the isoparametric coordinates are calculated
using the Newton-Raphson method, which usually converges in three or four iterations.
∂𝑁A ∂𝑁A ∂𝑁A
⎡𝑥A 𝑥A 𝑥A ⎤
⎢ ∂𝜉 ∂𝜂 ∂𝜁 ⎥
⎢ ∂𝑁 ∂𝑁A ⎥ ⎧Δ𝜉 ⎫ ⎧𝑥 − 𝑥 𝑁 ⎫
⎢ A ∂𝑁A ⎥ {Δ𝜂 } {𝑦s − 𝑦A 𝑁A }
⎢𝑦A 𝑦A 𝑦A ⎥ =⎨ s A A ⎬, (20.3.145)
∂𝜉 ∂𝜂 ∂𝜁 ⎥ ⎨
{ ⎬

⎢ ∂𝑁 ⎥ ⎩Δ𝜁 ⎭ ⎩𝑧s − 𝑧A 𝑁A }
} { ⎭
⎢𝑧 A ∂𝑁A ∂𝑁A ⎥
A 𝑧A 𝑧A
⎣ ∂𝜉 ∂𝜂 ∂𝜁 ⎦

𝜉 𝑖+1 = 𝜉 𝑖 + 𝛥𝜉 ,
𝜂𝑖+1 = 𝜂𝑖 + 𝛥𝜂, (20.3.146)
𝜁 𝑖+1 = 𝜁 𝑖 + 𝛥𝜁 .

LS-DYNA DEV 10/27/16 (r:8004) 18-27 (Simplified Arbitrary Lagrangian-Eulerian)


LS-DYNA Theory Manual Stress Update Overview

21
Stress Update Overview

21.1 Jaumann Stress Rate


Stresses for material which exhibit elastic-plastic and soil-like behavior (hypoelastic)
are integrated incrementally in time:
𝜎𝑖𝑗 (𝑡 + 𝑑𝑡) = 𝜎𝑖𝑗 (𝑡) + 𝜎̇ 𝑖𝑗 𝑑𝑡, (21.1)
Here, and in equations which follow, we neglect the contribution of the bulk viscosity to
the stress tensor. In Equation (21.1), the dot denotes the material time derivative given by
𝜎̇ 𝑖𝑗 = 𝜎𝑖𝑗∇ + 𝜎𝑖𝑘 𝜔𝑘𝑗 + 𝜎𝑗𝑘 𝜔𝑘𝑖 , (21.2)
in which
1 ∂𝑣 ∂𝑣𝑗
𝜔𝑖𝑗 = ( 𝑖 − ), (21.3)
2 ∂𝑥𝑗 ∂𝑥𝑖
is the spin tensor and
𝜎𝑖𝑗∇ = 𝐶𝑖𝑗𝑘𝑙 𝜀̇𝑘𝑙 , (21.4)
is the Jaumann (co-rotational) stress rate. In Equation (21.4), 𝐶𝑖𝑗𝑘𝑙 is the stress dependent
constitutive matrix, 𝑣𝑖 , is the velocity vector, and 𝜀̇𝑖𝑗 is the strain rate tensor:
1 ∂𝑣 ∂𝑣𝑗
𝜀̇𝑖𝑗 = ( 𝑖 + ). (21.5)
2 ∂𝑥𝑗 ∂𝑥𝑖

In the implementation of Equation (21.1) we first perform the stress rotation,


Equation (21.2), and then call a constitutive subroutine to add the incremental stress
components 𝜎𝑖𝑗𝛻 . This may be written as
𝛻𝑛+1⁄2 𝑛+1⁄2
𝜎𝑖𝑗𝑛+1 = 𝜎𝑖𝑗𝑛 + 𝑟𝑖𝑗𝑛 + 𝜎𝑖𝑗 Δ𝑡 , (21.6)

where
LS-DYNA DEV 10/27/16 (r:8004) 19-1 (Stress Update Overview)
Stress Update Overview LS-DYNA Theory Manual

𝛻𝑛+1⁄2 𝑛+1⁄2 𝑛+1⁄2


𝜎𝑖𝑗 Δ𝑡 = 𝐶𝑖𝑗𝑘𝑙 Δ𝜀𝑘𝑙 ,
(21.7)
𝑛+1⁄ 𝑛+1⁄ 𝑛+1⁄2
Δ𝜀𝑖𝑗 2 = 𝜀̇𝑖𝑗 2 Δ𝑡 ,

and 𝑟𝑖𝑗𝑛 gives the rotation of the stress at time 𝑡𝑛 to the configuration at 𝑡𝑛 + 1
𝑛+1⁄2 𝑛+1⁄2 𝑛+1⁄2
𝑟𝑖𝑗𝑛 = (𝜎𝑖𝑝
𝑛
𝜔𝑝𝑗 𝑛
+ 𝜎𝑗𝑝 𝜔𝑝𝑖 ) Δ𝑡 . (21.8)

In the implicit NIKE2D/3D [Hallquist 1981b] codes, which are used for low
frequency structural response, we do a half-step rotation, apply the constitutive law, and
complete the second half-step rotation on the modified stress. This approach has also been
adopted for some element formulations in LS-DYNA when the invariant stress update is
active. An exact or second order accurate rotation is performed rather than the
approximate one represented by Equation (21.3), which is valid only for small incremental
rotations. A typical implicit time step is usually 100 to 1000 or more times larger than the
explicit time step; consequently, the direct use of Equation (21.7) could lead to very
significant errors.

21.2 Jaumann Stress Rate Used With Equations of State


If pressure is defined by an equation of state as a function of relative volume, 𝑉, and
energy, 𝐸, or temperature, 𝑇,
𝑝 = 𝑝(𝑉, 𝐸) = 𝑝(𝑉, 𝑇), (21.9)
we update the deviatoric stress components
𝑛+1⁄2 𝑛+1⁄2
𝑠𝑛+1
𝑖𝑗 = 𝜎𝑖𝑗𝑛 + 𝑟𝑖𝑗𝑛 + 𝑝𝑛 𝛿𝑖𝑗 + 𝐶𝑖𝑗𝑘𝑙 𝜀̇′𝑘𝑙 𝛥𝑡 , (21.10)

𝑛+1⁄2
where 𝜀̇′𝑖𝑗 is the deviatoric strain rate tensor:
𝑛+1⁄2 1
𝜀̇′𝑖𝑗 = 𝜀̇𝑖𝑗 − 𝜀̇𝑘𝑘 𝛿. (21.11)
3

Before the equation of state, Equation (21.9), is evaluated, we compute the bulk
viscosity, 𝑞, and update the total internal energy 𝑒 of the element being processed to a trial
value 𝑒∗ :
1 𝑛−1⁄2 𝑛+1⁄2 𝑛+1⁄2 𝑛+1⁄2 𝑛+1⁄
𝑒∗𝑛+1 = 𝑒𝑛 − Δ𝑣 (𝑝𝑛 + 𝑞 +𝑞 )+𝑣 𝑠𝑖𝑗 Δ𝜀𝑖𝑗 2 , (21.12)
2
where 𝑣 is the element volume and

19-2 (Stress Update Overview) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Stress Update Overview

1
𝑛+1⁄2 𝑛+1 1
Δ𝑣 = 𝑣𝑛+1 − 𝑣𝑛 , 𝑣 = (𝑣𝑛 + 𝑣𝑛+1 ), 𝑠𝑖𝑗 2 = (𝑠𝑛𝑖𝑗 + 𝑠𝑛+1
𝑖𝑗 ).
(21.13)
2 2
The time-centering of the viscosity is explained by Noh [1976].

Assume we have an equation of state that is linear in internal energy of the form
𝑝𝑛+1 = 𝐴𝑛+1 + 𝐵𝑛+1 𝐸𝑛+1 , (21.14)
where

𝑛+1 𝑒𝑛+1
𝐸 = , (21.15)
𝑣0
and 𝜐0 is the initial volume of the element. Noting that
1
𝑒𝑛+1 = 𝑒∗𝑛+1 − Δ𝑣𝑝𝑛+1 , (21.16)
2
pressure can be evaluated exactly by solving the implicit form
(𝐴𝑛+1 + 𝐵𝑛+1 𝐸∗𝑛+1 )
𝑝𝑛+1 = , (21.17)
1
(1 + 𝐵 𝑛+1 Δ𝑣
2 𝑣0 )
and the internal energy can be updated in Equation (21.16). If the equation of state is not
linear in internal energy, a one-step iteration is used to approximate the pressure
𝑝∗𝑛+1 = 𝑝(𝑉 𝑛+1 , 𝐸∗𝑛+1 ). (21.18)
Internal energy is updated to 𝑛 + 1 using 𝑝∗𝑛+1 in Equation (21.16) and the final pressure is
then computed:
𝑝𝑛+1 = 𝑝(𝑉 𝑛+1 , 𝐸𝑛+1 ). (21.19)
This is also the iteration procedure used in KOVEC [Woodruff 1973]. All the equations of
state in LS-DYNA are linear in energy except the ratio of polynomials.

21.3 Green-Naghdi Stress Rate


The Green-Naghdi rate is defined as
𝜎𝑖𝑗𝛻 = 𝜎̇ 𝑖𝑗 + 𝜎𝑖𝑘 𝛺𝑘𝑗 + 𝜎𝑗𝑘 𝛺𝑘𝑖 = 𝑅𝑖𝑘 𝑅𝑗𝑙 𝜏̇ 𝑘𝑙 , (21.20)
where 𝛺𝑖𝑗 is defined as
𝛺𝑖𝑗 = 𝑅̇ 𝑖𝑘 𝑅𝑖𝑘 , (21.21)
and 𝐑 is found by application of the polar decomposition theorem
𝐹𝑖𝑗 = 𝑅𝑖𝑘 𝑈𝑘𝑗 = 𝑉𝑖𝑘 𝑅𝑘𝑗 , (21.22)

LS-DYNA DEV 10/27/16 (r:8004) 19-3 (Stress Update Overview)


Stress Update Overview LS-DYNA Theory Manual

𝐹𝑖𝑗 is the deformation gradient matrix and 𝑈𝑖𝑗 and 𝑉𝑖𝑗 are the positive definite right and left
stretch tensors:
𝜕𝑥𝑖
𝐹𝑖𝑗 = . (21.23)
𝜕𝑋𝑗

Stresses are updated for all materials by adding to the rotated Cauchy stress at time
n.
𝜏𝑖𝑗𝑛 = 𝑅𝑛𝑘𝑖 𝑅𝑛𝑙𝑗 𝜎𝑘𝑙
𝑛
, (21.24)
the stress increment obtained by an evaluation of the constitutive equations,
𝑛+1⁄2 𝑛+1⁄2
Δ𝜏𝑖𝑗 = 𝐶𝑖𝑗𝑘𝑙 Δ𝑑𝑘𝑙 , (21.25)

where
𝑛+1⁄2 𝑛+1⁄2 𝑛+1⁄2 𝑛+1⁄2
Δ𝑑𝑖𝑗 = 𝑅𝑘𝑖 𝑅𝑙𝑗 Δ𝜀𝑘𝑙 (21.26)

𝐶𝑖𝑗𝑘𝑙 = constitutive matrix


Δ𝜀𝑘𝑙 = increment in strain
and to obtain the rotated Cauchy stress at 𝑡𝑛+1 , i.e.,
𝑛+1⁄2
𝜏𝑖𝑗𝑛+1 = 𝜏𝑖𝑗𝑛 + Δ𝜏𝑖𝑗 . (21.27)

The desired Cauchy stress at 𝑛 + 1 can now be found


𝜎𝑖𝑗𝑛+1 = 𝑅𝑛+1 𝑛+1 𝑛+1
𝑖𝑘 𝑅𝑗𝑙 𝜏𝑘𝑙 . (21.28)
Because we evaluate our constitutive models in the rotated configuration, we avoid the
need to transform history variables such as the back stress that arises in kinematic
hardening.

In the computation of 𝐑, Taylor and Flanagan [1989] did an incremental update in


contrast with the direct polar decomposition used in the NIKE3D code. Following their
notation the algorithm is given by.

19-4 (Stress Update Overview) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Stress Update Overview

𝑧𝑖 = 𝑒𝑖𝑗𝑘 𝑉𝑗𝑚 𝜀̇𝑚𝑘 ,


−1
𝜛𝑖 = 𝑒𝑖𝑗𝑘 𝜔𝑗𝑘 − 2[𝑉𝑖𝑗 − 𝛿𝑖𝑗 𝑉𝑘𝑘 ] 𝑧𝑗 ,
1
𝛺𝑖𝑗 = 𝑒𝑖𝑗𝑘 𝜛𝑘 ,
2
(21.29)
𝛥𝑡 𝛥𝑡
(𝛿𝑖𝑘 − 𝛺𝑖𝑘 ) 𝑅𝑛+1 𝑘𝑗 = (𝛿𝑖𝑘 + 𝛺𝑖𝑘 ) 𝑅𝑛𝑘𝑗 ,
2 2
𝑉̇𝑖𝑗 = (𝜀̇𝑖𝑘 + 𝜔𝑖𝑘 )𝑉𝑘𝑗 − 𝑉𝑖𝑘 𝛺𝑘𝑗 ,

𝑉𝑖𝑗𝑛+1 = 𝑉𝑖𝑗𝑛 + 𝛥𝑡𝑉̇𝑖𝑗 .

We have adopted the PRONTO3D approach in LS-DYNA due to numerical


difficulties with the polar decomposition in NIKE3D. We believe the PRONTO3D
approach is reliable. Several disadvantages of the PRONTO3D approach include 300+
operations (at least that is the number we got), the requirement of 15 additional variables
per integration point, and if we rezone the material in the future the initial geometry will
need to be remapped and the 15 variables initialized.

21.4 Elastoplastic Materials

At low stress levels in elastoplastic materials the stresses, 𝜎𝑖𝑗 , depends only on the
state of strain; however, above a certain stress level, called the yield stress, 𝜎𝑦 (𝑎𝑖 ),

Initial uniaxial σy = σy(ai)


yield point σy0 experimental curve
L0
L

ε ε = ln(L/L0)
elastic plastic strain σ = P/A
strain
Figure 21.1. The uniaxial tension test demonstrates plastic behavior.

LS-DYNA DEV 10/27/16 (r:8004) 19-5 (Stress Update Overview)


Stress Update Overview LS-DYNA Theory Manual

σ3
yield surface σ 1 = σ2 = σ3
defined by
F(δij, k)

σ2

deviatoria plane
yield curve = intersection of the deviatoric
σ1 plane with the yield surface

Figure 21.2. The yield surface in principal stress space in pressure independent.

nonrecoverable plastic deformations are obtained. The yield stress changes with increasing
plastic deformations, which are measured by internal variables, 𝑎𝑖 .

In the uniaxial tension test, a curve like that in Figure 21.1 is generated where
logrithmic uniaxial strain is plotted against the uniaxial true stress which is defined as the
applied load 𝑃 divided by the actual cross-sectional area, 𝐴.

For the simple von Mises plasticity models the yield stress is pressure independent
and the yield surface is a cylinder in principal stress space as shown in Figure 21.2. With
isotropic hardening the diameter of the cylinder grows but the shape remains circular. In
kinematic hardening the diameter may remain constant but will translate in the plane as a
function of the plastic strain tensor, See Figure 21.3.

The equation describing the pressure independent yield surface, 𝐹, is a function of


the deviatoric stress tensor, 𝑠𝑖𝑗 , of a form given in Equation (1.30).
𝐹(𝑠𝑖𝑗 , 𝑎𝑖 ) = 𝑓 (𝑠𝑖𝑗 ) − 𝜎𝑦 (𝑎𝑖 ) = 0, (21.30)
𝑓 (𝑠𝑖𝑗 ) = determines the shape,
𝜎𝑦 (𝑎𝑖 ) = determines the translation and size.

The existence of a potential function, 𝑔, called the plastic potential, is assumed


𝑔 = 𝑔(𝑠𝑖𝑗 ). (21.31)
Stability and uniqueness require that:
p 𝜕𝑔
𝑑𝜀𝑖𝑗 = 𝜆 , (21.32)
𝜕𝑠𝑖𝑗
where 𝜆 is a proportionality constant.
19-6 (Stress Update Overview) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Stress Update Overview

σ3

initial yield
curve in the current
deviatoric yield
plane surface

σ1 σ2
Figure 21.3. With kinematic hardening the yield surface may shift as a function
of plastic strain.

p
As depicted in Figure 21.5 the plastic strain increments 𝑑𝜀𝑖𝑗 are normal to the plastic
potential function. This is the normality rule of plasticity.

The plastic potential 𝑔 is identical with the yield condition 𝐹(𝑠𝑖𝑗 )


𝑔 ≡ 𝐹. (21.33)
Hence:
𝑝 𝜕𝑓
𝑑𝜀𝑖𝑗 = 𝜆 = 𝜆 grad𝑓 (21.34)
𝜕𝑠𝑖𝑗
∂𝑔
and the stress increments 𝑑𝑠𝑖𝑗 are normal to the plastic flow ∂s .
𝑖𝑗

Post-yielding behavior from uniaxial tension tests typically show the following
behaviors illustrated in Figure 21.4:

  
hardening ideal softening

 0  0  0



Figure 21.4. Hardening, ideal, and softening plasticity models.

LS-DYNA DEV 10/27/16 (r:8004) 19-7 (Stress Update Overview)


Stress Update Overview LS-DYNA Theory Manual

σ2



σ1
σ3
Figure 21.5. The plastic strain is normal to the yield surface.

The behavior of these hardening laws are characterized in Table 18.1. below.
Although LS-DYNA permits softening to be defined and used, such softening behavior will
result in strain localization and nonconvergence with mesh refinement.

21.5 Hyperelastic Materials


Stresses for elastic and hyperelastic materials are path independent; consequently,
the stress update is not computed incrementally. The methods described here are well
known and the reader is referred to Green and Adkins [1970] and Ogden [1984] for more
details.

A retangular cartesian coordinate system is used so that the covariant and contravar-
iant metric tensors in the reference (undeformed) and deformed configuration are:
𝑔𝑖𝑗 = 𝑔𝑖𝑗 = 𝛿𝑖𝑗 ,
𝜕𝑥𝑘 𝜕𝑥𝑘
𝐺𝑖𝑗 = ,
𝜕𝑋𝑖 𝜕𝑋𝑗 (21.35)
𝜕𝑋𝑖 𝜕𝑋𝑗
𝐺𝑖𝑗 = .
𝜕𝑥𝑘 𝜕𝑥𝑘

The Green-St. Venant strain tensor and the principal strain invariants are defined as

19-8 (Stress Update Overview) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Stress Update Overview

1
𝛾𝑖𝑗 = (𝐺𝑖𝑗 − 𝛿𝑖𝑗 ), (21.36)
2

𝐼1 = 𝛿𝑖𝑗 𝐺𝑖𝑗 ,
1
𝐼2 = (𝛿𝑖𝑟 𝛿𝑗𝑠 𝐺𝑟𝑖 𝐺𝑠𝑗 − 𝛿𝑖𝑟 𝛿𝑗𝑠 𝐺𝑖𝑗 𝐺𝑟𝑠 ), (21.37)
2
𝐼3 = det(𝐺𝑖𝑗 ),

Hardening Ideal Softening

Behavior 𝜎𝑦 (𝑎𝑖 ) is monotonic 𝜎𝑦 (𝑎𝑖 ) is constant 𝜎𝑦 (𝑎𝑖 ) is monotonic


increasing decreasing

Stability yes yes No

Uniqueness yes yes No

Applications metals, crude idealization dense sand,


concrete, rock with for steel, plastics, concrete with
small deformations etc. large deformations
Table 18.1. Plastic hardening, ideal plasticity, and softening.

For a compressible elastic material the existence of a strain energy functional, 𝑊, is


assumed
𝑊 = 𝑊(𝐼1 , 𝐼2 , 𝐼3 ), (21.38)
which defines the energy per unit undeformed volume. The stress measured in the
deformed configuration is given as [Green and Adkins, 1970]:
𝑠𝑖𝑗 = 𝛷𝑔𝑖𝑗 + 𝛹𝐵𝑖𝑗 + 𝑝𝐺𝑖𝑗 , (21.39)
where
2 𝜕𝑊
𝛷= ,
√𝐼3 𝜕𝐼1
2 𝜕𝑊
𝛹= ,
√𝐼3 𝜕𝐼2 (21.40)
𝜕𝑊
𝑝 = 2√𝐼3 ,
𝜕𝐼3
𝐵𝑖𝑗 = 𝐼1 𝛿𝑖𝑗 − 𝛿𝑖𝑟 𝛿𝑗𝑠 𝐺𝑟𝑠 .

This stress is related to the second Piola-Kirchhoff stress tensor:

LS-DYNA DEV 10/27/16 (r:8004) 19-9 (Stress Update Overview)


Stress Update Overview LS-DYNA Theory Manual

𝑆𝑖𝑗 = 𝑠𝑖𝑗 √𝐼3 . (21.41)


Second Piola-Kirchhoff stresses are transformed to physical (Cauchy) stresses according to
the relationship:
𝜌 𝜕𝑥𝑖 𝜕𝑥𝑗
𝜎𝑖𝑗 = 𝑆 . (21.42)
𝜌0 𝜕𝑋𝑘 𝜕𝑋𝑙 𝑘𝑙

21.6 Layered Composites


The composite models for shell elements in LS-DYNA include models for elastic
behavior and inelastic behavior. The approach used here for updating the stresses also
applies to the airbag fabric model.

To allow for an arbitrary orientation of the shell elements within the finite element
mesh, each ply in the composite has a unique orientation angle, 𝛽, which measures the
offset from some reference in the element. Each integration point through the shell
thickness, typically though not limited to one point per ply, requires the definition of 𝛽at
that point. The reference is determined by the angle 𝛹which can be defined for each
element on the element card, and is measured from the 1-2 element side. Figures 21.6 and
21.7 depict these angles.

We update the stresses in the shell in the local shell coordinate system which is
defined by the 1-2 element side and the cross product of the diagonals. Thus to transform
the stress tensor into local system determined by the fiber directions entails a transfor-
mation that takes place in the plane of the shell.

In the implementation of the material model we first transform the Cauchy stress
and velocity strain tensor 𝑑𝑖𝑗 into the coordinate system of the material denoted by the
subscript L
𝛔L = 𝐪T 𝛔𝐪,
𝐪L = 𝐪T 𝐝𝐪,
𝜎11 𝜎12 𝜎13
𝛔L = ⎡
⎢𝜎21 𝜎22 𝜎23 ⎤
⎥, (21.43)
⎣𝜎32 𝜎32 𝜎33 ⎦
𝑑 𝑑12 𝑑13
⎡ 11 ⎤
𝛆L = ⎢𝑑21 𝑑22 𝑑23 ⎥,
⎣𝑑32 𝑑32 𝑑33 ⎦

19-10 (Stress Update Overview) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Stress Update Overview

θ = ψ+β

z y
θ

x
Figure 21.7. A multi-layer laminate can be defined. The angle i is defined for
the ith lamina.

The Arabic subscripts on the stress and strain (𝛔 and 𝛆) are used to indicate the
principal material directions where 1 indicates the fiber direction and 2 indicates the
transverse fiber direction (in the plane). The orthogonal 3 × 3 transformation matrix is
given by
cos𝜃 −sin𝜃 0
𝐪=⎡
⎢ sin𝜃 cos𝜃 0⎤⎥. (21.44)
⎣ 0 0 1⎦

In shell theory we assume a plane stress condition, i.e., that the normal stress, 𝜎33 , to the
mid-surface is zero. We can now incrementally update the stress state in the material
coordinates
𝑛+1⁄2
𝛔L𝑛+1 = 𝛔L𝑛 + Δ𝛔L , (21.45)

where for an elastic material

n4 n3
3 1
2 β
ψ
n1 X
n2
Figure 21.6. Orientation of material directions relative to the 1-2 side.

LS-DYNA DEV 10/27/16 (r:8004) 19-11 (Stress Update Overview)


Stress Update Overview LS-DYNA Theory Manual

Δ𝜎 𝑄 𝑄12 0 0 0 𝑑
⎡Δ𝜎11 ⎤ ⎡𝑄11 ⎡ 11 ⎤
⎢ 22 ⎥ ⎢ 12 𝑄22 0 0 0 ⎤⎥ ⎢𝑑22 ⎥
𝑛+1⁄2
Δ𝛔L = ⎢Δ𝜎12 ⎥ = ⎢ 0 0 𝑄44 0 0 ⎥ ⎢𝑑12 ⎥ Δ𝑡. (21.46)
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢Δ𝜎23 ⎥ ⎢ 0 0 0 𝑄55 0 ⎥ ⎢𝑑23 ⎥
⎣Δ𝜎31 ⎦ ⎣ 0 0 0 0 𝑄66 ⎦ ⎣𝑑31 ⎦L
The terms 𝑄𝑖𝑗 are referred to as reduced components of the lamina and are defined as
𝐸11
𝑄11 = ,
1 − 𝜈12 𝜈21
𝐸22
𝑄22 = ,
1 − 𝜈12 𝜈21
𝜈12 𝐸11 (21.47)
𝑄12 = ,
1 − 𝜈12 𝜈21
𝑄44 = 𝐺12 ,
𝑄55 = 𝐺23 ,
𝑄66 = 𝐺31 .
Because of the symmetry properties,
𝐸𝑗𝑗
𝜈𝑗𝑖 = 𝜈𝑖𝑗 , (21.48)
𝐸𝑖𝑖
where 𝜈𝑖𝑗 is Poisson’s ratio for the transverse strain in jth direction for the material
undergoing stress in the ith-direction, 𝐸𝑖𝑗 are the Young’s modulii in the ith direction, and
𝐺𝑖𝑗 are the shear modulii.

After completion of the stress update we transform the stresses back into the local
shell coordinate system.
𝛔 = 𝐪𝛔L 𝐪T . (21.49)

21.7 Constraints on Orthotropic Elastic Constants


The inverse of the constitutive matrix 𝐂l is generally defined in terms of the local
material axes for an orthotropic material is given in terms of the nine independent elastic
constants as

19-12 (Stress Update Overview) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Stress Update Overview

1 𝜐 21 𝜐 31
⎡ − − 0 0 0 ⎤
⎢ 𝐸 11 𝐸 22 𝐸 33 ⎥
⎢ 𝜐 1 𝜐 ⎥
⎢− 12 − 32 0 0 0 ⎥
⎢ 𝐸 11 𝐸 22 𝐸 33 ⎥
⎢ ⎥
⎢− 𝜐 13 𝜐
− 23
1
0 0 0 ⎥
⎢ 𝐸 𝐸 22 𝐸 33 ⎥
𝐂−1
l =⎢ 11 ⎥. (21.50)
⎢ 1 ⎥
⎢ 0 0 0 0 0 ⎥
⎢ 𝐺12 ⎥
⎢ 1 ⎥
⎢ 0 0 0 0 0 ⎥
⎢ 𝐺23 ⎥
⎢ ⎥
⎢ 0 1 ⎥
0 0 0 0
⎣ 𝐺31 ⎦

As discussed by Jones [1975], the constants for a thermodynamically stable material


must satisfy the following inequalities:
𝐸1 , 𝐸2 , 𝐸3 , 𝐺12 , 𝐺23 , 𝐺31 > 0,
𝐶11, 𝐶22 , 𝐶33 , 𝐶44 , 𝐶55 , 𝐶66 > 0,
(21.51)
(1 − 𝜈23 𝜈32 ), (1 − 𝜈13 𝜈31 ), (1 − 𝜈12 𝜈21 ) > 0,
1 − 𝜈12 𝜈21 − 𝜈23 𝜈32 − 𝜈31 𝜈13 − 2𝜈21 𝜈32 𝜈13 > 0.

Using Equation (21.48) and (21.51) leads to:


1 1
𝐸22 ⁄2 𝐸11 ⁄2
|𝜈21 | < ( ) |𝜈12 | < ( ) ,
𝐸11 𝐸22
⁄2 1 1⁄2
𝐸 𝐸
∣𝜈32 ∣ < ( 33 ) ∣𝜈23 ∣ < ( 22 ) , (21.52)
𝐸22 𝐸33
1 1
𝐸11 ⁄2 𝐸33 ⁄2
∣𝜈13 ∣ < ( ) ∣𝜈31 ∣ < ( ) .
𝐸33 𝐸11

21.8 Local Material Coordinate Systems in Solid Elements


In solid elements there is a number of different ways of defining a local coordinate
system. Perhaps the most general is by defining a triad for each element that is oriented in
the local material directions, See Figure 21.8. In this approach two vectors 𝐚 and 𝐝 are
defined. The local 𝐜 direction is found from the cross product, 𝐜 = 𝐚 × 𝐝, the local 𝐛
direction is the cross product 𝐛 = 𝐜 × 𝐚. This triad is stored as history data at each
integration point.

LS-DYNA DEV 10/27/16 (r:8004) 19-13 (Stress Update Overview)


Stress Update Overview LS-DYNA Theory Manual

c d

a
b

Figure 21.8. Local material directions are defined by a triad which can be input
for each solid element.

The biggest concern when dealing with local material directions is that the results
are not invariant with element numbering since the orientation of the local triad is fixed
with respect to the base of the brick element, nodes 1-4, in Figure 21.9. For Hyperelastic
materials where the stress tensor is computed in the initial configuration, this is not a
problem, but for materials like the honeycomb foams, the local directions can change due
to element distortion causing relative movement of nodes 1-4. In honeycomb foams we
assume that the material directions are orthogonal in the deformed configuration since the
stress update is performed in the deformed configuration.

21.9 General Erosion Criteria for Solid Elements

19-14 (Stress Update Overview) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Stress Update Overview

Several erosion criteria are available that are independent of the material models.
Each one is applied independently, and once any one of them is satisfied, the element is
deleted from the calculation. The criteria for failure are:
• 𝑃 ≥ 𝑃min where P is the pressure (positive in compression), and 𝑃min is the pressure
at failure.
• 𝜎1 ≥ 𝜎max , where 𝜎1 is the maximum principal stress, and 𝜎max is the principal
stress at failure.

• √32 𝜎𝑖𝑗′ 𝜎𝑖j′ ≥ 𝜎


̅̅̅̅max , where 𝜎𝑖𝑗′ are the deviatoric stress components, and 𝜎
̅̅̅̅max is the
equivalent stress at failure.
• 𝜀1 ≥ 𝜀max , where 𝜀1 is the maximum principal strain, and 𝜀max is the principal strain
at failure.
• 𝛾1 ≥ 𝛾max , where 𝛾1 is the shear strain, and 𝛾max is the shear strain at failure.
• The Tuler-Butcher criterion,
𝑡
∫ [max(0, 𝜎1 − 𝜎0 )]2 𝑑𝑡 ≥ 𝐾f , (21.53)
0

where 𝜎1 is the maximum principal stress, 𝜎0 is a specified threshold stress, 𝜎1 ≥ 𝜎0 ≥ 0,


and 𝐾f is the stress impulse for failure. Stress values below the threshold value are too low
to cause fracture even for very long duration loadings. Typical constants are given in Table
18.2below [Rajendran, 1989].

These failure models apply to solid elements with one point integration in 2 and 3

5 8

6 7

1 4

2 3

Figure 21.9. The orientation of the triad for the local material directions is stored
relative to the base of the solid element. The base is defined by nodes 1-4 of the
element connectivity.

LS-DYNA DEV 10/27/16 (r:8004) 19-15 (Stress Update Overview)


Stress Update Overview LS-DYNA Theory Manual

dimensions.

Material 𝜎0 (Kbar)
1020 Steel 10.0 2 12.5
OFHC Copper 3.60 2 10.0
C1008 14.0 2 0.38
HY100 15.7 2 61.0
7039-T64 8.60 2 3.00

Table 18.2. Typical constants for the Tuler-Bucher criterion.

21.10 Strain Output to the LS-DYNA Database


The strain tensors that are output to the LS-DYNA database from the solid, shell,
and beam elements are integrated in time. These strains are similar to the logarithmic
strain measure and are based on an integration of the strain rate tensor. Admittedly, the
shear strain components do not integrate as logarithmic strain measures, but in spite of
this, we have found that the strains output from LS-DYNA are far more useful than those
computed in LS-DYNA. The time integration of the strain tensor in LS-DYNA maintains
objectivity in the sense that rigid body motions do not cause spurious straining.

Recall, the spin tensor and strain rate tensor, Equations (21.3) and (21.5),
respectively:
1 𝜕𝑣𝑖 𝜕𝑣𝑗
𝜔𝑖𝑗 = ( − ), (21.54)
2 𝜕𝑥𝑗 𝜕𝑥𝑖

1 𝜕𝑣𝑖 𝜕𝑣𝑗
𝜀̇𝑖𝑗 = ( + ). (21.55)
2 𝜕𝑥𝑗 𝜕𝑥𝑖

In updating the strains from time 𝑛 to 𝑛 + 1, the following formula is used:


𝑛+1⁄2 𝑛+1⁄2
𝜀𝑛+1
𝑖𝑗 = 𝜀𝑛𝑖𝑗 + 𝜌𝑛𝑖𝑗 + 𝜀̇𝑖𝑗 Δ𝑡 , (21.56)

where 𝜌𝑛𝑖𝑗 gives the rotational correction that transforms the strain tensor at time 𝑡𝑛 into the
configuration at 𝑡𝑛 + 1
𝑛+1⁄2 𝑛+1⁄2 𝑛+1⁄2
𝜌𝑛𝑖𝑗 = (𝜀𝑛𝑖𝑝 𝜔𝑝𝑗 + 𝜀𝑛𝑗𝑝 𝜔𝑝𝑖 ) Δ𝑡 . (21.57)

19-16 (Stress Update Overview) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Stress Update Overview

For shell elements we integrate the strain tensor at the inner and outer integration
points and output two tensors per element. When the mid surface strains are plotted in LS-
PREPOST, these are the average values.

LS-DYNA DEV 10/27/16 (r:8004) 19-17 (Stress Update Overview)


Stress Update Overview LS-DYNA Theory Manual

21.12 Strain Rate Effects in Material Models


In a constitutive modeling context, the term rate effects is used to indicate that the material
response depends on the (time) rate of a certain quantity, here called 𝑒 for the sake of
generality. This quantity is usually some kind of strain measure, and an example is when
the stress depends on the rate of strain. In explicit simulations, high frequency strain
content yields a very noisy strain rate and calls for some kind of smoothing before being
used in the stress calculations. This section presents three of the common ways to do this
averaging and that are used frequently in the material models.

21.12.1 Running N-average option


One option is to do a running average of current and previous strain rates. With this
option the rate 𝑒 ̇ of a quantity 𝑒 is averaged according to the following algorithm
𝑒 ̃̇ = unfiltered rate (raw data) (21.1)
𝑛−1
𝑒 ̇̃ + ∑𝑖=𝑛−𝑁+1 𝑒𝑖̇
𝑒𝑛̇ = (21.2)
𝑁
𝑒 ̇ = 𝑒𝑛̇ = rate used in material routine (21.3)
In these equations, the subscript denotes stored history variables necessary for computing
the running average strain rates and n denotes the current cycle number. This requires
storage of 𝑁 − 1 history variables, for most materials 𝑁 = 12.

21.12.2 Last N-average option


The second option is to compute the rate as the average of the last 𝑁 computed rates. Here
the rate is evaluated according to
𝑒 ̇̃ = unfiltered rate (raw data) (21.4)

𝑒𝑛̇ = 𝑒 ̇̃ (21.5)
𝑛
∑𝑖=𝑛−𝑁+1 𝑒𝑖̇
𝑒̇= = rate used in material routine (21.6)
𝑁
As for the running average option, the subscript denotes stored history variables necessary
for computing the running average strain rates and n is the cycle number. This also
requires storage of 𝑁 − 1 history variables, for most materials 𝑁 = 12.

18.18
LS-DYNA Theory Manual Stress Update Overview

21.12.3 Averaging over a fixed amount of time 𝑻


This option was introduced in an attempt to suppress the time step dependence as much as
possible. With this option we use 𝑁 history variables to store the approximate values of
the quantity of interest from the time 𝑡𝑛 − 𝑇 to current time 𝑡𝑛 . That is, we set
𝑖𝑇
𝑒𝑛−𝑖 = 𝑒𝑛 (𝑡𝑛 − ) , 𝑖 = 0, … , 𝑁 − 1 (21.7)
𝑁−1
Here 𝑒𝑛 (𝑡) is an approximate function of the function of interest, 𝑒(𝑡), and will be given
more specifically below. Assuming 𝑒𝑛 (𝑡) is known, the rate used in the material routine in
cycle 𝑛 is simply
𝑒 − 𝑒𝑛−𝑁+1
𝑒̇= 𝑛 = rate used in material routine (21.8)
𝑇
For the update of history variables we assume that the 𝑁 points in (21.7) together with the
newly calculated quantity (raw data) at 𝑡𝑛+1 = 𝑡 + ∆𝑡,
𝑒𝑛+1 = 𝑒(𝑡𝑛+1 ), (21.9)
completely define the function 𝑒𝑛 (𝑡) from time 𝑡𝑛 − 𝑇 to time 𝑡𝑛+1 by linear interpolation
between each control point. That is, the function 𝑒𝑛 (𝑡) is extended to time 𝑡𝑛+1 by the new
value,
𝑒𝑛 (𝑡𝑛+1 ) = 𝑒𝑛+1 . (21.10)
This function is illustrated in the figure below by the dashed line and red control points.
The updated set of history variables are
𝑖𝑇
𝑒𝑛+1−𝑖 = 𝑒𝑛 (𝑡𝑛+1 − ) , 𝑖 = 0, … , 𝑁 − 1, (21.11)
𝑁−1
and these updated points define a new approximate function 𝑒𝑛+1 (𝑡), again by linear
interpolation between the control points. In Figure Error! Reference source not found., the
updated function is illustrated by the solid line and blue control points, note that the last
point is in fact both red and blue since 𝑒𝑛 (𝑡𝑛+1 ) = 𝑒𝑛+1 (𝑡𝑛+1 ) = 𝑒𝑛+1 . In sum, the function
used for calculated the effective rate is completely redefined between steps.
For a time step ∆𝑡 that is greater than or approximately equal to 𝑇/(𝑁 − 1), this option is
more or less exactly the average rate from 𝑡 − 𝑇 to 𝑡. For smaller time steps (which is
usually the case in explicit analysis) the previous values are in practice smoothed out since
only 𝑁 variables are available for storing this time history. That is, the high frequency
content of the history is lost and this could be an interesting alternative to the previous two
options that are otherwise commonly used for these types of situations.

18.19
Material Models LS-DYNA Theory Manual

22
Material Models

LS-DYNA accepts a wide range of material and equation of state models, each with
a unique number of history variables. Approximately 150 material models are
implemented, and space has been allotted for up to 10 user-specified models.

1 Elastic
2 Orthotropic Elastic
3 Kinematic/Isotropic Elastic-Plastic
4 Thermo-Elastic-Plastic
5 Soil and Crushable/Non-crushable Foam
6 Viscoelastic
7 Blatz - Ko Rubber
8 High Explosive Burn
9 Null Hydrodynamics
10 Isotropic-Elastic-Plastic-Hydrodynamic
11 Temperature Dependent, Elastoplastic, Hydrodynamic
12 Isotropic-Elastic-Plastic
13 Elastic-Plastic with Failure Model
14 Soil and Crushable Foam with Failure Model
15 Johnson/Cook Strain and Temperature Sensitive Plasticity
16 Pseudo TENSOR Concrete/Geological Model
17 Isotropic Elastic-Plastic Oriented Crack Model
18 Power Law Isotropic Plasticity
19 Strain Rate Dependent Isotropic Plasticity
20 Rigid
21 Thermal Orthotropic Elastic
22 Composite Damage Model
23 Thermal Orthotropic Elastic with 12 Curves
24 Piecewise Linear Isotropic Plasticity
25 Inviscid Two Invariant Geologic Cap Model

20-20 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

26 Metallic Honeycomb
27 Compressible Mooney-Rivlin Rubber
28 Resultant Plasticity
29 Force Limited Resultant Formulation
30 Closed-Form Update Shell Plasticity
31 Slightly Compressible Rubber Model
32 Laminated Glass Model
33 Barlat’s Anisotropic Plasticity Model
34 Fabric
35 Kinematic/Isotropic Elastic-Plastic Green-Naghdi Rate
36 Barlat’s 3-Parameter Plasticity Model
37 Transversely Anisotropic Elastic-Plastic
38 Blatz-Ko Compressible Foam
39 Transversely Anisotropic Elastic-Plastic with FLD
40 Nonlinear Elastic Orthotropic Material
41-50 User Defined Material Models
42 Planar Anisotropic Plasticity Model
48 Strain Rate Dependent Plasticity with Size Dependent Failure
51 Temperature and Rate Dependent Plasticity
52 Sandia’s Damage Model
53 Low Density Closed Cell Polyurethane Foam
54-55 Composite Damage Model
57 Low Density Urethane Foam
58 Laminated Composite Fabric
59 Composite Failure
60 Elastic with Viscosity
61 Maxwell/Kelvin Viscoelastic
62 Viscous Foam
63 Isotropic Crushable Foam
64 Strain Rate Sensitive Power-Law Plasticity
65 Modified Zerilli/Armstrong
66 Linear Stiffness/Linear Viscous 3D Discrete Beam
67 Nonlinear Stiffness/Viscous 3D Discrete Beam
68 Nonlinear Plastic/Linear Viscous 3D Discrete Beam
69 Side Impact Dummy Damper, SID Damper
70 Hydraulic/Gas Damper
71 Cable
72 Concrete Damage Model
73 Low Density Viscoelastic Foam
74 Elastic Spring for the Discrete Beam
75 Bilkhu/Dubois Foam Model
76 General Viscoelastic
77 Hyperviscoelastic Rubber
78 Soil/Concrete
79 Hysteretic Soil
LS-DYNA DEV 10/27/16 (r:8004) 20-21 (Material Models)
Material Models LS-DYNA Theory Manual

80 Ramberg-Osgood Plasticity
81 Plastic with Damage
82 Isotropic Elastic-Plastic with Anisotropic Damage
83 Fu-Chang’s Foam with Rate Effects
84-85 Winfrith Concrete
84 Winfrith Concrete Reinforcement
86 Orthotropic-Viscoelastic
87 Cellular Rubber
88 MTS Model
89 Plasticity Polymer
90 Acoustic
91 Soft Tissue
93 Elastic 6DOF Spring Discrete Beam
94 Inelastic Spring Discrete Beam
95 Inelastic 6DOF Spring Discrete Beam
96 Brittle Damage Model
97 General Joint Discrete Beam
100 Spot weld
101 GE Thermoplastics
102 Hyperbolic Sin
103 Anisotropic Viscoplastic
104 Damage 1
105 Damage 2
106 Elastic Viscoplastic Thermal
110 Johnson-Holmquist Ceramic Model
111 Johnson-Holmquist Concrete Model
112 Finite Elastic Strain Plasticity
113 Transformation Induced Plasticity
114 Layered Linear Plasticity
115 Elastic Creep Model
116 Composite Lay-Up Model
117-118 Composite Matrix
119 General Spring and Damper Model
120 Gurson Dilational-Plastic Model
120 Gurson Model with Rc-Dc
121 Generalized Nonlinear 1DOF Discrete Beam
122 Hill 3RC
123 Modified Piecewise Linear Plasticity
124 Tension-Compression Plasticity
126 Metallic Honeycomb
127 Arruda-Boyce rubber
128 Anisotropic heart tissue
129 Lung tissue
130 Special Orthotropic
131 Isotropic Smeared Crack
20-22 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Material Models

132 Orthotropic Smeared Crack


133 Barlat YLD2000
134 Viscoelastic Fabric
139 Modified Force Limited
140 Vacuum
141 Rate Sensitive Polymer
142 Transversely Anisotropic Crushable Foam
143 Wood Model
144 Pitzer Crushable Foam
145 Schwer Murray Cap Model
146 1DOF Generalized Spring
147 FHWA Soil Model
148 Gas Mixture
150 CFD
151 EMMI
154 Deshpande-Fleck Foam
156 Muscle
158 Rate Sensitive Composite Fabric
159 Continuous Surface Cap Model
161-162 Composite MSC
163 Modified Crushable Foam
164 Brain Linear Viscoelastic
166 Moment Curvature Beam
169 Arup Adhesive
170 Resultant Anisotropic
175 Viscoelastic Maxwell
176 Quasilinear Viscoelastic
177 Hill Foam
178 Viscoelastic Hill Foam
179 Low Density Synthetic Foam
181 Simplified Rubber/Foam
183 Simplified Rubber with Damage
184 Cohesive Elastic
185 Cohesive TH
191 Seismic Beam
192 Soil Brick
193 Drucker Prager
194 RC Shear Wall
195 Concrete Beam
196 General Spring Discrete Beam
197 Seismic Isolator
198 Jointed Rock

LS-DYNA DEV 10/27/16 (r:8004) 20-23 (Material Models)


Material Models LS-DYNA Theory Manual

In the table below, a list of the available material models and the applicable element types
are given. Some materials include strain rate sensitivity, failure, equations of state, and
thermal effects and this is also noted. General applicability of the materials to certain kinds
of behavior is suggested in the last column.

Notes:
Gn General
Cm Compo-
sites
Cr Ceram-
ics
Fl Fluids
Fm Foam
Gl Glass
Hy Hydro-

Strain-Rate Effects
dyn

Equation-of-State
Material Number

Thermal Effects
Material Title Mt Metal
Pl Plastic

Thick Shells
Thin Shells
Rb Rubber
Sl

Failure
Beams
Bricks

Soil/Co
nc

1 Elastic Y Y Y Y Gn, Fl
2 Orthotropic Elastic (Anisotropic - solids) Y Y Y Cm, Mt
3 Plastic Kinematic/Isotropic Y Y Y Y Y Y Cm, Mt, Pl
4 Elastic Plastic Thermal Y Y Y Y Y Mt, Pl
5 Soil and Foam Y Fm, Sl
6 Linear Viscoelastic Y Y Y Y Rb
7 Blatz-Ko Rubber Y Y Rb,
Polyurethane
8 High Explosive Burn Y Y Hy
9 Null Material Y Y Y Y Fl, Hy
10 Elastic Plastic Hydro(dynamic) Y Y Y Hy, Mt
11 Steinberg: Temp. Dependent Y Y Y Y Y Hy, Mt
Elastoplastic
12 Isotropic Elastic Plastic Y Y Y Mt
13 Isotropic Elastic Plastic with Failure Y Y Mt
14 Soil and Foam with Failure Y Y Fm, Sl
15 Johnson/Cook Plasticity Model Y Y Y Y Y Y Hy, Mt
16 Pseudo TENSOR Geological Model Y Y Y Y Sl
17 Oriented Crack (Elastoplastic with Y Y Y Hy, Mt, Pl
Fracture)
18 Power Law Plasticity (Isotropic) Y Y Y Y Y Mt, Pl

20-24 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

19 Strain Rate Dependent Plasticity Y Y Y Y Y Mt, Pl


20 Rigid Y Y Y Y

Notes:
Gn General
Cm Com-
posites
Cr Ceram-
ics
Fl Fluids
Fm Foam
Gl Glass
Hy Hydro-

Strain-Rate Effects
dyn

Equation-of-State
Material Number

Thermal Effects
Material Title Mt Metal
Pl Plastic

Thick Shells
Thin Shells
Rb Rubber
Sl

Failure
Beams
Bricks

Soil/Co
nc

21 Orthotropic Thermal (Elastic) Y Y Y Y Gn


22 Composite Damage Y Y Y Y Cm
23 Temperature Dependent Orthotropic Y Y Y Y Cm
24 Piecewise Linear Plasticity (Isotropic) Y Y Y Y Y Y Mt, Pl
25 Inviscid Two Invariant Geologic Cap Y Sl
26 Honeycomb Y Y Y Cm, Fm, Sl
27 Mooney-Rivlin Rubber Y Y Rb
28 Resultant Plasticity Y Y Mt
29 Force Limited Resultant Formulation Y
30 Closed Form Update Shell Plasticity Y Y Mt
31 Slightly Compressible Rubber Y Rb
32 Laminated Glass (Composite) Y Y Y Cm, Gl
33 Barlat Anisotropic Plasticity Y Y Y Cr, Mt
34 Fabric Y
35 Plastic Green-Naghdi Rate Y Y Mt
36 3-Parameter Barlat Plasticity Y Mt
37 Transversely Anisotropic Elastic Plastic Y Y Mt
38 Blatz-Ko Foam Y Y Fm, Pl
39 FLD Transversely Anisotropic Y Y Mt
40 Nonlinear Orthotropic Y Y Y Cm
41- User Defined Materials Y Y Y Y Y Y Y Y Gn
50
42 Planar Anisotropic Plasticity Model

LS-DYNA DEV 10/27/16 (r:8004) 20-25 (Material Models)


Material Models LS-DYNA Theory Manual

Notes:
Gn General
Cm Com-
posites
Cr Ceram-
ics
Fl Fluids
Fm Foam
Gl Glass
Hy Hydro-

Strain-Rate Effects
dyn

Equation-of-State
Material Number

Thermal Effects
Material Title Mt Metal
Pl Plastic

Thick Shells
Thin Shells
Rb Rubber
Sl

Failure
Beams
Bricks Soil/Co
nc

51 Bamman (Temp/Rate Dependent Y Y Y Y Y Gn


Plasticity)
52 Bamman Damage Y Y Y Y Y Y Mt
53 Closed Cell Foam (Low Density Y Fm
Polyurethane)
54 Composite Damage with Change Failure Y Y Cm
55 Composite Damage with Tsai-Wu Failure Y Y Cm
56
57 Low Density Urethane Foam Y Y Y Fm
58 Laminated Composite Fabric Y
59 Composite Failure (Plasticity Based) Y Y Y Cm, Cr
60 Elastic with Viscosity (Viscous Glass) Y Y Y Y Gl
61 Kelvin-Maxwell Viscoelastic Y Y Fm
62 Viscous Foam (Crash Dummy Foam) Y Y Fm
63 Isotropic Crushable Foam Y Y Fm
64 Rate Sensitive Powerlaw Plasticity Y Y Y Y Mt
65 Zerilli-Armstrong (Rate/Temp Plasticity) Y Y Y Y Y Mt
66 Linear Elastic Discrete Beam Y Y
67 Nonlinear Elastic Discrete Beam Y Y
68 Nonlinear Plastic Discrete Beam Y Y Y
69 SID Damper Discrete Beam Y Y
70 Hydraulic Gas Damper Discrete Beam Y Y
71 Cable Discrete Beam (Elastic) Y
72 Concrete Damage Y Y Y Y Sl

20-26 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

Notes:
Gn General
Cm Compo-
sites
Cr Ceramics
Fl Fluids
Fm Foam

Strain-Rate Effects
Gl Glass

Equation-of-State
Material Number

Thermal Effects
Hy Hydro-
dyn

Thick Shells
Thin Shells
Mt Metal
Material Title Pl Plastic

Failure
Beams
Bricks
Rb Rubber
Sl Soil/Conc

73 Low Density Viscous Foam Y Y Y Fm


74 Elastic Spring for the Discrete Beam
75 Bilkhu/Dubois Foam (Isotropic) Y Y Fm
76 General Viscoelastic (Maxwell Model) Y Y Rb
77 Hyperelastic and Ogden Rubber Y Rb
78 Soil Concrete Y Y Sl
79 Hysteretic Soil (Elasto-Perfectly Y Y Sl
Plastic)
80 Ramberg Osgood Plasticity
81 Plasticity with Damage (Elasto-Plastic) Y Y Y Y Y Y Mt, Pl
82 Isotropic Elastic-Plastic with
Anisotropic Damage
83 Fu Chang Foam Y Y Y Fm
84 Winfrith Concrete Reinforcement Y
85
86 Orthotropic Viscoelastic Y Y Rb
87 Cellular Rubber Y Y Rb
88 MTS Y Y Y Y Mt
89 Plasticity Polymer Y
90 Acoustic Y Fl
91 Soft Tissue Y Y
93 Elastic 6DOF Spring Discrete Beam Y

LS-DYNA DEV 10/27/16 (r:8004) 20-27 (Material Models)


Material Models LS-DYNA Theory Manual

94 Inelastic Spring Discrete Beam Y

Notes:
Gn General
Cm Compo-
sites
Cr Ceramics
Fl Fluids
Fm Foam
Gl Glass
Hy Hydro-

Strain-Rate Effects
dyn

Equation-of-State
Material Number

Thermal Effects
Mt Metal
Material Title Pl Plastic

Thick Shells
Thin Shells
Rb Rubber
Sl

Failure
Beams
Bricks

Soil/Con
c

95 Inelastic 6DOF Spring Discrete Beam Y


96 Brittle Damage Y Y Y
97 General Joint Discrete Beam Y
98 Simplified Johnson Cook Y Y Y Y
99 Simplified Johnson Cook Orthotropic
Damage
100 Spotweld Y
101 GEPLASTIC Strate2000a Y
102 Inv Hyperbolic Sin Y
103 Anisotropic Viscoplastic Y Y
104 Damage 1 Y Y
105 Damage 2 Y Y
106 Elastic Viscoplastic Thermal Y Y Y
107
108
109
110 Johnson Holmquist Ceramics Y
111 Johnson Holmquist Concrete Y

20-28 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

112 Finite Elastic Strain Plasticity Y


113 TRIP Y Y Y Mt
114 Layered Linear Plasticity Y Y
115 Unified Creep Y

Notes:
Gn General
Cm Compo-
sites
Cr Ceramics
Fl Fluids
Fm Foam
Gl Glass
Hy Hydro-

Strain-Rate Effects
dyn

Equation-of-State
Material Number

Thermal Effects
Mt Metal
Material Title Pl Plastic

Thick Shells
Thin Shells
Rb Rubber
Sl

Failure
Beams
Bricks

Soil/Con
c

116 Composite Layup Y


117 Composite Matrix Y
118 Composite Direct Y
119 General Nonlinear 6DOF Discrete Beam Y Y Y
120 Gurson Y
121 Generalized Nonlinear 1DOF Discrete Y
Beam
122 Hill 3RC
123 Modified Piecewise Linear Plasticity Y Y
124 Plasticity Compression Tension Y
126 Modified Honeycomb Y
127 Arruda Boyce Rubber Y
128 Heart Tissue Y
129 Lung Tissue Y
130 Special Orthotropic Y
131 Isotropic Smeared Crack Y Y Mt, Cm

LS-DYNA DEV 10/27/16 (r:8004) 20-29 (Material Models)


Material Models LS-DYNA Theory Manual

132 Orthotropic Smeared Crack Y Y Mt, Cm


133 Barlat YLD2000
139 Modified Force Limited Y
140 Vacuum
141 Rate Sensitive Polymer

Notes:
Gn General
Cm Compo-
sites
Cr Ceramics
Fl Fluids
Fm Foam

Strain-Rate Effects
Gl Glass

Equation-of-State
Material Number

Thermal Effects
Hy Hydro-
dyn
Thick Shells
Thin Shells
Mt Metal
Material Title Pl Plastic
Failure
Beams
Bricks

Rb Rubber
Sl Soil/Conc

142 Transversely Anisotropic Crushable


Foam
143 Wood
144 Pitzer Crushable Foam
145 Schwer Murray Cap Model
146 1DOF Generalized Spring
147 FHWA Soil
147 FHWA Soil Nebraska
148 Gas Mixture
150 CFD
151 EMMI Y Y Y Y Y Mt
154 Deshpande Fleck Foam
156 Muscle Y Y
158 Rate Sensitive Composite Fabric Y Y Y Y Cm
159 CSMC Y Y Y Y Sl

20-30 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

161 Composite MSC Y


163 Modified Crushable Foam
164 Brain Linear Viscoelastic
166 Moment Curvature Beam Y
169 Arup Adhesive Y Y Y Pb
170 Resultant Anisotropic Y Y Pl
175 Viscoelastic Thermal Y Y Y Y Y Rb
176 Quasilinear Viscoelastic

Notes:
Gn General
Cm Compo-
sites
Cr Ceramics
Fl Fluids
Fm Foam

Strain-Rate Effects
Gl Glass

Equation-of-State
Material Number

Thermal Effects
Hy Hydro-
dyn
Thick Shells
Thin Shells

Mt Metal
Material Title Pl Plastic
Failure
Beams
Bricks

Rb Rubber
Sl Soil/Conc

177 Hill Foam


178 Viscoelastic Hill Foam
179 Low Density Synthetic Foam
181 Simplified Rubber
183 Simplified Rubber with Damage Y Y Y Y Y Rb
184 Cohesive Elastic Y Y Cm, Mt
185 Cohesive TH Y Y Cm, Mt
191 Seismic Beam Y
192 Soil Brick Y
193 Drucker Prager Y
194 RC Shear Wall Y
195 Concrete Beam Y
196 General Spring Discrete Beam Y

LS-DYNA DEV 10/27/16 (r:8004) 20-31 (Material Models)


Material Models LS-DYNA Theory Manual

197 Seismic Isolator Y Y Mt


198 Jointed Rock Y Y
DS1 Spring Elastic (Linear) Y
DS2 Damper Viscous (Linear) Y Y
DS3 Spring Elastoplastic (Isotropic) Y
DS4 Spring Nonlinear Elastic Y Y
DS5 Damper Nonlinear Elastic Y Y
DS6 Spring General Nonlinear Y

Notes:
Gn General
Cm Compo-
sites
Cr Ceramics
Fl Fluids
Fm Foam

Strain-Rate Effects
Gl Glass

Equation-of-State
Material Number

Thermal Effects
Hy Hydro-
dyn
Thick Shells
Thin Shells

Mt Metal
Material Title Pl Plastic
Failure
Beams
Bricks

Rb Rubber
Sl Soil/Conc

DS7 Spring Maxwell (Three Parameter Y Y


Viscoelastic)
DS8 Spring Inelastic (Tension or Y
Compression)
DS1 Spring Trilinear Degrading
3
DS1 Spring Squat Shearwall
4
DS1 Spring Muscle
5
SB1 Seatbelt
T01 Thermal Isotropic Y Y Y
T02 Thermal Orthotropic Y Y Y
T03 Thermal Isotropic (Temp. Dependent) Y Y Y
T04 Thermal Orthotropic (Temp. Y Y Y
Dependent)
20-32 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Material Models

T05 Thermal Isotropic (Phase Change) Y Y Y


T06 Thermal Isotropic (Temp Dep-Load Y Y Y
Curve)
T11 Thermal User Defined Y Y Y

LS-DYNA DEV 10/27/16 (r:8004) 20-33 (Material Models)


Material Models LS-DYNA Theory Manual

22.1 Material Model 1: Elastic


In this elastic material we compute the co-rotational rate of the deviatoric Cauchy
stress tensor as
𝑛+1⁄2 𝑛+1⁄2
𝑠∇
𝑖𝑗 = 2𝐺𝜀̇′𝑖𝑗 , (22.1.1)

and pressure
𝑝𝑛+1 = −𝐾ln𝑉 𝑛+1 , (22.1.2)
where 𝐺 and 𝐾 are the elastic shear and bulk moduli, respectively, and 𝑉 is the relative
volume, i.e., the ratio of the current volume to the initial volume.

20-34 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.2 Material Model 2: Orthotropic Elastic


The material law that relates second Piola-Kirchhoff stress 𝐒 to the Green-St. Venant
strain 𝐄 is
𝐒 = 𝐂 ⋅ 𝐄 = 𝐓T 𝐂l 𝐓 ⋅ 𝐄, (22.2.1)
where 𝐓 is the transformation matrix [Cook 1974].
𝑙2 𝑚21 𝑛21 𝑙1 𝑚 1 𝑚 1 𝑛1 𝑛1 𝑙 1
⎡ 12 ⎤
⎢ 𝑙2 𝑚22 𝑛22 𝑙2 𝑚 2 𝑚 2 𝑛2 𝑛2 𝑙 2 ⎥
⎢ 2 ⎥

𝐓 = ⎢ 𝑙3 𝑚23 𝑛23 𝑙3 𝑚 3 𝑚 3 𝑛3 𝑛3 𝑙 3 ⎥
⎥, (22.2.2)
⎢2𝑙1 𝑙2 2𝑚1 𝑚2 2𝑛1 𝑛2 (𝑙1 𝑚2 + 𝑙1 𝑚1 ) (𝑚1 𝑛2 + 𝑚2 𝑛1 ) (𝑛1 𝑙2 + 𝑛2 𝑙1 )⎥
⎢ ⎥
⎢2𝑙2 𝑙3 2𝑚2 𝑚3 2𝑛2 𝑛3 (𝑙2 𝑚3 + 𝑙3 𝑚2 ) (𝑚2 𝑛3 + 𝑚3 𝑛2 ) (𝑛2 𝑙3 + 𝑛3 𝑙2 )⎥
⎣2𝑙3 𝑙1 2𝑚3 𝑚1 2𝑛3 𝑛1 (𝑙3 𝑚1 + 𝑙1 𝑚3 ) (𝑚3 𝑛1 + 𝑚1 𝑛3 ) (𝑛3 𝑙1 + 𝑛1 𝑙3 )⎦
𝑙𝑖 , 𝑚𝑖 , 𝑛𝑖 are the direction cosines
𝑥𝑖′ = 𝑙𝑖 𝑥1 + 𝑚𝑖 𝑥2 + 𝑛𝑖 𝑥3 , 𝑖 = 1, 2, 3, (22.2.3)
and 𝑥𝑖′ denotes the material axes. The constitutive matrix 𝐂l is defined in terms of the
material axes as
1 𝜐21 𝜐31
⎡ − − 0 0 0 ⎤
⎢ 𝐸11 𝐸22 𝐸33 ⎥
⎢ 𝜐 1 𝜐 ⎥
⎢− 12 − 32 0 0 0 ⎥
⎢ 𝐸11 𝐸22 𝐸33 ⎥
⎢ ⎥
⎢− 𝜐13 𝜐
− 23
1
0 0 0 ⎥
⎢ 𝐸 𝐸22 𝐸33 ⎥
𝐂−1
l =⎢ 11 ⎥, (22.2.4)
⎢ 1 ⎥
⎢ 0 0 0 0 0 ⎥
⎢ 𝐺12 ⎥
⎢ 1 ⎥
⎢ 0 0 0 0 0 ⎥
⎢ 𝐺23 ⎥
⎢ ⎥
⎢ 0 1 ⎥
0 0 0 0
⎣ 𝐺31 ⎦
where the subscripts denote the material axes, i.e.,
𝜐𝑖𝑗 = 𝜐𝑥′𝑖 𝑥′𝑗 and 𝐸𝑖𝑖 = 𝐸𝑥′𝑖 . (22.2.5)
Since 𝐂l is symmetric
𝜐12 𝜐21
= , etc. (22.2.6)
𝐸11 𝐸22

The vector of Green-St. Venant strain components is


𝐄T = [𝐸11 𝐸22 𝐸33 𝐸12 𝐸23 𝐸31 ]. (22.2.7)
LS-DYNA DEV 10/27/16 (r:8004) 20-35 (Material Models)
Material Models LS-DYNA Theory Manual

After computing 𝑆𝑖𝑗 , we use Equation (21.32) to obtain the Cauchy stress. This
model will predict realistic behavior for finite displacement and rotations as long as the
strains are small.

20-36 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.3 Material Model 3: Elastic Plastic with Kinematic


Hardening
Isotropic, kinematic, or a combination of isotropic and kinematic hardening may be
obtained by varying a parameter, called 𝛽 between 0 and 1. For 𝛽 equal to 0 and 1,
respectively, kinematic and isotropic hardening are obtained as shown in Figure 22.3.1.
Krieg and Key [1976] formulated this model and the implementation is based on their
paper.

In isotropic hardening, the center of the yield surface is fixed but the radius is a
function of the plastic strain. In kinematic hardening, the radius of the yield surface is
fixed but the center translates in the direction of the plastic strain. Thus the yield condition
is
2
1 𝜎𝑦
𝜙 = 𝜉𝑖𝑗 𝜉𝑖𝑗 − = 0, (22.3.1)
2 3
where
𝜉𝑖𝑗 = 𝑠𝑖𝑗 − 𝛼𝑖𝑗 (22.3.2)
p
𝜎𝑦 = 𝜎0 + 𝛽𝐸p 𝜀eff . (22.3.3)

The co-rotational rate of 𝛼𝑖𝑗 is


2 p
𝛼∇
𝑖𝑗 = (1 − 𝛽) 𝐸p 𝜀̇𝑖𝑗 . (22.3.4)
3
Hence,
𝑛+1⁄2 𝑛+1⁄2 𝑛+1⁄2 𝑛+1⁄2
𝛼𝑛+1
𝑖𝑗 = 𝛼𝑛𝑖𝑗 + (𝛼∇
𝑖𝑗 + 𝛼𝑛𝑖𝑘 𝛺𝑘𝑗 𝑛
+ 𝛼𝑗𝑘 𝛺𝑘𝑖 ) Δ𝑡 . (22.3.5)

Strain rate is accounted for using the Cowper-Symonds [Jones 1983] model which
scales the yield stress by a strain rate dependent factor
1
⎡ 𝜀̇ 𝑝 ⎤ p
𝜎𝑦 = ⎢1 + ( ) ⎥ (𝜎0 + 𝛽𝐸p 𝜀eff ), (22.3.6)
𝐶
⎣ ⎦
where 𝑝 and 𝐶 are user defined input constants and 𝜀̇ is the strain rate defined as:

𝜀̇ = √𝜀̇𝑖𝑗 𝜀̇𝑖𝑗 . (22.3.7)

LS-DYNA DEV 10/27/16 (r:8004) 20-37 (Material Models)


Material Models LS-DYNA Theory Manual

δx
Yield
Stress

⎛l⎛
ln ⎜ ⎜
⎝ l0 ⎝

β=0, kinematic hardening

β=1, isotropic hardening

Figure 22.3.1. Elastic-plastic behavior with isotropic and kinematic hardening


where l0 and l are the undeformed and deformed length of uniaxial tension
specimen, respectively.

The current radius of the yield surface, 𝜎𝑦 , is the sum of the initial yield strength, 𝜎0 ,
p
plus the growth 𝛽𝐸p 𝜀eff ,where 𝐸p is the plastic hardening modulus
𝐸t 𝐸
𝐸p = , (22.3.8)
𝐸 − 𝐸t
p
and 𝜀eff is the effective plastic strain
𝑡 1
p 2 p p ⁄2
𝜀eff = ∫ ( 𝜀̇𝑖𝑗 𝜀̇𝑖𝑗 ) 𝑑𝑡. (22.3.9)
3
0

The plastic strain rate is the difference between the total and elastic (right
superscript e) strain rates:
p
𝜀̇𝑖𝑗 = 𝜀̇𝑖𝑗 − 𝜀̇e𝑖𝑗 . (22.3.10)

In the implementation of this material model, the deviatoric stresses are updated
elastically, as described for model 1, but repeated here for the sake of clarity:
𝜎𝑖𝑗∗ = 𝜎𝑖𝑗𝑛 + 𝐶𝑖𝑗𝑘𝑙 Δ𝜀𝑘𝑙 , (22.3.11)
where
𝜎𝑖𝑗∗ is the trial stress tensor,
𝜎𝑖𝑗𝑛 is the stress tensor from the previous time step,
C𝑖𝑗𝑘𝑙 is the elastic tangent modulus matrix,

20-38 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

Δ𝜀𝑘𝑙 is the incremental strain tensor.


and, if the yield function is satisfied, nothing else is done. If, however, the yield function is
violated, an increment in plastic strain is computed, the stresses are scaled back to the yield
surface, and the yield surface center is updated.

Let s∗𝑖𝑗 represent the trial elastic deviatoric stress state at 𝑛 + 1


1 ∗
s∗𝑖𝑗 = σ𝑖𝑗∗ − σ𝑘𝑘 , (22.3.12)
3
and
ξ𝑖𝑗∗ = s∗𝑖𝑗 − α𝑖𝑗 . (22.3.13)

Define the yield function,


3 ≤0 for elastic or neutral loading
𝜙 = 𝜉𝑖𝑗∗ 𝜉𝑖𝑗∗ − 𝜎𝑦2 = 𝛬2 − 𝜎𝑦2 { , (22.3.14)
2 >0 for plastic harding
For plastic hardening then
p𝑛+1 p𝑛 𝛬 − 𝜎𝑦 p𝑛 p
𝜀eff = 𝜀eff + = 𝜀eff + Δ𝜀eff , (22.3.15)
3𝐺 + 𝐸p
scale back the stress deviators
p
3𝐺Δ𝜀eff ∗
𝜎𝑖𝑗𝑛+1 = 𝜎𝑖𝑗∗ − 𝜉𝑖𝑗 , (22.3.16)
𝛬
and update the center:
p
(1 − 𝛽)𝐸p Δ𝜀eff
𝛼𝑛𝑖𝑗 + 1 = 𝛼𝑛𝑖𝑗 + 𝜉𝑖𝑗∗ . (22.3.17)
𝛬

Plane Stress Plasticity


The plane stress plasticity options apply to beams, shells, and thick shells. Since the
stresses and strain increments are transformed to the lamina coordinate system for the
constitutive evaluation, the stress and strain tensors are in the local coordinate system.

The application of the Jaumann rate to update the stress tensor allows for the
possibility that the normal stress, 𝜎33 , will not be zero. The first step in updating the stress
tensor is to compute a trial plane stress update assuming that the incremental strains are
elastic. In the above, the normal strain increment Δ𝜀33 is replaced by the elastic strain
increment
𝜎33 + 𝜆(Δ𝜀11 + Δ𝜀22 )
Δ𝜀33 = − , (22.3.18)
𝜆 + 2𝜇
where 𝜆 and 𝜇 are Lamé’s constants.

LS-DYNA DEV 10/27/16 (r:8004) 20-39 (Material Models)


Material Models LS-DYNA Theory Manual

When the trial stress is within the yield surface, the strain increment is elastic and
the stress update is completed. Otherwise, for the plastic plane stress case, secant iteration
is used to solve Equation (22.3.16)for the normal strain increment (Δ𝜀33 ) required to
produce a zero normal stress:
p𝑖
𝑖 ∗
3𝐺Δ𝜀eff 𝜉33 (22.3.19)
𝜎33 = 𝜎33 − ,
𝛬
Here, the superscript 𝑖 indicates the iteration number.

The secant iteration formula for Δε33 (the superscript p is dropped for clarity) is
Δ𝜀𝑖33 − Δ𝜀𝑖−1
33
Δ𝜀𝑖+1 𝑖−1
33 = Δ𝜀33 − 𝑖 𝑖−1
𝑖−1
𝜎33 , (22.3.20)
𝜎33 − 𝜎33
where the two starting values are obtained from the initial elastic estimate and by
assuming a purely plastic increment, i.e.,
Δ𝜀133 = −(Δ𝜀11 − Δ𝜀22 ). (22.3.21)
These starting values should bound the actual values of the normal strain increment.

The iteration procedure uses the updated normal stain increment to update first the
deviatoric stress and then the other quantities needed to compute the next estimate of the
𝑖
normal stress in Equation (22.3.19). The iterations proceed until the normal stress 𝜎33 is
sufficiently small. The convergence criterion requires convergence of the normal strains:
∣Δ𝜀𝑖33 − Δ𝜀𝑖−1
33 ∣
< 10−4 . (22.3.22)
∣Δ𝜀𝑖+1
33 ∣

After convergence, the stress update is completed using the relationships given in
Equations (22.3.16) and (22.3.17)

20-40 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.4 Material Model 4: Thermo-Elastic-Plastic


This model was adapted from the NIKE2D [Hallquist 1979] code. A more complete
description of its formulation is given in the NIKE2D user’s manual.

Letting 𝑇 represent the temperature, we compute the elastic co-rotational stress rate
as
𝜎𝑖𝑗∇ = 𝐶𝑖𝑗𝑘𝑙 (𝜀̇𝑘𝑙 − ε̇T𝑘𝑙 ) + 𝜃𝑖𝑗̇ 𝑑𝑇, (22.4.1)
where
𝑑𝐶𝑖𝑗𝑘𝑙
𝜃𝑖𝑗̇ = 𝐶−1 𝜎̇ , (22.4.2)
𝑑𝑇 𝑘𝑙𝑚𝑛 𝑚𝑛
and 𝐶𝑖𝑗𝑘𝑙 is the temperature dependent elastic constitutive matrix:

⎡1 − 𝜐 𝜐 𝜐 0 0 0 ⎤
⎢ ⎥
⎢ ⎥
⎢ 𝜐 1−𝜐 𝜐 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 𝜐 𝜐 1−𝜐 0 0 0 ⎥
𝐸 ⎢ ⎥
𝐶𝑖𝑗𝑘𝑙 = ⎢ ⎥, (22.4.3)
(1 + 𝜐)(1 − 2𝜐) ⎢ 1 − 2𝜐
⎢ 0 0 0 0 0 ⎥⎥
⎢ 2 ⎥
⎢ 1 − 2𝜐 ⎥
⎢ 0 0 0 0 0 ⎥
⎢ 2 ⎥
⎢ 1 − 2𝜐⎥
⎣ 0 0 0 0 0
2 ⎦

where 𝜐 is Poisson’s ratio. The thermal strain rate can be written in terms of the coefficient
of thermal expansion 𝛼 as:
𝜀̇T𝑖𝑗 = 𝛼𝑇̇ 𝛿𝑖𝑗 , (22.4.4)

When treating plasticity, we use a procedure analogous to that for material 3. We


update the stresses elastically and check to see if we violate the isotropic yield function

1 𝜎𝑦 (𝑇)2
𝜙 = 𝑠𝑖𝑗 𝑠𝑖𝑗 − , (22.4.5)
2 3
where
p
𝜎𝑦 (𝑇) = 𝜎𝑜 (𝑇) + 𝐸p (𝑇)𝜀eff . (22.4.6)

LS-DYNA DEV 10/27/16 (r:8004) 20-41 (Material Models)


Material Models LS-DYNA Theory Manual

The initial yield, 𝜎o , and plastic hardening modulus, 𝐸p , are temperature dependent.
If the behavior is elastic we do nothing; otherwise, we scale back the stress deviators by the
factor 𝑓s :
𝑠𝑛+1
𝑖𝑗 = 𝑓s 𝑠∗𝑖𝑗 , (22.4.7)
where
𝜎𝑦
𝑓s = 1
,
⁄2 (22.4.8)
(3 𝑠∗𝑖𝑗 𝑠∗𝑖𝑗 )
2
and update the plastic strain by the increment
1
⁄2
(1 − 𝑓s )(3 𝑠∗𝑖𝑗 𝑠∗𝑖𝑗 )
p 2 (22.4.9)
Δ𝜀eff = .
𝐺 + 3𝐸p

20-42 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.5 Material Model 5: Soil and Crushable Foam


This model, due to Krieg [1972], provides a simple model for foam and soils whose
material properties are not well characterized. We believe the other foam models in LS-
DYNA are superior in their performance and are recommended over this model which
simulates the crushing through the volumetric deformations. If the yield stress is too low,
this foam model gives nearly fluid like behavior.

A pressure-dependent flow rule governs the deviatoric behavior:


1
𝜙s = s𝑖𝑗 s𝑖𝑗 − (𝑎0 + 𝑎1 𝑝 + 𝑎2 𝑝2 ), (22.5.1)
2
where 𝑎0 , 𝑎1 , and 𝑎2 are user-defined constants. Volumetric yielding is determined by a
tabulated curve of pressure versus volumetric strain. Elastic unloading from this curve is
assumed to a tensile cutoff as illustrated in Figure 22.5.1.

Implementation of this model is straightforward. One history variable, the


maximum volumetric strain in compression, is stored. If the new compressive volumetric
strain exceeds the stored value, loading is indicated. When the yield condition is violated,
the updated trial stresses, 𝑠∗𝑖𝑗 , are scaled back using a simple radial return algorithm:
Pressure

Loading and unloading (along the


grey arows) follows the input curve
when the volumetric crushing option
is off (VCR = 1.0)

tension compression
⎛V⎛
Volumetric Strain, ln ⎜ ⎜
⎝V0⎝
The bulk unloading modulus is used
Pressure Cutoff Value
if the volumetric crushing option is on
(VCR = 0). In thiscase the aterial's response
follows the black arrows.
Figure 22.5.1. Volumetric strain versus pressure curve for soil and crushable
foam model.

LS-DYNA DEV 10/27/16 (r:8004) 20-43 (Material Models)


Material Models LS-DYNA Theory Manual

1⁄
2

⎜ 𝑎0 + 𝑎1 𝑝 + a2 𝑝2 ⎞

s𝑛+1
𝑖𝑗 =⎜
⎜ ⎟
⎟ s∗𝑖𝑗 . (22.5.2)
1s s
⎝ 2 𝑖𝑗 𝑖𝑗 ⎠

If the hydrostatic tension exceeds the cutoff value, the pressure is set to the cutoff
value and the deviatoric stress tensor is zeroed.

20-44 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.6 Material Model 6: Viscoelastic


In this model, linear viscoelasticity is assumed for the deviatoric stress tensor
[Herrmann and Peterson 1968]:
𝑡 ∂𝜀′𝑖𝑗 (𝜏)
𝑠𝑖𝑗 = 2 ∫ 𝜙(𝑡 − 𝜏) 𝑑𝜏 , (22.6.1)
∂𝜏
0

where
𝜙(𝑡) = G∞ + (G0 − G∞ )𝑒−𝛽𝑡 , (22.6.2)
is the shear relaxation modulus. A recursion formula is used to compute the new value of
the hereditary integral at time 𝑡𝑛+1 from its value at time 𝑡𝑛 . Elastic bulk behavior is
assumed:
𝑝 = 𝐾ln𝑉, (22.6.3)
where pressure is integrated incrementally.

LS-DYNA DEV 10/27/16 (r:8004) 20-45 (Material Models)


Material Models LS-DYNA Theory Manual

22.7 Material Model 7: Continuum Rubber


The hyperelastic continuum rubber model was studied by Blatz and Ko [1962]. In
this model, the second Piola-Kirchhoff stress is given by
− 1
𝑆𝑖𝑗 = 𝐺 (𝑉 −1 𝐶𝑖𝑗 − 𝑉 1−2𝜐 𝛿 ),
𝑖𝑗 (22.7.1)

where 𝐺 is the shear modulus, 𝑉 is the relative volume, 𝜐 is Poisson’s ratio, and 𝐶𝑖𝑗 is the
right Cauchy-Green strain:
𝜕𝑥𝑘 𝜕𝑥𝑘
C𝑖𝑗 = , (22.7.2)
𝜕𝑋𝑖 𝜕𝑋𝑗
after determining 𝑆𝑖𝑗 , it is transformed into the Cauchy stress tensor, 𝜎𝑖𝑗 :
𝜌 𝜕𝑥𝑖 𝜕𝑥𝑗
𝜎𝑖𝑗 = 𝑆 , (22.7.3)
𝜌0 𝜕𝑋𝑘 𝜕𝑋𝑙 𝑘𝑙
where 𝜌0 and 𝜌 are the initial and current density, respectively. The default value of υ is
0.463.

20-46 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.8 Material Model 8: Explosive Burn


Burn fractions, which multiply the equations of states for high explosives, control
the release of chemical energy for simulating detonations. In the initialization phase, a
lighting time 𝑡1 is computed for each element by dividing the distance from the detonation
point to the center of the element by the detonation velocity 𝐷. If multiple detonation
points are defined, the closest point determines 𝑡1 . The burn fraction 𝐹 is taken as the
maximum
𝐹 = max(𝐹1 , 𝐹2 ), (22.8.1)
where
⎧ 2 (𝑡 − 𝑡𝑙 )𝐷
{
{ 𝑡 > 𝑡l
{ 𝑣
{ 3 ( e )
{ 𝐴emax
𝐹1 = ⎨ (22.8.2)
{
{ 0 𝑡 ≤ 𝑡l
{
{
{

1−𝑉
𝐹2 = , (22.8.3)
1 − 𝑉CJ
where 𝑉CJ is the Chapman-Jouguet relative volume and 𝑡 is current time. If 𝐹 exceeds 1, it
is reset to 1. This calculation of the burn fraction usually requires several time steps for 𝐹 to
reach unity, thereby spreading the burn front over several elements. After reaching unity,
𝐹 is held constant. This burn fraction calculation is based on work by Wilkins [1964] and is
also discussed by Giroux [1973].

As an option, the high explosive material can behave as an elastic perfectly-plastic


solid prior to detonation. In this case we update the stress tensor, to an elastic trial stress,
𝑠∗𝑛+1
𝑖𝑗 ,

𝑠∗𝑛+1
𝑖𝑗 = 𝑠𝑛𝑖𝑗 + 𝑠𝑖𝑝 𝛺𝑝𝑗 + 𝑠𝑗𝑝 𝛺𝑝𝑖 + 2𝐺𝜀̇′𝑖𝑗 𝑑𝑡, (22.8.4)
where 𝐺 is the shear modulus, and 𝜀̇′𝑖𝑗 is the deviatoric strain rate. The von Mises yield
condition is given by:
𝜎𝑦2
𝜙 = 𝐽2 − , (22.8.5)
3
where the second stress invariant, 𝐽2 , is defined in terms of the deviatoric stress
components as
1
𝐽2 = 𝑠𝑖𝑗 𝑠𝑖𝑗 , (22.8.6)
2

LS-DYNA DEV 10/27/16 (r:8004) 20-47 (Material Models)


Material Models LS-DYNA Theory Manual

and the yield stress is 𝜎𝑦 . If yielding has occurred, i.e., 𝜙 > 0, the deviatoric trial stress is
scaled to obtain the final deviatoric stress at time 𝑛 + 1:
𝜎𝑦 ∗𝑛+1
𝑠𝑛+1
𝑖𝑗 = 𝑠𝑖𝑗 , (22.8.7)
√3𝐽2
If 𝜙 ≤ 0, then
𝑠𝑛+1
𝑖𝑗 = 𝑠∗𝑛+1
𝑖𝑗 . (22.8.8)
Before detonation pressure is given by the expression
1
𝑝𝑛+1 = 𝐾 ( − 1). (22.8.9)
𝑉 𝑛+1
where K is the bulk modulus. Once the explosive material detonates:
𝑠𝑛+1
𝑖𝑗 = 0. (22.8.10)
and the material behaves like a gas.

The shadow burn option should be active when computing the lighting time if there
exist elements within the mesh for which there is no direct line of sight from the detonation
points. The shadow burn option is activated in the control section. The lighting time is
based on the shortest distance through the explosive material. If inert obstacles exist within
the explosive material, the lighting time will account for the extra time required for the
detonation wave to travel around the obstacles. The lighting times also automatically
accounts for variations in the detonation velocity if different explosives are used. No
additional input is required for the shadow option but care must be taken when setting up
the input. This option works for two and three-dimensional solid elements. It is
recommended that for best results:

1. Keep the explosive mesh as uniform as possible with elements of roughly the same
dimensions.

2. Inert obstacle such as wave shapers within the explosive must be somewhat larger
than the characteristic element dimension for the automatic tracking to function
properly. Generally, a factor of two should suffice. The characteristic element
dimension is found by checking all explosive elements for the largest diagonal

3. The detonation points should be either within or on the boundary of the explosive.
Offset points may fail to initiate the explosive.

4. Check the computed lighting times in the post processor LS-PrePost. The lighting
times may be displayed at time = 0, state 1, by plotting component 7 (a component
normally reserved for plastic strain) for the explosive material. The lighting times
are stored as negative numbers. The negative lighting time is replaced by the burn
fraction when the element ignites.

20-48 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

5. Line detonations may be approximated by using a sufficient number of


detonation points to define the line. Too many detonation points may result in
significant initialization cost.

LS-DYNA DEV 10/27/16 (r:8004) 20-49 (Material Models)


Material Models LS-DYNA Theory Manual

22.9 Material Model 9: Null Material


For solid elements equations of state can be called through this model to avoid
deviatoric stress calculations. A pressure cutoff may be specified to set a lower bound on
the pressure. This model has been very useful when combined with the reactive high
explosive model where material strength is often neglected. The null material should not
be used to delete solid elements.

An optional viscous stress of the form


𝜎𝑖𝑗 = 𝜇𝜀̇′𝑖𝑗 , (22.9.1)
is computed for nonzero 𝜇 where 𝜀̇′𝑖𝑗 is the deviatoric strain rate.

Sometimes it is advantageous to model contact surfaces via shell elements which are
not part of the structure, but are necessary to define areas of contact within nodal rigid
bodies or between nodal rigid bodies. Beams and shells that use this material type are
completely bypassed in the element processing. The Young’s modulus and Poisson’s ratio
are used only for setting the contact interface stiffnesses, and it is recommended that
reasonable values be input.

20-50 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.10 Material Model 10: Elastic-Plastic-Hydrodynamic


For completeness we give the entire derivation of this constitutive model based on
radial return plasticity.

The pressure, 𝑝, deviatoric strain rate, 𝜀̇′𝑖𝑗 , deviatoric stress rate, 𝑠𝑖𝑗̇ , volumetric strain
rate, and 𝜀̇v , are defined in Equation (1.1.1):
1 1
𝑝 = − 𝜎𝑖𝑗 𝛿𝑖𝑗 𝜀̇′𝑖𝑗 = 𝜀̇𝑖𝑗 − 𝜀̇v
3 3
𝑠𝑖𝑗 = 𝜎𝑖𝑗 + 𝑝𝛿𝑖𝑗 𝜀̇v = 𝜀̇𝑖𝑗 𝛿𝑖𝑗 (22.10.1)

𝑠∇ ′ ′
𝑖𝑗 = 2𝜇𝜀̇𝑖𝑗 = 2𝐺𝜀̇𝑖𝑗 .

The Jaumann rate of the deviatoric stress, 𝑠∇


𝑖𝑗 , is given by:

𝑠∇
𝑖𝑗 = 𝑠𝑖𝑗̇ − 𝑠𝑖𝑝 𝛺𝑝𝑗 − 𝑠𝑗𝑝 𝛺𝑝𝑖 . (22.10.2)

First we update s𝑛𝑖𝑗 to s𝑛+1


𝑖𝑗 elastically
∗ 𝑛+1
𝑠𝑖𝑗 = 𝑠𝑛𝑖𝑗 + 𝑠𝑖𝑝 𝛺𝑝𝑗 + 𝑠𝑗𝑝 𝛺𝑝𝑖 + 2𝐺𝜀̇′𝑖𝑗 𝑑𝑡 = 𝑠⏟
𝑛 ′
𝑖𝑗 + 𝑅𝑖𝑗 + 2𝐺𝜀̇
⏟ 𝑖𝑗 𝑑𝑡,
𝑛
(22.10.3)
𝑠𝑅
𝑖𝑗
2𝐺Δ𝜀′𝑖𝑗

where the left superscript, *, denotes a trial stress value. The effective trial stress is defined
by
⁄2 1
3 (22.10.4)
𝑠 = ( ∗𝑠𝑛+1
∗ ∗ 𝑛+1
𝑠 ) ,
2 𝑖𝑗 𝑖𝑗

and if 𝑠∗ exceeds yield stress 𝜎y , the Von Mises flow rule:


2
1 𝜎y
𝜙 = 𝑠𝑖𝑗 𝑠𝑖𝑗 − ≤ 0, (22.10.5)
2 3
is violated and we scale the trial stresses back to the yield surface, i.e., a radial return
𝜎𝑦 ∗ 𝑛+1
𝑠𝑛+1
𝑖𝑗 = 𝑠 = 𝑚 ∗𝑠𝑛+1
𝑖𝑗 . (22.10.6)
𝑠∗ 𝑖𝑗

The plastic strain increment can be found by subtracting the deviatoric part of the
1 𝑛
strain increment that is elastic, 2𝐺 (𝑠𝑛+1
𝑖𝑗 − 𝑠R ′
𝑖𝑗 ), from the total deviatoric increment, Δε𝑖𝑗 , i.e.,

p 1 𝑛
Δ𝜀𝑖𝑗 = Δ𝜀′𝑖𝑗 − (𝑠𝑛+1
𝑖𝑗 − 𝑠𝑖𝑗R ). (22.10.7)
2𝐺

LS-DYNA DEV 10/27/16 (r:8004) 20-51 (Material Models)


Material Models LS-DYNA Theory Manual

Recalling that,
𝑛
( ∗𝑠𝑛+1
𝑖𝑗 − 𝑠R
𝑖𝑗 )

Δ𝜀𝑖𝑗 = , (22.10.8)
2𝐺
and substituting Equation (22.10.8) into (22.10.7) we obtain,

p ( ∗𝑠𝑛+1
𝑖𝑗 − 𝑠𝑛+1
𝑖𝑗 )
Δ𝜀𝑖𝑗 = . (22.10.9)
2𝐺
Substituting Equation (22.10.6)
𝑠𝑛+1
𝑖𝑗 = 𝑚 ∗𝑠𝑛+1
𝑖𝑗 , (22.10.10)
into Equation (22.10.9) gives,
p 1 − 𝑚 ∗ 𝑛+1 1 − 𝑚 𝑛+1
Δ𝜀𝑖𝑗 = ( ) 𝑠𝑖𝑗 = 𝑠𝑖𝑗 = 𝑑λ𝑠𝑛+1
𝑖𝑗 . (22.10.11)
2𝐺 2𝐺𝑚
By definition an increment in effective plastic strain is
1
2 p p ⁄2
p (22.10.12)
Δ𝜀 = ( Δ𝜀𝑖𝑗 Δ𝜀𝑖𝑗 ) .
3
Squaring both sides of Equation (22.10.11) leads to:

p p 1 − 𝑚 2 ∗ 𝑛+1 ∗ 𝑛+1
Δ𝜀𝑖𝑗 Δ𝜀𝑖𝑗 = ( ) 𝑠𝑖𝑗 𝑠𝑖𝑗 (22.10.13)
2𝐺
or from Equations (22.10.4) and (22.10.12):
3 p2 1 − 𝑚 2 2 ∗2
Δ𝜀 = ( ) 𝑠 (22.10.14)
2 2𝐺 3
Hence,

p1 − 𝑚 ∗ 𝑠 − 𝜎y
Δ𝜀 = 𝑠 = (22.10.15)
3𝐺 3𝐺
where we have substituted for m from Equation (22.10.6)
𝜎y
𝑚= ∗ (22.10.16)
𝑠

If isotropic hardening is assumed then:


𝜎y𝑛+1 = 𝜎y𝑛 + 𝐸p Δ𝜀p (22.10.17)
and from Equation (22.10.15)
(𝑠∗ − 𝜎y𝑛+1 ) (𝑠∗ − 𝜎y𝑛 − 𝐸p Δ𝜀p )
Δ𝜀 = p
= . (22.10.18)
3𝐺 3𝐺

Thus,

20-52 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

(3𝐺 + 𝐸p )Δ𝜀p = (𝑠∗ − 𝜎y𝑛 ), (22.10.19)


and solving for the incremental plastic strain gives

p
(𝑠∗ − 𝜎y𝑛 )
Δ𝜀 = . (22.10.20)
(3𝐺 + 𝐸p )

The algorithm for plastic loading can now be outlined in five simple stress. If the
effective trial stress exceeds the yield stress then

1. Solve for the plastic strain increment:

p
(𝑠∗ − σy𝑛 )
Δ𝜀 = . (22.10.21)
(3𝐺 + 𝐸p )

2. Update the plastic strain:


𝑛+1 𝑛
𝜀p = 𝜀p + Δ𝜀p . (22.10.22)

3. Update the yield stress:


𝜎y𝑛+1 = 𝜎y𝑛 + 𝐸p Δ𝜀p . (22.10.23)

4. Compute the scale factor using the yield strength at time 𝑛 + 1:


𝜎y𝑛+1
𝑚= . (22.10.24)
𝑠∗

5. Radial return the deviatoric stresses to the yield surface:


𝑠𝑛+1
𝑖𝑗 = 𝑚 ∗𝑠𝑛+1
𝑖𝑗 . (22.10.25)

LS-DYNA DEV 10/27/16 (r:8004) 20-53 (Material Models)


Material Models LS-DYNA Theory Manual

22.11 Material Model 11: Elastic-Plastic With Thermal


Softening
Steinberg and Guinan [1978] developed this model for treating plasticity at high
strain rates (105 s-1) where enhancement of the yield strength due to strain rate effects is
saturated out.

Both the shear modulus 𝐺 and yield strength 𝜎y increase with pressure but decrease
with temperature. As a melt temperature is reached, these quantities approach zero. We
define the shear modulus before the material melts as
𝑓𝐸
1⁄ 𝐸 − 𝐸c −
(22.11.1)
𝐺 = 𝐺0 [1 + 𝑏𝑝𝑉 3 − ℎ( − 300)] 𝑒 𝐸m −𝐸 ,
3𝑅′
where 𝐺0 , 𝑏, ℎ, and 𝑓 are input parameters, 𝐸c is the cold compression energy:
𝑥
900𝑅′ exp(𝑎𝑥)
𝐸c (𝑋) = ∫ 𝑝𝑑𝑥 − , (22.11.2)
2(𝛾 −𝑎−1)
0 (1 − 𝑋) 𝑜 2
where,
𝑋 = 1 − 𝑉, (22.11.3)
and 𝐸m is the melting energy:
𝐸m (𝑋) = 𝐸c (𝑋) + 3𝑅′𝑇m (𝑋), (22.11.4)
which is a function of the melting temperature 𝑇m (𝑋):
𝑇mo exp(2𝑎𝑋)
𝑇m (𝑋) = , (22.11.5)
2(𝛾 −𝑎−1)
(1 − 𝑋) 𝑜 3
and the melting temperature 𝑇mo at 𝜌 = 𝜌0 . The constants 𝛾0 and a are input parameters.
In the above equation, 𝑅′ is defined by
𝑅𝜌0
, 𝑅′ = (22.11.6)
𝐴
where 𝑅 is the gas constant and A is the atomic weight. The yield strength 𝜎y is given by:
1 𝑓𝐸
𝐸 − 𝐸c −
(22.11.7)
𝜎y = 𝜎0′ [1 + 𝑏′𝑝𝑉 3 − ℎ( − 300)] 𝑒 𝐸m −𝐸 .
3𝑅′

If 𝐸m exceeds 𝐸𝑖 . Here, 𝜎0′ is given by:


𝜎0′ = 𝜎0 [1 + 𝛽(𝛾𝑖 + 𝜀̅𝑝 )]𝑛 . (22.11.8)

20-54 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

where 𝛾1 is the initial plastic strain, and 𝑏′ and 𝜎0′ are input parameters. Where 𝜎0′ exceeds
𝜎max , the maximum permitted yield strength, 𝜎0′ is set to equal to 𝜎max . After the material
melts, 𝜎y and 𝐺 are set to zero.

LS-DYNA fits the cold compression energy to a ten-term polynomial expansion:


9
𝐸c = ∑ 𝐸𝐶𝑖 𝜂𝑖 , (22.11.9)
𝑖=0
𝜌
where 𝐸𝐶𝑖 is the ith coefficient and 𝜂 =𝜌0 .
The least squares method is used to perform the
fit [Kreyszig 1972]. The ten coefficients may also be specified in the input.

Once the yield strength and shear modulus are known, the numerical treatment is
similar to that for material model 10.

LS-DYNA DEV 10/27/16 (r:8004) 20-55 (Material Models)


Material Models LS-DYNA Theory Manual

22.12 Material Model 12: Isotropic Elastic-Plastic


The von Mises yield condition is given by:
𝜎y2
𝜙 = 𝐽2 − , (22.12.1)
3
where the second stress invariant, 𝐽2 , is defined in terms of the deviatoric stress
components as
1
𝐽2 = 𝑠𝑖𝑗 𝑠𝑖𝑗 , (22.12.2)
2
p
and the yield stress, 𝜎y , is a function of the effective plastic strain, 𝜀eff , and the plastic
hardening modulus, 𝐸p :
p
𝜎y = 𝜎0 + 𝐸p 𝜀eff . (22.12.3)

The effective plastic strain is defined as


𝑡
p p
𝜀eff = ∫ 𝑑𝜀eff , (22.12.4)
0

p p p
where 𝑑𝜀eff = √23 𝑑𝜀𝑖𝑗 𝑑𝜀𝑖𝑗 , and the plastic tangent modulus is defined in terms of the input
tangent modulus, 𝐸t , as
𝐸𝐸t
𝐸p = . (22.12.5)
𝐸 − 𝐸t

Pressure is given by the expression


1
𝑝𝑛+1 = 𝐾 ( − 1), (22.12.6)
𝑉 𝑛+1
where 𝐾 is the bulk modulus. This is perhaps the most cost effective plasticity model.
p
Only one history variable, 𝜀eff , is stored with this model.

This model is not recommended for shell elements. In the plane stress implementa-
tion, a one-step radial return approach is used to scale the Cauchy stress tensor to if the
state of stress exceeds the yield surface. This approach to plasticity leads to inaccurate shell
thickness updates and stresses after yielding. This is the only model in LS-DYNA for plane
stress that does not default to an iterative approach.

20-56 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.13 Material Model 13: Isotropic Elastic-Plastic with


Failure
This highly simplistic failure model is occasionally useful. Material model 12 is
called to update the stress tensor. Failure is initially assumed to occur if either
𝑝𝑛+1 < 𝑝min , (22.13.1)
or
p p
𝜀eff > 𝜀max , (22.13.2)
p
where 𝑝min and 𝜀max are user-defined parameters. Once failure has occurred, pressure may
never be negative and the deviatoric components are set to zero:
𝑠𝑖𝑗 = 0 (22.13.3)
for all time. The failed element can only carry loads in compression.

LS-DYNA DEV 10/27/16 (r:8004) 20-57 (Material Models)


Material Models LS-DYNA Theory Manual

22.14 Material Model 14: Soil and Crushable Foam With


Failure
This material model provides the same stress update as model 5. However, if pressure ever
reaches its cutoff value, failure occurs and pressure can never again go negative. In
material model 5, the pressure is limited to its cutoff value in tension.

20-58 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.15 Material Model 15: Johnson and Cook Plasticity Model


Johnson and Cook express the flow stress as
𝑛
𝜎𝑦 = (𝐴 + 𝐵𝜀̅p )(1 + 𝑐ln𝜀̇∗ )(1 − 𝑇 ∗𝑚 ), (22.15.1)
where 𝐴, 𝐵, 𝐶, 𝑛, and 𝑚 are user defined input constants, and:

𝜀̅𝑝 = effective plastic strain

∗ 𝜀̅𝑝̇ 1
𝜀̇ = = effective plastic strain rate for 𝜀̇0 , in units of
𝜀̇0 [time]
𝑇 − 𝑇room
𝑇∗ =
𝑇melt − 𝑇room
Constants for a variety of materials are provided in Johnson and Cook [1983].

Due to the nonlinearity in the dependence of flow stress on plastic strain, an


accurate value of the flow stress requires iteration for the increment in plastic strain.
However, by using a Taylor series expansion with linearization about the current time, we
can solve for 𝜎𝑦 with sufficient accuracy to avoid iteration.
The strain at fracture is given by
𝜀f = [𝐷1 + 𝐷2 exp(𝐷3 𝜎 ∗ )][1 + 𝐷4 ln𝜀∗ ][1 + 𝐷5 𝑇 ∗ ], (22.15.2)
where 𝐷𝑖 , 𝑖 = 1, . . . ,5 are input constants and 𝜎 ∗ is the ratio of pressure divided by effective
stress:
𝑝
𝜎∗ = . (22.15.3)
𝜎eff

Fracture occurs when the damage parameter


Δ𝜀̅p
𝐷=∑ (22.15.4)
𝜀f
reaches the value 1.

A choice of three spall models is offered to represent material splitting, cracking,


and failure under tensile loads. The pressure limit model limits the minimum hydrostatic
pressure to the specified value, 𝑝 ≥ 𝑝min . If pressures more tensile than this limit are
calculated, the pressure is reset to 𝑝min . This option is not strictly a spall model since the
deviatoric stresses are unaffected by the pressure reaching the tensile cutoff and the
pressure cutoff value 𝑝min remains unchanged throughout the analysis. The maximum
principal stress spall model detects spall if the maximum principal stress, 𝜎max , exceeds the
limiting value 𝜎p . Once spall is detected with this model, the deviatoric stresses are reset to

LS-DYNA DEV 10/27/16 (r:8004) 20-59 (Material Models)


Material Models LS-DYNA Theory Manual

zero and no hydrostatic tension is permitted. If tensile pressures are calculated, they are
reset to 0 in the spalled material. Thus, the spalled material behaves as rubble. The
hydrostatic tension spall model detects spall if the pressure becomes more tensile than the
specified limit, 𝑝min . Once spall is detected, the deviatoric stresses are set to zero and the
pressure is required to be compressive. If hydrostatic tension is calculated then the
pressure is reset to 0 for that element.

In addition to the above failure criterion, this material model also supports a shell
element deletion criterion based on the maximum stable time step size for the element,
Δ𝑡max . Generally, Δ𝑡max goes down as the element becomes more distorted. To assure
stability of time integration, the global LS-DYNA time step is the minimum of the Δ𝑡max
values calculated for all elements in the model. Using this option allows the selective
deletion of elements whose time step Δ𝑡max has fallen below the specified minimum time
step, Δ𝑡crit . Elements which are severely distorted often indicate that material has failed
and supports little load, but these same elements may have very small time steps and
therefore control the cost of the analysis. This option allows these highly distorted
elements to be deleted from the calculation, and, therefore, the analysis can proceed at a
larger time step, and, thus, at a reduced cost. Deleted elements do not carry any load, and
are deleted from all applicable slide surface definitions. Clearly, this option must be
judiciously used to obtain accurate results at a minimum cost.

Material type 15 is applicable to the high rate deformation of many materials


including most metals. Unlike the Steinberg-Guinan model, the Johnson-Cook model
remains valid down to lower strain rates and even into the quasistatic regime. Typical
applications include explosive metal forming, ballistic penetration, and impact.

20-60 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.16 Material Model 16: Pseudo Tensor


This model can be used in two major modes - a simple tabular pressure-dependent
yield surface, and a potentially complex model featuring two yield versus pressure
functions with the means of migrating from one curve to the other. For both modes, load
curve N1 is taken to be a strain rate multiplier for the yield strength. Note that this model
must be used with equation-of-state type 8 or 9.

Response Mode I. Tabulated Yield Stress Versus Pressure

This model is well suited for implementing standard geologic models like the Mohr-
Coulomb yield surface with a Tresca limit, as shown in Figure 22.16.1. Examples of
converting conventional triaxial compression data to this type of model are found in (Desai
and Siriwardane, 1984). Note that under conventional triaxial compression conditions, the
LS-DYNA input corresponds to an ordinate of 𝜎1 − 𝜎3 rather than the more widely used
𝜎1 −𝜎3
2
, where 𝜎1 is the maximum principal stress and 𝜎3 is the minimum principal stress.

This material combined with equation-of-state type 9 (saturated) has been used very
successfully to model ground shocks and soil-structure interactions at pressures up to
100kbar.

To invoke Mode I of this model, set 𝑎0 , 𝑎1 , 𝑎2 , 𝑎0f , and 𝑎1f to zero, The tabulated
values of pressure should then be specified on cards 4 and 5, and the corresponding values
Mohr-Coulomb

Tresca

Friction Angle

Cohesion

Figure 22.16.1. Mohr-Coulomb surface with a Tresca limit.


LS-DYNA DEV 10/27/16 (r:8004) 20-61 (Material Models)
Material Models LS-DYNA Theory Manual

of yield stress should be specified on cards 6 and 7. The parameters relating to


reinforcement properties, initial yield stress, and tangent modulus are not used in this
response mode, and should be set to zero.

Simple tensile failure


Note that a1f is reset internally to 1/3 even though it is input as zero; this defines a
material failure curve of slope 3𝑝, where p denotes pressure (positive in compression). In
this case the yield strength is taken from the tabulated yield vs. pressure curve until the
maximum principal stress (𝜎1 ) in the element exceeds the tensile cut-off (𝜎cut ). For every
time step that 𝜎1 > 𝜎cut the yield strength is scaled back by a fraction of the distance
between the two curves until after 20 time steps the yield strength is defined by the failure
curve. The only way to inhibit this feature is to set σcut arbitrarily large.

Response Mode II. Two-Curve Model with Damage and Failure


This approach uses two yield versus pressure curves of the form
𝑝
𝜎y = 𝑎0 + . (22.16.1)
𝑎1 + 𝑎2 𝑝

The upper curve is best described as the maximum yield strength curve and the
lower curve is the material failure curve. There are a variety of ways of moving between
the two curves and each is discussed below.

MODE II.A: Simple tensile failure


Define 𝑎0 , 𝑎1 , 𝑎2 , 𝑎0f and 𝑎1f , set 𝑏1 to zero, and leave cards 4 through 7 blank. In this
case the yield strength is taken from the maximum yield curve until the maximum
principal stress (𝜎1 ) in the element exceeds the tensile cut-off (𝜎cut ). For every time step
that 𝜎1 > 𝜎cut the yield strength is scaled back by a fraction of the distance between the two
Yield

Pressure
Figure 22.2. Two-curve concrete model with damage and failure.

curves until after 20 time steps the yield strength is defined by the failure curve.

20-62 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

Mode II.B: Tensile failure plus plastic strain scaling


Define 𝑎0 , 𝑎1 , 𝑎2 , 𝑎0f and 𝑎1f , set 𝑏1 to zero, and user cards 4 through 7 to define a
scale factor, , versus effective plastic strain. LS-DYNA evaluates  at the current effective
plastic strain and then calculated the yield stress as
𝜎yield = 𝜎failed + 𝜂(𝜎max − 𝜎failed ), (22.16.2)
where 𝜎max and 𝜎failed are found as shown in Figure 19.16.2. This yield strength is then
subject to scaling for tensile failure as described above. This type of model allows the
description of a strain hardening or softening material such as concrete.

Mode II.C: Tensile failure plus damage scaling


The change in yield stress as a function of plastic strain arises from the physical
mechanisms such as internal cracking, and the extent of this cracking is affected by the
hydrostatic pressure when the cracking occurs. This mechanism gives rise to the
"confinement" effect on concrete behavior. To account for this phenomenon, a "damage"
function was defined and incorporated. This damage function is given the form:
𝜀p −𝑏1
𝑝
𝜆 = ∫ (1 + ) 𝑑𝜀p . (22.16.3)
𝜎cut
0

Define 𝑎0 , 𝑎1 , 𝑎2 , 𝑎0f and 𝑎1f , and 𝑏1 . Cards 4 through 7 now give 𝜂 as a function of 𝜆
and scale the yield stress as
𝜎yield = 𝜎failed + 𝜂(𝜎max − 𝜎failed ), (22.16.4)
and then apply any tensile failure criteria.

Mode II Concrete Model Options


Material Type 16 Mode II provides the option of automatic internal generation of a
simple "generic" model for concrete. If 𝑎0 is negative, then 𝜎cut is assumed to be the
unconfined concrete compressive strength, 𝑓c′ and −𝑎0 is assumed to be a conversion factor
from LS-DYNA pressure units to psi. (For example, if the model stress units are MPa, 𝑎0
should be set to –145.) In this case the parameter values generated internally are

LS-DYNA DEV 10/27/16 (r:8004) 20-63 (Material Models)


Material Models LS-DYNA Theory Manual

1
2
⎛ 𝑓c′ ⎞3
𝜎cut = 1.7 ⎜ ⎟
⎝ −𝑎 0 ⎠
𝑓c′
𝑎0 =
4
1 (22.16.5)
𝑎1 =
3
1
𝑎2 = ′
3𝑓c
𝑎0f = 0
𝑎1f = 0.385

Note that these 𝑎0f and 𝑎1f defaults will be overwritten by non-zero entries on Card
3. If plastic strain or damage scaling is desired, Cards 5 through 8 and b1 should be
specified in the input. When 𝑎0 is input as a negative quantity, the equation-of-state can be
given as 0 and a trilinear EOS Type 8 model will be automatically generated from the
unconfined compressive strength and Poisson's ratio. The EOS 8 model is a simple
pressure versus volumetric strain model with no internal energy terms, and should give
reasonable results for pressures up to 5kbar (approximately 72,500 psi).

Mixture model
A reinforcement fraction, 𝑓r , can be defined along with properties of the reinforcing
material. The bulk modulus, shear modulus, and yield strength are then calculated from a
simple mixture rule, i.e., for the bulk modulus the rule gives:
𝐾 = (1 − 𝑓r )𝐾m + 𝑓r 𝐾r , (22.16.6)
where 𝐾m and 𝐾r are the bulk moduli for the geologic material and the reinforcing
material, respectively. This feature should be used with caution. It gives an isotropic effect
in the material instead of the true anisotropic material behavior. A reasonable approach
would be to use the mixture elements only where reinforcing material exists and plain
elements elsewhere. When the mixture model is being used, the strain rate multiplier for
the principal material is taken from load curve N1 and the multiplier for the reinforcement
is taken from load curve N2.

20-64 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.17 Material Model 17: Isotropic Elastic-Plastic With


Oriented Cracks
This is an isotropic elastic-plastic material which includes a failure model with an
oriented crack. The von Mises yield condition is given by:
𝜎y2
𝜙 = 𝐽2 − , (22.17.1)
3
where the second stress invariant, 𝐽2 , is defined in terms of the deviatoric stress
components as
1
𝐽2 = 𝑠𝑖𝑗 𝑠𝑖𝑗 , (22.17.2)
2
p
and the yield stress, 𝜎y , is a function of the effective plastic strain, 𝜀eff , and the plastic
hardening modulus, 𝐸p :
p
𝜎y = 𝜎0 + 𝐸p 𝜀eff . (22.17.3)

The effective plastic strain is defined as:


𝑡
p p
𝜀eff = ∫ 𝑑𝜀eff , (22.17.4)
0

p p p
where 𝑑𝜀eff = √23 𝑑𝜀𝑖𝑗 𝑑𝜀𝑖𝑗 , and the plastic tangent modulus is defined in terms of the input
tangent modulus, 𝐸t , as
𝐸𝐸t
𝐸p = . (22.17.5)
𝐸 − 𝐸t

Pressure in this model is found from evaluating an equation of state. A pressure


cutoff can be defined such that the pressure is not allowed to fall below the cutoff value.

The oriented crack fracture model is based on a maximum principal stress criterion.
When the maximum principal stress exceeds the fracture stress, 𝜎f , the element fails on a
plane perpendicular to the direction of the maximum principal stress. The normal stress
and the two shear stresses on that plane are then reduced to zero. This stress reduction is
done according to a delay function that reduces the stresses gradually to zero over a small
number of time steps. This delay function procedure is used to reduce the ringing that may
otherwise be introduced into the system by the sudden fracture.

After a tensile fracture, the element will not support tensile stress on the fracture
plane, but in compression will support both normal and shear stresses. The orientation of
LS-DYNA DEV 10/27/16 (r:8004) 20-65 (Material Models)
Material Models LS-DYNA Theory Manual

this fracture surface is tracked throughout the deformation, and is updated to properly
model finite deformation effects. If the maximum principal stress subsequently exceeds the
fracture stress in another direction, the element fails isotropically. In this case the element
completely loses its ability to support any shear stress or hydrostatic tension, and only
compressive hydrostatic stress states are possible. Thus, once isotropic failure has
occurred, the material behaves like a fluid.

This model is applicable to elastic or elastoplastic materials under significant tensile


or shear loading when fracture is expected. Potential applications include brittle materials
such as ceramics as well as porous materials such as concrete in cases where pressure
hardening effects are not significant.

20-66 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.18 Material Model 18: Power Law Isotropic Plasticity


Elastoplastic behavior with isotropic hardening is provided by this model. The yield
stress, 𝜎y , is a function of plastic strain and obeys the equation:
𝑛
𝜎y = 𝑘𝜀𝑛 = 𝑘(𝜀yp + 𝜀̅p ) , (22.18.1)
where 𝜀yp is the elastic strain to yield and 𝜀̅p is the effective plastic strain (logarithmic).

A parameter, SIGY, in the input governs how the strain to yield is identified. If
SIGY is set to zero, the strain to yield if found by solving for the intersection of the linearly
elastic loading equation with the strain hardening equation:
𝜎 = 𝐸𝜀,
(22.18.2)
𝜎 = 𝑘𝜀𝑛 ,
which gives the elastic strain at yield as:
1
𝐸 𝑛−1 (22.18.3)
𝜀yp =( ) .
𝑘

If SIGY yield is nonzero and greater than 0.02 then:


1
𝜎y 𝑛
𝜀yp = ( ) . (22.18.4)
𝑘

Strain rate is accounted for using the Cowper-Symonds model which scales the yield
stress with the factor
1
𝜀̇ ⁄P (22.18.5)
1+( ) ,
𝐶
where 𝜀̇ is the strain rate. A fully viscoplastic formulation is optional with this model
which incorporates the Cowper-Symonds formulation within the yield surface. An
additional cost is incurred but the improvement allows for dramatic results.

LS-DYNA DEV 10/27/16 (r:8004) 20-67 (Material Models)


Material Models LS-DYNA Theory Manual

22.19 Material Model 19: Strain Rate Dependent Isotropic


Plasticity
In this model, a load curve is used to describe the yield strength 𝜎0 as a function of
effective strain rate 𝜀̅̇ where
1 ⁄2
2 (22.19.1)
𝜀̅̇ = ( 𝜀̇′𝑖𝑗 𝜀̇′𝑖𝑗 ) ,
3
and the prime denotes the deviatoric component. The yield stress is defined as
𝜎y = 𝜎0 (𝜀̅̇ ) + 𝐸p 𝜀̅p . (22.19.2)
where 𝜀̅p is the effective plastic strain and 𝐸p is given in terms of Young’s modulus and the
tangent modulus by
𝐸𝐸t
𝐸p = . (22.19.3)
𝐸 − 𝐸t

Both Young's modulus and the tangent modulus may optionally be made functions
of strain rate by specifying a load curve ID giving their values as a function of strain rate. If
these load curve ID's are input as 0, then the constant values specified in the input are used.

Note that all load curves used to define quantities as a function of strain rate must
have the same number of points at the same strain rate values. This requirement is used to
allow vectorized interpolation to enhance the execution speed of this constitutive model.

This model also contains a simple mechanism for modeling material failure. This
option is activated by specifying a load curve ID defining the effective stress at failure as a
function of strain rate. For solid elements, once the effective stress exceeds the failure stress
the element is deemed to have failed and is removed from the solution. For shell elements
the entire shell element is deemed to have failed if all integration points through the
thickness have an effective stress that exceeds the failure stress. After failure the shell
element is removed from the solution.

In addition to the above failure criterion, this material model also supports a shell
element deletion criterion based on the maximum stable time step size for the element,
Δ𝑡max . Generally, Δ𝑡max goes down as the element becomes more distorted. To assure
stability of time integration, the global LS-DYNA time step is the minimum of the Δ𝑡max
values calculated for all elements in the model. Using this option allows the selective
deletion of elements whose time step Δ𝑡max has fallen below the specified minimum time
step, Δ𝑡crit . Elements which are severely distorted often indicate that material has failed
and supports little load, but these same elements may have very small time steps and
therefore control the cost of the analysis. This option allows these highly distorted
20-68 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Material Models

elements to be deleted from the calculation, and, therefore, the analysis can proceed at a
larger time step, and, thus, at a reduced cost. Deleted elements do not carry any load, and
are deleted from all applicable slide surface definitions. Clearly, this option must be
judiciously used to obtain accurate results at a minimum cost.

LS-DYNA DEV 10/27/16 (r:8004) 20-69 (Material Models)


Material Models LS-DYNA Theory Manual

22.20 Material Model 20: Rigid


The rigid material type 20 provides a convenient way of turning one or more parts
comprised of beams, shells, or solid elements into a rigid body. Approximating a
deformable body as rigid is a preferred modeling technique in many real world
applications. For example, in sheet metal forming problems the tooling can properly and
accurately be treated as rigid. In the design of restraint systems the occupant can, for the
purposes of early design studies, also be treated as rigid. Elements which are rigid are
bypassed in the element processing and no storage is allocated for storing history variables;
consequently, the rigid material type is very cost efficient.

Two unique rigid part IDs may not share common nodes unless they are merged
together using the rigid body merge option. A rigid body may be made up of disjoint finite
element meshes, however. LS-DYNA assumes this is the case since this is a common
practice in setting up tooling meshes in forming problems.

All elements which reference a given part ID corresponding to the rigid material
should be contiguous, but this is not a requirement. If two disjoint groups of elements on
opposite sides of a model are modeled as rigid, separate part ID's should be created for
each of the contiguous element groups if each group is to move independently. This
requirement arises from the fact that LS-DYNA internally computes the six rigid body
degrees-of-freedom for each rigid body (rigid material or set of merged materials), and if
disjoint groups of rigid elements use the same part ID, the disjoint groups will move
together as one rigid body.

Inertial properties for rigid materials may be defined in either of two ways. By
default, the inertial properties are calculated from the geometry of the constituent elements
of the rigid material and the density specified for the part ID. Alternatively, the inertial
properties and initial velocities for a rigid body may be directly defined, and this overrides
data calculated from the material property definition and nodal initial velocity definitions.

Young's modulus, E, and Poisson's ratio, υ are used for determining sliding interface
parameters if the rigid body interacts in a contact definition. Realistic values for these
constants should be defined since unrealistic values may contribute to numerical problem
in contact.

20-70 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.21 Material Model 21: Thermal Orthotropic Elastic


In the implementation for three-dimensional continua a total Lagrangian
formulation is used. In this approach the material law that relates second Piola-Kirchhoff
stress 𝐒 to the Green-St. Venant strain 𝐄 is
𝐒 = 𝐂 ⋅ 𝐄 = 𝐓T 𝐂l 𝐓 ⋅ 𝐄, (22.21.1)
where 𝐓 is the transformation matrix [Cook 1974].
𝑙2 𝑚21 𝑛21 𝑙1 𝑚 1 𝑚 1 𝑛1 𝑛1 𝑙 1
⎡ 12 ⎤
⎢ 𝑙2 𝑚22 𝑛22 𝑙2 𝑚 2 𝑚 2 𝑛2 𝑛2 𝑙 2 ⎥
⎢ 2 ⎥

𝐓 = ⎢ 𝑙3 𝑚23 𝑛23 𝑙3 𝑚 3 𝑚 3 𝑛3 𝑛3 𝑙 3 ⎥
⎥,
⎢2𝑙1 𝑙2 2𝑚1 𝑚2 2𝑛1 𝑛2 (𝑙1 𝑚2 + 𝑙2 𝑚1 ) (𝑚1 𝑛2 + 𝑚2 𝑛1 ) (𝑛1 𝑙2 + 𝑛2 𝑙1 )⎥ (22.21.2)
⎢ ⎥
⎢2𝑙2 𝑙3 2𝑚2 𝑚3 2𝑛2 𝑛3 (𝑙2 𝑚3 + 𝑙3 𝑚2 ) (𝑚2 𝑛3 + 𝑚3 𝑛2 ) (𝑛2 𝑙3 + 𝑛3 𝑙2 )⎥
⎣2𝑙3 𝑙1 2𝑚3 𝑚1 2𝑛3 𝑛1 (𝑙3 𝑚1 + 𝑙1 𝑚3 ) (𝑚3 𝑛1 + 𝑚1 𝑛3 ) (𝑛3 𝑙1 + 𝑛1 𝑙3 )⎦

𝑙𝑖 , 𝑚𝑖 , 𝑛𝑖 are the direction cosines


𝑥𝑖′ = 𝑙𝑖 𝑥1 + 𝑚𝑖 𝑥2 + 𝑛𝑖 𝑥3 for 𝑖 = 1, 2, 3, (22.21.3)
and 𝑥𝑖′ denotes the material axes. The constitutive matrix 𝐂l is defined in terms of the
material axes as
1 𝜐21 𝜐31
⎡ − − 0 0 0 ⎤
⎢ 𝐸11 𝐸22 𝐸33 ⎥
⎢ 𝜐 1 𝜐 ⎥
⎢− 12 − 32 0 0 0 ⎥
⎢ 𝐸 11 𝐸22 𝐸33 ⎥
⎢ ⎥
⎢ − 𝜐13 𝜐
− 23
1
0 0 0 ⎥
⎢ 𝐸 𝐸22 𝐸33 ⎥
𝐂l−1 =⎢ 11 ⎥, (22.21.4)
⎢ 1 ⎥
⎢ 0 0 0 0 0 ⎥
⎢ 𝐺12 ⎥
⎢ 1 ⎥
⎢ 0 0 0 0 0 ⎥
⎢ 𝐺23 ⎥
⎢ ⎥
⎢ 0 1 ⎥
0 0 0 0
⎣ 𝐺31 ⎦
where the subscripts denote the material axes, i.e.,
υ𝑖𝑗 = υ𝑥′𝑖 𝑥′𝑗 and 𝐸𝑖𝑖 = 𝐸𝑥′𝑖 . (22.21.5)

Since 𝐂l is symmetric
υ12 υ
= 21 , etc. (22.21.6)
𝐸11 𝐸22

LS-DYNA DEV 10/27/16 (r:8004) 20-71 (Material Models)


Material Models LS-DYNA Theory Manual

The vector of Green-St. Venant strain components is


𝐄T = [𝐸11 𝐸22 𝐸33 𝐸12 𝐸23 𝐸31 ], (22.21.7)
which include the local thermal strains which are integrated in time:
𝜀𝑛+1 𝑛
𝑎𝑎 = 𝜀𝑎𝑎 + 𝛼𝑎 (𝑇
𝑛+1
− 𝑇 𝑛 ),
𝜀𝑛+1 𝑛
𝑏𝑏 = 𝜀𝑏𝑏 + 𝛼𝑏 (𝑇
𝑛+1
− 𝑇 𝑛 ), (22.21.8)
𝜀𝑛+1 𝑛
𝑐𝑐 = 𝜀𝑐𝑐 + 𝛼𝑐 (𝑇
𝑛+1
− 𝑇 𝑛 ).

After computing 𝑆𝑖𝑗 we use Equation (18.32) to obtain the Cauchy stress. This model
will predict realistic behavior for finite displacement and rotations as long as the strains are
small.

For shell elements, the stresses are integrated in time and are updated in the
corotational coordinate system. In this procedure the local material axes are assumed to
remain orthogonal in the deformed configuration. This assumption is valid if the strains
remain small.

20-72 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.22 Material Model 22: Chang-Chang Composite Failure


Model
For shells, five material parameters are used in the three failure criteria. These are
[Chang and Chang 1987a, 1987b]:
• 𝑆1 , longitudinal tensile strength
• 𝑆2 , transverse tensile strength
• 𝑆12 , shear strength
• 𝐶2 , transverse compressive strength
• 𝛼, nonlinear shear stress parameter.
𝑆1 , 𝑆2 , 𝑆12 , and 𝐶2 are obtained from material strength measurement. 𝛼 is defined by
material shear stress-strain measurements. In plane stress, the strain is given in terms of
the stress as
1
𝜀1 = (𝜎 − 𝜐1 𝜎2 ),
𝐸1 1
1
𝜀2 = (𝜎 − 𝜐2 𝜎1 ), (22.22.1)
𝐸2 2
1 3
2𝜀12 = 𝜏 + 𝛼𝜏12 .
𝐺12 12

The third equation defines the nonlinear shear stress parameter 𝛼. A fiber matrix
shearing term augments each damage mode:
2
𝜏12
+ 3 𝛼𝜏 4
2𝐺12 4 12
𝜏̅ = 2 , (22.22.2)
𝑆12 3
+ 𝛼𝑆4
2𝐺12 4 12
which is the ratio of the shear stress to the shear strength.

The matrix cracking failure criteria is determined from


𝜎2 2
𝐹matrix = ( ) + 𝜏̅ , (22.22.3)
𝑆2
where failure is assumed whenever 𝐹matrix > 1. If 𝐹matrix > 1, then the material constants
𝐸2 , 𝐺12 , 𝜐1 , and 𝜐2 are set to zero.

The compression failure criteria is given as

LS-DYNA DEV 10/27/16 (r:8004) 20-73 (Material Models)


Material Models LS-DYNA Theory Manual

𝜎2 2 ⎡ 𝐶2 2 𝜎2
𝐹comp =( ) + ⎢( ) − 1⎤
⎥ + 𝜏̅ , (22.22.4)
2𝑆12 ⎣ 2𝑆12 ⎦ 𝐶2
where failure is assumed whenever 𝐹comb > 1. If 𝐹comb > 1, then the material constants 𝐸2 ,
𝜐1 , and 𝜐2 are set to zero.

The final failure mode is due to fiber breakage.


𝜎1 2
𝐹fiber = ( ) + 𝜏̅ , (22.22.5)
S1
Failure is assumed whenever 𝐹fiber > 1. If 𝐹fiber > 1, then the constants 𝐸1 , 𝐸2 , 𝐺12 𝜐1 and
𝜐2 are set to zero.

For solids, a fourth failure mode corresponding to delamination is computed as


max(0.0, 𝜎3 ) 2 𝜏 2 𝜏 2
𝐹delam = ( ) + ( 23 ) + ( 31 )
S3 S23 S31
This involves three additional material parameters.
• 𝑆3 , normal tensile strength (22.22.140)
• 𝑆23 , transverse shear strength
• 𝑆31 , transverse shear strength

20-74 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.23 Material Model 23: Thermal Orthotropic Elastic with


12 Curves
In the implementation for three-dimensional continua a total Lagrangian
formulation is used. In this approach the material law that relates second Piola-Kirchhoff
stress 𝐒 to the Green-St. Venant strain 𝐄 is
𝐒 = 𝐂 ⋅ 𝐄 = 𝐓T 𝐂l 𝐓 ⋅ 𝐄, (22.23.1)
where 𝐓 is the transformation matrix [Cook 1974].
𝑙2 𝑚21 𝑛21 𝑙1 𝑚1 𝑚1 𝑛1 𝑛1 𝑙1
⎡ 12 ⎤
⎢ 𝑙2 𝑚22 𝑛22 𝑙2 𝑚2 𝑚2 𝑛2 𝑛2 𝑙2 ⎥
⎢ 2 ⎥

𝐓 = ⎢ 𝑙3 𝑚23 𝑛23 𝑙3 𝑚3 𝑚3 𝑛3 𝑛3 𝑙3 ⎥
⎥, (22.23.2)
⎢2𝑙1 𝑙2 2𝑚1 𝑚2 2𝑛1 𝑛2 (𝑙1 𝑚2 + 𝑙1 𝑚1 ) (𝑚1 𝑛2 + 𝑚2 𝑛1 ) (𝑛1 𝑙2 + 𝑛2 𝑙1 )⎥
⎢ ⎥
⎢2𝑙2 𝑙3 2𝑚2 𝑚3 2𝑛2 𝑛3 (𝑙2 𝑚3 + 𝑙3 𝑚2 ) (𝑚2 𝑛3 + 𝑚3 𝑛2 ) (𝑛2 𝑙3 + 𝑛3 𝑙2 )⎥
⎣2𝑙3 𝑙1 2𝑚3 𝑚1 2𝑛3 𝑛1 (𝑙3 𝑚1 + 𝑙1 𝑚3 ) (𝑚3 𝑛1 + 𝑚1 𝑛3 ) (𝑛3 𝑙1 + 𝑛1 𝑙3 )⎦
𝑙𝑖 , 𝑚𝑖 , 𝑛𝑖 are the direction cosines
𝑥𝑖′ = 𝑙𝑖 𝑥1 + 𝑚𝑖 𝑥2 + 𝑛𝑖 𝑥3 for 𝑖 = 1, 2, 3, (22.23.3)
and 𝑥𝑖′ denotes the material axes. The temperature dependent constitutive matrix 𝐂l is
defined in terms of the material axes as
1 𝜐21 (𝑇) 𝜐31 (𝑇)
⎡ − − 0 0 0 ⎤
⎢ 𝐸11 (𝑇) 𝐸22 (𝑇) 𝐸33 (𝑇) ⎥
⎢ 𝜐 (𝑇) 1 𝜐 (𝑇) ⎥
⎢− 12 − 32 0 0 0 ⎥
⎢ 𝐸11 (𝑇) 𝐸22 (𝑇) 𝐸33 (𝑇) ⎥
⎢ ⎥
⎢− 𝜐13 (𝑇) 𝜐 (𝑇)
− 23
1
0 0 0 ⎥
⎢ 𝐸 (𝑇) 𝐸 22 (𝑇) 𝐸33 (𝑇) ⎥
𝐂−1
l =⎢ 11 ⎥, (22.23.4)
⎢ 1 ⎥
⎢ 0 0 0 0 0 ⎥
⎢ 𝐺12 (𝑇) ⎥
⎢ 1 ⎥
⎢ 0 0 0 0 0 ⎥
⎢ 𝐺23 (𝑇) ⎥
⎢ ⎥
⎢ 1 ⎥
0 0 0 0 0
⎣ 𝐺31 (𝑇)⎦
where the subscripts denote the material axes, i.e.,
𝜐𝑖𝑗 = υ𝑥′𝑖 𝑥′𝑗 and 𝐸𝑖𝑖 = 𝐸𝑥′𝑖 . (22.23.5)

Since 𝐂l is symmetric
𝜐12 𝜐21
= , etc. (22.23.6)
𝐸11 𝐸22

LS-DYNA DEV 10/27/16 (r:8004) 20-75 (Material Models)


Material Models LS-DYNA Theory Manual

The vector of Green-St. Venant strain components is


𝐄T = [𝐸11 𝐸22 𝐸33 𝐸12 𝐸23 𝐸31 ], (22.23.7)
which include the local thermal strains which are integrated in time:
𝑛+1
𝜀𝑛+1 𝑛
𝑎𝑎 = 𝜀𝑎𝑎 + 𝛼𝑎 (𝑇
2 ) [𝑇 𝑛+1 − 𝑇 𝑛 ],

𝑛+1
𝜀𝑛+1 𝑛
𝑏𝑏 = 𝜀𝑏𝑏 + 𝛼𝑏 (𝑇
2 ) [𝑇 𝑛+1 − 𝑇 𝑛 ], (22.23.8)

𝑛+1
𝜀𝑛+1 𝑛
𝑐𝑐 = 𝜀𝑐𝑐 + 𝛼𝑐 (𝑇
2 ) [𝑇 𝑛+1 − 𝑇 𝑛 ].

After computing 𝑆𝑖𝑗 we use Equation (16.32) to obtain the Cauchy stress. This model
will predict realistic behavior for finite displacement and rotations as long as the strains are
small.

For shell elements, the stresses are integrated in time and are updated in the
corotational coordinate system. In this procedure the local material axes are assumed to
remain orthogonal in the deformed configuration. This assumption is valid if the strains
remain small.

20-76 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.24 Material Model 24: Piecewise Linear Isotropic


Plasticity
This plasticity treatment in this model is quite similar to Model 10, but unlike 10, it
includes strain rate effects and does not use an equation of state. Deviatoric stresses are
determined that satisfy the yield function
2
1 𝜎y
𝜙 = 𝑠𝑖𝑗 𝑠𝑖𝑗 − ≤ 0, (22.24.1)
2 3
where
p
σy = 𝛽[𝜎0 + 𝑓h (𝜀eff )], (22.24.2)
p
where the hardening function 𝑓h (𝜀eff ) can be specified in tabular form as an option.
Otherwise, linear hardening of the form
p p
𝑓h (𝜀eff ) = 𝐸p (𝜀eff ), (22.24.3)
p
is assumed where 𝐸p and 𝜀eff are given in Equations (22.3.6) and (8.61), respectively. The
parameter 𝛽 accounts for strain rate effects. For complete generality a table defining the
yield stress versus plastic strain may be defined for various levels of effective strain rate.

In the implementation of this material model, the deviatoric stresses are updated
elastically (see material model 1), the yield function is checked, and if it is satisfied the
deviatoric stresses are accepted. If it is not, an increment in plastic strain is computed:
1⁄
(3 𝑠∗𝑖𝑗 𝑠∗𝑖𝑗 ) − 𝜎y
2
(22.24.4)
= 2
p
Δ𝜀eff ,
3𝐺 + 𝐸p
is the shear modulus and 𝐸p is the current plastic hardening modulus. The trial deviatoric
stress state 𝑠∗𝑖𝑗 is scaled back:
𝜎𝑦
𝑠𝑛+1
𝑖𝑗 = 𝑠∗ .
1⁄ 𝑖𝑗 (22.24.5)
(3 𝑠∗𝑖𝑗 𝑠∗𝑖𝑗 )
2
2

For shell elements, the above equations apply, but with the addition of an iterative
loop to solve for the normal strain increment, such that the stress component normal to the
mid surface of the shell element approaches zero.

Three options to account for strain rate effects are possible:

LS-DYNA DEV 10/27/16 (r:8004) 20-77 (Material Models)


Material Models LS-DYNA Theory Manual

1. Strain rate may be accounted for using the Cowper-Symonds model which scales
the yield stress with the factor
1
𝜀̇ ⁄𝑝 (22.24.6)
𝛽=1+( ) .
𝐶
where 𝜀̇ is the strain rate.

2. For complete generality a load curve, defining 𝛽, which scales the yield stress may
be input instead. In this curve the scale factor versus strain rate is defined.

3. If different stress versus strain curves can be provided for various strain rates, the
option using the reference to a table definition can be used. See Figure 19.24.1.

A fully viscoplastic formulation is optional which incorporates the different options


above within the yield surface. An additional cost is incurred over the simple scaling but
the improvement is results can be dramatic.

If a table ID is specified a curve ID is given for each strain rate, see Section 23.
Intermediate values are found by interpolating between curves. Effective plastic strain
versus yield stress is expected. If the strain rate values fall out of range, extrapolation is not
used; rather, either the first or last curve determines the yield stress depending on whether
the rate is low or high, respectively.

20-78 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.25 Material Model 25: Kinematic Hardening Cap Model


The implementation of an extended two invariant cap model, suggested by Stojko
[1990], is based on the formulations of Simo, et al. [1988, 1990] and Sandler and Rubin
[1979]. In this model, the two invariant cap theory is extended to include nonlinear
kinematic hardening as suggested by Isenberg, Vaughn, and Sandler [1978]. A brief
discussion of the extended cap model and its parameters is given below.

The cap model is formulated in terms of the invariants of the stress tensor. The
square root of the second invariant of the deviatoric stress tensor, √𝐽2D is found from the
deviatoric stresses 𝐒 as

1
√𝐽2D ≡ √ 𝑠𝑖𝑗 𝑠𝑖𝑗 , (22.25.1)
2
and is the objective scalar measure of the distortional or shearing stress. The first invariant
of the stress, 𝐽1 , is the trace of the stress tensor.

The cap model consists of three surfaces in √𝐽2D − 𝐽1 space, as shown in Figure
22.25.1. First, there is a failure envelope surface, denoted 𝑓1 in the figure. The functional
form of 𝑓1 is
𝑓1 = √𝐽2D − min(𝐹e (𝐽1 ), 𝑇mises ), (22.25.2)
where 𝐹e is given by
𝐹e (𝐽1 ) ≡ 𝛼 − 𝛾exp(−𝛽𝐽1 ) + 𝜃𝐽1 . (22.25.3)

5
4
3
2

σy 1

εpeff
Figure 22.25.1. Rate effects may be accounted for by defining a table of curves.

LS-DYNA DEV 10/27/16 (r:8004) 20-79 (Material Models)


Material Models LS-DYNA Theory Manual

J 2D
J 2D = Fe

J2D = Fc

f1
f2

f3
J1
T O X( )
Figure 22.25.2. The yield surface of the two-invariant cap model in
pressure √𝐽2D − 𝐽1 space Surface 𝑓1 is the failure envelope, 𝑓2 is the cap surface,
and 𝑓3 is the tension cutoff.
and 𝑇mises ≡ |𝑋(𝜅𝑛 ) − 𝐿(𝜅𝑛 )|. This failure envelope surface is fixed in √𝐽2D − 𝐽1 space, and
therefore does not harden unless kinematic hardening is present. Next, there is a cap
surface, denoted 𝑓2 in the figure, with 𝑓2 given by
𝑓2 = √𝐽2D − 𝐹c (𝐽1 , 𝜅), (22.25.4)
where 𝐹c is defined by
1
𝐹c (𝐽1 , 𝜅) ≡ √[𝑋(𝜅) − 𝐿(𝜅)] 2 − [𝐽1 − 𝐿(𝜅)] 2 , (22.25.5)
𝑅
𝑋(𝜅) is the intersection of the cap surface with the 𝐽1 axis
𝑋(𝜅) = 𝜅 + 𝑅𝐹e (𝜅), (22.25.6)
and 𝐿(𝜅) is defined by
𝜅 𝜅>0
𝐿(𝜅) ≡ { . (22.25.7)
0 𝜅≤0

p
The hardening parameter 𝜅 is related to the plastic volume change 𝜀v through the
hardening law
p
𝜀v = W{1 − exp[−𝐷(𝑋(κ) − 𝑋0 )]}. (22.25.8)

Geometrically, 𝜅 is seen in the figure as the 𝐽1 coordinate of the intersection of the


cap surface and the failure surface. Finally, there is the tension cutoff surface, denoted 𝑓3 in
the figure. The function 𝑓3 is given by
𝑓3 + 𝑇 − 𝐽1 , (22.25.9)
where 𝑇 is the input material parameter which specifies the maximum hydrostatic tension
sustainable by the material. The elastic domain in √𝐽2D − 𝐽1 space is then bounded by the

20-80 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

failure envelope surface above, the tension cutoff surface on the left, and the cap surface on
the right.

An additive decomposition of the strain into elastic and plastic parts is assumed:
𝜀 = 𝜀 + 𝜀P , (22.25.10)
where 𝜀e is the elastic strain and 𝜀p is the plastic strain. Stress is found from the elastic
strain using Hooke’s law,
𝜎 = 𝐶(𝜀 − 𝜀P ), (22.25.11)
where 𝜎 is the stress and 𝐶 is the elastic constitutive tensor.

The yield condition may be written


𝑓1 (𝜎) ≤ 0,
𝑓2 (𝜎, 𝜅) ≤ 0, (22.25.12)
𝑓3 (𝜎) ≤ 0,
and the plastic consistency condition requires that
𝜆̇ 𝑘 𝑓𝑘 = 0
𝑘 = 1, 2, 3, (22.25.13)
𝜆̇ 𝑘 ≥ 0
where 𝜆𝑘 is the plastic consistency parameter for surface 𝑘. If 𝑓𝑘 < 0, then 𝜆̇ 𝑘 = 0 and the
response is elastic. If 𝑓𝑘 > 0, then surface k is active and 𝜆̇ 𝑘 is found from the requirement
that 𝑓𝑘̇ = 0.

Associated plastic flow is assumed, so using Koiter’s flow rule the plastic strain rate
is given as the sum of contribution from all of the active surfaces,
3
∂𝑓𝑘
𝜀̇ = ∑ 𝜆̇ 𝑘
p
. (22.25.14)
𝑘=1
∂𝑠

One of the major advantages of the cap model over other classical pressure-
dependent plasticity models is the ability to control the amount of dilatency produced
under shear loading. Dilatency is produced under shear loading as a result of the yield
surface having a positive slope in √𝐽2D − 𝐽1 space, so the assumption of plastic flow in the
direction normal to the yield surface produces a plastic strain rate vector that has a
component in the volumetric (hydrostatic) direction (see Figure 22.25.1). In models such as
the Drucker-Prager and Mohr-Coulomb, this dilatency continues as long as shear loads are
applied, and in many cases produces far more dilatency than is experimentally observed in
material tests. In the cap model, when the failure surface is active, dilatency is produced
just as with the Drucker-Prager and Mohr-Columb models. However, the hardening law
permits the cap surface to contract until the cap intersects the failure envelope at the stress
point, and the cap remains at that point. The local normal to the yield surface is now
vertical, and therefore the normality rule assures that no further plastic volumetric strain
LS-DYNA DEV 10/27/16 (r:8004) 20-81 (Material Models)
Material Models LS-DYNA Theory Manual

(dilatency) is created. Adjustment of the parameters that control the rate of cap
contractions permits experimentally observed amounts of dilatency to be incorporated into
the cap model, thus producing a constitutive law which better represents the physics to be
modeled. Another advantage of the cap model over other models such as the Drucker-
Prager and Mohr-Coulomb is the ability to model plastic compaction. In these models all
purely volumetric response is elastic. In the cap model, volumetric response is elastic until
the stress point hits the cap surface. Therefore, plastic volumetric strain (compaction) is
generated at a rate controlled by the hardening law. Thus, in addition to controlling the
amount of dilatency, the introduction of the cap surface adds another experimentally
observed response characteristic of geological material into the model.

The inclusion of kinematic hardening results in hysteretic energy dissipation under


cyclic loading conditions. Following the approach of Isenberg, et al., [1978] a nonlinear
kinematic hardening law is used for the failure envelope surface when nonzero values of
and N are specified. In this case, the failure envelope surface is replaced by a family of
yield surfaces bounded by an initial yield surface and a limiting failure envelope surface.
Thus, the shape of the yield surfaces described above remains unchanged, but they may
translate in a plane orthogonal to the J axis.

Translation of the yield surfaces is permitted through the introduction of a “back


stress” tensor, α. The formulation including kinematic hardening is obtained by replacing
the stress σ with the translated stress tensor 𝜂 ≡ 𝜎 − 𝛼 in all of the above equation. The
history tensor α is assumed deviatoric, and therefore has only 5 unique components. The
evolution of the back stress tensor is governed by the nonlinear hardening law
𝛼 = c𝐹̅ ̅(𝜎, 𝛼) 𝜀̇p , (22.25.15)
where c ̅ is a constant, 𝐹̅ is a scalar function of 𝜎 and 𝛼 and 𝜀̇p is the rate of deviator plastic
strain. The constant may be estimated from the slope of the shear stress - plastic shear
strain curve at low levels of shear stress.

The function 𝐹̅ is defined as


(𝜎 − 𝛼)𝛼
𝐹̅ ≡ max (0,1 − ), (22.25.16)
2N𝐹𝑒 (𝐽1 )
where N is a constant defining the size of the yield surface. The value of N may be
interpreted as the radial distant between the outside of the initial yield surface and the
inside of the limit surface. In order for the limit surface of the kinematic hardening cap
model to correspond with the failure envelope surface of the standard cap model, the scalar
parameter a must be replaced α − N in the definition 𝐹e .

The cap model contains a number of parameters which must be chosen to represent
a particular material, and are generally based on experimental data. The parameters 𝛼, 𝛽,𝜃
and 𝛾 are usually evaluated by fitting a curve through failure data taken from a set of
triaxial compression tests. The parameters 𝑊, 𝐷, and X0 define the cap hardening law.
The value W represents the void fraction of the uncompressed sample and 𝐷 governs the
20-82 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Material Models

slope of the initial loading curve in hydrostatic compression. The value of R is the ration of
major to minor axes of the quarter ellipse defining the cap surface. Additional details and
guidelines for fitting the cap model to experimental data are found in [Chen and Baladi,
1985].

LS-DYNA DEV 10/27/16 (r:8004) 20-83 (Material Models)


Material Models LS-DYNA Theory Manual

22.26 Material Model 26: Crushable Foam


This orthotropic material model does the stress update in the local material system
denoted by the subscripts, 𝑎, 𝑏, and 𝑐. The material model requires the following input
parameters:
• E, Young’s modulus for the fully compacted material;
• 𝜈, Poisson’s ratio for the compacted material;
• 𝜎y , yield stress for fully compacted honeycomb;
• LCA, load curve number for sigma-aa versus either relative volume or volumetric
strain (see Figure 19.26.1.);
• LCB, load curve number for sigma-bb versus either relative volume or volumetric
strain (default: LCB = LCA);
• LCC, the load curve number for sigma-cc versus either relative volume or volumet-
ric strain (default: LCC = LCA);
• LCS, the load curve number for shear stress versus either relative volume or
volumetric strain (default LCS = LCA);
• 𝑉f , relative volume at which the honeycomb is fully compacted;
• 𝐸𝑎𝑎u , elastic modulus in the uncompressed configuration;
• 𝐸𝑏𝑏u , elastic modulus in the uncompressed configuration;
• 𝐸𝑐𝑐u , elastic modulus in the uncompressed configuration;
• 𝐺𝑎𝑏u , elastic shear modulus in the uncompressed configuration;
• 𝐺𝑏𝑐u , elastic shear modulus in the uncompressed configuration;
• 𝐺𝑐𝑎u , elastic shear modulus in the uncompressed configuration;
• LCAB, load curve number for sigma-ab versus either relative volume or volumetric
strain (default: LCAB = LCS);
• LCBC, load curve number for sigma-bc versus either relative volume or volumetric
strain default: LCBC = LCS);
• LCCA, load curve number for sigma-ca versus either relative volume or volumetric
strain (default: LCCA = LCS);
• LCSR, optional load curve number for strain rate effects.

The behavior before compaction is orthotropic where the components of the stress
tensor are uncoupled, i.e., an 𝑎 component of strain will generate resistance in the local 𝑎

20-84 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

direction with no coupling to the local 𝑏 and 𝑐 directions. The elastic moduli vary linearly
with the relative volume from their initial values to the fully compacted values:
𝐸𝑎𝑎 = 𝐸𝑎𝑎u + 𝛽(𝐸 − 𝐸𝑎𝑎u ),
𝐸𝑏𝑏 = 𝐸𝑏𝑏u + 𝛽(𝐸 − 𝐸𝑏𝑏u ),
𝐸𝑐𝑐 = 𝐸𝑐𝑐u + 𝛽(𝐸 − 𝐸𝑐𝑐u ),
(22.26.1)
𝐺𝑎𝑏 = 𝐺𝑎𝑏u + 𝛽(𝐺 − 𝐺𝑎𝑏u ),
𝐺𝑏𝑐 = 𝐺𝑏𝑐u + 𝛽(𝐺 − 𝐺𝑏𝑐u ),
𝐺𝑐𝑎 = 𝐺𝑐𝑎u + 𝛽(𝐺 − 𝐺𝑐𝑎u ),
where
1 − 𝑉min
𝛽 = max [min ( , 1) ,0], (22.26.2)
1 − 𝑉𝑓
and 𝐺 is the elastic shear modulus for the fully compacted honeycomb material
𝐸
𝐺= . (22.26.3)
2(1 + 𝜈)

The relative volume V is defined as the ratio of the current volume over the initial
volume; typically, 𝑉 = 1 at the beginning of a calculation. The relative volume, 𝑉min , is the
minimum value reached during the calculation.

Curve extends into negative volumetric


strain quadrant since LS-DYNA will
extrapolate using the two end points. It
is important that the extropolation does
not extend into the negative stress
σij region.

unloading and
reloading path

0 strain: -εij
Unloading is based on the interpolated Young’s
moduli which must provide an unloading
tangent that exceeds the loading tangent.
Figure 22.26.1. Stress quantity versus volumetric strain. Note that the “yield
stress” at a volumetric strain of zero is nonzero. In the load curve definition, the
“time” value is the volumetric strain and the “function” value is the yield stress.

LS-DYNA DEV 10/27/16 (r:8004) 20-85 (Material Models)


Material Models LS-DYNA Theory Manual

The load curves define the magnitude of the average stress as the material changes
density (relative volume). Each curve related to this model must have the same number of
points and the same abscissa values. There are two ways to define these curves: as a
function of relative volume V, or as a function of volumetric strain defined as:
𝜀𝑉 = 1 − 𝑉. (22.26.4)

In the former, the first value in the curve should correspond to a value of relative
volume slightly less than the fully compacted value. In the latter, the first value in the
curve should be less than or equal to zero corresponding to tension and should increase to
full compaction. When defining the curves, care should be taken that the extrapolated
values do not lead to negative yield stresses.

At the beginning of the stress update we transform each element’s stresses and
strain rates into the local element coordinate system. For the uncompacted material, the
trial stress components are updated using the elastic interpolated moduli according to:
trial
𝑛+1 𝑛
𝜎𝑎𝑎 = 𝜎𝑎𝑎 + 𝐸𝑎𝑎 Δ𝜀𝑎𝑎 ,
trial
𝑛+1 𝑛
𝜎𝑏𝑏 = 𝜎𝑏𝑏 + 𝐸𝑏𝑏 Δ𝜀𝑏𝑏 ,
trial
𝑛+1 𝑛
𝜎𝑐𝑐 = 𝜎𝑐𝑐 + 𝐸𝑐𝑐 Δ𝜀𝑐𝑐 ,
(22.26.5)
𝑛+1trial 𝑛
𝜎𝑎𝑏 = 𝜎𝑎𝑏 + 2𝐺𝑎𝑏 Δ𝜀𝑎𝑏 ,
trial
𝑛+1 𝑛
𝜎𝑏𝑐 = 𝜎𝑏𝑐 + 2𝐺𝑏𝑐 Δ𝜀𝑏𝑐 ,
trial
𝑛+1 𝑛
𝜎𝑐𝑎 = 𝜎𝑐𝑎 + 2𝐺𝑐𝑎 Δ𝜀𝑐𝑎 = 1.

Then we independently check each component of the updated stresses to ensure that
they do not exceed the permissible values determined from the load curves, e.g., if
trial
∣𝜎𝑖𝑗𝑛+1 ∣ > 𝜆𝜎𝑖𝑗 (𝑉min ), (22.26.6)

then
trial
𝜆𝜎𝑖𝑗𝑛+1
𝜎𝑖𝑗𝑛+1 = 𝜎𝑖𝑗 (𝑉min ) trial
. (22.26.7)
∣𝜎𝑖𝑗𝑛+1 ∣

The parameter 𝜆 is either unity or a value taken from the load curve number, LCSR,
that defines 𝜆 as a function of strain rate. Strain rate is defined here as the Euclidean norm
of the deviatoric strain rate tensor.

For fully compacted material we assume that the material behavior is elastic-
perfectly plastic and updated the stress components according to

20-86 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

𝑛+1⁄2
𝑠trial
𝑖𝑗 = 𝑠𝑛𝑖𝑗 + 2𝐺Δ𝜀dev
𝑖𝑗 , (22.26.8)

where the deviatoric strain increment is defined as


1
Δ𝜀dev
𝑖𝑗 = Δ𝜀𝑖𝑗 − Δ𝜀𝑘𝑘 𝛿𝑖𝑗 . (22.26.9)
3

We next check to see if the yield stress for the fully compacted material is exceeded
by comparing
1
3 trial trial ⁄2 (22.26.10)
𝑠trial
eff = ( 𝑠𝑖𝑗 𝑠𝑖𝑗 ) .
2
the effective trial stress, to the yield stress 𝜎y . If the effective trial stress exceeds the yield
stress, we simply scale back the stress components to the yield surface:
𝜎y
𝑠𝑛+1
𝑖𝑗 = trial 𝑠trial
𝑖𝑗 . (22.26.11)
𝑠eff

We can now update the pressure using the elastic bulk modulus, 𝐾:
𝑛+1⁄2
𝑝𝑛+1 = 𝑝𝑛 − 𝐾Δ𝜀𝑘𝑘 ,
(22.26.12)
𝐸
𝐾= ,
3(1 − 2𝜈)
and obtain the final value for the Cauchy stress
𝜎𝑖𝑗𝑛+1 = 𝑠𝑛+1
𝑖𝑗 − 𝑝𝑛+1 𝛿𝑖𝑗 . (22.26.13)

After completing the stress update, we transform the stresses back to the global
configuration.

LS-DYNA DEV 10/27/16 (r:8004) 20-87 (Material Models)


Material Models LS-DYNA Theory Manual

22.27 Material Model 27: Incompressible Mooney-Rivlin


Rubber
The Mooney-Rivlin material model is based on a strain energy function, 𝑊, as
follows
1
𝑊 = A(𝐼1 − 3) + B(𝐼2 − 3) + C( − 1) + D(𝐼3 − 1)2 . (22.27.1)
𝐼32

A and B are user defined constants, whereas C and D are related to A and B as
follows
1
C = A + B,
2
(22.27.2)
A(5𝜐 − 2) + B(11𝜐 − 5)
D= .
2(1 − 2𝜐)

The derivation of the constants C and D is straightforward [Feng, 1993] and is


included here since we were unable to locate it in the literature. The principal components
of Cauchy stress, 𝜎𝑖 , are given by [Ogden, 1984]
∂𝑊
𝐽𝜎𝑖 = 𝜆𝑖 . (22.27.3)
∂𝜆𝑖
For uniform dilation
𝜆1 = 𝜆2 = 𝜆3 = 𝜆, (22.27.4)
thus the pressure, 𝑝, is obtained (please note the sign convention),
2 𝜕𝑊 𝜕𝑊 𝜕𝑊
𝑝 = 𝜎1 = 𝜎2 = 𝜎3 = 3
(𝜆2 + 2𝜆4 + 𝜆6 ). (22.27.5)
𝜆 𝜕𝐼1 𝜕𝐼2 𝜕𝐼3

The relative volume, 𝑉, can be defined in terms of the stretches as:


new volume
𝑉 = 𝜆3 = . (22.27.6)
old volume

For small volumetric deformations the bulk modulus, 𝐾, can be defined as the ratio
of the pressure over the volumetric strain as the relative volume approaches unity:
𝑝
𝐾 = lim ( ). (22.27.7)
𝑉→1 𝑉 − 1

The partial derivatives of 𝑊 lead to:

20-88 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

∂𝑊
= A,
∂𝐼1
𝜕𝑊
= B,
𝜕𝐼2
𝜕𝑊
= −2C𝐼3−3 + 2D(𝐼3 − 1) = −2C𝜆−18 + 2D(𝜆6 − 1), (22.27.8)
𝜕𝐼3
2
𝑝= 3
{A𝜆2 + 2𝜆4 B + 𝜆6 [−2C𝜆−18 + 2D(𝜆6 − 1)]}
𝜆
2
= 3 {A𝜆2 + 2𝜆4 B − 2C𝜆−12 + 2D(𝜆12 − 𝜆6 )}.
𝜆

In the limit as the stretch ratio approaches unity, the pressure must approach zero:
lim 𝑝 = 0. (22.27.9)
𝜆→1

Therefore, A + 2B − 2C = 0 and
C = 0.5A + B. (22.27.10)

To solve for D we note that:


𝑝
𝐾 = lim ( )
𝑉→1 𝑉 − 1

2 {A𝜆2 + 2𝜆4 B − 2C𝜆−12 + 2D(𝜆12 − 𝜆6 )}


𝜆 3
= lim
λ→1 𝜆3 − 1
A𝜆2 + 2𝜆4 B − 2C𝜆−12 + 2D(𝜆12 − 𝜆6 )
= 2lim
λ→1 𝜆6 − 𝜆3 (22.27.11)
3 −13 11 5
2A𝜆 + 8𝜆 B + 24C𝜆 + 2D(12𝜆 − 6𝜆 )
= 2lim
λ→1 6𝜆5 − 3𝜆2
2
= (2A + 8B + 24C + 12D)
3
2
= (14A + 32B + 12D).
3

We therefore obtain:
3 3 2𝐺(1 + 𝜐) 2(A + B)(1 + 𝜐)
14A + 32B + 12D = 𝐾 = ( )= . (22.27.12)
2 2 3(1 − 2𝜐) (1 − 2𝜐)
Solving for D we obtain the desired equation:

LS-DYNA DEV 10/27/16 (r:8004) 20-89 (Material Models)


Material Models LS-DYNA Theory Manual

A(5𝜐 − 2) + B(11𝜐 − 5)
D= . (22.27.13)
2(1 − 2𝜐)

The invariants 𝐼1 − 𝐼3 are related to the right Cauchy-Green tensor C as


𝐼1 = 𝐶𝑖𝑖 ,
1 1
𝐼2 = 𝐶𝑖𝑖 2 − 𝐶𝑖𝑗 𝐶𝑖𝑗 , (22.27.14)
2 2
𝐼3 = det(𝐶𝑖𝑗 ).

The second Piola-Kirchhoff stress tensor, S, is found by taking the partial derivative
of the strain energy function with respect to the Green-Lagrange strain tensor, E.
∂𝑊 ∂𝑊 ∂𝐼 ∂𝐼 2C ∂I
𝑆𝑖𝑗 = =2 = 2 [A 1 + B 2 + (2D(𝐼3 − 1) − 2 ) 3 ]. (22.27.15)
∂𝐸𝑖𝑗 ∂𝐶𝑖𝑗 ∂𝐶𝑖𝑗 ∂𝐶𝑖𝑗 𝐼3 ∂𝐶𝑖𝑗

The derivatives of the invariants 𝐼1 − 𝐼3 are


∂𝐼1
= 𝛿𝑖𝑗 ,
∂𝐶𝑖𝑗
∂𝐼2
= 𝐼1 𝛿𝑖𝑗 − 𝐶𝑖𝑗 , (22.27.16)
∂𝐶𝑖𝑗
∂𝐼3
= 𝐼3 𝐶𝑖𝑗−1 .
∂𝐶𝑖𝑗

Inserting Equation (22.27.16) into Equation (22.27.15) yields the following expression
for the second Piola-Kirchhoff stress:
1
𝑆𝑖𝑗 = 2A𝛿𝑖𝑗 + 2B(𝐼1 𝛿𝑖𝑗 − 𝐶𝑖𝑗 ) − 4C 𝐶𝑖𝑗−1 + 4D(𝐼3 − 1)I3 𝐶𝑖𝑗−1 . (22.27.17)
𝐼32
Equation (22.27.17) can be transformed into the Cauchy stress by using the push forward
operation
1
𝜎𝑖𝑗 = 𝐹𝑖𝑘 𝑆𝑘𝑙 𝐹𝑗𝑙 . (22.27.18)
𝐽
where 𝐽 = det(𝐹𝑖𝑗 ).

22.27.1 Stress Update for Shell Elements


As a basis for discussing the algorithmic tangent stiffness for shell elements in
Section 19.27.3, the corresponding stress update as it is done in LS-DYNA is shortly

20-90 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

recapitulated in this section. When dealing with shell elements, the stress (as well as
constitutive matrix) is typically evaluated in corotational coordinates after which it is
transformed back to the standard basis according to
𝜎𝑖𝑗 = 𝑅𝑖𝑘 𝑅𝑗𝑙 𝜎̂ 𝑘𝑙 . (22.27.19)

Here 𝑅𝑖𝑗 is the rotation matrix containing the corotational basis vectors. The so-
called corotated stress 𝜎̂ 𝑖𝑗 is evaluated using Equation 19.27.21 with the exception that the
deformation gradient is expressed in the corotational coordinates, i.e.,
1
̂ 𝑆𝑘𝑙 𝐹𝑗𝑙̂ ,
𝜎̂ 𝑖𝑗 = 𝐹𝑖𝑘 (22.27.20)
𝐽
where 𝑆𝑖𝑗 is evaluated using Equation (22.27.17). The corotated deformation gradient is
incrementally updated with the aid of a time increment Δ𝑡, the corotated velocity gradient
̂𝑖𝑗 with which the embedded coordinate system is rotating.
𝐿̂𝑖𝑗 , and the angular velocity 𝛺
̂𝑖𝑘 )𝐹𝑘𝑗
𝐹𝑖𝑗̂ = (𝛿𝑖𝑘 + Δ𝑡𝐿̂𝑖𝑘 − Δ𝑡𝛺 ̂ . (22.27.21)

The primary reason for taking a corotational approach is to facilitate the


maintenance of a vanishing normal stress through the thickness of the shell, something that
is achieved by adjusting the corresponding component of the corotated velocity gradient
𝐿̂33 accordingly. The problem can be stated as to determine 𝐿̂33 such that when updating
the deformation gradient through Equation (22.27.21) and subsequently the stress through
Equation (22.27.20), 𝜎̂ 33 = 0. To this end, it is assumed that
𝐿̂33 = 𝛼(𝐿̂11 + 𝐿̂22 ), (22.27.22)
for some parameter α that is determined in the following three step procedure. In the first
(0)
two steps, 𝛼 = 0 and 𝛼 = −1, respectively, resulting in two trial normal stresses 𝜎̂ 33 and
(−1)
𝜎̂ 33 . Then it is assumed that the actual normal stress depends linearly on 𝛼, meaning that
the latter can be determined from
(𝛼) (0) (0) (−1)
0 = 𝜎33 = 𝜎33 + 𝛼(𝜎33 − 𝜎33 ). (22.27.23)

In LS-DYNA, α is given by
⎧ (0)
{ 𝜎̂ 33 (−1) (0)
{ (−1) (0)
∣𝜎̂ 33 − 𝜎̂ 33 ∣ ≥ 10−4
{𝜎̂ − 𝜎̂ 33
𝛼 = ⎨ 33 , (22.27.24)
{
{ −1 otherwise
{

and the stresses are determined from this value of α. Finally, to make sure that the normal
stress through the thickness vanishes, it is set to 0 (zero) before exiting the stress update
routine.

LS-DYNA DEV 10/27/16 (r:8004) 20-91 (Material Models)


Material Models LS-DYNA Theory Manual

22.27.2 Derivation of the Continuum Tangent Stiffness


This section will describe the derivation of the continuum tangent stiffness for the
Mooney-Rivlin material. For solid elements, the continuum tangent stiffness is chosen in
favor of an algorithmic (consistent) tangential modulus as the constitutive equation at hand
is smooth and a consistent tangent modulus is not required for good convergence
properties. For shell elements however, this stiffness must ideally be modified in order to
account for the zero normal stress condition. This modification, and its consequences, are
discussed in the next section.

The continuum tangent modulus in the reference configuration is per definition,


∂𝑆𝑖𝑗 ∂𝑆𝑖𝑗
𝐸PK
𝑖𝑗𝑘𝑙 = =2 . (22.27.25)
∂𝐸𝑘𝑙 ∂𝐶𝑘𝑙

Splitting up the differentiation of Equation (22.27.17) we get


∂(𝐼1 𝛿𝑖𝑗 − 𝐶𝑖𝑗 ) 1
= 𝛿𝑘𝑙 𝛿𝑖𝑗 − (𝛿𝑖𝑘 𝛿𝑗𝑙 + 𝛿𝑖𝑙 𝛿𝑗𝑘 ) (22.27.26)
∂𝐶𝑘𝑙 2

𝜕 ( 12 𝐶𝑖𝑗−1 )
𝐼3 2 −1 −1 1 −1 −1 (22.27.27)
= − 2 𝐶𝑘𝑙 𝐶𝑖𝑗 − 2 (𝐶𝑘𝑗 𝐶𝑖𝑙 + 𝐶𝑙𝑗−1 𝐶𝑖𝑘
−1
)
𝜕𝐶𝑘𝑙 𝐼3 2𝐼3

𝜕(𝐼3 (𝐼3 − 1)𝐶𝑖𝑗−1 ) −1 −1 1 −1 −1


= 𝐼3 (2𝐼3 − 1)𝐶𝑘𝑙 𝐶𝑖𝑗 − 𝐼3 (𝐼3 − 1)(𝐶𝑘𝑗 𝐶𝑖𝑙 + 𝐶𝑙𝑗−1 𝐶𝑖𝑘
−1
). (22.27.28)
𝜕𝐶𝑘𝑙 2

Since LS-DYNA needs the tangential modulus for the Cauchy stress, it is a good idea
to transform the terms in Equation (22.27.27) before summing them up. The push forward
pk
operation for the fourth-order tensor 𝐸𝑖𝑗𝑘𝑙 is
1
𝐸TC PK
𝑖𝑗𝑘𝑙 = 𝐹𝑖𝑎 𝐹𝑗𝑏 𝐹𝑘𝑐 𝐹𝑙𝑑 𝐸𝑎𝑏𝑐𝑑 . (22.27.29)
𝐽

Since the right Cauchy-Green tensor is 𝐂 = 𝐅T 𝐅 and the left Cauchy-Green tensor is
𝐛 = 𝐅T 𝐅, and the determinant and trace of the both stretches are equal, the transformation
is in practice carried out by interchanging
𝐶𝑖𝑗−1 → 𝛿𝑖𝑗 , (22.27.30)

𝛿𝑖𝑗 → 𝑏𝑖𝑗 . (22.27.31)

The end result is then

20-92 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

1 4C
𝐽𝐸TC
𝑖𝑗𝑘𝑙 = 4B [𝑏𝑘𝑙 𝑏𝑖𝑗 − (𝑏𝑖𝑘 𝑏𝑗𝑙 + 𝑏𝑖𝑙 𝑏𝑗𝑘 )] + 2 [4𝛿𝑖𝑗 𝛿𝑘𝑙 + (𝛿𝑘𝑗 𝛿𝑖𝑙 + 𝛿𝑙𝑗 𝛿𝑖𝑚 )] +
2 I3
(22.27.32)
1
8D𝐼3 [(2𝐼3 − 1)𝛿𝑖𝑗 𝛿𝑘𝑙 − (𝐼3 − 1)(𝛿𝑘𝑗 𝛿𝑖𝑙 + 𝛿𝑙𝑗 𝛿𝑖𝑘 )].
2

22.27.3 The Algorithmic Tangent Stiffness for Shell Elements


The corotated tangent stiffness matrix is given by Equation (22.27.32) with the
exception that the left Cauchy-Green tensor and deformation gradient are given in
corotational coordinates, i.e.,
1 ̂ ̂ 4C
𝐽𝐸̂TC ̂ ̂ ̂ ̂
𝑖𝑗𝑘𝑙 = 4B [𝑏𝑘𝑙 𝑏𝑖𝑗 − (𝑏𝑖𝑘 𝑏𝑗𝑙 + 𝑏𝑖𝑙 𝑏𝑗𝑘 )] + 2 [4𝛿𝑖𝑗 𝛿𝑘𝑙 + (𝛿𝑘𝑗 𝛿𝑖𝑙 + 𝛿𝑙𝑗 𝛿𝑖𝑚 )]
2 𝐼3
(22.27.33)
1
+ 8D𝐼3 [(2𝐼3 − 1)𝛿𝑖𝑗 𝛿𝑘𝑙 − (𝐼3 − 1)(𝛿𝑘𝑗 𝛿𝑖𝑙 + 𝛿𝑙𝑗 𝛿𝑖𝑘 )].
2

Using this exact expression for the tangent stiffness matrix in the context of shell
elements is not adequate since it does not take into account that the normal stress is zero
and it must be modified appropriately. To this end, we assume that the tangent moduli in
̂ 𝑖𝑗 to the corotated rate
Equation (22.27.33) relates the corotated rate-of-deformation tensor 𝐷

of stress 𝜎̂ 𝑖𝑗 ,

𝜎̂ 𝑖𝑗• = 𝐸̂TC ̂
𝑖𝑗𝑘𝑙 𝐷𝑘𝑙 . (22.27.34)

Even though this is not completely true, we believe that attempting a more thorough
treatment would hardly be worth the effort. The objective can now be stated as to find a
TCalg
modified tangent stiffness matrix 𝐸̂ijkl such that
•alg TCalg ̂
𝜎̂ 𝑖𝑗 = 𝐸̂𝑖𝑗𝑘𝑙 𝐷𝑘𝑙 , (22.27.35)
alg
where 𝜎̂ 𝑖𝑗 is the stress as it is evaluated in LS-DYNA. The stress update, described in
Section 19.27.1, is performed in a rather ad hoc way which probably makes the stated
objective unachievable. Still we attempt to extract relevant information from it that enables
us to come somewhat close.

An example of a modification of this tangent moduli is due to Hughes and Liu


[1981] and given by
𝐸̂TC ̂TC
𝑖𝑗33 𝐸33𝑘𝑙
TCalg
𝐸̂𝑖𝑗𝑘𝑙 = 𝐸̂TC
𝑖𝑗𝑘𝑙 − . (22.27.36)
𝐸̂TC
3333

LS-DYNA DEV 10/27/16 (r:8004) 20-93 (Material Models)


Material Models LS-DYNA Theory Manual

̂ 33 from the equation


This matrix is derived by eliminating the thickness strain 𝐷

𝜎̂ 33
= 0 in Equation (22.27.35) as an unknown. This modification is unfortunately not
consistent with how the stresses are updated in LS-DYNA. When consulting Section
̂ 33 instead can be eliminated from
19.27.1, it is suggested that 𝐷
̂ 33 = 𝛼(𝐷
𝐷 ̂ 11 + 𝐷
̂ 22 ), (22.27.37)
using the α determined from the stress update. Unfortunately, by the time when the
tangent stiffness matrix is calculated, the exact value of α is not known. From experimental
observations however, we have found that α is seldom far from being equal to −1. The fact
that α = −1 represents incompressibility strengthen this hypothesis. This leads to a
TCalg
modified tangent stiffness 𝐸̂𝑖𝑗𝑘𝑙 that is equal to 𝐸̂TC𝑖𝑗𝑘𝑙 except for the following modifica-
tions,
TCalg ̂TC ̂TC ̂TC
𝐸̂𝑖𝑖𝑗𝑗 = 𝐸̂TC
𝑖𝑖𝑗𝑗 − 𝐸𝑖𝑖33 − 𝐸33𝑗𝑗 + 𝐸3333 ,
(22.27.38)
TCalg TCalg
𝐸̂33𝑖𝑗 = 𝐸̂𝑖𝑗33 = 0, 𝑖 ≠ 𝑗.

TCalg
To preclude the obvious singularity, a small positive value is assigned to 𝐸̂3333 ,
TCalg TCalg TCalg
𝐸̂3333 = 10−4 (∣𝐸̂1111 ∣ + ∣𝐸̂2222 ∣). (22.27.39)

As with the Hughes-Liu modification, this modification preserves symmetry and


positive definiteness of the tangent moduli, which together with the stress update
“consistency” makes it intuitively attractive.

20-94 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.28 Material Model 28: Resultant Plasticity


This plasticity model, based on resultants as illustrated in Figure 22.28.1, is very cost
effective but not as accurate as through-thickness integration. This model is available only
with the C0 triangular, Belytschko-Tsay shell, and the Belytschko beam element since these
elements, unlike the Hughes-Liu elements, lend themselves very cleanly to a resultant
formulation.
(a) (b)
Membrane σy
Bending σy

Figure 22.28.1. Full section yield using resultant plasticity.

In applying this model to shell elements the resultants are updated incrementally
using the midplane strains 𝜀m and curvatures 𝜅:
Δ𝑛 = Δ𝑡𝐶𝜀m (22.28.1)

ℎ3
Δ𝑚 = Δ𝑡 𝐶𝜅, (22.28.2)
12
where the plane stress constitutive matrix is given in terms of Young’s Modulus 𝐸 and
Poisson’s ratio 𝜈 as:
̅̅̅̅̅ = 𝑚2𝑥𝑥 − 𝑚𝑥𝑥 𝑚𝑦𝑦 + 𝑚2𝑦𝑦 + 3𝑚2𝑥𝑦 .
𝑚 (22.28.3)

Defining
𝑛̅ = 𝑛2𝑥𝑥 − 𝑛𝑥𝑥 𝑛𝑦𝑦 + 𝑛2𝑦𝑦 + 3𝑛2𝑥𝑦 , (22.28.4)

̅̅̅̅̅ = 𝑚2𝑥𝑥 − 𝑚𝑥𝑥 𝑚𝑦𝑦 + 𝑚2𝑦𝑦 + 3𝑚2𝑥𝑦 ,


𝑚 (22.28.5)

1 1
𝑚
̅̅̅̅̅𝑛̅ = 𝑚𝑥𝑥 𝑛𝑥𝑥 − 𝑚𝑥𝑥 𝑛𝑦𝑦 − 𝑛𝑥𝑥 𝑚𝑦𝑦 + 𝑚𝑦 𝑛𝑦 + 3𝑚𝑥𝑦 𝑛𝑥𝑦 , (22.28.6)
2 2
the Ilyushin yield function becomes
4|𝑚
̅̅̅̅̅𝑛̅| 16𝑚̅̅̅̅̅
𝑓 (𝑚, 𝑛) = 𝑛̅ + + 2
≤ 𝑛2y = ℎ2 𝜎y2 . (22.28.7)
ℎ √3 ℎ
LS-DYNA DEV 10/27/16 (r:8004) 20-95 (Material Models)
Material Models LS-DYNA Theory Manual

In our implementation we update the resultants elastically and check to see if the yield
condition is violated:
𝑓 (𝑚, 𝑛) > 𝑛2y . (22.28.8)
If so, the resultants are scaled by the factor 𝛼:

𝑛2y
𝛼=√ . (22.28.9)
𝑓 (𝑚, 𝑛)
We update the yield stress incrementally:
𝜎y𝑛+1 = 𝜎y𝑛 + 𝐸P Δ𝜀eff
plastic , (22.28.10)

where 𝐸P is the plastic hardening modulus which in incremental plastic strain is


approximated by

√𝑓 (𝑚, 𝑛) − 𝑛y
Δ𝜀eff (22.28.11)
plastic = 𝑝 .
ℎ(3𝐺 + 𝐸 )

Kennedy, et. al., report that this model predicts results that may be too stiff; users of
this model should proceed cautiously.

In applying this material model to the Belytschko beam, the flow rule changes to
̂ 2𝑦 𝐴
4𝑚 ̂ 2𝑧 𝐴
4𝑚
𝑓 (𝑚, 𝑛) = 𝑓𝑥2̂ + + ≤ 𝑛2y = 𝐴2 𝜎y2 , (22.28.12)
3𝐼𝑦𝑦 3𝐼𝑧𝑧
have been updated elastically according to Equations (4.16)-(4.18). The yield condition is
checked with Equation (22.28.8), and if it is violated, the resultants are scaled as described
above.

This model is frequently applied to beams with non-rectangular cross sections. The
accuracy of the results obtained should be viewed with some healthy suspicion. No work
hardening is available with this model.

20-96 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.29 Material Model 29: FORCE LIMITED Resultant


Formulation
This material model is available for the Belytschko beam element only. Plastic
hinges form at the ends of the beam when the moment reaches the plastic moment. The
moment-versus-rotation relationship is specified by the user in the form of a load curve
and scale factor. The point pairs of the load curve are (plastic rotation in radians, plastic
moment). Both quantities should be positive for all points, with the first point pair being
(zero, initial plastic moment). Within this constraint any form of characteristic may be used
including flat or falling curves. Different load curves and scale factors may be specified at
each node and about each of the local s and t axes.

Axial collapse occurs when the compressive axial load reaches the collapse load.
The collapse load-versus-collapse deflection is specified in the form of a load curve. The
points of the load curve are (true strain, collapse force). Both quantities should be entered as
positive for all points, and will be interpreted as compressive i.e., collapse does not occur in
tension. The first point should be the pair (zero, initial collapse load).

The collapse load may vary with end moment and with deflection. In this case,
several load-deflection curves are defined, each corresponding to a different end moment.
Each load curve should have the same number of point pairs and the same deflection
values. The end moment is defined as the average of the absolute moments at each end of
the beam, and is always positive. It is not possible to make the plastic moment vary with
axial load.

A co-rotational technique and moment-curvature relations are used to compute the


internal forces. The co-rotational technique is treated in Section 4 in and will not be treated
here as we will focus solely on the internal force update and computing the tangent
stiffness. For this we use the notation

LS-DYNA DEV 10/27/16 (r:8004) 20-97 (Material Models)


Material Models LS-DYNA Theory Manual

M8

M7

M6

M5

M4
Axial Force

M3

M2

M1

Strain (or change in length, see AOPT)

Figure 22.29.1. The force magnitude is limited by the applied end moment. For
an intermediate value of the end moment, LS-DYNA interpolates between the
curves to determine the allowable force.

𝐸 = Young′ smodulus
𝐺 = Shear modulus
𝐴 = Cross sectional area
𝐴s = Effective area in shear
𝑙𝑛 = Reference length of beam
𝑙𝑛+1 = Current length of beam
(22.29.1)
𝐼𝑦𝑦 = Second moment of inertia about 𝑦
𝐼𝑧𝑧 = Second moment of inertia about 𝑧
𝐽 = Polar moment of inertia
𝑒𝑖 = 𝑖th local base vector in the current configuration
𝑦𝐼 = nodal vector in y direction at node I in the current configuration
𝑧𝐼 = nodal vector in z direction at node I in the current configuration

20-98 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

We emphasize that the local 𝑦 and 𝑧 base vectors in the reference configuration
always coincide with the corresponding nodal vectors. The nodal vectors in the current
configuration are updated using the Hughes-Winget formula while the base vectors are
computed from the current geometry of the element and the current nodal vectors.

22.29.1 Internal Forces


Elastic Update
In the local system for a beam connected by nodes I and J, the axial force is updated
as
𝐟ael = 𝐟an + Kael 𝛅, (22.29.2)
where
𝐸𝐴
𝐾ael = ,
𝑙𝑛 (22.29.3)
𝑛+1 𝑛
𝛿=𝑙 −𝑙 .

The torsional moment is updated as


𝑚el n el
t = 𝑚t + 𝐾t 𝜃t , (22.29.4)
where
𝐺𝐽
𝐾tel = ,
𝑙𝑛
(22.29.5)
1
𝜃t = 𝐞T1 (𝐲I × 𝐲J + 𝐳I × 𝐳J ).
2
The bending moments are updated as
𝐦𝑦el = 𝐦𝑦n + 𝐀𝑦el 𝛉𝑦 (22.29.6)

𝐦𝑧el = 𝐦𝑧n + 𝐀𝑧el 𝛉𝑧 , (22.29.7)


where
1 𝐸𝐼∗∗ 4 + 𝜑∗ 2 − 𝜑∗
𝐀∗el = [ ] (22.29.8)
1 + 𝜑∗ 𝑙n 2 − 𝜑∗ 4 + 𝜑∗

12𝐸𝐼∗∗
𝜑∗ = (22.29.9)
𝐺𝐴s ln ln

𝛉Ty = −𝐞T3 (𝐲I × 𝐳I 𝐲J × 𝐳J ) (22.29.10)

𝛉Tz = 𝐞T2 (𝐲I × 𝐳I 𝐲J × 𝐳J ). (22.29.11)

LS-DYNA DEV 10/27/16 (r:8004) 20-99 (Material Models)


Material Models LS-DYNA Theory Manual

In the following we refer to 𝐀∗el as the (elastic) moment-rotation matrix.

Plastic Correction
After the elastic update the state of force is checked for yielding as follows. As a
preliminary note we emphasize that whenever yielding does not occur the elastic
stiffnesses and forces are taken as the new stiffnesses and forces.

The yield moments in direction 𝑖 at node 𝐼 as functions of plastic rotations are


denoted 𝑚Y P
𝑖𝐼 (𝜃𝑖𝐼 ). This function is given by the user but also depends on whether a plastic
hinge has been created. The theory for plastic hinges is given in the LS-DYNA Keyword
User’s Manual [Hallquist 2003] and is not treated here. Whenever the elastic moment
exceeds the plastic moment, the plastic rotations are updated as
∣𝑚el Y
𝑖𝐼 ∣ − 𝑚𝑖𝐼
𝜃𝑖𝐼P(𝑛+1) = 𝜃𝑖𝐼P(𝑛) + ,
∂𝑚Y (22.29.12)
max (0.001, 𝐴el
𝑖(𝐼𝐼) + P𝑖𝐼 )
∂𝜃𝑖𝐼
and the moment is reduced to the yield moment
P(𝑛+1)
𝑚𝑛+1
𝑖𝐼 = 𝑚Y
𝑖𝐼 (𝜃𝑖𝐼 )sgn(𝑚el
𝑖𝐼 ). (22.29.13)

The corresponding diagonal component in the moment-rotation matrix is reduced as


⎜ el


el ⎜
⎜ 𝐴 𝑖(𝐼𝐼)


𝐴𝑛+1
𝑖(II) = 𝐴𝑖(𝐼𝐼) ⎜

⎜ 1 − 𝛼 Y ⎟
⎟,
⎟ (22.29.14)

⎜ ∂𝑚 𝑖𝐼 ⎟⎟
max (0.001, 𝐴el
𝑖(𝐼𝐼) + )
⎝ ∂𝜃𝑖𝐼P ⎠
where α ≤ 1 is a parameter chosen such that the moment-rotation matrix remains positive
definite.

The yield moment in torsion is given by 𝑚Y P


t (𝜃t ) and is provided by the user. If the
elastic torsional moment exceeds this value, the plastic torsional rotation is updated as
∣𝑚el Y
t ∣ − 𝑚t
𝜃tP(𝑛+1) = 𝜃tP(𝑛) + ,
∂𝑚Y (22.29.15)
max (0.001, 𝐾tel + Pt )
∂𝜃t
and the moment is reduced to the yield moment
P(𝑛+1)
𝑚𝑛+1
t = 𝑚Y
t (𝜃t )sgn(𝑚el
t ). (22.29.16)

The torsional stiffness is modified as

20-100 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models


⎜ ⎞


⎜ 𝐾tel ⎟

𝐾tn+1 = 𝐾tel ⎜
⎜1
⎜ −α Y

⎟,
⎟ (22.29.17)

⎜ ∂𝑚 t ⎟
𝐾t + P ⎟
el
⎝ ∂𝜃t ⎠
where again α ≤ 1 is chosen so that the stiffness is positive.

Axial collapse is modeled by limiting the axial force by 𝑓aY (𝜀, 𝑚), i.e., a function of the
axial strains and the magnitude of bending moments. If the axial elastic force exceeds this
value it is reduced to yield
𝑓a𝑛+1 = 𝑓aY (𝜀𝑛+1 , 𝑚𝑛+1 )sgn(𝑓ael ), (22.29.18)
and the axial stiffness is given by
Y
el ∂𝑓a
𝐾a𝑛+1 = max (0.05𝐾a , ). (22.29.19)
∂𝜀
We neglect the influence of change in bending moments when computing this parameter.

Damping
Damping is introduced by adding a viscous term to the internal force on the form
δ
𝑑⎡ 𝜃t ⎤
𝐟v = 𝐃 ⎢ ⎢ ⎥,
⎥ (22.29.20)
𝑑𝑡 ⎢𝜃𝑦 ⎥
⎣𝜃𝑧 ⎦
el
⎡𝐾a ⎤
⎢ 𝐾tel ⎥
𝐃 = 𝛾⎢

⎥,
⎥ (22.29.21)
⎢ 𝐴el
𝑦 ⎥
⎣ 𝐴el
𝑧 ⎦

where γ is a damping parameter.

Transformation
The internal force vector in the global system is obtained through the transformation
𝐟g𝑛+1 = 𝐒𝐟l𝑛+1 , (22.29.22)
where
−𝑒 0 −𝑒3 /𝑙𝑛+1 −𝑒3 /𝑙𝑛+1 𝑒2 /𝑙𝑛+1 𝑒2 /𝑙𝑛+1
⎡ 1 ⎤
⎢ 0 −𝑒1 𝑒2 0 𝑒3 0 ⎥
𝐒=⎢ 𝑛+1 ⎥
, (22.29.23)
⎢ 𝑒1 0 𝑒3 /𝑙𝑛+1 𝑒3 /𝑙𝑛+1 −𝑒2 /𝑙𝑛+1 −𝑒2 /𝑙 ⎥
⎣ 0 𝑒1 0 𝑒2 0 𝑒3 ⎦

LS-DYNA DEV 10/27/16 (r:8004) 20-101 (Material Models)


Material Models LS-DYNA Theory Manual

f 𝑛+1
⎡ a 𝑛+1 ⎤
⎢ 𝑚t ⎥
𝐟l𝑛+1 =⎢
⎢𝑚𝑛+1
⎥.
⎥ (22.29.24)
⎢ y ⎥
⎣𝑚𝑛+1z ⎦

22.29.2 Tangent Stiffness


Derivation
The tangent stiffness is derived from taking the variation of the internal force
δ𝐟g𝑛+1 = δ𝐒𝐟l𝑛+1 + 𝐒δ𝐟l𝑛+1 , (22.29.25)
which can be written
δ𝐟g𝑛+1 = 𝐊geo δ𝐮 + 𝐊mat δ𝐮, (22.29.26)
where
T
δ𝐮 = [δx𝐼T δωT𝐼 δx𝐽T δωT𝐽 ] . (22.29.27)

There are two contributions to the tangent stiffness, one geometrical and one
material contribution. The geometrical contribution is given (approximately) by
1
𝐊geo = 𝐑(𝐟l𝑛+1 ⊗ 𝐈)𝐖 − 𝐓𝐟l𝑛+1 𝐋, (22.29.28)
l𝑛+1 l𝑛+1
where
𝑅 0 𝑅3 /𝑙𝑛+1 𝑅3 /𝑙𝑛+1 −𝑅2 /𝑙𝑛+1 −𝑅2 /𝑙𝑛+1
⎡ 1 ⎤
⎢ 0 𝑅1 −𝑅2 0 −𝑅3 0 ⎥
𝐑=⎢ 𝑛+1 ⎥
, (22.29.29)
⎢−𝑅1 0 −𝑅3 /𝑙𝑛+1 −𝑅3 /𝑙𝑛+1 𝑅2 /𝑙𝑛+1 𝑅2 /𝑙 ⎥
⎣ 0 −𝑅1 0 −𝑅2 0 −𝑅3 ⎦

𝐖 = [−𝑅1 /𝑙𝑛+1 𝐞1 𝐞T1 /2 𝑅1 /𝑙𝑛+1 𝐞1 𝐞T1 /2], (22.29.30)

0 0 −𝑒3 −𝑒3 𝑒2 𝑒2
⎡0 0 0 0 0 0 ⎤
𝐓=⎢
⎢0
⎥, (22.29.31)
0 𝑒3 𝑒3 −𝑒2 −𝑒2 ⎥
⎣0 0 0 0 0 0 ⎦

𝐋 = [−𝐞T1 0 𝐞T1 0], (22.29.32)


and 𝐈 is the 3 by 3 identity matrix. We use ⊗ as the outer matrix product and define
Ri v = 𝐞i × 𝐯. (22.29.33)

The material contribution can be written as

20-102 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

𝐊mat = 𝐒𝐊𝐒T , (22.29.34)


where
𝐾 𝑛+1
⎡ a ⎤
⎢ 𝐾t𝑛+1 ⎥ 1
𝐊=⎢

⎥ + 𝐃.
⎥ Δ𝑡 (22.29.35)
⎢ 𝐴𝑛+1
y ⎥
⎣ 𝐴𝑛+1
z ⎦

LS-DYNA DEV 10/27/16 (r:8004) 20-103 (Material Models)


Material Models LS-DYNA Theory Manual

22.30 Material Model 30: Closed-Form Update Shell


Plasticity
This section presents the mathematical details of the shape memory alloy material in
LS-DYNA. The description closely follows the one of Auricchio and Taylor [1997] with
appropriate modifications for this particular implementation.

22.30.1 Mathematical Description of the Material Model


The Kirchhoff stress 𝛕 in the shape memory alloy can be written
𝛕 = 𝑝𝐢 + 𝐭, (22.30.1)
where 𝐢 is the second order identity tensor and
𝑝 = 𝐾(𝜃 − 3α𝜉S 𝜀L ), (22.30.2)

𝐭 = 2𝐺(𝐞 − 𝜉S 𝜀L 𝐧). (22.30.3)

Here 𝐾 and 𝐺 are bulk and shear modulii, 𝜃 and e are volumetric and shear
logarithmic strains and 𝛼 and 𝜀L are constant material parameters. There is an option to
define the bulk and shear modulii as functions of the martensite fraction according to
𝐾 = 𝐾A + 𝜉S (𝐾S − 𝐾A ),
(22.30.4)
𝐺 = 𝐺A + 𝜉S (𝐺S − 𝐺A ),
in case the stiffness of the martensite differs from that of the austenite. Furthermore, the
unit vector 𝐧 is defined as
𝐧 = 𝐞/(‖𝐞‖ + 10−12 ), (22.30.5)
and a loading function is introduced as
𝐹 = 2𝐺‖𝐞‖ + 3𝛼𝐾𝜃 − 𝛽𝜉S , (22.30.6)
where
𝛽 = (2𝐺 + 9𝛼2 𝐾)𝜀L . (22.30.7)

For the evolution of the martensite fraction 𝜉S in the material, the following rule is
adopted
𝐹 − 𝑅ASs > 0⎫
} 𝐹̇
𝐹̇ > 0 ⎬ ⇒ 𝜉Ṡ = −(1 − 𝜉S ) (22.30.8)
𝜉S < 1 } 𝐹 − 𝑅AS
⎭ f

20-104 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

𝐹 − 𝑅SAs < 0⎫
} 𝐹̇
𝐹̇ < 0 ⎬ ⇒ 𝜉Ṡ = 𝜉S . (22.30.9)
𝜉S > 0 } 𝐹 − 𝑅SA
⎭ f

AS SA
Here 𝑅AS SA
s , 𝑅f , 𝑅s and 𝑅f are constant material parameters. The Cauchy stress is
finally obtained as
𝛕
𝛔= , (22.30.10)
𝐽
where 𝐽 is the Jacobian of the deformation.

22.30.2 Algorithmic Stress Update


For the stress update we assume that the martensite fraction 𝜉S𝑛 and the value of the
loading function F𝑛 is known from time 𝑡𝑛 and the deformation gradient at time 𝑡𝑛+1 , F, is
known. We form the left Cauchy-Green tensor as 𝐁 = 𝐅𝐅T which is diagonalized to obtain
the principal values and directions 𝚲 = diag(λi ) and 𝐐. The volumetric and principal
shear logarithmic strains are given by
𝜃 = log(𝐽) ,

⎛𝜆𝑖 ⎞ (22.30.11)
𝑒𝑖 = log ⎜

⎜ 1⎟
⎟,

⎝𝐽 3 ⎠
where
𝐽 = 𝜆1 𝜆2 𝜆3 . (22.30.12)
is the total Jacobian of the deformation. Using Equation (22.30.6) with 𝜉S = 𝜉S𝑛 , a value𝐹trial
of the loading function can be computed. The discrete counterpart of Equation (22.30.8)
becomes
𝐹trial − 𝑅AS
s > 0⎫
}
𝐹 trial
− 𝐹 > 0 ⎬ ⇒ Δ𝜉S
n

𝜉Sn < 1 }

(22.30.13)
trial AS
𝐹 − 𝛽Δ𝜉S − min(max(𝐹 , 𝑅AS
𝑛
s ), 𝑅f )
= −(1 − 𝜉S𝑛 − Δ𝜉S )
𝐹trial − 𝛽Δ𝜉S − 𝑅AS
f

𝐹trial − 𝑅SA
s < 0⎫ 𝐹trial − 𝛽Δ𝜉S − min(max(𝐹n , 𝑅SA SA
} 𝑛 f ) , 𝑅s )
𝐹 trial n
−𝐹 <0 ⎬ ⇒ Δ𝜉S = (𝜉S + Δ𝜉𝑆 ) . (22.30.14)
n } 𝐹trial − 𝛽Δ𝜉S − 𝑅SA
f
𝜉S > 0 ⎭

If none of the two conditions to the left are satisfied, set 𝜉S𝑛+1 = 𝜉S𝑛 , 𝐹𝑛+1 = 𝐹trial and
compute the stress 𝜎 𝑛+1 using Equations (22.30.1), (22.30.2), (22.30.5), (22.30.10) and 𝜉S =

LS-DYNA DEV 10/27/16 (r:8004) 20-105 (Material Models)


Material Models LS-DYNA Theory Manual

𝜉S𝑛 . When phase transformation occurs according to a condition to the left, the
corresponding equation to the right is solved for Δ𝜉S . If the bulk and shear modulii are
constant this is an easy task. Otherwise 𝐹trial as well as 𝛽 depends on this parameter and
makes things a bit more tricky. We have that
𝐸 S − 𝐸A
𝐹trial = 𝐹ntrial (1 + Δ𝜉S ) ,
𝐸n
(22.30.15)
𝐸 − 𝐸A
𝛽 = 𝛽n (1 + S Δ𝜉S ),
𝐸n
where 𝐸S and 𝐸A are Young’s modulii for martensite and austenite, respectively. The
subscript 𝑛 is introduced for constant quantities evaluated at time 𝑡𝑛 . To simplify the
upcoming expressions, these relations are written
𝐹trial = 𝐹ntrial + Δ𝐹trial Δ𝜉S ,
(22.30.16)
𝛽 = 𝛽n + Δ𝛽Δ𝜉S .

Inserting these expressions into Equation (19.30.7) results in

𝑓 (Δ𝜉S ) = Δ𝛽(1 − 𝜉S𝑛 )Δ𝜉S2 + (𝑅AS ̃𝑛


f − 𝐹AS + (𝛽n − Δ𝐹
trial
)(1 − 𝜉S𝑛 )) Δ𝜉S +
(22.30.17)
(1 − 𝜉S𝑛 )(𝐹AS
̃𝑛 − 𝐹𝑛trial ) = 0.

and
̃𝑛 − 𝑅SA
𝑓 (Δ𝜉S ) = Δ𝛽𝜉S𝑛 Δ𝜉S2 + (𝐹SA f + (𝛽𝑛 − Δ𝐹
trial 𝑛
)𝜉S )Δ𝜉S +
𝑛 ̃𝑛 trial
(22.30.18)
𝜉S (𝐹SA − 𝐹𝑛 ) = 0.
respectively, where we have for simplicity set
̃𝑛 = min(max(𝐹𝑛 , 𝑅AS AS
𝐹AS s ) , 𝑅f ) ,
(22.30.19)
̃𝑛 = min(max(𝐹𝑛 , 𝑅SA
𝐹SA SA
f ), 𝑅s ).

The solutions to these equations are approximated with two Newton iterations
starting in the point Δ𝜉S = 0. Now set 𝜉S𝑛+1 = min(1, max(0, 𝜉S𝑛 + Δ𝜉S )) and compute 𝜎 𝑛+1
and 𝐹𝑛+1 according to Equations (22.30.1), (22.30.2), (22.30.5), (22.30.6), (22.30.10) and 𝜉S =
𝜉S𝑛+1 .

22.30.3 Tangent Stiffness Matrix


An algorithmic tangent stiffness matrix relating a change in true strain to a
corresponding change in Kirchhoff stress is derived in the following. Taking the variation
of Equation (22.30.2) results in
δ𝑝 = 𝐾(δ𝜃 − 3𝛼δ𝜉S 𝜀L ) + δ𝐾(𝜃 − 3𝛼𝜉S 𝜀L ), (22.30.20)

20-106 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

δ𝐭 = 2𝐺(δ𝐞 − δ𝜉S 𝜀L 𝐧 − 𝜉S 𝜀L δ𝐧) + 2δ𝐺(𝐞 − 𝜉S 𝜀L 𝐧). (22.30.21)

The variation of the unit vector in Equation (22.30.5) can be written


1
δ𝐧 = (𝐈 − 𝐧 ⊗ 𝐧)δ𝐞, (22.30.22)
‖𝐞‖ + 10−12
where 𝐈 is the fourth order identity tensor. For the variation of martensite fraction we
introduce the indicator parameters 𝐻 AS and 𝐻 SA that should give information of the
probability of phase transformation occurring in the next stress update. Set initially𝐻 AS =
𝐻 SA = 0 and change them according to
𝐹trial − 𝑅AS
s > 0⎫}
𝐹trial − 𝐹𝑛 > 0 ⎬ ⇒ 𝐻 AS = 1, (22.30.23)
𝜉S𝑛 + Δ𝜉S ≤ 1 } ⎭

𝐹trial − 𝑅SA
s < 0⎫}
𝐹trial − 𝐹𝑛 < 0 ⎬ ⇒ 𝐻 SA = 1, (22.30.24)
𝜉S𝑛 + Δ𝜉S ≥ 0 } ⎭
using the quantities computed in the previous stress update. For the variation of the
martensite fraction we take variations of Equations (22.30.17) and (22.30.18) with
δ𝐹ntrial = 2𝐺𝐧: δ𝐞 + 3𝛼𝐾δ𝜃, (22.30.25)
which results in
δ𝜉S = γ(2𝐺𝐧: δ𝐞 + 3𝛼𝐾δ𝜃), (22.30.26)
where
(1 − 𝜉S𝑛 )𝐻 AS 𝜉S𝑛 𝐻 SA
𝛾= + . (22.30.27)
𝑅AS ̃𝑛 trial 𝑛 ̃𝑛 − 𝑅SA + (𝛽𝑛 − Δ𝐹𝑛trial )𝜉S𝑛
f − 𝐹AS + (𝛽𝑛 − Δ𝐹𝑛 )(1 − 𝜉S ) 𝐹SA f

As can be seen, we use the value of 𝛾 obtained in the previous stress update since
this is easier to implement and will probably give a good indication of the current value of
this parameter.

The variation of the material parameters 𝐾 and 𝐺 results in


δK = (𝐾S − 𝐾A )δ𝜉S ,
(22.30.28)
δG = (𝐺S − 𝐺A )δ𝜉S ,
and, finally, using the identities
𝐧: δ𝐞 = 𝐧: δ𝛆, (22.30.29)

δ𝜃 = 𝐢: δ𝛆, (22.30.30)

δ𝛕 = 𝐢δ𝑝 + δ𝐭, (22.30.31)


LS-DYNA DEV 10/27/16 (r:8004) 20-107 (Material Models)
Material Models LS-DYNA Theory Manual

results in
𝜉S 𝜀L
δ𝛕 = {2𝐺 (1 − ) 𝐈dev
‖𝐞‖ + 10−12

+ 𝐾[1 − 9𝛼2 𝐾𝛾𝜀L + 3𝛼𝛾(𝐾S − 𝐾A )(𝜃 − 3𝛼𝜉S 𝜀L )]𝐢 ⊗ 𝐢

+ 2γ𝐺(𝐾S − 𝐾A )(𝜃 − 3𝛼𝜉S 𝜀L )𝐢 ⊗ 𝐧

+ 6𝛾𝛼𝐾(𝐺S − 𝐺A )(‖𝐞‖ − 𝜉S 𝜀L )𝐧 ⊗ 𝐢 (22.30.32)

𝜉S 𝜀L
+ 2𝐺 [ − 2𝐺𝛾𝜀L + 2𝛾(𝐺S − 𝐺A )(‖𝐞‖ − 𝜉S 𝜀L )] 𝐧 ⊗ 𝐧
‖𝐞‖ + 10−12

− 6𝐾𝐺𝛼𝛾𝜀L (𝐢 ⊗ 𝐧 + 𝐧 ⊗ 𝐢)} δ𝜀.

where 𝐈dev is the fourth order deviatoric identity tensor. In general this tangent is not
symmetric because of the terms on the second line in the expression above. We simply use
a symmetrization of the tangent stiffness above in the implementation. Furthermore, we
transform the tangent to a tangent closer related to the one that should be used in the LS-
DYNA implementation,
𝜉S 𝜀L
𝐂 = 𝐽 −1 {2𝐺 (1 − ) 𝐈dev
‖𝐞‖ + 10−12

+ 𝐾[1 − 9𝛼2 𝐾𝛾𝜀L + 3𝛼𝛾(𝐾S − 𝐾A )(𝜃 − 3𝛼𝜉S 𝜀L )]𝐢 ⊗ 𝐢


(22.30.33)
+ 3𝛾𝛼𝐾(𝐺S − 𝐺A )(‖𝐞‖ − 𝜉S 𝜀L ) − 6𝐾𝐺𝛼𝛾𝜀L ))(𝐢 ⊗ 𝐧 + 𝐧 ⊗ 𝐢)

𝜉S 𝜀L
+ 2G [ − 2𝐺𝛾𝜀L + 2𝛾(𝐺S − 𝐺A )(‖𝐞‖ − 𝜉S 𝜀L )] 𝐧 ⊗ 𝐧} δ𝜀.
‖𝐞‖ + 10−12

20-108 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.31 Material Model 31: Slightly Compressible Rubber


Model
This model implements a modified form of the hyperelastic constitutive law first
described in [Kenchington 1988].

The strain energy functional, 𝑈, is defined in terms of the input constants as:
𝑈 = C100 𝐼1 + C200 𝐼12 + C300 𝐼13 + C400 𝐼14 + C110 𝐼1 𝐼2 +
(22.31.1)
C210 𝐼12 𝐼2 + C010 𝐼2 + C020 𝐼22 + 𝑓 (𝐽),
where the strain invariants can be expressed in terms of the deformation gradient matrix,
𝐹𝑖𝑗 , and the Green-St. Venant strain tensor, 𝐸𝑖𝑗 :
𝐽 = ∣𝐹𝑖𝑗 ∣
𝐼1 = 𝐸𝑖𝑖 (22.31.2)
1 𝑖𝑗
𝐼2 = 𝛿𝑝𝑞 𝐸𝑝𝑖 𝐸𝑞𝑗 .
2!

The derivative of 𝑈 with respect to a component of strain gives the corresponding


component of stress
∂𝑈
𝑆𝑖𝑗 = , (22.31.3)
∂𝐸𝑖𝑗
where, 𝑆𝑖𝑗 , is the second Piola-Kirchhoff stress tensor which is transformed into the Cauchy
stress tensor:
𝜌 𝜕𝑥𝑖 𝜕𝑥𝑗
𝜎𝑖𝑗 = 𝑆 , (22.31.4)
𝜌0 𝜕𝑋𝑘 𝜕𝑋𝑙 𝑘𝑙
where 𝜌0 and 𝜌 are the initial and current density, respectively.

LS-DYNA DEV 10/27/16 (r:8004) 20-109 (Material Models)


Material Models LS-DYNA Theory Manual

22.32 Material Model 32: Laminated Glass Model


This model is available for modeling safety glass. Safety glass is a layered material
of glass bonded to a polymer material which can undergo large strains.

The glass layers are modeled by isotropic hardening plasticity with failure based on
exceeding a specified level of plastic strain. Glass is quite brittle and cannot withstand
large strains before failing. Plastic strain was chosen for failure since it increases
monotonically and, therefore, is insensitive to spurious numerical noise in the solution.

The material to which the glass is bonded is assumed to stretch plastically without
failure. The user defined integration rule option must be used with this material. The user
defined rule specifies the thickness of the layers making up the safety glass. Each
integration point is flagged with a zero if the layer is glass and with a one if the layer is
polymer.

An iterative plane stress plasticity algorithm is used to enforce the plane stress
condition.

20-110 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.33 Material Model 33: Barlat’s Anisotropic Plasticity


Model

This model was developed by Barlat, Lege, and Brem [1991] for modeling material
behavior in forming processes. The finite element implementation of this model is
described in detail by Chung and Shah [1992] and is used here.

The yield function 𝛷 is defined as


𝛷 = |𝑆1 − 𝑆2 |𝑚 + |𝑆2 − 𝑆3 |𝑚 + |𝑆3 − 𝑆1 |𝑚 = 2𝑚 , (22.33.1)
where 𝜎
̅̅̅̅ is the effective stress, and 𝑆𝑖 for 𝑖 = 1, 2, 3 are the principal values of the symmetric
matrix 𝑆𝛼𝛽 ,
[𝑐(𝜎𝑥𝑥 − 𝜎𝑦𝑦 ) − 𝑏(𝜎𝑧𝑧 − 𝜎𝑥𝑥 )]
𝑆𝑥𝑥 =
3
[𝑎(𝜎𝑦𝑦 − 𝜎𝑧𝑧 ) − 𝑐(𝜎𝑥𝑥 − 𝜎𝑦𝑦 )]
𝑆𝑦𝑦 =
3
[𝑏(𝜎𝑧𝑧 − 𝜎𝑥𝑥 ) − 𝑎(𝜎𝑦𝑦 − 𝜎𝑧𝑧 )] (22.33.2)
𝑆𝑧𝑧 =
3
𝑆𝑦𝑧 = 𝑓 𝜎𝑦𝑧
𝑆𝑧𝑥 = 𝑔𝜎𝑧𝑥
𝑆𝑥𝑦 = ℎ𝜎𝑥𝑦 .

The material constants 𝑎, 𝑏, 𝑐, 𝑓 , 𝑔 and ℎ represent anisotropic properties. When 𝑎 =


𝑏 = 𝑐 = 𝑓 = 𝑔 = ℎ = 1, the material is isotropic and the yield surface reduces to the Tresca
yield surface for 𝑚 = 1 and von Mises yield surface for 𝑚 = 2 or 4. For face-centered-cubic
(FCC) materials 𝑚 = 8 is recommended and for body-centered-cubic (BCC) materials𝑚 = 6
is used.

The yield strength of the material is


𝜎y = 𝑘(1 + 𝜀0 )𝑛 . (22.33.3)

LS-DYNA DEV 10/27/16 (r:8004) 20-111 (Material Models)


Material Models LS-DYNA Theory Manual

22.34 Material Model 34: Fabric


The fabric model is a variation on the Layered Orthotropic Composite material
model (Material 22) and is valid for only 3 and 4 node membrane elements. This material
model is strongly recommended for modeling airbags and seatbelts. In addition to being a
constitutive model, this model also invokes a special membrane element formulation that is
better suited to the large deformations experienced by fabrics. For thin fabrics, buckling
(wrinkling) can occur with the associated inability of the structure to support compressive
stresses; a material parameter flag is included for this option. A linear elastic liner is also
included which can be used to reduce the tendency for these material/elements to be
crushed when the no-compression option is invoked.

If the airbag material is to be approximated as an isotropic elastic material, then only


one Young’s modulus and Poisson’s ratio should be defined. The elastic approximation is
very efficient because the local transformations to the material coordinate system may be
skipped. If orthotropic constants are defined, it is very important to consider the
orientation of the local material system and use great care in setting up the finite element
mesh.

If the reference configuration of the airbag is taken as the folded configuration, the
geometrical accuracy of the deployed bag will be affected by both the stretching and the
compression of elements during the folding process. Such element distortions are very
difficult to avoid in a folded bag. By reading in a reference configuration such as the final
unstretched configuration of a deployed bag, any distortions in the initial geometry of the
folded bag will have no effect on the final geometry of the inflated bag. This is because the
stresses depend only on the deformation gradient matrix:
𝜕𝑥𝑖
𝐹𝑖𝑗 = , (22.34.1)
𝜕𝑋𝑗
where the choice of 𝑋𝑗 may coincide with the folded or unfold configurations. It is this
unfolded configuration which may be specified here. When the reference geometry is used
then the no-compression option should be active. With the reference geometry it is
possible to shrink the airbag and then perform the inflation. Although the elements in the
shrunken bag are very small, the time step can be based on the reference geometry so a
very reasonable time step size is obtained. The reference geometry based time step size is
optional in the input.

The parameters fabric leakage coefficient, FLC, fabric area coefficient, FAC, and
effective leakage area, ELA, for the fabric in contact with the structure are optional for the
Wang-Nefske and hybrid inflation models. It is possible for the airbag to be constructed of
multiple fabrics having different values of porosity and permeability. The gas leakage
through the airbag fabric then requires an accurate determination of the areas by part ID

20-112 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

available for leakage. The leakage area may change over time due to stretching of the
airbag fabric or blockage when the outer surface of the bag is in contact with the structure.
LS-DYNA can check the interaction of the bag with the structure and split the areas into
regions that are blocked and unblocked depending on whether the regions are in contact or
not, respectively. Typically, the parameters, FLC and FAC, must be determined
experimentally and their variation with time and pressure are optional inputs that allow
for maximum modeling flexibility.

LS-DYNA DEV 10/27/16 (r:8004) 20-113 (Material Models)


Material Models LS-DYNA Theory Manual

22.35 Material Model 35: Kinematic/Isotropic Plastic Green-


Naghdi Rate
The reader interested in an detailed discussion of the relative merits of various stress
rates, especially Jaumann [1911] and Green-Naghdi [1965], is urged to read the work of
Johnson and Bammann [1984]. A mathematical description of these two stress rates, and
how they are implemented in LS-DYNA, is given in the section entitled Stress Update
Overview in this manual.

In the cited work by Johnson and Bammann, they conclude that the Green-Naghdi
stress rate is to be preferred over all other proposed stress rates, including the most widely
used Jaumann rate, because the Green-Naghdi stress rate is based on the notions of
invariance under superimposed rigid-body motions. However, implementation of the
Green-Naghdi stress rate comes at a significant computational cost compared to the
Jaumann stress rate, e.g., see the discussion in this manual in the section entitled Green-
Naghdi Stress Rate.

Perhaps more importantly, in practical applications there is little if any noted


difference in the results generated using either Jaumann or Green-Naghdi stress rate. This
is in part due to the fact that the Jaumann stress rate only produces erroneous results2
when linear kinematic hardening is used; the results for isotropic hardening are not
affected by the choice of either of these two stress rates. Also in practical applications, the
shear strains are rather small, compared to the extensional strains, and if they are not small
it is usually the case that the material description, i.e., constitutive model, is not valid for
such large shear strains.

2 The results of a simple shear simulation, monotonically increasing shear deformation, produce sinusoidal
stress response.
20-114 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Material Models

22.36 Material Model 36: Barlat’s 3-Parameter Plasticity


Model
Material model 36 in LS-DYNA aims at modeling sheets with anisotropy under
plane stress conditions. It allows the use of the Lankford parameters for the definition of
the anisotropic yield surface. The yield condition can be written
𝑓 (𝛔, 𝜀p ) = 𝜎eff (𝜎11 , 𝜎22 , 𝜎12 ) − 𝜎Y (𝜀p ) ≤ 0, (22.36.1)
where
𝑎 𝑎 𝑐 1/𝑚
𝜎eff (𝜎11 , 𝜎22 , 𝜎12 ) = ( |𝐾1 + 𝐾2 |𝑚 + |𝐾1 − 𝐾2 |𝑚 + |2𝐾2 |𝑚 )
2 2 2
𝜎 + ℎ𝜎22
𝐾1 = 𝐾1 (𝜎11 , 𝜎22 , 𝜎12 ) = 11 (22.36.2)
2
2
𝜎11 − ℎ𝜎22
𝐾2 = 𝐾2 (𝜎11 , 𝜎22 , 𝜎12 ) = √( 2
) + 𝑝2 𝜎12 ,
2
and the hardening of the yield surface is either linear, exponential or determined by a load
curve. In the above, the stress components 𝜎11 , 𝜎22 and 𝜎12 are with respect to the material
coordinate system and 𝜀p denotes the effective plastic strain. The material parameters 𝑎, 𝑐,
ℎ and 𝑝 can be determined from the Lankford parameters as described in the LS-DYNA
Keyword User’s Manual [Hallquist 2003]. The Lankford parameters, or R-values, are
defined as the ratio of instantaneous width change to instantaneous thickness change. That
is, assume that the width 𝑊 and thickness 𝑇 are measured as function of strain. Then the
corresponding R-value is given by

𝑑𝑊
𝑑𝜀
/𝑊
𝑅= 𝑑𝑇
.
𝑑𝜀
/𝑇

The gradient of the yield surface is denoted

LS-DYNA DEV 10/27/16 (r:8004) 20-115 (Material Models)


Material Models LS-DYNA Theory Manual

∂𝑓 𝜕𝑓

⎜ (𝛔)⎞⎟ ⎛
⎜ (𝜎 , 𝜎 , 𝜎 )⎞

⎜ ∂𝜎11 ⎟
⎟ ⎜
⎜ 𝜕𝜎11 11 22 12 ⎟ ⎟


⎜ ∂𝑓 ⎟
⎟ ⎜
⎜ 𝜕𝑓 ⎟


⎜ ⎟

(𝛔)⎟ ⎜ ⎜
⎜ (𝜎11 , 𝜎22 , 𝜎12 )⎟⎟

⎜ ∂𝜎22 ⎟ ⎜ 𝜕𝜎22 ⎟


⎜ ⎟
⎟ ⎜
⎜ ⎟

∂𝑓 ⎜
⎜ ∂𝑓 ⎟
⎟ ⎜
⎜ 𝜕𝑓 ⎟

(𝛔) = ⎜

⎜ (𝛔) ⎟

⎟ = ⎜

⎜ (𝜎11 , 𝜎22 , 𝜎12 ) ⎟

⎟ , (22.36.3)
∂𝛔 ⎜ ∂𝜎12 ⎟ ⎜ 𝜕𝜎12 ⎟

⎜ ⎟
⎟ ⎜
⎜ ⎟


⎜ ⎟
⎟ ⎜
⎜ ⎟


⎜ 0 ⎟
⎟ ⎜
⎜ 0 ⎟


⎜ ⎟
⎟ ⎜
⎜ ⎟


⎜ ⎟
⎟ ⎜
⎜ ⎟

⎜ 0 ⎟ ⎜ 0 ⎟
⎝ ⎠ ⎝ ⎠
where
𝜕𝑓 ∂𝜎 𝜎 (𝜎 , 𝜎 , 𝜎 )1−𝑚
(𝜎11 , 𝜎22 , 𝜎12 ) = eff (𝜎11 , 𝜎22 , 𝜎12 ) = eff 11 22 12 ⋅
𝜕𝜎11 ∂𝜎11 2
1 𝜎 − ℎ𝜎22
{𝑎(𝐾1 − 𝐾2 )|𝐾1 − 𝐾2 |𝑚−2 ( − 11 )+
2 4𝐾2
(22.36.4)
1 𝜎 − ℎ𝜎22
𝑎(𝐾1 + 𝐾2 )|𝐾1 + 𝐾2 |𝑚−2 ( + 11 )+
2 4𝐾2
𝜎 − ℎ𝜎22
𝑐2𝑚 𝐾2𝑚−1 11 },
4𝐾2

𝜕𝑓 𝜕𝜎eff 𝜎eff (𝜎11 , 𝜎22 , 𝜎12 )1−𝑚


(𝜎 , 𝜎 , 𝜎 ) = (𝜎 , 𝜎 , 𝜎 ) = ℎ⋅
𝜕𝜎22 11 22 12 𝜕𝜎22 11 22 12 2
1 𝜎 − ℎ𝜎22
{𝑎(𝐾1 − 𝐾2 )|𝐾1 − 𝐾2 |𝑚−2 ( + 11 )+
2 4𝐾2
(22.36.5)
1 𝜎 − ℎ𝜎22
𝑎(𝐾1 + 𝐾2 )|𝐾1 + 𝐾2 |𝑚−2 ( − 11 )−
2 4𝐾2
𝜎 − ℎ𝜎22
𝑐2𝑚 𝐾2𝑚−1 11 },
4𝐾2
and
𝜕𝑓 𝜕𝜎 𝜎 (𝜎 , 𝜎 , 𝜎 )1−𝑚 𝑝2 𝜎12
(𝜎11 , 𝜎22 , 𝜎12 ) = eff (𝜎11 , 𝜎22 , 𝜎12 ) = eff 11 22 12 ⋅
𝜕𝜎12 𝜕𝜎12 2 𝐾2 (22.36.6)
{−𝑎(𝐾1 − 𝐾2 )|𝐾1 − 𝐾2 |𝑚−2 + 𝑎(𝐾1 + 𝐾2 )|𝐾1 + 𝐾2 |𝑚−2 + 𝑐2𝑚 𝐾2𝑚−1 }.

22.36.1 Material Tangent Stiffness


Since the plastic model is associative, the general expression for tangent relating the
total strain rate to total stress rate can be found in standard textbooks. Since this situation
is rather special we derive it here for the plane stress model presented in the previous
section. The elastic stress-strain relation can be written

20-116 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

1 𝜈

⎜ 𝜈 1 ⎞
⎟ p
𝜀̇11 − 𝜀̇11
𝜎̇ 11 ⎜
⎜ ⎟
⎟ ⎛ ⎞
⎜ 1−𝜈 ⎟ ⎜
⎜ p ⎟

⎜𝜎̇ 22 ⎞
⎟ ⎜ ⎟ ⎜ 𝜀̇22 − 𝜀̇22 ⎟ ⎟

⎜ ⎟
⎟ 𝐸 ⎜ ⎜
⎜ 2





⎜ p ⎟


𝛔̇ = ⎜

⎜𝜎̇ 12 ⎟

⎟ = ⎜
⎜ ⎟
⎟ ⎜
⎜ 2(𝜀̇12 − 𝜀̇ )
12 ⎟⎟
⎜ ⎟ 1 − ν 2 ⎜ 1−𝜈 ⎟ ⎜ p ⎟
⎜𝜎̇ 23 ⎟ ⎜
⎜ ⎟
⎟ ⎜
⎜ 2(𝜀̇ − 𝜀̇ ) ⎟ (22.36.7)

⎜ 2 ⎟
⎟ ⎜ 23 23 ⎟⎟
⎝𝜎̇ 13 ⎠ ⎜
⎜ 1 − 𝜈⎟
⎟ p
⎝2(𝜀̇13 − 𝜀̇13 )⎠
⎝ 2 ⎠
= 𝐂ps (𝛆̇ − 𝛆̇p ).

where 𝐸 is the Young’s modulus, 𝜈 is the Poisson’s ratio and 𝐂ps denotes the plane stress
elastic tangential stiffness matrix. The associative flow rule for the plastic strain can be
written
∂𝑓
𝛆̇p = λ̇ , (22.36.8)
∂𝛔
and the consistency condition results in
∂𝑓 T ∂𝑓
𝛔̇ + 𝛆̇ = 0. (22.36.9)
∂𝛔 ∂𝜀p p

For algorithmic consistency, the effective plastic strain rate is defined as 𝜀̇p = 𝜆̇ .
∂𝑓
Multiplying Equation (22.36.7) with ∂𝛔 and using Equation (22.36.8) and Equation (22.36.9)
gives
T
∂𝑓
𝐂ps 𝛆̇
λ̇ = ∂𝛔 .
T (22.36.10)
∂𝑓 ps ∂𝑓 ∂𝑓
𝐂 −
∂𝛔 ∂𝛔 ∂𝜀p

Inserting
T
∂𝑓
𝐂ps 𝛆̇ ∂𝑓
𝛆̇p = ∂𝛔 ,
T ∂𝛔 (22.36.11)
∂𝑓 ∂𝑓 ∂𝑓
𝐂ps −
∂𝛔 ∂𝛔 ∂𝜀p
into Equation (22.36.7) results in
T
⎛ ∂𝑓 ∂𝑓 ⎞


⎜ {𝐂ps } {𝐂ps } ⎟


∂𝛔 ∂𝛔
𝛔̇ = ⎜

⎜ 𝐂ps − ⎟

⎟ 𝛆̇. (22.36.12)

⎜ ∂𝑓
T
∂𝑓 ∂𝑓 ⎟

⎜ 𝐂ps − ⎟
⎝ ∂𝛔 ∂𝛔 ∂𝜀p ⎠

LS-DYNA DEV 10/27/16 (r:8004) 20-117 (Material Models)


Material Models LS-DYNA Theory Manual

To get the elastic-plastic tangent stiffness tensor in the element coordinate system it
needs to be transformed back. Since the elastic tangential stiffness tensor is isotropic with
respect to the axis of rotation, the plastic tangent stiffness tensor can be written


⎜ ps ∂𝑓 ps ∂𝑓 ⎞
⎜ {𝐐𝐂 } {𝐐𝐂 }⎟
ps ⎜
⎜ ps ∂𝛔 ∂𝛔 ⎟


𝐂plastic =⎜
⎜ 𝐂 − ⎟
⎟ , (22.36.13)

⎜ ∂𝑓
T
∂𝑓 ∂𝑓 ⎟

⎜ 𝐂ps − ⎟
⎝ ∂𝛔 ∂𝛔 ∂𝜀 p ⎠
where 𝐐 is the rotation matrix in Voigt form.
𝜎22 𝒏(𝑅𝜑 ) 𝜕𝑓 𝒏(𝑅45 ) 𝜎𝑌
𝜕𝝈 𝜎𝑌90 (𝜀𝑝 )
𝜕𝑓
𝜕𝑓 𝜎𝑌45 (𝜀𝑝 )
𝜕𝝈
𝜕𝝈
𝜎𝑌00 (𝜀𝑝 )
𝒏(𝑅90 )
𝑓 (𝝈, 𝜀𝑝 ) ≤ 0
𝜎𝑌 (𝜀𝑝 )
𝛼00 , 𝛼45 , 𝛼90
changes

𝜕𝑓 𝜎11 𝜀𝑝
𝒏(𝑅00 ) 𝜕𝝈

20-1 Plastic flow direction (left) and hardening (right) illustrated for variable
R-values and hardening. Changes in 𝛼00 , 𝛼45 and 𝛼90 come from changes in

22.36.2 Load curves in different directions


A project by Fleischer et.al. [2007] resulted in the option HR = 7, see *MAT_036 in LS-
DYNA Keyword User’s Manual, which allows for specification of different hardening
curves in the three directions corresponding to 0, 45 and 90 degrees. In addition, the R-
values in these directions can be specified as functions of plastic strains, both features make
up an interesting extension to the standard form of the material model. This section is
devoted to the theory of the hardening option while the theory for the R-values follows in
the next section.

22.36.2.1 An introductory remark


This material typically fits three Lankford parameters and the yield stress in the 00
direction. This fit will result in a non-intuitive effective stress-strain relationship for
uniaxial tension in other directions. To explain this we assume that we pull in the 𝜑
direction giving a uniaxial stress value of 𝜎𝜑 and a corresponding plastic strain component
𝑝
𝜀𝜑 . The relation between the stress value 𝜎𝜑 and effective stress 𝜎eff is given by

20-118 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

𝜎eff = 𝑘𝜑 𝜎𝜑
where 𝑘𝜑 = 1 if 𝜑 = 0 but not in general. The plastic work relation, which defines the
effective plastic strain for the current material, gives the following expression for the
effective plastic strain
𝑝
𝜀 𝜑 𝜎𝜑
𝜀𝑝 = 𝜎eff .
This means that there is a relationship with a stress-strain hardening curve using the
effective stress and strain and a corresponding stress-strain hardening curve using the
actual stress and strain values. Assume that a test reveals that the hardening is given by
the curve
𝑝
𝜎𝜑 = 𝜎𝜑 (𝜀𝜑 )
and we want to determine the hardening curve used by LS-DYNA
𝜎eff = 𝜎eff (𝜀𝑝 ),
then using the relationships above yields
𝜎eff (𝜀𝑝 ) = 𝑘𝜑 𝜎𝜑 (𝑘𝜑 𝜀𝑝 ).
Consequently a user input hardening curve must internally be transformed to an effective
hardening curve to be used in the material model to get the desired behavior. Still, the
effective plastic strain is not going to be equal to the plastic strain component in the tensile
direction and validation of the hardening behavior is not straightforward. Therefore we
introduce a new effective plastic strain 𝜀̃𝑝 with evolution given by
𝑑𝜀̃𝑝 𝑑𝜀𝑝
𝑑𝑡
= 𝑘𝜑 𝑑𝑡
that can be used to verify the hardening relationship. This is actually the von Mises plastic
strain in the work hardening sense and is output to the d3plot database as history variable
#2 for post-processing.

22.36.2.2 The model


The load curve hardening option can be generalized to allow different hardening curves in
the 00, 45 and 90 directions, as well as balanced biaxial and shear loading. To this end we
let the yield value be given as a convex combination of the hardening curves in each
direction as
𝜎𝑌 (𝝈, 𝜀𝑝 ) = 𝛼00 𝜎𝑌00 (𝜀𝑝 ) + 𝛼45 𝜎𝑌45 (𝜀𝑝 ) + 𝛼90 𝜎𝑌90 (𝜀𝑝 ) + 𝛼bi 𝜎𝑌bi (𝜀𝑝 ) + 𝛼sh 𝜎𝑌sh (𝜀𝑝 )
where 𝜎𝑌00 (𝜀𝑝 ) is the user defined hardening curve in direction 00, and the others are the
internal hardening curves in the other directions (see the section 22.36.2.1 and figure 20-1).
The convex parameters must fulfill
0 ≤ 𝛼00 ≤ 1, 0 ≤ 𝛼45 ≤ 1, 0 ≤ 𝛼90 ≤ 1, 0 ≤ 𝛼bi ≤ 1, 0 ≤ 𝛼sh ≤ 1,
𝛼00 + 𝛼45 + 𝛼90 + 𝛼bi + 𝛼sh = 1,
and depend on the stress state. Furthermore 𝛼00 = 1 must mean that the stress is uniaxial
and is directed in the 00 direction, and that the same thing holds for the other directions.
To accomplish this we reason as follows.
Let 𝜎̂ 𝑖𝑗 be the normalized in-plane stress components, i.e.,
𝜎𝑖𝑗
𝜎̂ 𝑖𝑗 =
2 +𝜎 2 +2𝜎 2
√𝜎11 22 12

LS-DYNA DEV 10/27/16 (r:8004) 20-119 (Material Models)


Material Models LS-DYNA Theory Manual

and let 𝜎 be the largest eigenvalue to this matrix and 𝑞𝑖 the associated eigenvector
components. Furthermore we define 𝜎𝑣 = (𝜎̂ 11 + 𝜎̂ 22 )/2 as the normalized volumetric
(𝜎 ̂22 )2
̂11 −𝜎
stress and 𝜎𝑑 = √ 4
2
+ 𝜎̂ 12 the normalized shear stress, and make a note that 0 ≤
𝜎𝑑 ≤ 1/√2 and
𝜎𝑑 = 0 → biaxial stress state
𝜎𝑑 = 1/2 → uniaxial stress state
𝜎𝑑 = 1/√2 → shear stress state.
If 𝑎 = 2𝜎𝑣2 is the fraction of stress that is volumetric and 𝑏 = 𝜎 2 an indicator of uniaxial
stress state, then 𝑐 = 𝑎(1 − 4{𝑏 − 1/2}2 ) is a normalized measure that indicates when the
stress is deviatoric/uniaxial or volumetric. That is, 𝑐 = 0 means that the stress is deviatoric
or uniaxial and 𝑐 = 1 means that it is volumetric.
Now, let 𝑞 = 4𝑞21 (1 − 𝑞21 ) be the fraction of the eigenvector 𝑞𝑖 that points in the 00 or 90
direction. That is, 𝑞 = 1 means that the eigenvector points in the 45 direction and 𝑞 = 0
means that it is pointing in either the 00 or 90 direction. Moreover, the same way of
reasoning is valid for the eigenvector associated to the smallest eigenvalue.
To determine in which direction of 00 or 90 a certain eigenvector is pointing we introduce
𝑑 = 𝑏𝑞21 + (1 − 𝑏)(1 − 𝑞21 ), and deduce that 𝑑 = 1 means that the eigenvector 𝑞𝑖 points in the
00 direction and 𝑑 = 0 means that it is pointing in the 90 direction.
We are now ready to give partial expressions for the three uniaxial convex parameters
𝛼̃00 = (1 − 𝑐)𝑑(1 − 𝑞) + 𝑐/4
𝛼̃45 = (1 − 𝑐)𝑞 + 𝑐/2
𝛼̃00 = (1 − 𝑐)(1 − 𝑑)(1 − 𝑞) + 𝑐/4
and these are completed by means of adding the biaxial and shear parts
𝛼bi = max(0,1 − 4𝜎𝑑2 )
𝛼sh = max(0,4𝜎𝑑2 − 1)
𝛼00 = 𝛼̃00 (1 − 𝛼bi − 𝛼sh )
𝛼45 = 𝛼̃45 (1 − 𝛼bi − 𝛼sh )
𝛼90 = 𝛼̃90 (1 − 𝛼bi − 𝛼sh )
This set of parameters fulfills the requirements mentioned above and allows for a decent
expression for a directional dependent yield stress.
In the consistency condition we do not consider the derivatives of the convex parameters
with respect to the stress, as we assume that these will not have a major impact on
convergence.

22.36.3 Variable Lankford parameters


The R-values are supposed to be variable with deformation, and we let 𝑅00 (𝜀𝑝 ),
𝑅45 (𝜀𝑝 ) and 𝑅90 (𝜀𝑝 ) be the internal load curves that are transformed from the ones given
by the user. Then we define the directional dependent R-value according to
𝑅(𝝈, 𝜀𝑝 ) = 𝛼00 𝑅00 (𝜀𝑝 ) + 𝛼45 𝑅45 (𝜀𝑝 ) + 𝛼90 𝑅90 (𝜀𝑝 )
using the same set of convex parameters as in the previous section. A generalized relation
for the R-value in terms of the stress can be given as
(𝜎̂ 22 + {𝜎̂ 11 + 𝜎̂ 22 }𝑅)𝑛1 + (𝜎̂ 11 + {𝜎̂ 11 + 𝜎̂ 22 }𝑅)𝑛2 − 𝜎̂ 12 𝑛4 = 0,

20-120 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

where 𝜎̂ 𝑖𝑗 are the normalized stress components, 𝑛𝑖 is the direction of plastic flow and
where we have suppressed the dependence of stress and plastic strain in 𝑅. By setting
𝜕𝑓 𝜕𝑓 𝜕𝑓
∆𝑛1 = 𝑛1 − 𝜕𝜎 , ∆𝑛2 = 𝑛2 − 𝜕𝜎 , ∆𝑛4 = 𝑛4 − 𝜕𝜎
11 22 12
and
∆𝑅 = 𝑅(𝜀𝑝 ) − 𝑅(0)
we can simplify this equation as
(𝜎̂ 22 + {𝜎̂ 11 + 𝜎̂ 22 }𝑅)∆𝑛1 + (𝜎̂ 11 + {𝜎̂ 11 + 𝜎̂ 22 }𝑅)∆𝑛2 − 𝜎̂ 12 ∆𝑛4 =
𝜕𝑓 𝜕𝑓
−(𝜕𝜎 + 𝜕𝜎 ){𝜎̂ 11 + 𝜎̂ 22 }∆𝑅
11 22
assuming that the relation already holds for the yield surface normal and R-value in the
reference configuration.
This equation is complemented with a consistency condition of the plastic flow
𝜎̂ 11 ∆𝑛1 + 𝜎̂ 22 ∆𝑛2 + 𝜎̂ 12 ∆𝑛4 = 0.
These two equations are linearly independent if and only if
(𝜎
̂ −𝜎̂ ) 2
{𝜎̂ 11 + 𝜎̂ 22 }√ 11 4 22 + 𝜎̂ 12
2
≠0
and then the equation
−𝜎̂ 12 ∆𝑛1 + 𝜎̂ 12 ∆𝑛2 + (𝜎̂ 11 − 𝜎̂ 22 )∆𝑛4 = 0
can be used to complement the previous two. This defines a system of equations that can
be used to solve in least square sense for the perturbation ∆𝑛𝑖 of the yield surface normal to
get the R-value of interest. To avoid numerical problems and make the perturbation
continuous with respect to the stress, the right hand side of the first equation is changed to
𝜕𝑓 𝜕𝑓 2
−(𝜕𝜎 + 𝜕𝜎 ){𝜎̂ 11 + 𝜎̂ 22 }((𝜎̂ 11 − 𝜎̂ 22 )2 + 4𝜎̂ 12 )∆𝑅.
11 22
This results in a non-associated flow rule, meaning that the plastic flow is not in the
direction of the yield surface normal. Again, we don’t take any special measures into
account for the stress return algorithm as we believe that the perturbation of the normal
is small enough not to deteriorate convergence. In figure 20-1 the plastic flow direction
is illustrated as function of the stress on the yield surface.

LS-DYNA DEV 10/27/16 (r:8004) 20-121 (Material Models)


Material Models LS-DYNA Theory Manual

22.37 Material Model 37: Transversely Anisotropic Elastic-


Plastic
This fully iterative plasticity model is available only for shell elements. The input
parameters for this model are: Young’s modulus 𝐸; Poisson’s ratio 𝜐; the yield stress; the
tangent modulus 𝐸t ; and the anisotropic hardening parameter 𝑅.

Consider Cartesian reference axes which are parallel to the three symmetry planes of
anisotropic behavior. Then the yield function suggested by Hill [1948] can be written
2
F(𝜎22 − 𝜎33 )2 + G(𝜎33 − 𝜎11 )2 + H(𝜎11 − 𝜎22 )2 + 2L𝜎23 2
+ 2M𝜎31 2
+ 2N𝜎12 − 1 = 0, (22.37.1)
where 𝜎y1 , 𝜎y2 , and 𝜎y3 , are the tensile yield stresses and 𝜎y12 , 𝜎y23 , and 𝜎y31 are the shear
yield stresses. The constants F, G, H, L, M, and N are related to the yield stress by
1
2L = 2
𝜎y23
1
2M = 2
𝜎y31
1
2N = 2
𝜎y12
(22.37.2)
1 1 1
2F = 2
+ 2
− 2
𝜎y2 𝜎y3 𝜎y1
1 1 1
2G = 2
+ 2
− 2
𝜎y3 𝜎y1 𝜎y2
1 1 1
2H = 2
+ 2
− 2
.
𝜎y1 𝜎y2 𝜎y3

The isotropic case of von Mises plasticity can be recovered by setting


1
F=G=H= , (22.37.3)
2𝜎y2
and
1
L=M=N= . (22.37.4)
2𝜎y2

For the particular case of transverse anisotropy, where properties do not vary in the
𝑥1 − 𝑥2 plane, the following relations hold:
20-122 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Material Models

1
F = 2G = 2
𝜎y3
2 1
2H = 2
− 2 (22.37.5)
𝜎y 𝜎y3
2 1 1
N= 2
− 2
,
𝜎y 2 𝜎y3

where it has been assumed that 𝜎y1 = 𝜎y2 = 𝜎y .

𝜎y
Letting K = 𝜎y3, the yield criterion can be written

𝑭(𝛔) = 𝜎e = 𝜎y , (22.37.6)
where
2 2
𝐹(𝛔) ≡ [𝜎11 + 𝜎22 2
+ K2 𝜎33 − K2 𝜎33 (𝜎11 + 𝜎22 ) − (2 − K2 )𝜎11 𝜎22
1 (22.37.7)
1 2 2 ⁄2
+2L𝜎y2 (𝜎23
2
+ 2
𝜎31 ) + 2 (2 − K ) 𝜎12 ] .
2

p
The rate of plastic strain is assumed to be normal to the yield surface so 𝜀̇𝑖𝑗 is found
from
p ∂𝐹
𝜀̇𝑖𝑗 = λ . (22.37.8)
∂𝜎𝑖𝑗

Now consider the case of plane stress, where 𝜎33 = 0. Also, define the anisotropy
input parameter 𝑅 as the ratio of the in-plane plastic strain rate to the out-of-plane plastic
strain rate:
p
𝜀̇22
𝑅= p . (22.37.9)
𝜀̇33
It then follows that
2
𝑅= − 1. (22.37.10)
K2

Using the plane stress assumption and the definition of 𝑅, the yield function may
now be written
1
2 2 2R 2R + 1 2 ⁄2 (22.37.11)
F(𝛔) = [𝜎11 + 𝜎22 − 𝜎11 𝜎22 + 2 𝜎 ] .
R+1 R + 1 12

LS-DYNA DEV 10/27/16 (r:8004) 20-123 (Material Models)


Material Models LS-DYNA Theory Manual

22.38 Material Model 38: Blatz-Ko Compressible Foam

𝜇 𝐼2
𝑊(𝐼1 , 𝐼2 , 𝐼3 ) = ( + 2√𝐼3 − 5), (22.38.1)
2 𝐼3
where 𝜇 is the shear modulus and 𝐼1 , 𝐼2 , and 𝐼3 are the strain invariants. Blatz and Ko
[1962] suggested this form for a 47 percent volume polyurethane foam rubber with a
Poisson’s ratio of 0.25. The second Piola-Kirchhoff stresses are given as
1 𝐼
𝑆𝑖𝑗 = 𝜇 [(𝐼𝛿𝑖𝑗 − 𝐺𝑖𝑗 ) + (√𝐼3 − 2 ) 𝐺𝑖𝑗 ], (22.38.2)
𝐼3 𝐼3
𝜕𝑥 𝜕𝑥 𝜕𝑋𝑖 𝜕𝑋𝑗
where 𝐺𝑖𝑗 = 𝜕𝑋𝑘 𝜕𝑋𝑘 , 𝐺𝑖𝑗 = 𝜕𝑥𝑘 𝜕𝑥𝑘
, after determining 𝑆𝑖𝑗 , it is transformed into the Cauchy
𝑖 𝑗

stress tensor:
𝜌 𝜕𝑥𝑖 𝜕𝑥𝑗
σ𝑖𝑗 = 𝑆 , (22.38.3)
𝜌0 𝜕𝑋𝑘 𝜕𝑋𝑙 𝑘𝑙
where 𝜌0 and 𝜌 are the initial and current density, respectively.

20-124 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.39 Material Model 39: Transversely Anisotropic Elastic-


Plastic With FLD
See Material Model 37 for the similar model theoretical basis. The first history
variable is the maximum strain ratio defined by:
𝜀majorworkpiece
(22.39.1)
εmajorfld
corresponding to 𝜀minorworkpiece . This history variable replaces the effective plastic strain in
the output. Plastic strains can still be obtained but one additional history variable must be
written into the D3PLOT database.

The strains on which these calculations are based are integrated in time from the
strain rates:

εmjr = 0

εmjr
Plane Strain
80

70

60
% Major Strain

50

40
εmnr εmjr
30 εmnr
20
Draw
εmjr
10
Stretch

-50 -40 -30 -20 -10 0 10 20 30 40 50


% Minor Strain

Figure 22.39.1. Flow limit diagram.

∇𝑛+1⁄2 𝑛+1⁄2
𝜀𝑛+1
𝑖𝑗 = 𝜀𝑛𝑖𝑗 + 𝜀𝑖𝑗 Δ𝑡 , (22.39.2)

and are stored as history variables. The resulting strain measure is logarithmic.

LS-DYNA DEV 10/27/16 (r:8004) 20-125 (Material Models)


Material Models LS-DYNA Theory Manual

22.42 Material Model 42: Planar Anisotropic Plasticity Model


This model is built into LS-DYNA as a user material model for modeling plane stress
anisotropic plasticity in shells. Please note that only three cards are input here. The
orthotropic angles must be defined later as for all materials of this type. This model is
currently not vectorized.

This is an implementation of an anisotropic plasticity model for plane stress where


the flow rule, see Material Type 37, simplifies to:
𝐹(𝜎22 )2 + 𝐺(𝜎11 )2 + 𝐻(𝜎11 − 𝜎22 )2 + 2𝑁𝜎12
2
− 1 = 0. (22.42.1)

The anisotropic parameters R00, R45, and R90 are defined in terms of 𝐹, 𝐺, 𝐻, and𝑁
as [Hill, 1989]:
𝐻
2𝑅00 = ,
𝐺
2𝑁
2𝑅45 = − 1, (22.42.2)
(𝐹 + 𝐺)
𝐻
2𝑅90 = .
𝐹

The yield function for this model is given as:


𝜎y = 𝐴𝜀𝑚 𝜀̇𝑛 . (22.42.3)

To avoid numerical problems the minimum strain rate, 𝜀̇min must be defined and the
initial yield stress 𝜎0 is calculated as
𝜎0 = 𝐴𝜀𝑚 𝑛
0 𝜀̇min = 𝐸𝜀0 , (22.42.4)
1
𝐸 𝑚−1
(22.42.5)
𝜀0 = ( 𝑛 ) .
𝐴𝜀̇min

20-126 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.51 Material Model 51: Temperature and Rate Dependent


Plasticity
The kinematics associated with the model are discussed in references [Hill 1948,
Bammann and Aifantis 1987, Bammann 1989]. The description below is taken nearly
verbatim from Bammann [Hill 1948].

With the assumption of linear elasticity we can write,


o
𝛔 = 𝜆tr(𝐃e )𝟏 + 2𝜇𝐃e , (22.51.1)
where, the Cauchy stress 𝛔 is convected with the elastic spin 𝐖e as,
o
𝛔 = 𝛔̇ − 𝐖e 𝛔 + 𝛔𝐖e . (22.51.2)

This is equivalent to writing the constitutive model with respect to a set of directors
whose direction is defined by the plastic deformation [Bammann and Aifantis 1987,
Bammann and Johnson 1987]. Decomposing both the skew symmetric and symmetric
parts of the velocity gradient into elastic and plastic parts we write for the elastic stretching
𝐃e and the elastic spin 𝐖e ,
𝐃e = 𝐃 − 𝐃p − 𝐃th , 𝐖e = 𝐖 = 𝐖p . (22.51.3)

Within this structure it is now necessary to prescribe an equation for the plastic spin
𝐖 in addition to the normally prescribed flow rule for 𝐃p and the stretching due to the
p

thermal expansion 𝐃th . As proposed, we assume a flow rule of the form,


|𝛏| − 𝜅 − 𝑌(𝑇) 𝛏′
𝐃p = 𝑓 (𝑇) sinh [ ] , (22.51.4)
𝑉(𝑇) ∣𝛏′∣
where 𝑇 is the temperate, 𝜅 is the scalar hardening variable, 𝛏′ is the difference between the
deviatoric Cauchy stress 𝛔′ and the tensor variable 𝛂′,
𝛏 ′ = 𝛔′ − 𝛂 ′ , (22.51.5)
and 𝑓 (𝑇), 𝑌(𝑇), 𝑉(𝑇) are scalar functions whose specific dependence upon the temperature
is given below. Assuming isotropic thermal expansion, and introducing the expansion
coefficient 𝐴,̇ the thermal stretching can be written,
𝐃th = 𝐴𝑇
̇ ̇ 𝟏. (22.51.6)

The evolution of the internal variables 𝛼 and 𝜅 are prescribed in a hardening minus
recovery format as,
o
𝛂 = ℎ(𝑇)𝐃p − [𝑟d (T) ∣𝐃p ∣ + 𝑟s (𝑇)] |𝛂|𝛂, (22.51.7)

LS-DYNA DEV 10/27/16 (r:8004) 20-127 (Material Models)


Material Models LS-DYNA Theory Manual

𝜅̇ = 𝐻(𝑇)𝐃p − [𝑅d (𝑇) ∣𝐃p ∣ − 𝑅s (𝑇)]𝜅 2 , (22.51.8)


where ℎ and 𝐻 are the hardening moduli, 𝑟𝑠 (𝑇) and 𝑅s (𝑇) are scalar functions describing
the diffusion controlled ‘static’ or ‘thermal’ recovery, and 𝑟d (𝑇) and 𝑅d (𝑇) are the functions
describing dynamic recovery.

If we assume that 𝐖p = 0, we recover the Jaumann stress rate which results in the
prediction of an oscillatory shear stress response in simple shear when coupled with a
Prager kinematic hardening assumption [Johnson and Bammann 1984]. Alternatively we
can choose,
𝐖p = 𝐑T 𝐔̇ 𝐔−1 𝐑, (22.51.9)
which recovers the Green-Naghdi rate of Cauchy stress and has been shown to be
equivalent to Mandel’s isoclinic state [Bammann and Aifantis 1987]. The model employing
this rate allows a reasonable prediction of directional softening for some materials but in
general under-predicts the softening and does not accurately predict the axial stresses
which occur in the torsion of the thin walled tube.

The final equation necessary to complete our description of high strain rate
deformation is one which allows us to compute the temperature change during the
deformation. In the absence of a coupled thermomechanical finite element code we assume
adiabatic temperature change and follow the empirical assumption that 90 - 95% of the
plastic work is dissipated as heat. Hence,
0.9
𝑇̇ = (𝛔 ⋅ 𝐃p ), (22.51.10)
𝜌𝐶v
where 𝜌 is the density of the material and 𝐶v the specific heat.

In terms of the input parameters the functions defined above become:


𝑉(𝑇) = C1 ∙ exp(−C2/𝑇) ℎ(𝑇) = C9 ∙ exp(C10/𝑇)
𝑌(𝑇) = C3 ∙ exp(C4/𝑇) 𝑟𝑠 (𝑇) = C11 ∙ exp(−C12/𝑇)
𝑓 (𝑇) = C5 ∙ exp(−C6/𝑇) 𝑅𝑑 (𝑇) = C13 ∙ exp(−C14/𝑇)
𝑟𝑑 (𝑇) = C7 ∙ exp(−C8/𝑇) 𝐻(𝑇) = C15 ∙ exp(C16/𝑇)
𝑅𝑠 (𝑇) = C17 ∙ exp(−C18/𝑇)
and the heat generation coefficient is
0.9
𝐻𝐶 = . (22.51.11)
𝜌𝐶𝑉

20-128 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.52 Material Model 52: Sandia’s Damage Model


The evolution of the damage parameter, 𝜙, is defined by [Bammann, et. al., 1990]
∣Dp ∣
1
𝜙̇ = 𝛽 [ − (1 − 𝜙)] , (22.52.1)
(1 − 𝜙)𝑁
in which
2(2𝑁 − 1)𝑝
𝛽 = sin [ ], (22.52.2)
(2𝑁 − 1)𝜎 ̅̅̅̅
where 𝑝 is the pressure and ̅𝜎̅̅̅ is the effective stress.

LS-DYNA DEV 10/27/16 (r:8004) 20-129 (Material Models)


Material Models LS-DYNA Theory Manual

22.53 Material Model 53: Low Density Closed Cell


Polyurethane Foam
A rigid, low density, closed cell, polyurethane foam model developed at Sandia
Laboratories [Neilsen et al., 1987] has been recently implemented for modeling impact
limiters in automotive applications. A number of such foams were tested at Sandia and
reasonable fits to the experimental data were obtained.

In some respects this model is similar to the crushable honeycomb model type 26 in
that the components of the stress tensor are uncoupled until full volumetric compaction is
achieved. However, unlike the honeycomb model this material possesses no directionality
but includes the effects of confined air pressure in its overall response characteristics.
𝜎𝑖𝑗 = 𝜎𝑖𝑗sk − δ𝑖𝑗 𝜎 air , (22.53.1)

where 𝜎𝑖𝑗sk is the skeletal stress and 𝜎 air is the air pressure computed from the equation:
𝑝0 𝛾
𝜎 air = − , (22.53.2)
1+𝛾−𝜙
where 𝑝0 is the initial foam pressure usually taken as the atmospheric pressure and 𝛾
defines the volumetric strain
𝛾 = 𝑉 − 1 + 𝛾0 , (22.53.3)
where 𝑉 is the relative volume and 𝛾0 is the initial volumetric strain which is typically
zero. The yield condition is applied to the principal skeletal stresses which are updated
independently of the air pressure. We first obtain the skeletal stresses:
𝜎𝑖𝑗sk = 𝜎𝑖𝑗 + 𝜎𝑖𝑗 𝜎 air , (22.53.4)

and compute the trial stress, 𝛔𝑖skt


𝜎𝑖𝑗skt = 𝜎𝑖𝑗sk + 𝐸𝜀̇𝑖𝑗 Δ𝑡, (22.53.5)
where 𝐸 is Young’s modulus. Since Poisson’s ratio is zero, the update of each stress
component is uncoupled and 2𝐺 = 𝐸 where 𝐺 is the shear modulus. The yield condition is
applied to the principal skeletal stresses such that if the magnitude of a principal trial stress
component, 𝛔𝑖skt , exceeds the yield stress, 𝜎y , then

𝛔𝑖skt
𝛔𝑖sk = min(𝜎y , ∣𝛔𝑖skt ∣) . (22.53.6)
∣𝛔𝑖skt ∣

The yield stress is defined by


𝜎y = 𝑎 + 𝑏(1 + 𝑐𝛾), (22.53.7)

20-130 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

where 𝑎, 𝑏, and 𝑐 are user defined input constants. After scaling the principal stresses they
are transformed back into the global system and the final stress state is computed
𝜎𝑖𝑗 = 𝜎𝑖𝑗sk − 𝛿𝑖𝑗 𝜎 air . (22.53.8)

LS-DYNA DEV 10/27/16 (r:8004) 20-131 (Material Models)


Material Models LS-DYNA Theory Manual

22.54 Material Models 54 and 55: Enhanced Composite


Damage Model
These models are very close in their formulations. Material 54 uses the Chang
matrix failure criterion (as Material 22), and material 55 uses the Tsay-Wu criterion for
matrix failure.

Arbitrary orthothropic materials, e.g., unidirectional layers in composite shell


structures can be defined. Optionally, various types of failure can be specified following
either the suggestions of [Chang and Chang, 1984] or [Tsai and Wu, 1981]. In addition
special measures are taken for failure under compression. See [Matzenmiller and
Schweizerhof, 1990]. This model is only valid for thin shell elements.

The Chang/Chang criteria is given as follows: for the tensile fiber mode,
𝜎aa 2 𝜎
𝜎aa > 0 then 𝑒f2 = ( ) + 𝛽 ( ab ) − 1 {≥ 0 failed , (22.54.1)
Xt Sc < 0 elastic

Ea = Eb = Gab = νba = νab = 0, (22.54.2)


for the compressive fiber mode,
𝜎aa 2
𝜎aa < 0 then 𝑒c2 = ( ) − 1 {≥ 0 failed , (22.54.3)
Xc < 0 elastic

Ea = νba = νab = 0, (22.54.4)


for the tensile matrix mode,
𝜎bb 2 𝜎 2
𝜎bb > 0 then 2
𝑒m =( ) + ( ab ) − 1 {≥ 0 failed , (22.54.5)
Yt Sc < 0 elastic

Eb = νba = 0 → Gab = 0, (22.54.6)


and for the compressive matrix mode,
𝜎bb 2 ⎡ Yc 2 𝜎bb 𝜎ab 2
𝜎bb < 0 then 𝑒d2 =( ) + ⎢( ) − 1⎤
⎥ + ( ) − 1 {≥ 0 failed , (22.54.7)
2Sc ⎣ 2S c Y
⎦ c S c < 0 elastic

Gab = 0
Eb = νba = νab = 0 ⇒ . (22.54.8)
XC = 2Yc , for 50% fiber volume

In the Tsay/Wu criteria the tensile and compressive fiber modes are treated as in the
Chang/Chang criteria. The failure criterion for the tensile and compressive matrix mode
is given as:

20-132 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

2 2
𝜎bb 𝜎 (Y − Yt ) 𝜎bb
2
𝑒md = + ( ab ) + c − 1 {≥ 0 failed . (22.54.9)
Yc Yt Sc Yc Yt < 0 elastic
For 𝛽 = 1 we get the original criterion of Hashin [1980] in the tensile fiber mode. For 𝛽 = 0,
we get the maximum stress criterion which is found to compare better to experiments.

Failure can occur in any of four different ways:

1. If DFAILT is zero, failure occurs if the Chang/Chang failure criterion is satisfied in


the tensile fiber mode.

2. If DFAILT is greater than zero, failure occurs if the tensile fiber strain is greater
than DFAILT or less than DFAILC.

3. If EFS is greater than zero, failure occurs if the effective strain is greater than EFS.

4. If TFAIL is greater than zero, failure occurs according to the element time step as
described in the definition of TFAIL above.

When failure has occurred in all the composite layers (through-thickness integration
points), the element is deleted. Elements which share nodes with the deleted element
become “crashfront” elements and can have their strengths reduced by using the SOFT
parameter with TFAIL greater than zero.

Information about the status in each layer (integration point) and element can be
plotted using additional integration point variables. The number of additional integration
point variables for shells written to the LS-DYNA database is input by the
*DATABASE_BINARY definition as variable NEIPS. For Models 54 and 55 these
additional variables are tabulated below (i = shell integration point):

History Description Value LS-PREPOST


Variable History
Variable
1. 𝑒𝑓 (𝑖) tensile fiber mode 1
2. 𝑒𝑐(𝑖) compressive fiber mode 1 – elastic 2
3. 𝑒𝑚(𝑖) tensile matrix mode 0 – failed 3
4. 𝑒𝑑(𝑖) compressive matrix mode 4
5. 𝑒𝑓𝑎𝑖l max[𝑒𝑓 (𝑖𝑝)] 5
-1 - element intact
6. 𝑑𝑎𝑚 damage parameter 10-8 - element in crashfront 6
+1 - element failed

The following components, defined by the sum of failure indicators over all
through-thickness integration points, are stored as element component 7 instead of the
effective plastic strain.:
LS-DYNA DEV 10/27/16 (r:8004) 20-133 (Material Models)
Material Models LS-DYNA Theory Manual

Description Integration point


𝑛𝑖𝑝
1
∑ 𝑒𝑓 (𝑖) 1
𝑛𝑖𝑝 𝑖=1
𝑛𝑖𝑝
1
∑ 𝑒𝑐(𝑖) 2
𝑛𝑖𝑝 𝑖=1
𝑛𝑖𝑝
1
∑ 𝑐𝑚(𝑖) 2
𝑛𝑖𝑝 𝑖=1

20-134 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.57 Material Model 57: Low Density Urethane Foam


The urethane foam model is available to model highly compressible foams such as
those used in seat cushions and as padding on the Side Impact Dummy (SID). The
compressive behavior is illustrated in Figure 22.57.1 where hysteresis on unloading is
shown. This behavior under uniaxial loading is assumed not to significantly couple in the
transverse directions. In tension the material behaves in a linear fashion until tearing
occurs. Although our implementation may be somewhat unusual, it was first motivated by
Shkolnikov [1991] and a paper by Storakers [1986]. The recent additions necessary to
model hysteretic unloading and rate effects are due to Chang, et al., [1994]. These latter
additions have greatly expanded the usefulness of this model.

The model uses tabulated input data for the loading curve where the nominal
stresses are defined as a function of the elongations, 𝜀𝑖 , which are defined in terms of the
principal stretches, 𝜆𝑖 , as:
𝜀𝑖 = 𝜆 𝑖 − 1. (22.57.1)

The stretch ratios are found by solving for the eigenvalues of the left stretch tensor,
𝑉𝑖𝑗 , which is obtained via a polar decomposition of the deformation gradient matrix, 𝐹𝑖𝑗 :
𝐹𝑖𝑗 = 𝑅𝑖𝑘 𝑈𝑘𝑗 = 𝑉𝑖𝑘 𝑅𝑘𝑗 . (22.57.2)

The update of 𝑉𝑖𝑗 follows the numerically stable approach of Taylor and Flanagan
[1989]. After solving for the principal stretches, the elongations are computed and, if the
elongations are compressive, the corresponding values of the nominal stresses, 𝜏𝑖 are
interpolated. If the elongations are tensile, the nominal stresses are given by
𝜏𝑖 = 𝐸𝜀𝑖 . (22.57.3)

Typical unloading Typical unloading for


curves determined by a large shape factor, e.g.
the hysteretic unloading 5.0-8.0, and a small
factor. With the shape hystereticfactor, e.g., 0.010.
factor equal to unity.
σ σ

Unloading
curves

Strain Strain
Figure 22.57.1. Behavior of the low-density urethane foam model.

LS-DYNA DEV 10/27/16 (r:8004) 20-135 (Material Models)


Material Models LS-DYNA Theory Manual

The Cauchy stresses in the principal system become


𝜏𝑖
𝜎𝑖 = . (22.57.4)
𝜆𝑗 𝜆𝑘
The stresses are then transformed back into the global system for the nodal force
calculations.

When hysteretic unloading is used, the reloading will follow the unloading curve if
the decay constant, 𝛽, is set to zero. If 𝛽 is nonzero the decay to the original loading curve
is governed by the expression:
1 − 𝑒−𝛽𝑡 . (22.57.5)

The bulk viscosity, which generates a rate dependent pressure, may cause an
unexpected volumetric response and, consequently, it is optional with this model.

Rate effects are accounted for through linear viscoelasticity by a convolution integral
of the form
𝑡 ∂𝜀𝑘𝑙
𝜎𝑖𝑗r = ∫ 𝑔𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) 𝑑𝜏, (22.57.6)
0 ∂𝜏
where 𝑔𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) is the relaxation function. The stress tensor, 𝜎𝑖𝑗r , augments the stresses
determined from the foam, 𝜎𝑖𝑗f ; consequently, the final stress, 𝜎𝑖𝑗 , is taken as the summation
of the two contributions:
𝜎𝑖𝑗 = 𝜎𝑖𝑗f + 𝜎𝑖𝑗r . (22.57.7)

Since we wish to include only simple rate effects, the relaxation function is
represented by one term from the Prony series:
N
𝑔(𝑡) = 𝛼0 + ∑ α𝑚 𝑒−𝛽𝑡 , (22.57.8)
𝑚=1

given by,
𝑔(𝑡) = 𝐸d 𝑒−𝛽1𝑡 . (22.57.9)

This model is effectively a Maxwell fluid which consists of a damper and spring in
series. We characterize this in the input by a Young's modulus, 𝐸d , and decay constant, 𝛽1 .
The formulation is performed in the local system of principal stretches where only the
principal values of stress are computed and triaxial coupling is avoided. Consequently, the
one-dimensional nature of this foam material is unaffected by this addition of rate effects.
The addition of rate effects necessitates twelve additional history variables per integration
point. The cost and memory overhead of this model comes primarily from the need to
“remember” the local system of principal stretches.

20-136 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.58 Material Model 58: Laminated Composite Fabric


Parameters to control failure of an element layer are: ERODS, the maximum effective
strain, i.e., maximum 1 = 100% straining. The layer in the element is completely removed
after the maximum effective strain (compression/tension including shear) is reached.

The stress limits are factors used to limit the stress in the softening part to a given
value,
𝜎min = SLIMxx ⋅ strength, (22.58.1)
thus, the damage value is slightly modified such that elastoplastic like behavior is
achieved with the threshold stress. As a factor for SLIMxx a number between 0.0 and 1.0 is
possible. With a factor of 1.0, the stress remains at a maximum value identical to the
strength, which is similar to ideal elastoplastic behavior. For tensile failure a small value
for SLIMTx is often reasonable; however, for compression SLIMCx = 1.0 is preferred. This
is also valid for the corresponding shear value. If SLIMxx is smaller than 1.0 then
localization can be observed depending on the total behavior of the lay-up. If the user is
intentionally using SLIMxx < 1.0, it is generally recommended to avoid a drop to zero and
set the value to something in between 0.05 and 0.10. Then elastoplastic behavior is
achieved in the limit which often leads to less numerical problems. Defaults for
SLIMXX = 1.0E-8.

The crashfront-algorithm is started if and only if a value for TSIZE (time step size,
with element elimination after the actual time step becomes smaller than TSIZE) is input .

The damage parameters can be written to the postprocessing database for each
integration point as the first three additional element variables and can be visualized.

Material models with FS = 1 or FS = −1 are favorable for complete laminates and


fabrics, as all directions are treated in a similar fashion.

For material model FS = 1 an interaction between normal stresses and shear stresses
is assumed for the evolution of damage in the a- and b- directions. For the shear damage is
always the maximum value of the damage from the criterion in a- or b- direction is taken.

For material model FS = −1 it is assumed that the damage evolution is independent


of any of the other stresses. A coupling is present only via the elastic material parameters
and the complete structure.

In tensile and compression directions and in a- as well as in b- direction, different


failure surfaces can be assumed. The damage values, however, increase only when the
loading direction changes.

LS-DYNA DEV 10/27/16 (r:8004) 20-137 (Material Models)


Material Models LS-DYNA Theory Manual

SC

TAU1

SLIMS*SC

GAMMA1 GMS
γ
Figure 22.58.1. Stress-strain diagram for shear.

Special control of shear behavior of fabrics


For fabric materials a nonlinear stress strain curve for the shear part of failure
surface FS = −1 can be assumed as given below. This is not possible for other values of FS.

The curve, shown in Figure 22.58.1, is defined by three points:


• the origin (0,0) is assumed,
• the limit of the first slightly nonlinear part (must be input), stress (TAU1) and strain
(GAMMA1), see below.
• the shear strength at failure and shear strain at failure.

In addition a stress limiter can be used to keep the stress constant via the SLIMS
parameter. This value must be less than or equal to 1.0 and positive, which leads to an
elastoplastic behavior for the shear part. The default is 1.0E-08, assuming almost brittle
failure once the strength limit SC is reached.

20-138 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.60 Material Model 60: Elastic With Viscosity


This material model was developed to simulate the forming of glass products (e.g.,
car windshields) at high temperatures. Deformation is by viscous flow but elastic
deformations can also be large. The material model, in which the viscosity may vary with
temperature, is suitable for treating a wide range of viscous flow problems and is
implemented for brick and shell elements.

Volumetric behavior is treated as linear elastic. The deviatoric strain rate is


considered to be the sum of elastic and viscous strain rates:
𝛔̇ ′ 𝛔̇ ′
𝛆̇′total = 𝛆̇′elastic + 𝛆̇′ viscous = + , (22.60.1)
2𝐺 2𝜈
where 𝐺 is the elastic shear modulus, 𝜈 is the viscosity coefficient. The stress increment
over one time step 𝑑𝑡 is
𝐺
𝑑𝛔′ = 2𝐺𝛆̇′total 𝑑𝑡 − 𝑑𝑡𝛔′ . (22.60.2)
𝜐

The stress before the update is used for 𝛔′. For shell elements, the through-thickness
strain rate is calculated as follows
𝐺
𝑑𝑡σ′ 33 ,
𝑑𝜎33 = 0 = 𝐾(𝜀̇11 + 𝜀̇22 + 𝜀̇33 )𝑑𝑡 + 2𝐺𝜀̇′33 𝑑𝑡 − (22.60.3)
𝜐
where the subscript 𝑖𝑗 = 33 denotes the through-thickness direction and 𝐾 is the elastic bulk
modulus. This leads to:
𝜀̇33 = −a(𝜀̇11 + 𝜀̇22 ) + 𝑏𝑝, (22.60.4)

𝐾 − 2𝐺
𝑎= 3 , (22.60.5)
𝐾 + 4𝐺
3
𝐺𝑑𝑡
𝑏= , (22.60.6)
𝜐(𝐾 + 4 𝐺)
3
in which 𝑝 is the pressure defined as the negative of the hydrostatic stress.

LS-DYNA DEV 10/27/16 (r:8004) 20-139 (Material Models)


Material Models LS-DYNA Theory Manual

22.61 Material Model 61: Maxwell/Kelvin Viscoelastic with


Maximum Strain
The shear relaxation behavior is described for the Maxwell model by:
𝐺(𝑡) = G∞ + (G0 − G∞ )𝑒−𝛽𝑡 . (22.61.1)

A Jaumann rate formulation is used


∇ 𝑡
𝑠′𝑖𝑗 = 2 ∫ 𝐺(𝑡 − 𝜏)𝜀′̇ 𝑖𝑗 (𝜏)𝑑𝑡, (22.61.2)
0

where the prime denotes the deviatoric part of the stress rate, 𝑠′𝑖𝑗 , and 𝜀̇′𝑖𝑗 is the deviatoric
strain rate.

For the Kelvin model the stress evolution equation is defined as:
1 G
𝑠𝑖𝑗̇ + 𝑠𝑖𝑗 = (1 + 𝛿𝑖𝑗 )G0 𝜀′̇ 𝑖𝑗 + (1 + 𝛿𝑖𝑗 ) ∞ 𝜀′𝑖𝑗 , (22.61.3)
𝜏 𝜏
where 𝛿𝑖𝑗 is the Kronecker delta, G0 is the instantaneous shear modulus, G∞ is the long term
shear modulus, and τ is the decay constant.

The pressure is determined from the bulk modulus and the volumetric strain:
𝑝 = −𝐾𝜀v , (22.61.4)
where
𝑉
𝜀v = ln ( ), (22.61.5)
𝑉0
defines the logarithmic volumetric strain from the relative volume.

Bandak’s [1991] calculation of the total strain tensor, 𝜀𝑖𝑗 , for output uses an
incremental update based on Jaumann rate:
𝛻𝑛+1⁄2 𝑛+1⁄2
𝜀𝑛+1
𝑖𝑗 = 𝜀𝑛𝑖𝑗 + 𝑟𝑖𝑗𝑛 + 𝜀𝑖𝑗 𝛥𝑡 , (22.61.6)

where
𝑛+1⁄2 𝑛+1⁄2 𝑛+1⁄2
𝛥𝜀𝑖𝑗 = 𝜀̇𝑖𝑗 𝛥𝑡 , (22.61.7)

and 𝑟𝑖𝑗𝑛 gives the rotation of the stain tensor at time 𝑡𝑛 to the configuration at 𝑡𝑛+1
𝑛+1⁄2 𝑛+1⁄2 𝑛+1⁄2
𝑟𝑖𝑗𝑛 = (𝜀𝑛𝑖𝑝 𝜔𝑝𝑗 𝑛
+ 𝜀𝑗𝑝 𝜔𝑝𝑖 ) 𝛥𝑡 . (22.61.8)

20-140 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.62 Material Model 62: Viscous Foam


This model was written to represent the energy absorbing foam found on certain
crash dummies, i.e., the ‘Confor Foam’ covering the ribs of the Eurosid dummy.

The model consists of a nonlinear elastic stiffness in parallel with a viscous damper.
A schematic is shown in Figure 22.62.1. The elastic stiffness is intended to limit total crush
while the viscous damper absorbs energy. The stiffness 𝐸2 prevents timestep problems.

Both 𝐸1 and 𝑉2 are nonlinear with crush as follows:


𝐸𝑡1 = 𝐸1 (𝑉 −n1 ),
n (22.62.1)
𝑉2𝑡 = 𝑉2 (abs(1 − 𝑉)) 2 ,
where 𝑉 is the relative volume defined by the ratio of the current to initial volume. Typical
values are (units of N, mm, s)
𝐸1 = 0.0036,
𝑛1 = 4.0,
𝑉2 = 0.0015,
(22.62.2)
𝐸2 = 100.0,
𝑛2 = 0.2,
𝜈 = 0.05.

E1

V1
E2

Figure 22.62.1. Schematic of Material Model 62.

LS-DYNA DEV 10/27/16 (r:8004) 20-141 (Material Models)


Material Models LS-DYNA Theory Manual

22.63 Material Model 63: Crushable Foam


The intent of this model is model crushable foams in side impact and other
applications where cyclic behavior is unimportant.

This isotropic foam model crushes one-dimensionally with a Poisson’s ratio that is
essentially zero. The stress versus strain behavior is depicted in Figure 22.63.1 where an
example of unloading from point a to the tension cutoff stress at b then unloading to point
c and finally reloading to point d is shown. At point the reloading will continue along the
loading curve. It is important to use nonzero values for the tension cutoff to prevent the
disintegration of the material under small tensile loads. For high values of tension cutoff
the behavior of the material will be similar in tension and compression.

In the implementation we assume that Young’s modulus is constant and update the
stress assuming elastic behavior.
𝑛+1⁄2 𝑛+1⁄2
𝜎𝑖𝑗trial = 𝜎𝑖𝑗𝑛 + 𝐸𝜀̇𝑖𝑗 Δ𝑡 . (22.63.1)

The magnitudes of the principal values, 𝜎𝑖trial , 𝑖 = 1,3 are then checked to see if the yield
stress, 𝜎y , is exceeded and if so they are scaled back to the yield surface:

𝜎𝑖trial
if 𝜎y < ∣𝜎𝑖trial ∣ then 𝜎𝑖𝑛+1 = 𝜎y . (22.63.2)
∣σ𝑖trial ∣

After the principal values are scaled, the stress tensor is transformed back into the
global system. As seen in Figure 22.63.1, the yield stress is a function of the natural

a
σij

d
E

c b
Volumetric strain - In V
Figure 22.63.1. Yield stress versus volumetric strain curve for the crushable
foam.

20-142 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

logarithm of the relative volume, 𝑉, i.e., the volumetric strain.

LS-DYNA DEV 10/27/16 (r:8004) 20-143 (Material Models)


Material Models LS-DYNA Theory Manual

22.64 Material Model 64: Strain Rate Sensitive Power-Law


Plasticity
This material model follows a constitutive relationship of the form:
𝜎 = 𝑘𝜀𝑚 𝜀̇𝑛 (22.64.1)
where 𝜎 is the yield stress, 𝜀 is the effective plastic strain, 𝜀̇ is the effective plastic strain
rate, and the constants 𝑘, 𝑚, and 𝑛 can be expressed as functions of effective plastic strain or
can be constant with respect to the plastic strain. The case of no strain hardening can be
obtained by setting the exponent of the plastic strain equal to a very small positive value,
i.e., 0.0001.

This model can be combined with the superplastic forming input to control the
magnitude of the pressure in the pressure boundary conditions in order to limit the
effective plastic strain rate so that it does not exceed a maximum value at any integration
point within the model.

A fully viscoplastic formulation is optional. An additional cost is incurred but the


improvement in results can be dramatic.

20-144 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.65 Material Model 65: Modified Zerilli/Armstrong


The Armstrong-Zerilli Material Model expresses the flow stress as follows. For fcc
metals,
1⁄ ∗ 𝜇(𝑇)
𝜎 = C1 + {C2 (𝜀p ) 2 [𝑒(−C3 +C4 ln(𝜀̇ ))𝑇 ] + C5 } ( ), (22.65.1)
𝜇(293)
𝜀p = effective plastic strain
𝜀̇
𝜀̇∗ = effective plastic strain rate
𝜀̇0
where 𝜀̇0 = 1,1𝑒 − 3,1𝑒 − 6 for time units of seconds, milliseconds, and microseconds,
respectively.

For bcc metals,


∗ 𝜇(𝑇)
𝜎 = C1 + C2 𝑒(−C3+C4ln(𝜀̇ ))𝑇 + [C5 (𝜀p )𝑛 + C6 ] ( ), (22.65.2)
𝜇(293)
where
𝜇(𝑇)
( ) = B1 + B 2 𝑇 + B 3 𝑇 2 . (22.65.3)
𝜇(293)

The relationship between heat capacity (specific heat) and temperature may be
characterized by a cubic polynomial equation as follows:
Cp = G 1 + G 2 𝑇 + G 3 𝑇 2 + G 4 𝑇 3 . (22.65.4)

A fully viscoplastic formulation is optional. An additional cost is incurred but the


improvement in results can be dramatic.

LS-DYNA DEV 10/27/16 (r:8004) 20-145 (Material Models)


Material Models LS-DYNA Theory Manual

22.66 Material Model 66: Linear Stiffness/Linear Viscous 3D


Discrete Beam
The formulation of the discrete beam (Type 6) assumes that the beam is of zero
length and requires no orientation node. A small distance between the nodes joined by the
beam is permitted. The local coordinate system which determines (𝑟, 𝑠, 𝑡) is given by the
coordinate ID in the cross sectional input where the global system is the default. The local
coordinate system axes rotate with the average of the rotations of the two nodes that define
the beam.

For null stiffness coefficients, no forces corresponding to these null values will
develop. The viscous damping coefficients are optional.

20-146 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.67 Material Model 67: Nonlinear Stiffness/Viscous 3D


Discrete Beam
The formulation of the discrete beam (Type 6) assumes that the beam is of zero
length and requires no orientation node. A small distance between the nodes joined by the
beam is permitted. The local coordinate system which determines (𝑟, 𝑠, 𝑡) is given by the
coordinate ID in the cross sectional input where the global system is the default. The local
coordinate system axes rotate with the average of the rotations of the two nodes that define
the beam.

For null load curve ID’s, no forces are computed. The force resultants are found
from load curves (See Figure 22.67.1) that are defined in terms of the force resultant versus
the relative displacement in the local coordinate system for the discrete beam.

(a.) (b.)
R
E
S R
U E
L S
T U
A L
N T
T A
N
DISPLACEMENT T

|DISPLACEMENT|

Figure 22.67.1. The resultant forces and moments are determined by a table
lookup. If the origin of the load curve is at [0,0] as in (b.) and tension and
compression responses are symmetric.

LS-DYNA DEV 10/27/16 (r:8004) 20-147 (Material Models)


Material Models LS-DYNA Theory Manual

22.68 Material Model 68: Nonlinear Plastic/Linear Viscous


3D Discrete Beam
The formulation of the discrete beam (Type 6) assumes that the beam is of zero
length and requires no orientation node. A small distance between the nodes joined by the
beam is permitted. The local coordinate system which determines (𝑟, 𝑠, 𝑡) is given by the
coordinate ID in the cross sectional input where the global system is the default. The local
coordinate system axes rotate with the average of the rotations of the two nodes that define
the beam. Each force resultant in the local system can have a limiting value defined as a
function of plastic displacement by using a load curve (See Figure 22.68.1). For the degrees
of freedom where elastic behavior is desired, the load curve ID is simply set to zero.

Catastrophic failure, based on force resultants, occurs if the following inequality is


satisfied:
2 2 2 2 2 2
𝐹r 𝐹s 𝐹t 𝑀r 𝑀s 𝑀t
( fail ) + ( fail ) + ( fail ) + ( fail ) + ( fail ) + ( fail ) − 1. ≥ 0. (22.68.1)
𝐹r 𝐹s 𝐹t 𝑀r 𝑀s 𝑀t
Likewise, catastrophic failure based on displacement resultants occurs if:
2 2 2 2 2 2
𝑢r 𝑢s 𝑢t 𝜃r 𝜃s 𝜃t
( ) + ( ) + ( ) + ( ) + ( ) + ( ) − 1. ≥ 0. (22.68.2)
𝑢fail
r 𝑢fail
s 𝑢fail
t 𝜃r
fail 𝜃s
fail 𝜃t
fail

After failure, the discrete element is deleted. If failure is included, either one or both
of the criteria may be used.

R
E
S
U
L
T
A
N
T

PLASTIC DISPLACEMENT
Figure 22.68.1. The resultant forces and moments are limited by the yield
definition. The initial yield point corresponds to a plastic displacement of zero.

20-148 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.69 Material Model 69: Side Impact Dummy Damper (SID


Damper)
The side impact dummy uses a damper that is not adequately treated by nonlinear
force versus relative velocity curves, since the force characteristics are also dependent on
the displacement of the piston. As the damper moves, the fluid flows through the open
orifices to provide the necessary damping resistance. While moving as shown in Figure
22.69.1, the piston gradually blocks off and effectively closes the orifices. The number of
orifices and the size of their openings control the damper resistance and performance. The
damping force is computed from the equation:

{ ⎧
{𝐶1 𝐴p 2 ⎫
} ⎫
}
𝐹 = 𝑆𝐹 ⎨𝐾𝐴p 𝑉p ⎨ 𝑡 + 𝐶2 ∣𝐕p ∣𝜌fluid ⎢( 𝑡 ) − 1⎤

⎥⎬ − 𝑓 (𝑠 + 𝑠 0 ) + 𝑉p 𝑔(𝑠 + 𝑠 0 ⎬, (22.69.1)
)
{
⎩ {
⎩𝐴0 ⎣ 𝐶𝐴0 ⎦}⎭ }

where 𝐾 is a user defined constant or a tabulated function of the absolute value of the
relative velocity, 𝐕p is the piston's relative velocity, 𝐶 is the discharge coefficient, 𝐴p is the
piston area, 𝐴𝑡0 is the total open areas of orifices at time 𝑡, 𝜌fluid is the fluid density, 𝐶1 is the
coefficient for the linear term, and 𝐶2 is the coefficient for the quadratic term.

d4
d3
d2 Piston

d1

Vp

ith Piston Orifice

Orifice Opening Controller

2Ri - h

Figure 22.69.1. Mathematical model for the Side Impact Dummy damper.

LS-DYNA DEV 10/27/16 (r:8004) 20-149 (Material Models)


Material Models LS-DYNA Theory Manual

F
O
R
C
E Last orifice
closes.

Force increases as orifice


is gradually covered.

DISPLACEMENT

Figure 22.69.2. Force versus displacement as orifices are covered at a constant


relative velocity. Only the linear velocity term is active.

In the implementation, the orifices are assumed to be circular with partial covering
by the orifice controller. As the piston closes, the closure of the orifice is gradual. This
gradual closure is taken into account to insure a smooth response. If the piston stroke is
exceeded, the stiffness value, 𝑘, limits further movement, i.e., if the damper bottoms out in
tension or compression, the damper forces are calculated by replacing the damper by a
bottoming out spring and damper, k and c, respectively. The piston stroke must exceed the
initial length of the beam element. The time step calculation is based in part on the stiffness
value of the bottoming out spring. A typical force versus displacement curve at constant
relative velocity is shown in Figure 22.69.2. The factor, SF, which scales the force defaults
to 1.0 and is analogous to the adjusting ring on the damper.

20-150 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.70 Material Model 70: Hydraulic/Gas Damper


This special purpose element represents a combined hydraulic and gas-filled
damper which has a variable orifice coefficient. A schematic of the damper is shown in
Figure 22.70.1. Dampers of this type are sometimes used on buffers at the end of railroad
tracks and as aircraft undercarriage shock absorbers. This material can be used only as a
discrete beam element.

As the damper is compressed two actions contribute to the force that develops.
First, the gas is adiabatically compressed into a smaller volume. Secondly, oil is forced
through an orifice. A profiled pin may occupy some of the cross-sectional area of the
orifice; thus, the orifice area available for the oil varies with the stroke. The force is
assumed proportional to the square of the velocity and inversely proportional to the
available area. The equation for this element is:
𝑛
𝑉 2 𝐶0
𝐹 = SCLF ⋅ {𝐾h ( ) + [𝑃0 ( ) − 𝑃a ] ⋅ 𝐴p }, (22.70.1)
𝑎0 𝐶0 − 𝑆
where 𝑆 is the element deflection and 𝑉 is the relative velocity across the element.

Orifice

Oil Profiled Pin Gas


Figure 22.70.1. Schematic of Hydraulic/Gas damper.

LS-DYNA DEV 10/27/16 (r:8004) 20-151 (Material Models)


Material Models LS-DYNA Theory Manual

22.71 Material Model 71: Cable


This material can be used only as a discrete beam element. The force, 𝐹, generated
by the cable is nonzero only if the cable is in tension. The force is given by:
𝐹 = 𝐾 ⋅ max(Δ𝐿, 0. ), (22.71.1)
where Δ𝐿 is the change in length
Δ𝐿 = current length − (initial length-offset), (22.71.2)
and the stiffness is defined as:
𝐸 ⋅ area
𝐾= . (22.71.3)
(initial length- offset)

The area and offset are defined on either the cross section or element cards in the LS-
DYNA input. For a slack cable the offset should be input as a negative length. For an
initial tensile force the offset should be positive. If a load curve is specified, the Young’s
modulus will be ignored and the load curve will be used instead. The points on the load
curve are defined as engineering stress versus engineering strain, i.e., the change in length
over the initial length. The unloading behavior follows the loading.

20-152 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.73 Material Model 73: Low Density Viscoelastic Foam


This viscoelastic foam model is available to model highly compressible viscous
foams. The hyperelastic formulation of this model follows that of material 57. Rate effects
are accounted for through linear viscoelasticity by a convolution integral of the form
𝑡 𝜕𝜀𝑘𝑙
𝜎𝑖𝑗r = ∫ 𝑔𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) 𝑑𝜏, (22.73.1)
0 𝜕𝜏
where 𝑔𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) is the relaxation function. The stress tensor, 𝜎𝑖𝑗r , augments the stresses
determined from the foam, 𝜎𝑖𝑗f ; consequently, the final stress, 𝜎𝑖𝑗 , is taken as the summation
of the two contributions:
𝜎𝑖𝑗 = 𝜎𝑖𝑗f + 𝜎𝑖𝑗r . (22.73.2)

Since we wish to include only simple rate effects, the relaxation function is
represented by up to six terms of the Prony series:
𝑁
𝑔(𝑡) = 𝛼0 + ∑ 𝛼𝑚 𝑒−𝛽𝑡 . (22.73.3)
𝑚=1

This model is effectively a Maxwell fluid which consists of a dampers and springs in
series. The formulation is performed in the local system of principal stretches where only
the principal values of stress are computed and triaxial coupling is avoided. Consequently,
the one-dimensional nature of this foam material is unaffected by this addition of rate
effects. The addition of rate effects necessitates 42 additional history variables per
integration point. The cost and memory overhead of this model comes primarily from the
need to “remember” the local system of principal stretches and the evaluation of the
viscous stress components.

LS-DYNA DEV 10/27/16 (r:8004) 20-153 (Material Models)


Material Models LS-DYNA Theory Manual

22.74 Material Model 74: Elastic Spring for the Discrete


Beam
This model permits elastic springs with damping to be combined and represented
with a discrete beam element type 6. Linear stiffness and damping coefficients can be
defined, and, for nonlinear behavior, a force versus deflection and force versus rate curves
can be used. Displacement based failure and an initial force are optional.

If the linear spring stiffness is used, the force, F, is given by:


𝐹 = 𝐹0 + 𝐾Δ𝐿 + 𝐷Δ𝐿̇, (22.74.1)
where K is the stiffness constant, and D is the viscous damping coefficient.

If the load curve ID for 𝑓 (Δ𝐿) is specified, nonlinear behavior is activated. For this
case the force is given by:
∣Δ𝐿̇∣
𝐹 = 𝐹0 + 𝐾 𝑓 (Δ𝐿) [1 + C1 ⋅ Δ𝐿̇ + C2 ⋅ sgn(Δ𝐿̇)ln (max {1. , })]
DLE (22.74.2)
+𝐷Δ𝐿̇ + 𝑔(Δ𝐿)ℎ(Δ𝐿̇),
where C1 and C2 are damping coefficients for nonlinear behavior, DLE is a factor to scale
time units, and 𝑔(Δ𝐿) is an optional load curve defining a scale factor versus deflection for
load curve ID, ℎ(𝑑Δ𝐿/𝑑𝑡).

In these equations, Δ𝐿 is the change in length


Δ𝐿 = current length-initial length. (22.74.3)

Failure can occur in either compression or tension based on displacement values of


CDF and TDF, respectively. After failure no forces are carried. Compressive failure does
not apply if the spring is initially zero length.

The cross sectional area is defined on the section card for the discrete beam
elements, in *SECTION_BEAM. The square root of this area is used as the contact
thickness offset if these elements are included in the contact treatment.

20-154 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.75 Material Model 75: Bilkhu/Dubois Foam Model


This model uses uniaxial and triaxial test data to provide a more realistic treatment
of crushable foam. The Poisson’s ratio is set to zero for the elastic response. The
volumetric strain is defined in terms of the relative volume, 𝑉, as:
𝛾 = −ln(𝑉). (22.75.1)

In defining the curves, the stress and strain pairs should be positive values starting
with a volumetric strain value of zero.
Max Stress

Uniaxial Yield Stress

Pressure Yield

Volumetric Strain
Figure 22.75.1. Behavior of crushable foam. Unloading is elastic.

LS-DYNA DEV 10/27/16 (r:8004) 20-155 (Material Models)


Material Models LS-DYNA Theory Manual

22.76 Material Model 76: General Viscoelastic


22.76.1 Introduction
Material type 76 in LS-DYNA is a general viscoelastic Maxwell model having up to 18
terms in the prony series expansion and is useful for modeling dense continuum rubbers
and solid explosives. It is characterized in the input by bulk and shear modulii, 𝐾𝑚 and
𝑔
𝐺𝑚 , and associated decay constants, 𝛽𝑘𝑚 and 𝛽𝑚 . Either the coefficients of the prony series
expansion can be used directly, or a relaxation curve may be specified to define the
viscoelastic deviatoric and bulk behavior.

22.76.2 Constitutive Model


The model is a hypoelastic version of the model given by Christensen and can be stated as

𝝈 ∇ = ∑(𝐾𝑚 𝑡𝑚 𝒊 + 2𝐺𝑚 𝒔∇
𝑚 ), (22.76.1)
𝑚

where 𝑡𝑚 and 𝒔𝑚 are (strain) quantities governed by the following evolution in time
𝑔
𝒔∇
𝑚 = 𝑫dev − 𝛽𝑚 𝒔𝑚 (22.76.2)
and

𝑡𝑚 = 𝐷vol − 𝛽𝑘𝑚 𝑡𝑚 (22.76.3)
𝑔
Here 𝐾𝑚 and 𝐺𝑚 are bulk and shear moduli, respectively, 𝛽𝑘𝑚 and 𝛽𝑚 are the corresponding
decay coefficients, 𝐷vol and 𝑫dev are the volumetric and deviatoric strain rates, 𝒊 is the 2nd
order identity tensor and ∇ denotes the Jaumann objective rate. It should immediately be
noted that for all decay coefficients equal to 0 (zero), the model is reduced to a time
independent elastic model,
𝝈 ∇ = 𝐾𝐷vol 𝒊 + 2𝐺𝑫dev (22.76.4)
with bulk and shear modulus given by 𝐾 = ∑𝑚 𝐾𝑚 and 𝐺 = ∑𝑚 𝐺𝑚 . For small
displacement theory, the stress can be integrated to be given by
𝑡 𝑘 𝑡 𝑔
𝝈(𝑡) = ∑ (𝐾𝑚 ∫ 𝑒−𝛽𝑚 (𝑡−𝜏) 𝐷vol (𝜏) 𝑑𝜏𝒊 + 2𝐺𝑚 ∫ 𝑒−𝛽𝑚(𝑡−𝜏) 𝑫dev (𝜏) 𝑑𝜏) (22.76.5)
𝑚 0 0

For shell elements, the same theory applies except that the objective rate ∇ is the
corotational time derivative instead of the Jaumann rate.

22.76.3 Tangent Modulus


For the implicit tangent modulus, we note that the internal force contribution depends on
the displacement and time, which we denote 𝒇int = 𝒇int (𝒖, 𝑡). The time derivative of this
vector is the sum of a material and a geometric contribution. The material contribution is
given by

20-156 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

̇
mat
𝒇int = ∫ 𝑩𝑇 𝝈 ∇𝑇 𝑑Ω (22.76.6)

where 𝑩 is the strain displacement matrix, the integration is over the current configuration
Ω and ∇𝑇 stands for Truesdell rate. This expression should later be identified with
mat
𝜕𝒇int 𝜕𝒇 mat
̇
mat
𝒇int = 𝒖̇ + int (22.76.7)
𝜕𝒖 𝜕𝑡
in order to determine the tangent modulus. Neglecting discrepancies between the
Jaumann and Truesdell rates, we can use (22.76.2) and (22.76.3) in (22.76.6) to get
̇
mat 𝑔
𝒇int = ∫ 𝑩𝑇 ∑(𝐾𝑚 𝒊⨂𝒊 + 2𝐺𝑚 𝑰dev ) 𝑩𝑑𝛺 𝒖̇ − ∫ 𝑩𝑇 ∑(𝐾𝑚 𝛽𝑘𝑚 𝑡𝑚 𝒊 + 2𝐺𝑚 𝛽𝑚 𝒔𝑚 ) 𝑑𝛺, (22.76.8)
𝑚 𝑚

where 𝑰dev is the 4th order deviatoric identity tensor. Comparing this expression with
(22.76.7) one can conclude that
mat
𝜕𝒇int
= ∫ 𝑩𝑇 ∑(𝐾𝑚 𝒊⨂𝒊 + 2𝐺𝑚 𝑰dev ) 𝑩𝑑𝛺, (22.76.9)
𝜕𝒖 𝑚

and hence the tangent modulus is


𝑪 = ∑(𝐾𝑚 𝒊⨂𝒊 + 2𝐺𝑚 𝑰dev ). (22.76.10)
𝑚
i.e., independent of deformation and time. In fact, this tangent modulus is equal to the classical
elastic tangent modulus for a hypoelastic material, cf. (22.76.4).

22.76.4 Using the Relaxation Curve

LS-DYNA DEV 10/27/16 (r:8004) 20-157 (Material Models)


Material Models LS-DYNA Theory Manual

Instead of inputting the stiffness and relaxation parameters, one can input a relaxation
curve from test according to Figure 22.76.1. The time scale is determined by BSTART and
LS-DYNA will determine all parameters as to best fit the curve.

Stress relaxation curve

10n 10n+1 10n+2 time


Optional ramp time for loading
Figure 22.76.1. Relaxation curve. This curve defines stress versus time where
time is defined on a logarithmic scale. For best results, the points defined in the
load curve should be equally spaced on the logarithmic scale. Furthermore, the
load curve should be smooth and defined in the positive quadrant. If nonphysical
values are determined by least squares fit, LS-DYNA will terminate with an error
message after the initialization phase is completed. If the ramp time for loading is
included, then the relaxation which occurs during the loading phase is taken into
account. This effect may or may not be important.

20-158 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.77 Material Model 77: Hyperviscoelastic Rubber


Material type 77 in LS-DYNA consists of two hyperelastic rubber models, a general
hyperelastic rubber model and an Ogden rubber model, that can be combined optionally
with a viscoelastic stress contribution. As for the rate independent part, the constitutive
law is determined by a strain energy function which in this case advantageously can be
expressed in terms of the principal stretches, i.e., 𝑊 = 𝑊(𝜆1 , 𝜆2 , 𝜆3 ). To obtain the Cauchy
stress 𝜎𝑖𝑗 , as well as the constitutive tensor of interest, 𝐷TC
𝑖𝑗𝑘𝑙 , they are first calculated in the
principal basis after which they are transformed back to the “base frame”, or standard
basis. The complete set of formulas is given by Crisfield [1997] and is for the sake of
completeness recapitulated here.
The principal Kirchoff stress components are given by
𝜕𝑊
𝜏𝑖𝑖E = 𝜆𝑖 (no sum), (22.77.1)
𝜕𝜆𝑖
that are transformed to the standard basis using the standard formula
E
𝜏𝑖𝑗 = 𝑞𝑖𝑘 𝑞𝑗𝑙 𝜏𝑘𝑙 . (22.77.2)

The 𝑞𝑖𝑗 are the components of the orthogonal tensor containing the eigenvectors of
the principal basis. The Cauchy stress is then given by
𝜎𝑖𝑗 = 𝐽 −1 𝜏𝑖𝑗 , (22.77.3)
where 𝐽 = 𝜆1 𝜆2 𝜆3 is the relative volume change.

The constitutive tensor that relates the rate of deformation to the Truesdell
(convected) rate of Kirchoff stress can in the principal basis be expressed as
𝜕𝜏𝑖𝑖E
𝐷TKE
𝑖𝑖𝑗𝑗 = 𝜆𝑗 − 2𝜏𝑖𝑖E 𝛿𝑖𝑗
𝜕𝜆𝑗
𝜆2𝑗 𝜏𝑖𝑖E − 𝜆2𝑖 𝜏𝑗𝑗E
𝐷TKE
𝑖𝑗𝑖𝑗 = , 𝑖 ≠ 𝑗, 𝜆𝑖 ≠ 𝜆𝑗 (no sum). (22.77.4)
𝜆2𝑖 − 𝜆2𝑗
𝜆𝑖 𝜕𝜏𝑖𝑖E 𝜕𝜏𝑖𝑖E
𝐷TKE
𝑖𝑗𝑖𝑗 = ( − ), 𝑖 ≠ 𝑗, 𝜆𝑖 = 𝜆𝑗
2 𝜕𝜆𝑖 𝜕𝜆𝑗

These components are transformed to the standard basis according to


𝐷TK TKE
𝑖𝑗𝑘𝑙 = 𝑞𝑖𝑝 𝑞𝑗𝑞 𝑞𝑘𝑟 𝑞𝑙𝑠 𝐷𝑝𝑞𝑟𝑠 , (22.77.5)
and finally the constitutive tensor relating the rate of deformation to the Truesdell rate of
Cauchy stress is obtained through

LS-DYNA DEV 10/27/16 (r:8004) 20-159 (Material Models)


Material Models LS-DYNA Theory Manual

𝐷TC −1 TK
𝑖𝑗𝑘𝑙 = 𝐽 𝐷𝑖𝑗𝑘𝑙 . (22.77.6)

When dealing with shell elements, the tangent moduli in the corotational
coordinates is of interest. This matrix is given by
̂ TC
𝐷 TC −1 TK −1
𝑖𝑗𝑘𝑙 = 𝑅𝑝𝑖 𝑅𝑞𝑗 𝑅𝑟𝑘 𝑅𝑠𝑙 𝐷𝑝𝑞𝑟𝑠 = 𝐽 𝑅𝑝𝑖 𝑅𝑞𝑗 𝑅𝑟𝑘 𝑅𝑠𝑙 𝐷𝑝𝑞𝑟𝑠 = 𝐽 𝑞𝑖𝑝 ̂ 𝑞𝑙𝑠̂ 𝐷TKE
̂ 𝑞𝑗𝑞̂ 𝑞𝑘𝑟 𝑝𝑞𝑟𝑠 , (22.77.7)
where 𝑅𝑖𝑗 is the matrix containing the unit basis vectors of the corotational system and𝑞𝑖𝑗̂ =
𝑅𝑘𝑖 𝑞𝑘𝑗 . The latter matrix can be determined as the eigenvectors of the co-rotated left
Cauchy-Green tensor (or the left stretch tensor). In LS-DYNA, the tangent stiffness matrix
is after assembly transformed back to the standard basis according to standard
transformation formulae.

22.77.1 General Hyperelastic Rubber Model


The strain energy function for the general hyperelastic rubber model is given by
𝑛
𝑝 𝑞 1
𝑊 = ∑ 𝐶𝑝𝑞 𝑊1 𝑊2 + 𝐾(𝐽 − 1)2 , (22.77.8)
𝑝,𝑞=0
2

where 𝐾 is the bulk modulus,


−1
𝑊1 = 𝐼1 𝐼3 3 −3
(22.77.9)
𝑊2 = 𝐼2 𝐼3−2/3 − 3,
and
𝐼1 = 𝜆21 + 𝜆22 + 𝜆23
𝐼2 = 𝜆21 𝜆22 + 𝜆22 𝜆23 + 𝜆21 𝜆23 (22.77.10)
𝐼3 = 𝜆21 𝜆22 𝜆23 ,
are the invariants in terms of the principal stretches. To apply the formulas in the previous
section, we require
𝑛
𝜕𝑊 𝑝−1 ′ 𝑞 𝑝 𝑞−1 ′
𝜏𝑖𝑖E = 𝜆𝑖 = ∑ 𝐶𝑝𝑞 (𝑝𝑊1 𝑊1𝑖 𝑊2 + 𝑞𝑊1 𝑊2 𝑊2𝑖 ) + 𝐾𝐽(𝐽 − 1), (22.77.11)
𝜕𝜆𝑖 𝑝,𝑞=0

where

′ 𝜕𝑊1 2 2 −1
𝑊1𝑖 ≔ 𝜆𝑖 = (2𝜆𝑖 − 𝐼1 ) 𝐼3 3
𝜕𝜆𝑖 3
(22.77.12)
′ 𝜕𝑊2 4 −2
𝑊2𝑖 : = 𝜆𝑖 = (2𝜆2𝑖 (𝐼1 − 𝜆2𝑖 ) − 𝐼2 ) 𝐼3 3 .
𝜕𝜆𝑖 3

20-160 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

If C𝑝𝑞 is nonzero only for 𝑝𝑞 = 01,10,11,20,02,30, then Equation (22.77.11) can be


written as
𝜏𝑖𝑖E = (𝐶10 + 𝐶11 𝑊2 + 2𝐶20 𝑊1 + 3𝐶30 𝑊12 )𝑊1𝑖

+
(22.77.13)

(𝐶01 + 𝐶11 𝑊1 + 2𝐶02 𝑊2 )𝑊2𝑖 + 𝐾𝐽(𝐽 − 1).

Proceeding with the constitutive tensor, we have


𝜕𝜏𝑖𝑖E 𝑛
𝑝−2 ′ ′ 𝑞 𝑝−1 ′′ 𝑞 𝑝−1 ′ 𝑞−1 ′
𝜆𝑗 = ∑ 𝐶𝑝𝑞 (𝑝(𝑝 − 1)𝑊1 𝑊1𝑖 𝑊1𝑗 𝑊2 + 𝑝𝑊1 𝑊1𝑖𝑗 𝑊2 + 𝑝𝑞𝑊1 𝑊1𝑖 𝑊2 𝑊2𝑗
𝜕𝜆𝑗 𝑝,𝑞=0
𝑝−1 𝑞−1 𝑝 𝑞−2 𝑝 𝑞−1
(22.77.14)
′ ′ ′ ′ ′′
+𝑞𝑝𝑊1 𝑊1𝑗 𝑊2 𝑊2𝑖 + 𝑞(𝑞 − 1)𝑊1 𝑊2 𝑊2𝑖 𝑊2𝑗 + 𝑞𝑊1 𝑊2 𝑊2𝑖𝑗 )
+𝐾𝐽(2𝐽 − 1),
where

𝜕𝑊1𝑖 4 4
′′
𝑊1𝑖𝑗 : = 𝜆𝑗 = (4𝜆2𝑖 𝛿𝑖𝑗 − (𝜆2𝑖 + 𝜆2𝑗 ) + 𝐼1 )𝐼3−1/3
𝜕𝜆𝑗 3 9
′ (22.77.15)
′′ 𝜕𝑊2𝑖 2 4 2 2 8 2 2 2 2 16 −2/3
𝑊2𝑖𝑗 : = 𝜆𝑗 = ((4𝜆𝑖 𝐼1 − 8𝜆𝑖 )𝛿𝑖𝑗 + 4𝜆𝑖 𝜆𝑗 − (𝜆𝑖 (𝐼1 − 𝜆𝑖 ) + 𝜆𝑗 (𝐼1 − 𝜆𝑗 )) + 𝐼2 )𝐼3
𝜕𝜆𝑗 3 9

Again, using only the nonzero coefficients mentioned above, Equation (22.77.14) is
reduced to
𝜕𝜏𝑖𝑖E ′ ′ ′ ′ ′ ′ ′ ′
𝜆𝑗 = 𝐶11 (𝑊1𝑖 𝑊2𝑗 + 𝑊1𝑗 𝑊2𝑖 ) + 2(𝐶20 + 3𝐶30 𝑊1 )𝑊1𝑗 𝑊1𝑖 + 2𝐶02 𝑊2𝑖 𝑊2𝑗 +
𝜕𝜆𝑗
(22.77.16)
(𝐶10 + 𝐶11 𝑊2 + 2𝐶20 𝑊1 + 3𝐶30 𝑊12 )𝑊1𝑖𝑗
′′ ′′
+ (𝐶01 + 𝐶11 𝑊1 + 2𝐶02 𝑊2 )𝑊2𝑖𝑗 +
𝐾𝐽(2𝐽 − 1).

22.77.2 Ogden Rubber Model


The strain energy function for the Ogden rubber model is given by
𝑛
𝜇𝑚 𝛼 𝑚 𝛼 𝛼 1
𝑊= ∑ (𝜆̃ 1 + 𝜆̃ 2𝑚 + 𝜆̃ 3𝑚 − 3) + 𝐾(𝐽 − 1)2 , (22.77.17)
𝛼
𝑚=1 𝑚
2
where
𝜆𝑖
𝜆̃ 𝑖 = , (22.77.18)
𝐽 1/3
are the volumetric independent principal stretches, and 𝜇𝑚 and 𝛼𝑚 are material parameters.
To apply the formulas in the previous section, we require
𝑛
𝜕𝑊 1
𝜏𝑖𝑖E = 𝜆𝑖 = ∑ 𝜇𝑚 (𝜆̃ 𝛼𝑖 𝑚 − 𝑎𝑚 ) + 𝐾𝐽(𝐽 − 1), (22.77.19)
𝜕𝜆𝑖 𝑚=1 3

LS-DYNA DEV 10/27/16 (r:8004) 20-161 (Material Models)


Material Models LS-DYNA Theory Manual

where
𝛼 𝛼 𝛼
𝑎𝑚 = 𝜆̃ 1𝑚 + 𝜆̃ 2𝑚 + 𝜆̃ 3𝑚 . (22.77.20)

Proceeding with the constitutive tensor, we have


𝜕𝜏𝑖𝑖E 𝑛
𝜇 𝛼 1 𝛼 𝛼 𝛼
𝜆𝑗 = ∑ 𝑚 𝑚 ( 𝑎𝑚 + 3𝜆̃ 𝑖 𝑚 𝛿𝑖𝑗 − 𝜆̃ 𝑖 𝑚 − 𝜆̃ 𝑗 𝑚 ) + 𝐾𝐽(2𝐽 − 1). (22.77.21)
𝜕𝜆𝑗 𝑚=1
3 3

22.77.3 The Viscoelastic Contribution


As mentioned above, this material model is accompanied with a viscoelastic stress
contribution. The rate form of this constitutive law can in co-rotational coordinates be
written
𝑛 𝑛 𝑡

𝜎̂ 𝑖𝑗ve ̂ dev
= ∑ 2𝐺𝑚 𝐷 𝑖𝑗 − ∑ 2𝛽𝑚 𝐺𝑚 ∫ 𝑒
−𝛽𝑚 (𝑡−𝜏) ̂ dev
𝐷𝑖𝑗 (𝜏)𝑑𝜏 . (22.77.22)
𝑚=1 𝑚=1 0

Here 𝑛 is a number less than or equal to 6, 𝜎̂ 𝑖𝑗ve is the co-rotated viscoelastic stress,
̂ dev
𝐷 𝑖𝑗 is the deviatoric co-rotated rate-of-deformation and 𝐺𝑚 and 𝛽𝑚 are material
parameters. The parameters 𝐺𝑚 can be thought of as shear moduli and 𝛽𝑚 as decay
coefficients determining the relaxation properties of the material. This rate form can be
integrated in time to form the corotated viscoelastic stress
6 𝑡
𝜎̂ 𝑖𝑗ve = ∑ 2𝐺𝑚 ∫ 𝑒−𝛽𝑚(𝑡−𝜏) 𝐷
̂ dev
𝑖𝑗 (𝜏)𝑑𝜏 . (22.77.23)
𝑚=1 0

For the constitutive matrix, we refer to Borrvall [2002] and here simply state that it is
equal to
𝑛
1 1
̂ TCve
𝐷 𝑖𝑗𝑘𝑙 = ∑ 2𝐺𝑚 ( (𝛿𝑖𝑘 𝛿𝑗𝑙 + 𝛿𝑖𝑙 𝛿𝑗𝑘 ) − 𝛿𝑖𝑗 𝛿𝑘𝑙 ). (22.77.24)
𝑚=1
2 3

22.77.4 Stress Update for Shell Elements


In principal, the stress update for material 77 and shell elements follows closely the
one that is implemented for material 27. The stress is evaluated in corotational coordinates
after which it is transformed back to the standard basis according to
𝜎𝑖𝑗 = 𝑅𝑖𝑘 𝑅𝑗𝑙 𝜎̂ 𝑘𝑙 , (22.77.25)

20-162 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

or equivalently the internal force is assembled in the corotational system and then
transformed back to the standard basis according to standard transformation formulae.
Here 𝑅𝑖𝑗 is the rotation matrix containing the corotational basis vectors. The so-called
corotated stress 𝜎̂ 𝑖𝑗 is evaluated as the sum of the stresses given in Sections 19.77.1 and
19.77.4.

The viscoelastic stress contribution is incrementally updated with aid of the


corotated rate of deformation. To be somewhat more precise, the values of the 12 integrals
in Equation (22.77.23) are kept as history variables that are updated in each time step. Each
integral is discretized in time and the mean value theorem is used in each time step to
determine their values.

For the hyperelastic stress contribution, the principal stretches are needed and here
taken as the square root of the eigenvalues of the co-rotated left Cauchy-Green tensor 𝑏𝑖𝑗̂ .
The corotated left Cauchy-Green tensor is incrementally updated with the aid of a time
increment Δ𝑡, the corotated velocity gradient 𝐿̂𝑖𝑗 , and the angular velocity 𝛺̂𝑖𝑗 with which
the embedded coordinate system is rotating,
𝑏̂𝑖𝑗 = 𝑏̂𝑖𝑗 + Δ𝑡(𝐿̂𝑖𝑘 − 𝛺
̂𝑖𝑘 )𝑏̂𝑘𝑗 + Δ𝑡𝑏𝑖𝑘
̂ (𝐿̂𝑖𝑘 − 𝛺
̂𝑖𝑘 ). (22.77.26)

The primary reason for taking a corotational approach is to facilitate the


maintenance of a vanishing normal stress through the thickness of the shell, something that
is achieved by adjusting the corresponding component of the corotated velocity gradient
𝐿̂33 accordingly. The problem can be stated as to determine L̂33 such that when updating
the left Cauchy-Green tensor through Equation (22.77.26) and subsequently the stress
through formulae in Sections 19.77.1 and 19.77.4, 𝜎̂ 33 = 0. To this end, it is assumed that
𝐿̂33 = α(𝐿̂11 + 𝐿̂22 ), (22.77.27)
for some parameter α that is determined in the following three step procedure. In the first
(0)
two steps, α = 0 and α = −1, respectively, resulting in two trial normal stresses 𝜎̂ 33 and
(−1)
𝜎̂ 33 . Then it is assumed that the actual normal stress depends linearly on α, meaning that
the latter can be determined from
(α) (0) (0) (−1)
0 = 𝜎33 = 𝜎33 + α(𝜎33 − 𝜎33 ). (22.77.28)

In the current implementation, α is given by


⎧ (0)
{ 𝜎̂ 33 (−1) (0)
{ (−1) (0)
if ∣𝜎̂ 33 − 𝜎̂ 33 ∣ ≥ 10−4
{𝜎̂ 33 − 𝜎̂ 33
𝛼=⎨ (22.77.29)
{
{ −1 otherwise
{

LS-DYNA DEV 10/27/16 (r:8004) 20-163 (Material Models)


Material Models LS-DYNA Theory Manual

and the stresses are determined from this value of 𝛼. Finally, to make sure that the normal
stress through the thickness vanishes, it is set to 0 (zero) before exiting the stress update
routine.

20-164 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.78 Material Model 78: Soil/Concrete


Concrete pressure is positive in compression. Volumetric strain is defined as the
natural log of the relative volume and is positive in compression where the relative
volume, 𝑉, is the ratio of the current volume to the initial volume. The tabulated data
should be given in order of increasing compression. If the pressure drops below the cutoff
value specified, it is reset to that value and the deviatoric stress state is eliminated.

If the load curve ID is provided as a positive number, the deviatoric perfectly plastic
pressure dependent yield function 𝜙, is described in terms of the second invariant, 𝐽2 , the
pressure, 𝑝, and the tabulated load curve, 𝐹(𝑝), as
𝜙 = √3𝐽2 − 𝐹(𝑝) = σ𝑦 − 𝐹(𝑝), (22.78.1)
where 𝐽2 is defined in terms of the deviatoric stress tensor as:
1
𝐽2 = 𝑆𝑖𝑗 𝑆𝑖𝑗 , (22.78.2)
2
assuming that if the ID is given as negative, then the yield function becomes:
𝜙 = 𝐽2 − 𝐹(𝑝), (22.78.3)
being the deviatoric stress tensor.

If cracking is invoked, the yield stress is multiplied by a factor f which reduces with
plastic stain according to a trilinear law as shown in Figure 22.78.1.

1.0

ε1 ε2 εp

Figure 22.78.1. Strength reduction factor.

LS-DYNA DEV 10/27/16 (r:8004) 20-165 (Material Models)


Material Models LS-DYNA Theory Manual

ε
ε2

ε1

P
Figure 22.78.2. Cracking strain versus pressure.

b = residual strength factor


𝜀1 = plastic stain at which cracking begins.
𝜀2 = plastic stain at which residual strength is reached.

𝜀1 and 𝜀2 are tabulated functions of pressure that are defined by load curves (see Figure
22.78.2). The values on the curves are pressure versus strain and should be entered in
order of increasing pressure. The strain values should always increase monotonically with
pressure.

By properly defining the load curves, it is possible to obtain the desired strength and
ductility over a range of pressures. See Figure 22.78.3.

p3
p2
Yield
stress p1

Plastic strain
Figure 22.78.3. Example Caption

20-166 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.79 Material Model 79: Hysteretic Soil


This model is a nested surface model with five superposed “layers” of elasto-
perfectly plastic material, each with its own elastic modulii and yield values. Nested
surface models give hysteretic behavior, as the different “layers” yield at different stresses.

The constants 𝑎0 , 𝑎1 , 𝑎2 govern the pressure sensitivity of the yield stress. Only the
ratios between these values are important - the absolute stress values are taken from the
stress-strain curve.

The stress strain pairs (𝛾1, 𝜏1), ... (𝛾5, 𝜏5) define a shear stress versus shear strain
curve. The first point on the curve is assumed by default to be (0,0) and does not need to be
entered. The slope of the curve must decrease with increasing 𝛾. Not all five points need
be to be defined. This curve applies at the reference pressure; at other pressures the curve
varies according to 𝑎0 , 𝑎1 , and 𝑎2 as in the soil and crushable foam model, Material 5.

The elastic moduli 𝐺 and 𝐾 are pressure sensitive.


𝐺 = 𝐺0 (𝑝 − 𝑝0 )𝑏 ,
(22.79.1)
𝐾 = 𝐾0 (𝑝 − 𝑝0 )𝑏 ,
where 𝐺0 and 𝐾0 are the input values, 𝑝 is the current pressure, 𝑝0 the cut-off or reference
pressure (must be zero or negative). If 𝑝 attempts to fall below 𝑝0 (i.e., more tensile) the
shear stresses are set to zero and the pressure is set to 𝑝0 . Thus, the material has no
stiffness or strength in tension. The pressure in compression is calculated as follows:
1⁄
𝑝 = [−𝐾0 ln(𝑉)] 1−𝑏 , (22.79.2)
where 𝑉 is the relative volume, i.e., the ratio between the original and current volume.

LS-DYNA DEV 10/27/16 (r:8004) 20-167 (Material Models)


Material Models LS-DYNA Theory Manual

22.80 Material Model 80: Ramberg-Osgood Plasticity


The Ramberg-Osgood equation is an empirical constitutive relation to represent the
one-dimensional elastic-plastic behavior of many materials, including soils. This model
allows a simple rate independent representation of the hysteretic energy dissipation
observed in soils subjected to cyclic shear deformation. For monotonic loading, the stress-
strain relationship is given by:
𝛾 𝜏 𝜏 𝑟
= + 𝛼∣ ∣ if 𝛾 ≥ 0,
𝛾𝑦 𝜏𝑦 𝜏𝑦
(22.80.1)
𝛾 𝜏 𝜏 𝑟
= − 𝛼∣ ∣ if 𝛾 < 0,
𝛾𝑦 𝜏𝑦 𝜏𝑦
where 𝛾 is the shear and 𝜏 is the stress. The model approaches perfect plasticity as the
stress exponent 𝑟 → ∞. These equations must be augmented to correctly model unloading
and reloading material behavior. The first load reversal is detected by 𝛾𝛾̇ < 0. After the
first reversal, the stress-strain relationship is modified to
(𝛾 − 𝛾0 ) (𝜏 − 𝜏0 ) (𝜏 − 𝜏0 ) 𝑟
= + 𝛼∣ ∣ if 𝛾 ≥ 0,
2𝛾𝑦 2𝜏𝑦 2𝜏𝑦
(22.80.2)
(𝛾 − 𝛾0 ) (𝜏 − 𝜏0 ) (𝜏 − 𝜏0 ) 𝑟
= − 𝛼∣ ∣ if 𝛾 < 0,
2𝛾𝑦 2𝜏𝑦 2𝜏𝑦
where 𝛾0 and 𝜏0 represent the values of strain and stress at the point of load reversal.
Subsequent load reversals are detected by (𝛾 − 𝛾0 )𝛾̇ < 0.

The Ramberg-Osgood equations are inherently one-dimensional and are assumed to


apply to shear components. To generalize this theory to the multidimensional case, it is
assumed that each component of the deviatoric stress and deviatoric tensorial strain is
independently related by the one-dimensional stress-strain equations. A projection is used
to map the result back into deviatoric stress space if required. The volumetric behavior is
elastic, and, therefore, the pressure 𝑝 is found by
𝑝 = −𝐾𝜀𝑣 , (22.80.3)
where 𝜀𝑣 is the volumetric strain.

20-168 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.81 Material Models 81 and 82: Plasticity with Damage


and Orthotropic Option
With this model an elasto-viscoplastic material with an arbitrary stress versus strain
curve and arbitrary strain rate dependency can be defined. Damage is considered before
rupture occurs. Also, failure based on a plastic strain or a minimum time step size can be
defined.

An option in the keyword input, ORTHO, is available, which invokes an orthotropic


damage model. This option is an extension to include orthotropic damage as a means of
treating failure in aluminum panels. Directional damage begins after a defined failure
strain is reached in tension and continues to evolve until a tensile rupture strain is reached
in either one of the two orthogonal directions.

The stress versus strain behavior may be treated by a bilinear stress strain curve by
defining the tangent modulus, ETAN. Alternately, a curve similar to that shown in Figure
22.81.1 is expected to be defined by (EPS1,ES1) - (EPS8,ES8); however, an effective stress
versus effective plastic strain curve (LCSS) may be input instead if eight points are
insufficient. The cost is roughly the same for either approach. The most general approach
yield stress versus
effective plastic strain
σyield for undamaged material

Failure Begins

nominal stress
after failure

damage increases
linearly with plastic
strain after failure rupture

0 ω=0 ω=1 εpeff


Figure .22.81.1. Stress strain behavior when damage is included.

is to use the table definition (LCSS) discussed below.

Two options to account for strain rate effects are possible. Strain rate may be
accounted for using the Cowper-Symonds model which scales the yield stress with the
factor,

LS-DYNA DEV 10/27/16 (r:8004) 20-169 (Material Models)


Material Models LS-DYNA Theory Manual

1
𝜀̇ ⁄𝑝 (22.81.1)
1+( ) ,
𝐶
where 𝜀̇ is the strain rate, 𝜀̇ = √𝜀̇𝑖𝑗 𝜀̇𝑖𝑗 . If the viscoplastic option is active, VP = 1.0, and if
SIGY is > 0 then the dynamic yield stress is computed from the sum of the static stress,
p
𝜎ys (𝜀eff ), which is typically given by a load curve ID, and the initial yield stress, SIGY,
multiplied by the Cowper-Symonds rate term as follows:
p 1⁄
p
p p p 𝜀̇ (22.81.2)
𝜎𝑦 (𝜀eff , 𝜀̇eff ) = 𝜎𝑦s (𝜀eff ) + SIGY ⋅ ( eff ) ,
C
where the plastic strain rate is used. With this latter approach similar results can be
obtained between this model and material model: *MAT_ANISOTROPIC_VISCOPLASTIC.
p
If SIGY = 0, the following equation is used instead where the static stress, 𝜎ys (𝜀eff ), must be
defined by a load curve:
p 1⁄
p
p p p ⎡ 𝜀̇ ⎤
𝜎y (𝜀eff , 𝜀̇eff ) = 𝜎ys (𝜀eff ) ⎢
⎢1 + ( eff ) ⎥.
⎥ (22.81.3)
C
⎣ ⎦

This latter equation is always used if the viscoplastic option is off. For complete
generality a load curve (LCSR) to scale the yield stress may be input instead. In this curve
the scale factor versus strain rate is defined.

The constitutive properties for the damaged material are obtained from the
undamaged material properties. The amount of damage evolved is represented by the
constant, 𝜔, which varies from zero if no damage has occurred to unity for complete
rupture. For uniaxial loading, the nominal stress in the damaged material is given by
𝑃
𝜎nominal =
, (22.81.4)
𝐴
where P is the applied load and A is the surface area. The true stress is given by:
𝑃
𝜎true = , (22.81.5)
𝐴 − 𝐴loss
where 𝐴loss is the void area. The damage variable can then be defined:
𝐴loss
𝜔= , 0 ≤ 𝜔 ≤ 1. (22.81.6)
𝐴

In this model damage is defined in terms of plastic strain after the failure strain is
exceeded:
p p
𝜀eff − 𝜀failure p p p
𝜔= p p if 𝜀failure ≤ 𝜀eff ≤ 𝜀rupture . (22.81.7)
𝜀rupture − 𝜀failure

20-170 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

After exceeding the failure strain softening begins and continues until the rupture
strain is reached.

By default, deletion of a shell element occurs when all integration points in the shell
have failed. A parameter is available, NUMINT, that defines the number of through
thickness integration points for shell element deletion. The default of all integration points
is not recommended since shells undergoing large strain are often not deleted due to nodal
fiber rotations which limit strains at active integration points after most points have failed.
Better results are obtained if NUMINT is set to 1 or a number less than one half of the
number of through thickness points. For example, if four through thickness points are
used, NUMINT should not exceed 2, even for fully integrated shells which have 16
integration points.

22.81.1 Material Model 82: Isotropic Elastic-Plastic with Anisotropic Damage


Material 82 is an isotropic elastic-plastic material model with anisotropic damage..
The stress update in the case of shell elements is performed as follows. For a given stress
state 𝜎𝑖𝑗𝑡 and damage parameters 𝐷𝑡𝑖 , 𝑖 = 1, 2 at time 𝑡, the local stress is obtained as
𝑡
𝜎𝑖𝑗𝑙 = 𝑞𝑘𝑖 𝑞𝑙𝑗 𝜎𝑘𝑙 , (22.81.8)
where 𝑞𝑖𝑗 is an orthogonal matrix determining the direction of the damage. The directions
are determined as follows. The first direction is taken as the one in which the plastic strain
first reaches the plastic strain at impending failure, see below. The other direction is
orthogonal to the first and in the plane of the shell.

For this local stress, the undamaged stress is computed as


1
damage

0 failure
ε peff- f s

Figure 22.81.2. A nonlinear damage curve is optional. Note that the origin of the
curve is at (0,0). It is permissible to input the failure strain, fs, as zero for this
option. The nonlinear damage curve is useful for controlling the softening
behavior after the failure strain is reached.

l
u 𝜎11
𝜎11 = , (22.81.9)
1 − D1t

LS-DYNA DEV 10/27/16 (r:8004) 20-171 (Material Models)


Material Models LS-DYNA Theory Manual

l
u 𝜎22
𝜎22 = ,
1 − D2𝑡
l
u 2𝜎12
𝜎12 = ,
2 − 𝐷𝑡1 − 𝐷𝑡2
l
u 𝜎23
𝜎23 = ,
1 − 𝐷𝑡2
l
u 𝜎13
𝜎13 = .
1 − 𝐷𝑡1

A new undamaged stress 𝜎𝑖𝑗u+ is then computed following a standard elastic-plastic


stress update. The damage at the next time step is computed according to
p
𝑡 𝜀𝑖𝑖 − 𝜀f
𝐷𝑡+
𝑖 = max (𝐷𝑖 , ), 𝑖 = 1, 2, (22.81.10)
𝜀r − 𝜀f
p
where 𝜀f is the plastic strain at impending failure, 𝜀r is the plastic strain at rupture and 𝜀𝑖𝑖 is
the current plastic strain in the local 𝑖 direction. There is also an option of defining a
nonlinear damage curve, with this option the new damage is computed as
p
𝐷𝑡+ 𝑡
𝑖 = max(𝐷𝑖 , 𝑓 (𝜀𝑖𝑖 − 𝜀f )), 𝑖 = 1, 2, (22.81.11)
for a user-defined load curve 𝑓 .

The new local (damaged) stress is given by


l+ u+
𝜎11 = 𝜎11 (1 − 𝐷𝑡+
1 ),
l+ u+
𝜎22 = 𝜎22 (1 − 𝐷𝑡+
2 ),

l+ u+ 2 − 𝐷t+ t+
1 − 𝐷2
𝜎12 = 𝜎12 , (22.81.12)
2
l+ u+
𝜎23 = 𝜎23 (1 − 𝐷t+
2 ),
l+ u+
𝜎13 = 𝜎13 (1 − 𝐷t+
1 ),

which is transformed back to the local system to obtain the new global damaged stress as
𝑙+
𝜎𝑖𝑗𝑡+ = 𝑞𝑖𝑘 𝑞𝑗𝑙 𝜎𝑘𝑙 . (22.81.13)

An integration point is completely failed, i.e., it is removed from the calculations,


when max(𝐷1 , 𝐷2 ) > 0.999. The element is removed from the model when a user specified
number of integration points in that element have failed.

There are options of using visco-plasticity in the current model. The details of this
part of the stress update is omitted here.
20-172 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Material Models

The Rc-Dc Damage Model


The Rc-Dc model is defined as the following, see the report on the Fundamental
Study of Crack Initiation and Propagation [2003]. The damage 𝐷 is updated as
𝐷𝑡+ = 𝐷𝑡 + 𝜔1 𝜔2 Δ𝜀p (22.81.14)
where Δ𝜀p is the plastic strain increment and
𝜔1 = (1 + 𝛾𝑝)−𝛼 ,
(22.81.15)
𝜔2 = (2 − 𝐴𝐷 )𝛽 .

Here 𝑝 is the pressure, 𝛼, 𝛽 and 𝛾 are material parameters and



{
{1.9999 if max(𝑆1 , 𝑆2 ) ≤ 0
𝐴𝐷 = ⎨ . (22.81.16)
{ 𝑆1 𝑆2
{min (∣ ∣ , ∣ ∣) otherwise
⎩ 𝑆2 𝑆1
where 𝑆1 and 𝑆2 are the in-plane principal stress values. Fracture is initiated when the
accumulation of damage is greater than a critical damage 𝐷c given by
𝐷c = 𝐷0 (1 + 𝑏‖∇𝐷‖𝜆 ). (22.81.17)
Here 𝐷0 , 𝑏 and λ are material parameters and ∇D is the spatial gradient of damage. We
have added an option to use a non-local formulation with 𝐷 as the non-local variable and a
characteristic length 𝑙. More information on this can be found in the LS-DYNA Keyword
User’s Manual [Hallquist 2003]. With this option we compute 𝐷c as,
𝐷c = 𝐷0 , (22.81.18)
hence the parameters 𝑏 and 𝜆 are not used. A fracture fraction given by
𝐷 − 𝐷c
𝐹= (22.81.19)
𝐷s
defines the degradations of the material by the Rc-Dc model. Here 𝐷s is yet another
parameter determined by the user. The stress update of material 82 is modified
accordingly.

Upon entry the stress is divided by the factor 1 − 𝐹𝑡 to account for the Rc-Dc
damage. Before exiting the routine, the stress is multiplied by the new Rc-Fc (reversed)
fracture fraction 1 − 𝐹𝑡+ . An integration point is considered failed when min(1 − 𝐷1 , 1 −
𝐷2 )(1 − 𝐹) < 0.001.

LS-DYNA DEV 10/27/16 (r:8004) 20-173 (Material Models)


Material Models LS-DYNA Theory Manual

22.83 Material Model 83: Fu-Chang’s Foam With Rate


Effects
This model allows rate effects to be modeled in low and medium density foams, see
Figure 22.83.1. Hysteretic unloading behavior in this model is a function of the rate
sensitivity with the most rate sensitive foams providing the largest hysteresis and visa
versa. The unified constitutive equations for foam materials by Fu-Chang [1995] provide
the basis for this model. This implementation incorporates the coding in the reference in
modified form to ensure reasonable computational efficiency. The mathematical
description given below is excerpted from the reference.

The strain is divided into two parts: a linear part and a non-linear part of the strain

𝐄(𝑡) = 𝐄L (𝑡) + 𝐄N (𝑡), (22.83.1)


and the strain rate becomes
𝐄̇ (𝑡) = 𝐄̇ L (𝑡) + 𝐄̇ N (𝑡). (22.83.2)
𝐄̇ N is an expression for the past history of 𝐄N . A postulated constitutive equation may be
written as:

1-V
Figure 22.83.1. Rate effects in Fu Chang’s foam model.

20-174 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models


𝛔(𝑡) = ∫ [𝐄N
𝑡 (𝜏), 𝐒(𝑡)] 𝑑𝜏, (22.83.3)
𝜏=0

where 𝐒(𝑡) is the state variable and ∫𝜏=0 is a functional of all values of 𝜏 in 𝑇𝜏 : 0 ≤ 𝜏 ≤ ∞
and
𝐄N N
𝑡 (𝜏) = 𝐄 (𝑡 − 𝜏), (22.83.4)
where 𝜏 is the history parameter:
𝐄N
𝑡 (𝜏 = ∞) ⇔ the virgin material. (22.83.5)

It is assumed that the material remembers only its immediate past, i.e., a
neighborhood about 𝜏 = 0. Therefore, an expansion of 𝐄N
𝑡 (𝜏) in a Taylor series about 𝜏 =
0 yields:
𝜕𝐄N
𝑡
𝐄N N
𝑡 (𝜏) = 𝐄 (0) + (0)𝑑𝑡. (22.83.6)
𝜕𝑡
Hence, the postulated constitutive equation becomes:

𝛔(𝑡) = 𝛔∗ (𝐄N (𝑡), 𝐄̇ N (𝑡), 𝐒(𝑡)), (22.83.7)


∂𝐄N
where we have replaced 𝑡
∂𝑡
by 𝐄̇ N , and 𝛔∗ is a function of its arguments.

For a special case,

𝛔(𝑡) = 𝛔∗ (𝐄̇ N (𝑡), 𝐒(𝑡)), (22.83.8)

we may write
𝐄̇ N
𝑡 = 𝑓 (𝐒(𝑡), 𝐬(𝑡)), (22.83.9)
which states that the nonlinear strain rate is the function of stress and a state variable which
represents the history of loading. Therefore, the proposed kinetic equation for foam
materials is:
𝛔 tr(𝛔𝐒) 2𝑛0
𝐄̇ N = 𝐷 exp [−𝑐0 ( ) ], (22.83.10)
‖𝛔‖ 0 (‖𝛔‖)2
where 𝐷0 , 𝑐0 , and 𝑛0 are material constants, and 𝐒 is the overall state variable. If either
𝐷0 = 0 or 𝑐0 → ∞ then the nonlinear strain rate vanishes.
𝑛2
𝑆𝑖𝑗̇ = [𝑐1 (𝑎𝑖𝑗 𝑅 − 𝑐2 𝑆𝑖𝑗 )𝑃 + 𝑐3 𝑊 𝑛1 (∥𝐸̇𝑁 ∥) 𝐼𝑖𝑗 ]𝑅 (22.83.11)

𝑛3
∥𝐄̇ N ∥

𝑅 = 1 + 𝑐4 ⎜ ⎞
− 1⎟ (22.83.12)
⎝ 𝑐5 ⎠

𝑃 = tr(𝛔𝐄̇ N ) (22.83.13)

LS-DYNA DEV 10/27/16 (r:8004) 20-175 (Material Models)


Material Models LS-DYNA Theory Manual

W = ∫ tr(𝛔𝑑𝐄), (22.83.14)

where 𝑐1 , 𝑐2 , 𝑐3 , 𝑐4 , 𝑐5 , 𝑛1 , 𝑛2 , 𝑛3 , and 𝑎𝑖𝑗 are material constants and:


1
‖𝛔‖ = (𝜎𝑖𝑗 𝜎𝑖𝑗 )2 ,
1
∥𝐄̇ ∥ = (𝐸̇𝑖𝑗 𝐸̇𝑖𝑗 )2 , (22.83.15)
1
∥𝐄̇ N ∥ = (𝐸̇N ̇N 2
𝑖𝑗 𝐸𝑖𝑗 ) .

In the implementation by Fu Chang the model was simplified such that the input
constants 𝑎𝑖𝑗 and the state variables 𝑆𝑖𝑗 are scalars.

20-176 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.84 Material Model 84 and 85: Winfrith Concrete


Pressure is positive in compression; volumetric strain is given by the natural log of
the relative volume and is negative in compression. The tabulated data are given in order
of increasing compression, with no initial zero point.

If the volume compaction curve is omitted, the following scaled curve is


automatically used where 𝑝1 is the pressure at uniaxial compressive failure computed from:
𝜎
𝑝1 = 𝑐 , (22.84.1)
3
and 𝐾 is the unloading bulk modulus computed from
𝐸s
𝐾= , (22.84.2)
3(1 − 2𝑣)
where 𝐸s is the input tangent modulus for concrete and 𝑣 is Poisson's ratio.

Volumetric Strain Pressure (MPa)


−𝑝1 /K 1.00 × 𝑝1
-0.002 1.50 × 𝑝1
-0.004 3.00 × 𝑝1
-0.010 4.80 × 𝑝1
-0.020 6.00 × 𝑝1
-0.030 7.50 × 𝑝1
-0.041 9.45 × 𝑝1
-0.051 11.55 × 𝑝1
-0.062 14.25 × 𝑝1
-0.094 25.05 × 𝑝1
Table 19.84.1.Default pressure versus volumetric strain curve for concrete if the curve is
not defined.

LS-DYNA DEV 10/27/16 (r:8004) 20-177 (Material Models)


Material Models LS-DYNA Theory Manual

22.87 Material Model 87: Cellular Rubber


This material model provides a cellular rubber model combined with linear
viscoelasticity as outlined by Christensen [1980].

Rubber is generally considered to be fully incompressible since the bulk modulus


greatly exceeds the shear modulus in magnitude. To model the rubber as an unconstrained
material a hydrostatic work term, 𝑊𝐻 (𝐽), is included in the strain energy functional which
is function of the relative volume, 𝐽, [Ogden, 1984]:
𝑛
𝑊(𝐽1 , 𝐽2 , 𝐽) = ∑ 𝐶𝑝𝑞 (𝐽1 − 3)𝑝 (𝐽2 − 3)𝑞 + 𝑊𝐻 (𝐽)
𝑝,𝑞=0

−1⁄3 (22.87.1)
𝐽1 = 𝐼1 𝐼3
−2⁄3
𝐽2 = 𝐼2 𝐼3

In order to prevent volumetric work from contributing to the hydrostatic work the
first and second invariants are modified as shown. This procedure is described in more
detail by Sussman and Bathe [1987].

The effects of confined air pressure in its overall response characteristics are
included by augmenting the stress state within the element by the air pressure.
𝜎𝑖𝑗 = 𝜎𝑖𝑗sk − 𝛿𝑖𝑗 𝜎 air , (22.87.2)

where 𝜎𝑖𝑗sk is the bulk skeletal stress and σair is the air pressure computed from the
equation:
𝑝0 𝛾
𝜎 air = − , (22.87.3)
1+𝛾−𝜙
where 𝑝0 is the initial foam pressure usually taken as the atmospheric pressure and 𝛾
defines the volumetric strain
𝛾 = 𝑉 − 1 + 𝛾0 , (22.87.4)
where 𝑉 is the relative volume of the voids and 𝛾0 is the initial volumetric strain which is
typically zero. The rubber skeletal material is assumed to be incompressible.

Rate effects are taken into account through linear viscoelasticity by a convolution
integral of the form:
𝑡 𝜕𝜀𝑘𝑙
𝜎𝑖𝑗 = ∫ 𝑔𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) 𝑑𝜏, (22.87.5)
0 𝜕𝜏

20-178 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

or in terms of the second Piola-Kirchhoff stress, 𝑆𝑖𝑗 , and Green's strain tensor, 𝐸𝑖𝑗 ,
𝑡 𝜕𝐸𝑘𝑙
𝑆𝑖𝑗 = ∫ 𝐺𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) 𝑑𝜏, (22.87.6)
0 𝜕𝜏
where 𝑔𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) and 𝐺𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) are the relaxation functions for the different stress
measures. This stress is added to the stress tensor determined from the strain energy
functional.

Since we wish to include only simple rate effects, the relaxation function is
represented by one term from the Prony series:
𝑁
𝑔(𝑡) = 𝛼0 + ∑ 𝛼𝑚 𝑒−𝛽𝑡 , (22.87.7)
𝑚=1

given by,
𝑔(𝑡) = 𝐸𝑑 𝑒−𝛽1𝑡 . (22.87.8)

This model is effectively a Maxwell fluid which consists of a damper and spring in
series. We characterize this in the input by a shear modulus, 𝐺, and decay constant, 𝛽1 .

The Mooney-Rivlin rubber model is obtained by specifying 𝑛 = 1. In spite of the


differences in formulations with Model 27, we find that the results obtained with this
model are nearly identical with those of 27 as long as large values of Poisson’s ratio are
used.

Rubber Block with Entrapped Air

Air

Figure 22.87.1. Cellular rubber with entrapped air. By setting the initial air
pressure to zero, an open cell, cellular rubber can be simulated.
LS-DYNA DEV 10/27/16 (r:8004) 20-179 (Material Models)
Material Models LS-DYNA Theory Manual

22.88 Material Model 88: MTS Model


The Mechanical Threshhold Stress (MTS) model is due to Mauldin, Davidson, and
Henninger [1990] and is available for applications involving large strains, high pressures
and strain rates. As described in the foregoing reference, this model is based on dislocation
mechanics and provides a better understanding of the plastic deformation process for
ductile materials by using an internal state variable called the mechanical threshold stress.
This kinematic quantity tracks the evolution of the material’s microstructure along some
arbitrary strain, strain rate, and temperature-dependent path using a differential form that
balances dislocation generation and recovery processes. Given a value for the mechanical
threshold stress, the flow stress is determined using either a thermal-activation-controlled
or a drag-controlled kinetics relationship. An equation-of-state is required for solid
elements and a bulk modulus must be defined below for shell elements.

The flow stress 𝜎 is given by:


𝐺
𝜎 = 𝜎̂ a + [𝑠 𝜎̂ + 𝑠th,𝑖 𝜎̂ 𝑖 + 𝑠th,𝑠 𝜎̂ s ]. (22.88.1)
𝐺0 th
The first product in the equation for 𝜎 contains a micro-structure evolution variable, i.e., 𝜎̂ ,
called the Mechanical Threshold Stress (MTS), that is multiplied by a constant-structure
deformation variable 𝑠th :𝑠th is a function of absolute temperature 𝑇 and the plastic strain-
rates 𝜀̇P . The evolution equation for 𝜎̂ is a differential hardening law representing
dislocation-dislocation interactions:

⎡ tanh (α σ̂ )⎤
∂σ̂ ⎢ 𝜎̂ εs ⎥
p ≡ Θo ⎢1 − ⎥. (22.88.2)
∂𝜀 ⎢ tanh(𝛼) ⎥
⎣ ⎦

∂𝜎
̂
The term, ∂𝜀 p , represents the hardening due to dislocation generation and the stress
𝜎
̂
ratio, 𝜎
̂εs
, represents softening due to dislocation recovery. The threshold stress at zero
strain-hardening 𝜎̂ εs is called the saturation threshold stress. Relationships for 𝛩𝑜 , 𝜎̂ εs are:
𝜀̇p 𝜀̇p
𝛩𝑜 = 𝑎𝑜 + 𝑎1 ln ( ) + 𝑎2 √ , (22.88.3)
𝜀0 𝜀0
which contains the material constants 𝑎o , 𝑎1 , and 𝑎2 . The constant, 𝜎̂ εs , is given as:
𝑘𝑇/𝐺𝑏3 𝐴
𝜀̇𝑝
𝜎̂ εs = 𝜎̂ εso ( ) , (22.88.4)
𝜀̇εso

20-180 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

which contains the input constants: 𝜎̂ εso , 𝜀̇εso , 𝑏, 𝐴, and 𝑘. The shear modulus 𝐺 appearing
in these equations is assumed to be a function of temperature and is given by the
correlation.
𝑏1
𝐺 = 𝐺0 − 𝑏2
, (22.88.5)
𝑒𝑇 −1
which contains the constants: 𝐺0, 𝑏1 , and 𝑏2 . For thermal-activation controlled deformation
𝑠th is evaluated via an Arrhenius rate equation of the form:
1
1 𝑝
⎡ 𝜀̇ 𝑞⎤
⎢ ⎛
⎜ 𝑘𝑇ln ( p0 )⎞ ⎥
𝜀̇ ⎟
𝑠th = ⎢1 − ⎜



⎟ ⎥ . (22.88.6)
⎢ ⎜ 𝐺𝑏3 𝑔0 ⎟ ⎟ ⎥
⎢ ⎥
⎝ ⎠
⎣ ⎦

The absolute temperature is given as:


𝑇 = 𝑇ref + 𝜌𝑐p 𝐸, (22.88.7)
where 𝐸 in the internal energy density per unit initial volume.

LS-DYNA DEV 10/27/16 (r:8004) 20-181 (Material Models)


Material Models LS-DYNA Theory Manual

22.89 Material Model 89: Plasticity Polymer


Unlike other LS-DYNA material models, both the input stress-strain curve and the
strain to failure are defined as total true strain, not plastic strain. The input can be defined
from uniaxial tensile tests; nominal stress and nominal strain from the tests must be
converted to true stress and true strain. The elastic component of strain must not be
subtracted out.

The stress-strain curve is permitted to have sections steeper (i.e. stiffer) than the
elastic modulus. When these are encountered the elastic modulus is increased to prevent
spurious energy generation.

20-182 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.90 Material Model 90: Acoustic


This model is appropriate for tracking low-pressure stress waves in an acoustic
media such as air or water and can be used only with the acoustic pressure element
formulation. The acoustic pressure element requires only one unknown per node. This
element is very cost effective.

LS-DYNA DEV 10/27/16 (r:8004) 20-183 (Material Models)


Material Models LS-DYNA Theory Manual

22.91 Material Model 91: Soft Tissue


The overall strain energy W is "uncoupled" and includes two isotropic deviatoric
matrix terms, a fiber term F, and a bulk term:
1
𝑊 = 𝐶1 (𝐼1̃ − 3) + 𝐶2 (𝐼2̃ − 3) + 𝐹(𝜆) + 𝐾[ln(𝐽)]2 . (22.91.1)
2
Here, 𝐼1̃ and 𝐼2̃ are the deviatoric invariants of the right Cauchy deformation tensor, 𝜆 is the
deviatoric part of the stretch along the current fiber direction, and 𝐽 = det𝐅 is the volume
ratio. The material coefficients 𝐶1 and 𝐶2 are the Mooney-Rivlin coefficients, while 𝐾 is the
effective bulk modulus of the material (input parameter XK).

The derivatives of the fiber term 𝐹 are defined to capture the behavior of crimped
collagen. The fibers are assumed to be unable to resist compressive loading - thus the
model is isotropic when 𝜆 < 1. An exponential function describes the straightening of the
fibers, while a linear function describes the behavior of the fibers once they are straightened
past a critical fiber stretch level 𝜆 ≥ 𝜆∗ (input parameter XLAM):
⎧0 𝜆<1
{
{ 𝐶
∂𝐹 { 3 [exp(𝐶4 (𝜆 − 1)) − 1] 𝜆 < 𝜆∗
= 𝜆 . (22.91.2)
∂λ ⎨
{ 1
{
{ (𝐶5 𝜆 + 𝐶6 )
⎩𝜆 𝜆 ≥ 𝜆∗

Coefficients 𝐶3 , 𝐶4 , and 𝐶5 must be defined by the user. 𝐶6 is determined by LS-DYNA to


ensure stress continuity at 𝜆 = 𝜆∗ . Sample values for the material coefficients 𝐶1 − 𝐶5 and
𝜆∗ for ligament tissue can be found in Quapp and Weiss [1998]. The bulk modulus K
should be at least 3 orders of magnitude larger than 𝐶1 to ensure near-incompressible
material behavior.

Viscoelasticity is included via a convolution integral representation for the time-


dependent second Piola-Kirchoff stress 𝐒(𝐂, 𝑡):
𝑡
∂W
𝐒(𝐂, 𝑡) = 𝐒e (𝐂) + ∫ 2𝐺(𝑡 − 𝑠) 𝑑𝑠. (22.91.3)
∂𝐂(s)
0

Here, 𝐒e is the elastic part of the second PK stress as derived from the strain energy, and
𝐺(𝑡 − 𝑠) is the reduced relaxation function, represented by a Prony series:
6
𝑡
𝐺(t) = ∑ S𝑖 exp ( ). (22.91.4)
𝑖=1
𝑇𝑖
Puso and Weiss [1998] describe a graphical method to fit the Prony series coefficients to
relaxation data that approximates the behavior of the continuous relaxation function
proposed by Y-C. Fung, as quasilinear viscoelasticity.

20-184 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.94 Material Model 94: Inelastic Spring Discrete Beam


The yield force is taken from the load curve:
𝐹Y = 𝐹y (Δ𝐿plastic ), (22.94.1)

where 𝐿plastic is the plastic deflection. A trial force is computed as:


𝐹T = 𝐹n + 𝐾 ⋅ Δ𝐿̇ ⋅ Δ𝑡, (22.94.2)
and is checked against the yield force to determine F:
Y T Y
𝐹 = {𝐹T if 𝐹T > 𝐹Y . (22.94.3)
𝐹 if 𝐹 ≤ 𝐹

The final force, which includes rate effects and damping, is given by:
∣Δ𝐿̇∣
𝐹𝑛+1 = 𝐹 ⋅ [1 + 𝐶1 ⋅ Δ𝐿̇ + 𝐶2 ⋅ sgn(Δ𝐿̇)ln (max {1. , })] + DΔ𝐿̇ + 𝑔(𝛥𝐿)ℎ(𝛥𝐿̇), (22.94.4)
DLE
where 𝐶1, 𝐶2 are damping coefficients, DLE is a factor to scale time units.

Unless the origin of the curve starts at (0,0), the negative part of the curve is used
when the spring force is negative where the negative of the plastic displacement is used to
interpolate, 𝐹y . The positive part of the curve is used whenever the force is positive. In
these equations, Δ𝐿 is the change in length
Δ𝐿 = current length-initial length. (22.94.5)

LS-DYNA DEV 10/27/16 (r:8004) 20-185 (Material Models)


Material Models LS-DYNA Theory Manual

22.96 Material Model 96: Brittle Damage Model


A full description of the tensile and shear damage parts of this material model is
given in Govindjee, Kay and Simo [1994,1995]. It is an anisotropic brittle damage model
designed primarily for concrete, though it can be applied to a wide variety of brittle
materials. It admits progressive degradation of tensile and shear strengths across smeared
cracks that are initiated under tensile loadings. Compressive failure is governed by a
simplistic J2 flow correction that can be disabled if not desired. Damage is handled by
treating the rank 4 elastic stiffness tensor as an evolving internal variable for the material.
Softening induced mesh dependencies are handled by a characteristic length method
[Oliver 1989].

Description of properties:

1. 𝐸 is the Young's modulus of the undamaged material also known as the virgin
modulus.

2. 𝜐 is the Poisson's ratio of the undamaged material also known as the virgin
Poisson's ratio.

3. 𝑓𝑛 is the initial principal tensile strength (stress) of the material. Once this stress
has been reached at a point in the body a smeared crack is initiated there with a
normal that is co-linear with the 1st principal direction. Once initiated, the crack is
fixed at that location, though it will convect with the motion of the body. As the
loading progresses the allowed tensile traction normal to the crack plane is pro-
gressively degraded to a small machine dependent constant.

The degradation is implemented by reducing the material's modulus normal to the


smeared crack plane according to a maximum dissipation law that incorporates
exponential softening. The restriction on the normal tractions is given by
𝜙t = (𝐧 ⊗ 𝐧): 𝛔 − 𝑓n + (1 − 𝜀)𝑓n (1 − exp[−𝐻𝛼]) ≤ 0, (22.96.1)

where 𝐧 is the smeared crack normal, 𝜀 is the small constant, 𝐻 is the softening
modulus, and 𝛼 is an internal variable. 𝐻 is set automatically by the program; see
𝑔c below. 𝛼 measures the crack field intensity and is output in the equivalent plas-
tic strain field, 𝜀̅p , in a normalized fashion.

The evolution of alpha is governed by a maximum dissipation argument. When


the normalized value reaches unity it means that the material's strength has been
reduced to 2% of its original value in the normal and parallel directions to the
smeared crack. Note that for plotting purposes, it is never output greater than 5.

20-186 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

4. 𝑓s is the initial shear traction that may be transmitted across a smeared crack plane.
The shear traction is limited to be less than or equal to 𝑓s (1 − 𝛽)(1 − exp[−𝐻𝛼]),
through the use of two orthogonal shear damage surfaces. Note that the shear
degradation is coupled to the tensile degradation through the internal variable
alpha which measures the intensity of the crack field. 𝛽 is the shear retention factor
defined below. The shear degradation is taken care of by reducing the material's
shear stiffness parallel to the smeared crack plane.

5. 𝑔c is the fracture toughness of the material. It should be entered as fracture energy


per unit area crack advance. Once entered the softening modulus is automatically
calculated based on element and crack geometries.

6. 𝛽 is the shear retention factor. As the damage progresses the shear tractions
allowed across the smeared crack plane asymptote to the product 𝛽𝑓s .

7. 𝜂 represents the viscosity of the material. Viscous behavior is implemented as a


simple Perzyna regularization method. This allows for the inclusion of first order
rate effects. The use of some viscosity is recommend as it serves as regularizing
parameter that increases the stability of calculations.

8. 𝜎y is a uniaxial compressive yield stress. A check on compressive stresses is made


using the J2 yield function s: s − √23 𝜎y ≤ 0, where s is the stress deviator. If violat-
ed, a J2 return mapping correction is executed. This check is executed when (1) no
damage has taken place at an integration point yet, (2) when damage has taken
place at a point but the crack is currently closed, and (3) during active damage
after the damage integration (ie. as an operator split). Note that if the crack is
open, the plasticity correction is done in the plane-stress subspace of the crack
plane.

Remark: A variety of experimental data has been replicated using this model from
quasi-static to explosive situations. Reasonable properties for a standard grade concrete
would be 𝐸 = 3.15 × 106 psi, 𝑓n = 450 psi, 𝑓s = 2100 psi, 𝑣 = 0.2, 𝑔c = 0.8 lbs/in, 𝛽 = 0.03,𝜂 =
0.0 psi-sec, 𝜎y = 4200 psi. For stability, values of 𝜂 between 104 to 106 psi/sec are
recommended. Our limited experience thus far has shown that many problems require
nonzero values of 𝜂 to run to avoid error terminations. Various other internal variables
such as crack orientations and degraded stiffness tensors are internally calculated but
currently not available for output.

LS-DYNA DEV 10/27/16 (r:8004) 20-187 (Material Models)


Material Models LS-DYNA Theory Manual

22.97 Material Model 97: General Joint Discrete Beam


For explicit calculations, the additional stiffness due to this joint may require
additional mass and inertia for stability. Mass and rotary inertia for this beam element is
based on the defined mass density, the volume, and the mass moment of inertia defined in
the *SECTION_ BEAM input.

The penalty stiffness applies to explicit calculations. For implicit calculations,


constraint equations are generated and imposed on the system equations; therefore, these
constants, RPST and RPSR, are not used.

20-188 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.98 Material Model 98: Simplified Johnson Cook


Johnson and Cook express the flow stress as
p𝑛
𝜎𝑦 = (𝐴 + 𝐵𝜀̅ ) (1 + 𝐶ln𝜀̇∗ ), (22.98.1)

where
𝐴, 𝐵, 𝐶 and 𝑛 are input constants
𝜀̅p effective plastic strain
̇
𝜀̇∗ = 𝜀̇𝜀̅ effective strain rate for 𝜀̇0 = 1s−1
0

The maximum stress is limited by SIGMAX and SIGSAT by:


p𝑛
𝜎y = min {min [𝐴 + 𝐵𝜀̅ , SIGMAX] (1 + 𝐶ln𝜀̇∗ ), SIGSAT}. (22.98.2)

Failure occurs when the effective plastic strain exceeds PSFAIL.

If the viscoplastic option is active, VP = 1.0, the parameters SIGMAX and SIGSAT
are ignored since these parameters make convergence of the viscoplastic strain iteration
loop difficult to achieve. The viscoplastic option replaces the plastic strain in the forgoing
equations by the viscoplastic strain and the strain rate by the viscoplastic strain rate.
Numerical noise is substantially reduced by the viscoplastic formulation.

LS-DYNA DEV 10/27/16 (r:8004) 20-189 (Material Models)


Material Models LS-DYNA Theory Manual

22.100 Material Model 100: Spot Weld


This material model applies to beam element type 9 for spot welds. These beam
elements may be placed between any two deformable shell surfaces, see Figure 22.100.1,
and tied with type 7 constraint contact which eliminates the need to have adjacent nodes at
spot weld locations. Beam spot welds may be placed between rigid bodies and
rigid/deformable bodies by making the node on one end of the spot weld a rigid body
node which can be an extra node for the rigid body. In the same way, rigid bodies may
also be tied together with this spot weld option.

It is advisable to include all spot welds, which provide the slave nodes, and spot
welded materials, which define the master segments, within a single type 7 tied interface.
As a constraint method, multiple type 7 interfaces are treated independently which can
lead to significant problems if such interfaces share common nodal points. The offset
option, “o 7”, should not be used with spot welds.

The DAMAGE-FAILURE option causes one additional line to be read with the
damage parameter and a flag that determines how failure is computed from the resultants.
On this line the parameter, DMG, if nonzero, invokes damage mechanics combined with
the plasticity model to achieve a smooth drop off of the resultant forces prior to the
removal of the spot weld. The parameter FOPT determines the method used in computing
resultant based failure, which is unrelated to damage.

The weld material is modeled with isotropic hardening plasticity coupled to two
failure models. The first model specifies a failure strain which fails each integration point

SPOTWELD ELEMENT r
n2
Trr
t
Mtt
Frr s
Frt
n1
Frs Mss

Figure 22.100.1. Deformable spotwelds can be arbitrarily placed within the


structure.
20-190 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Material Models

in the spot weld independently. The second model fails the entire weld if the resultants are
outside of the failure surface defined by:
2 2 2 2 2 2
𝑁 𝑁 𝑁 𝑀 𝑀 𝑇
( 𝑟𝑟 ) + ( 𝑟𝑠 ) + ( 𝑟𝑡 ) + ( 𝑟𝑟 ) + ( 𝑠𝑠 ) + ( 𝑟𝑟 ) − 1 = 0, (22.100.1)
𝑁𝑟𝑟F 𝑁𝑟𝑠F 𝑁𝑟𝑡F 𝑀𝑟𝑟F 𝑀𝑠𝑠F 𝑇𝑟𝑟F
where the numerators in the equation are the resultants calculated in the local coordinates
of the cross section, and the denominators are the values specified in the input. If the user
defined parameter, NF, which the number of force vectors stored for filtering, is nonzero
the resultants are filtered before failure is checked. The default value is set to zero which is
generally recommended unless oscillatory resultant forces are observed in the time history
databases. Even though these welds should not oscillate significantly, this option was
added for consistency with the other spot weld options. NF affects the storage since it is
necessary to store the resultant forces as history variables.

If the failure strain is set to zero, the failure strain model is not used. In a similar
manner, when the value of a resultant at failure is set to zero, the corresponding term in the
failure surface is ignored. For example, if only N𝑟𝑟F is nonzero, the failure surface is
reduced to |N𝑟𝑟 | = N𝑟𝑟F . None, either, or both of the failure models may be active
depending on the specified input values.

The inertias of the spot welds are scaled during the first time step so that their stable
time step size is Δ𝑡. A strong compressive load on the spot weld at a later time may reduce
the length of the spot weld so that stable time step size drops below Δ𝑡. If the value of Δ𝑡 is
zero, mass scaling is not performed, and the spot welds will probably limit the time step
size. Under most circumstances, the inertias of the spot welds are small enough that
scaling them will have a negligible effect on the structural response and the use of this
option is encouraged.

Spotweld force history data is written into the SWFORC ASCII file. In this database
the resultant moments are not available, but they are in the binary time history database.

The constitutive properties for the damaged material are obtained from the
undamaged material properties. The amount of damage evolved is represented by the
constant, ω, which varies from zero if no damage has occurred to unity for complete
rupture. For uniaxial loading, the nominal stress in the damaged material is given by
𝑃
𝜎nominal =
, (22.100.2)
𝐴
where 𝑃 is the applied load and 𝐴 is the surface area. The true stress is given by:
𝑃
𝜎true = , (22.100.3)
𝐴 − 𝐴loss
where 𝐴loss is the void area. The damage variable can then be defined:
𝐴loss (22.100.4)
𝜔= , 0 ≤ 𝜔 ≤ 1.
𝐴
LS-DYNA DEV 10/27/16 (r:8004) 20-191 (Material Models)
Material Models LS-DYNA Theory Manual

In this model damage is defined in terms of plastic strain after the failure strain is
exceeded:
p p
𝜀eff − 𝜀failure p p p
𝜔= p p if 𝜀failure ≤ 𝜀eff ≤ 𝜀rupture . (22.100.5)
𝜀rupture − 𝜀failure

After exceeding the failure strain softening begins and continues until the rupture strain is
reached.

20-192 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.101 Material Model 101: GE Thermoplastics


The constitutive model for this approach is:
𝜀̇p = 𝜀̇0 exp(𝐴{𝜎 − 𝑆(𝜀p )}) × exp(−𝑝𝛼𝐴), (22.101.1)
where 𝜀̇0 and A are rate dependent yield stress parameters, 𝑆(𝜀𝑝 ) internal resistance (strain
hardening) and 𝛼 is a pressure dependence parameter.

In this material the yield stress may vary throughout the finite element model as a
function of strain rate and hydrostatic stress. Post yield stress behavior is captured in
material softening and hardening values. Finally, ductile or brittle failure measured by
plastic strain or maximum principal stress respectively is accounted for by automatic
element deletion.

Although this may be applied to a variety of engineering thermoplastics, GE Plastics


have constants available for use in a wide range of commercially available grades of their
engineering thermoplastics.

LS-DYNA DEV 10/27/16 (r:8004) 20-193 (Material Models)


Material Models LS-DYNA Theory Manual

22.102 Material Model 102: Hyperbolic Sine


Resistance to deformation is both temperature and strain rate dependent. The flow
stress equation is:
1
1 ⎛ 𝑍 N⎞
𝜎 = sinh−1 ⎜⎜[ ] ⎟⎟, (22.102.1)
α 𝐴
⎝ ⎠
where 𝑍, the Zener-Holloman temperature compensated strain rate, is:
𝑄
𝑍 = 𝜀̇exp ( ). (22.102.2)
𝐺𝑇

The units of the material constitutive constants are as follows: 𝐴 (1/sec), N


(dimensionless), 𝛼 (1/MPa), the activation energy for flow, 𝑄 (J/mol), and the universal gas
constant, 𝐺 [J/(mol ⋅ K)]. The value of 𝐺 will only vary with the unit system chosen.
Typically it will be either 8.3145 J/(mol ⋅ K), or 40.8825 lb ⋅ in/mol ⋅ R.

The final equation necessary to complete the description of high strain rate
deformation herein is one that allows computation of the temperature change during the
deformation. In the absence of a coupled thermo-mechanical finite element code we
assume adiabatic temperature change and follow the empirical assumption that 90-95% of
the plastic work is dissipated as heat. Thus the heat generation coefficient is
0.9
HC ≈ , (22.102.3)
𝜌𝐶𝑣
where 𝜌 is the material density and 𝐶𝑣 is the specific heat.

20-194 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.103 Material Model 103: Anisotropic Viscoplastic


The uniaxial stress-strain curve is given on the following form
𝑝 𝑝 𝑝 𝑝
𝜎(𝜀eff , 𝜀̇eff ) = 𝜎0 + 𝑄𝑟1 [(1 − 𝑒𝑥𝑝(−𝐶𝑟1 𝜀eff ))] + 𝑄𝑟2 [1 − 𝑒𝑥𝑝(−𝐶𝑟2 𝜀eff )]
𝑝 𝑝 𝑝 𝑉𝑚 (22.103.1)
+ 𝑄𝜒1 [(1 − 𝑒𝑥𝑝(−𝐶𝜒1 𝜀eff ))] + 𝑄𝜒2 [(1 − 𝑒𝑥𝑝(−𝐶𝜒2 𝜀eff ))] + 𝑉𝑘 𝜀̇eff ,

For bricks the following yield criteria is used


2
𝐹(𝜎22 − 𝜎33 )2 + 𝐺(𝜎33 − 𝜎11 )2 + 𝐻(𝜎11 − 𝜎22 )2 + 2𝐿𝜎23 2
+ 2𝑀𝜎31 2
+ 2𝑁𝜎12
𝑝 𝑝 2 (22.103.2)
= [𝜎(𝜀eff , 𝜀̇eff )] ,
𝑝 𝑝
where 𝜀eff is the effective plastic strain and 𝜀̇eff is the effective plastic strain rate. For shells
the anisotropic behavior is given by 𝑅00 , 𝑅45 and 𝑅90 . When 𝑉𝑘 = 0 the material will
behave elasto-plastically. Default values are given by:
1
𝐹=𝐺=𝐻= , (22.103.3)
2
3
𝐿=𝑀=𝑁= , (22.103.4)
2
𝑅00 = 𝑅45 = 𝑅90 = 1. (22.103.5)

Strain rate is accounted for using the Cowper-Symonds model which, e.g., model 3,
scales the yield stress with the factor:
1
𝜀̇ ⁄𝑝 (22.103.6)
1+( ) .
𝐶
To convert these constants set the viscoelastic constants, 𝑉𝑘 and 𝑉𝑚 , to the following
values:
1
1 𝑝
𝑉𝑘 = 𝜎 ( ) ,
𝐶
(22.103.7)
1
𝑉𝑚 = .
𝑝

This model properly treats rate effects and should provide superior results to
models 3 and 24.

LS-DYNA DEV 10/27/16 (r:8004) 20-195 (Material Models)


Material Models LS-DYNA Theory Manual

22.104 Material Model 104: Continuum Damage Mechanics


Model
Anisotropic Damage model (FLAG = −1). At each thickness integration points, an
anisotorpic damage law acts on the plane stress tensor in the directions of the principal
total shell strains, ε1 and 𝜀2 , as follows:
𝜎11 = (1 − 𝐷1 (𝜀1 ))𝜎110 ,
𝜎22 = (1 − 𝐷2 (𝜀2 ))𝜎220 ,
(22.104.1)
𝐷 + 𝐷2
𝜎12 = (1 − 1 ) 𝜎120 .
2

The transverse plate shear stresses in the principal strain directions are assumed to
be damaged as follows:
𝐷1
𝜎13 = (1 − ) 𝜎130 ,
2
(22.104.2)
𝐷
𝜎23 = (1 − 2 ) 𝜎230 .
2

In the anisotropic damage formulation, 𝐷1 (𝜀1 ) and 𝐷2 (𝜀2 ) are anisotropic damage
functions for the loading directions 1 and 2, respectively. Stresses 𝜎110 , 𝜎220 , 𝜎120 , 𝜎130 and
𝜎230 are stresses in the principal shell strain directions as calculated from the undamaged
elastic-plastic material behavior. The strains 𝜀1 and 𝜀2 are the magnitude of the principal
strains calculated upon reaching the damage thresholds. Damage can only develop for
tensile stresses, and the damage functions 𝐷1 (𝜀1 ) and 𝐷2 (𝜀2 ) are identical to zero for
negative strains 𝜀1 and 𝜀2 . The principal strain directions are fixed within an integration
point as soon as either principal strain exceeds the initial threshold strain in tension. A
more detailed description of the damage evolution for this material model is given in the
description of material 82.

The Continuum Damage Mechanics (CDM) model (FLAG≥0) is based on a CDM


model proposed by Lemaitre [1992]. The effective stress 𝜎̃ , which is the stress calculated
over the section that effectively resist the forces and reads.
𝜎
𝜎̃ = , (22.104.3)
1−𝐷
where 𝐷 is the damage variable. The evolution equation for the damage variable is defined
as

20-196 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

⎧ 𝑌
{
{𝑆(1 − 𝐷) 𝑟 ̇ for 𝑟 > 𝑟𝐷 and 𝜎1 > 0
𝐷̇ = ⎨ . (22.104.4)
{
{ 0 otherwise

where 𝑟𝐷 is the damage threshold, 𝑌 is a positive material constant, 𝑆 is the strain energy
release rate, and 𝜎1 is the maximal principal stress. The strain energy density release rate is
2
1 𝜎vm 𝑅𝑣
𝑌 = 𝐞e : 𝐂: 𝐞e = , (22.104.5)
2 2𝐸(1 − 𝐷)2
where 𝜎vm is the equivalent von Mises stress. The triaxiality function 𝑅𝑣 is defined as
2 𝜎 2
𝑅𝑣 = (1 + 𝜈) + 3(1 − 2𝜈) ( H ) . (22.104.6)
3 𝜎vm

The uniaxial stress-strain curve is given in the following form


p p 𝑉𝑚
𝜎(𝑟, 𝜀̇eff ) = 𝜎0 + 𝑄1 (1 − exp(−𝐶1 𝑟)) + 𝑄2 (1 − exp(−𝐶2 𝑟)) + 𝑉𝑘 𝜀̇eff , (22.104.7)
where 𝑟 is the damage accumulated plastic strain, which can be calculated by
p
𝑟 ̇ = 𝜀̇eff (1 − 𝐷). (22.104.8)

For bricks the following yield criteria is used


2
𝐹(𝜎̃ 22 − 𝜎̃ 33 )2 + 𝐺(𝜎̃ 33 − 𝜎̃ 11 )2 + 𝐻(𝜎̃ 11 − 𝜎̃ 22 )2 + 2𝐿𝜎̃ 23 2
+ 2𝑀𝜎̃ 31 2
+ 2𝑁𝜎̃ 12
p (22.104.9)
= 𝜎(𝑟, 𝜀̇eff ),
p
where 𝑟 is the damage effective viscoplastic strain and 𝜀̇eff is the effective viscoplastic strain
rate. For shells the anisotropic behavior is given by the R-values: 𝑅00 , 𝑅45 , and 𝑅90 . When
𝑉𝑘 = 0 the material will behave as an elastoplastic material without rate effects. Default
values for the anisotropic constants are given by:
1
𝐹=𝐺=𝐻= , (22.104.10)
2

3
𝐿=𝑀=𝑁= , (22.104.11)
2
𝑅00 = 𝑅45 = 𝑅90 = 1, (22.104.12)
so that isotropic behavior is obtained.

Strain rate is accounted for using the Cowper-Symonds model which scales the yield
stress with the factor:

LS-DYNA DEV 10/27/16 (r:8004) 20-197 (Material Models)


Material Models LS-DYNA Theory Manual

1
𝜀̇ ⁄p (22.104.13)
1+( ) .
𝐶

To convert these constants, set the viscoelastic constants, 𝑉𝑘 and 𝑉𝑚 , to the following
values:
1
1 𝑝
𝑉𝑘 = 𝜎 ( ) ,
𝐶
(22.104.14)
1
𝑉𝑚 = .
𝑝

20-198 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.106 Material Model 106: Elastic Viscoplastic Thermal


If LCSS is not given any value the uniaxial stress-strain curve has the form
p p p
𝜎(𝜀eff ) = 𝜎0 + 𝑄𝑟1 (1 − exp(−𝐶𝑟1 𝜀eff )) + 𝑄𝑟2 (1 − exp(−𝐶𝑟2 𝜀eff ))
p p (22.106.1)
+𝑄χ1 (1 − exp(−𝐶χ1 𝜀eff )) + Qχ2 (1 − exp(−Cχ2 𝜀eff )).

Viscous effects are accounted for using the Cowper-Symonds model, which scales
the yield stress with the factor:
p 1⁄
p
𝜀̇ (22.106.2)
1 + ( eff ) .
𝐶

LS-DYNA DEV 10/27/16 (r:8004) 20-199 (Material Models)


Material Models LS-DYNA Theory Manual

22.110 Material Model 110: Johnson-Holmquist Ceramic


Model
The Johnson-Holmquist plasticity damage model is useful for modeling ceramics,
glass and other brittle materials. A more detailed description can be found in a paper by
Johnson and Holmquist [1993].

The equivalent stress for a ceramic-type material is given in terms of the damage
parameter 𝐷 by
𝜎 ∗ = 𝜎i∗ − 𝐷(𝜎i∗ − 𝜎f∗ ). (22.110.1)
Here,
𝜎i∗ = 𝑎(𝑝∗ + 𝑡∗ )𝑛 (1 + 𝑐ln𝜀̇∗ ), (22.110.2)
represents the intact, undamaged behavior. The superscript, '*', indicates a normalized
quantity. The stresses are normalized by the equivalent stress at the Hugoniot elastic limit
(see below), the pressures are normalized by the pressure at the Hugoniot elastic limit, and
the strain rate by the reference strain rate defined in the input. In this equation 𝑎 is the
intact normalized strength parameter, 𝑐 is the strength parameter for strain rate
dependence, 𝜀̇∗ is the normalized plastic strain rate, and,
𝑇
𝑡∗ = ,
PHEL
(22.110.3)
∗ 𝑝
𝑝 = ,
PHEL
where 𝑇 is the maximum tensile pressure strength, PHEL is the pressure component at the

Hugoniot elastic limit, and p is the pressure.

p
𝐷 = ∑ Δ𝜀p /𝜀f , (22.110.4)
represents the accumulated damage based upon the increase in plastic strain per
computational cycle and the plastic strain to fracture
p
𝜀f = 𝑑1 (𝑝∗ + 𝑡∗ )𝑑2 , (22.110.5)
where 𝑑1 and 𝑑2 are user defined input parameters. The equation:
𝜎f∗ = 𝑏(𝑝∗ )𝑚 (1 + 𝑐ln𝜀̇∗ ) ≤ SFMAX, (22.110.6)
represents the damaged behavior where 𝑏 is an input parameter and SFMAX is the
maximum normalized fracture strength. The parameter, 𝑑1 , controls the rate at which
damage accumulates. If it approaches 0, full damage can occur in one time step, i.e.,

20-200 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

instantaneously. This rate parameter is also the best parameter to vary if one attempts to
reproduce results generated by another finite element program.

In undamaged material, the hydrostatic pressure is given by


𝑝 = 𝑘 1 𝜇 + 𝑘 2 𝜇2 + 𝑘 3 𝜇3 , (22.110.7)
where 𝜇 = 𝜌/𝜌0 − 1. When damage starts to occur, there is an increase in pressure. A
fraction defined in the input, between 0 and 1, of the elastic energy loss, 𝛽, is converted into
hydrostatic potential energy, which results in an increase in pressure. The details of this
pressure increase are given in the reference.

Given HEL and the shear modulus, 𝑔, 𝜇hel can be found iteratively from
4 𝜇
HEL = 𝑘1 𝜇hel + 𝑘2 𝜇2hel + 𝑘3 𝜇3hel + 𝑔 ( hel ), (22.110.8)
3 1 + 𝜇hel
and, subsequently, for normalization purposes,
PHEL = 𝑘1 𝜇hel + 𝑘2 𝜇2hel + 𝑘3 𝜇3hel , (22.110.9)
and
𝜎hel = 1.5(HEL − PHEL). (22.110.10)
These are calculated automatically by LS-DYNA if PHEL is zero on input.

LS-DYNA DEV 10/27/16 (r:8004) 20-201 (Material Models)


Material Models LS-DYNA Theory Manual

22.111 Material Model 111: Johnson-Holmquist Concrete


Model
This model can be used for concrete subjected to large strains, high strain rates, and
high pressures. The equivalent strength is expressed as a function of the pressure, strain
rate, and damage. The pressure is expressed as a function of the volumetric strain and
includes the effect of permanent crushing. The damage is accumulated as a function of the
plastic volumetric strain, equivalent plastic strain and pressure. A more detailed
description of this model can be found in the paper by Holmquist, Johnson, and Cook
[1993]

The normalized equivalent stress is defined as


𝜎
𝜎 ∗ = ′, (22.111.1)
𝑓c
where 𝜎 is the actual equivalent stress, and 𝑓c′ is the quasi-static uniaxial compressive
strength. The yield stress is defined in terms of the input parameters 𝑎, 𝑏, 𝑐, and 𝑛 as:
𝑛
𝜎 ∗ = [𝑎(1 − 𝐷) + 𝑏𝑝∗ ][1 − 𝑐ln(𝜀̇∗ )], (22.111.2)
where 𝐷 is the damage parameter, 𝑝∗ = 𝑝/𝑓c′ is the normalized pressure and 𝜀̇∗ = 𝜀̇/𝜀̇0 is the
dimensionless strain rate. The model accumulates damage both from equivalent plastic
strain and plastic volumetric strain, and is expressed as
Δ𝜀p + Δ𝜇p
𝐷=∑ , (22.111.3)
𝐷1 (𝑝∗ + 𝑇 ∗ )𝐷2
where Δ𝜀p and Δ𝜇p are the equivalent plastic strain and plastic volumetric strain, 𝐷1 and
𝐷2 are material constants and 𝑇 ∗ = 𝑇/𝑓c′ is the normalized maximum tensile hydrostatic
pressure where 𝑇 is the maximum tensile hydrostatic pressure.

The pressure for fully dense material is expressed as:


𝑃 = 𝐾1 𝜇̅̅̅ + 𝐾2 𝜇̅̅̅2 + 𝐾3 𝜇̅̅̅3 , (22.111.4)
where 𝐾1 , 𝐾2 and 𝐾3 are material constants and the modified volumetric strain is defined
as
𝜇 − 𝜇lock
𝜇̅̅̅ = , (22.111.5)
1 + 𝜇lock
where 𝜇lock is the locking volumetric strain.

20-202 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.115 Material Model 115: Elastic Creep Model


The effective creep strain, 𝜀̅c , given as:
𝜀̅c = 𝐴𝜎
̅̅̅̅ 𝑛 𝑡𝑚̅ , (22.115.1)
where 𝐴, 𝑛, and 𝑚 are constants and 𝑡 ̅ is the effective time. The effective stress, 𝜎
̅̅̅̅, is
defined as:

3
̅̅̅̅ = √ 𝜎𝑖𝑗 𝜎𝑖𝑗 .
𝜎 (22.115.2)
2

The creep strain, therefore, is only a function of the deviatoric stresses. The
volumetric behavior for this material is assumed to be elastic. By varying the time constant
𝑚 primary creep (𝑚 < 1), secondary creep (𝑚 = 1), and tertiary creep (𝑚 > 1) can be
modeled. This model is described by Whirley and Henshall (1992).

LS-DYNA DEV 10/27/16 (r:8004) 20-203 (Material Models)


Material Models LS-DYNA Theory Manual

22.116 Material Model 116: Composite Layup


This material is for modeling the elastic responses of composite lay-ups that have an
arbitrary number of layers through the shell thickness. A pre-integration is used to
compute the extensional, bending, and coupling stiffness for use with the Belytschko-Tsay
resultant shell formulation. The angles of the local material axes are specified from layer to
layer in the *SECTION_SHELL input. This material model must be used with the user
defined integration rule for shells, which allows the elastic constants to change from
integration point to integration point. Since the stresses are not computed in the resultant
formulation, the stress output to the binary databases for the resultant elements are zero.

This material law is based on standard composite lay-up theory. The implementa-
tion, [Jones 1975], allows the calculation of the force, 𝐍, and moment, 𝐌, stress resultants
from:

⎧𝑁 𝑥 ⎫ 𝐴11 𝐴12 𝐴16 ⎧ {𝜀0𝑥 ⎫


} ⎡𝐵11 𝐵12 𝐵16 ⎧ 𝜅𝑥 ⎫
{ } { 0} { }
𝑁𝑦 = ⎡ 𝐴26 ⎤
⎥ ⎨𝜀𝑦 ⎬ + ⎢𝐵21 𝐵22 𝐵26 ⎤
⎥ ⎨ 𝜅 𝑦 ⎬,
⎨ ⎬ ⎢𝐴21 𝐴22 (22.116.1)
{𝑁 } ⎣ 𝐴
⎩ 𝑥𝑦 ⎭ 16 𝐴26 𝐴66 ⎦ {
{𝜀0 }}
⎩ 𝑧 ⎭ ⎣𝐵16 𝐵26 𝐵66 ⎦ {
⎩𝜅𝑥𝑦 }⎭

⎧𝑀𝑥 ⎫ 𝐵11 𝐵12 𝐵16 ⎧ { 𝜀0𝑥 ⎫


} ⎡𝐷11 𝐷12 𝐷16 ⎧ 𝜅𝑥 ⎫
{ } { 0} { }
𝑀𝑦 = ⎡ 𝐵26 ⎤ 𝐷26 ⎤
⎥ ⎨ 𝜅 𝑦 ⎬,
⎨ ⎬ ⎢𝐵21 𝐵22 ⎥ ⎨𝜀𝑦 ⎬ + ⎢𝐷21 𝐷22 (22.116.2)
{𝑀 } ⎣𝐵
⎩ 𝑥𝑦 ⎭ 16 𝐵26 𝐵66 ⎦ {
{𝜀0 } }
⎩ 𝑧 ⎭ ⎣𝐷16 𝐷26 𝐷66 ⎦ {
⎩𝜅𝑥𝑦 }⎭

where 𝐴𝑖𝑗 is the extensional stiffness, 𝐷𝑖𝑗 is the bending stiffness, and 𝐵𝑖𝑗 is the coupling
stiffness, which is a null matrix for symmetric lay-ups. The mid-surface strains and
curvatures are denoted by 𝜀0𝑖𝑗 and 𝜅𝑖𝑗 , respectively. Since these stiffness matrices are
symmetric, 18 terms are needed per shell element in addition to the shell resultants, which
are integrated in time. This is considerably less storage than would typically be required
with through thickness integration which requires a minimum of eight history variables
per integration point, e.g., if 100 layers are used 800 history variables would be stored. Not
only is memory much less for this model, but the CPU time required is also considerably
reduced.

20-204 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.117 Material Model 117-118: Composite Matrix


This material is used for modeling the elastic responses of composites where pre-
integration, which is done outside of LS-DYNA unlike the lay-up option above, is used to
compute the extensional, bending, and coupling stiffness coefficients for use with the
Belytschko-Tsay and the assumed strain resultant shell formulations. Since the stresses are
not computed in the resultant formulation, the stresses output to the binary databases for
the resultant elements are zero.

The calculation of the force, 𝑁𝑖𝑗 , and moment, 𝑀𝑖𝑗 , stress resultants is given in terms
of the membrane strains, 𝜀0𝑖 , and shell curvatures, 𝜅𝑖 , as:
⎧ 𝑁𝑥 ⎫ 𝐶11 𝐶12 𝐶13 𝐶14 𝐶15 𝐶16 ⎧ { 𝜀0𝑥 ⎫
}
{ } ⎡
{
{ 𝑁 }
𝑦 }
⎢𝐶21 𝐶22 𝐶23 𝐶24 𝐶25 𝐶26 ⎤ { 𝜀
⎥{ 𝑦 }
{ 0}
{
{ 𝑁𝑥𝑦 }
} ⎢𝐶31 𝐶32 𝐶33 𝐶34 𝐶35 𝐶36 ⎥ { } }
⎢ ⎥ 𝜀0𝑧 ,
⎨ 𝑀𝑥 ⎬ = ⎢𝐶41 𝐶42 𝐶43 𝐶44 𝐶45 𝐶46 ⎥ ⎨ ⎬ (22.117.1)
{
{ } ⎢ ⎥{ 𝜅𝑥 }
{ 𝑀𝑦 }} ⎢𝐶51 𝐶52 𝐶53 𝐶54 𝐶55 𝐶56 ⎥ { 𝜅𝑦 }
{
{ }
{𝑀 } ⎣𝐶61 { } }
⎩ 𝑥𝑦 ⎭ 𝐶62 𝐶63 𝐶64 𝐶65 𝐶66 ⎦ { 𝜅
⎩ 𝑥𝑦 ⎭
}

where 𝐶𝑖𝑗 = 𝐶𝑗𝑖 . In this model this symmetric matrix is transformed into the element local
system and the coefficients are stored as element history variables.

In a variation of this model, *MAT_COMPOSITE_DIRECT, the resultants are


already assumed to be given in the element local system which reduces the storage since
the 21 coefficients are not stored as history variables as part of the element data. The shell
thickness is built into the coefficient matrix and, consequently, within the part ID, which
references this material ID, the thickness must be uniform.

LS-DYNA DEV 10/27/16 (r:8004) 20-205 (Material Models)


Material Models LS-DYNA Theory Manual

22.119 Material Model 119: General Nonlinear 6DOF


Discrete Beam
Catastrophic failure, which is based on displacement resultants, occurs if either of
the following inequalities are satisfied:
2 2 2 2 2 2
𝑢r 𝑢s 𝑢t 𝜃r 𝜃s 𝜃t
( tfail ) + ( tfail ) + ( tfail ) + ( tfail ) + ( tfail ) + ( tfail ) − 1. ≥ 0, (22.119.1)
𝑢r 𝑢s 𝑢t 𝜃r 𝜃s 𝜃t
2 2 2 2 2 2
𝑢r 𝑢s 𝑢t 𝜃r 𝜃s 𝜃t
( ) +( ) +( ) +( ) +( ) +( ) − 1. ≥ 0. (22.119.2)
𝑢cfail
r 𝑢cfail
s 𝑢cfail
t 𝜃rcfail 𝜃scfail 𝜃tcfail

After failure the discrete element is deleted. If failure is included either the tension
failure or the compression failure or both may be used.

Unloading curve
Unload = 0 Unload = 1
Resultant

loading-unloading Resultant
curve

Displacement Displacement

Unload = 2 Unload = 3
Resultant
Resultant

Unloading curve
Unloading curve

Displacement Displacement

Umin OFFSET x Umin

Figure 22.119.1. Load and unloading behavior.

20-206 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.120 Material Model 120: Gurson


The Gurson flow function is defined as:
2
𝜎M 3𝑞2 𝜎H
𝛷= + 2𝑞1 𝑓 ∗ cosh ( ) − 1 − (𝑞1 𝑓 ∗ )2 = 0, (22.120.1)
𝜎Y2 2𝜎Y
where 𝜎M is the equivalent von Mises stress, 𝜎Y is the Yield stress, 𝜎H is the mean
hydrostatic stress. The effective void volume fraction is defined as

{
{
{ 𝑓 𝑓 ≤ 𝑓c
𝑓 ∗ (𝑓 ) = ⎨ . (22.120.2)
{
{ 1/𝑞 1 − 𝑓 c
{𝑓c + (𝑓 − 𝑓c ) 𝑓 > 𝑓c
⎩ 𝑓F − 𝑓c

The growth of void volume fraction is defined as


𝑓 ̇ = 𝑓Ġ + 𝑓Ṅ , (22.120.3)
where the growth of existing voids is defined as
p
𝑓Ġ = (1 − 𝑓 )𝜀̇𝑘𝑘 , (22.120.4)
and the nucleation of new voids is defined as
𝑓Ṅ = 𝐴𝜀̇p , (22.120.5)
where
𝑓N 1 𝜀p − 𝜀N 2
𝐴= exp (− ( ) ). (22.120.6)
𝑆N √2π 2 𝑆N

LS-DYNA DEV 10/27/16 (r:8004) 20-207 (Material Models)


Material Models LS-DYNA Theory Manual

22.120 Material Model 120: Gurson RCDC


The Rc-Dc model is defined as follows. The damage 𝐷 is given by

𝐷 = ∫ 𝜔1 𝜔2 𝑑𝜀p , (22.120.1)

where 𝜀p is the equivalent plastic strain,


𝛼
1
𝜔1 = ( ) , (22.120.2)
1 − 𝛾𝜎m
is the triaxial stress weighting term and
𝜔2 = (2 − 𝐴D )𝛽 , (22.120.3)
is the asymmetric strain weighting term. In the above 𝜎m is the mean stress and
𝑆 𝑆
𝐴D = min (∣ 2 ∣ , ∣ 3 ∣). (22.120.4)
𝑆3 𝑆2

Fracture is initiated when the accumulation of damage satisfies


𝐷
> 1, (22.120.5)
𝐷c
where 𝐷c is the critical damage given by
𝐷c = 𝐷0 (1 + 𝑏|∇𝐷|λ ). (22.120.6)

20-208 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.124 Material Model 124: Tension-Compression Plasticity


This is an isotropic elastic-plastic material where a unique yield stress versus plastic
strain curve can be defined for compression and tension. Failure can occur based on plastic
strain or a minimum time step size. Rate effects are modeled by using the Cowper-
Symonds strain rate model.

The stress-strain behavior follows one curve in compression and another in tension.
The sign of the mean stress determines the state where a positive mean stress (i.e., a
negative pressure) is indicative of tension. Two load curves, 𝑓t (𝑝) and 𝑓c (𝑝), are defined,
which give the yield stress, 𝜎y , versus effective plastic strain for both the tension and
compression regimes. The two pressure values, 𝑝t and 𝑝c , when exceeded, determine if the
tension curve or the compressive curve is followed, respectively. If the pressure, 𝑝, falls
between these two values, a weighted average of the two curves are used:
⎧ 𝑝c − 𝑝
{scale =
if − 𝑝t ≤ 𝑝 ≤ 𝑝c ⎨ 𝑝c + 𝑝t . (22.124.1)
{ 𝜎 = scale ⋅ 𝑓 (𝑝) + (1 − scale) ⋅ 𝑓 (𝑝)
⎩ y t c

Strain rate is accounted for using the Cowper and Symonds model, which scales the
yield stress with the factor
1
𝜀̇ ⁄p (22.124.2)
1+( ) ,
𝐶
where 𝜀̇ is the strain rate 𝜀̇ = √𝜀̇𝑖𝑗 𝜀̇𝑖𝑗 .

LS-DYNA DEV 10/27/16 (r:8004) 20-209 (Material Models)


Material Models LS-DYNA Theory Manual

22.126 Material Model 126: Metallic Honeycomb


For efficiency it is strongly recommended that the load curve ID’s: LCA, LCB, LCC,
LCS, LCAB, LCBC, and LCCA, contain exactly the same number of points with
corresponding strain values on the abscissa. If this recommendation is followed the cost of
the table lookup is insignificant. Conversely, the cost increases significantly if the abscissa
strain values are not consistent between load curves.

The behavior before compaction is orthotropic where the components of the stress
tensor are uncoupled, i.e., a component of strain will generate resistance in the local a-
direction with no coupling to the local 𝑏 and 𝑐 directions. The elastic modulii vary from
their initial values to the fully compacted values linearly with the relative volume:
𝐸𝑎𝑎 = 𝐸𝑎𝑎𝑢 + 𝛽𝑎𝑎 (𝐸 − 𝐸𝑎𝑎𝑢 ), 𝐺𝑎𝑏 = 𝐺𝑎𝑏𝑢 + 𝛽(𝐺 − 𝐺𝑎𝑏𝑢 ),
𝐸𝑏𝑏 = 𝐸𝑏𝑏𝑢 + 𝛽𝑏𝑏 (𝐸 − 𝐸𝑏𝑏𝑢 ), 𝐺𝑏𝑐 = 𝐺𝑏𝑐𝑢 + 𝛽(𝐺 − 𝐺𝑏𝑐𝑢 ), (22.126.1)
𝐸𝑐𝑐 = 𝐸𝑐𝑐𝑢 + 𝛽𝑐𝑐 (𝐸 − 𝐸𝑐𝑐𝑢 ), 𝐺𝑐𝑎 = 𝐺𝑐𝑎𝑢 + 𝛽(𝐺 − 𝐺𝑐𝑎𝑢 ),
where
1−𝑉
𝛽 = max [min ( , 1) , 0], (22.126.2)
1 − 𝑉𝑓
and 𝐺 is the elastic shear modulus for the fully compacted honeycomb material
𝐸
𝐺= . (22.126.3)
2(1 + 𝜈)

The relative volume, 𝜈, is defined as the ratio of the current volume over the initial
volume, and typically, 𝜈 = 1 at the beginning of a calculation.

The load curves define the magnitude of the stress as the material undergoes
deformation. The first value in the curve should be less than or equal to zero correspond-
ing to tension and increase to full compaction. Care should be taken when defining the
curves so the extrapolated values do not lead to negative yield stresses.

At the beginning of the stress update we transform each element’s stresses and
strain rates into the local element coordinate system. For the uncompacted material, the
trial stress components are updated using the elastic interpolated modulii according to:
trial trial
𝑛+1 𝑛 𝑛+1 𝑛
𝜎𝑎𝑎 = 𝜎𝑎𝑎 + 𝐸𝑎𝑎 Δ𝜀𝑎𝑎 , 𝜎𝑎𝑏 = 𝜎𝑎𝑏 + 2𝐺𝑎𝑏 Δ𝜀𝑎𝑏 ,
trial trial
𝑛+1 𝑛 𝑛+1 𝑛 (22.126.4)
𝜎𝑏𝑏 = 𝜎𝑏𝑏 + 𝐸𝑏𝑏 Δ𝜀𝑏𝑏 , 𝜎𝑏𝑐 = 𝜎𝑏𝑐 + 2𝐺𝑏𝑐 Δ𝜀𝑏𝑐 ,
trial trial
𝑛+1 𝑛 𝑛+1 𝑛
𝜎𝑐𝑐 = 𝜎𝑐𝑐 + 𝐸𝑐𝑐 Δ𝜀𝑐𝑐 , 𝜎𝑐𝑎 = 𝜎𝑐𝑎 + 2𝐺𝑐𝑎 Δ𝜀𝑐𝑎 ,

20-210 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

We then independently check each component of the updated stresses to ensure that
they do not exceed the permissible values determined from the load curves, e.g., if
trial
∣𝜎𝑖𝑗𝑛+1 ∣ > 𝜆𝜎𝑖𝑗 (𝜀𝑖𝑗 ), (22.126.5)

then
trial
𝜆𝜎𝑖𝑗𝑛+1
𝜎𝑖𝑗𝑛+1 = 𝜎𝑖𝑗 (𝜀𝑖𝑗 ) trial
. (22.126.6)
∣𝜎𝑖𝑗𝑛+1 ∣

The components of 𝜎𝑖𝑗 (𝜀𝑖𝑗 ) are defined by load curves. The parameter 𝜆 is either
unity or a value taken from the load curve number, LCSR, that defines 𝜆 as a function of
strain-rate. Strain-rate is defined here as the Euclidean norm of the deviatoric strain-rate
tensor.

For fully compacted material we assume that the material behavior is elastic-
perfectly plastic and updated the stress components according to:
𝑛+1⁄2
𝑠trial
𝑖𝑗 = 𝑠𝑛𝑖𝑗 + 2𝐺Δ𝜀dev
𝑖𝑗 , (22.126.7)

where the deviatoric strain increment is defined as


1
Δ𝜀dev
𝑖𝑗 = Δ𝜀𝑖𝑗 − Δ𝜀𝑘𝑘 𝛿𝑖𝑗 . (22.126.8)
3

We now check to see if the yield stress for the fully compacted material is exceeded
by comparing
⁄21
3 (22.126.9)
𝑠trial = ( 𝑠trial 𝑠 trial
) ,
eff
2 𝑖𝑗 𝑖𝑗
the effective trial stress to the yield stress, σy . If the effective trial stress exceeds the yield
stress, we simply scale back the stress components to the yield surface
𝜎y
𝑠𝑛+1
𝑖𝑗 = trial 𝑠trial
𝑖𝑗 . (22.126.10)
𝑠eff

We can now update the pressure using the elastic bulk modulus, K
𝑛+1⁄2
𝑝𝑛+1 = 𝑝𝑛 − 𝐾Δ𝜀𝑘𝑘 ,
(22.126.11)
𝐸
𝐾= ,
3(1 − 2𝜈)
and obtain the final value for the Cauchy stress
𝜎𝑖𝑗𝑛+1 = 𝑠𝑛+1
𝑖𝑗 − 𝑝𝑛+1 𝛿𝑖𝑗 . (22.126.12)

LS-DYNA DEV 10/27/16 (r:8004) 20-211 (Material Models)


Material Models LS-DYNA Theory Manual

After completing the stress update we transform the stresses back to the global
configuration.

22.126.1 Stress Update


If LCA < 0, a transversely anisotropic yield surface is obtained where the uniaxial
limit stress, 𝜎 y (𝜑, 𝜀vol ), can be defined as a function of angle 𝜑 with the strong axis and
volumetric strain, 𝜀vol . Elastically, the new material model is assumed to behave exactly as
material 126 (with the restriction 𝐸22𝑢 = 𝐸33𝑢 and 𝐺12𝑢 = 𝐺13𝑢 ), see the LS-DYNA Keyword
User’s Manual [Hallquist 2003]. As for the plastic behavior, a natural question that arises is
how to define the limit stress for a general multiaxial stress state that reduces to the
uniaxial limit stress requirement when the stress is uniaxial. Having given it some thought,
we feel that it is most convenient to work with the principal stresses and the corresponding
directions to achieve this goal.

Assume that the elastic update results in a trial stress 𝜎 trial in the material coordinate
system. This stress tensor is diagonalized to obtain the principal stresses 𝜎1trial , 𝜎2trial and
𝜎3trial and the corresponding principal directions 𝐪1 , 𝐪2 and 𝐪3 relative to the material
coordinate system. The angle that each direction makes with the strong axis of anisotropy
𝐞1 is given by

Curve extends into negative strain


quadrant since LS-DYNA will
extrapolate using the two end points.
It is important that the extropolation
does not extend into the negative stress
region.

σij

unloading and
reloading path

0 Strain: -εij
Unloading is based on the interpolated Young’s
moduli which must provide an unloading
tangent that exceeds the loading tangent.
Figure 22.126.1. Stress quantity versus strain. Note that the “yield stress” at a
strain of zero is nonzero. In the load curve definition the “time” value is the
directional strain and the “function” value is the yield stress.

20-212 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

𝜑𝑖 = arccos∣𝐪𝑖 ⋅ 𝐞1 ∣, 𝑖 = 1, 2, 3.
(22.126.13)

Now a limit stress in the direction of a general multiaxial stress is determined as a


convex combination of the uniaxial limit stress in each principal direction
3
∑𝑗=1 𝜎 Y (𝜙j , 𝜀vol )𝜎𝑗trial 𝜎𝑗trial
Y trial (22.126.14)
𝜎 (𝜎 )= 3
,
∑𝑘=1 𝜎𝑘trial 𝜎𝑘trial
Each of the principal stresses is updated as


⎜ 𝜎 Y (𝜎 trial ) ⎞

𝜎𝑖 = 𝜎𝑖trial min ⎜

⎜ 1, ⎟

⎟ , (22.126.15)
⎜ 3 trial trial

⎝ √∑𝑘=1 𝜎𝑘 𝜎𝑘 ⎠
and the new stress is transformed back to the material coordinate system3.

This stress update is not uniquely defined when the stress tensor possesses multiple
eigenvalues, thus the following simple set of rules is applied. If all principal stresses are
equal, one of the principal directions is chosen to coincide with the strong axis of
anisotropy. If two principal stresses are equal, then one of the directions corresponding to
this stress value is chosen perpendicular to the strong axis of anisotropy.

22.126.2 Support for Independent Shear and Hydrostatic Yield Stress Limit
The model just described turned out to be weak in shear [Okuda 2003] and there
were no means of adding shear resistance without changing the behavior in pure uniaxial
compression. We propose the following modification of the model where the user can
prescribe the shear and hydrostatic resistance in the material without changing the uniaxial
behavior.

Assume that the elastic update results in a trial stress 𝜎 trial in the material coordinate
system. This stress tensor is diagonalized to obtain the principal stresses 𝜎1trial , 𝜎2trial and
𝜎3trial and the corresponding principal directions 𝐪1 , 𝐪2 and 𝐪3 relative to the material
coordinate system. For this discussion we assume that the principal stress values are
ordered so that 𝜎1trial ≤ 𝜎2trial ≤ 𝜎3trial . Two cases need to be treated.

3 Since each component of the stress tensor is scaled by the same factor in Equation 19.126.14, the stress is in
practice not transformed back but the scaling is performed on the stress in the material coordinate system.
LS-DYNA DEV 10/27/16 (r:8004) 20-213 (Material Models)
Material Models LS-DYNA Theory Manual

If 𝜎1trial ≤ −𝜎3trial then the principal stress value with largest magnitude is 𝜎1trial , and
consequently 𝜎1trial ≤ 0. Let
𝜎𝑢 = 𝜎1trial + max(∣𝜎2trial ∣, ∣𝜎3trial ∣),
1 (22.126.16)
𝜎p = {−max(∣𝜎2trial ∣, ∣𝜎3trial ∣) + 𝜎2trial + 𝜎3trial },
3
and finally

𝜎d1 = − max(∣𝜎2trial ∣, ∣𝜎3trial ∣) − 𝜎p

𝜎d2 = 𝜎2trial − 𝜎p , (22.126.17)

𝜎d3 = 𝜎3trial − 𝜎p .

The total stress is the sum of a uniaxial stress represented by 𝜎u , a hydrostatic stress
represented by 𝜎p and a deviatoric stress represented by 𝜎d1 , 𝜎d2 and 𝜎d3 . The angle that the
direction of 𝜎u makes with the strong axis of anisotropy 𝐞1 is given by 𝜙 = arccos∣𝐪1 ⋅ 𝐞1 ∣.

Now a limit stress for the general multiaxial stress is determined as a convex
combination of the three stress contributions as follows
𝜎 Y (𝜎 trial )

𝜎uY (𝜙, 𝜀vol )𝜎u2 + 3√3𝜎pY (𝜀vol )𝜎p2 + √2𝜎dY (𝜀vol ){(𝜎d1 )2 + (𝜎d2 )2 + (𝜎d3 )2 } (22.126.18)
= .
𝜎u2 + 3𝜎p2 + (𝜎d1 )2 + (𝜎d2 )2 + (𝜎d3 )2

Here 𝜎uY (𝜙, εvol ) is the prescribed uniaxial stress limit, 𝜎pY (𝜀vol ) is the hydrostatic stress
limit and 𝜎dY (𝜀vol ) is the stress limit in simple shear. The input for the first of these is
exactly as for the old model. The other two functions are for now written
𝜎pY (𝜀vol ) = 𝜎pY + 𝜎 S (𝜀vol ),
(22.126.19)
𝜎dY (𝜀vol ) = 𝜎dY S
+ 𝜎 (𝜀 vol
),
where 𝜎pY and 𝜎dY are user prescribed constant stress limits and 𝜎 S is the function
describing the densification of the material when loaded in the direction of the strong
material axis. We use the keyword parameters ECCU and GCAU for the new input as
follows.
• ECCU σdY , average stress limit (yield) in simple shear.
• GCAU σpY , average stress limit (yield) in hydrostatic compression (pressure).
Both of these parameters should be positive.

Each of the principal stresses is updated as

20-214 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

⎛⎜ ⎞
𝜎 Y (𝜎 trial )

𝜎𝑖 = ⎜1,
𝜎𝑖trial min ⎜ ⎟,
⎟ (22.126.20)

⎜ ⎟

3 trial trial
⎝ √ ∑ 𝜎
𝑘=1 𝑘
𝜎𝑘 ⎠

and the new stress is transformed back to the material coordinate system.

If σ3trial ≥ −σ1trial then the principal stress value with largest magnitude is σ3trial and
consequently σ3trial ≥ 0. Let

𝜎u = 𝜎3trial − max(∣𝜎2trial ∣, ∣𝜎1trial ∣),


1 (22.126.21)
𝜎p = {max(∣𝜎2trial ∣, ∣𝜎1trial ∣) + 𝜎2trial + 𝜎1trial },
3
and finally
𝜎d1 = 𝜎1trial − 𝜎p ,

𝜎d2 = 𝜎2trial − 𝜎p , (22.126.22)

𝜎d3 = max(∣𝜎2trial ∣, ∣𝜎1trial ∣) − 𝜎p .

The angle that the direction of 𝜎u makes with the strong axis of anisotropy 𝐞1 is
given by 𝜙 = arccos∣𝐪3 ⋅ 𝐞1 ∣. The rest of the treatment is the same as for the case when
𝜎1trial ≤ −𝜎3trial . To motivate the model, let us consider three states of stress.

1. For a uniaxial stress 𝜎, we have 𝜎u = 𝜎 and 𝜎p = 𝜎d1 = 𝜎d2 = 𝜎d3 = 0. This leads us
to 𝜎 Y (𝜎 trial ) = 𝜎uY (𝜙, 𝜀vol ) and hence the stress level will be limited by the user
prescribed uniaxial stress limit.

2. For a simple shear 𝜎, we have 𝜎d1 = −𝜎d3 = −𝜎 and 𝜎p = 𝜎d2 = 𝜎u = 0. Hence


𝜎 Y (𝜎 trial ) = √2𝜎dY (𝜀vol ) and the stress level will be limited by the user prescribed
shear stress limit.

3. For a pressure 𝜎, we have 𝜎p = 𝜎 and 𝜎u = 𝜎d1 = 𝜎d2 = 𝜎d3 = 0. Hence


𝜎 Y (𝜎 trial ) = √3𝜎pY (𝜀vol ) and the stress level will be limited by the user prescribed
hydrostatic stress limit.

LS-DYNA DEV 10/27/16 (r:8004) 20-215 (Material Models)


Material Models LS-DYNA Theory Manual

22.127 Material Model 127: Arruda-Boyce Hyperviscoelastic


Rubber
This material model, described in the paper by Arruda and Boyce [1993], provides a
rubber model that is optionally combined with linear viscoelasticity. Rubber is generally
considered to be fully incompressible since the bulk modulus greatly exceeds the shear
modulus in magnitude; therefore, to model the rubber as an unconstrained material, a
hydrostatic work term, 𝑊H (𝐽), is included in the strain energy functional which is a
function of the relative volume, 𝐽, [Ogden, 1984]:

1 1 11
𝑊(𝐽1 , 𝐽2 , 𝐽) = 𝑛𝑘𝜃 [ (𝐽1 − 3) + (𝐽12 − 9) + (𝐽 3 − 27)]
2 20𝑁 1050𝑁 2 1
19 519
+ 𝑛𝑘𝜃 [ 3
(𝐽14 − 81) + 4
(𝐽15 − 243)] + 𝑊H (𝐽),
7000𝑁 673750𝑁 (22.127.1)
−1⁄3
𝐽1 = 𝐼1 𝐽 ,
𝐽2 = 𝐼2 𝐽.

The hydrostatic work term is expressed in terms of the bulk modulus, 𝐾, and 𝐽, as:
𝐾
(𝐽 − 1)2 .
𝑊H (𝐽) = (22.127.2)
2
Rate effects are taken into account through linear viscoelasticity by a convolution integral
of the form:
𝑡 𝜕𝜀𝑘𝑙
𝜎𝑖𝑗 = ∫ 𝑔𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) 𝑑𝜏, (22.127.3)
0 𝜕𝜏
or in terms of the second Piola-Kirchhoff stress, 𝑆𝑖𝑗 , and Green's strain tensor, 𝐸𝑖𝑗 ,
𝑡 𝜕𝐸𝑘𝑙
𝑆𝑖𝑗 = ∫ 𝐺𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) 𝑑𝜏, (22.127.4)
0 𝜕𝜏
where 𝑔𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) and 𝐺𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) are the relaxation functions for the different stress
measures. This stress is added to the stress tensor determined from the strain energy
functional.

If we wish to include only simple rate effects, the relaxation function is represented
by six terms from the Prony series:
𝑁
𝑔(𝑡) = 𝛼0 + ∑ 𝛼𝑚 𝑒−𝛽𝑡 , (22.127.5)
𝑚=1

given by,

20-216 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

𝑛
𝑔(𝑡) = ∑ 𝐺𝑖 𝑒−𝛽𝑖𝑡 . (22.127.6)
𝑖=1

This model is effectively a Maxwell fluid which consists of a dampers and springs in series.
We characterize this in the input by shear modulii, 𝐺𝑖 , and decay constants, 𝛽𝑖 . The
viscoelastic behavior is optional and an arbitrary number of terms may be used.

LS-DYNA DEV 10/27/16 (r:8004) 20-217 (Material Models)


Material Models LS-DYNA Theory Manual

22.128 Material Model 128: Heart Tissue


This material model provides a tissue model described in the paper by Guccione,
McCulloch, and Waldman [1991]

The tissue model is described in terms of the energy functional in terms of the Green
strain components, 𝐸𝑖𝑗 ,
𝐶 𝑄
𝑊(𝐸, 𝐽) = (𝑒 − 1) + 𝑊H (𝐽),
2 (22.128.1)
𝑄= 𝑏1 𝐸211 + 𝑏2 (𝐸222 + 𝐸233 + 𝐸223 + 𝐸232 ) + 𝑏3 (𝐸212 + 𝐸221 + 𝐸213 + 𝐸231 ),

where the hydrostatic work term is in terms of the bulk modulus, 𝐾, and the third invariant
𝐽, as:
𝐾
(𝐽 − 1)2 .
𝑊H (𝐽) = (22.128.2)
2
The Green components are modified to eliminate any effects of volumetric work following
the procedures of Ogden.

20-218 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.129 Material Model 129: Isotropic Lung Tissue


This material model provides a lung tissue model described in the paper by Vawter
[1980].

The material is described by a strain energy functional expressed in terms of the


invariants of the Green Strain:
𝐶 (𝛼𝐼12+𝛽𝐼2) 12𝐶1
𝑊(𝐼1 , 𝐼2 ) = 𝑒 + [𝐴(1+𝐶2) − 1],
2𝛥 𝛥(1 + 𝐶2 )
(22.129.1)
4
𝐴2 = (𝐼1 + 𝐼2 ) − 1,
3
where the hydrostatic work term is in terms of the bulk modulus, 𝐾, and the third invariant
𝐽, as:
𝐾
(𝐽 − 1)2 ,
𝑊H (𝐽) = (22.129.2)
2
Rate effects are taken into account through linear viscoelasticity by a convolution integral
of the form:
𝑡 𝜕𝜀𝑘𝑙
𝜎𝑖𝑗 = ∫ 𝑔𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) 𝑑𝜏, (22.129.3)
0 𝜕𝜏
or in terms of the second Piola-Kirchhoff stress, 𝑆𝑖𝑗 , and Green's strain tensor, 𝐸𝑖𝑗 ,
𝑡 𝜕𝐸𝑘𝑙
𝑆𝑖𝑗 = ∫ 𝐺𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) 𝑑𝜏, (22.129.4)
0 𝜕𝜏
where 𝑔𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) and 𝐺𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) are the relaxation functions for the different stress
measures. This stress is added to the stress tensor determined from the strain energy
functional.

If we wish to include only simple rate effects, the relaxation function is represented
by six terms from the Prony series:
𝑁
𝑔(𝑡) = 𝛼0 + ∑ 𝛼𝑚 𝑒−𝛽𝑡 , (22.129.5)
𝑚=1

given by,
𝑛
𝑔(𝑡) = ∑ 𝐺𝑖 𝑒−𝛽𝑖𝑡 . (22.129.6)
𝑖=1

LS-DYNA DEV 10/27/16 (r:8004) 20-219 (Material Models)


Material Models LS-DYNA Theory Manual

This model is effectively a Maxwell fluid which consists of a dampers and springs in
series. We characterize this in the input by shear moduli, 𝐺𝑖 , and decay constants, 𝛽𝑖 . The
viscoelastic behavior is optional and an arbitrary number of terms may be used.

20-220 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.130 Material Model 130: Special Orthotropic


The in-plane elastic matrix for in-plane plane stress behavior is given by:
𝑄 𝑄12p 0 0 0
⎡ 11p ⎤
⎢𝑄12p 𝑄22p 0 0 0 ⎥
⎢ ⎥
𝐂in plane =⎢ 0 0 𝑄44p 0 0 ⎥, (22.130.1)
⎢ 0 0 0 𝑄55p 0 ⎥
⎢ ⎥
⎣ 0 0 0 0 𝑄66p ⎦
where the terms 𝑄𝑖𝑗p are defined as:
𝐸11p
𝑄11p = ,
1 − ν12p ν21p
𝐸22p
𝑄22p = ,
1 − ν12p ν21p
ν12p E11p (22.130.2)
𝑄12p = ,
1 − ν12p ν21p
𝑄44p = 𝐺12p ,
𝑄55p = 𝐺23p ,
Q66p = 𝐺31p .
The elastic matrix for bending behavior is given by:
𝑄11b 𝑄12b 0
𝐂bending ⎡
= ⎢𝑄12b 𝑄22b 0 ⎤ ⎥, (22.130.3)
⎣ 0 0 𝑄44b ⎦
where the terms 𝑄𝑖𝑗b are similarly defined.

LS-DYNA DEV 10/27/16 (r:8004) 20-221 (Material Models)


Material Models LS-DYNA Theory Manual

22.131 Material Model 131: Isotropic Smeared Crack


The following documentation is taken nearly verbatim from the documentation of
that by Lemmen and Meijer [2001].

Three methods are offered to model progressive failure. The maximum principal
stress criterion detects failure if the maximum (most tensile) principal stress exceeds 𝜎max .
Upon failure, the material can no longer carry stress.

The second failure model is the smeared crack model with linear softening stress-
strain curve using equivalent uniaxial strains. Failure is assumed to be perpendicular to
the principal strain directions. A rotational crack concept is employed in which the crack
directions are related to the current directions of principal strain. Therefore crack
directions may rotate in time. Principal stresses are expressed as
𝜎1 𝐸̅̅̅̅ 0 0 𝜀̃ ̅̅̅̅1 𝜀̃1
𝐸
⎡ 1 ⎤ ⎛ 1⎞ ⎛ ⎜ ⎞
⎛ ⎞
⎜𝜎2 ⎟ = ⎢
⎢0 ̅̅̅̅2
𝐸 0 ⎥ ⎥ ⎜𝜀̃2 ⎟ = ⎜
⎜𝐸̅̅̅̅2 𝜀̃2 ⎟

⎟, (22.131.1)
⎝𝜎3 ⎠ ⎣0 0 ̅𝐸̅̅̅3 ⎦ ⎝𝜀̃3 ⎠ ⎝𝐸 ̅̅̅̅3 𝜀̃3 ⎠
̅̅̅̅1 , 𝐸
with 𝐸 ̅̅̅̅2 and 𝐸
̅̅̅̅3 being secant stiffness in the terms that depend on internal variables.

In the model developed for DYCOSS it has been assumed that there is no interaction
between the three directions in which case stresses simply follow

{𝐸𝜀̃𝑗 if 0 ≤ 𝜀̃𝑗 ≤ 𝜀̃𝑗,ini
{
{
{ 𝜀̃𝑗 − 𝜀̃𝑗,ini
{
𝜎𝑗 (𝜀̃𝑗 ) = ⎨(1 − )𝜎
̅̅̅̅ if 𝜀̃𝑗,ini < 𝜀̃𝑗 ≤ 𝜀̃𝑗,ult , (22.131.2)
{ 𝜀̃𝑗,ult − 𝜀̃𝑗,ini
{
{
{0 if 𝜀̃𝑗 > 𝜀̃𝑗,ult
{

with 𝜎
̅̅̅̅ the ultimate stress, 𝜀̃𝑗,ini the damage threshold, and 𝜀̃𝑗,ult the ultimate strain in 𝑗-
direction. The damage threshold is defined as
̅̅̅̅
𝜎
𝜀̃𝑗,ini = . (22.131.3)
𝐸

The ultimate strain is obtained by relating the crack growth energy and the
dissipated energy

∫∫𝜎
̅̅̅̅ 𝑑𝜀̃𝑗,ult 𝑑𝑉 = 𝐺𝐴, (22.131.4)

with 𝐺 the energy release rate, 𝑉 the element volume and 𝐴 the area perpendicular to the
principal strain direction. The one-point elements in LS-DYNA have a single integration

20-222 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

point and the integral over the volume may be replaced by the volume. For linear
softening it follows
2𝐺𝐴
𝜀̃𝑗,ult = , (22.131.5)
𝑉𝜎̅̅̅̅

The above formulation may be regarded as a damage equivalent to the maximum


principle stress criterion.

The third model is a damage model presented by Brekelmans et. al. [1991]. Here
the Cauchy stress tensor 𝜎 is expressed as
𝜎 = (1 − 𝐷)𝐸𝜀, (22.131.6)
where 𝐷 represents the current damage and the factor (1 − 𝐷) is the reduction factor
caused by damage. The scalar damage variable is expressed as a function of a so-called
damage equivalent strain 𝜀d
𝜀ini (𝜀ult − 𝜀d )
𝐷 = 𝐷(𝜀d ) = 1 − , (22.131.7)
𝜀d (𝜀ult − 𝜀ini )
with ultimate and initial strains as defined by
2
𝑘−1 1 𝑘−1 6𝑘 (22.131.8)
𝜀d = 𝐽1 + √ ( 𝐽1 ) + 𝐽 ,
2𝑘(1 − 2𝑣) 2𝑘 1 − 2𝑣 (1 + 𝑣)2 2
where the constant 𝑘 represents the ratio of the strength in tension over the strength in
compression
𝜎ult,tension
𝑘= , (22.131.9)
𝜎ult,compression
𝐽1 and 𝐽2 are the first and the second invariants of the strain tensor representing the
volumetric and the deviatoric straining, respectively
𝐽1 = 𝜀𝑖𝑗 ,
(22.131.10)
𝐽2 = 𝜀′𝑖𝑗 𝜀′𝑖𝑗 ,
where 𝜀′𝑖𝑗 denotes the deviatoric part of the strain tensor.

If the compression and tension strength are equal the dependency on the volumetric
strain vanishes in Equation (22.131.8) and failure is shear dominated. If the compressive
strength is much larger than the strength in tension, 𝑘 becomes large and the 𝐽1 terms in
Equation (22.131.8) dominate the behavior.

LS-DYNA DEV 10/27/16 (r:8004) 20-223 (Material Models)


Material Models LS-DYNA Theory Manual

22.133 Material Model 133: Barlat_YLD2000


The yield condition for this material can be written
𝑓 (𝜎, 𝛼, 𝜀p ) = 𝜎eff (𝜎𝑥𝑥 − 2𝛼𝑥𝑥 − 𝛼𝑦𝑦 , 𝜎𝑦𝑦 − 2𝛼𝑦𝑦 − 𝛼𝑥𝑥 , 𝜎𝑥𝑦 − 𝛼𝑥𝑦 ) −
(22.133.1)
𝜎Yt (𝜀p , 𝜀̇p , 𝛽) ≤ 0,
where
1
1 a
𝜎eff (𝑠𝑥𝑥 , 𝑠𝑦𝑦 , 𝑠𝑥𝑦 ) = ⎛
⎜ (𝜑′ + 𝜑′′ )⎞
⎟ ,
⎝2 ⎠ (22.133.2)
𝜑′ = ∣𝑋′1 − 𝑋′2 ∣𝑎 ,
𝜑′′ = ∣2𝑋 ′′1 + 𝑋 ′′2 ∣𝑎 + ∣𝑋 ′′1 + 2𝑋 ′′2 ∣𝑎 .

The 𝑋′𝑖 and 𝑋′′𝑖 are the eigenvalues of 𝑋′𝑖𝑗 and 𝑋′′𝑖𝑗 and are given by
1
𝑋′1 = (𝑋′11 + 𝑋′22 + √(𝑋′11 − 𝑋′22 )2 + 4𝑋′212 ) ,
2
(22.133.3)
1
𝑋′2 = (𝑋′11 + 𝑋′22 − √(𝑋′11 − 𝑋′22 )2 + 4𝑋′212 ),
2
and
1
𝑋 ′′1 = (𝑋 ′′11 + 𝑋 ′′22 + √(𝑋′′11 − 𝑋′′22 )2 + 4𝑋′′212 ) ,
2
(22.133.4)
1
𝑋 ′′2 = (𝑋 ′′11 + 𝑋 ′′22 − √(𝑋′′11 − 𝑋′′22 )2 + 4𝑋′′212 ),
2
respectively. The 𝑋′𝑖𝑗 and 𝑋′′𝑖𝑗 are given by4
𝑋′ 𝐿′ 𝐿′12 0 𝑠𝑥𝑥
⎛ 11 ⎞ ⎛ 11 ⎞ ⎛𝑠 ⎞
⎜𝑋′22 ⎟ = ⎜𝐿′21 𝐿′22 0 ⎟ ⎜ 𝑦𝑦 ⎟ ,
⎝𝑋′12 ⎠ ⎝0 0 𝐿′33 ⎠ ⎝𝑠𝑥𝑦 ⎠
(22.133.5)
𝑋 ′′11
𝐿 ′′11
𝐿′′12
0 𝑠𝑥𝑥
⎛𝑋 ′′22 ⎟
⎜ ⎛𝐿′′21
⎞=⎜ 𝐿′′22 0 ⎟ ⎞ ⎜𝑠𝑦𝑦 ⎞
⎛ ⎟,
⎝𝑋 ′′12 ⎠ ⎝0 ′′33 ⎠ ⎝𝑠𝑥𝑦 ⎠
0 𝐿
where
𝐿′11 2 0 0

⎜ ⎞

⎜𝐿′ ⎟ ⎛−1 0 0⎞ 𝛼1


12 ⎟
⎟ 1⎜

⎜0


⎟ ⎛ ⎞,

⎜𝐿′ 21 ⎟
⎟ = ⎜
⎜ −1 ⎟
0⎟ ⎜ 𝛼2 ⎟ (22.133.6)

⎜ ⎟ 3⎜ ⎜0 ⎟

0 ⎝𝛼7 ⎠
⎜𝐿′22 ⎟⎟ 2
⎝0 0 3⎠
⎝𝐿′33 ⎠

20-224 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

𝐿′′11 −2 2 8 −2 0 𝛼3

⎜ ′′12 ⎞

𝐿 ⎟ 1⎜ 1 ⎛
⎜ −4 −4 4 0⎞⎟ ⎛
⎜ 𝛼4 ⎞


⎜ ⎟= ⎜ ⎟
⎟ ⎜
⎜ ⎟
⎟.

⎜𝐿′′21 ⎟
⎟ 9⎜
⎜ 4 −4 −4 1 0⎟⎟ ⎜
⎜ 𝛼5⎟⎟

⎜𝐿′′22 ⎟ ⎜
⎜−2 ⎟ ⎜ ⎟
⎟ 8 2 −2 0⎟ ⎜𝛼6 ⎟
⎝𝐿′′33 ⎠ ⎝0 0 0 0 9⎠ ⎝𝛼8 ⎠
where 𝛼1 to 𝛼8 are the parameters that determine the shape of the yield surface.

The yield stress is expressed as

𝜎Yt (𝜀p , 𝜀̇p , 𝛽) = 𝜎Yv (𝜀p , 𝜀̇p ) + 𝛽(𝜎0 − 𝜎Yv (𝜀p , 𝜀̇p )), (22.133.7)

where 𝛽 determines the fraction kinematic hardening and 𝜎0 is the initial yield stress. The
yield stress for purely isotropic hardening is given by
1
⎛ 𝜀̇p p ⎞
𝜎Yv (𝜀p , 𝜀̇p )⎜
= 𝜎Y (𝜀p ) ⎜1 + { } ⎟
⎜ ⎟, (22.133.8)
𝐶 ⎟
⎝ ⎠
where 𝐶 and 𝑝 are the Cowper-Symonds material parameters for strain rate effects.

The evolution of back stress is given by

∂𝜎Y ⎛ 𝜀̇p 1/p ⎞ 1 𝜀̇p 1/𝑝−1 ⎞



⎜ ⎜1 + { } ⎟ + 𝜎Y { } ⎟
⎝ ∂𝜀p ⎝ 𝐶 ⎠ 𝑝𝐶Δ𝑡 𝐶 ⎠ ∂𝜎eff ,
𝛼̇ = 𝜆̇ 𝛽 (22.133.9)
∂𝜎eff ∂𝜎eff ∂𝑠

∂𝑠 ∂𝑠
where Δ𝑡 is the current time step size and 𝜆̇ is the rate of plastic strain multiplier.

For the plastic return algorithms, the gradient of effective stress is computed as
∂𝜎eff 𝜕𝜙′

⎜ ⎞
⎟ ⎛
⎜ ⎞

⎜ ∂𝑠 xx ⎟
⎟ ⎜
⎜ 𝜕𝑋′11 ⎟⎟


⎜ ⎟
⎟ 𝑎−1 L′11 L′21 0 ⎜ ⎜ ⎟

⎜ ∂𝜎 eff ⎟ 𝑎𝜎eff ⎛ ⎞ ⎜ 𝜕𝜙′ ⎟

⎜ ⎟
⎟ = ⎜ ⎟ ⎜ ⎟

⎜ ∂𝑠 ⎟
⎟ ⎜L′12 L′22 0 ⎟⎜ ⎜ ⎟

⎜ yy ⎟ 2 ⎜ 𝜕𝑋′ 22 ⎟

⎜ ⎟
⎟ ⎝ 0 0 ⎜
L′33 ⎠ ⎜ ⎟

⎜∂𝜎eff ⎟
⎜ ⎟ ⎜
⎜ 𝜕𝜙′ ⎟

⎝ ∂𝑠xy ⎠ ⎝𝜕𝑋′12 ⎠
(22.133.10)
∂𝜙′′

⎜ ⎞


⎜ ∂X′′ 11 ⎟

𝑎−1 L′′11 L′′21 0 ⎜
⎜ ⎟

𝑎𝜎eff ⎛
⎜ ⎞
⎟ ⎜
⎜ ∂𝜙′′ ⎟

+ ⎜L′′12 L′′22 0 ⎟⎜ ⎜ ⎟
⎟ ,
2 ⎜ ∂X′′ 22 ⎟
⎝ 0 0 ⎜
L′′33 ⎠ ⎜ ⎟


⎜ ∂𝜙′′ ⎟ ⎟
⎝∂X′′12 ⎠

with the aid of

LS-DYNA DEV 10/27/16 (r:8004) 20-225 (Material Models)


Material Models LS-DYNA Theory Manual

𝜕𝜑′ 𝜕(𝑋′1 − 𝑋′2 )


= 𝑎(𝑋′1 − 𝑋′2 )𝑎−1 , (22.133.11)
𝜕𝑋′𝑖𝑗 𝜕𝑋′𝑖𝑗

𝜕𝜑′′ 𝜕(2𝑋′′1 + 𝑋′′2 )


= 𝑎∣2𝑋′′1 + 𝑋′′2 ∣𝑎−1 sgn(2𝑋′′1 + 𝑋′′2 ) +
𝜕𝑋′′𝑖𝑗 𝜕𝑋′′𝑖𝑗
′′2
(22.133.12)
′′2 ′′1 𝑎−1 ′′2 ′′1 𝜕(2𝑋 + 𝑋 ′′1 )
𝑎∣2𝑋 + 𝑋 ∣ sgn(2𝑋 + 𝑋 ) ′′ ,
𝜕𝑋 𝑖𝑗
and
𝜕(𝑋′1 − 𝑋′2 ) 𝑋′11 − 𝑋′22
= , (22.133.13)
𝜕𝑋′11 2 2
√(𝑋′11 − 𝑋′22 ) + 4𝑋′12

𝜕(𝑋′1 − 𝑋′2 ) 𝑋′22 − 𝑋′11


= , (22.133.14)
𝜕𝑋′22 √(𝑋′11 − 𝑋′22 )2 + 4𝑋′212

𝜕(𝑋′1 − 𝑋′2 ) 4𝑋′12


= , (22.133.15)
𝜕𝑋′12 √(𝑋′ 2 2
11 − 𝑋′22 ) + 4𝑋′12

𝜕(2𝑋′′1 + 𝑋′′2 ) 3 1 𝑋′′11 − 𝑋′′22


= + , (22.133.16)
𝜕𝑋′′11 2 2 √(𝑋′′ − 𝑋′′ )2 + 4𝑋′′2
11 22 12

𝜕(2𝑋′′1 + 𝑋′′2 ) 3 1 𝑋′′22 − 𝑋′′11


= + , (22.133.17)
𝜕𝑋′′22 2 2 √(𝑋′′ − 𝑋′′ )2 + 4𝑋′′2
11 22 12

𝜕(2𝑋′′1 + 𝑋′′2 ) 2𝑋′′12


= , (22.133.18)
𝜕𝑋′′12 √(𝑋′′ )2 4𝑋′′212
11 − 𝑋′′22 +

𝜕(2𝑋′′2 + 𝑋′′1 ) 3 1 𝑋 ′′11 − 𝑋 ′′22


= − , (22.133.19)
𝜕𝑋′′11 2 2 √(𝑋′′ − 𝑋′′ )2 + 4𝑋′′2
11 22 12

𝜕(2𝑋′′2 + 𝑋′′1 ) 3 1 𝑋 ′′22 − 𝑋 ′′11


= − , (22.133.20)
𝜕𝑋′′22 2 2 √(𝑋′′ − 𝑋′′ )2 + 4𝑋′′2
11 22 12

𝜕(2𝑋′′2 + 𝑋′′1 ) 2𝑋 ′′12


=− . (22.133.21)
𝜕𝑋′′12 √(𝑋′′ 2 2
11 − 𝑋′′22 ) + 4𝑋′′12

The algorithm for the plane stress update as well as the formula for the tangent
modulus is given in detail in Section 19.36.1 and is not repeated here.
20-226 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Material Models

22.133.1 Closest point projection algorithm


This section describes shortly the closest point projection algorithm that was
implemented to improve accuracy, hence the implicit performance, of the model. The
closest point projection comes down to solving the following system of equations
t t
𝑓1 = 𝑡 + 𝐀𝛼 + (𝛔Y (Δ𝜆) − 𝛔Y (0))ℎ − 𝛔trial (Δ𝜀33 ) + 2𝐺Δ𝜆D∇𝜎eff (t) = 0, (22.133.22)
t
𝑓2 = −𝛔eff (t) + 𝛔Y (Δ𝜆) = 0, (22.133.23)

trial 1 2
𝑓3 = 𝜎33 (Δ𝜀33 ) + 2𝐺Δ𝜆(∇𝜎eff (t) + ∇𝜎eff (t)) = 0, (22.133.24)
where
𝛽
ℎ= 𝐁∇𝜎eff (𝑡), (22.133.25)
∇𝜎eff (𝑡)T 𝐁∇𝜎eff (𝑡)
in terms of the unknown variables 𝐭 (stress), Δε33 (thickness strain increment) and Δ𝜆
(plastic strain increment). In the above
1 2 1 2 1 t
∂σY

𝐃=⎢ 1 ⎤, ⎡
𝐀 = ⎢1 2 ⎤, ⎡
𝐁 = ⎢1 2 ⎤, 𝐻= . (22.133.26)
⎥ ⎥ ⎥
⎣ 0.5⎦ ⎣ 1 ⎦ ⎣ 0.5⎦ ∂εp

This system of equations is solved using a Newton method with an additional line
search for robustness. Using the notation
𝑓 t
⎡ 1⎤
𝐟 = ⎢𝑓2 ⎥ , 𝐱=⎡
⎢Δ𝜆 ⎥,
⎤ (22.133.27)
⎣𝑓3 ⎦ ⎣Δ𝜀33 ⎦
a Newton step is completed as
−1
∂𝑓
𝑥+ = 𝑥− − 𝑠 ( ) f (22.133.28)
∂𝑥
for a step size 𝑠 ≤ 1 chosen such that the norm of the objective function is decreasing. The
gradient of the objective function is given by
∇𝑓 = ∇𝑓1−β + ∇𝑓β (22.133.29)
where
I + 2𝐺Δ𝜆𝐷∇2 𝜎eff 2𝐺𝐷∇𝜎eff −𝐶3
⎡ ⎤
∇f1−β =⎢
⎢ −(∇𝜎eff )T 𝐻 0 ⎥⎥ (22.133.30)
⎣ 2𝐺Δ𝜆𝐞T ∇2 𝛔eff T
2Ge ∇σeff C33 ⎦

LS-DYNA DEV 10/27/16 (r:8004) 20-227 (Material Models)


Material Models LS-DYNA Theory Manual

∂h
⎡Δ𝜆𝐻 𝐻ℎ 0⎤
∇fβ = ⎢
⎢ ∂𝑡 ⎥
⎥ (22.133.31)
⎢0T 0 0⎥
⎣0 0 0⎦
and
1 2G 4G
𝐞=⎡
⎢1⎤⎥, C3 = (K − ) e, C33 = (K + ), (22.133.32)
⎣0⎦ 3 3

𝐺 and 𝐾 stands for the shear and bulk modulus, respectively. This algorithm requires
computation of the effective stress hessian. The derivation of this is quite straightforward
but the expression for it is rather long and is hence omitted in this report.

20-228 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.134 Material Model 134: Viscoelastic Fabric


The viscoelastic fabric model is a variation on the general viscoelastic Material
Model 76. This model is valid for 3 and 4 node membrane elements only and is strongly
recommended for modeling isotropic viscoelastic fabrics where wrinkling may be a
problem. For thin fabrics, buckling can result in an inability to support compressive
stresses; thus, a flag is included for this option. If bending stresses are important, use a
shell formulation with Model 76.

Rate effects are taken into account through linear viscoelasticity by a convolution
integral of the form:
𝑡 𝜕𝜀𝑘𝑙
𝜎𝑖𝑗 = ∫ 𝑔𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) 𝑑𝜏, (22.134.1)
0 𝜕𝜏
where 𝑔𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) is the relaxation function.

If we wish to include only simple rate effects for the deviatoric stresses, the
relaxation function is represented by six terms from the Prony series:
𝑁
𝑔(𝑡) = ∑ 𝐺𝑚 𝑒−𝛽𝑚 𝑡 . (22.134.2)
𝑚=1

We characterize this in the input by shear modulii, 𝐺𝑖 , and decay constants, 𝛽𝑖 . An


arbitrary number of terms, up to 6, may be used when applying the viscoelastic model.

For volumetric relaxation, the relaxation function is also represented by the Prony
series in terms of bulk modulii:
𝑁
𝑘(𝑡) = ∑ 𝐾𝑚 𝑒−𝛽𝑘𝑚 𝑡 . (22.134.3)
𝑚=1

LS-DYNA DEV 10/27/16 (r:8004) 20-229 (Material Models)


Material Models LS-DYNA Theory Manual

22.139 Material Model 139: Modified Force Limited


This material model is available for the Belytschko resultant beam element only.
Plastic hinges form at the ends of the beam when the moment reaches the plastic moment.
The plastic moment versus rotation relationship is specified by the user in the form of a
load curve and scale factor. The points of the load curve are (plastic rotation in radians,
plastic moment). Both quantities should be positive for all points, with the first point being
(zero, initial plastic moment). Within this constraint any form of characteristic may be
used, including flat or falling curves. Different load curves and scale factors may be
specified at each node and about each of the local s and t axes.

Axial collapse occurs when the compressive axial load reaches the collapse load.
Collapse load versus collapse deflection is specified in the form of a load curve. The points
of the load curve are either (true strain, collapse force) or (change in length, collapse force).
Both quantities should be entered as positive for all points, and will be interpreted as
compressive. The first point should be (zero, initial collapse load).

The collapse load may vary with end moment as well as with deflections. In this
case several load-deflection curves are defined, each corresponding to a different end
moment. Each load curve should have the same number of points and the same deflection
values. The end moment is defined as the average of the absolute moments at each end of
the beam and is always positive.

Stiffness-proportional damping may be added using the damping factor 𝜆. This is


defined as follows:
2∗𝜉
𝜆= , (22.139.1)
𝜔
where 𝜉 is the damping factor at the reference frequency 𝜔 (in radians per second). For
example if 1% damping at 2Hz is required
2 ∗ 0.01
𝜆= = 0.001592. (22.139.2)
2π ∗ 2
If damping is used, a small time step may be required. LS-DYNA does not check this so to
avoid instability it may be necessary to control the timestep via a load curve. As a guide,
the timestep required for any given element is multiplied by 0.3𝐿/𝑐𝜆 when damping is
present (𝐿 = element length, 𝑐 = sound speed).

Moment Interaction
Plastic hinges can form due to the combined action of moments about the three axes.
This facility is activated only when yield moments are defined in the material input. A
hinge forms when the following condition is first satisfied.

20-230 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

2 2 2
𝑀r 𝑀s 𝑀t
( ) +( ) +( ) ≥ 1, (22.139.3)
𝑀ryield 𝑀syield 𝑀tyield
where,
𝑀r , 𝑀s , 𝑀t = current moments
𝑀ryield , 𝑀syield , 𝑀tyield = yield moments
Note that scale factors for hinge behavior defined in the input will also be applied to the
yield moments: for example, 𝑀syield in the above formula is given by the input yield
moment about the local axis times the input scale factor for the local s-axis. For strain-
softening characteristics, the yield moment should generally be set equal to the initial peak
of the moment-rotation load curve.

On forming a hinge, upper limit moments are set. These are given by
𝑀
𝑀rupper = MAX ⎜⎛𝑀r , ryield ⎟
⎞, (22.139.4)
⎝ 2 ⎠
and similarly for 𝑀s and 𝑀t . Thereafter the plastic moments will be given by:
𝑀rp = min(𝑀rcurve , 𝑀rcurve )
and similarly for 𝑠 and 𝑡, where
𝑀rp = current plastic moment
𝑀rcurve = moment taken from load curve at the current rotation scaled according to
the scale factor.

The effect of this is to provide an upper limit to the moment that can be generated; it
represents the softening effect of local buckling at a hinge site. Thus if a member is bent
about its local s-axis it will then be weaker in torsion and about its local 𝑡-axis. For
moment-softening curves, the effect is to trim off the initial peak (although if the curves
subsequently harden, the final hardening will also be trimmed off).

It is not possible to make the plastic moment vary with the current axial load, but it
is possible to make hinge formation a function of axial load and subsequent plastic moment
a function of the moment at the time the hinge formed. This is discussed in the next
section.

Independent plastic hinge formation


In addition to the moment interaction equation, Cards 7 through 18 allow plastic
hinges to form independently for the s-axis and t-axis at each end of the beam and also for
the torsional axis. A plastic hinge is assumed to form if any component of the current
moment exceeds the yield moment as defined by the yield moment vs. axial force curves
input on cards 7 and 8. If any of the 5 curves is omitted, a hinge will not form for that
component. The curves can be defined for both compressive and tensile axial forces. If the
axial force falls outside the range of the curve, the first or last point in the curve will be
used. A hinge forming for one component of moment does not affect the other
components.
LS-DYNA DEV 10/27/16 (r:8004) 20-231 (Material Models)
Material Models LS-DYNA Theory Manual

Upon forming a hinge, the magnitude of that component of moment will not be
permitted to exceed the current plastic moment. The current plastic moment is obtained by
interpolating between the plastic moment vs. plastic rotation curves input on cards 10, 12,
14, 16, or 18. Curves may be input for up to 8 hinge moments, where the hinge moment is
defined as the yield moment at the time that the hinge formed. Curves must be input in
order of increasing hinge moment and each curve should have the same plastic rotation
values. The first or last curve will be used if the hinge moment falls outside the range of
the curves. If no curves are defined, the plastic moment is obtained from the curves on
cards 4 through 6. The plastic moment is scaled by the scale factors on lines 4 to 6.

A hinge will form if either the independent yield moment is exceeded or if the
moment interaction equation is satisfied. If both are true, the plastic moment will be set to
the minimum of the interpolated value and 𝑀rp .

M8
M7
M6
M5
Axial Force

M4

M3

M2

M1

Strain (or change in length, see AOPT)

Figure 22.139.1. The force magnitude is limited by the applied end moment. For
an intermediate value of the end moment LS-DYNA interpolates between the
curves to determine the allowable force value.

20-232 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.141 Material Model 141: Rate Sensitive Polymer

1 𝑍2 𝑆𝑖𝑗 − 𝛺𝑖𝑗
𝜀𝑖𝑗 = 𝐷o exp [− ( o )] ⎛
⎜ ⎞
⎟, (22.141.1)
2 3𝐾2 ⎝ √𝐾2 ⎠

where 𝐷o is the maximum inelastic strain rate, 𝑍o is the isotropic initial hardness of
material, 𝛺𝑖𝑗 is the internal stress, 𝑆𝑖𝑗 is the deviatoric stress component, and 𝐾2 is defined
as follows:
1
𝐾2 = (𝑆𝑖𝑗 − 𝛺𝑖𝑗 )(𝑆𝑖𝑗 − 𝛺𝑖𝑗 ), (22.141.2)
2
and represent the second invariant of the overstress tensor. The elastic components of the
strain are added to the inelastic strain to obtain the total strain. The following relationship
defines the internal stress variable rate:
2
𝛺𝑖𝑗 = 𝑞𝛺m 𝜀̇I𝑖𝑗 − 𝑞𝛺𝑖𝑗 𝜀̇Ie , (22.141.3)
3
where 𝑞 is a material constant, 𝛺m is a material constant that represents the maximum
value of the internal stress, and 𝜀̇Ie is the effective inelastic strain.

LS-DYNA DEV 10/27/16 (r:8004) 20-233 (Material Models)


Material Models LS-DYNA Theory Manual

22.142 Material Model 142: Transversely Anisotropic


Crushable Foam
A new material model for low density, transversely isotropic crushable foams, has
been developed at DaimlerChrysler by Hirth, Du Bois, and Weimar. Hirth, Du Bois, and
Weimar determined that material model 26, MAT_HONEYCOMB, which is commonly
used to model foams, can systematically over estimate the stress when it is loaded off-axis.
Their new material model overcomes this problem without requiring any additional input.
Their new model can possibly replace the MAT_HONEYCOMB material, which is
currently used in the frontal offset and side impact barriers.

Many polymers used for energy absorption are low density, crushable foams with
no noticeable Poisson effect. Frequently manufactured by extrusion, they are transversely
isotropic. This class of material is used to enhance automotive safety in low velocity
(bumper impact) and medium velocity (interior head impact) applications. These materials
require a transversely isotropic, elastoplastic material with a flow rule allowing for large
permanent volumetric deformations.

The MAT_HONEYCOMB model uses a local coordinate system defined by the user.
One of the axes of the local system coincides with the extrusion direction of the honeycomb
in the undeformed configuration. As an element deforms, its local coordinate system
rotates with its mean rigid body motion. Each of the six stress components is treated
independently, and each has its own law relating its flow stress to its plastic strain.

The effect of off-axis loading on the MAT_HONEYCOMB model can be estimated


by restricting our considerations to plane strain in two dimensions. Our discussion is
restricted to the response of the foam before it becomes fully compacted. After compaction,
its response is modeled with conventional 𝐽2 plasticity. The model reduces to
y
|σ11 | ≤ 𝜎11 (𝜀V ),
y
|𝜎22 | ≤ 𝜎22 (𝜀V ), (22.142.1)
y
|𝜎12 | ≤ 𝜎12 (𝜀V ),
where 𝜀V is the volumetric strain. For a fixed value of volumetric strain, the individual
stress components respond in an elastic-perfectly plastic manner, i.e., the foam doesn’t
have any strain hardening.

In two dimensions, the stress tensor transforms according to


[𝜎] = [𝑅(𝜃)]T [𝜎𝜃 ][𝑅(𝜃)],
cos(𝜃) − sin(𝜃) (22.142.2)
[𝑅(𝜃)] = [ ],
sin(𝜃) cos(𝜃)

20-234 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

where 𝜃 is the angle of the local coordinate system relative to the global system. For
uniaxial loading along the global 1-axis, the stress will be (accounting for the sign of the
volume strain),
y y y
𝜎11 = {[cos(𝜃)]2 𝜎11 + [sin(𝜃)]2 𝜎22 + 2 ⋅ sin(𝜃)cos(𝜃)𝜎12 }sgn(𝜀V )}, (22.142.3)
assuming the strain is large enough to cause yielding in both directions.

y y
If the shear strength is neglected, 𝜎11 will vary smoothly between 𝜎11 and 𝜎22 and
never exceed the maximum of the two yield stresses. This behavior is intuitively what we
would like to see. However, if the value of shear yield stress isn’t zero, 𝜎11 will be greater
y y y y
than either 𝜎11 or 𝜎22 . To illustrate, if 𝜎11 and 𝜎22 are equal (a nominally isotropic
response) the magnitude of the stress is
y y
|𝜎11 | = 𝜎11 + 2 ⋅ sin(𝜃)cos(𝜃)𝜎12 , (22.142.4)
and achieves a maximum value at 45 degrees of
y y
|𝜎11 | = 𝜎11 + 𝜎12 . (22.142.5)

For cases where there is anisotropy, the maximum occurs at a different angle and
will have a different magnitude, but it will exceed the maximum uniaxial yield stress. In
fact, a simple calculation using Mohr’s circle shows that the maximum value will be
y 1 y y 1 y y 2 y
𝜎max = (𝜎11 + 𝜎22 ) + √(𝜎11 − 𝜎22 ) + 4𝜎12 , (22.142.6)
2 2

To correct for the systematic overestimation of the off-axis strength by MAT_


HONEYCOMB, MAT_TRANSVERSELY_ISOTROPIC_CRUSHABLE_FOAM has been
implemented in LS-DYNA. It uses a single yield surface, calculated dynamically from the
six yield stresses specified by the user. The yield surface hardens and softens as a function
of the volumetric strain through the yield stress functions. While the cost of the model is
higher than for MAT_HONEYCOMB, its superior response off-axis makes it the model of
choice for critical applications involving many types of low-density foams.

LS-DYNA DEV 10/27/16 (r:8004) 20-235 (Material Models)


Material Models LS-DYNA Theory Manual

22.143 Material Model 143: Wood Model


The wood model is a transversely isotropic material and is available for solid
elements. The development of this model was done by Murray [2002], who provided the
documentation that follows, under a contract from the FHWA.

The general constitutive relation for an orthotropic material, written in terms of the
principal material directions [Bodig & Jayne, 1993] is:
𝜎1 𝐶 𝐶12 𝐶13 0 0 0 𝜀1
⎡𝜎2 ⎤ ⎡𝐶11 0 ⎤
⎢𝜎 ⎥ ⎢ 21 𝐶22 𝐶23 0 0 ⎥ ⎢ 𝜀2 ⎤

⎢ 3⎥ = ⎢
⎢𝐶31 𝐶32 𝐶33 0 0 0 ⎥ ⎥=⎢ 𝜀3 ⎥
⎥.
⎢𝜎4 ⎥ (22.143.1)
⎢ ⎥ ⎢ 0 0 0 2𝐶44 0 0 ⎥ ⎢ 𝜀 ⎥
⎥ ⎢
4⎥

⎢𝜎5 ⎥ ⎢ 0 0 0 0 2𝐶55 0 ⎥ ⎢ 𝜀5 ⎥
⎣𝜎6 ⎦ ⎣ 0 0 0 0 0 2𝐶66 ⎦ ⎣𝜀6 ⎦
The subscripts 1, 2, and 3 refer to the longitudinal, tangential, and radial, stresses and
strains (1 = 11 , 2 = 22 , 3 = 33 , 1 = 11 , 2 = 22 , 3 = 33 ), respectively. The
subscripts 4, 5, and 6 are in a shorthand notation that refers to the shearing stresses and
strains 4 = 12 , 5 = 13 , 6 = 23 , 4 = 12 , 5 = 13 , 6 = 23 ). As an alternative notation
for wood, it is common to substitute L (longitudinal) for 1, T (tangential) for 2, and R
(radial) for 3. The components of the constitutive matrix, 𝐶𝑖𝑗 , are listed here in terms of the
nine independent elastic constants of an orthotropic material:
E11 (1 − ν23 ν32 )
C11 = ,
Δ
E (1 − ν31 ν13 )
C22 = 22 ,
Δ
E (1 − ν12 ν21 )
C33 = 33 ,
Δ
(ν + ν31 ν23 )E11
C12 = 21 ,
Δ
(ν + ν21 ν32 )E11 (22.143.2)
C13 = 31 ,
Δ
(ν + ν12 ν31 )E22
C23 = 32 ,
Δ
C44 = G12 ,
C55 = G13 ,
C66 = G23 ,
Δ = 1 − ν12 ν21 − ν23 ν32 − ν31 ν13 − 2ν21 ν32 ν13 .

20-236 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

The following identity, relating the dependent (minor Poisson’s ratios ν21 , ν31 , and
ν32 ) and independent elastic constants, is obtained from symmetry considerations of the
constitutive matrix:
𝜈𝑖𝑗 𝜈𝑗𝑖
= for 𝑖, 𝑗 = 1, 2, 3. (22.143.3)
𝐸𝑖 𝐸𝑗

One common assumption is that wood materials are transversely isotropic. This
means that the properties in the tangential and radial directions are modeled the same, i.e.
𝐸22 = 𝐸33 , 𝐺12 = 𝐺13 , and 12 = 13 . This reduces the number of independent elastic
constants to five, 𝐸11 , 𝐸22 , 12 , 𝐺12 , and 𝐺23 . Further, Poisson's ratio in the isotropic plane,
23 , is not an independent quantity. It is calculated from the isotropic relation:  = (𝐸 −
2𝐺)/2𝐺 where E = 𝐸22 = 𝐸33 and 𝐺 = 𝐺23 . Transverse isotropy is a reasonable assumption
because the difference between the tangential and radial properties of wood (particularly
Southern yellow pine and Douglas fir) is small in comparison with the difference between
the tangential and longitudinal properties.

The yield surfaces parallel and perpendicular to the grain are formulated from six
ultimate strength measurements obtained from uniaxial and pure-shear tests on wood
specimens:
XT Tensile strength parallel to the grain
XC Compressive strength parallel to the grain
YT Tensile strength perpendicular to the grain
YC Compressive strength perpendicular to the grain
S|| Shear strength parallel to the grain
S Shear strength perpendicular to the grain

Here 𝑋 and 𝑌 are the strengths parallel and perpendicular to the grain, respectively,
and S are the shear strengths. The formulation is based on the work of Hashin [1980].

For the parallel modes, the yield criterion is composed of two terms involving two of
the five stress invariants of a transversely isotropic material. These invariants are 𝐼1 = 𝜎11
2 2
and 𝐼4 = 𝜎12 + 𝜎13 This criterion predicts that the normal and shear stresses are mutually
weakening, i.e. the presence of shear stress reduces the strength below that measured in
uniaxial stress tests. Yielding occurs when 𝑓|| ≥ 0, where:
2 2 2
𝜎11 (𝜎12 + 𝜎13 ) 𝑋𝑡 for 𝜎11 > 0
𝑓|| = + −1 𝑋={ . (22.143.4)
𝑋2 𝑆2|| 𝑋𝑐 for 𝜎11 < 0

For the perpendicular modes, the yield criterion is also composed of two terms
involving two of the five stress invariants of a transversely isotropic material. These
invariants are 𝐼2 = 22 + 33 and 𝐼3 = 𝜎23
2
− 𝜎22 𝜎33 . Yielding occurs when 𝑓  0, where:

LS-DYNA DEV 10/27/16 (r:8004) 20-237 (Material Models)


Material Models LS-DYNA Theory Manual

Figure 22.143.1. The yield criteria for wood produces smooth surfaces in stress
space.

2
(𝜎22 + 𝜎33 )2 (𝜎23 − 𝜎22 𝜎33 ) 𝑌t for 𝜎22 + 𝜎33 > 0
𝑓⊥ = + −1 𝑌={ (22.143.5)
𝑌 2 𝑆2⊥ 𝑌c for 𝜎22 + 𝜎33 < 0
Each yield criterion is plotted in 3D in Figure 22.143.1 in terms of the parallel and
perpendicular stresses. Each criterion is a smooth surface (no corners).

The plasticity algorithms limit the stress components once the yield criteria in
[Murry 2002] are satisfied. This is done by returning the trial elastic stress state back to the
yield surface. The stress and strain tensors are partitioned into elastic and plastic parts.
Partitioning is done with a return mapping algorithm which enforces the plastic
consistency condition.

Separate plasticity algorithms are formulated for the parallel and perpendicular
modes by enforcing separate consistency conditions. The solution of each consistency
condition determines the consistency parameters,  and  . The  solutions, in turn,
determine the stress updates. No input parameters are required.

The stresses are readily updated from the total strain increments and the consistency
parameters, as follows:
𝜕𝑓
̅𝜎̅̅̅𝑖𝑗𝑛+1 = 𝜎𝑖𝑗∗𝑛+1 − 𝐶𝑖𝑗𝑘𝑙 Δ𝜆 ∣ (22.143.6)
𝜕𝜎𝑘𝑙 𝑛
20-238 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Material Models

𝜎𝑖𝑗∗𝑛+1 = 𝜎𝑖𝑗𝑛 + 𝐶𝑖𝑗𝑘𝑙 Δ𝜀𝑘𝑙 (22.143.7)


Here 𝑛 denotes the nth time step in the finite element analysis, and 𝜎𝑖𝑗∗ are the trial elastic
stress updates calculated from the total strain increments, Δ𝜀𝑖𝑗 , prior to application of
plasticity. Each normal stress update depends on the consistency parameters and yield
surface functions for both the parallel ( = || and 𝑓 = 𝑓|| ) and perpendicular ( =
 and 𝑓 = 𝑓 ) modes. Each shear stress update depends on just one consistency
parameter and yield surface function. If neither parallel nor perpendicular yielding occurs
(𝑓||∗ < 0 and 𝑓⊥∗ < 0) then  = 0 and the stress update is trivial: 𝜎̂ 𝑖𝑗𝑛+1 = 𝜎𝑖𝑗∗𝑛+1 .

Wood exhibits pre-peak nonlinearity in compression parallel and perpendicular to


the grain. Separate translating yield surface formulations are modeled for the parallel and
perpendicular modes, which simulate gradual changes in moduli. Each initial yield surface
hardens until it coincides with the ultimate yield surface. The initial location of the yield
surface determines the onset of plasticity. The rate of translation determines the extent of
the nonlinearity.

For each mode (parallel and perpendicular), the user inputs two parameters: the
initial yield surface location in uniaxial compression, 𝑁, and the rate of translation, 𝑐. Say
the user wants pre-peak nonlinearity to initiate at 70% of the peak strength. The user will
input 𝑁 = 0.3 so that 1 − 𝑁 = 0.7. If the user wants to harden rapidly, then a large value of
𝑐 is input, like 𝑐 = 1 msec. If the user wants to harden gradually, then a small value of 𝑐 is
input, like 𝑐 = 0.2 msec.

The state variable that defines the translation of the yield surface is known as the
back stress, and is denoted by 𝑖𝑗 . Hardening is modeled in compression, but not shear, so
the only back stress required for the parallel modes is 11 . The value of the back stress is
11 = 0 upon initial yield and 11 = −𝑁|| 𝑋𝑐 at ultimate yield (in uniaxial compression).
The maximum back stress occurs at ultimate yield and is equal to the total translation of the
longitudinal stress

XT
Tension Ultimate Yield Strength
Xc
CII large
|longitudinal stress|

CII small
Increasing Rate of Translation NIIXc
(1-NII)Xc
Initial Yield Surface Initial Yield Strength
-(1-NII)Xc
-xc
Ultimate Yield Surace

Compression
SII
Strain
Square root of parallel shear invariant

(a) Initial and ultimate yield surfaces (b) Stress-strain behavior


Figure 22.143.2. Pre-peak nonlinearity in compression is modeled with
translating yield surfaces that allow the user to specify the hardening response.
LS-DYNA DEV 10/27/16 (r:8004) 20-239 (Material Models)
Material Models LS-DYNA Theory Manual

yield surface in stress space. The back stress components required for the perpendicular
modes are 22 and 33 . The value of the backstress sum is 22 + 33 = 0 upon initial yield
and 22 + 33 = −𝑁 𝑌𝑐 at ultimate yield (biaxial compression without shear).

Separate damage formulations are modeled for the parallel and perpendicular
modes. These formulations are loosely based on the work of Simo and Ju [1987]. If failure
occurs in the parallel modes, then all six stress components are degraded uniformly. This
is because parallel failure is catastrophic, and will render the wood useless. If failure
occurs in the perpendicular modes, then only the perpendicular stress components are
degraded. This is because perpendicular failure is not catastrophic: we expect the wood to
continue to carry load in the parallel direction. Based on these assumptions, the following
degradation model is implemented:
𝑑m = max(𝑑(𝜏|| ), 𝑑(𝜏⊥ )) ,
𝑑|| = 𝑑(𝜏|| ),
𝜎11 = (1 − 𝑑|| )𝜎
̅̅̅̅11 ,
𝜎22 = (1 − 𝑑m )𝜎
̅̅̅̅22 ,
(22.143.8)
𝜎33 = (1 − 𝑑m )𝜎
̅̅̅̅33 ,
σ12 = (1 − 𝑑|| )𝜎
̅̅̅̅12 ,
𝜎13 = (1 − 𝑑|| )𝜎
̅̅̅̅13 ,
𝜎23 = (1 − 𝑑m )𝜎
̅̅̅̅23 .
Here, each scalar damage parameter, 𝑑, transforms the stress tensor associated with the
undamaged state, 𝜎 ̅̅̅̅𝑖𝑗 , into the stress tensor associated with the damaged state, 𝑖𝑗 . The
stress tensor 𝜎
̅̅̅̅𝑖𝑗 is calculated by the plasticity algorithm prior to application of the damage
model. Each damage parameter ranges from zero for no damage and approaches unity for
maximum damage. Thus 1 − 𝑑 is a reduction factor associated with the amount of damage.
Each damage parameter evolves as a function of a strain energy-type term. Mesh size
dependency is regulated via a length scale based on the element size (cube root of volume).
Damage-based softening is brittle in tension, less brittle in shear, and ductile (no softening)
in compression, as demonstrated in Figure 22.143.1.

Element erosion occurs when an element fails in the parallel mode, and the parallel
damage parameter exceeds 𝑑|| = 0.99. Elements do not automatically erode when an
element fails in the perpendicular mode. A flag is available, which, when set, allows
elements to erode when the perpendicular damage parameter exceeds 𝑑 = 0.99. Setting
this flag is not recommended unless excessive perpendicular damage is causing
computational difficulties.

20-240 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

(a) Tensile softening.

(b) Shear softening.

(c) Compressive yielding.

Figure 19.143.3. Softening response modeled for parallel modes of Southern yellow
pine.

Data available in the literature for pine [Reid & Peng, 1997] indicates that dynamic
strength enhancement is more pronounced in the perpendicular direction than in the
parallel direction. Therefore, separate rate effects formulations are modeled for the parallel

LS-DYNA DEV 10/27/16 (r:8004) 20-241 (Material Models)


Material Models LS-DYNA Theory Manual

and perpendicular modes. The formulations increase strength with increasing strain rate
by expanding each yield surface:
𝜎11 = 𝑋 + 𝐸11 𝜀̇ 𝜂|| Parallel
. (22.143.9)
𝜎22 = 𝑌 + 𝐸22 𝜀̇ 𝜂⊥ Perpendicular
Here 𝑋 and 𝑌 are the static strengths, 11 and 22 are the dynamic strengths, and 𝐸11 𝜀̇ 𝜂||
and 𝐸22 𝜀̇ 𝜂⊥ are the excess stress components. The excess stress components depend on the
value of the fluidity parameter, , as well as the stiffness and strain rate. The user inputs
two values, 0 and 𝑛, to define each fluidity parameter:
𝜂0||
𝜂|| = 𝑛|| ,
𝜀̇
(22.143.10)
𝜂0⊥
𝜂 ⊥ = 𝑛⊥ .
𝜀̇

The two parameter formulation [Murray, 1997] allows the user to model a nonlinear
variation in dynamic strength with strain rate. Setting 𝑛 = 0 allows the user to model a
linear variation in dynamic strength with strain rate.

20-242 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.144 Material Model 144: Pitzer Crushable Foam


The logarithmic volumetric strain is defined in terms of the relative volume, 𝑉, as:
𝛾 = −ln(𝑉). (22.144.1)
In defining the curves the stress and strain pairs should be positive values starting with a
volumetric strain value of zero.

LS-DYNA DEV 10/27/16 (r:8004) 20-243 (Material Models)


Material Models LS-DYNA Theory Manual

22.147 Material Model 147: FHWA Soil Model


A brief discussion of the FHWA soil model is given. The elastic properties of the soil
are isotropic. The implementation of the modified Mohr-Coulomb plasticity surface is
based on the work of Abbo and Sloan [1995]. The model is extended to include excess pore
water effects, strain softening, kinematic hardening, strain rate effects, and element
deletion.

The modified yield surface is a hyperbola fitted to the Mohr-Coulomb surface. At


the crossing of the pressure axis (zero shear strength) the modified surface is a smooth
surface and it is perpendicular to the pressure axis. The yield surface is given as

𝐹 = −𝑃sin𝜙 + √𝐽2 𝐾(𝜃)2 + AHYP2 sin2 𝜙 − 𝑐cos𝜙 = 0, (22.147.1)

where 𝑃 is the pressure, 𝜙 is the internal friction angle, 𝐾(𝜃) is a function of the angle in

𝐹 = −𝑃sin𝜑 + √𝐽2 𝐾(𝜃)2 + AHYP2 sin2 𝜑 − 𝑐cos𝜑 = 0, (22.147.2)

deviatoric plane, √𝐽2 is the square root of the second invariant of the stress deviator, 𝑐 is the
amount of cohesion and

3√3𝐽3
cos3𝜃 = 3
, (22.147.3)
2𝐽22
J3 is the third invariant of the stress deviator, AHYP is a parameter for determining how
close to the standard Mohr-Coulomb yield surface the modified surface is fitted. If the user
defined parameter, AHYP, is input as zero, the standard Mohr-Coulomb surface is
recovered. The parameter aℎypshould be set close to zero, based on numerical
considerations, but always less than 𝑐 cot𝜙. It is best not to set the cohesion, 𝑐, to very small
values as this causes excessive iterations in the plasticity routines.

To generalize the shape in the deviatoric plane, we have changed the standard
Mohr- Coulomb 𝐾(𝜃) function to a function used by Klisinski [1985]
4(1 − 𝑒2 )cos2 𝜃 + (2𝑒 − 1)2
𝐾(𝜃) = 1
, (22.147.4)
2(1 − 𝑒2 )cos𝜃 + (2𝑒 − 1)[4(1 − 𝑒2 )cos2 𝜃 + 5𝑒2 − 4𝑒]2
where 𝑒 is a material parameter describing the ratio of triaxial extension strength to triaxial
compression strength. If e is set equal to 1, then a circular cone surface is formed. If 𝑒 is set
to 0.55, then a triangular surface is found, 𝐾(𝜃) is defined for 0.5 < 𝑒 ≤ 1.0.

To simulate non-linear strain hardening behavior the friction, angle 𝜙 is increased as


a function of the effective plastic strain,

20-244 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

Figure 22.147.1. Pressure versus volumetric strain showing the effects of D1


parameter.

𝜑 − 𝜑init
Δ𝜑 = 𝐸t (1 − ) Δ𝜀eff plas . (22.147.5)
𝐴𝑛 𝜑max
where 𝜀eff plas is the effective plastic strain. 𝐴𝑛 is the fraction of the peak strength internal
friction angle where nonlinear behavior begins, 0 < 𝐴𝑛 ≤ 1. The input parameter 𝐸𝑡
determines the rate of the nonlinear hardening.

To simulate the effects of moisture and air voids including excess pore water
pressure, both the elastic and plastic behaviors can be modified. The bulk modulus is
𝐾𝑖
𝐾= . (22.147.6)
1 + 𝐾𝑖 𝐷1 𝑛cur
where
𝐾𝑖 = initial bulk modulus
𝑛cur = current porosity = Max[0, (𝑤 − 𝜀v )]
𝑤 = volumetric strain corresponding to the volume of air voids = 𝑛(1 − 𝑆)
𝜀v = total volumetric strain
𝐷1 = material constant controlling the stiffness before the air voids are collapsed
𝑒
𝑛 = porosity of the soil =
1+𝑒
γsp (1 + mc )
𝑒 = void ratio = −1
ρ

LS-DYNA DEV 10/27/16 (r:8004) 20-245 (Material Models)


Material Models LS-DYNA Theory Manual

u
p

Figure 22.147.2. The effect on pressure due to pore water pressure.

𝜌𝑚c
𝑆 = degree of saturation =
𝑛(1 + 𝑚c )
and 𝜌, 𝛾, 𝑚c are the soil density, specific gravity, and moisture content, respectively.

Figure 22.147.1 shows the effect of the 𝐷1 parameter on the pressure-volumetric


strain relationship (bulk modulus). The bulk modulus will always be a monotonically
increasing value, i.e.,
⎧ 𝐾𝑖
{ if 𝜀𝑣 𝑗+1 > 𝜀𝑣𝑗
𝐾𝑗+1 = ⎨1 + 𝐾𝑖 𝐷1 𝑛cur . (22.147.7)
{𝐾 if 𝜀𝑣 𝑗+1 ≤ 𝜀𝑣𝑗
⎩ 𝑗

Note that the model is following the standard practice of assuming compressive
stresses and strains are positive. If the input parameter 𝐷1 is zero, then the standard linear
elastic bulk modulus behavior is used.

To simulate the loss of shear strength due to excess pore water effects, the model
uses a standard soil mechanics technique [Holtz and Kovacs, 1981] of reducing the total
pressure, 𝑃, by the excess pore water pressure, 𝑢, to get an “ effective pressure”, 𝑃′;
therefore,
𝑃′ = 𝑃 − 𝑢. (22.147.8)

Figure 22.147.2 shows pore water pressure will affect the algorithm for the plasticity
surface. The excess pore water pressure reduces the total pressure, which will lower the
shear strength, √𝐽2 . A large excess pore water pressure can cause the effective pressure to
become zero.

20-246 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

To calculate the pore water pressure, 𝑢, the model uses an equation similar to the
equation used for the moisture effects on the bulk modulus.
𝐾sk
𝑢= 𝜀 , (22.147.9)
1 + 𝐾sk 𝐷2 𝑛cur 𝑣
where
𝐾sk = bulk modulus for soil without air voids (skeletal bulk modulus)
𝑛cur = current porosity = Max[0, (𝑤 − 𝜀𝑣 )]
𝑤 = volumetric strain corresponding to the volume of air voids = 𝑛(1 − 𝑆)
𝜀v = total volumetric strain
𝐷2 = material constant controlling the pore water pressure before
the air voids are collapsed to 𝐷2 ≥ 0
𝑒
𝑛 = porosity of the soil =
1+𝑒
γsp (1 + mc )
𝑒 = void ratio = −1
ρ
𝜌𝑚c
𝑆 = degree of saturation =
𝑛(1 + 𝑚c )
and 𝜌, 𝛾, 𝑚c are the soil density, specific gravity, and moisture content, respectively. The
pore water pressure will not be allowed to become negative, 𝑢 ≥ 0.

Figure 22.147.3 is a plot of the pore pressure versus volumetric strain for different
parameter values. With the 𝐷2 parameter set relatively high compared to 𝐾sk there is no
pore pressure until the volumetric strain is greater than the strains associated with the air
voids. However, as 𝐷2 is lowered, the pore pressure starts to increase before the air voids
are totally collapsed. The 𝐾sk parameter affects the slope of the post-void collapse pressure
- volumetric behavior.

LS-DYNA DEV 10/27/16 (r:8004) 20-247 (Material Models)


Material Models LS-DYNA Theory Manual

The parameter 𝐷2 can be found from Skempton pore water pressure parameter 𝐵,
where 𝐵 is defined as [Holtz and Kovacs, 1981]:
1
𝐵= ,
𝐾
1 + 𝑛 sk
𝐾v (22.147.10)
1−𝐵
𝐷2 = .
𝐵𝐾sk [𝑛(1 − 𝑆)]

To simulate strain softening behavior the FHWA soil model uses a continuum
damage algorithm. The strain-based damage algorithm is based on the work of J. W. Ju
and J. C. Simo [1987, 1989]. They proposed a strain based damage criterion, which is
uncoupled from the plasticity algorithm.

For the damage criterion,


1
𝜉 =− ̅̅̅̅𝑑𝜀pv ,
∫𝑃 (22.147.11)
𝐾𝑖
̅̅̅̅ is the pressure and 𝜀pv is the plastic volumetric strain, the damaged stress is found
where 𝑃
from the undamaged stresses.
𝜎 = (1 − 𝑑)𝜎
̅̅̅̅, (22.147.12)
where 𝑑 is the isotropic damage parameter. The damage parameter is found at step 𝑗 + 1

Figure 22.147.3. The effects of 𝐷2 and 𝐾sk parameters on pore water pressure.

20-248 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

as:
𝑑𝑗+1 = 𝑑𝑗 if 𝜉𝑗+1 ≤ 𝑟𝑗
𝜉𝑗+1 − 𝜉0 , (22.147.13)
𝑑𝑗+1 = if 𝜉𝑗+1 > 𝑟𝑗
𝛼 − 𝜉0
where 𝑟t is a damage threshold surface, 𝑟𝑗+1 = max{𝑟𝑗 , 𝜉𝑗+1 ), and 𝜉0 = 𝑟0 (DINT). The mesh
sensitivity parameter, 𝛼, will be described below.

Typically, the damage, 𝑑, varies from 0 to a maximum of 1. However, some soils can
have a residual strength that is pressure dependent. The residual strength is represented
by 𝜙res , the minimum internal friction angle.

The maximum damage allowed is related to the internal friction angle of residual
strength by:
sin𝜙 − sin𝜙res
𝑑max = , (22.147.14)
sin𝜙
If 𝜙res > 0, then 𝑑max , the maximum damage, will not reach 1, and the soil will have some
residual strength.

When material models include strain softening, special techniques must be used to
prevent mesh sensitivity. Mesh sensitivity is the tendency of the finite element
model/analysis to produce significantly different results as the element size is reduced.
The mesh sensitivity occurs because the softening in the model concentrates in one
element. As the element size is reduced the failure becomes localized in smaller volumes,
which causes less energy to be dissipated by the softening leading to instabilities or at least
mesh sensitive behavior.

To eliminate or reduce the effects of strain softening mesh sensitivity, the softening
parameter, α (the strain at full damage), must be modified as the element size changes.
The FHWA soil model uses an input parameter, “void formation”, 𝐺f , that is like fracture
energy material property for metals. The void formation parameter is the area under the
softening region of the pressure volumetric strain curve times the cube root of the element
1
volume, 𝑉 3 .
1
1 𝛼 𝑃peak (𝛼 − 𝜉0 )𝑉 3 (22.147.15)
𝐺f = 𝑉 3 ∫ 𝑃 𝑑𝜀v = ,
𝜉0 2
with 𝜉0 , the volumetric strain at peak pressure (strain at initial damage, DINT). Then 𝛼 can
be found as a function of the volume of the element 𝑉:
2𝐺f
𝛼= 1⁄
+ 𝜉0 . (22.147.16)
𝐾𝜉0 𝑉 3

LS-DYNA DEV 10/27/16 (r:8004) 20-249 (Material Models)


Material Models LS-DYNA Theory Manual

1⁄
If 𝐺f is made very small relative to 𝐾𝜉0 𝑉 3,
then the softening behavior will be
brittle.
Strain-rate enhanced strength is simulated by a two-parameter Devaut-Lions viscoplastic
update algorithm, developed by Murray [1997]. This algorithm interpolates between the
elastic trial stress (beyond the plasticity surface) and the inviscid stress. The inviscid
stresses (𝜎
̅̅̅̅) are on the plasticity surface.
𝜎
̅̅̅̅vp = (1 − 𝜍)𝜎
̅̅̅̅ + 𝜍𝜎
̅̅̅̅trial , (22.147.17)
𝜂 𝛾
where 𝜍 = Δ𝑡+𝜂, and 𝜂 = ( 𝜀̇r)(𝑣𝑛−1)/𝑣𝑛 .

As 𝜁 approaches 1, then the viscoplastic stress becomes the elastic trial stress.
Setting the input value 𝛾r = 0 eliminates any strain-rate enhanced strength effects.

The model allows element deletion, if needed. As the strain softening (damage)
increases, the effective stiffness of the element can get very small, causing severe element
distortion and hourglassing. The element can be “deleted” to remedy this behavior. There
are two input parameters that affect the point of element deletion. DAMLEV is the damage
threshold where element deletion will be considered. EPSPRMAX is the maximum
principal strain where element will be deleted. Therefore,
𝑑 ≥ DAMLEV and 𝜀prmax > EPSPRMAX, (22.147.18)
for element deletion to occur. If DAMLEV is set to zero, there is no element deletion. Care
must be taken when employing element deletion to assure that the internal forces are very
small (element stiffness is zero) or significant errors can be introduced into the analysis.

The keyword option, NEBRASKA, gives the soil parameters used to validate the
material model with experiments performed at University of Nebraska at Lincoln. The
units for this default inputs are milliseconds, kilograms, and millimeters. There are no
required input parameters except material id (MID). If different units are desired the unit
conversion factors that need to multiply the default parameters can be input.

20-250 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.154 Material Model 154: Deshpande-Fleck Foam

𝛷 = 𝜎̂ − 𝜎Y , (22.154.1)
The equivalent stress, 𝜎̂ , is given by:
2
𝜎VM 2
+ 𝛼2 𝜎m
𝜎̂ 2 = , (22.154.2)
1 + (𝛼/3)2
where, 𝜎VM , is the von Mises effective stress,

3
𝜎VM = √ 𝛔dev : 𝛔dev , (22.154.3)
2
and, 𝜎m and 𝛔dev , is the mean and deviatoric stress
𝜎m = tr(𝛔)
(22.154.4)
𝛔dev = 𝛔 − σm 𝐈.

The yield stress 𝜎Y can be expressed as


𝜀̂ 1
𝜎Y = 𝜎p + 𝛾 + 𝛼2 ( ), (22.154.5)
𝜀D 1 − (𝜀̂/𝜀D )𝛽
Here, 𝜎p , 𝛼2 , 𝛾 and 𝛽 are material parameters. The densification strain, 𝜀D , is defined as
𝜌f
𝜀D = −ln ( ), (22.154.6)
𝜌f0
where 𝜌f is the foam density and 𝜌f0 is the density of the virgin material.

LS-DYNA DEV 10/27/16 (r:8004) 20-251 (Material Models)


Material Models LS-DYNA Theory Manual

22.156 Material Model 156: Muscle


The material behavior of the muscle model is adapted from
*MAT_SPRING_MUSCLE, the spring muscle model and treated here as a standard
material. The initial length of muscle is calculated automatically. The force, relative length
and shortening velocity are replaced by stress, strain and strain rate. A new parallel
damping element is added.

The strain and normalized strain rate are defined respectively as


𝑙
𝜀= −1=𝐿−1
𝑙𝑜
(22.156.1)
𝑙̇ 𝑉M 𝑉M 𝑉M
𝜀̇ = = = = = 𝑉,
𝑙o 𝜀̇max 𝑙𝑜 ∗ (SRM ∗ SFR) (𝑙𝑜 ∗ SRM) ∗ SFR 𝑉max ∗ SFR
where 𝑙𝑜 is the original muscle length.

From the relation above, it is known:


𝑙0
𝑙𝑜 = , (22.156.2)
1 + 𝜀0
where 𝜀0 = SNO; 𝑙0 = muscle length at time 0. Stress of Contractile Element is:
𝜎1 = 𝜎max 𝑎(𝑡)𝑓 (𝜀) 𝑔(𝜀̇), (22.156.3)
where 𝜎max =PIS; 𝑎(𝑡) =ALM; 𝑓 (𝜀) =SVS; 𝑔(𝜀̇) =SVR. Stress of Passive Element is:
𝜎2 = 𝜎max ℎ(ε). (22.156.4)
For exponential relationship:

{
{ 0 𝜀≤0
{
{
{ 1 𝑐𝜀
ℎ(ε) = ⎨ [ exp ( )−1] 𝜀>0 𝑐≠0 (22.156.5)
{ exp(𝑐) − 1 𝐿max
{
{
{
{ 𝜀⁄𝐿max 𝜀>0 𝑐=0

where 𝐿max = 1 + SSM; and 𝑐 =CER. Stress of Damping Element is:
𝜎3 = 𝐷𝜀̇max 𝜀̇ (22.156.6)

Total Stress is:


𝜎 = 𝜎1 + 𝜎2 + 𝜎3 . (22.156.7)

20-252 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.158 Material Model 158: Rate Sensitive Composite Fabric


See material type 58, Laminated Composite Fabric, for the treatment of the
composite material.

Rate effects are taken into account through a Maxwell model using linear
viscoelasticity by a convolution integral of the form:
𝑡 𝜕𝜀𝑘𝑙
𝜎𝑖𝑗 = ∫ 𝑔𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) 𝑑𝜏 , (22.158.1)
0 𝜕𝜏
where 𝑔𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) is the relaxation function for different stress measures. This stress is
added to the stress tensor determined from the strain energy functional. Since we wish to
include only simple rate effects, the relaxation function is represented by six terms from the
Prony series:
𝑁
𝑔(𝑡) = ∑ 𝐺𝑚 𝑒−𝛽𝑚 𝑡 . (22.158.2)
𝑚=1

We characterize this in the input by the shear moduli, 𝐺𝑖 , and the decay constants,
𝛽𝑖 . An arbitrary number of terms, not exceeding 6, may be used when applying the
viscoelastic model. The composite failure is not directly affected by the presence of the
viscous stress tensor.

LS-DYNA DEV 10/27/16 (r:8004) 20-253 (Material Models)


Material Models LS-DYNA Theory Manual

22.159 Material Model 159: Continuous Surface Cap Model


This is a cap model with a smooth intersection between the shear yield surface and
hardening cap, as shown in Figure 22.159.1. The initial damage surface coincides with the
yield surface. Rate effects are modeled with viscoplasticity.

Stress Invariants. The yield surface is formulated in terms of three stress invariants: 𝐽1 is
the first invariant of the stress tensor, 𝐽′2 is the second invariant of the deviatoric stress
tensor, and 𝐽′3 is the third invariant of the deviatoric stress tensor. The invariants are
defined in terms of the deviatoric stress tensor, 𝑆𝑖𝑗 and pressure, 𝑃, as follows:
𝐽1 = 3𝑃,
1
𝐽 ′ 2 = 𝑆𝑖𝑗 𝑆𝑖𝑗 , (22.159.1)
2
1
𝐽 ′ 3 = 𝑆𝑖𝑗 𝑆𝑗𝑘 𝑆𝑘𝑖 .
3

Plasticity Surface. The three invariant yield function is based on these three invariants,
and the cap hardening parameter, 𝑘, as follows:
𝑓 (𝐽1 , 𝐽′2 , 𝐽′3 , 𝜅) = 𝐽′2 − ℜ2 𝐹f2 𝐹c . (22.159.2)
Here 𝐹f is the shear failure surface, 𝐹c is the hardening cap, and  is the Rubin three-
invariant reduction factor. The cap hardening parameter 𝜅 is the value of the pressure
invariant at the intersection of the cap and shear surfaces.

Trial elastic stress invariants are temporarily updated via the trial elastic stress
tensor, 𝜎 T  These are denoted 𝐽1𝑇 , 𝐽2′𝑇 and 𝐽3′𝑇 . Elastic stress states are modeled when

Figure 22.159.1. General Shape of the concrete model yield surface in two-
dimensions.

20-254 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

𝑓 (𝐽1 , 𝐽′2 , 𝐽′3 , 𝜅 𝑇 ) < 0. Elastic-plastic stress states are modeled when 𝑓 (𝐽1 , 𝐽′2 , 𝐽′3 , 𝜅 𝑇 ) > 0. In
this case, the plasticity algorithm returns the stress state to the yield surface such that
𝑓 (𝐽1𝑃 , 𝐽′𝑃2 , 𝐽′𝑃3 , 𝜅 𝑃 ) = 0. This is accomplished by enforcing the plastic consistency condition
with associated flow.

Shear Failure Surface. The strength of concrete is modeled by the shear surface in the
tensile and low confining pressure regimes:
𝐹f (𝐽1 ) = 𝛼 − 𝜆𝑒−𝛽𝐽1 + 𝜃𝐽1 . (22.159.3)
Here the values of 𝛼, 𝛽, 𝜆, and 𝜃 are selected by fitting the model surface to strength
measurements from triaxial compression (txc) tests conducted on plain concrete cylinders.

Rubin Scaling Function. Concrete fails at lower values of √3𝐽2′ (principal stress difference)
for triaxial extension (txe) and torsion (tor) tests than it does for txc tests conducted at the
same pressure. The Rubin scaling function ℜ determines the strength of concrete for any
state of stress relative to the strength for txc, via ℜ𝐹𝑓 . Strength in torsion is modeled as
𝑄1 𝐹f . Strength in txe is modeled as 𝑄2 𝐹f , where:
𝑄1 = 𝛼1 − 𝜆1 𝑒−𝛽1𝐽1 + 𝜃1 𝐽1 ,
(22.159.4)
𝑄2 = 𝛼2 − 𝜆2 𝑒−𝛽2𝐽1 + 𝜃2 𝐽1 .

Cap Hardening Surface. The strength of concrete is modeled by a combination of the cap
and shear surfaces in the low to high confining pressure regimes. The cap is used to model
plastic volume change related to pore collapse (although the pores are not explicitly
modeled). The isotropic hardening cap is a two-part function that is either unity or an
ellipse:
[𝐽1 − 𝐿 (𝜅)][|𝐽1 − 𝐿(𝜅)| + 𝐽1 − 𝐿(𝜅)]
𝐹c ( 𝐽1 , 𝜅 ) = 1 − , (22.159.5)
2[𝑋(𝜅) − 𝐿 (𝜅)] 2
where 𝐿(𝜅) is defined as:
𝜅 if 𝜅 > 𝜅0
L(κ) = { . (22.159.6)
𝜅0 otherwise

The equation for 𝐹c is equal to unity for 𝐽1  𝐿(𝜅). It describes the ellipse for 𝐽1 >
𝐿(𝜅). The intersection of the shear surface and the cap is at 𝐽1 = 𝜅. 𝜅0 is the value of 𝐽1 at
the initial intersection of the cap and shear surfaces before hardening is engaged (before the
cap moves). The equation for 𝐿(𝜅) restrains the cap from retracting past its initial location
at 𝜅0 .

The intersection of the cap with the 𝐽1 axis is at 𝐽1 = 𝑋(𝜅). This intersection depends
upon the cap ellipticity ratio 𝑅, where 𝑅 is the ratio of its major to minor axes:
𝑋(𝜅) = 𝐿(𝜅) + 𝑅𝐹f (𝐿(𝜅)). (22.159.7)

LS-DYNA DEV 10/27/16 (r:8004) 20-255 (Material Models)


Material Models LS-DYNA Theory Manual

The cap moves to simulate plastic volume change. The cap expands (𝑋(𝜅) and𝜅 increase)
to simulate plastic volume compaction. The cap contracts (𝑋(𝜅) and 𝜅 decrease) to
simulate plastic volume expansion, called dilation. The motion (expansion and
contraction) of the cap is based upon the hardening rule:
p 2
𝜀𝑣 = 𝑊(1 − 𝑒−𝐷1(𝑋−𝑋0)−𝐷2(𝑋−𝑋0) ). (22.159.8)
p
Here 𝜀v the plastic volume strain, 𝑊 is the maximum plastic volume strain, and 𝐷1 and 𝐷2
are model input parameters. 𝑋0 is the initial location of the cap when 𝜅 = 𝜅0 .

The five input parameters (𝑋0 , 𝑊, 𝐷1 , 𝐷2 , and 𝑅) are obtained from fits to the
pressure-volumetric strain curves in isotropic compression and uniaxial strain. 𝑋0
determines the pressure at which compaction initiates in isotropic compression. 𝑅
combined with 𝑋0 , determines the pressure at which compaction initiates in uniaxial strain.
𝐷1 and 𝐷2 determine the shape of the pressure-volumetric strain curves. 𝑊 determines the
maximum plastic volume compaction.

Shear Hardening Surface. In unconfined compression, the stress-strain behavior of


concrete exhibits nonlinearity and dilation prior to the peak. Such behavior is be modeled
with an initial shear yield surface, 𝑁H 𝐹f , which hardens until it coincides with the ultimate
shear yield surface, 𝐹f . Two input parameters are required. One parameter, 𝑁H , initiates
hardening by setting the location of the initial yield surface. A second parameter, 𝐶H ,
determines the rate of hardening (amount of nonlinearity).

Damage. Concrete exhibits softening in the tensile and low to moderate compressive
regimes.
vp
𝜎𝑖𝑗𝑑 = (1 − 𝑑)𝜎𝑖𝑗 , (22.159.9)
A scalar damage parameter, 𝑑, transforms the viscoplastic stress tensor without damage,
denoted 𝜎 vp , into the stress tensor with damage, denoted 𝜎 d . Damage accumulation is
based upon two distinct formulations, which we call brittle damage and ductile damage.
The initial damage threshold is coincident with the shear plasticity surface, so the threshold
does not have to be specified by the user.

Ductile Damage. Ductile damage accumulates when the pressure (𝑃) is compressive and
an energy-type term, 𝜏c , exceeds the damage threshold, 𝜏0c . Ductile damage accumulation
depends upon the total strain components, 𝜀𝑖𝑗 , as follows:

1
𝜏c = √ 𝜎𝑖𝑗 𝜀𝑖𝑗 (22.159.10)
2
The stress components 𝜎𝑖𝑗 are the elasto-plastic stresses (with kinematic hardening)
calculated before application of damage and rate effects.

20-256 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

Brittle Damage. Brittle damage accumulates when the pressure is tensile and an energy-
type term, 𝜏t , exceeds the damage threshold, 𝜏0t . Brittle damage accumulation depends
upon the maximum principal strain, 𝜀max , as follows:

𝜏t = √𝐸𝜀2max . (22.159.11)

Softening Function. As damage accumulates, the damage parameter 𝑑 increases from an


initial value of zero, towards a maximum value of one, via the following formulations:
Brittle Damage:
0.999 1+𝐷
𝑑(𝜏1 ) = ( − 1). (22.159.12)
𝐷 1 + 𝐷exp[−𝐶(𝜏𝑡 − 𝜏0𝑡 )]

Ductile Damage:
𝑑max 1+𝐵
𝑑(𝜏1 ) = ( − 1). (22.159.13)
𝐵 1 + 𝐵exp[−𝐴(𝜏c − 𝜏0c )]

The damage parameter that is applied to the six stresses is equal to the current
maximum of the brittle or ductile damage parameter. The parameters 𝐴 and 𝐵 or 𝐶 and 𝐷
set the shape of the softening curve plotted as stress-displacement or stress-strain. The
parameter 𝑑max is the maximum damage level that can be attained. It is internally
calculated and is less than one at moderate confining pressures. The compressive softening
parameter, 𝐴, may also be reduced with confinement, using the input parameter PMOD, as
follows:
𝐴 = 𝐴(𝑑max + 0.001)PMOD . (22.159.14)

Regulating Mesh Size Sensitivity. The concrete model maintains constant fracture
energy, regardless of element size. The fracture energy is defined here as the area under
the stress-displacement curve from peak strength to zero strength. This is done by
internally formulating the softening parameters 𝐴 and 𝐶 in terms of the element length, 𝑙
(cube root of the element volume), the fracture energy, 𝐺f , the initial damage threshold,𝜏0t
or 𝜏0c , and the softening shape parameters, 𝐷 or 𝐵.

The fracture energy is calculated from up to five user-specified input parameters


(𝐺fc , 𝐺ft , 𝐺fs , pwrc, pwrc). The user specifies three distinct fracture energy values. These
are the fracture energy in uniaxial tensile stress, 𝐺ft , pure shear stress, 𝐺fs , and uniaxial
compressive stress, 𝐺fc . The model internally selects the fracture energy from equations
which interpolate between the three fracture energy values as a function of the stress state
(expressed via two stress invariants). The interpolation equations depend upon the user-
specified input powers PWRC and PWRT, as follows.
if the pressure is tensile (22.159.15)

LS-DYNA DEV 10/27/16 (r:8004) 20-257 (Material Models)


Material Models LS-DYNA Theory Manual

PWRT

⎜ −𝐽1 ⎞⎟
𝐺f = 𝐺fs + trans(𝐺ft − 𝐺fs ) where trans = ⎜
⎜ ⎟


⎝√3𝐽 2 ⎠
if the pressure is compressive
PWRC
⎛ 𝐽1 ⎞
𝐺f = 𝐺fs + trans(𝐺fc − 𝐺fs ) where trans = ⎜

⎜ ⎟


⎝ √3𝐽′ 2⎠

The internal parameter trans is limited to range between 0 and 1.

Element Erosion. An element loses all strength and stiffness as 𝑑 → 1. To prevent


computational difficulties with very low stiffness, element erosion is available as a user
option. An element erodes when 𝑑 > 0.99 and the maximum principal strain is greater than
a user supplied input value, 1-erode.

Viscoplastic Rate Effects. At each time step, the viscoplastic algorithm interpolates
p
between the elastic trial stress, 𝜎𝑖𝑗T , and the inviscid stress (without rate effects), 𝜎𝑖𝑗 , to set
vp
the viscoplastic stress (with rate effects), 𝜎𝑖𝑗 :
vp p
𝜎𝑖𝑗 = (1 − 𝛾)𝜎𝑖𝑗T + 𝛾𝜎𝑖𝑗 , (22.159.16)
Δ𝑡/𝜂
with 𝛾 = 1+Δ𝑡/𝜂.

This interpolation depends upon the effective fluidity coefficient, , and the time
step, 𝑡. The effective fluidity coefficient is internally calculated from five user-supplied
input parameters and interpolation equations:
if the pressure is tensile
pwrt
⎛ −𝐽1 ⎞
𝜂 = 𝜂s + trans(𝜂t − 𝜂s ) trans = ⎜

⎜ ⎟


√3𝐽 ′
⎝ 2⎠

if the pressure is compressive (22.159.17)


pwrc

⎜ 𝐽1 ⎞ ⎟
𝜂 = 𝜂s + trans(𝜂c − 𝜂s ) trans = ⎜
⎜ ⎟

√3𝐽 ′
⎝ 2⎠
𝜂0t 𝜂0c
𝜂t = Nt
𝜂c = 𝜂s = SRATE 𝜂t
𝜀̇ 𝜀̇Nc
The input parameters are 𝜂0t and 𝑁t for fitting uniaxial tensile stress data, 𝜂0c and 𝑁c for
fitting the uniaxial compressive stress data, and SRATE for fitting shear stress data. The
effective strain rate is 𝜀̇.

20-258 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

This viscoplastic model may predict substantial rate effects at high strain rates (𝜀̇ >
100). To limit rate effects at high strain rates, the user may input overstress limits in tension
(OVERT) and compression (OVERC). These input parameters limit calculation of the
fluidity parameter, as follows:

If 𝐸𝜀̇𝜂 > OVER, then  = over


𝛦𝜀̇
(22.159.18)

where OVER = OVERT when the pressure is tensile, and OVER = OVERC when the
pressure is compressive.

The user has the option of increasing the fracture energy as a function of effective
strain rate via the REPOW input parameter, as follows:
REPOW
𝐸𝜀̇𝜂
𝐺rate
f = 𝐺f (1 + ) (22.159.19)
𝑓′
Here 𝐺rate
f is the fracture energy enhanced by rate effects, and 𝑓′ is the yield strength before
application of rate effects (which is calculated internally by the model). The term in
brackets is greater than, or equal to one, and is the approximate ratio of the dynamic to
static strength.

LS-DYNA DEV 10/27/16 (r:8004) 20-259 (Material Models)


Material Models LS-DYNA Theory Manual

22.161 Material Models 161 and 162: Composite MSC


The unidirectional and fabric layer failure criteria and the associated property
degradation models for material 161 are described as follows. All the failure criteria are
expressed in terms of stress components based on ply level stresses (𝜎𝑎 , 𝜎𝑏 , 𝜎𝑐 , 𝜏𝑎𝑏 , 𝜏𝑏𝑐 , 𝜏𝑐𝑎 )
and the associated elastic moduli are (𝐸𝑎 , 𝐸𝑏 , 𝐸𝑐 , 𝐺𝑎𝑏 , 𝐺𝑏𝑐 , 𝐺𝑐𝑎 ). Note that for the
unidirectional model, 𝑎, 𝑏 and 𝑐 denote the fiber, in-plane transverse and out-of-plane
directions, respectively, while for the fabric model, 𝑎, 𝑏 and 𝑐 denote the in-plane fill, in-
plane warp and out-of-plane directions, respectively.

Unidirectional Lamina Model


Three criteria are used for fiber failure, one in tension/shear, one in compression
and another one in crush under pressure. They are chosen in terms of quadratic stress
forms as follows:

Tensile/shear fiber mode:


2
〈𝜎𝑎 〉 2 𝜏𝑎𝑏 2
+ 𝜏𝑐𝑎
𝑓1 = ( ) +( ) − 1 = 0. (22.161.1)
𝑆𝑎T 𝑆2FS
Compression fiber mode:
2
〈𝜎𝑎′ 〉 𝜎𝑏 + 𝜎𝑐
𝑓2 = ( ) − 1 = 0, 𝜎𝑎′ = −𝜎𝑎 + ⟨− ⟩. (22.161.2)
𝑆𝑎C 2
Crush mode:
〈𝑝〉 2 𝜎𝑎 + 𝜎𝑏 + 𝜎𝑐
𝑓3 = ( ) − 1 = 0, 𝑝=− . (22.161.3)
𝑆FC 3
where ⟨ ⟩ are Macaulay brackets, 𝑆𝑎T and 𝑆𝑎C are the tensile and compressive strengths in
the fiber direction, and 𝑆FS and 𝑆FC are the layer strengths associated with the fiber shear
and crush failure, respectively.

Matrix mode failures must occur without fiber failure, and hence they will be on
planes parallel to fibers. For simplicity, only two failure planes are considered: one is
perpendicular to the planes of layering and the other one is parallel to them. The matrix
failure criteria for the failure plane perpendicular and parallel to the layering planes,
respectively, have the forms:
Perpendicular matrix mode:
2
⟨𝜎𝑏 ⟩ 2 𝜏 𝜏 2
𝑓4 = ( ) + ( ′𝑏𝑐 ) + ( 𝑎𝑏 ) − 1 = 0. (22.161.4)
𝑆𝑏T 𝑆𝑏𝑐 𝑆𝑎𝑏

20-260 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

Parallel matrix mode (Delamination):


⎧ 2 2
{ ⟨𝜎 ⟩ 𝜏 𝜏 2⎫ }
𝑓5 = 𝑆2 ⎨( 𝑐 ) + ( "𝑏𝑐 ) + ( 𝑐𝑎 ) ⎬ − 1 = 0, (22.161.5)
{ 𝑆
⎩ 𝑏T 𝑆𝑏𝑐 𝑆𝑐𝑎 }

where 𝑆𝑏T is the transverse tensile strength. Based on the Coulomb-Mohr theory, the shear
strengths for the transverse shear failure and the two axial shear failure modes are assumed
to be the forms,
𝑆𝑎𝑏 = 𝑆(0)
𝑎𝑏 + tan(𝜑)⟨−𝜎𝑏 ⟩,

𝑆′𝑏𝑐 = 𝑆(0)
𝑏𝑐 + tan(𝜑)⟨−𝜎𝑏 ⟩,
(22.161.6)
𝑆𝑐𝑎 = 𝑆(0)
𝑐𝑎 + tan(𝜑)⟨−𝜎𝑐 ⟩,

𝑆"𝑏𝑐 = 𝑆(0)
𝑏𝑐 + tan(𝜑)⟨−𝜎𝑐 ⟩,

where 𝜑 is a material constant as tan(𝜑) is similar to the coefficient of friction, and 𝑆(0) (0)
𝑎𝑏 , 𝑆𝑐𝑎
and 𝑆(0)
𝑏𝑐 are the shear strength values of the corresponding tensile modes.

Failure predicted by the criterion of 𝑓4 can be referred to as transverse matrix failure,


while the matrix failure predicted by 𝑓5 , which is parallel to the layer, can be referred as the
delamination mode when it occurs within the elements that are adjacent to the ply
interface. Note that a scale factor 𝑆 is introduced to provide better correlation of
delamination area with experiments. The scale factor 𝑆 can be determined by fitting the
analytical prediction to experimental data for the delamination area.

When fiber failure in tension/shear mode is predicted in a layer by 𝑓1 , the load


carrying capacity of that layer is completely eliminated. All the stress components are
reduced to zero instantaneously (100 time steps to avoid numerical instability). For
compressive fiber failure, the layer is assumed to carry a residual axial load, while the
transverse load carrying capacity is reduced to zero. When the fiber compressive failure
mode is reached due to 𝑓2 , the axial layer compressive strength stress is assumed to reduce
to a residual value SRC (=SFFC ∗ SAC ). The axial stress is then assumed to remain constant,
i.e., 𝜎𝑎 = −𝑆RC , for continuous compressive loading, while the subsequent unloading curve
follows a reduced axial modulus to zero axial stress and strain state. When the fiber crush
failure occurs, the material is assumed to behave elastically for compressive pressure, 𝑝 >
0, and to carry no load for tensile pressure, 𝑝 < 0.

When a matrix failure (delamination) in the a-b plane is predicted, the strength
(0)
values for 𝑆(0)
𝑐𝑎 and 𝑆𝑏𝑐 are set to zero. This results in reducing the stress components 𝜎𝑐 ,𝜏𝑏𝑐
and 𝜏𝑐𝑎 to the fractured material strength surface. For tensile mode, 𝜎𝑐 > 0, these stress
components are reduced to zero. For compressive mode, 𝜎𝑐 < 0, the normal stress 𝜎𝑐 is
assumed to deform elastically for the closed matrix crack. Loading on the failure envelop,
the shear stresses are assumed to ‘slide’ on the fractured strength surface (frictional shear
stresses) like in an ideal plastic material, while the subsequent unloading shear stress-strain

LS-DYNA DEV 10/27/16 (r:8004) 20-261 (Material Models)


Material Models LS-DYNA Theory Manual

path follows reduced shear moduli to the zero shear stress and strain state for both 𝜏𝑏𝑐 and
𝜏𝑐𝑎 components.

The post failure behavior for the matrix crack in the a-c plane due to 𝑓4 is modeled in
the same fashion as that in the a-b plane as described above. In this case, when failure
occurs, 𝑆(0) (0)
𝑎𝑏 and 𝑆𝑏𝑐 are reduced to zero instantaneously. The post fracture response is then
governed by failure criterion of f5 with 𝑆(0) (0)
𝑎𝑏 = 0 and 𝑆𝑏𝑐 = 0. For tensile mode, 𝜎𝑏 , , 𝜏𝑎𝑏 and
𝜏𝑏𝑐 are zero. For compressive mode, 𝜎𝑏 < 0, 𝜎𝑏 is assumed to be elastic, while 𝜏𝑎𝑏 and 𝜏𝑏𝑐
‘slide’ on the fracture strength surface as in an ideal plastic material, and the unloading
path follows reduced shear moduli to the zero shear stress and strain state. It should be
noted that 𝜏𝑏𝑐 is governed by both the failure functions and should lie within or on each of
these two strength surfaces.

Fabric Lamina Model


The fiber failure criteria of Hashin for a unidirectional layer are generalized to
characterize the fiber damage in terms of strain components for a plain weave layer. The
fill and warp fiber tensile/shear failure are given by the quadratic interaction between the
associated axial and shear stresses, i.e.
2 2
⟨𝜎𝑎 ⟩ 2 (𝜏𝑎𝑏 + 𝜏𝑐𝑎 )
𝑓6 = ( ) + − 1 = 0, (22.161.7)
𝑆𝑎T 𝑆2𝑎FS

2 2
⟨𝜎𝑏 ⟩ 2 (𝜏𝑎𝑏 + 𝜏𝑏𝑐 )
𝑓7 = ( ) + 2
− 1 = 0, (22.161.8)
𝑆𝑏T 𝑆𝑏FS
where 𝑆𝑎T and 𝑆𝑏T are the axial tensile strengths in the fill and warp directions,
respectively, and 𝑆𝑎FS and 𝑆𝑏FS are the layer shear strengths due to fiber shear failure in the
fill and warp directions. These failure criteria are applicable when the associated 𝜎𝑎 or 𝜎𝑏 is
positive. It is assumed 𝑆aFS = SFS, and
𝑆𝑏T
𝑆𝑏FS = SFS ∗ . (22.161.9)
𝑆𝑎T

When 𝜎𝑎 or 𝜎𝑏 is compressive, it is assumed that the in-plane compressive failure in


both the fill and warp directions are given by the maximum stress criterion, i.e.
2
⟨𝜎 ′ ⟩
𝑓8 = [ 𝑎 ] − 1 = 0, 𝜎𝑎′ = −𝜎𝑎 + ⟨−𝜎𝑐 ⟩, (22.161.10)
𝑆𝑎C
2
⟨𝜎 ′ ⟩
𝑓9 = [ 𝑏 ] − 1 = 0, 𝜎𝑏′ = −𝜎𝑏 + ⟨−𝜎𝑐 ⟩. (22.161.11)
𝑆𝑏C
where 𝑆𝑎C and 𝑆𝑏C are the axial compressive strengths in the fill and warp directions,
respectively. The crush failure under compressive pressure is

20-262 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

2
⟨𝑝⟩ 𝜎𝑎 + 𝜎𝑏 + 𝜎𝑐
𝑓10 =( ) − 1 = 0, 𝑝=− . (22.161.12)
𝑆FC 3

A plain weave layer can fail under in-plane shear stress without the occurrence of
fiber breakage. This in-plane matrix failure mode is given by
𝜏𝑎𝑏 2
𝑓11 = ( ) − 1 = 0, (22.161.13)
𝑆𝑎𝑏
where 𝑆𝑎𝑏 is the layer shear strength due to matrix shear failure.

Another failure mode, which is due to the quadratic interaction between the
thickness stresses, is expected to be mainly a matrix failure. This through the thickness
matrix failure criterion is

2 ⟨𝜎𝑐 ⟩ 2 𝜏𝑏𝑐 2 𝜏𝑐𝑎 2


𝑓12 = 𝑆 {( ) + ( ) + ( ) } − 1 = 0, (22.161.14)
𝑆𝑐𝑇 𝑆𝑏𝑐 𝑆𝑐𝑎
where 𝑆𝑐T is the through the thickness tensile strength, and 𝑆𝑏𝑐 , and 𝑆𝑐𝑎 are the shear
strengths assumed to depend on the compressive normal stress sc, i.e.,
𝑆 𝑆(0)
{ 𝑐𝑎 } = { 𝑐𝑎 } + tan(𝜑)⟨−𝜎𝑐 ⟩. (22.161.15)
𝑆𝑏𝑐 𝑆(0)
𝑏𝑐

When failure predicted by this criterion occurs within elements that are adjacent to
the ply interface, the failure plane is expected to be parallel to the layering planes, and,
thus, can be referred to as the delamination mode. Note that a scale factor 𝑆 is introduced
to provide better correlation of delamination area with experiments. The scale factor 𝑆 can
be determined by fitting the analytical prediction to experimental data for the delamination
area.

Similar to the unidirectional model, when fiber tensile/shear failure is predicted in a


layer by 𝑓6 or 𝑓7 , the load carrying capacity of that layer in the associated direction is
completely eliminated. For compressive fiber failure due to 𝑓8 or 𝑓9 , the layer is assumed to
carry a residual axial load in the failed direction, while the load carrying capacity
transverse to the failed direction is assumed unchanged. When the compressive axial stress
in a layer reaches the compressive axial strength 𝑆𝑎C or 𝑆𝑏C , the axial layer stress is
assumed to be reduced to the residual strength 𝑆aRC or 𝑆bRC where 𝑆aRC = SFFC ∗ SaC
and SbRC = SFFC ∗ SbC . The axial stress is assumed to remain constant, i.e., 𝜎𝑎 = −𝑆aCR or
𝜎𝑏 = −SbCR , for continuous compressive loading, while the subsequent unloading curve
follows a reduced axial modulus. When the fiber crush failure has occurred, the material is
assumed to behave elastically for compressive pressure, 𝑝 > 0, and to carry no load for
tensile pressure, 𝑝 < 0.

LS-DYNA DEV 10/27/16 (r:8004) 20-263 (Material Models)


Material Models LS-DYNA Theory Manual

When the in-plane matrix shear failure is predicted by f11 the axial load carrying
capacity within a failed element is assumed unchanged, while the in-plane shear stress is
assumed to be reduced to zero.

For through the thickness matrix (delamination) failure given by equations 𝑓12 , the
in-plane load carrying capacity within the element is assumed to be elastic, while the
(0)
strength values for the tensile mode, 𝑆(0)
𝑐𝑎 and 𝑆𝑏𝑐 , are set to zero. For tensile mode, 𝜎𝑐 > 0,
the through the thickness stress components are reduced to zero. For compressive mode,
𝜎𝑐 < 0, 𝜎𝑐 is assumed to be elastic, while 𝜏𝑏𝑐 and 𝜏𝑐𝑎 ‘slide’ on the fracture strength surface
as in an ideal plastic material, and the unloading path follows reduced shear moduli to the
zero shear stress and strain state.

The effect of strain-rate on the layer strength values of the fiber failure modes is
modeled by the strain-rate dependent functions for the strength values {IRT } as
{ε̅}̇
{SRT } = {S0 } ( 1 + Crate1 ln ), (22.161.16)
ε̇0

⎧ ∣𝜀̇𝑎 ∣ ⎫
⎧ 𝑆𝑎T ⎫ { }
{ } { ∣𝜀̇𝑎 ∣ }
{ 𝑆𝑎C } { }
{ } { ∣𝜀̇ ∣ }
{ 𝑆𝑏T } { 𝑏 }
{𝑆RT } = ⎨ ⎬ and {𝜀̅
} ̇ = ⎨ ∣𝜀̇ ∣ ⎬, (22.161.17)
{ 𝑆𝑏C }
𝑏
{ ∣𝜀̇𝑐 ∣ }
{𝑆 } { }
{ FC } { }
{ } { 1}
⎩ 𝑆FS ⎭ { 2 2 }
⎩(𝜀̇𝑐𝑎 + 𝜀̇𝑏𝑐 )2 ⎭
where 𝐶rate is the strain-rate constants, and {𝑆0 }are the strength values of {𝑆RT } at the
reference strain-rate 𝜀̇0 .

Damage Model
The damage model is a generalization of the layer failure model of Material 161 by
adopting the MLT damage mechanics approach, Matzenmiller et al. [1995], for
characterizing the softening behavior after damage initiation. Complete model description
is given in Yen [2001]. The damage functions, which are expressed in terms of ply level
engineering strains, are converted from the above failure criteria of fiber and matrix failure
modes by neglecting the Poisson’s effect. Elastic moduli reduction is expressed in terms of
the associated damage parameters 𝜛𝑖 :
E′𝑖 = (1 − ϖ𝑖 )E𝑖 (22.161.18)
ϖ𝑖 = 1 − exp(−𝑟𝑖𝑚𝑖 /𝑚𝑖 ) 𝑟𝑖 ≥ 0 𝑖 = 1, . . . ,6, (22.161.19)

where 𝐸𝑖 are the initial elastic moduli, 𝐸𝑖 are the reduced elastic moduli, 𝑟𝑖 are the damage
thresholds computed from the associated damage functions for fiber damage, matrix
damage and delamination, and mi are material damage parameters, which are currently
assumed to be independent of strain-rate. The damage function is formulated to account
for the overall nonlinear elastic response of a lamina including the initial ‘hardening’ and
the subsequent softening beyond the ultimate strengths.

20-264 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

In the damage model (Material 162), the effect of strain-rate on the nonlinear stress-
strain response of a composite layer is modeled by the strain-rate dependent functions for
the elastic moduli {𝐸RT } as
{𝜀̅}̇
{𝐸RT } = {𝐸0 } ( 1 + {𝐶rate } ln ), (22.161.20)
𝜀̇0

⎧ 𝐸𝑎 ⎫ ⎧ ∣𝜀̇𝑎 ∣ ⎫ ⎧𝐶rate2 ⎫
{ { } { }
{ 𝐸𝑏 }} { ∣𝜀̇𝑏 ∣ }
{ } {𝐶rate2 }
{ }
{
{ 𝐸𝑐 }} { ∣𝜀̇𝑐 ∣ } {𝐶rate4 }
̇
{𝐸RT } = ⎨ ⎬ , {𝜀̅} = ⎨ and {𝐶rate } = ⎨ ,
{ 𝐺𝑎𝑏 } { ∣𝜀̇𝑎𝑏 ∣⎬
} {𝐶rate3 ⎬
} (22.161.21)
{
{ 𝐺 } { ∣𝜀̇ ∣} {𝐶 }
{ 𝑏𝑐 }} {
{ 𝑏𝑐 }
}
{ rate3 }
{ }
⎩𝐺𝑐𝑎 ⎭ ⎩ ∣𝜀̇𝑐𝑎 ∣⎭ ⎩𝐶rate3 ⎭

where {𝐶rate } are the strain-rate constants. {𝐸0 } are the modulus values of {𝐸RT } at the
reference strain-rate 𝜀̇0 .

LS-DYNA DEV 10/27/16 (r:8004) 20-265 (Material Models)


Material Models LS-DYNA Theory Manual

22.163 Material Model 163: Modified Crushable Foam


The volumetric strain is defined in terms of the relative volume, 𝑉, as: 𝛾 = 1. −𝑉.
The relative volume is defined as the ratio of the current to the initial volume. In place of
the effective plastic strain in the D3PLOT database, the integrated volumetric strain is
output.

This material is an extension of material 63, *MAT_CRUSHABLE_FOAM. It allows


the yield stress to be a function of both volumetric strain rate and volumetric strain. Rate
effects are accounted for by defining a table of curves using *DEFINE_TABLE. Each curve
defines the yield stress versus volumetric strain for a different strain rate. The yield stress
is obtained by interpolating between the two curves that bound the strain rate.

To prevent high frequency oscillations in the strain rate from causing similar high
frequency oscillations in the yield stress, a modified volumetric strain rate is used when
interpolating to obtain the yield stress. The modified strain rate is obtained as follows. If
NCYCLE is > 1, then the modified strain rate is obtained by a time average of the actual
strain rate over NCYCLE solution cycles. For SRCLMT > 0, the modified strain rate is
capped so that during each cycle, the modified strain rate is not permitted to change more
than SRCLMT multiplied by the solution time step.

1-V
Figure 22.163.1. Rate effects are defined by a family of curves giving yield stress
versus volumetric strain.

20-266 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.164 Material Model 164: Brain Linear Viscoelastic


The shear relaxation behavior is described by the Maxwell model as:
𝐺(𝑡) = 𝐺 + (𝐺0 − 𝐺∞ )𝑒−𝛽𝑡 . (22.164.1)

A Jaumann rate formulation is used


𝛻 𝑡
𝜎′𝑖𝑗 = 2 ∫ 𝐺(𝑡 − 𝜏) 𝐷′𝑖𝑗 (𝜏)𝑑𝑡. (22.164.2)
0
𝛻
where the prime denotes the deviatoric part of the stress rate, 𝜎𝑖𝑗 , and the strain rate 𝐷𝑖𝑗 .
For the Kelvin model the stress evolution equation is defined as:
1 𝐺
𝑠𝑖𝑗̇ + 𝑠𝑖𝑗 = (1 + 𝛿𝑖𝑗 ) (𝐺0 + ∞ ) 𝑒𝑖𝑗̇ . (22.164.3)
𝜏 𝜏
The strain data as written to the LS-DYNA database may be used to predict damage, see
[Bandak 1991].

LS-DYNA DEV 10/27/16 (r:8004) 20-267 (Material Models)


Material Models LS-DYNA Theory Manual

22.166 Material Model 166: Moment Curvature Beam


Curvature rate can be decomposed into elastic part and plastic part:
𝜀̇ 𝜀̇p 𝜀̇p
𝜀̇ = 𝜀̇e + 𝜀̇p ⇒ = + ⇒ 𝜅̇ = 𝜅̇e + 𝜅 p . (22.166.1)
𝑦 𝑦 𝑦
Moment rate is the product of elastic bending stiffness and elastic curvature:

𝑀̇ = ∫ 𝜎̇ 𝑦𝑑𝐴 = ∫ 𝐸e 𝜀̇e 𝑦𝑑𝐴 = ∫ 𝐸e 𝜅̇e 𝑦2 𝑑𝐴 = ∫ 𝐸e (𝜅̇ − 𝜅̇p )𝑦2 𝑑𝐴


𝐴 𝐴 𝐴 𝐴
(22.166.2)
= 𝐸e (𝜅̇ − 𝜅̇p ) ∫ 𝑦2 𝑑𝐴 = (𝐸𝐼)e (𝜅̇ − 𝜅̇p ).
𝐴

Plastic flow rule: 𝜓 = |𝑀| (Isotropic hardening)


𝜕𝜓
𝜅̇p = 𝜆̇ = 𝜆̇ sign(𝑀), 𝜅̅ṗ = √𝜅̇p 𝜅̇p = 𝜆̇ . (22.166.3)
𝜕𝑀

Yield condition:
𝑓 = |𝑀| − 𝑀Y (𝜅̅p ) = 0. (22.166.4)
Loading and unloading conditions:
𝜆̇ ≥ 0, 𝑓 ≤ 0, 𝜆̇ 𝑓 = 0. (22.166.5)
Consistency condition:
∂𝑀Y p
𝑓 ̇ = 0 ⇒ 𝑀̇ sign(𝑀) − 𝜅̅ = 0
∂𝜅̅p
⇒ 𝜆̇ ≡ 𝜅̅ṗ
𝑀sign(𝑀) (𝐸𝐼)e p
= p = p (𝜅̇ − 𝜅̇ ) sign(𝑀)
(𝐸𝐼) (𝐸𝐼) (22.166.6)
e
(𝐸𝐼)
= [𝜅̇ − 𝜆̇ sign(𝑀)]sign(𝑀)
(𝐸𝐼)p
(𝐸𝐼)e 𝜅̇ sign(𝑀)
⇒ 𝜆̇ ≡ 𝜅̅̇ =
(𝐸𝐼)p + (𝐸𝐼)e

Moment rate is also the product of tangential bending stiffness and total curvature:
𝑀̇ = (𝐸𝐼)ep 𝜅̇. (22.166.7)
Elastic, plastic, and tangential stiffnesses are obtained from user-defined curves:
𝑑𝑀 𝑑𝑀
(𝐸𝐼)ep = , (𝐸𝐼)p = . (22.166.8)
𝑑𝜅 𝑑𝜅̅p

20-268 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

Both are obtained from user-defined curves.

e (𝐸𝐼)ep (𝐸𝐼)p
(𝐸𝐼) = . (22.166.9)
(𝐸𝐼)p − (𝐸𝐼)ep
For Torsion-Twist, simply replace 𝑀 by 𝑇, 𝜅 by 𝛽, (𝐸𝐼) by (𝐺𝐽). For Force-Strain, simply
replace 𝑀 by 𝑁, 𝜅 by 𝜀, (𝐸𝐼) by (𝐸𝐴).

LS-DYNA DEV 10/27/16 (r:8004) 20-269 (Material Models)


Material Models LS-DYNA Theory Manual

22.169 Material Model 169: Arup Adhesive


The through-thickness direction is identified from the smallest dimension of each
element. It is expected that this dimension will be much smaller than in-plane dimensions
(typically 2mm compared with 10mm).

In-plane stresses are set to zero: it is assumed that the stiffness and strength of the
substrate is large compared with that of the adhesive, given the relative thicknesses. If the
substrate is modeled with shell elements, it is expected that these will lie at the mid-surface
of the substrate geometry. Therefore the solid elements representing the adhesive will be
thicker than the actual bond. The yield and failure surfaces are treated as a power-law
combination of direct tension and shear across the bond:(𝜎/𝜎max )PWRT + (𝜏/𝜏max )PWRS =
1.0 at yield. The stress-displacement curves for tension and shear are shown in the
diagrams below. In both cases, 𝐺c is the area under the curve. Because of the algorithm
used, yielding in tension across the bond does not require strains in the plane of the bond –
unlike the plasticity models, plastic flow is not treated as volume-conserving.

The Plastic Strain output variable has a special meaning:


• 0 < ps < 1: ps is the maximum value of the yield function experienced since time
zero
• 1 < ps < 2: the element has yielded and the strength is reducing towards
failure – yields at ps = 1, fails at ps = 2.

Stress Stress

SHRMAX dp = SHRP.dfs
TENMAX
Area = Gcten Area = Gcshr

Failure Failure

Displacement dp Displacement
dft dfs
Tension Shear
Figure 22.169.1.

20-270 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.170 Material Model 170: Resultant Anisotropic


The in-plane elastic matrix for in-plane plane stress behavior is given by:
𝑄 𝑄12p 0 0 0
⎡ 11p ⎤
⎢𝑄12p 𝑄22p 0 0 0 ⎥
⎢ ⎥
𝐂in plane =⎢ 0 0 𝑄44p 0 0 ⎥. (22.170.1)
⎢ 0 0 0 𝑄55p 0 ⎥
⎢ ⎥
⎣ 0 0 0 0 𝑄66p ⎦

The terms Q𝑖𝑗p are defined as:


𝐸11p
𝑄11p = ,
1 − 𝜈12p 𝜈21p
𝐸22p
𝑄22p = ,
1 − 𝜈12p 𝜈21p
𝜈12p 𝐸11p (22.170.2)
𝑄12p = ,
1 − 𝜈12p 𝜈21p
𝑄44p = 𝐺12p ,
𝑄55p = 𝐺23p ,
𝑄66p = 𝐺31p .

The elastic matrix for bending behavior is given by:


𝑄11b 𝑄12b 0
𝐂bending ⎡
= ⎢𝑄12b 𝑄22b 0 ⎤ ⎥. (22.170.3)
⎣ 0 0 𝑄44b ⎦
The terms 𝑄𝑖𝑗b are similarly defined.

LS-DYNA DEV 10/27/16 (r:8004) 20-271 (Material Models)


Material Models LS-DYNA Theory Manual

22.175 Material Model 175: Viscoelastic Maxwell


Rate effects are taken into accounted through linear viscoelasticity by a convolution
integral of the form:
𝑡 𝜕𝜀𝑘𝑙
𝜎𝑖𝑗 = ∫ 𝑔𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) 𝑑𝜏 , (22.175.1)
0 𝜕𝜏
where 𝑔𝑖𝑗𝑘𝑙 (𝑡 − 𝜏) is the relaxation function for different stress measures. This stress is
added to the stress tensor determined from the strain energy functional.

If we wish to include only simple rate effects, the relaxation function is represented
by six terms from the Prony series:
𝑁
𝑔(𝑡) = ∑ 𝐺𝑚 𝑒−𝛽𝑚 𝑡 . (22.175.2)
𝑚=1

We characterize this in the input by shear moduli, 𝐺𝑖 , and the decay constants, 𝛽𝑖 .
An arbitrary number of terms, up to 6, may be used when applying the viscoelastic model.

For volumetric relaxation, the relaxation function is also represented by the Prony
series in terms of bulk moduli:
𝑁
𝑘(𝑡) = ∑ 𝐾𝑚 𝑒−𝛽𝑘𝑚 𝑡 . (22.175.3)
𝑚=1

The Arrhenius and Williams-Landau-Ferry (WLF) shift functions account for the
effects of the temperature on the stress relaxation. A scaled time, 𝑡′ ,
𝑡
𝑡′ = ∫ 𝛷(𝑇)𝑑𝑡, (22.175.4)
0

is used in the relaxation function instead of the physical time. The Arrhenius shift function
is
1 1
𝛷(𝑇) = exp (−𝐴 { − }), (22.175.5)
𝑇 𝑇REF
and the Williams-Landau-Ferry shift function is
𝑇 − TREF
𝛷(𝑇) = exp (−𝐴 ). (22.175.6)
𝐵 + 𝑇 − 𝑇REF

20-272 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.176 Material Model 176: Quasilinear Viscoelastic


The equations for this model are given as:
𝑡
𝜕𝜎𝜀 [𝜀(𝜏)] 𝜕𝜀
𝜎(𝑡) = ∫ 𝐺(𝑡 − 𝜏) 𝑑𝜏,
𝜕𝜀 𝜕𝜏
0
𝑛
𝐺(𝑡) = ∑ 𝐺𝑖 𝑒−𝛽𝑡 , (22.176.1)
𝑖=1
𝑘
𝜎𝜀 (𝜀) = ∑ 𝐶𝑖 𝜀𝑖 ,
𝑖=1

where G is the shear modulus. In place of the effective plastic strain in the D3PLOT
database, the effective strain is output:

2
𝜀effective = √ 𝜀𝑖𝑗 𝜀𝑖𝑗 . (22.176.2)
3
The polynomial for instantaneous elastic response should contain only odd terms if
symmetric tension-compression response is desired.

LS-DYNA DEV 10/27/16 (r:8004) 20-273 (Material Models)


Material Models LS-DYNA Theory Manual

22.177 Material Models 177 and 178: Hill Foam and


Viscoelastic Hill Foam
22.177.1 Hyperelasticity Using the Principal Stretch Ratios
Material types 177 and 178 in LS-DYNA are highly compressible Ogden models
combined with viscous stress contributions. The latter model also allows for an additive
viscoelastic stress contribution. As for the rate independent part, the constitutive law is
determined by a strain energy function that is expressed in terms of the principal stretches,
i.e., 𝑊 = 𝑊(𝜆1 , 𝜆2 , 𝜆3 ). To obtain the Cauchy stress 𝜎𝑖𝑗 , as well as the constitutive tensor of
TC
interest, 𝐶𝑖𝑗𝑘𝑙 , they are first calculated in the principal basis after which they are
transformed back to the “base frame”, or standard basis. The complete set of formulas is
given by Crisfield [1997] and is for the sake of completeness recapitulated here.

The principal Kirchhoff stress components are given by


𝜕𝑊
𝜏𝑖𝑖E = 𝜆𝑖 (no sum), (22.177.1)
𝜕𝜆𝑖
that are transformed to the standard basis using the standard formula
E
𝜏𝑖𝑗 = 𝑞𝑖𝑘 𝑞𝑗𝑙 𝜏𝑘𝑙 . (22.177.2)

The 𝑞𝑖𝑗 are the components of the orthogonal tensor containing the eigenvectors of
the principal basis. The Cauchy stress is then given by
𝜎𝑖𝑗 = 𝐽 −1 𝜏𝑖𝑗 , (22.177.3)
where 𝐽 = 𝜆1 𝜆2 𝜆3 is the relative volume change.

The constitutive tensor that relates the rate of deformation to the Truesdell
(convected) rate of Kirchhoff stress in the principal basis can be expressed as

TKE 𝜕𝜏𝑖𝑖E ⎫
}
𝐶𝑖𝑖𝑗𝑗 = 𝜆𝑗 − 2𝜏𝑖𝑖E 𝛿𝑖𝑗 }
𝜕𝜆𝑗 }
}
2 E 2 E
𝜆𝑗 𝜏𝑖𝑖 − 𝜆𝑖 𝜏𝑗𝑗 }
TKE
}
𝐶𝑖𝑗𝑖𝑗 = 2 2
, 𝑖 ≠ 𝑗, 𝜆𝑖 ≠ 𝜆𝑗 ⎬ (no sum). (22.177.4)
𝜆𝑖 − 𝜆𝑗 }
}
E E }
TKE 𝜆𝑖 𝜕𝜏𝑖𝑖 𝜕𝜏𝑖𝑖 }
𝐶𝑖𝑗𝑖𝑗 = ( − ), 𝑖 ≠ 𝑗, 𝜆𝑖 = 𝜆𝑗 }
}
2 𝜕𝜆𝑖 𝜕𝜆𝑗 ⎭

These components are transformed to the standard basis according to


TK TKE
𝐶𝑖𝑗𝑘𝑙 = 𝑞𝑖𝑝 𝑞𝑗𝑞 𝑞𝑘𝑟 𝑞𝑙𝑠 𝐶𝑝𝑞𝑟𝑠 , (22.177.5)

20-274 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

and finally the constitutive tensor relating the rate of deformation to the Truesdell rate of
Cauchy stress is obtained through
TC TK
𝐶𝑖𝑗𝑘𝑙 = 𝐽 −1 𝐶𝑖𝑗𝑘𝑙 . (22.177.6)

22.177.2 Hill’s Strain Energy Function


The strain energy function for materials 177 and 178 is given by
𝑛
𝜇𝑚 𝛼𝑚 𝛼 𝛼 1
𝑊= ∑ [𝜆1 + 𝜆2𝑚 + 𝜆3𝑚 − 3 + (𝐽 −𝑛𝛼𝑚 − 1)]. (22.177.7)
𝛼
𝑚=1 𝑚
𝑛
where 𝑛, 𝜇𝑚 and 𝛼𝑚 are material parameters. To apply the formulas in the previous
section, we require
𝑛
𝜕𝑊 𝜇 𝛼
𝜏𝑖𝑖E = 𝜆𝑖 = ∑ 𝑚 (𝜆𝑖 𝑚 − 𝐽 −𝑛𝛼𝑚 ). (22.177.8)
𝜕𝜆𝑖 𝑚=1 𝐽

Proceeding with the constitutive tensor, we have


𝜕𝜏𝑖𝑖E 𝑛
𝜆𝑗 = ∑ 𝜇𝑚 𝛼𝑚 (𝜆𝛼𝑖 𝑚 𝛿𝑖𝑗 + 𝑛𝐽 −𝑛𝛼𝑚 ). (22.177.9)
𝜕𝜆𝑗 𝑚=1

In addition to the hyperelastic stress described above, a viscous stress is added.


Converting to Voigt notation, this stress can be written,
𝛔 = 𝐂𝐃, (22.177.10)
where 𝛔 denotes Cauchy stress, 𝐃 is the rate-of-deformation and 𝐂 is an isotropic
constitutive matrix representing the viscosity. In element m, the constitutive matrix
depends on the element deformation according to
𝑑𝑚
𝐂= 𝐂, (22.177.11)
𝐽 𝟎
where 𝑑𝑚 is the diameter4 of element m and 𝐂𝟎 is a constitutive matrix that depends only
on the material parameters. The stress contribution to the internal force can be written

𝑓 int = ∫ 𝐁T 𝛔𝑑𝛺𝑚 , (22.177.12)


𝛺𝑚

and the corresponding material time derivative is


̇
𝑓 mat = ∫ 𝐁T 𝛔∇𝑇 𝑑𝛺𝑚 . (22.177.13)
𝛺𝑚

4 Experiments indicate that d is the smallest dimension of the element.


LS-DYNA DEV 10/27/16 (r:8004) 20-275 (Material Models)
Material Models LS-DYNA Theory Manual

Here 𝛺𝑚 is the current configuration of element m, 𝐁 is the strain-displacement matrix and


∇𝑇 denotes the Truesdell rate of Cauchy stress. The aim is to identify the material tangent
modulus through
̇
𝑓 mat = ∫ 𝐁T 𝐂mat 𝐁𝑑𝛺𝑚 𝑢̇, (22.177.14)
𝛺𝑚

for the viscous stress with u̇ being the nodal velocity. The Truesdell rate of the viscous
stress can be written,
𝛔𝛁𝐓 = 𝐂𝐃
̇ + 𝐂𝐃̇ + tr(𝐃)𝛔 − 𝐋𝛔 − 𝛔𝐋𝐓 , (22.177.15)
where Lis the velocity gradient. The terms on the right hand side can be treated as follows.

For the first term, we can assume that 𝑑𝑚 ∝ J1/3 and then approximate
2
𝐂̇ = − tr(𝐃)𝐂. (22.177.16)
3

Using Equation (22.177.10), Equation (22.177.13), the first term on the right hand side
of Equation (22.177.15), Equation (22.177.16) and the expression
𝐃 = 𝐁𝐮̇, (22.177.17)
a material tangent modulus contribution can be identified in Equation (22.177.14) as
2
− 𝛔𝛅T , (22.177.18)
3
where 𝛅 denotes the identity matrix in Voigt notation.

For the second term in Equation (22.177.15), we differentiate Equation (22.177.17) to


see that
𝐃̇ = 𝐁̇𝐮̇ + 𝐁𝐮̈. (22.177.19)

Post-poning the treatment of the first term, the second of these two terms can be
treated easily as this gives the following contribution to the material time derivative
𝛾
∫ 𝐁T 𝐂𝐁𝑑𝛺𝑚 𝑢̇ , (22.177.20)
𝛽Δ𝑡
𝛺𝑚

where γ and 𝛽 are parameters in the Newmark scheme and Δ𝑡 is the time step. From this
expression, a material tangent modulus can through Equation (22.177.14) be identified as,
𝛾
𝐂mat = 𝐂. (22.177.21)
𝛽Δ𝑡

The third term in Equation (22.177.15) contributes to the material tangent modulus
as

20-276 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

𝛔𝛅T (22.177.22)
resulting in a material tangent modulus given so far by
𝛾 1
𝐂 + 𝛔𝛅T . (22.177.23)
𝛽Δ𝑡 3

22.177.3 Viscous Stress


From the remaining terms, i.e., the last two terms in Equation (22.177.15) and the
first term in Equation (22.177.19), we see it impossible to identify contributions to a
material tangent modulus. We believe that these terms must be treated in some other
manner. We are thus left with two choices, either to approximate these terms within the
existing framework or to attempt a thorough implementation of the correct tangent
stiffness using a different, and probably demanding, approach. We reason as follows.

Since this stress contribution is viscous and proportional to the mesh size, it is our
belief that it serves as a stabilizing stress in the occurrence of a coarse mesh and/or large
deformation rates, and really has little or nothing to do with the actual material models. If
only the simulation process is slow (which it often is in an implicit analysis) and/or the
mesh is sufficiently fine, this stress should be negligible compared to the other stress(es).
With this in mind, we feel that it is not crucial to derive an exact tangent for this stress but
we can be satisfied with an approximation. Even if attempting a more thorough derivation
of the tangent stiffness, we would most certainly have to make approximations along the
way. Hence we do not see this as an attractive approach.

In the implementation we have simply neglected all terms involving stresses since
the experience from earlier work is that such terms generally have a negative effect on the
tangent if they are not absolutely correct. In addition, most of the terms involving stresses
contribute to a nonsymmetric tangent stiffness, which cannot be supported by LS-DYNA at
the moment. Hence the material tangent modulus for the viscous stress is given by
Equation (22.177.21). We are aware of that this may be a crude approximation, and if
experiments show that it is a poor one, we will take a closer look at it.

In material type 178, the viscous stress acts only in the direction of the principal
stretches and in compression. With C being an isotropic tensor, we evaluate the tangent
stiffness modulus in the principal basis according to Equation (22.177.21), modify it to
account for the mentioned conditions and then transform it back to the global frame of
reference.

22.177.4 Viscoelastic Stress Contribution


For material 178, an optional viscoelastic stress contribution can be added. The
evolution of this stress in time can be stated as

LS-DYNA DEV 10/27/16 (r:8004) 20-277 (Material Models)


Material Models LS-DYNA Theory Manual

12
𝜎𝑖𝑗∇ = ∑ 2𝐺𝑚 𝑠m∇
𝑖𝑗 , (22.177.24)
𝑚=1

where
𝑠m∇ m
𝑖𝑗 = 𝐷𝑖𝑗 − 𝛽𝑚 𝑠𝑖𝑗 , (22.177.25)
Here 𝐺𝑚 and 𝛽𝑚 are material constants, and 𝐷𝑖𝑗 is the rate-of-deformation tensor. Referring
to Borrvall [2002], we state that the tangent stiffness modulus for this stress contribution
can be written
12
𝐶𝑖𝑗𝑘𝑙 = ∑ 𝐺𝑚 (𝛿𝑖𝑘 𝛿𝑗𝑙 + 𝛿𝑖𝑙 𝛿𝑗𝑘 ). (22.177.26)
𝑚=1

Just as for the viscous stress, this stress acts only in the direction of the principal stretches.
Hence the tangent modulus is formed in the principal basis, modified to account for this
condition and then transformed back to the global frame of reference.

22.177.5 Material Tangent Modulus for the Fully Integrated Brick


To avoid locking tendencies for the fully integrated brick element in LS-DYNA, the
stress is modified as
𝛔S/R = 𝛔 + (𝑝 − 𝑝̅)𝐈, (22.177.27)
where 𝑝 is the pressure and 𝑝̅ is the mean pressure in the element. This affects the tangent
stiffness since one has to take into account that the pressure is constant in the element.
Deriving the material time derivative of the internal force results in
̇
𝑓 mat = ∫ 𝐁T 𝐂𝐁𝑑𝛺𝑚 𝑢̇ + ∫ (𝑝 − 𝑝̅)𝐁T (𝐈 ⊗ 𝐈)𝐁𝑑𝛺𝑚 𝑢̇
𝛺𝑚 𝛺𝑚
(22.177.28)
−2 ∫ (𝑝 − 𝑝̅)𝐁T 𝐁𝑑𝛺𝑚 𝑢̇ + ∫ (ṗ − p
̅̅̅̅̇ )BT dΩm .
𝛺𝑚 Ωm

To implement this tangent, the last term is the most difficult to deal with as it
involves the time derivative (or variation) of the pressure. For certain types of material
models, for instance material type 77 in LS-DYNA, the pressure is a function of the relative
volume
𝑝 = 𝑝(𝐽), (22.177.29)
and with the approximation
𝑝̅ = 𝑝(𝐽 )̅ , (22.177.30)
the last term can be evaluated to

∫ 𝐽𝑝′(𝐽)𝐁T (𝐈 ⊗ 𝐈)𝐁𝑑𝛺𝑚 𝑢̇ − ∫ 𝐽 𝑝̅ ′ (𝐽 )̅ 𝐁
̅̅̅̅̅T (𝐈 ⊗ 𝐈)𝐁
̅̅̅̅̅𝑑𝛺𝑚 𝑢̇,
(22.177.31)
𝛺𝑚 𝛺𝑚

20-278 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

and a symmetric tangent stiffness can quite easily be implemented. We have here used 𝐽 ̅
and 𝐁̅̅̅̅̅ for the mean values of 𝐽 and 𝐁, respectively. For other types of material models,
such as the ones described in this document or material type 27 in LS-DYNA, the
expression for the pressure is more complicated. A characterizing feature is that a non-zero
pressure can occur under constant volume. This will in general complicate the
implementation of the last term and will also contribute to a non-symmetric tangent
stiffness that cannot be handled in LS-DYNA at the moment. For material 27, neglecting
this had a tremendous impact on the performance of the implicit solution procedure, (see
material type 27). For the current material models, it seems to be of less importance, and
we believe that this is due to the higher compressibility allowed.

LS-DYNA DEV 10/27/16 (r:8004) 20-279 (Material Models)


Material Models LS-DYNA Theory Manual

22.179 Material Models 179 and 180: Low Density Synthetic


Foam
Material types 179 and 180 in LS-DYNA are highly compressible synthetic foam
models with no Poisson’s ratio effects combined with an optional visco-elastic and a
stabilizing viscous stress contribution. The tensile behavior of the materials is linear where
the stress cannot exceed a user prescribed cutoff stress. In compression the materials show
a hysteresis on unloading similar to material 57. In addition, the first load cycle damages
the material so that the stress level on reloading is significantly reduced. For material 179
the damage is isotropic while it is orthotropic for material 180.

22.179.1 Hyperelasticity Using the Principal Stretch Ratios


As for the rate independent part of the stress, the constitutive law is mainly
determined by a strain energy function that is expressed in terms of the principal stretches,
i.e., 𝑊 = 𝑊(𝜆1 , 𝜆2 , 𝜆3 ). To obtain the Cauchy stress 𝜎𝑖𝑗 , as well as the constitutive tensor of
TC
interest, 𝐶𝑖𝑗𝑘𝑙 , they are first calculated in the principal basis after which they are
transformed back to the “base frame”, or standard basis. The complete set of formulas is
given by Crisfield [1997] and is for the sake of completeness recapitulated here.

The principal Kirchhoff stress components are given by


𝜕𝑊
𝜏𝑖𝑖E = 𝜆𝑖 (no sum), (22.179.1)
𝜕𝜆𝑖
that are transformed to the standard basis using the standard formula
E
𝜏𝑖𝑗 = 𝑞𝑖𝑘 𝑞𝑗𝑙 𝜏𝑘𝑙 . (22.179.2)

The 𝑞𝑖𝑗 are the components of the orthogonal tensor containing the eigenvectors of
the principal basis. The Cauchy stress is then given by ,
𝜎𝑖𝑗 = 𝐽 −1 𝜏𝑖𝑗 , (22.179.3)
where 𝐽 = 𝜆1 𝜆2 𝜆3 is the relative volume change.

The constitutive tensor that relates the rate of deformation to the Truesdell
(convected) rate of Kirchhoff stress can in the principal basis be expressed as

20-280 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

TKE 𝜕τEii
𝐶𝑖𝑖𝑗𝑗 = 𝜆𝑗 − 2𝜏𝑖𝑖E 𝛿𝑖𝑗
𝜕𝜆𝑗

TKE
𝜆2𝑗 𝜏𝑖𝑖E − 𝜆2𝑖 𝜏𝑗𝑗E
𝐶𝑖𝑗𝑖𝑗 = , 𝑖 ≠ 𝑗, 𝜆𝑖 ≠ 𝜆𝑗 (no sum). (22.179.4)
𝜆2𝑖 − 𝜆2𝑗
TKE 𝜆𝑖 𝜕𝜏𝑖𝑖E 𝜕𝜏𝑖𝑖E
𝐶𝑖𝑗𝑖𝑗 = ( − ), 𝑖 ≠ 𝑗, 𝜆𝑖 = 𝜆𝑗
2 𝜕𝜆𝑖 𝜕𝜆𝑗

These components are transformed to the standard basis according to


TK TKE
𝐶𝑖𝑗𝑘𝑙 = 𝑞𝑖𝑝 𝑞𝑗𝑞 𝑞𝑘𝑟 𝑞𝑙𝑠 𝐶𝑝𝑞𝑟𝑠 , (22.179.5)
and finally the constitutive tensor relating the rate of deformation to the Truesdell rate of
Cauchy stress is obtained through.
TC TK
𝐶𝑖𝑗𝑘𝑙 = 𝐽 −1 𝐶𝑖𝑗𝑘𝑙 . (22.179.6)

22.179.2 Strain Energy Function


The strain energy function for materials 179 and 180 is given by
3
W = ∑ 𝑤(λm ), (22.179.7)
m=1

where
⎧𝑠 (𝜆 − 1 − 𝑠 ) if 𝜆 ≥
𝑠
+1
{
{ 2𝐸 𝐸
{
{
{𝐸 2 𝑠
𝑤(𝜆) = ⎨ 2 (𝜆 − 1) if 1 ≤ 𝜆 < + 1
𝐸 (22.179.8)
{
{ 𝜆
{
{
{ 𝑓s (1 − 𝜇)𝑑𝜇
∫ otherwise
⎩1
Here s is the nominal tensile cutoff stress and 𝐸 is the stiffness coefficient relating a change
in principal stretch to a corresponding change in nominal stress. The function 𝑓s (≤ 0) gives
the nominal compressive stress as a function of the strain in compression for the second
and all subsequent load cycles and is supplied by the user. To apply the formulas in the
previous section, we require
⎧𝑠𝜆 𝑠
{ 𝑖 if 𝜆𝑖 ≥ + 1
𝜕𝑤 { { 𝐸
E
𝜏𝑖𝑖 = 𝜆𝑖 = ⎨𝐸𝜆 (𝜆 − 1) if 1 ≤ 𝜆 < 𝑠 + 1 (22.179.9)
𝜕𝜆𝑖 { 𝑖 𝑖 𝑖
𝐸
{
{
⎩𝜆𝑖 𝑓𝑠 (1 − 𝜆𝑖 ) otherwise

Proceeding with the constitutive tensor, we have

LS-DYNA DEV 10/27/16 (r:8004) 20-281 (Material Models)


Material Models LS-DYNA Theory Manual

⎧𝑠𝜆 𝑠
{ 𝑖 if 𝜆𝑖 ≥ + 1
𝜕𝜏 E {
{ 𝐸
𝜆𝑗 𝑖𝑖 = 𝛿𝑖𝑗 ⎨𝐸𝜆 (2𝜆 − 1) 𝑠 (22.179.10)
𝜕𝜆𝑗 { 𝑖 𝑖 if 1 ≤ 𝜆 𝑖 < +1
{ 𝐸
{
⎩𝜆𝑖 (𝑓𝑠 (1 − 𝜆𝑖 ) − 𝜆𝑖 𝑓 ′𝑠 (1 − 𝜆𝑖 )) otherwise

22.179.3 Modeling of the Hysteresis


The hyperelastic part of the Cauchy stress is scaled by a factor 𝜅 given by
𝐸
𝜅= , (22.179.11)
̅̅̅̅
𝐸
where

𝐸 = ∫ 𝐽𝛔: 𝑑𝛆, (22.179.12)

is the stored energy in the material and


̅̅̅̅ = 𝐸max exp(−𝛽(𝑡 − 𝑠)).
𝐸 (22.179.13)
Here 𝑠 stands for the time point when E has its maximum 𝐸max in the interval [0, 𝑡]. The
factor κ is introduced to model the hysteresis that characterizes this material (and material
57). The decay coefficient 𝛽 is introduced to get a reloading curve similar to the original
loading curve.

This factor κ is treated as a constant in the determination of the tangent stiffness


matrix.

22.179.4 Viscous Stress


In addition to the hyperelastic stress described above, a viscous stress is added.
Converting to Voigt notation, this stress can be written
𝛔 = 𝐂𝐃, (22.179.14)
where 𝛔 denotes Cauchy stress, 𝐃 is the rate-of-deformation and 𝐂 is an isotropic
constitutive matrix representing the viscosity. In element m, the constitutive matrix
depends on the element deformation according to
𝑑𝑚
𝐂= 𝐂 , (22.179.15)
𝐽 0
where 𝑑𝑚 is the diameter5 of element m and 𝐂0 is a constitutive matrix that depends only
on the material parameters. Following material models 177 and 178 we use the following
material tangent stiffness for this stress contribution

5 Experiments indicate that d is the smallest dimension of the element.


20-282 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Material Models

𝛾
𝐂mat = 𝐂, (22.179.16)
𝛽Δ𝑡
where 𝛾 and 𝛽 are parameters in the Newmark scheme and Δ𝑡 is the time step.

22.179.5 Viscoelastic Stress Contribution


An optional viscoelastic stress contribution can be added. The evolution of this
stress in time can be stated as
𝜎𝑖𝑗∇ = 𝐸d 𝑠∇
𝑖𝑗 , (22.179.17)
where
𝑠∇
𝑖𝑗 = 𝐷𝑖𝑗 − 𝛽1 𝑠𝑖𝑗 . (22.179.18)
Here 𝐸d and 𝛽1 are material constants, 𝐷𝑖𝑗 is the rate-of-deformation tensor and ∇ stands
for an objective rate. Referring to material models 177 and 178, we state that the tangent
stiffness modulus for this stress contribution can be written
𝐸d
𝐶𝑖𝑗𝑘𝑙 = (𝛿 𝛿 + 𝛿𝑖𝑙 𝛿𝑗𝑘 ). (22.179.19)
2 𝑖𝑘 𝑗𝑙

This stress acts only in the direction of the principal stretches. Hence the tangent
modulus is formed in the principal basis, modified to account for this condition and then
transformed back to the global frame of reference.

22.179.6 Stress Corresponding to First Load Cycle


We define a contribution to the principal Kirchhoff stress as
1
𝜏𝑖𝑖E = 𝜆𝑖 {𝑔s (1 − 𝜆𝑖 ) − 𝑓s (1 − 𝜆𝑖 )}𝜉 . (22.179.20)

When the damage is isotropic the factor 𝜉 is given by


𝜀h
𝜉 = max (0,1 − ). (22.179.21)
0.0001 + 𝜀m
where 𝜀h is the damage parameter that is initially zero and 𝜀m is the maximum compressive
volumetric strain during the entire simulation thus far. Damage evolves when the material
is in compression and unloads
0 if 𝐽 ≥ 1
Δ𝜀h = { , (22.179.22)
max(0, Δ𝐽) otherwise
where 𝐽 is the jacobian of the deformation. The first load cycle will result in a total stress
that follows load curve 𝑔𝑠 since there is no damage. After a complete load cycle, i.e.,
unloading has occurred, the material is completely damaged, i.e., 𝜀h ≈ 𝜀m , and the nominal
stress will for the second and subsequent load cycles be given by the load curve 𝑓s .

LS-DYNA DEV 10/27/16 (r:8004) 20-283 (Material Models)


Material Models LS-DYNA Theory Manual

In the orthotropic case the principal Kirchhoff stress contribution is instead given by
1
𝜏𝑖𝑖E = 𝜆𝑖 {𝑔𝑠 (1 − 𝜆𝑖 ) − 𝑓𝑠 (1 − 𝜆𝑖 )}𝜉𝑖 . (22.179.23)

For the damage to be orthotropic we introduce a symmetric and positive definite


𝑖𝑗
damage tensor 𝜀h . This tensor is initially the zero tensor corresponding to no damage. The
evolution of damage begins with a half step Jaumann rotation of the tensor to maintain
objectivity. After that the local increment is performed. As for the isotropic case, damage
evolves in compression in combination with unloading. We introduce the local damage
increment as
𝑖𝑗 0 if 𝜆𝑖 ≥ 1
Δ𝜀loc = 𝛿𝑖𝑗 { , (22.179.24)
max(0, Δ𝜆𝑖 ) otherwise
which is a diagonal tensor. The global damage tensor increment is given by
𝑖𝑗
Δ𝜀ℎ = 𝑞𝑖𝑘 𝑞𝑗𝑙 Δ𝜀𝑘𝑙
loc , (22.179.25)
𝑖𝑗
which is used to increment the damage tensor 𝜀ℎ . The factor 𝜉𝑖 is now given by
𝑞𝑘𝑖 𝑞𝑙𝑖 𝜀𝑘𝑙

𝜉𝑖 = max ⎛
⎜0,1 − ⎞
⎟. (22.179.26)
⎝ 0.0001 + 𝜀 m⎠

where the quantity 𝜀m in the orthotropic case is the maximum compressive principal strain
in any direction during the simulation thus far. As for the isotropic case, the material is
completely damaged after one load cycle and reloading will follow load curve 𝑓s . In
addition, the directions corresponding to no loading will remain unaffected.

The factors 𝜉 and 𝜉𝑖 are treated as constants in the determination of the tangent
stiffness so the contribution is regarded as hyperelastic and follows the exposition given in
Section 19.179.1.

The reason for not differentiating the coefficients 𝜅, 𝜉 and 𝜉𝑖 is that they are always
non-differentiable. Their changes depend on whether the material is loaded or unloaded,
i.e., the direction of the load. Even if they were differentiable their contributions would
occasionally result in a non-symmetric tangent stiffness matrix and any attempt to
symmetrize this would probably destroy its properties. After all, we believe that the one-
dimensional nature and simplicity of this foam will be enough for good convergence
properties even without differentiating these coefficients.

20-284 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.181 Material Model 181: Simplified Rubber/Foam


Material type 181 in LS-DYNA is a simplified “quasi”-hyperelastic rubber or foam
model defined by a single uniaxial load curve or by a family of curves at discrete strain
rates. The term “quasi” is used because there is really no strain energy function for
determining the stresses used in this model. Rather the stress response mimics the gradient
of the strain energy potential in the Ogden rubber (see material 77). For deriving the
tangent stiffness matrix we use the formulas as if a strain energy function were present,
with appropriate modifications.
This model is equipped with various features related to dissipation and damage, but not all
of those are described in detail.

𝑃
𝑔(𝜆) = 𝑃/𝐴
𝐴 𝜆 = 𝜆1 = 𝑙/𝐿
𝜆2 = 𝜆3 = 𝑑/𝐷
𝑑

𝑙
𝐷

20-122 Uniaxial test parameters.

22.181.1 Hyperelasticity Using the Principal Stretch Ratios


A hyperelastic constitutive law is determined by a strain energy function that we
assume is expressed in terms of the principal stretches, i.e., 𝑊 = 𝑊(𝜆1 , 𝜆2 , 𝜆3 ). To obtain
the Cauchy stress 𝜎𝑖𝑗 this is first calculated in the principal basis after which it is
transformed back to the “base frame”, or standard basis. The complete set of formulas is
given by Crisfield [1997] and is for the sake of completeness recapitulated here. For the
following discussion we refer to figure 20-122

The principal Kirchhoff stress components are given by


𝜕𝑊
𝜏𝑖𝑖E = 𝜆𝑖 (no sum), (22.181.1)
𝜕𝜆𝑖
that are transformed to the standard basis using the standard formula

LS-DYNA DEV 10/27/16 (r:8004) 20-285 (Material Models)


Material Models LS-DYNA Theory Manual

E
𝜏𝑖𝑗 = 𝑞𝑖𝑘 𝑞𝑗𝑙 𝜏𝑘𝑙 . (22.181.2)

The 𝑞𝑖𝑗 are the components of the orthogonal tensor containing the eigenvectors of
the principal basis. The Cauchy stress is then given by
𝜎𝑖𝑗 = 𝐽 −1 𝜏𝑖𝑗 , (22.181.3)
where 𝐽 = 𝜆1 𝜆2 𝜆3 is the relative volume change.

Now, the Ogden strain energy potential results in a Kirchhoff stress on the form
1 3
𝜏𝑖𝑖𝐸 = 𝑓 (𝜆̃ 𝑖 ) + 𝐾𝑚 (𝐽 − 1) − ∑ 𝑓 (𝜆̃ 𝑘 ) (22.4)
3 𝑘=1
for a (large) bulk modulus 𝐾𝑚 and where 𝜆̃ 𝑖 = 𝜆𝑖 /𝐽 1/3 are the isochoric stretches. In the
Ogden material, 𝑓 has a specific form a priori that requires a least square approximation for
fitting test data. This is of course a restriction and the idea in material 181 is to let 𝑓 be
determined directly from input data. The ansatz for the compressible foam option is to let
− 𝜈
𝜏𝑖𝑖𝐸 = 𝑓 (𝜆𝑖 ) − 𝑓 (𝐽 1−2𝜈 )

(22.5)

for a given Poisson’s ratio 𝜈, a decision that will be made clear below. So assume that 𝑔(𝜆)
is the curve providing the engineering stress as function of stretch in a uniaxial test, see
figure 20-122, then the principal Kirchhoff stresses are
𝐸
𝜏11 = 𝜆𝑔(𝜆)
𝐸 𝐸
𝜏22 = 𝜏33 =0

(22.6)

What follows is the determination of the internal function 𝑓 for the rubber and foam option.

22.181.1.1 Determination of f, rubber option


For incompressibility we deduce that the principal stretches are
𝜆1 = 𝜆
𝜆2 = 𝜆3 = 𝜆−1/2 (22.7)

20-286 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

which coincide with the isochoric counterparts. Using these expressions when equating
(22.4) and (22.6), one must determine 𝑓 from
2
𝜆𝑔(𝜆) = (𝑓 (𝜆) − 𝑓 (𝜆−1/2 )) + 𝐾𝑚 (𝐽 − 1)
3
1
0 = (𝑓 (𝜆−1/2 ) − 𝑓 (𝜆)) + 𝐾𝑚 (𝐽 − 1).
3
(22.8)

By subtracting these two equations to eliminate the influence of the pressure we get

−1
𝜆𝑔(𝜆) = 𝑓 (𝜆) − 𝑓 (𝜆 2 )

(22.9)

that can be rewritten as


𝑓 (𝜆) = 𝜆𝑔(𝜆) + 𝑓 (𝜆−1/2 ).
(22.10)
This in turn can be recursively expanded as
𝑛
𝑓 (𝜆) = 𝜆𝑔(𝜆) + 𝜆−1/2 𝑔(𝜆−1/2 ) + 𝜆1/4 𝑔(𝜆1/4 ) + ⋯ + 𝑓 (𝜆(−1/2) )
(22.11)
)

and by letting 𝑛 be large enough the function 𝑓 can be determined since the last term tends
to zero.

22.181.1.2 Determination of f, foam option


Similarly, for a given Poisson’s ratio 𝜈, the principal stretches in a uniaxial tension test are
𝜆1 = 𝜆
𝜆2 = 𝜆3 = 𝜆−𝜈 (22.12)

LS-DYNA DEV 10/27/16 (r:8004) 20-287 (Material Models)


Material Models LS-DYNA Theory Manual

and using (22.5) and (22.6) the equation to solve is now


𝜆𝑔(𝜆) = 𝑓 (𝜆) − 𝑓 (𝜆−𝜈 )

(22.13)

Note that the equation corresponding to the second of (22.6) vanishes because of the ansatz
in (22.5). The same technique as for the rubber option is used, recursive expansion gives
2 2 𝑛
𝑓 (𝜆) = 𝜆𝑔(𝜆) + 𝜆−𝜈 𝑔(𝜆−𝜈 ) + 𝜆𝜈 𝑔(𝜆𝜈 ) + ⋯ + 𝑓 (𝜆(−𝜈) )
(22.14)

which for a large 𝑛 gives a sufficiently accurate representation of 𝑓 .

22.181.2 Some Remarks

22.181.2.1 Strain rates


The function 𝑓 introduced in the previous section depends not only on the stretches
but for some choices of input also on the strain rate. In this case each test curve 𝑔𝑖 (𝜆)
corresponding to a particular strain rate 𝜀̇𝑖 is converted to an internal function 𝑓𝑖 (𝜆)
following the procedure described in the previous section. These internal functions are
then used for determining the response for a given strain rate 𝜀̇ by interpolation. Strain
rates are treated in various ways depending on user defined parameters and we refer to
Section and the Keyword Manual for more info.

22.181.2.2 Modeling of the Frequency Independent Damping


An elastic-plastic stress 𝜎𝑑 is added to model the frequency independent damping
properties of rubber. This stress is deviatoric and determined by the shear modulus 𝐺 and
the yield stress 𝜎𝑌 . This part of the stress is updated incrementally as
̃𝑑𝑛+1 = 𝝈𝑑𝑛 + 2𝑮𝑰 dev 𝛥𝜺,
𝝈 (22.181.15)
where 𝛥𝜺 is the strain increment. The trial stress is then radially scaled (if necessary) to the
yield surface according to
𝜎
𝜎𝑑𝑛+1 = 𝜎̃ 𝑑𝑛+1 min (1, 𝑌 ), (22.181.16)
𝜎eff
where 𝜎eff is the effective von Mises stress for the trial stress 𝜎̃ 𝑑𝑛+1 .

20-288 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

22.187 Material Model 187: Semi-Analytical Model for the


Simulation of Polymers
22.187.1 Material law formulation
Choice of a yield surface formulation
All plastics are to some degree anisotropic. The anisotropic characteristic can be due
to fibre reinforcement, to the moulding process or it can be load induced in which case the
material is at least initially isotropic. Therefore a quadratic form in the stress tensor is often
used to describe the yield surface. We restrict the scope of this work to isotropic
formulations. However, the choice of this yield surface was made in view of later
anisotropic generalisations. In the isotropic case the most general quadratic yield surface
can be written as
𝑓 = 𝛔T 𝐅𝛔 + 𝐁𝛔 + 𝐹0 ≤ 0, (22.187.1)

where
σ𝑥𝑥

⎜ σ𝑦𝑦 ⎞⎟

⎜ ⎟

⎜ σ 𝑧𝑧 ⎟
𝛔=⎜

⎜ σ


⎟ ,
⎜ 𝑥𝑦 ⎟

⎜σ𝑦𝑧 ⎟ ⎟
σ
⎝ 𝑧𝑥 ⎠
𝐹11 𝐹12 𝐹12 0 0 0

⎜ 𝐹12 𝐹11 𝐹12 0 0 0 ⎞⎟

⎜ ⎟

⎜ 𝐹12 𝐹12 𝐹11 0 0 0 ⎟
𝐅=⎜



⎟, (22.187.2)

⎜ 0 0 0 𝐹44 0 0 ⎟⎟


⎜ 0 0 0 0 𝐹44 0 ⎟⎟
⎝ 0 0 0 0 0 𝐹44 ⎠
𝐹1 0 0 0 0 0

⎜ 0 𝐹1 0 0 0 0⎞⎟

⎜ ⎟

⎜ 0 0 𝐹1 0 0 0⎟
𝐁=⎜



⎟.

⎜ 0 0 0 0 0 0⎟⎟

⎜0 0 0 0 0 0⎟
⎝0 0 0 0 0 0⎠
Some restrictions apply to the choice of the coefficients. The existence of a stress-free state
and the equivalence of pure shear and biaxial tension/compression require respectively
𝐹0 ≤ 0 and 𝐹44 = 2(𝐹11 − 𝐹12 ). (22.187.3)
Although 4 independent coefficients remain in the expression for the isotropic yield surface
at this point, however the yield condition is not affected if all coefficients are multiplied by
a constant. Consequently only 3 coefficients can be freely chosen and 3 experiments under
different states of stress can be fitted by this formulation.
LS-DYNA DEV 10/27/16 (r:8004) 20-289 (Material Models)
Material Models LS-DYNA Theory Manual

Figure 22.1. Recommended tests for material data in SAMP

Without loss of generality the expression for the yield surface can be reformulated in
terms of the first two stress invariants: pressure and von Mises stress:
σ𝑥𝑥 + σ𝑦𝑦 + σ𝑧𝑧
𝑝=− ,
3
. (22.187.4)
3 2
σvm = √ ((σ𝑥𝑥 + 𝑝) + (σ𝑦𝑦 + 𝑝) + (σ𝑧𝑧 + 𝑝) + 2σ𝑥𝑦 + 2σ𝑦𝑧 + 2σ𝑧𝑥 )
2 2 2 2 2
2
The expression for the yield surface then becomes
2
𝑓 = σvm − 𝐴0 − 𝐴1 𝑝 − 𝐴2 𝑝2 ≤ 0, (22.187.5)
and identification of the coefficients gives
𝐴0 = −𝐹0 , 𝐴1 = 3𝐹1 and 𝐴2 = 9(1 − 𝐹11 ), (22.187.6)
or equivalently
𝐴1 𝐴 𝐹 1 𝐴
𝐹0 = −𝐴0 , 𝐹1 = , 𝐹11 = 1 − 2 , 𝐹44 = 3 and 𝐹12 = 𝐹11 − 44 = − ( + 2 ). (22.187.7)
3 9 2 2 9
Since there is no loss of generality, the simpler formulation in invariants is adopted from
this point on. In principle the coefficients of the yield surface can now be determined from
3 experiments. Typically we would perform uniaxial tension, uniaxial compression and
simple shear tests:

This allows computation of the coefficients in function of the test results:


3σs2 = 𝐴0 ⎫⎧𝐴0 = 3σs2
}{
{
2} { 2 σc − σt
2 2 σ t σt }
}{ 𝐴 = 9σ ( )
σt = 3σs − 𝐴1 + 𝐴2 ⇒
1 s
σ σ . (22.187.8)
3 9⎬ ⎨ c t
2} { σc σt − 3σs2
σ σ } {
σc2 = 3σs2 + 𝐴1 c + 𝐴2 c }
} { {𝐴2 = 9 ( σ σ )
3 9⎭ ⎩ c t

Alternatively we can also compute the coefficients relating to the formulation in stress
space:

20-290 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

⎫ ⎧ 1 1
𝐹0 + 𝐹1 𝜎t + 𝐹11 σt2 = 0 }
} {
{ 𝐹1 = 𝐹0 ( − )
} { σc σt
}
} {
{ 𝐹0
𝐹0 − 𝐹1 σc + 𝐹11 σc2 = 0⎬ ⇒ ⎨𝐹11 = − . (22.187.9)
} { σt σc
}
} {
{ 𝐹0
𝐹0 + 𝐹44 σs2 = 0 }
} {{𝐹44 = − 2
⎭ ⎩ σs
Both are easily seen to be equivalent.

Conditions for convexity of the yield surface


Usually the yield surface is required to be convex, i.e.
𝑓 (σ1 ) ≤ 0 ⎫
}
𝑓 (σ2 ) ≤ 0 ⎬ ⇒ 𝑓 (ασ1 + (1 − α)σ2 ) ≤ 0. (22.187.10)
}
0 ≤ α ≤ 1⎭
The second derivative of 𝑓 is computed as

T ∂2 𝑓
𝑓 = 𝛔 𝐅𝛔 + 𝐁𝛔 + 𝐹0 → = 2𝐅 (22.187.11)
∂σ2
A sufficient condition for convexity in 6D stress space is then that the matrix F should be
positive semidefinite. This means all eigenvalues of F should be positive or zero. The
conditions for convexity will now be examined in physical terms for two cases: plane stress
and general 3D.

The plane stress case


In the plane stress case the yield condition reduces to:
𝑓 = 𝛔T 𝐅𝛔 + 𝐁𝛔 + 𝐹0 , (22.187.12)
where
σ𝑥𝑥 𝐹11 𝐹12 0 𝐹1 0 0
⎛ ⎞ ⎛ 0 ⎞⎟ 𝐁=⎛
𝛔= ⎜ σ 𝑦𝑦 ⎟ ⎜
𝐅 = 𝐹12 𝐹11 ⎜0 𝐹1 0⎞⎟, (22.187.13)
⎝σ𝑥𝑦 ⎠ ⎝ 0 0 𝐹44 ⎠ ⎝0 0 0⎠
And convexity requires the eigenvalues of F to be non-negative:
𝐹11 + 𝐹12 ≥ 0⎫ 2
}
𝐹11 − 𝐹12 ≥ 0⎬ ⇒ {4σs ≥ σt σc . (22.187.14)
𝐹44 ≥ 0 } −𝐹0 ≥ 0

The 3D case
In the full 3D case, the convexity condition is generally more stringent. Again we
require the eigenvalues of F to be non-negative, where F is now the full 6 by 6 matrix:
𝐹11 + 2𝐹12 ≥ 0⎫ 2
}
𝐹11 − 𝐹12 ≥ 0 ⎬ ⇒ {3σs ≥ σt σc . (22.187.15)
𝐹44 ≥ 0 } −𝐹0 ≥ 0

LS-DYNA DEV 10/27/16 (r:8004) 20-291 (Material Models)


Material Models LS-DYNA Theory Manual

Leading to
√σt σc √σt σc
σs ≥ > . (22.187.16)
√3 2

Alternatively a yield surface containing a linear rather than a quadratic term was
implemented in SAMP-1.
𝑓 = σvm − 𝐴0 − 𝐴1 𝑝 − 𝐴2 𝑝2 ≤ 0. (22.187.17)
As it will be difficult in general to guarantee a reasonable flow behaviour from three
independent measurements in shear, tension and compression, a simplified flow rule has
been implemented as the default in SAMP-1. The generally non-associated flow surface is
given as:
2
𝑔 = σvm + α𝑝2 . (22.187.18)
This flow rule is associated if:
𝐴1 = 0,
(22.187.19)
𝐴2 = −α (= cte).
And clearly leads to a constant value for the plastic Poisson ratio:
9 − 2α 9 1 − 2ν𝑝
ν𝑝 = ⇒ α= . (22.187.20)
18 + 2α 2 1 + ν𝑝
Plausible flow behaviour just means that:
9
0≤α≤ ⇒ 0 ≤ ν𝑝 ≤ 0.5. (22.187.21)
2
In SAMP-1 the value of the plastic Poisson coefficient is given by the user, either as a
constant or as a load curve in function of the uniaxial plastic strain. This allows adjusting
the flow rule of the material to measurements of transversal deformation during uniaxial
tensile or compressive testing. This can be important for plastics since often a non-
isochoric behaviour is measured.

20-292 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

Figure 22.2. Influence of the flow rule on the plastic Poisson ratio

The possible values for the plastic Poisson ratio and the resulting flow behaviour are
illustrated in Figure 22.2.

In SAMP-1 the formulation is slightly modified and based on a flow rule given as:

𝑔′ = √σvm
2 + α𝑝2 . (22.187.22)

The plastic strain rate computation is not normalized:


∂𝑔′
ε̇p = λ̇ . (22.187.23)
∂σ
The volumetric and deviatoric plastic strain rates in this case are given as :
λ̇ (−2α𝑝)
ε̇vp = λ̇ (−2α𝑝)⁄2𝑔′ = ,
2 + 4α𝑝2
√4σvm
(22.187.24)
λ̇ 3s
ε̇dp = λ̇ 3s⁄2𝑔′ = ,
2
√4σvm + 4α𝑝2
which amounts to a different definition of the plastic consistency parameter which of
course has to be considered when equivalent plastic strain values are computed.

22.187.2 Hardening formulation


The hardening formulation is the attractive part of SAMP-1. The formulation is fully
tabulated and consequently the user can directly input measurement results from uniaxial
tension, uniaxial compression and simple shear tests in terms of load curves giving the
yield stress as a function of the corresponding plastic strain. No fitting of coefficients is
required. The test results that are reflected in the load curves will be used exactly by
SAMP-1 without fitting to any analytical expression. Consequently the hardening will be
dependent upon the state of stress and not only upon the plastic strain.

LS-DYNA DEV 10/27/16 (r:8004) 20-293 (Material Models)


Material Models LS-DYNA Theory Manual

Figure 22.3. Tensile hardening curve from dynamic tensile tests

22.187.3 Rate effects


Plastics are usually highly rate dependent. A proper viscoplastic consideration of
the rate effects is therefore important in the numerical treatment of the material law. Data
to determine the rate dependency are based on uniaxial dynamic testing. If dynamic tests
are available, then the load curve defining the yield stress in uniaxial tension is simply
replaced by a table definition containing multiple load curves corresponding to different
values of the plastic strain rate. This is illustrated in the Figure 22.3.

22.187.4 Damage and failure


Numerous damage models can be found in the literature. Probably the simplest
concept is elastic damage where the damage parameter (usually written as 𝑑) is a function
of the elastic energy and effectively reduces the elastic modulus of the material. In the case
of ductile damage, 𝑑 is a function of plastic straining and affects the yield stress rather than
the elastic modulus. This is equivalent to plastic softening. In more sophisticated damage
models, d depends on both the plastic straining and the elastic energy (and maybe other
factors) and effects yield stress as well as elastic modulus.

A simple damage model was added to the SAMP-1 material law where the damage
parameter d is a function of plastic strain only. A load curve must be provided by the user
giving d as a function of the (true) plastic strain under uniaxial tension. The value of the
critical damage Dc leading to rupture is then the only other required additional input. The
implemented damage model is isotropic.

The implemented model then uses the notion of effective cross section, which is the
true cross section of the material minus the cracks that have developed. We will use the
following notation:
𝐴0 → undeformed cross section

20-294 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Material Models

Figure 22.4. Damage parameter from uniaxial tensile test

𝐴 → deformed or current cross section


𝐴0 → undeformed cross section

We define the effective stress as the force divided by the effective cross section:
𝑓
σ= ,
𝐴
(22.25)
𝑓 𝑓 σ
σeff = = = ,
𝐴eff 𝐴(1 − 𝑑) 1 − 𝑑
which allows defining an effective yield stress:
σy
σy,eff = . (22.26)
1−𝑑

LS-DYNA DEV 10/27/16 (r:8004) 20-295 (Material Models)


Material Models LS-DYNA Theory Manual

22.196 Material Model 196: General Spring Discrete Beam


If TYPE = 0, elastic behavior is obtained. In this case, if the linear spring stiffness
is used, the force, 𝐹, is given by:
𝐹 = 𝐹0 + 𝐾Δ𝐿 + 𝐷Δ𝐿̇, (22.196.1)
but if the load curve ID is specified, the force is then given by:

⎧ ̇ ⎫
𝐹 = 𝐹0 + 𝐾 𝑓 (Δ𝐿) ⎡ ⎛max {1. , ∣Δ𝐿∣ }⎟
⎢1 + 𝐶1 ⋅ Δ𝐿̇ + 𝐶2 ⋅ sgn(Δ𝐿̇)ln ⎜ ⎞⎤
⎥ + 𝐷Δ𝐿̇

{ 𝐷𝐿𝐸 ⎬
}
⎣ ⎝ ⎩ ⎭⎠⎦ (22.196.2)
+ 𝑔(Δ𝐿)ℎ(Δ𝐿̇).

In these equations, Δ𝐿 is the change in length


Δ𝐿 = current length − initial length. (22.196.3)

If TYPE = 1, inelastic behavior is obtained. In this case, the yield force is taken
from the load curve:
𝐹Y = 𝐹y (Δ𝐿plastic ), (22.196.4)

where 𝐿plastic is the plastic deflection. A trial force is computed as:


𝐹T = 𝐹n + KΔ𝐿̇Δ𝑡, (22.196.5)
and is checked against the yield force to determine F:
Y T Y
𝐹 = {𝐹T if 𝐹T > 𝐹Y . (22.196.6)
𝐹 if 𝐹 ≤ 𝐹

The final force, which includes rate effects and damping, is given by:



{ ∣Δ𝐿̇∣ ⎫
}⎞⎤
𝐹𝑛+1 = 𝐹 ⋅ ⎡
⎢ 1 + 𝐶1 ⋅ Δ𝐿̇ + 𝐶2 ⋅ sgn(Δ𝐿̇ )ln ⎜ max ⎨ 1. , ⎬ ⎟⎥ + 𝐷Δ𝐿̇
⎣ ⎝ {
⎩ 𝐷𝐿𝐸 }
⎭⎠⎦ (22.196.7)
+ 𝑔(Δ𝐿)ℎ(Δ𝐿̇).
Unless the origin of the curve starts at (0,0), the negative part of the curve is used when
the spring force is negative where the negative of the plastic displacement is used to
interpolate, 𝐹y . The positive part of the curve is used whenever the force is positive.

The cross sectional area is defined on the section card for the discrete beam
elements, See *SECTION_BEAM. The square root of this area is used as the contact
thickness offset if these elements are included in the contact treatment.

20-296 (Material Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Equation of State Models

23
Equation of State Models

LS-DYNA has 10 equation of state models which are described in this section.

1. Linear Polynomial
2. JWL High Explosive
3. Sack “Tuesday” High Explosive
4. Gruneisen
5. Ratio of Polynomials
6. Linear Polynomial With Energy Deposition
7. Ignition and Growth of Reaction in High Explosives
8. Tabulated Compaction
9. Tabulated
10. Propellant-Deflagration

The forms of the first five equations of state are given in the KOVEC user’s manual
[Woodruff 1973] as well as below.

23.1 Equation of State Form 1: Linear Polynomial


This polynomial equation of state, linear in the internal energy per initial volume, 𝐸,
is given by
𝑝 = 𝐶0 + 𝐶1 𝜇 + 𝐶2 𝜇2 + 𝐶3 𝜇3 + (𝐶4 + 𝐶5 𝜇 + 𝐶6 𝜇2 )𝐸 (23.1.1)
Here 𝐶0 , 𝐶1 , 𝐶2 , 𝐶3 , 𝐶4 , 𝐶5 and 𝐶6 are user defined constants and
1
𝜇= − 1. (23.1.2)
𝑉
where 𝑉 is the relative volume. In expanded elements, the coefficients of 𝜇2 are set to zero,
i.e.,
𝐶2 = 𝐶6 = 0. (23.1.3)
LS-DYNA DEV 10/27/16 (r:8004) 21-1 (Equation of State Models)
Equation of State Models LS-DYNA Theory Manual

The linear polynomial equation of state may be used to model gas with the gamma
law equation of state. This may be achieved by setting:
𝐶0 = 𝐶1 = 𝐶2 = 𝐶3 = 𝐶6 = 0, (23.1.4)
and
𝐶4 = 𝐶5 = 𝛾 − 1, (23.1.5)
where 𝛾 is the ratio of specific heats. The pressure is then given by:
𝜌
𝑝 = (𝛾 − 1) 𝐸. (23.1.6)
𝜌0
Note that the units of 𝐸 are the units of pressure.

23.2 Equation of State Form 2: JWL High Explosive


The JWL equation of state defines pressure as a function of relative volume, 𝑉, and
internal energy per initial volume, 𝐸, as
𝜔 𝜔 𝜔𝐸
𝑝 = 𝐴 (1 − ) 𝑒−𝑅 1𝑉 + 𝐵 (1 − ) 𝑒−𝑅 2𝑉 + , (23.2.7)
𝑅 1𝑉 𝑅 2𝑉 𝑉
where 𝜔, A, 𝐵, 𝑅1 and 𝑅2 are user defined input parameters. The JWL equation of state is
used for determining the pressure of the detonation products of high explosives in
applications involving metal accelerations. Input parameters for this equation are given by
Dobratz [1981] for a variety of high explosive materials.

This equation of state is used with the explosive burn (material model 8) material
model which determines the lighting time for the explosive element.

23.3 Equation of State Form 3: Sack “Tuesday” High


Explosives
Pressure of detonation products is given in terms of the relative volume, 𝑉, and
internal energy per initial volume, 𝐸, as [Woodruff 1973]:
𝐴3 𝐵1 𝐵
𝑝= 𝐴1
𝑒−𝐴 2𝑉 (1 − ) + 2 𝐸, (23.3.8)
𝑉 𝑉 𝑉
where 𝐴1 , 𝐴2 , 𝐴3 , 𝐵1 and 𝐵2 are user-defined input parameters.

21-2 (Equation of State Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Equation of State Models

This equation of state is used with the explosive burn (material model 8) material
model which determines the lighting time for the explosive element.

23.4 Equation of State Form 4: Gruneisen


The Gruneisen equation of state with cubic shock velocity-particle velocity defines
pressure for compressed material as
𝛾
𝜌0 𝐶2 𝜇[1 + (1 − 0 )𝜇 − 𝑎 𝜇 2 ]
𝑝= 2 2 + (𝛾0 + 𝛼𝜇)𝐸,
𝜇2 𝜇3 (23.4.9)
[1 − (𝑆1 − 1)𝜇 − 𝑠2 − 𝑠3 ]
𝜇+1 (𝜇 + 1)2
where 𝐸 is the internal energy per initial volume, 𝐶 is the intercept of the 𝑢s − 𝑢p curve, 𝑆1 ,
𝑆2 , and 𝑆3 are the coefficients of the slope of the 𝑢s − 𝑢p curve, 𝛾0 is the Gruneisen gamma,
and a is the first order volume correction to 𝛾0 . Constants 𝐶, 𝑆1 , 𝑆2 , 𝑆3 , 𝛾0 and 𝑎 are user
defined input parameters. The compression is defined in terms of the relative volume, 𝑉,
as:
1
𝜇= − 1. (23.4.10)
𝑉

For expanded materials as the pressure is defined by:


𝑝 = 𝜌0 𝐶 2 𝜇 + (𝛾0 + 𝛼𝜇)𝐸. (23.4.11)

23.5 Equation of State Form 5: Ratio of Polynomials


The ratio of polynomials equation of state defines the pressure as
𝐹1 + 𝐹2 𝐸 + 𝐹3 𝐸2 + 𝐹4 𝐸3
𝑝= (1 + 𝛼𝜇), (23.5.12)
𝐹5 + 𝐹6 𝐸 + 𝐹7 𝐸2
where
𝑛
𝐹𝑖 = ∑ 𝐴𝑖𝑗 𝑚𝑗 , 𝑛 = 4 if 𝑖 < 3, 𝑛 = 3 if 𝑖 ≥ 3 (23.5.13)
𝑗= 0

𝜌
𝜇= . (23.5.14)
𝜌0 − 1

LS-DYNA DEV 10/27/16 (r:8004) 21-3 (Equation of State Models)


Equation of State Models LS-DYNA Theory Manual

In expanded zoned 𝐹1 is replaced by 𝐹′1 = 𝐹1 + 𝛽𝜇2 Constants 𝐴𝑖𝑗 , 𝛼, and 𝛽 are user
input.

23.6 Equation of State Form 6: Linear With Energy


Deposition
This polynomial equation of state, linear in the internal energy per initial volume, 𝐸,
is given by
𝑝 = 𝐶0 + 𝐶1 𝜇 + 𝐶2 𝜇2 + 𝐶3 𝜇3 + (𝐶4 + 𝐶5 𝜇 + 𝐶6 𝜇2 )𝐸, (23.6.15)
Here 𝐶0 , 𝐶1 , 𝐶2 , 𝐶3 , 𝐶4 , 𝐶5 and 𝐶6 are user defined constants and
1
𝜇= − 1, (23.6.16)
𝑉
where 𝑉 is the relative volume. In expanded elements, we set the coefficients of 𝜇2 to zero,
i.e.,
𝐶2 = 𝐶6 = 0. (23.6.17)

Internal energy, 𝐸, is increased according to an energy deposition rate versus time


curve whose ID is defined in the input.

23.7 Equation of State Form 7: Ignition and Growth Model


A JWL equation of state defines the pressure in the unreacted high explosive as
𝜔𝑒 𝜔𝑒 𝜔𝐸
𝑃𝑒 = 𝐴𝑒 (1 − ) 𝑒−𝑅1𝑒𝑉𝑒 + 𝐵𝑒 (1 − ) 𝑒−𝑅2 𝑒𝑉𝑒 + 𝑒 , (23.7.18)
𝑅1𝑒 𝑉𝑒 𝑅2 𝑒 𝑉𝑒 𝑉𝑒
where 𝑉𝑒 is the relative volume, 𝐸𝑒 is the internal energy, and the constants 𝐴𝑒 , 𝐵𝑒 , 𝜔𝑒 , 𝑅1𝑒
and 𝑅2𝑒 are input constants. Similarly, the pressure in the reaction products is defined by
another JWL form
𝜔𝑝 −𝑅1𝑝 𝑉𝑝 𝜔𝑒 −𝑅2 𝑝 𝑉𝑝
𝜔 𝑝𝐸
𝑃𝑝 = 𝐴𝑝 (1 − )𝑒 + 𝐵𝑝 (1 − )𝑒 + . (23.7.19)
𝑅1𝑝 𝑉𝑝 𝑅2 𝑝 𝑉𝑝 𝑉𝑝

The mixture of unreacted explosive and reaction products is defined by the fraction
reacted 𝐹 (𝐹 = 0 implies no reaction, 𝐹 = 1 implies complete conversion from explosive to

21-4 (Equation of State Models) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Equation of State Models

products). The pressures and temperature are assumed to be in equilibrium, and the
relative volumes are assumed to be additive:
𝑉 = (1 − 𝐹)𝑉𝑒 + 𝐹𝑉𝑝 . (23.7.20)

The rate of reaction is defined as


∂𝐹 3 𝑚
= 𝐼(FCRIT − 𝐹)𝑦 (𝑉𝑒−1 − 1) [1 + 𝐺(𝑉 −1 𝑦 𝑥 𝑧 −1
𝑒 − 1)] + 𝐻(1 − 𝐹) 𝐹 𝑃 (𝑉𝑝 − 1) , (23.7.21)
∂𝑡
where 𝐼, 𝐺, 𝐻, 𝑥, 𝑦, 𝑧 and 𝑚 (generally 𝑚 = 0) are input constants.

The JWL equations of state and the reaction rates have been fitted to one- and two-
dimensional shock initiation and detonation data for four explosives: PBX-9404, RX-03-BB,
PETN, and cast TNT. The details of calculational method are described by Cochran and
Chan [1979]. The detailed one-dimensional calculations and parameters for the four
explosives are given by Lee and Tarver [1980]. Two-dimensional calculations with this
model for PBX 9404 and LX-17 are discussed by Tarver and Hallquist [1981].

23.8 Equation of State Form 8: Tabulated Compaction


Pressure is positive in compression, and volumetric strain 𝜀𝑉 is positive in tension.
The tabulated compaction model is linear in internal energy per unit volume. Pressure is
defined by
𝑝 = 𝐶(𝜀𝑉 ) + 𝛾𝑇(𝜀𝑉 )𝐸, (23.8.22)
during loading (compression). Unloading occurs at a slope corresponding to the bulk
modulus at the peak (most compressive) volumetric strain, as shown in Figure 23.1.
Reloading follows the unloading path to the point where unloading began, and then
continues on the loading path described by Equation (23.8.22).

23.9 Equation of State Form 9: Tabulated


The tabulated equation of state model is linear in internal energy. Pressure is
defined by
𝑝 = 𝐶(𝜀𝑉 ) + 𝛾𝑇(𝜀𝑉 )𝐸, (23.9.23)
The volumetric strain 𝜀𝑉 is given by the natural algorithm of the relative volume. Up to 10
points and as few as 2 may be used when defining the tabulated functions. The pressure is
extrapolated if necessary. Loading and unloading are along the same curve unlike
equation of state form 8.

LS-DYNA DEV 10/27/16 (r:8004) 21-5 (Equation of State Models)


Equation of State Models LS-DYNA Theory Manual

C
B
Pressure P A

kεAV kεVB kεCV

εAV εVB εCV

Volumetric Strain (-εV)


Figure 23.1. Pressure versus volumetric strain curve for equation of state form 8
with compaction. In the compacted states, the bulk unloading modulus depends
on the peak volumetric strain.

23.10 Equation of State Form 10: Propellant-Deflagration


A deflagration (burn rate) reactive flow model requires an unreacted solid equation-
of-state, a reaction product equation-of-state, a reaction rate law and a mixture rule for the
two (or more) species. The mixture rule for the standard ignition and growth model [Lee
and Tarver 1980] assumes that both pressures and temperatures are completely
equilibrated as the reaction proceeds. However, the mixture rule can be modified to allow
no thermal conduction or partial heating of the solid by the reaction product gases. For this
relatively slow process of airbag propellant burn, the thermal and pressure equilibrium
assumptions are valid. The equations-of-state currently used in the burn model are the
JWL, Gruneisen, the van der Waals co-volume, and the perfect gas law, but other
equations-of-state can be easily implemented. In this propellant burn, the gaseous nitrogen
produced by the burning sodium azide obeys the perfect gas law as it fills the airbag but
may have to be modeled as a van der Waal’s gas at the high pressures and temperatures
produced in the propellant chamber. The chemical reaction rate law is pressure, particle
geometry and surface area dependant, as are most high-pressure burn processes. When the
temperature profile of the reacting system is well known, temperature dependent
Arrhenius chemical kinetics can be used.
Since the airbag propellant composition and performance data are company private
information, it is very difficult to obtain the required information for burn rate modeling.
However, Imperial Chemical Industries (ICI) Corporation supplied pressure exponent,
21-6 (Equation of State Models) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Equation of State Models

particle geometry, packing density, heat of reaction, and atmospheric pressure burn rate
data which allowed us to develop the numerical model presented here for their NaN3 +
Fe2 O3 driver airbag propellant. The deflagration model, its implementation, and the
results for the ICI propellant are presented in the are described by [Hallquist, et. al., 1990].

The unreacted propellant and the reaction product equations-of-state are both of the
form:
𝜔𝐶v 𝑇
𝑝 = 𝐴 𝑒−𝑅1𝑉 + 𝐵𝑒−𝑅2𝑉 + , (23.10.24)
𝑉−𝑑
where 𝑝 is pressure (in Mbars), 𝑉 is the relative specific volume (inverse of relative
density), 𝜔 is the Gruneisen coefficient, 𝐶v is heat capacity (in Mbars -cc/cc°K), 𝑇 is
temperature in °𝐾, 𝑑 is the co-volume, and 𝐴, 𝐵, 𝑅1 and 𝑅2 are constants. Setting𝐴 = 𝐵 = 0
yields the van der Waal’s co-volume equation-of-state. The JWL equation-of-state is
generally useful at pressures above several kilobars, while the van der Waal’s is useful at
pressures below that range and above the range for which the perfect gas law holds. Of
course, setting 𝐴 = 𝐵 = 𝑑 = 0 yields the perfect gas law. If accurate values of 𝜔 and 𝐶v
plus the correct distribution between “cold” compression and internal energies are used,
the calculated temperatures are very reasonable and thus can be used to check propellant
performance.

The reaction rate used for the propellant deflagration process is of the form:
∂𝐹
= 𝑍(1 − 𝐹)𝑦 𝐹𝑥 𝑝𝑤 + 𝑉(1 − 𝐹)𝑢 𝐹𝑟𝑝𝑠
∂𝑡
(23.10.25)
for 0 < 𝐹 < 𝐹limit1
for 𝐹limit2 < 𝐹 < 1
where 𝐹 is the fraction reacted (𝐹 = 0 implies no reaction, 𝐹 = 1 is complete reaction), 𝑡 is
time, and 𝑝 is pressure (in Mbars), 𝑟, 𝑠, 𝑢, 𝑤, 𝑥, 𝑦, 𝐹limit1 and 𝐹limit2 are constants used to
describe the pressure dependence and surface area dependence of the reaction rates. Two
(or more) pressure dependent reaction rates are included in case the propellant is a mixture
or exhibited a sharp change in reaction rate at some pressure or temperature. Burning
surface area dependences can be approximated using the (1 − 𝐹)𝑦 𝐹𝑥 terms. Other forms of
the reaction rate law, such as Arrhenius temperature dependent 𝑒−𝐸/𝑅𝑇 type rates, can be
used, but these require very accurate temperatures calculations. Although the theoretical
justification of pressure dependent burn rates at kilobar type pressures is not complete, a
vast amount of experimental burn rate versus pressure data does demonstrate this effect
and hydrodynamic calculations using pressure dependent burn accurately simulate such
experiments.

The deflagration reactive flow model is activated by any pressure or particle velocity
increase on one or more zone boundaries in the reactive material. Such an increase creates
pressure in those zones and the decomposition begins. If the pressure is relieved, the
reaction rate decreases and can go to zero. This feature is important for short duration,

LS-DYNA DEV 10/27/16 (r:8004) 21-7 (Equation of State Models)


Equation of State Models LS-DYNA Theory Manual

partial decomposition reactions. If the pressure is maintained, the fraction reacted


eventually reaches one and the material is completely converted to product molecules. The
deflagration front rates of advance through the propellant calculated by this model for
several propellants are quite close to the experimentally observed burn rate versus pressure
curves.

To obtain good agreement with experimental deflagration data, the model requires
an accurate description of the unreacted propellant equation-of-state, either an analytical fit
to experimental compression data or an estimated fit based on previous experience with
similar materials. This is also true for the reaction products equation-of-state. The more
experimental burn rate, pressure production and energy delivery data available, the better
the form and constants in the reaction rate equation can be determined.

Therefore the equations used in the burn subroutine for the pressure in the
unreacted propellant
𝑅3 ⋅ 𝑇𝑢
𝑃𝑢 = 𝑅1 ⋅ 𝑒−𝑅5⋅𝑉𝑢 + 𝑅2 ⋅ 𝑒−𝑅6⋅𝑉𝑢 + , (23.10.26)
𝑉𝑢 − FRER
where 𝑉𝑢 and 𝑇𝑢 are the relative volume and temperature respectively of the unreacted
propellant. The relative density is obviously the inverse of the relative volume. The
pressure 𝑃p in the reaction products is given by:

−𝑋𝑃1⋅𝑉𝑝 −𝑋𝑃2⋅𝑉𝑝
𝐺 ⋅ 𝑇𝑝
𝑃p = 𝐴 ⋅ 𝑒 +𝐵⋅𝑒 + . (23.10.27)
𝑉𝑝 − CCRIT
As the reaction proceeds, the unreacted and product pressures and temperatures are
assumed to be equilibrated (𝑇𝑢 = 𝑇𝑝 = 𝑇, 𝑝 = 𝑃𝑢 = 𝑃𝑝 ) and the relative volumes are
additive:
𝑉 = (1 − 𝐹) ⋅ 𝑉𝑢 + 𝐹 ⋅ 𝑉𝑝 (23.10.28)
where 𝑉 is the total relative volume. Other mixture assumptions can and have been used
in different versions of DYNA2D/3D. The reaction rate law has the form:
∂𝐹
= GROW1(𝑝 + FREQ)𝑒𝑚 (𝐹 + FMXIG)𝑎𝑟1 (1 − 𝐹 + FMXIG)𝑒𝑠1
∂𝑡 (23.10.29)
+GROW2(𝑝 + FREQ)𝑒𝑛 (𝐹 + FMXIG)

If 𝐹 exceeds FMXGR, the GROW1 term is set equal to zero, and, if 𝐹 is less
thanFMNGR, the GROW2 term is zero. Thus, two separate (or overlapping) burn rates can
be used to describe the rate at which the propellant decomposes.

This equation-of-state subroutine is used together with a material model to describe


the propellant. In the airbag propellant case, a null material model (type #10) can be used.
Material type #10 is usually used for a solid propellant or explosive when the shear
modulus and yield strength are defined. The propellant material is defined by the material
model and the unreacted equation-of-state until the reaction begins. The calculated
21-8 (Equation of State Models) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Equation of State Models

mixture states are used until the reaction is complete and then the reaction product
equation-of-state is used. The heat of reaction, ENQ, is assumed to be a constant and the
same at all values of 𝐹 but more complex energy release laws could be implemented.

LS-DYNA DEV 10/27/16 (r:8004) 21-9 (Equation of State Models)


LS-DYNA Theory Manual Artificial Bulk Viscosity

24
Artificial Bulk Viscosity

Bulk viscosity is used to treat shock waves. Proposed in one spatial dimension by
von Neumann and Richtmyer [1950], the bulk viscosity method is now used in nearly all
wave propagation codes. A viscous term q is added to the pressure to smear the shock
discontinuities into rapidly varying but continuous transition regions. With this method
the solution is unperturbed away from a shock, the Hugoniot jump conditions remain valid
across the shock transition, and shocks are treated automatically. In our discussion of bulk
viscosity we draw heavily on works by Richtmyer and Morton [1967], Noh [1976], and
Wilkins [1980]. The following discussion of the bulk viscosity applies to solid elements
since strong shocks are not normally encountered in structures modeled with shell and
beam elements.

24.1 Shock Waves


Shock waves result from the property that sound speed increases with increasing
pressure. A smooth pressure wave can gradually steepen until it propagates as a
discontinuous disturbance called a shock. See Figure 24.1. Shocks lead to jumps in
pressure, density, particle velocity, and energy.

Consider a planar shock front moving through a material. Mass, momentum, and
energy are conserved across the front. Application of these conservation laws leads to the
well-known Rankine-Hugoniot jump conditions
𝜌(𝑢 − 𝑢0 )
𝑢𝑠 = , (24.1)
𝜌 − 𝜌0

𝜌 − 𝜌0 = 𝜌0 𝑢𝑠 (𝑢 − 𝑢0 ), (24.2)
𝑝 − 𝑝0 𝜌 − 𝜌0
𝑒 − 𝑒0 = , (24.3)
2 𝜌0 𝜌

LS-DYNA DEV 10/27/16 (r:8004) 22-1 (Artificial Bulk Viscosity)


Artificial Bulk Viscosity LS-DYNA Theory Manual

σ
ε

x
Figure 24.1. If the sound speed increases as the stress increases the traveling
wave above will gradually steepen as it moves along the x-coordinate to form a
shock wave.

where Equation (24.3) is an expression of the energy jump condition using the results of
mass conservation, Equation (24.1), and momentum conservation, Equation (24.2). Here,𝑢𝑠
is the shock velocity, 𝑢 is the particle velocity, 𝜌 is the density, 𝑒 is the specific internal
energy, 𝑝 is the pressure, and the subscript, 0 , indicates the state ahead of the shock.

The energy equation relating the thermodynamic quantities density, pressure, and
energy must be satisfied for all shocks. The equation of state
𝑝 = 𝑝(𝜌, 𝑒), (24.4)
which defines all equilibrium states that can exist in a material and relating the same
quantities as the energy equation, must also be satisfied. We may use this equation to
eliminate energy from
Equation (24.3) and obtain a unique relationship between pressure and compression. This
relation, called the Hugoniot, determines all pressure-compression states achievable behind
the shock. Shocking takes place along the Rayleigh line and not the Hugoniot (Figure 24.1)
and because the Hugoniot curve closely approximates an isentrope, we may often assume
the unloading follows the Hugoniot. Combining Equations (24.1) and (24.2), we see that
the slope of the Rayleigh line is related to the shock speed:
1⁄
2
1 ⎡𝑝1 − 𝑝0 ⎤
𝑢𝑠 = ⎢ ⎥ . (24.5)
𝜌0 ⎢ 1 − 1 ⎥
⎣ 𝜌0 𝜌 ⎦
For the material of Figure 24.2, increasing pressure increases shock speed.

Consider a 𝛾-law gas with the equation of state

22-2 (Artificial Bulk Viscosity) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Artificial Bulk Viscosity

p1

p0

Figure 24.2. Shocking takes place along the Rayleigh line, and release closely
follows the Hugoniot. The cross-hatched area is the difference between the
internal energy behind the shock and the internal energy lost on release.

𝑝 = (𝛾 − 1)𝜌𝑒, (24.6)
where 𝛾 is the ratio of specific heats. Using the energy jump condition, we can eliminate e
and obtain the Hugoniot
𝑝 2𝑉 + (𝛾 − 1)(𝑉0 − 𝑉)
= 𝑝∗ = 0 , (24.7)
𝑝0 2𝑉 − (𝛾 − 1)(𝑉0 − 𝑉)
where 𝑉 is the relative volume. Figure 24.3 shows a plot of the Hugoniot and adiabat
where it is noted that for 𝑝∗ = 1, the slopes are equal. Thus for weak shocks, Hugoniot and
adiabat agree to the first order and can be ignored in numerical calculations. However,
special treatment is required for strong shocks, and in numerical calculations this special
treatment takes the form of bulk viscosity.

24.2 Bulk Viscosity


In the presence of shocks, the governing partial differential equations can give
multiple weak solutions. In their discussion of the Rankine-Hugoniot jump conditions,
Richtmyer and Morton [1967] report that the unmodified finite difference (element)
equations often will not produce even approximately correct answers. One possible
solution is to employ shock fitting techniques and treat the shocks as interior boundary
conditions. This technique has been used in one spatial dimension but is too complex to

LS-DYNA DEV 10/27/16 (r:8004) 22-3 (Artificial Bulk Viscosity)


Artificial Bulk Viscosity LS-DYNA Theory Manual

Limiting compression

Hugoniot

Adiabat

Slopes of Hugoniot and adiabat


are equat at

Figure 24.3. Hugoniot curve and adiabat for a g-law gas (from [Noh 1976]).

extend to multi-dimensional problems of arbitrary geometry. For these reasons the


pseudo-viscosity method was a major breakthrough for computational mechanics.

The viscosity proposed by von Neumann and Richtmyer [1950] in one spatial
dimension has the form
2
2 ∂𝑥̇ ∂𝑥̇
𝑞 = 𝐶0 𝜌(Δ𝑥) ( ) if <0
∂𝑥 ∂𝑥 (24.8)
∂𝑥̇
𝑞=0 if ≥0
∂𝑥
where 𝐶0 is a dimensionless constant and 𝑞 is added to the pressure in both the momentum
and energy equations. When 𝑞 is used, they proved the following for steady state shocks:
the hydrodynamic equations possess solutions without discontinuities; the shock thickness
is independent of shock strength and of the same order as the Δ𝑥 used in the calculations;
the q term is insignificant outside the shock layer; and the jump conditions are satisfied.
According to Noh, it is generally believed that these properties: “hold for all shocks, and
this has been borne out over the years by countless numerical experiments in which
excellent agreement has been obtained either with exact solutions or with hydrodynamical
experiments.”

In 1955, Landshoff [1955] suggested the addition of a linear term to the 𝑞 of von
Neumann and Richtmyer leading to a 𝑞 of the form
2
2 𝜕𝑥̇ 𝜕𝑥̇
𝑞 = 𝐶0 𝜌(Δ𝑥) ( ) if <0
𝜕𝑥 𝜕𝑥 (24.9)
2
2 ∂ẋ 𝜕𝑥̇
𝑞 = 𝐶0 𝜌(Δ𝑥) ( ) if ≥ 0,
∂x 𝜕𝑥
where 𝐶1 is a dimensionless constant and 𝑎 is the local sound speed. The linear term
rapidly damps numerical oscillations behind the shock front (Figure 24.3). A similar form
was proposed independently by Noh about the same time.
22-4 (Artificial Bulk Viscosity) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Artificial Bulk Viscosity

In an interesting aside, Wilkins [1980] discusses work by Kuropatenko who, given


an equation of state, derived a q by solving the jump conditions for pressure in terms of a
change in the particle velocity, Δ𝑢. For an equation of state of the form,
𝜌
𝑝 = 𝐾 (− − 1), (24.10)
𝜌0
pressure across the shock front is given by [Wilkins 1980]
1⁄
(Δ𝑢)2 (Δ𝑢)2 2
(24.11)
𝑝 = 𝑝0 + 𝜌 0 + 𝜌0 |Δ𝑢| [ + 𝑎 2] ,
2 2
where 𝑎 is a sound speed
1
𝐾 ⁄2
𝑎=( ) . (24.12)
𝜌0

For a strong shock, Δ𝑢2 >> 𝑎2 , we obtain the quadratic form


𝑞 = 𝜌0 Δ𝑢2 , (24.13)
and for a weak shock, Δ𝑢2 << 𝑎2 , the linear form
𝑞 = 𝜌0 𝑎Δ𝑢, (24.14)
Thus linear and quadratic forms for 𝑞 can be naturally derived. According to Wilkins, the
particular expressions for 𝑞 obtained by Kuropatenko offer no particular advantage over
the expressions currently used in most computer programs.

In extending the one-dimensional viscosity formulations to multi-dimensions, most


code developers have simply replaced the divergence of the velocity with, 𝜀̇𝑘𝑘 , the trace of
the strain rate tensor, and the characteristic length with the square root of the area A, in
two dimensions and the cubic root of the volume v in three dimensions. These changes
also give the default viscosities in the LS-DYNA codes:
𝑞 = 𝜌 𝑙(𝐶0 𝑙𝜀̇2𝑘𝑘 − 𝐶𝑙 𝑎𝜀̇𝑘𝑘 ) if 𝜀̇𝑘𝑘 < 0
(24.15)
𝑞=0 if 𝜀̇𝑘𝑘 ≥ 0
where 𝐶0 and 𝐶1 are dimensionless constants which default to 1.5 and 0.06, respectively,
where 1 = √𝐴 in 2D, and √
3
𝑣 in 3D, a is the local sound speed, 𝐶0 defaults to 1.5 and 𝐶1
defaults to 0.06.

In converging two- and three-dimensional geometries, the strain rate 𝜀̇𝑘𝑘 is negative
and the 𝑞 term in Equation (24.15) is nonzero, even though no shocks may be generated.
This results in nonphysical 𝑞 heating. When the aspect ratios of the elements are poor (far
from unity), the use of a characteristic length based on √𝐴 or √ 3
𝑣 can also result in
nonphysical 𝑞 heating and even occasional numerical instabilities. Wilkins uses a bulk
viscosity that is based in part on earlier work by Richards [1965] that extends the von
LS-DYNA DEV 10/27/16 (r:8004) 22-5 (Artificial Bulk Viscosity)
Artificial Bulk Viscosity LS-DYNA Theory Manual

Neumann and Richtmyer formulations in a way that avoids these problems. This latter 𝑞
may be added in the future if the need arises.
Wilkins’ 𝑞 is defined as:
2
𝑑𝑠 𝑑𝑠
𝑞 = 𝐶0 𝜌𝑙 ( ) − 𝐶𝑙 𝜌𝑙𝑎∗
2
if 𝜀̇𝑘𝑘 < 0
𝑑𝑡 𝑑𝑡 (24.16)
𝑑𝑠
𝑞=0 if ≥0
𝑑𝑡
where 𝑙 and 𝑑𝑠
𝑑𝑡
are the thickness of the element and the strain rate in the direction of the
acceleration, respectively, and 𝑎∗ is the sound speed defined by (p/𝜌)1/2 if 𝑝 > 0. We use
the local sound speed in place of 𝑎∗ to reduce the noise at the low stress levels that are
typical of our applications.

Two disadvantages are associated with Equation (24.16). To compute the length
parameter and the strain rate, we need to know the direction of the acceleration through
the element. Since the nodal force vector becomes the acceleration vector in the explicit
integration scheme, we have to provide extra storage to save the direction. In three
dimensions our present storage capacity is marginal at best and sacrificing this storage for
storing the direction would make it even more so. Secondly, we need to compute l and 𝑑𝑠 𝑑𝑡
which results in a noticeable increase in computer cost even in two dimensions. For most
problems the additional refinement of Wilkins is not needed. However, users must be
aware of the pitfalls of Equation (24.15), i.e., when the element aspect ratios are poor or the
deformations are large, an anomalous 𝑞 may be generated.

22-6 (Artificial Bulk Viscosity) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Time Step Control

25
Time Step Control

During the solution we loop through the elements to update the stresses and the
right hand side force vector. We also determine a new time step size by taking the
minimum value over all elements.
Δ𝑡𝑛+1 = 𝑎 ⋅ min{Δ𝑡1 , Δ𝑡2 , Δ𝑡3 , . . . , Δ𝑡𝑁 }, (25.1)
where 𝑁 is the number of elements. For stability reasons the scale factor 𝑎 is typically set to
a value of .90 (default) or some smaller value.

25.1 Time Step Calculations for Solid Elements


A critical time step size, Δ𝑡𝑒 , is computed for solid elements from
𝐿𝑒
Δ𝑡𝑒 = (25.2)
{[𝑄 + (𝑄2 + 𝑐2 )1/2 ]}
where 𝑄 is a function of the bulk viscosity coefficients 𝐶0 and 𝐶1 :
𝐶1 𝑐 + 𝐶0 𝐿𝑒 |𝜀̇𝑘𝑘 | for 𝜀̇𝑘𝑘 ≤ 0
𝑄={ (25.3)
0 for 𝜀̇𝑘𝑘 > 0
𝐿𝑒 is a characteristic length:
𝜐𝑒
8 node solids: 𝐿𝑒 = 𝐴
𝑒 max
4 node tetrahedras: 𝐿𝑒 = minimum altitude
𝜐𝑒 is the element volume, 𝐴𝑒 max is the area of the largest side, and 𝑐 is the adiabatic sound
speed:
1⁄
4𝐺 ∂𝑝 2
𝑐=[ + )] , (25.4)
3𝜌0 ∂𝜌 𝑠
where 𝜌 is the specific mass density. Noting that:

22.1
Time Step Control LS-DYNA Theory Manual

∂𝑝 ∂𝑝 ∂𝑝 ∂𝑈
) = ) + ) ), (25.5)
∂𝜌 𝑠 ∂𝜌 𝑈 ∂𝑈 𝜌 ∂𝜌 𝑠
and that along an isentrope the incremental energy, 𝑈, in the units of pressure is the
product of pressure, 𝑝, and the incremental relative volume, 𝑑𝑉:
𝑑𝑈 = −𝑝𝑑𝑉, (25.6)
we obtain
1⁄
⎡ 4𝐺 ∂𝑝 𝑝𝑉 2 ∂𝑝 ⎤ 2
𝑐=⎢ + ) + ) ⎥ . (25.7)
⎣3𝜌0 ∂𝜌 𝑈 𝜌0 ∂𝑈 𝜌 ⎦

For elastic materials with a constant bulk modulus the sound speed is given by:
𝐸(1 − 𝜐)
𝑐=√ (25.8)
(1 + 𝜐)(1 − 2𝜐)𝜌
where 𝐸 is Young’s modulus, and 𝜐 is Poisson’s ratio.

25.2 Time Step Calculations for Beam and Truss Elements


For the Hughes-Liu beam and truss elements, the time step size is given by:
𝐿
Δ𝑡𝑒 = (25.9)
𝑐
where 𝐿 is the length of the element and c is the wave speed:
𝐸
𝑐=√ . (25.10)
𝜌

For the Belytschko beam the time step size given by the longitudinal sound speed is
used (Equation (25.9)), unless the bending-related time step size given by [Belytschko and
Tsay 1982]
0.5𝐿
Δ𝑡𝑒 =
3 (25.11)
𝑐√3𝐼 [ + 1 ]
12𝐼 + 𝐴𝐿2 𝐴𝐿2
is smaller, where 𝐼 and 𝐴 are the maximum value of the moment of inertia and area of the
cross section, respectively.

Comparison of critical time steps of the truss versus the elastic solid element shows
that it if Poisson's ratio, 𝜐 is nonzero the solid elements give a considerably smaller stable
time step size. If we define the ratio, 𝛼, as:

22.2
LS-DYNA Theory Manual Time Step Control

Δ𝑡continuum 𝐶rod (1 + 𝜐)(1 − 2𝜐)


𝛼= = =√ , (25.12)
Δ𝑡rod 𝐶continuum 1−𝜐
we obtain the results in Table 22.1 where we can see that as  approaches .5 𝛼 → 0.

0 0.2 0.3 0.4 0.45 0.49 0.50


1. 0.949 0.862 0.683 0.513 0.242 0.0

Table 22.1. Comparison of critical time step sizes for a truss versus a solid element.

25.3 Time Step Calculations for Shell Elements


For the shell elements, the time step size is given by:
𝐿𝑠
Δ𝑡𝑒 = (25.13)
𝑐
where 𝐿𝑠 is the characteristic length and 𝑐 is the sound speed:

𝐸
𝑐=√ . (25.14)
𝜌(1 − 𝜈 2 )

Three user options exists for choosing the characteristic length. In the default or first
option the characteristic length is given by:
(1 + 𝛽)𝐴𝑠
𝐿𝑠 = (25.15)
max(𝐿1 , 𝐿2 , 𝐿3 , (1 − 𝛽)𝐿4 )
where 𝛽 = 0 for quadrilateral and 1 for triangular shell elements, 𝐴𝑠 is the area, and𝐿𝑖 , (𝑖 =
1. . . .4) is the length of the sides defining the shell elements. In the second option a more
conservative value of 𝐿𝑠 is used:
(1 + 𝛽)𝐴𝑠
𝐿𝑠 = , (25.16)
max(𝐷1 , 𝐷2 )
where 𝐷𝑖 (𝑖 = 1,2) is the length of the diagonals. The third option provides the largest time
step size and is frequently used when triangular shell elements have very short altitudes.
The bar wave speed, Equation (21.10), is used to compute the time step size and 𝐿𝑠 is given
by
(1 + 𝛽)𝐴𝑠
𝐿𝑠 = max [ , min(𝐿1 , 𝐿2 , 𝐿3 , 𝐿4 + 𝛽1020 )]. (25.17)
max(𝐿1 , 𝐿2 , 𝐿3 , (1 − 𝛽)𝐿4 )

A comparison of critical time steps of truss versus shells is given in Table 22.2 with𝛽
defined as:
22.3
Time Step Control LS-DYNA Theory Manual

Δ𝑡2D-continuum 𝐶rod
𝛽= = = √1 − 𝜐 2 . (25.18)
Δ𝑡rod 𝐶

0 0.2 0.3 0.4 0.5


1.0 0.98 0.954 0.917 0.866

Table 22.2. Comparison of critical time step sizes for a truss versus a shell element.

25.4 Time Step Calculations for Solid Shell Elements


A critical time step size, Δ𝑡𝑒 is computed for solid shell elements from
𝜐𝑒
Δ𝑡𝑒 = , (25.19)
𝑐𝐴𝑒max
where 𝜐𝑒 is the element volume, 𝐴𝑒max is the area of the largest side, and 𝑐 is the plane stress
sound speed given in Equation (25.14).

25.5 Time Step Calculations for Discrete Elements


For spring elements such as that in Figure 25.1 there is no wave propagation speed𝑐
to calculate the critical time step size.

The eigenvalue problem for the free vibration of spring with nodal masses 𝑚1 and
𝑚2 , and stiffness, 𝑘, is given by:

[𝑘 −𝑘 ] [𝑢1 ] − 𝜔2 [𝑚1 0 𝑢
] [𝑢1 ] = [0]. (25.20)
−𝑘 𝑘 𝑢2 0 𝑚2 2 0
Since the determinant of the characteristic equation must equal zero, we can solve for the
maximum eigenvalue:
𝑘 − 𝜔 2 𝑚1 −𝑘 2 𝑘(𝑚1 + 𝑚2 )
det [ ]= 0 → 𝜔max = , (25.21)
−𝑘 𝑘 − 𝜔 2 𝑚2 𝑚1 ⋅ 𝑚2

22.4
LS-DYNA Theory Manual Time Step Control

m1 = 0.5M1 ; M1 = nodal mass

m2 = 0.5M2 ; M2 = nodal mass


Figure 25.1. Lumped spring mass system.

Recalling the critical time step of a truss element:


ℓ⎫
Δ𝑡 ≤ }
𝑐} Δ𝑡 ≤
2
, (25.22)
2𝑐 ⎬ 𝜔max
𝜔max = } }
ℓ ⎭
and approximating the spring masses by using 1/2 the actual nodal mass, we obtain:
𝑚 𝑚 1
Δ𝑡 = 2√ 1 2 . (25.23)
𝑚1 + 𝑚2 𝑘

Therefore, in terms of the nodal mass we can write the critical time step size as:
2𝑀1 𝑀2
Δ𝑡𝑒 = 2√ . (25.24)
𝑘(𝑀1 + 𝑀2 )
The springs used in the contact interface are not checked for stability.

22.5
LS-DYNA Theory Manual Boundary and Loading Conditions

26
Boundary and Loading Conditions

26.1 Pressure Boundary Conditions


Consider pressure loadings on boundary ∂b1 in Equation (2.4). To carry out the
surface integration indicated by the integral

∫ 𝑁 𝑡 𝑡𝑑𝑠, (26.1)
∂𝑏1

a Gaussian quadrature rule is used. To locate any point of the surface under consideration,
a position vector, 𝑟, is defined:
𝑟 = 𝑓1 (𝜉 , 𝜂)𝑖1 + 𝑓2 ((𝜉 , 𝜂)𝑖2 + 𝑓3 (𝜉 , 𝜂)𝑖3 , (26.2)
where
4
𝑗
𝑓𝑖 (𝜉 , 𝜂) = ∑ 𝜙𝑗 𝑥𝑖 , (26.3)
𝑗=1

and 𝑖1 , 𝑖2 , 𝑖3 are unit vectors in the 𝑥1 , 𝑥2 , 𝑥3 directions (see Figure 26.1).

Nodal quantities are interpolated over the four-node linear surface by the functions
1
𝜙𝑖 = (1 + 𝜉 𝜉𝑖 )(1 + 𝜂𝜂𝑖 ), (26.4)
4
so that the differential surface area 𝑑𝑠 may be written in terms of the curvilinear
coordinates as
𝑑𝑠 = |𝐽|𝑑𝜉𝑑𝜂, (26.5)
where |𝐽| is the surface Jacobian defined by
∂𝑟 ∂𝑟 1⁄
|𝐽| = ∣ × ∣ = (𝐸𝐺 − 𝐹2 ) 2 , (26.6)
∂𝜉 ∂𝜂
in which

LS-DYNA DEV 10/27/16 (r:8004) 24-1 (Boundary and Loading Conditions)


Boundary and Loading Conditions LS-DYNA Theory Manual

3
2
x3

ξ
x2
4
i3 r(ξ,η) 1
i2

i1 x1

Figure 26.1. Parametric representation of a surface segment.

∂𝑟 ∂𝑟
𝐸= ⋅ ,
∂𝜉 ∂𝜉
∂𝑟 ∂𝑟
𝐹= ⋅ , (26.7)
∂𝜉 ∂𝜂
∂𝑟 ∂𝑟
𝐺= ⋅ .
∂𝜂 ∂𝜂

A unit normal vector 𝐧 to the surface segment is given by


∂𝐫 ∂𝐫
𝐧 = |𝐽| −1 ( × ), (26.8)
∂𝜉 ∂𝜂
and the global components of the traction vector can now be written
4
𝑡𝑖 = 𝑛𝑖 ∑ 𝜙𝑗 𝑝𝑗 , (26.9)
𝑗=1

where 𝑝𝑗 is the applied pressure at the jth node.

The surface integral for a segment is evaluated as:


1 1
∫ ∫ 𝑁 𝑡 𝑡|𝐽|𝑑𝜉 𝑑𝜂 . (26.10)
−1 −1

One such integral is computed for each surface segment on which a pressure loading acts.
Note that the Jacobians cancel when Equations (26.8) and (26.7) are put into Equation
(26.10). Equation (26.10) is evaluated with one-point integration analogous to that
employed in the volume integrals. The area of an element side is approximated by 4|𝐽|
where |𝐽| = |𝐽(0, 0)|.

24-2 (Boundary and Loading Conditions) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Boundary and Loading Conditions

26.2 Transmitting Boundaries


Transmitting boundaries are available only for problems that require the modeling
of semi-infinite or infinite domains with solid elements and therefore are not available for
beam or shell elements. Applications of this capability include problems in geomechanics
and underwater structures.
The transmitting or silent boundary is defined by providing a complete list of
boundary segments. In the approach used, discussed by Cohen and Jennings [1983] who in
turn credit the method to Lysmer and Kuhlemeyer [1969], viscous normal shear stresses in
Equation (23.11) are applied to the boundary segments:
𝛔normal = −𝜌𝑐𝑑 𝐕normal (26.11)

𝛔shear = −𝜌𝑐𝑠 𝐕tan , (26.12)


where 𝜌, 𝑐𝑑 ,and 𝑐𝑠 are the material density, dilatational wave speed, and the shear wave
speed of the transmitting media respectively. The magnitude of these stresses is
proportional to the particle velocities in the normal, 𝐕normal , and tangential, 𝐕tan ,
directions. The material associated with each transmitting segment is identified during
initialization so that unique values of the constants 𝜌, 𝑐𝑑 , and 𝑐𝑠 can be defined
automatically.

26.3 Kinematic Boundary Conditions


In this subsection, the kinematic constraints are briefly reviewed. LS-DYNA tracks
reaction forces for each type of kinematic constraint and provides this information as
output if requested. For the prescribed boundary conditions, the input energy is integrated
and included in the external work.

26.4 Displacement Constraints


Translational and rotational boundary constraints are imposed either globally or
locally by setting the constrained acceleration components to zero. If nodal single point
constraints are employed, the constraints are imposed in a local system. The user defines
the local system by specifying a vector 𝐮1 in the direction of the local x-axis 𝐱l , and a local
in-plane vector 𝐯l . After normalizing 𝐮1 , the local 𝐱𝑙 , 𝐲𝑙 and 𝐳l axes are given by:

LS-DYNA DEV 10/27/16 (r:8004) 24-3 (Boundary and Loading Conditions)


Boundary and Loading Conditions LS-DYNA Theory Manual

𝐮𝑙
𝐱𝑙 = (26.13)
‖𝐮𝑙 ‖
𝐱𝑙 × 𝐯𝑙
𝐳𝑙 = (26.14)
‖𝐱𝑙 × 𝐯𝑙 ‖

𝐲𝑙 = 𝐳𝑙 × 𝐱𝑙 . (26.15)

A transformation matrix 𝐪 is constructed to transform the acceleration components


to the local system:
T
⎡𝐱𝑙T ⎤
𝐪=⎢
⎢𝐲𝑙 ⎥⎥, (26.16)
T
⎣𝐳𝑙 ⎦
and the nodal translational and rotational acceleration vectors 𝐚𝐼 and 𝛚̇ 𝐼 , for node I are
transformed to the local system:
𝐚𝐼1 = 𝐪𝐚𝐼 (26.17)

𝛚̇ 𝐼𝑙 = 𝐪𝛚̇ 𝐼 , (26.18)
and the constrained components are zeroed. The modified vectors are then transformed
back to the global system:
𝐚𝐼 = 𝐪T 𝐚𝐼1 (26.19)

𝛚̇ 𝐼 = 𝐪T 𝛚̇ 𝐼𝑙 (26.20)

26.5 Prescribed Displacements, Velocities, and


Accelerations
Prescribed displacements, velocities, and accelerations are treated in a nearly
identical way to displacement constraints. After imposing the zero displacement
constraints, the prescribed values are imposed as velocities at time, 𝑡𝑛+1/2 . The acceleration
versus time curve is integrated or the displacement versus time curve is differentiated to
generate the velocity versus time curve. The prescribed nodal components are then set.

24-4 (Boundary and Loading Conditions) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Boundary and Loading Conditions

26.6 Body Force Loads


Body force loads are used in many applications. For example, in structural analysis
the base accelerations can be applied in the simulation of earthquake loadings, the gun
firing of projectiles, and gravitational loads. The latter is often used with dynamic
relaxation to initialize the internal forces before proceeding with the transient response
calculation. In aircraft engine design the body forces are generated by the application of an
angular velocity of the spinning structure. The generalized body force loads are available if
only part of the structure is subjected to such loadings, e.g., a bird striking a spinning fan
blade.
For base accelerations and gravity we can fix the base and apply the loading as part
of the body force loads element by element according to Equation (22.18)

𝐟𝑒 body = ∫ 𝜌𝐍T 𝐍𝐚base 𝑑𝜐 = 𝐦𝑒 𝐚base , (26.21)


𝜐𝑚

where 𝐚base is the base acceleration and 𝐦𝑒 is the element (lumped) mass matrix.

LS-DYNA DEV 10/27/16 (r:8004) 24-5 (Boundary and Loading Conditions)


LS-DYNA Theory Manual Time Integration

27
Time Integration

27.1 Background
Consider the single degree of freedom damped system in Figure 27.1.

p(t)

u(t) - displacements
Figure 27.1. Single degree of freedom damped system.

Forces acting on mass m are shown in Figure 27.2.

The equations of equilibrium are obtained from d'Alembert’s principle


𝑓𝐼 + 𝑓𝐷 + 𝑓int = 𝑝(𝑡) (27.1)

fI inertia force
elastic force
fs
m p(t) external forces
fD
damping forces
Figure 27.2. Forces acting on mass, m

LS-DYNA DEV 10/27/16 (r:8004) 25-1 (Time Integration)


Time Integration LS-DYNA Theory Manual

𝑑2 𝑢
𝑓𝐼 = 𝑚𝑢̈; 𝑢̈ = acceleration
𝑑𝑡2
𝑑𝑢 (27.2)
𝑓𝐷 = 𝑐𝑢̇; 𝑢̇ = velocity
𝑑𝑡
𝑓int = 𝑘 ⋅ 𝑢; 𝑢 displacement
where 𝑐 is the damping coefficient, and k is the linear stiffness. For critical damping 𝑐 = ccr .
The equations of motion for linear behavior lead to a linear ordinary differential equation,
o.d.e.:
𝑚𝑢̈ + 𝑐𝑢̇ + 𝑘𝑢 = 𝑝(𝑡) (27.3)
but for the nonlinear case the internal force varies as a nonlinear function of the
displacement, leading to a nonlinear o.d.e.:
𝑚𝑢̈ + 𝑐𝑢̇ + 𝑓int (𝑢) = 𝑝(𝑡) (27.4)

Analytical solutions of linear ordinary differential equations are available, so instead


we consider the dynamic response of linear system subjected to a harmonic loading. It is
convenient to define some commonly used terms:
Harmonic loading: 𝑝(𝑡) = 𝑝0 sin𝜛𝑡
Circular frequency: 𝜔 = √𝑚𝑘 for single degree of freedom
𝜔
Natural frequency: 𝑓 = 2𝜋 = 𝑇1 𝑇 = period
Damping ratio: 𝜉 = 𝑐𝑐𝑐𝑟 = 2𝑚𝜔
𝑐

Damped vibration frequency: 𝜔0 = 𝜔√1 − 𝜉 2


Applied load frequency: 𝛽=𝜔
̅̅̅̅̅
𝜔

The closed form solution is:


𝑢̇0 𝑝 1
𝑢(𝑡) = 𝑢0 cos𝜔𝑡 + sin𝜔𝑡 + 0 (sin𝜔
̅̅̅̅𝑡 − 𝛽sin𝜔𝑡)
𝜔 𝑘 1 − 𝛽2
(27.5)
homogeneous solution steady state transient
particular solution
with the initial conditions:

𝑢0 = initial displacement
𝑢̇0 = initial velocity
𝑝0
= static displacement
𝑘

For nonlinear problems, only numerical solutions are possible. LS-DYNA uses the
explicit central difference scheme to integrate the equations of motion.

25-2 (Time Integration) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Time Integration

27.2 The Central Difference Method


The semi-discrete equations of motion at time n are:
𝐌𝐚𝑛 = 𝐏𝑛 − 𝐅𝑛 + 𝐇𝑛 , (27.6)
where 𝐌 is the diagonal mass matrix, 𝐏𝑛 accounts for external and body force loads, 𝑭 𝑛 is
the stress divergence vector, and 𝐇𝑛 is the hourglass resistance. To advance to time 𝑡𝑛+1 ,
we use central difference time integration:
𝐚𝑛 = 𝐌−1 (𝐏𝑛 − 𝐅𝑛 + 𝐇𝑛 ), (27.7)

𝑛+1⁄2 𝑛−1⁄2 (27.8)


𝐯 =𝐯 + 𝐚𝑛 Δ𝑡𝑛 ,

𝑛+1⁄2 𝑛+1⁄2 (27.9)


𝐮𝑛+1 = 𝐮𝑛 + 𝐯 Δ𝑡 ,
where
(Δ𝑡𝑛 + Δ𝑡𝑛+1 )
𝑛+1⁄2
Δ𝑡 = , (27.10)
2
and 𝐯 and 𝐮 are the global nodal velocity and displacement vectors, respectively. We
update the geometry by adding the displacement increments to the initial geometry:
𝐱𝑛+1 = 𝐱0 + 𝐮𝑛+1 . (27.11)
We have found that, although more storage is required to store the displacement vector the
results are much less sensitive to round-off error.

27.3 Stability of Central Difference Method


The stability of the central difference scheme is determined by looking at the
stability of a linear system. The system of linear equations in uncoupled into the modal
equations where the modal matrix of eigenvectors, 𝛟, are normalized with respect to the
mass and linear stiffness matrices 𝐊, and 𝐌, respectively, such that:
𝛟T 𝐌𝛟 = I
(27.12)
𝛟T 𝐊𝛟 = ω2 .
With this normalization, we obtain for viscous proportional damping the decoupling of the
damping matrix, 𝐂:
𝛟T 𝐂𝛟 = 2𝜉𝜔 (27.13)
The equations of motion in the modal coordinates 𝐱 are:
𝑥̈ + 2𝜉ω𝑥̇ + ω2 𝑥 = 𝛟𝐓
⏟𝐩. (27.14)
=Y

LS-DYNA DEV 10/27/16 (r:8004) 25-3 (Time Integration)


Time Integration LS-DYNA Theory Manual

With central differences we obtain for the velocity and acceleration:


𝑥 − 𝑥𝑛−1
𝑥̇𝑛 = 𝑛+1 (27.15)
2Δ𝑡
𝑥𝑛+1 − 2𝑥𝑛 + 𝑥𝑛−1
𝑥̈𝑛 = . (27.16)
Δ𝑡2

Substituting 𝑥̇𝑛 and 𝑥̈𝑛 into equation of motion at time 𝑡𝑛 leads to:
2 − 𝜔2 Δ𝑡2 1 − 2𝜉𝜔Δ𝑡 Δ𝑡2
𝑥𝑛+1 = 𝑥 − 𝑥 + 𝑌 , (27.17)
1 + 2𝜉𝜔Δ𝑡2 𝑛 1 + 2𝜉𝜔Δ𝑡 𝑛−1 1 + 2𝜉𝜔Δ𝑡2 𝑛

𝑥𝑛 = 𝑥𝑛 , (27.18)
which in matrix form leads to

𝑥𝑛+1 2 − 𝜔2 Δ𝑡2 1 − 2𝜉𝜔Δ𝑡 𝑥 Δ𝑡2


⎡ − ⎤ 𝑛 ⎡ ⎤
[𝑥 ] = ⎢1 + 2𝜉𝜔Δ𝑡 1 + 2𝜉𝜔Δ𝑡⎥ [𝑥𝑛−1 ] + ⎢1 + 2𝜉𝜔Δ𝑡2 ⎥ 𝑌𝑛 , (27.19)
𝑛
⎣ 1 0 ⎦ ⎣ 0 ⎦
or
𝐱̂𝑛+1 = 𝐀𝐱̂𝑛 + 𝐋𝐘𝑛 , (27.20)
where, 𝐀 is the time integration operator for discrete equations of motion. After 𝑚 time
steps with 𝐋 = 0 we obtain:
𝐱̂𝑚 = 𝐀𝑚 𝐱̂0 . (27.21)
As 𝑚 approaches infinity, 𝐀 must remain bounded.

A spectral decomposition of 𝐴 gives:


𝑚
𝐀𝑚 = (𝐏T 𝐉𝐏) = 𝐏T 𝐉𝑚 𝐏, (27.22)
where, 𝐏, is the orthonormal matrix containing the eigenvectors of 𝐀, and 𝐉 is the Jordan
form with the eigenvalues on the diagonal. The spectral radius, 𝜌(𝐀), is the largest
eigenvalue of 𝐀 = max [diag. (J)]. We know that 𝐉𝑚 , is bounded if and only if:
∣𝜌(𝐀)∣ ≤ 1. (27.23)

Consider the eigenvalues of 𝐴 for the undamped equation of motion


2 2
Det [∣2 − 𝜔 Δ𝑡 −1∣ − 𝜆∣1 −1∣] = 0, (27.24)
1 0 1 0

−(2 − 𝜔2 Δ𝑡2 − 𝜆) ⋅ 𝜆 + 1 = 0, (27.25)

2 − 𝜔2 Δ𝑡2 (2 − 𝜔2 Δ𝑡2 )2 (27.26)


𝜆= ±√ − 1.
2 4

25-4 (Time Integration) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Time Integration

The requirement that |𝜆| ≤ 1 leads to:


2
Δ𝑡 ≤ , (27.27)
𝜔max
as the critical time step. For the damped equations of motion we obtain:
2
Δ𝑡 ≤ (√1 + 𝜉 2 − 𝜉 ). (27.28)
𝜔max

Thus, damping reduces the critical time step size. The time step size is bounded by
the largest natural frequency of the structure which, in turn, is bounded by the highest
frequency of any individual element in the finite element mesh.

27.4 Subcycling (Mixed Time Integration)


The time step size, Δ𝑡, is always limited by a single element in the finite element
mesh. The idea behind subcycling is to sort elements based on their step size into groups
whose step size is some even multiple of the smallest element step size, 2(𝑛−1) Δ𝑡, for integer
values of 𝑛 greater than or equal to 1. For example, in Figure 27.3 the mesh on the right
because of the thin row of elements is three times more expensive than the mesh on the left

The subcycling in LS-DYNA is based on the linear nodal interpoation partition


subcycling algorithm of Belytschko, Yen, and Mullen [1979], and Belytschko [1980]. In their
implementation the steps are:

1. Assign each node, 𝑖, a time step size, Δ𝑡𝑗 , according to:


Δ𝑡𝑖 = min (2⁄𝜔𝑗 ) over all elements 𝑗, connected to node 𝑖
2. Assign each element, 𝑗, a time step size, Δ𝑡𝑗 , according to:

Figure 27.3. The right hand mesh is much more expensive to compute than the
left hand due to the presence of the thinner elements.

LS-DYNA DEV 10/27/16 (r:8004) 25-5 (Time Integration)


Time Integration LS-DYNA Theory Manual

Δ𝑡𝑗 = min(Δ𝑡𝑖 ) over all nodes, 𝑖, of element, 𝑗


3. Group elements into blocks by time step size for vectorization.

In LS-DYNA we desire to use constant length vectors as much as possible even if it


means updating the large elements incrementally with the small time step size. We have
found that doing this decreases costs and stability is unaffected.

Hulbert and Hughes [1988] reviewed seven subcycling algorithms in which the
partitioning as either node or element based. The major differences within the two subcy-
cling methods lie in how elements along the interface between large and small elements are
handled, a subject which is beyond the scope of this theoretical manual. Nevertheless, they
concluded that three of the methods including the linear nodal interpolation method
chosen for LS-DYNA, provide both stable and accurate solutions for the problems they
studied. However, there was some concern about the lack of stability and accuracy proofs
for any of these methods when applied to problems in structural mechanics.

The implementation of subcycling currently includes the following element classes


and contact options:

• Solid elements
• Beam elements
• Shell elements
• Brick shell elements
• Penalty based contact algorithms.

but intentionally excludes discrete elements since these elements generally contribute
insignificantly to the calculational costs. The interface stiffnesses used in the contact
algorithms are based on the minimum value of the slave node or master segment stiffness
and, consequently, the time step size determination for elements on either side of the
interface is assumed to be decoupled; thus, scaling penalty values to larger values when
subcycling is active can be a dangerous exercise indeed. Nodes that are included in
constraint equations, rigid bodies, or in contact with rigid walls are always assigned the
smallest time step sizes.

To explain the implementation and the functioning of subcycling, we consider the


beam shown in Figure 27.4 where the beam elements on the right (material group 2) have a
Courant time step size exactly two times greater than the elements on the left. The nodes
attached to material group 2 will be called minor cycle nodes and the rest, major cycle
nodes. At time step 𝑛 = 𝑚𝑘 all nodal displacements and element stresses are known, where
𝑚 is the ratio between the largest and smallest time step size, 𝑘 is the number of major time
steps, and 𝑛 is the number of minor time steps. In Figures 27.5 and 27.6, the update of the
state variables to the next major time step 𝑘 + 1 is depicted. The stress state in the element
on the material interface in group 1 is updated during the minor cycle as the displacement
of the major cycle node shared by this element is assumed to vary linearly during the

25-6 (Time Integration) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Time Integration

E2 = 4E1 F(t)
Material Group 1
A2 = A1
Material Group 2
ρ2 = ρ1
Figure 27.4. Subcycled beam problem from Hulbert and Hughes [1988].

minor cycle update. This linear variation of the major cycle nodal displacements during
the update of the element stresses improves accuracy and stability.

In the timing study results in Table 24.1, fifty solid elements were included in each
group for the beam depicted in Figure 27.4, and the ratio between 𝐸2 to 𝐸1 was varied from
1 to 128 giving a major time step size greater than 10 times the minor. Note that as the ratio
between the major and minor time step sizes approaches infinity the reduction in cost
should approach 50 percent for the subcycled case, and we see that this is indeed the case.
The effect of subcycling for the more expensive fully integrated elements is greater as may
be expected. The overhead of subcycling for the case where 𝐸1 = 𝐸2 is relatively large.
This provides some insight into why a decrease in speed is often observed when subcycling
is active. For subcycling to have a significant effect, the ratio of the major to minor time
step size should be large and the number of elements having the minor step size should be
small. In crashworthiness applications the typical mesh is very well planned and
generated to have uniform time step sizes; consequently, subcycling will probably give a
net increase in costs.

LS-DYNA DEV 10/27/16 (r:8004) 25-7 (Time Integration)


Time Integration LS-DYNA Theory Manual

t
u2

v2
u1

v1
σ σ σ σ σ σ σ σ σ

Solve for accelerations, velocities, and displacements

t
u2

v2 σ σ σ σ σ σ σ
u1

v1
σ σ

Solve for minor cycle stresses


Figure 27.5. Timing diagram for subcycling algorithm based on linear nodal
interpolations.

Case 1 one point


integration with Number of cycles cpu time(secs)
elastic material
model
𝐸2 = 𝐸1 178 4.65
178 5.36 (+15.%)
𝐸2 = 4𝐸1 367 7.57
367 7.13 (-6.0%)
𝐸2 = 16𝐸1 714 12.17
715 10.18 (-20.%)
𝐸2 = 64𝐸1 1417 23.24
1419 16.39 (-29.%)
𝐸2 = 128𝐸1 2003 31.89
2004 22.37 (-30.%)

Case 2 eight point


integration with Number of cycles cpu time(secs)
orthotropic
material model
25-8 (Time Integration) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Time Integration

u2 u2

v2
σ σ σ σ σ σ σ
v2 u1

v1
σ σ

Solve for minor cycle accelerations, velocities, and displacements

t
σ σ σ σ σ σ σ σ σ u2
u2

v2
v2 u1

v1

Update stress for all elements


Figure 27.6. Timing diagram for subcycling algorithm based on linear nodal
interpolations.

𝐸2 = 𝐸1 180 22.09
180 22.75 (+3.0%)
𝐸2 = 4𝐸1 369 42.91
369 34.20 (-20.%)
𝐸2 = 16𝐸1 718 81.49
719 54.75 (-33.%)
𝐸2 = 64𝐸1 1424 159.2
1424 97.04 (-39.%)
𝐸2 = 128𝐸1 2034 226.8
2028 135.5 (-40.%)

Table 24.1. Timing study showing effects of the ratio of the major to minor time step
size.

The impact of the subcycling implementation in the software has a very significant
effect on the internal structure. The elements in LS-DYNA are now sorted three times

LS-DYNA DEV 10/27/16 (r:8004) 25-9 (Time Integration)


Time Integration LS-DYNA Theory Manual

Time Integration Loop

update velocities
update displacements
write databases
and new geometry

kinematic based contact update current time and


and rigid walls check for termination
Start

update accelerations and apply force boundary


apply kinematic b.c.'s conditions

process penalty based process brick,beam,


contact interfaces shell elements
process discrete
elements
Figure 27.7. The time integration loop in LS-DYNA.

• By element number in ascending order.


• By material number for large vector blocks.
• By connectivity to insure disjointness for right hand side vectorization which is very
important for efficiency.

Sorting by Δ𝑡, interact with the second and third sorts and can result in the creation
of much smaller vector blocks and result in higher cost per element time step. During the
simulation elements can continuously change in time step size and resorting may be
required to maintain stability; consequently, we must check for this continuously. Sorting
cost, though not high when spread over the entire calculation, can become a factor that
results in higher overall cost if done too frequently especially if the factor, m, is relatively
small and the ratio of small to large elements is large.

25-10 (Time Integration) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Rigid Body Dynamics

28
Rigid Body Dynamics

A detailed discussion of the rigid body algorithm is presented by Benson and


Hallquist [1986] and readers are referred to this publication for more information. The
equations of motion for a rigid body are given by
𝑀𝜌 𝒙̈ = 𝒇𝜌𝑥 , (28.1)

𝑱𝜌 𝝎̇ + 𝝎 × 𝑱𝜌 𝝎 = 𝒇𝜌𝜔 , (28.2)
where 𝑀𝜌 is the physical mass, 𝑱𝜌 is the physical inertia tensor, 𝒙 is the location of the center
of mass, 𝝎 is the angular velocity of the body, and 𝒇𝜌𝑥 and 𝒇𝜌𝜔 are the forces and torques
applied to the rigid body through *LOAD_RIGID_BODY. These are equations that can be
found in a standard text book on rigid body mechanics. The physical properties of a rigid
body may come from three sources, these are

1.Integration of the mass density 𝜌 over a region 𝑉 occupied by the rigid body, for
which 𝑀𝜌 = ∫ 𝜌𝑑𝑉, and 𝑱𝜌 = ∫ 𝜌(𝒚 − 𝒙)⨂(𝒚 − 𝒙)𝑑𝑉. The initial rigid body coordi-
∫ 𝜌𝒚𝑑𝑉
nate 𝒙 is in this case determined from 𝒙 = 𝑀𝜌
. Here 𝒚 is the integrand variable.
2.Specifying properties using *PART_INERTIA, for which 𝑀𝜌 , 𝑱𝜌 and initial coordinate
𝒙 is simply specified in the keyword input deck.
3.For a nodal rigid body, *CONSTRAINED_NODAL_RIGID_BODY, the physical
properties vanish, i.e., 𝑀𝜌 = 0 and 𝑱𝜌 = 𝟎, and the position is arbitrary.

All rigid bodies possess slave nodes, which play a role when rigid bodies in LS-DYNA
interact with their surroundings. Slave nodes may come from the following.

1.The nodes in the finite element mesh for the part specified as rigid through
*MAT_RIGID.
2.Extra nodes definitions through *CONSTRAINED_EXTRA_NODES.
3.The set used for a nodal rigid body in *CONSTRAINED_NODAL_RIGID_BODY.

LS-DYNA DEV 10/27/16 (r:8004) 26-1 (Rigid Body Dynamics)


Rigid Body Dynamics LS-DYNA Theory Manual

We use 𝑆 to denote the set of slave nodes to the rigid body, and these are constrained to the
rigid body through the equations
𝒙𝑖 = 𝒙 + 𝑸(𝒙𝑖0 − 𝒙 0 ), (28.3)

𝑸𝑖 = 𝑸, (28.4)
for all 𝑖 ∈ 𝑆. Here we have introduced the orientations 𝑸 and 𝑸𝑖 of the rigid body and
slave node 𝑖, respectively. Furthermore, 𝒙𝑖 is the coordinate of slave node 𝑖 and we use
superscript 0 to denote a quantity at time zero. The time evolution of 𝑸 is
𝑸̇ = 𝜴𝑸, (28.5)
where 𝛀𝒓 = 𝝎 × 𝒓 for an arbitrary vector 𝒓 and 𝑸 = 𝑰 (identity) at time zero. So equations
(28.3) and (28.4) can equivalently be put in rate form
𝒙̇𝑖 = 𝒙̇ + 𝝎 × 𝒓𝑖 , (28.6)

𝝎𝑖 = 𝝎, (28.7)
where 𝒓𝑖 = 𝒙𝑖 − 𝒙 and understandably 𝝎𝑖 rotational velocity of slave node 𝑖. This also
determines the space of admissible virtual displacement for the slave nodes in the context
of work principles, and for this reason we use a compact notation for this equation
𝒙̇ 𝑰 −𝑹𝑖 𝒙̇
[ 𝑖] = [ ] [ ]. (28.8)
𝝎𝑖 𝟎 𝑰 𝝎
where 𝑹𝑖 𝒓 = 𝒓𝑖 × 𝒓 for an arbitrary vector 𝒓.

Slave nodes may have masses 𝑚𝑖 , inertias 𝑱𝑖 and forces 𝒇𝑖𝑥 and 𝒇𝑖𝜔 associated with
them. The inertia properties may come from

1.Mass contributions from deformable elements connected to the rigid body, either
through *CONSTRAINED_EXTRA_NODES or
*CONSTRAINED_NODAL_RIGID_BODY or simply merged mesh.
2.Lumped masses through *ELEMENT_MASS or *ELEMENT_INERTIA.

and the forces may come from

1.External loads through *LOAD_NODE or *LOAD_SEGMENT.


2.Contacts or fluid structure interaction (FSI) or similar.
3.Internal forces on deformable nodes of a contiguous part.

Note that these quantities exclude any contributions from and on the rigid body itself, as
these are all collected in 𝑀𝜌 , 𝑱𝜌 , 𝒇𝜌𝑥 and 𝒇𝜌𝜔 . The motion of the slave nodes is governed by
their own equations of motion
𝑚𝑖 𝒙̈𝑖 = 𝒇𝑖𝑥 , (28.9)

𝑱𝑖 𝝎̇ 𝑖 + 𝝎𝑖 × 𝑱𝑖 𝝎𝑖 = 𝒇𝑖𝜔 , (28.10)

26-2 (Rigid Body Dynamics) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Rigid Body Dynamics

for 𝑖 ∈ 𝑆. We seek a set of equations for the rigid body that combines (28.1)-(28.2) and
(28.9)-(28.10) by condensing out the dependence of the slave nodes through (28.6)-(28.7).
Differentiating (28.6)-(28.7) with respect to time yields
𝒙̈𝑖 = 𝒙̈ + 𝝎̇ × 𝒓𝑖 + 𝝎 × 𝝎 × 𝒓𝑖 , (28.11)

𝝎̇ 𝑖 = 𝝎̇ , (28.12)
which can be inserted into (28.9)-(28.10) to yield
𝑚𝑖 [𝒙̈ + 𝝎̇ × 𝒓𝑖 ] = 𝒇𝑖𝑥 − 𝑚𝑖 𝝎 × 𝝎 × 𝒓𝑖 , (28.13)

𝑱𝑖 𝝎̇ = 𝒇𝑖𝜔 − 𝝎 × 𝑱𝑖 𝝎. (28.14)
This can be compactly written as
𝑚 −𝑚𝑖 𝑹𝑖 𝒙̈ 𝒇𝑖𝑥 − 𝑚𝑖 𝝎 × 𝝎 × 𝒓𝑖
[ 𝑖 ][ ] = [ 𝜔 ]. (28.15)
𝟎 𝑱𝑖 𝝎̇ 𝒇 𝑖 − 𝝎 × 𝑱𝑖 𝝎

It remains to use the principle of virtual work, employing (28.8), to reduce the number of
equations (rigid body and slave nodes) to the generalized rigid body equations. The result
of this endeavor is

(𝑀𝜌 + ∑ 𝑚𝑖 )𝒙̈ − (∑ 𝑚𝑖 𝑹𝑖 )𝝎̇ = 𝒇𝜌𝑥 + ∑ (𝒇𝑖𝑥 − 𝑚𝑖 𝝎 × 𝝎 × 𝒓𝑖 ), (28.16)


𝑖∈𝑆 𝑖∈𝑆 𝑖∈𝑆

−(∑ 𝑚𝑖 𝑹𝑖𝑇 )𝒙̈ + (𝑱𝜌 + ∑ 𝑱𝑖 + ∑ 𝑚𝑖 𝑹𝑖𝑇 𝑹𝑖 )𝝎̇


𝑖∈𝑆 𝑖∈𝑆 𝑖∈𝑆
(28.17)
= 𝒇𝜌𝜔 + ∑ (𝒇𝑖𝜔 − 𝝎 × 𝑱𝑖 𝝎) − ∑ 𝑹𝑖𝑇 (𝒇𝑖𝑥 − 𝑚𝑖 𝝎 × 𝝎 × 𝒓𝑖 ) − 𝝎 × 𝑱𝜌 𝝎,
𝑖∈𝑆 𝑖∈𝑆

A simplified expression can be obtained through the variable substitution


1
𝒛=𝒙+ ∑ 𝑚𝒓. (28.18)
𝑀 𝑖∈𝑆 𝑖 𝑖
This yields 𝒙̈ = 𝒛̈ + 𝑹𝑧−𝑥 𝝎̇ − 𝝎 × 𝝎 × 𝒓𝑧−𝑥 where 𝒓𝑧−𝑥 = 𝒛 − 𝒙 and 𝑹𝑧−𝑥 𝒓 = 𝒓𝑧−𝑥 × 𝒓 for an
arbitrary vector 𝒓. This can be inserted into (28.16) and (28.17) to provide
𝑀𝒛̈ = 𝒇 𝑥 , (28.19)

𝑱𝝎̇ = 𝒇 𝜔 . (28.20)
where

𝑀 = 𝑀𝜌 + ∑ 𝑚𝑖 , (28.21)
𝑖∈𝑆

𝒇 𝑥 = 𝒇𝜌𝑥 + ∑ 𝒇𝑖𝑥 , (28.22)


𝑖∈𝑆

𝑱 = 𝑱𝜌 + ∑ 𝑱𝑖 + ∑ 𝑚𝑖 𝑹𝑖𝑇 𝑹𝑖 − 𝑀𝑹𝑧−𝑥
𝑇
𝑹𝑧−𝑥 , (28.23)
𝑖∈𝑆 𝑖∈𝑆

LS-DYNA DEV 10/27/16 (r:8004) 26-3 (Rigid Body Dynamics)


Rigid Body Dynamics LS-DYNA Theory Manual

𝒇 𝜔 = −𝝎 × 𝑱𝝎 + 𝒇𝜌𝜔 + ∑ (𝒇𝑖𝜔 + 𝒓𝑖 × 𝒇𝑖𝑥 ) − 𝒓𝑧−𝑥 × 𝒇 𝑥 . (28.24)


𝑖∈𝑆

Equations (28.19)-(28.24) are the generalized rigid body equations to be solved for 𝒙
and 𝑸, cf. (28.5) and (28.18). The mass in (28.21) and inertia tensor in (28.23) are the
physical mass and physical inertia augmented by slave node properties; nodal masses 𝑚𝑖 ,
inertias 𝑱𝑖 and locations 𝒙𝑖 . We denote 𝑀 the algorithmic mass, which may not reflect what
the user intuitively expects when using *MAT_RIGID to make a part rigid. Similarly 𝑱 is
the algorithmic inertia tensor, and it is worth noting that a nodal rigid body must therefore
be connected to deformable elements or otherwise 𝑀 = 0 and 𝑱 = 𝟎 and its whereabouts
will be impossible to determine. For no mass scaling, all these properties are constant
(except for rotational updates of the inertia tensors) and can essentially be calculated at
time zero. If mass scaling is active, the slave nodal masses and inertias include the added
mass due to mass scaling and therefore change over time. This means that inertia
properties should be recomputed every time step to account for these changes, but the
default behavior is that this is done only for nodal rigid bodies and not for regular rigid
bodies. Presumably this is based on the assumption that the influence from slave nodes is
significant for nodal rigid bodies and not so much for regular rigid bodies, which is
probably true as long as the number of contiguous nodes is small compared to the total
number of nodes in the rigid body. Nevertheless, with RBSMS = 1 on *CONTROL_RIGID,
these extra masses are accounted for and equations (28.19)-(28.24) are solved as expressed
herein. This amounts to transforming 𝒙 to 𝒛 before the update, then update 𝒛, and
transform back to obtain the new 𝒙. As we now turn to the algorithmic update of the rigid
body location, we restrict ourselves to a special case for the sake of simplifying the
exposition;

1.We neglect mass scaling.


2.Physical properties are not defined by *PART_INERTIA.
3.The rigid body is not connected to deformable elements.
4.No lumped masses are present.

which means that 𝒛 = 𝒙.

From (28.19)-(28.20) can readily solve for the rigid body accelerations
𝒇𝑥
𝒙̈ = , (28.25)
𝑀

𝝎̇ = 𝑱 −1 𝒇 𝜔 . (28.26)
It turns out that the algorithmic mass 𝑀 can be calculated as
𝑀=∑ 𝑀𝑖 , (28.27)
𝑖∈𝑆

26-4 (Rigid Body Dynamics) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Rigid Body Dynamics

where 𝑀𝑖 is the mass of node 𝑖 as obtained from the mass of its associated elements
(integrating material density 𝜌 by shape functions 𝜑𝑖 over element domain). Furthermore𝒙
can be approximated from
𝑀𝒙 = ∑ 𝑀 𝑖 𝒙𝑖 . (28.28)
𝑖∈𝑆

Likewise, the inertia tensor is approximated by a nodal summation of the product of the
point masses with their moment arms
𝑱=∑ 𝑀𝑖 𝑹𝑖𝑇 𝑹𝑖 . (28.29)
𝑖∈𝑆

The initial velocities of the slave nodes are readily calculated for a rigid body from (28.6).
For arbitrary orientations of the body, the inertia tensor is transformed each time step
based on the incremental rotations using the standard rules of second-order tensors:
𝑱 𝑛+1 = 𝑨𝑱 𝑛 𝑨𝑇 (28.30)
where 𝑱 𝑛+1 is the updated inertia tensor components in the global frame. The transfor-
mation matrix 𝑨 is not stored since the formulation is incremental, but recomputed as
explained below. After calculating the rigid body accelerations from Equation (28.25) and
(28.26), the rigid body translational and rotational increment, ∆𝒙 and ∆𝜽, can be calculated
using the time step ∆𝑡 and the explicit time integration update. The translational
coordinate is then updated as
𝒙 𝑛+1 = 𝒙 𝑛 + ∆𝒙 (28.31)
and ∆𝜽 is used to calculate 𝑨 in (28.30) using the Hughes-Winget algorithm,
1 1
𝑨=𝑰+ 𝑇 ∆𝜽
(𝑰 + 𝛥𝑺) 𝛥𝑺, (28.32)
∆𝜽 2
1+
4
𝛥𝑺𝒓 = ∆𝜽 × 𝒓, ∀𝒓. (28.33)
The coordinates of the slave nodes are incrementally updated
𝒙𝑖𝑛+1 = 𝒙𝑖𝑛 + ∆𝒙 + (𝑨 − 𝑰)𝒓𝑖𝑛 , (28.34)
and the velocities of the nodes are calculated by differencing the coordinates
(𝒙𝑖𝑛+1 − 𝒙𝑖𝑛 )
𝒙̇𝑖 = . (28.35)
𝛥𝑡
A direct integration of the rigid body accelerations into velocity and displacements is not
used for two reasons: (1) calculating the rigid body accelerations of the nodes is more
expensive than the current algorithm, and (2) the second-order accuracy of the central
difference integration method would introduce distortion into the rigid bodies. Since the
accelerations are not needed within the program, they are calculated by a post-processor
using a difference scheme similar to the above.

LS-DYNA DEV 10/27/16 (r:8004) 26-5 (Rigid Body Dynamics)


Rigid Body Dynamics LS-DYNA Theory Manual

28.1 Rigid Body Joints


The joints for the rigid bodies in LS-DYNA, see Figure 28.1, are implemented using the
penalty method. Given a constraint equation 𝐶(𝑥𝑖 , 𝑥𝑗 ) = 0, for nodes 𝑖 and 𝑗, the penalty
function added to the Lagrangian of the system is −1/2𝑘𝐶2 (𝑥𝑖 , 𝑥𝑗 ). The resulting nodal
forces are:
∂𝐶(𝑥𝑖 , 𝑥𝑗 )
𝑓𝑖 = −𝑘𝐶(𝑥𝑖 , 𝑥𝑗 ) , (28.36)
∂𝑥𝑖

∂𝐶(𝑥𝑖 , 𝑥𝑗 )
𝑓𝑗 = −𝑘𝐶(𝑥𝑖 , 𝑥𝑗 ) . (28.37)
∂𝑥𝑗

The forces acting at the nodes have to convert into forces acting on the rigid bodies.
Recall that velocities of a node i is related to the velocity of the center of mass of a rigid
body by Equation (28.6). By using Equation (28.6) and virtual power arguments, it may be
shown that the generalized forces are:
𝐹𝑖𝑥 = 𝑓𝑖 , (28.38)

𝐹𝑖𝜔 = 𝑒𝑖𝑗𝑘 𝑥̅𝑗 𝑓𝑘 , (28.39)


which are the forces and moments about the center of mass.

The magnitude of the penalty stiffness 𝑘 is chosen so that it does not control the
stable time step size. For the central difference method, the stable time step Δ𝑡 is restricted
by the condition that,
2
Δ𝑡 = , (28.40)
𝛺
where 𝛺 is the highest frequency in the system. The six vibrational frequencies associated
with each rigid body are determined by solving their eigenvalue problems assuming 𝑘 = 1.
For a body with 𝑚 constraint equations, the linearized equations of the translational
degrees of freedom are
𝑀𝑋̈ + 𝑚𝑘𝑋 = 0, (28.41)

and the frequency is √𝑚𝑘/𝑀 where 𝑀 is the mass of the rigid body. The corresponding
rotational equations are
𝐉𝛉̈ + 𝐊𝛉 = 0, (28.42)
𝐉 is the inertia tensor and 𝐊 is the stiffness matrix for the moment contributions from the
penalty constraints. The stiffness matrix is derived by noting that the moment contribution
of a constraint may be approximated by
𝐅𝑥 = −k𝐫𝑖 × (𝛉 × 𝐫𝑖 ), (28.43)

26-6 (Rigid Body Dynamics) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Rigid Body Dynamics

𝐫𝑖 = 𝐱𝑖 − 𝐗cm , (28.44)
and noting the identity,
𝐀 × (𝐁 × 𝐂) = |𝐀 ⋅ 𝐂 − 𝐀 ⊗ 𝐂|𝐁, (28.45)
so that
𝑚
𝐾 = ∑ 𝑘[𝐫𝑖 ⋅ 𝐫𝑖 − 𝐫𝑖 ⊗ 𝐫𝑖 ]. (28.46)
𝑖=1

The rotational frequencies are the roots of the equation det∣𝐊 − 𝛺2 𝐉∣ = 0, which is
cubic in 𝛺2 . Defining the maximum frequency over all rigid bodies for 𝑘 = 1 as 𝛺max , and
introducing a time step scale factor TSSF, the equation for 𝑘 is
2TSSF 2
𝑘≤( ) , (28.47)
Δ𝑡Ωmax

The joint constraints are defined in terms of the displacements of individual nodes.
Regardless of whether the node belongs to a solid element or a structural element, only its
translational degrees of freedom are used in the constraint equations.

A spherical joint is defined for nodes 𝑖 and 𝑗 by the three constraint equations,
𝑥1𝑖 − 𝑥1𝑗 = 0 𝑥2𝑖 − 𝑥2𝑗 = 0 𝑥3𝑖 − 𝑥3𝑗 = 0, (28.48)
and a revolute joint, which requires five constraints, is defined by two spherical joints, for a
total of six constraint equations. Since a penalty formulation is used, the redundancy in the
joint constraint equations is unimportant. A cylindrical joint is defined by taking a revolute
joint and eliminating the penalty forces along the direction defined by the two spherical
joints. In a similar manner, a planar joint is defined by eliminating the penalty forces that
are perpendicular to the two spherical joints.

The translational joint is a cylindrical joint that permits sliding along its axis, but not
rotation. An additional pair of nodes is required off the axis to supply the additional
constraint. The only force active between the extra nodes acts in the direction normal to the
plane defined by the three pairs of nodes.

The universal joint is defined by four nodes. Let the nodes on one body be 𝑖 and 𝑘,
and the other body, 𝑗 and 𝑙. Two of them, 𝑖 and 𝑗, are used to define a spherical joint for the
first three constraint equations. The fourth constraint equation is,
𝐶(𝑥𝑖 , 𝑥𝑗 , 𝑥𝑘 , 𝑥𝑙 ) = (𝑥𝑘 − 𝑥𝑖 ) ⋅ (𝑥𝑖 − 𝑥𝑗 ) = 0, (28.49)
∂𝐶
and is differentiated to give the penalty forces 𝑓𝑛 = −𝑘𝐶 ∂𝑥 , where 𝑛 ranges over the four
𝑛
nodes numbers.

LS-DYNA DEV 10/27/16 (r:8004) 26-7 (Rigid Body Dynamics)


Rigid Body Dynamics LS-DYNA Theory Manual

Figure 28.1. Rigid body joints in LS-DYNA.

28.2 Deformable to Rigid Material Switching


Occasionally in practice, long duration, large rigid body motions arise that are
prohibitively expensive to simulate if the elements in the model are deformable. Such a
case could occur possibly in automotive rollover where the time duration of the rollover
would dominate the cost relative to the post impact response that occurs much later.

To permit such simulations to be efficiently handled a capability to switch a subset


of materials from deformable to rigid and back to deformable is available. In practice the
suspension system and tires would remain deformable A flag is set in the input to let LS-
DYNA know that all materials in the model have the potential to become rigid bodies
sometime during the calculation. When this flag is set a cost penalty is incurred on vector
machines since the blocking of materials in the element loops will be based on the part ID
rather than the material type ID. Normally this cost is insignificant relative to the cost
reduction due to this unique feature.

26-8 (Rigid Body Dynamics) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Rigid Body Dynamics

For rigid body switching to work properly the choice of the shell element
formulation is critical. The Hughes-Liu elements cannot currently be used for two reasons.
First, since these elements compute the strains from the rotations of the nodal fiber vectors
from one cycle to the next, the nodal fiber vectors should be updated with the rigid body
motions and this is not done. Secondly, the stresses are stored in the global system as
opposed to the co-rotational system. Therefore, the stresses would also need to be
transformed with the rigid body motions or zeroed out. The co-rotational elements of
Belytschko and co-workers do not reference nodal fibers for the strain computations and
the stresses are stored in the co-rotational coordinate system which eliminates the need for
the transformations; consequently, these elements can be safely used. The membrane
elements and airbag elements are closely related to the Belytschko shells and can be safely
used with the switching options.

The beam elements have nodal triads that are used to track the nodal rotations and
to calculate the deformation displacements from one cycle to the next. These nodal triads
are updated every cycle with the rigid body rotations to prevent non-physical behavior
when the rigid body is switched back to deformable. This applies to all beam element
formulations in LS-DYNA. The Belytschko beam formulations are preferred for the
switching options for like the shell elements, the Hughes-Liu beams keep the stresses in the
global system. Truss elements like the membrane elements are trivially treated and pose
no difficulties.

The brick elements store the stresses in the global system and upon switching the
rigid material to deformable the element stresses are zeroed to eliminate spurious behavior.
The implementation addresses many potential problems and has worked well in practice.
The current restrictions can be eliminated if the need arises and anyway should pose no
insurmountable problems. We will continue to improve this capability if we find that it is
becoming a popular option.

28.3 Rigid Body Welds


The weld capability in LS-DYNA is based on rigid body dynamics. Each weld is
defined by a set of nodal points which moves rigidly with six degrees of freedom until a
failure criteria is satisfied. Five weld options are implemented including:

•Spot weld.
•Fillet weld
•Butt weld
•Cross fillet weld
•General weld

LS-DYNA DEV 10/27/16 (r:8004) 26-9 (Rigid Body Dynamics)


Rigid Body Dynamics LS-DYNA Theory Manual

z z

node 2 node 3

node 1 node 2
y y
2 node spotweld 3 node spotweld
node 1
x x
z

node n

node n-1

n node spotweld

y
node 2

x
node 1

Figure 28.2. Nodal ordering and orientation of the local coordinate system is
important for determining spotweld failure.

Welds can fail three ways: by ductile failure which is based on the effective plastic
strain, by brittle failure which is based on the force resultants acting on the rigid body
weld, and by a failure time which is specified in the input. When effective plastic strain is
used the weld fails when the nodal plastic strain exceeds the input value. A least squares
algorithm is used to generate the nodal values of plastic strains at the nodes from the
element integration point values. The plastic strain is integrated through the element and
the average value is projected to the nodes with a least square fit. In the resultant based
brittle failure the resultant forces and moments on each node of the weld are computed.
These resultants are checked against a failure criterion which is expressed in terms of these
resultants. The forces may be averaged over a user specified number of time steps to
eliminate breakage due to spurious noise. After all nodes of a weld are released the rigid
body is removed from the calculation.

26-10 (Rigid Body Dynamics) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Rigid Body Dynamics

Spotwelds are shown in Figure 28.2. Spotweld failure due to plastic straining occurs
p
when the effective nodal plastic strain exceeds the input value, 𝜀fail . This option can model
the tearing out of a spotweld from the sheet metal since the plasticity is in the material that
surrounds the spotweld, not the spotweld itself. This option should only be used for the
material models related to metallic plasticity and can result is slightly increased run times.
Brittle failure of the spotwelds occurs when:
𝑛 𝑚
max(𝑓𝑛 , 0) ∣𝑓 ∣
( ) + ( 𝑠 ) ≥ 1, (28.50)
𝑆𝑛 𝑆𝑠
where 𝑓𝑛 and 𝑓𝑠 are the normal and shear interface force. Component 𝑓𝑛 contributes for
tensile values only. When the failure time, 𝑡f , is reached the nodal rigid body becomes
inactive and the constrained nodes may move freely. In Figure 28.2 the ordering of the
nodes is shown for the 2 and 3 noded spotwelds. This order is with respect to the local
coordinate system where the local 𝑧 axis determines the tensile direction. The nodes in the
spotweld may coincide but if they are offset the local system is not needed since the 𝑧-axis
is automatically oriented based on the locations of node 1, the origin, and node 2. The
failure of the 3 noded spotweld may occur gradually with first one node failing and later
the second node may fail. For 𝑛 noded spotwelds the failure is progressive starting with
the outer nodes (1 and 𝑛) and then moving inward to nodes 2 and 𝑛 − 1. Progressive
failure is necessary to preclude failures that would create new rigid bodies.

Ductile fillet weld failure, due to plastic straining, is treated identically to spotweld
failure. Brittle failure of the fillet welds occurs when:

z local coordinate z
system
α
2 2 node fillet weld
1 x y

a L

3 node fillet weld


3
2
1

Figure 28.3. Nodal ordering and orientation of the local coordinate system is
shown for fillet weld failure.

LS-DYNA DEV 10/27/16 (r:8004) 26-11 (Rigid Body Dynamics)


Rigid Body Dynamics LS-DYNA Theory Manual

𝛽√𝜎𝑛2 + 3(𝜏𝑛2 + 𝜏𝑡2 ) ≥ 𝜎𝑓 , (28.51)

where
𝜎𝑛 = normal stress
𝜏𝑛 = shear stress in direction of weld (local 𝑦)
𝜏𝑡 = shear stress normal to weld (local 𝑥)
𝜎𝑓 = failure stress
𝛽 = failure parameter

Component 𝜎𝑛 is nonzero for tensile values only. When the failure time, 𝑡𝑓 , is
reached the nodal rigid body becomes inactive and the constrained nodes may move freely.
In Figure 28.3 the ordering of the nodes is shown for the 2 node and 3 node fillet welds.
This order is with respect to the local coordinate system where the local z axis determines
the tensile direction. The nodes in the fillet weld may coincide. The failure of the 3 node
fillet weld may occur gradually with first one node failing and later the second node may
fail.

In Figure 28.4 the butt weld is shown. Ductile butt weld failure, due to plastic
straining, is treated identically to spotweld failure. Brittle failure of the butt welds occurs
when:

𝛽√𝜎𝑛2 + 3(𝜏𝑛2 + 𝜏𝑡2 ) ≥ 𝜎𝑓 , (28.52)

2 tied noes that can


d be coincident

2 tied nodes
Lt

L y

4 tied nodes

Figure 28.4. Orientation of the local coordinate system and nodal ordering is
shown for butt weld failure.

where
26-12 (Rigid Body Dynamics) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Rigid Body Dynamics

2
3
1

z1 z2 z3
x2 x3
y1
x1 y2 2 y3
3 3
1

Figure 28.5. A simple cross fillet weld illustrates the required input. Here
NFW = 3 with nodal pairs (A = 2, B = 1), (A = 3, B = 1), and (A = 3, B = 2). The
local coordinate axes are shown. These axes are fixed in the rigid body and are
referenced to the local rigid body coordinate system which tracks the rigid body
rotation.

𝜎𝑛 = normal stress
𝜏𝑛 = shear stress in direction of weld (local y)
𝜏𝑡 = shear stress normal to weld (local z)
𝜎𝑓 = failure stress
𝛽 = failure parameter

Component 𝜎𝑛 is nonzero for tensile values only. When the failure time, 𝑡𝑓 , is reached the
nodal rigid body becomes inactive and the constrained nodes may move freely. The nodes
in the butt weld may coincide.

The cross fillet weld and general weld are shown in Figures 28.5 and 28.6,
respectively. The treatment of failure for these welds is based on the formulation for the
fillet and butt welds.

LS-DYNA DEV 10/27/16 (r:8004) 26-13 (Rigid Body Dynamics)


Rigid Body Dynamics LS-DYNA Theory Manual

Figure 28.6. A general weld is a mixture of fillet and butt welds.

26-14 (Rigid Body Dynamics) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

29
Contact-Impact Algorithm

29.1 Introduction
The treatment of sliding and impact along interfaces has always been an important
capability in DYNA3D and more recently in LS-DYNA. Three distinct methods for
handling this have been implemented, which we will refer to as the kinematic constraint
method, the penalty method, and the distributed parameter method. Of these, the first
approach is now used for tying interfaces. The relative merits of each approach are
discussed below.

Interfaces can be defined in three dimensions by listing in arbitrary order all


triangular and quadrilateral segments that comprise each side of the interface. One side of
the interface is designated as the slave side, and the other is designated as the master side.
Nodes lying in those surfaces are referred to as slave and master nodes, respectively. In the
symmetric penalty method, this distinction is irrelevant, but in the other methods the slave
nodes are constrained to slide on the master surface after impact and must remain on the
master surface until a tensile force develops between the node and the surface.

Today, automatic contact definitions are commonly used. In this approach the slave
and master surfaces are generated internally within LS-DYNA from the part ID's given for
each surface. For automotive crash models it is quite common to include the entire vehicle
in one single surface contact definition where the all the nodes and elements within the
interface can interact.

29.2 Kinematic Constraint Method


The kinematic constraint method which uses the impact and release conditions of
Hughes et al., [1976] was implemented first in DYNA2D [Hallquist 1976b] and finally
extended to three dimensions in DYNA3D. Constraints are imposed on the global
LS-DYNA DEV 10/27/16 (r:8004) 27-1 (Contact-Impact Algorithm)
Contact-Impact Algorithm LS-DYNA Theory Manual

equations by a transformation of the nodal displacement components of the slave nodes


along the contact interface. This transformation has the effect of eliminating the normal
degree of freedom of nodes. To preserve the efficiency of the explicit time integration, the
mass is lumped to the extent that only the global degrees of freedom of each master node
are coupled. Impact and release conditions are imposed to insure momentum
conservation. The release conditions are of academic interest and were quickly removed
from the coding.

Problems arise with this method when the master surface zoning is finer than the
slave surface zoning as shown in two dimensions in Figure 29.1. Here, certain master
nodes can penetrate through the slave surface without resistance and create a kink in the
slide line. Such kinks are relatively common with this formulation, and, when interface
pressures are high, these kinks occur whether one or more quadrature points are used in
the element integration. It may be argued, of course, that better zoning would minimize
such problems; but for many problems that are of interest, good zoning in the initial
configuration may be very poor zoning later. Such is the case, for example, when gaseous
products of a high explosive gas expand against the surface of a structural member.

29.3 Penalty Method


The penalty method is used in the explicit programs DYNA2D and DYNA3D as well
as in the implicit programs NIKE2D and NIKE3D. The method consists of placing normal
interface springs between all penetrating nodes and the contact surface. With the exception
of the spring stiffness matrix which must be assembled into the global stiffness matrix, the
implicit and explicit treatments are similar. The NIKE2D/3D and DYNA2D/3D programs
compute a unique modulus for the element in which it resides. In our opinion, pre-
empting user control over this critical parameter greatly increases the success of the
method.

slave surface
master surface

Indicates nodes treated as free surface nodes


Figure 29.1. Nodes of the master slide surface designated with an “x” are treated
as free surface nodes in the nodal constraint method.

27-2 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

Quite in contrast to the nodal constraint method, the penalty method approach is
found to excite little if any mesh hourglassing. This lack of noise is undoubtedly
attributable to the symmetry of the approach. Momentum is exactly conserved without the
necessity of imposing impact and release conditions. Furthermore, no special treatment of
intersecting interfaces is required, greatly simplifying the implementation.

Currently three implementations of the penalty algorithm are available:


• Standard Penalty Formulation
• Soft Constraint Penalty Formulation, which has been implemented to treat contact
between bodies with dissimilar material properties (e.g. steel-foam). Stiffness
calculation and its update during the simulation differs from the Standard Penalty
Formulation.
• Segment-based Penalty Formulation, it is a powerful contact algorithm whose logic
is a slave segment-master segment approach instead of a traditional slave node-
master segment approach. This contact has proven very useful for airbag self-
contact during inflation and complex contact conditions.

In the standard penalty formulation, the interface stiffness is chosen to be


approximately the same order of magnitude as the stiffness of the interface element normal
to the interface. Consequently the computed time step size is unaffected by the existence of
the interfaces. However, if interface pressures become large, unacceptable penetration may
occur. By scaling up the stiffness and scaling down the time step size, we may still solve
such problems using the penalty approach. Since this increases the number of time steps
and hence the cost, a sliding-only option has been implemented for treating explosive-
structure interaction problems thereby avoiding use of the penalty approach. This latter
option is based on a specialization of the third method described below.

29.4 Distributed Parameter Method


This method is used in DYNA2D, and a specialization of it is the sliding only option
in DYNA3D. Motivation for this approach came from the TENSOR [Burton et. al., 1982]
and HEMP [Wilkins 1964] programs which displayed fewer mesh instabilities than
DYNA2D with the nodal constraint algorithm. The first DYNA2D implementation of this
last algorithm is described in detail by Hallquist [1978]. Since this early publication, the
method has been moderately improved but the major ideas remain the same.

In the distributed parameter formulation, one-half the slave element mass of each
element in contact is distributed to the covered master surface area. Also, the internal
stress in each element determines a pressure distribution for the master surface area that
receives the mass. After completing this distribution of mass and pressure, we can update

LS-DYNA DEV 10/27/16 (r:8004) 27-3 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

b10 ∂b1 ∂b2


b20

∂B2
∂B 1
B20
B10

Figure 29.2. Reference and deformed configuration.

the acceleration of the master surface. Constraints are then imposed on slave node
accelerations and velocities to insure their movement along the master surface. Unlike the
finite difference hydro programs, we do not allow slave nodes to penetrate; therefore we
avoid “put back on” logic. In another simplification, our calculation of the slave element
relative volume ignores any intrusion of the master surfaces. The HEMP and TENSOR
codes consider the master surface in this calculation.

29.5 Preliminaries

Consider the time-dependent motion of two bodies occupying regions B1 and B2 in


their undeformed configuration at time zero. Assume that the intersection
B1 ∩ B2 = 0, (29.1)
is satisfied. Let 𝜕B1 and ∂B2 denote the boundaries of B1 and B2 , respectively. At some
later time, these bodies occupy regions b1 and b2 bounded by ∂b1 and ∂b2 as shown in
Figure 29.2. Because the deformed configurations cannot penetrate,
(b1 − ∂b1 ) ∩ b2 = 0. (29.2)
As long as (∂b1 ∩ ∂b2 ) = 0, the equations of motion remain uncoupled. In the foregoing
and following equations, the right superscript 𝛼 (= 1,2) denotes the body to which the
quantity refers.

27-4 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

S1
ns
S4
ms
S2

S3
X3
X2

X1
Figure 29.3. In this figure, four master segments can harbor slave node 𝑛𝑠 given
that 𝑚𝑠 is the nearest master node.

Before a detailed description of the theory is given, some additional statements


should be made concerning the terminology. The surfaces ∂b1 and ∂b2 of the discretized
bodies b1 and b2 become the master and slave surfaces respectively. Choice of the master
and slave surfaces is arbitrary when the symmetric penalty treatment is employed.
Otherwise, the more coarsely meshed surface should be chosen as the master surface unless
there is a large difference in mass densities in which case the side corresponding to the
material with the highest density is recommended. Nodal points that define ∂b1 are called
master nodes and nodes that define ∂b2 are called slave nodes. When (∂b1 ∩ ∂b2 ) ≠ 0, the
constraints are imposed to prevent penetration. Right superscripts are implied whenever a
variable refers to either the master surface ∂b1 , or slave surface, ∂b2 ; consequently, these
superscripts are dropped in the development which follows.

29.6 Slave Search


The slave search is common to all interface algorithms implemented in DYNA3D.
This search finds for each slave node its nearest point on the master surface. Lines drawn
from a slave node to its nearest point will be perpendicular to the master surface, unless the
point lies along the intersection of two master segments, where a segment is defined to be a
3- or 4-node element of a surface.

Consider a slave node, 𝑛𝑠 , sliding on a piecewise smooth master surface and assume
that a search of the master surface has located the master node, 𝑚𝑠 , lying nearest to 𝑛𝑠 .
Figure 29.3 depicts a portion of a master surface with nodes 𝑚𝑠 and 𝑛𝑠 labeled. If 𝑚𝑠 and 𝑛𝑠
do not coincide, 𝑛𝑠 can usually be shown to lie in a segment 𝑠1 via the following tests:
(𝐜𝑖 × 𝐬) ⋅ (𝐜𝑖 × 𝐜𝑖+1 ) > 0, (29.3)

LS-DYNA DEV 10/27/16 (r:8004) 27-5 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

(𝐜𝑖 × 𝐬) ⋅ (𝐬 × 𝐜𝑖+1 ) > 0,


where vector 𝐜𝑖 and 𝐜𝑖+1 are along edges of 𝑠1 and point outward from 𝑚𝑠 . Vector 𝐬 is the
projection of the vector beginning at 𝑚𝑠 , ending at 𝑛𝑠 , and denoted by 𝐠, onto the plane
being examined (see Figure 29.4).
𝐬 = 𝐠 − (𝐠 ⋅ 𝐦)𝐦, (29.4)
where for segment 𝑠1
𝐜𝑖 × 𝐜𝑖+1
𝐦= . (29.5)
∣𝐜𝑖 × 𝐜𝑖+1 ∣

Since the sliding constraints keep 𝑛𝑠 close but not necessarily on the master surface
and since 𝑛𝑠 may lie near or even on the intersection of two master segments, the
inequalities of Equation (29.3) may be inconclusive, i.e., they may fail to be satisfied or
more than one may give positive results. When this occurs 𝑛𝑠 is assumed to lie along the
intersection which yields the maximum value for the quantity
𝐠 ⋅ 𝐜𝑖
𝑖 = 1,2,3,4, .. (29.6)
|𝐜𝑖 |
When the contact surface is made up of badly shaped elements, the segment apparently
identified as containing the slave node actually may not, as shown in Figure 29.5.

Assume that a master segment has been located for slave node 𝑛𝑠 and that 𝑛𝑠 is not
identified as lying on the intersection of two master segments. Then the identification of
the contact point, defined as the point on the master segment which is nearest to 𝑛𝑠 ,
becomes nontrivial. For each master surface segment, 𝑠1 is given the parametric
representation of Equation (1.7), repeated here for clarity:

ns
s
g
ci+1 ms

X3
X2

X1
Figure 29.4. Projection of g onto master segment 𝑠1

27-6 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

1
2 8

7
6
3
5
4

Figure 29.5. When the nearest node fails to contain the segment that harbors the
slave node, segments numbered 1-8 are searched in the order shown.

𝐫 = 𝑓1 (𝜉 , 𝜂)𝐢1 + 𝑓2 (𝜉 , 𝜂)𝐢2 + 𝑓3 (𝜉 , 𝜂)𝐢3 , (29.7)


where
4
𝑗
𝑓𝑖 (𝜉 , 𝜂) = ∑ 𝜙𝑗 𝑥𝑖 . (29.8)
𝑗=1

Note that r1 is at least once continuously differentiable and that


∂𝐫 ∂𝐫
× ≠ 0, (29.9)
∂𝜉 ∂𝜂
Thus 𝐫 represents a master segment that has a unique normal whose direction depends
continuously on the points of s1 .

Let t be a position vector drawn to slave node ns and assume that the master surface
segment s1 has been identified with ns . The contact point coordinates (𝜉c , 𝜂c ) on s1 must
satisfy
∂𝐫
(𝜉 , 𝜂 ) ⋅ [𝐭 − 𝐫(𝜉𝑐 , 𝜂𝑐 )] = 0, (29.10)
∂𝜉 𝑐 𝑐

∂𝐫
(𝜉 , 𝜂 ) ⋅ [𝐭 − 𝐫(𝜉𝑐 , 𝜂𝑐 )] = 0. (29.11)
∂𝜂 𝑐 𝑐

The physical problem is illustrated in Figure 29.6, which shows ns lying above the
master surface. Equations (29.10) and (29.11) are readily solved for 𝜉c and 𝜂c . One way to
accomplish this is to solve Equation (29.10) for 𝜉c in terms of 𝜂c , and substitute the results

LS-DYNA DEV 10/27/16 (r:8004) 27-7 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

3 2
ns

ξ
X3 t
4
1
r

X2

X1
Figure 29.6. Location of contact point when ns lies above master segment.

into Equation (29.11). This yields a cubic equation in 𝜂c which is presently solved
numerically in LS-DYNA. In the near future, we hope to implement a closed form solution
for the contact point.

The equations are solved numerically. When two nodes of a bilinear quadrilateral
are collapsed into a single node for a triangle, the Jacobian of the minimization problem is
singular at the collapsed node. Fortunately, there is an analytical solution for triangular
segments since three points define a plane. Newton-Raphson iteration is a natural choice
for solving these simple nonlinear equations. The method diverges with distorted elements
unless the initial guess is accurate. An exact contact point calculation is critical in post-
buckling calculations to prevent the solution from wandering away from the desired
buckling mode.

Three iterations with a least-squares projection are used to generate an initial guess:
𝜉0 = 0, 𝜂0 = 0,
𝐫,𝜉 Δ𝜉 𝐫,𝜉
[𝐫 ] [𝐫,𝜉 𝐫,𝜂 ] { } = [𝐫 ] {𝐫(𝜉𝑖, 𝜂𝑖 ) − 𝐭}, (29.12)
,𝜂 Δ𝜂 ,𝜂

𝜉𝑖+1 = 𝜉𝑖 + Δ𝜉 , 𝜂𝑖+1 = 𝜂𝑖 + Δ𝜂,


followed by the Newton-Raphson iterations which are limited to ten iterations, but which
usually converges in four or less.
Δ𝜉 𝐫,𝜉
[H] { } = − {𝐫 } {𝐫(𝜉𝑖, 𝜂𝑖 ) − 𝐭},
Δ𝜂 ,𝜂
(29.13)
𝐫,ξ 0 𝐫 ⋅ 𝐫,𝜉𝜂
[H] = {𝐫 } [𝐫,𝜉 𝐫,𝜂 ] + [ ],
,η 𝐫 ⋅ 𝐫,𝜉𝜂 0

27-8 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

𝜉𝑖+1 = 𝜉𝑖 + Δ𝜉 , 𝜂𝑖+1 = 𝜂𝑖 + Δ𝜂,

In concave regions, a slave node may have isoparametric coordinates that lie outside
of the [−1, +1] range for all of the master segments, yet still have penetrated the surface. A
simple strategy is used for handling this case, but it can fail. The contact segment for each
node is saved every time step. If the slave node contact point defined in terms of the
isoparametric coordinates of the segment, is just outside of the segment, and the node
penetrated the isoparametric surface, and no other segment associated with the nearest
neighbor satisfies the inequality test, then the contact point is assumed to occur on the edge
of the segment. In effect, the definition of the master segments is extended so that they
overlap by a small amount. In the hydrocode literature, this approach is similar to the slide
line extensions used in two dimensions. This simple procedure works well for most cases,
but it can fail in situations involving sharp concave corners.

29.7 Sliding With Closure and Separation


29.7.1 Standard Penalty Formulation
Because this is perhaps the most general and most used interface algorithm, we
choose to discuss it first. In applying this penalty method, each slave node is checked for
penetration through the master surface. If the slave node does not penetrate, nothing is
done. If it does penetrate, an interface force is applied between the slave node and its
contact point. The magnitude of this force is proportional to the amount of penetration.
This may be thought of as the addition of an interface spring.

Penetration of the slave node ns through the master segment which contains its
contact point is indicated if
𝑙 = 𝐧𝑖 × [𝐭 − 𝐫(𝜉𝑐 , 𝜂𝑐 )] < 0, (29.14)
where
𝐧𝑖 = 𝐧𝑖 (𝜉𝑐 , 𝜂𝑐 ) (29.15)
is normal to the master segment at the contact point.

If slave node ns has penetrated through master segment 𝑠𝑖 , we add an interface force
vector 𝐟s :
𝐟𝑠 = −𝑙𝑘𝑖 𝐧𝑖 if 𝑙 < 0 (29.16)
to the degrees of freedom corresponding to ns and
𝑓𝑚𝑖 = 𝜙𝑖 (𝜉𝑐 , 𝜂𝑐 )𝑓𝑠 if 𝑙 < 0 (29.17)

LS-DYNA DEV 10/27/16 (r:8004) 27-9 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

to the four nodes (𝑖 = 1,2,3,4) that comprise master segment 𝑠𝑖 . The stiffness factor 𝑘𝑖 for
master segment 𝑠𝑖 is given in terms of the bulk modulus 𝐾𝑖 , the volume 𝑉𝑖 , and the face
area 𝐴𝑖 of the element that contains 𝑠𝑖 as
𝑓𝑠𝑖 𝐾𝑖 𝐴2𝑖
𝑘𝑖 = (29.18)
𝑉𝑖
for brick elements and
𝑓𝑠𝑖 𝐾𝑖 𝐴𝑖
𝑘𝑖 = (29.19)
max(shell diagonal)
for shell elements where 𝑓𝑠𝑖 is a scale factor for the interface stiffness and is normally
defaulted to .10. Larger values may cause instabilities unless the time step size is scaled
back in the time step calculation.

In LS-DYNA, a number of options are available for setting the penalty stiffness
value. This is often an issue since the materials in contact may have drastically different
bulk modulii. The calculational choices are:

• Minimum of the master segment and slave node stiffness. (default)


• Use master segment stiffness
• Use slave node value
• Use slave node value, area or mass weighted.
• As above but inversely proportional to the shell thickness.

The default may sometimes fail due to an excessively small stiffness. When this
occurs it is necessary to manually scale the interface stiffness. Care must be taken not to
induce an instability when such scaling is performed. If the soft material also has a low
density, it may be necessary to reduce the scale factor on the computed stable time step.

29.7.2 Soft Constraint Penalty Formulation


Very soft materials have an undesired effect on the contact stiffness, lowering its
value and ultimately causing excessive penetration. An alternative to put a scale factor on
the contact stiffness for SOFT = 0 is to use a Soft Constraint Penalty Formulation. The idea
behind this option is to eliminate the excessive penetration by using a different formulation
for the contact stiffness.

In addition to the master and slave contact stiffness, an additional stiffness is


calculated, which is based on the stability (Courant’s criterion) of the local system
comprised of two masses (segments) connected by a spring. This is the stability contact
stiffness 𝑘cs and is calculated by:

27-10 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

2
1
𝑘cs (𝑡) = 0.5 ⋅ SOFSCL ⋅ 𝑚∗ ⋅ ( ) , (29.20)
Δ𝑡𝑐 (𝑡)
where SOFSCL on Optional Card A of *CONTROL_CONTACT is the scale factor for the
Soft Constraint Penalty Formulation, 𝑚∗ is a function of the mass of the slave node and of
the master nodes. Δ𝑡cs is set to the initial solution timestep. If the solution time step grows,
Δ𝑡c is reset to the current time step to prevent unstable behavior.

A comparative check against the contact stiffness calculated with the traditional
penalty formulation, 𝑘soft=0 , and in general the maximum stiffness between the two is
taken,
𝑘soft=1 = max{𝑘cs , 𝑘soft=0 }. (29.21)

29.7.3 Segment-based Penalty Formulation


Segment based contact is a general purpose shell and solid element penalty type
contact algorithm. Segment based contact uses a contact stiffness similar to the SOFT = 1
stiffness option, but the details are quite different.

{SFS ⎫
⎧ 2
} 𝑚 𝑚 1
𝑘cs (𝑡) = 0.5 ⋅ SLSFAC ⋅ ⎨or ⎬ ( 1 2 ) ( ) . (29.22)
{
⎩SFM}⎭ 𝑚1 + 𝑚2 Δ𝑡𝑐 (𝑡)

Segment masses are used rather than nodal masses. Segment mass is equal to the
element mass for shell segments and half the element mass for solid element segments.
Like the Soft Constraint Penalty Formulation, 𝑑𝑡 is set to the initial solution time step which
is updated if the solution time step grows larger to prevent unstable behavior. However, it
differs from SOFT = 1 in how 𝑑𝑡 is updated. 𝑑𝑡 is updated only if the solution time step
grows by more than 5%. This allows 𝑑𝑡 to remain constant in most cases, even if the
solution time step slightly grows.

29.8 Recent Improvements in Surface-to-Surface Contact


A number of recent changes have been made in the surface-to-surface contact
including contact searching, accounting for thickness, and contact damping. These changes
have been implemented primarily to aid in the analysis of sheet metal forming problems.

29.8.1 Improvements to the Contact Searching

LS-DYNA DEV 10/27/16 (r:8004) 27-11 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

closet nodal point

slave node
Figure 29.7. Failure to find the contact segment can be caused by poor aspect
ratios in the finite element mesh.

In metal forming applications, problems with the contact searching were found
when the rigid body stamping dies were meshed with elements having very poor aspect
ratios. The nearest node algorithm described above can break down since the nearest node
is not necessarily anywhere near the segment that harbors the slave node as is assumed in
Figure 29.5 (see Figure 29.7). Such distorted elements are commonly used in rigid bodies in
order to define the geometry accurately.

To circumvent the problem caused by bad aspect ratios, an expanded searching


procedure is used in which we attempt to locate the nearest segment rather than the nearest
nodal point. We first sort the segments based on their centroids as shown in Figure 29.8
using a one-dimensional bucket sorting technique.

Once a list of possible candidates is identified for a slave node, it is necessary to


centroids of master contact segments

search 3 bins for this slave node


Figure 29.8. One-dimensional bucket sorting identifies the nearest segments for
each slave node.

locate the possible segments that contain the slave node of interest. For each quadrilateral
27-12 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Contact-Impact Algorithm

Figure 29.9. Interior points are constructed in the segments for determining the
closest point to the slave node.

segment, four points are constructed at the centroids of the four triangles each defined by 3
nodes as shown in Figure 29.9 where the black point is the centroid of the quadrilateral.
These centroids are used to find the nearest point to the slave node and hence the nearest
segment. The nodes of the three nearest segments are then examined to identify the three
nearest nodes. Just one node from each segment is allowed to be a nearest node.

When the nearest segment fails to harbor the slave node, the adjacent segments are
checked. The old algorithm checks the segments labeled 1-3 (Figure 29.10), which do not
contain the slave node, and fails.

29.8.2 Accounting For the Shell Thickness


Shell thickness effects are important when shell elements are used to model sheet

closet nodal point


2
1
3

slave node
segment identified as containing slave node
Figure 29.10. In case the stored segment fails to contain the node, the adjacent
segments are checked.

LS-DYNA DEV 10/27/16 (r:8004) 27-13 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

length of projection vector


Projected contact surface is 1/2 the shell thickness

Figure 29.11. Contact surface is based on mid-surface normal projection vectors.

metal. Unless thickness is considered in the contact, the effect of thinning on frictional
interface stresses due to membrane stretching will be difficult to treat. In the treatment of
thickness we project both the slave and master surfaces based on the mid-surface normal
projection vectors as shown in Figure 29.11. The surfaces, therefore, must be offset by an
amount equal to 1/2 their total thickness (Figure 29.12). This allows DYNA3D to check the
node numbering of the segments automatically to ensure that the shells are properly
oriented.

Thickness changes in the contact are accounted for if and only if the shell thickness
change option is flagged in the input. Each cycle, as the shell elements are processed, the
nodal thicknesses are stored for use in the contact algorithms. The interface stiffness may
change with thickness depending on the input options used.

Type 5 contact considers nodes interacting with a surface. This algorithm calls
exactly the same subroutines as surface-to-surface but not symmetrically: i.e., the
subroutines are called once, not twice. To account for the nodal thickness, the maximum
shell thickness of any shell connected to the node is taken as the nodal thickness and is

Figure 29.12. The slave and master surfaces must be offset in the input by one-
half the total shell thickness. This also allows the segments to be oriented
automatically.

27-14 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

updated every cycle. The projection of the node is done normal to the contact surface as
shown in Figure 29.13.

29.8.3 Initial Contact Interpenetrations

The need to offset contact surfaces to account for the thickness of the shell elements
contributes to initial contact interpenetrations. These interpenetrations can lead to severe
numerical problems when execution begins so they should be corrected if LS-DYNA is to
run successfully. Often an early growth of negative contact energy is one sign that initial
interpenetrations exist. Currently, warning messages are printed to the terminal, the
D3HSP file, and the MESSAG file to report interpenetrations of nodes through contact
segments and the modifications to the geometry made by LS-DYNA to eliminate the
interpenetrations. Sometimes such corrections simply move the problem elsewhere since it
is very possible that the physical location of the shell mid-surface and possibly the shell
thickness are incorrect. In the single surface contact algorithms any nodes still
interpenetrating on the second time step are removed from the contact with a warning
message.

In some geometry's interpenetrations cannot be detected since the contact node


interpenetrates completely through the surface at the beginning of the calculation. This is
illustrated in Figure 29.14. Another case contributing to initial interpenetrations occurs
when the edge of a shell element is on the surface of a solid material as seen in Figure 29.15.
Currently, shell edges are rounded with a radius equal to one-half the shell thickness.

To avoid problems with initial interpenetrations, the following recommendations


should be considered:
•Adequately offset adjacent surfaces to account for part thickness during the mesh
1/2 thickness of node

length of projection vector


Projected contact surface is 1/2 the shell thickness
Figure 29.13. In a type 5 contact, thickness can also be taken into account.

generation phase.
LS-DYNA DEV 10/27/16 (r:8004) 27-15 (Contact-Impact Algorithm)
Contact-Impact Algorithm LS-DYNA Theory Manual

Detected Penetration Undetected Penetration

Figure 29.14. Undetected interpenetration. Such interpenetrations are frequently


due to the use of coarse meshes.

•Use consistently refined meshes on adjacent parts which have significant curva-
tures.
•Be very careful when defining thickness on shell and beam section definitions --
especially for rigid bodies.
•Scale back part thickness if necessary. Scaling a 1.5mm thickness to .75mm should
not cause problems but scaling to .075mm might. Alternatively, define a smaller
contact thickness by part ID. Warning: if the part is too thin contact failure will
probably occur
•Use spot welds instead of merged nodes to allow the shell mid surfaces to be
offset.

29.8.4 Contact Energy Calculation


Contact energy, 𝐸contact , is incrementally updated from time 𝑛 to time 𝑛 + 1 for each
contact interface as:
𝑛𝑠𝑛 𝑛𝑚𝑛 𝑛+1
2
𝐸𝑛+1 𝑛 slave
contact = 𝐸contact + [∑ Δ𝐹𝑖 × Δ𝑑𝑖𝑠𝑡𝑖slave + ∑ Δ𝐹𝑖master × Δ𝑑𝑖𝑠𝑡𝑖master ] , (29.23)
𝑖=1 𝑖=1

Brick

shell

Inner penetration if edge is


too close
Figure 29.15. Undetected interpenetration due to rounding the edge of the shell
element.

27-16 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

Figure 29.16. Hemispherical deep drawing problem.

where 𝑛𝑠𝑛 is the number of slave nodes, 𝑛𝑚𝑛 is the number of master nodes, Δ𝐹𝑖slave is the
interface force between the ith slave node and the contact segment Δ𝐹𝑖master is the interface
force between the ith master node and the contact segment, Δ𝑑𝑖𝑠𝑡𝑖slave is the incremental
distance the ith slave node has moved during the current time step, and Δ𝑑𝑖𝑠𝑡𝑖master is the
incremental distance the ith master node has moved during the current time step. In the
absence of friction the slave and master side energies should be close in magnitude but
opposite in sign. The sum, 𝐸contact , should equal the stored energy. Large negative contact
energy is usually caused by undetected penetrations. Contact energies are reported in the
SLEOUT file. In the presence of friction and damping discussed below the interface energy
can take on a substantial positive value especially if there is, in the case of friction,
substantial sliding.

29.8.5 Contact Damping


Viscous contact damping has been added to all contact options including single
surface contact. The original intent was to damp out oscillations normal to the contact
surfaces during metal forming operations; however, it was later found to work effectively
in removing high frequency noise in problems which involve impact. The input requires a
damping value as a percentage of critical, 2𝑚, where 𝑚 is the mass and 𝜔 is the natural
frequency. Letting 𝑘 denote the interface stiffness, we compute the natural frequency for an
interface slave node from Equation 26.15.
𝑘(𝑚slave + 𝑚master )
𝜔=√ 𝑚 = min{𝑚slave , 𝑚master }. (29.24)
𝑚slave 𝑚master

LS-DYNA DEV 10/27/16 (r:8004) 27-17 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

Figure 29.17. Reaction forces with and without contact damping.

The master mass 𝑚master is interpolated from the master nodes of the segment containing
the slave node using the basis functions evaluated at the contact point of the slave node.

Force oscillations often occur as curved surfaces undergo relative motion. In these
cases contact damping will eliminate the high frequency content in the contact reaction
forces but will be unable to damp the lower frequency oscillations caused by nodes moving
from segment to segment when there is a large angle change between the segments. This is
shown in the hemispherical punch deep drawing in Figure 29.16. The reaction forces with
and without contact damping in Figure 29.17 show only minor differences since the
oscillations are not due to the dynamic effects of explicit integration. However, refining the
mesh as shown in Figure 29.18 to include more elements around the die corner as in Figure
29.18 greatly reduces the oscillations as shown in Figure 29.19. This shows the importance
of using an adequate mesh density in applications where significant relative motion is
expected around sharp corners.

27-18 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

Figure 29.18. Refinement of die radius.

Figure 29.19. The oscillations are effectively eliminated by the mesh refinement.

LS-DYNA DEV 10/27/16 (r:8004) 27-19 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

Friction
Friction in LS-DYNA is based on a Coulomb formulation. Let 𝐟∗ be the trial force,𝐟𝑛
the normal force, 𝑘 the interface stiffness, 𝜇 the coefficient of friction, and 𝐟𝑛 the frictional
force at time 𝑛. The frictional algorithm, outlined below, uses the equivalent of an elastic
plastic spring. The steps are as follows:

1. Compute the yield force, 𝐹𝑦 :


𝐹𝑦 = 𝜇 |𝐟𝑛 | (29.25)
2. Compute the incremental movement of the slave node
Δ𝐞 = 𝐫 𝑛+1 (𝜉𝑐𝑛+1 , 𝜂𝑐𝑛+1 ) − 𝐫 𝑛+1 (𝜉𝑐𝑛 , 𝜂𝑐𝑛 ) (29.26)
3. Update the interface force to a trial value:
𝐟∗ = 𝐟𝑛 − 𝑘Δ𝐞 (29.27)
4. Check the yield condition:
𝐟𝑛+1 = 𝐟∗ if ∣𝐟∗ ∣ ≤ 𝐹𝑦 (29.28)
5. Scale the trial force if it is too large:

𝑛+1
𝐹𝑦 𝐟∗
𝐟 = if ∣𝐟∗ ∣ > 𝐹𝑦 (29.29)
|𝐟∗ |

An exponential interpolation function smooths the transition between the static, 𝜇𝑠 ,


and dynamic, 𝜇𝑑 , coefficients of friction where 𝐯 is the relative velocity between the slave
node and the master segment:
𝜇 = 𝜇𝑑 + (𝜇𝑠 − 𝜇𝑑 ) 𝑒−𝑐|𝐯| , (29.30)
where
Δ𝐞
𝐯=
, (29.31)
Δ𝑡
Δ𝑡 is the time step size, and 𝑐 is a decay constant.

The interface shear stress that develops as a result of Coulomb friction can be very
large and in some cases may exceed the ability of the material to carry such a stress. We
therefore allow another limit to be placed on the value of the tangential force:
𝑛+1
𝑓 𝑛+1 = min(𝑓Coulomb , 𝜅𝐴master ), (29.32)
where 𝐴master is the area of the master segment and 𝜅 is the viscous coefficient. Since more
than one node may contribute to the shear stress of a segment, we recognize that the stress
may still in some cases exceed the limit 𝜅.

Typical values of friction, see Table 26.1, can be found in Marks Engineering
Handbook.
27-20 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Contact-Impact Algorithm

MATERIALS STATIC SLIDING


Hard steel on hard steel 0.78 (dry) 08 (greasy), .42 (dry)
Mild steel on mild steel 0.74 (dry) 10 (greasy), .57 (dry)
Aluminum on mild steel 0.61 (dry) 47 (dry)
Aluminum on aluminum1 05 (dry) 1.4 (dry)
Tires on pavement (40psi) 0.90 (dry). 69(wet), .85(dry)

Table 26.1. Typical values of Coulomb Friction [Marks]

29.9 Tied Interfaces


Sudden transitions in zoning are permitted with the tied interfaces as shown in
Figure 29.20 where two meshes of solid elements are joined. This feature can often
decrease the amount of effort required to generate meshes since it reduces the need to
match nodes across interfaces of merged parts.

Tied interfaces include four interface options of which three are in the Sliding
Interface Definition Section in the LS-DYNA User’s Manual. These are:

• Type 2 for tying surfaces with translational degrees of freedom.


• Type 6 for tying translational degrees of freedom of nodes to a surface

Tied interface permits


mesh transitions
Figure 29.20. Tied interface used for a mesh transition.

LS-DYNA DEV 10/27/16 (r:8004) 27-21 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

• Type 7 for tying both translational and rotational degrees of freedom of nodes

The fourth option is in the “Tie-Breaking Shell Definitions” Section of the user’s
manual and is meant as a way of tying edges of adjacent shells together. Unlike Type 7 this
latter option does not require a surface definition, simply nodal lines, and includes a failure
model based on plastic strain which can be turned off by setting the plastic failure strain to
a high value. The first two options, which are equivalent in function but differ in the input
definition, can be properly applied to nodes of elements which lack rotational degrees of
freedom. The latter options must be used with element types that have rotational degrees
of freedom defined at their nodes such as the shell and beam elements. One important
application of Type 7 is that it allows edges of shells to be tied to shell surfaces. In such
transitions the shell thickness is not considered. Exceptions from these latter statements is
in case of invoking the implicit accuracy option, see *CONTROL_ACCURACY, for which a
node with rotational degrees of freedom can tie to any element with or without offset. In
this case moments are consistently transferred based on the kinematics of the chosen tied
interface, the theory for this is presented in Section.

Since the constraints are imposed only on the slave nodes, the more coarsely meshed
side of the interface is recommended as the master surface. Ideally, each master node
should coincide with a slave node to ensure complete displacement compatibility along the
interface, but in practice this is often difficult if not impossible to achieve. In other words,
master nodes that do not coincide with a slave node can interpenetrate through the slave
surface.

Implementation of tied interface constraints is straightforward. Each time step we


loop through the tied interfaces and update each one independently. First, we distribute
the nodal forces and nodal mass of each slave node to the master nodes which define the
segment containing the contact point, i.e., the increments in mass and forces
Δ𝑓𝑚𝑖 = 𝜙𝑖 (𝜉𝑐, 𝜂𝑐 )𝑓𝑠 (29.33)
are added to the mass and force vector of the master surface. After the summation over all
slave nodes is complete, we can compute the acceleration of the master surface. The
acceleration of each slave node 𝑎𝑖𝑠 is then interpolated from the master segment containing
its contact points:
4
𝑗
𝑎𝑖𝑠 = ∑ 𝜙𝑗 (𝜉𝑐 , 𝜂𝑐 )𝑎𝑖 . (29.34)
𝑗=1

Velocities and displacements are now updated normally.

The interpolated contact point, (𝜉𝑐 , 𝜂𝑐 ), for each slave node is computed once, since
its relative position on the master segment is constant for the duration of the calculation. If
the closest point projection of the slave node to the master surface is non-orthogonal,
values of (𝜉𝑐 , 𝜂𝑐 ) greater than unity will be computed. To allow for slight errors in the
mesh definition, the slave node is left unconstrained if the magnitude of the contact point
27-22 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Contact-Impact Algorithm

exceeds 1.02. Great care should be exercised in setting up tied interfaces to ensure that the
slave nodes are covered by master segments.

Conflicting constraints must be avoided. Care should be taken not to include nodes
that are involved in a tied interfaces in another tied interface, in constraint sets such as
nodal constraint sets, in linear constraint equations, and in spot welds. Furthermore, tied
interfaces between rigid and deformable bodies are not permitted. LS-DYNA checks for
conflicting constraints on nodal points and if such conflicts are found, the calculation will
terminate with an error message identifying the conflict. Nodes in tied interfaces should
not be included as slave nodes in rigid wall definitions since interactions with stonewalls
will cause the constraints that were applied in the tied interface logic to be violated. We do
not currently check for this latter condition is LS-DYNA.

Tied interfaces require coincident surfaces and for shell element this means that the
mid-surfaces must be coincident. Consider Figure 29.21 where identical slave and master
surfaces are offset. In this case the tied constraints require that translational velocities of
tied nodes be identical, i.e.,
𝐯𝑠 = 𝐯𝑚 . (29.35)
Consequently, if the nodes are offset, rotations are not possible. The velocity of a tied slave
node in Figure 29.21 should account for the segment rotation:
𝐯𝑠 = 𝐯𝑚 − 𝑧̂ 𝐞3 × 𝛚, (29.36)
where 𝑧̂ is the distance to the slave node, 𝐞3 is the normal vector to the master surface at the
contact point, and 𝛚 is the angular velocity. Since this is not the case in the tied interfaces
logic, 𝑧̂ must be of zero length.

LS-DYNA projects tied slave nodes back to the master surface if possible and prints
warning messages for all projected offset nodes or nodes too far away to tie. This
projection eliminates the problems with rotational constraints but creates other difficulties:

• Geometry is modified
• Tied interfaces must be excluded from automatic generation since tied surfaces
cannot be mixed with automatic contact with thickness offsets.

LS-DYNA DEV 10/27/16 (r:8004) 27-23 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

Slave Surface

Master Surface
Figure 29.21. Offset tied interface.

An offset capability has been added to the tied interfaces which uses a penalty
approach. The penalty approach removes the major limitations of the constraint
formulation since with the offset option:

• Multiple tied interfaces cannot share common nodes.


• Rigid body nodes can be constrained.
• Tied interface nodes can have other constraints applied and can be subjected to
prescribed motions.

29.10 Strongly Objective Tied Contacts


In implicit calculations, non-physical results observed when some tied contact formulations
are combined with automatic single point constraints on solid element rotational degrees of
freedom (AUTOSPC on *CONTROL_IMPLICIT_SOLVER) have motivated an extension in
this context. A goal is to provide a universal tied contact formulation in implicit that works
well in most situations, thus preventing the user from having to think too hard about
which interface is best suited for the application at hand. To this end, a small selection of
tied interfaces have been singled out that all represent this universal contact, these are

*CONTACT_TIED_NODES_TO_SURFACE_CONSTRAINED_OFFSET
*CONTACT_TIED_NODES_TO_SURFACE_OFFSET
*CONTACT_TIED_SHELL_EDGE_TO_SURFACE_CONSTRAINED_OFFSET
*CONTACT_TIED_SHELL_EDGE_TO_SURFACE_BEAM_OFFSET

The first of these two do not consider rotational degrees of freedom, whereas the other two
do. Furthermore, the first and third are constraint based and the other two are penalty
based, so all in all these four cover much of what a user expects from a tied interface. By
setting IACC to 1 on *CONTROL_ACCURACY any of the tied contact options mentioned
above (and the non-offset counterparts as a side effect, i.e.,
*CONTACT_TIED_NODES_TO_SURFACE and
*CONTACT_TIED_SHELL_EDGE_TO_SURFACE) are treated with this strongly objective
formulation. In addition to being strongly objective, i.e., forces and moments transform
correctly under superposed rigid body motions in a single implicit step, this formulation

27-24 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

applies rotational constraints consistently when and only when necessary. This means not
only that slave nodes without rotational degrees of freedom are not rotationally
constrained, but also that bending and torsional rotations are constrained to the master
segment’s rotational motion in a way that is physically justified. To be more specific, slave
node bending rotations (i.e., rotations in the plane of the master segment) are constrained
to the master segment rotational degrees of freedom if this happens to stem from a shell
element, otherwise they are constrained to the master segment rotation as determined from
its individual nodal translations. The slave node torsional rotations (i.e., rotations with
respect to the normal of the master segment) are always constrained according to this latter
philosophy, thus avoiding a torsional constraint to the relatively weak drilling mode of
shells. So this tied contact formulation properly treats bending and torsional rotations, here
slave node rotational degrees of freedom typically come from shell or beam elements. So in
effect, it is in a sense sufficient to only consider

*CONTACT_TIED_SHELL_EDGE_TO_SURFACE_CONSTRAINED_OFFSET
*CONTACT_TIED_SHELL_EDGE_TO_SURFACE_BEAM_OFFSET

for most situations (choosing between a constraint or penalty formulation) but the other
two

*CONTACT_TIED_NODES_TO_SURFACE_CONSTRAINED_OFFSET
*CONTACT_TIED_NODES_TO_SURFACE_OFFSET

“non-rotational” formulations are included in the event of not wanting to constrain


rotations whatsoever. Referring to Figure 2927-22, the following is the mathematical
formulation of this tied contact formulation.

LS-DYNA DEV 10/27/16 (r:8004) 27-25 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

𝑮 𝑬
𝑮0 𝑬0

𝑠 𝒅
𝛼𝜆 Tying
𝑚4
point
𝜆 𝑚3 𝑭
𝑭0

𝑀
𝑝
𝑡>0
𝑚1
𝑡=0
𝑚2

Figure 2927-22 Kinematics of the implicit strong objective tied contact


f l ti

29.10.1 Kinematics
We consider a slave node 𝑠 tied to a master segment 𝑀 with an offset 𝜆, the slave node
projection onto the master segment is denoted 𝑝. Let 𝒙𝑠 denote the slave node coordinate
and
4 𝑖
𝒙𝑝 = ∑ ℎ𝑖 𝒙𝑚 (29.37)
𝑖=1
be the slave node projection on the master segment, where ℎ𝑖 are the constant isoparametric
weights. To each of 𝑠, 𝑝 and 𝑀 we associate orthonormal bases (coordinate systems)
represented by
𝑬 = {𝒆 1 𝒆 2 𝒆 3 } (29.38)

𝑭 = {𝒇1 𝒇2 𝒇3 } (29.39)

𝑮 = {𝒈1 𝒈2 𝒈3 }, (29.40)
respectively. The orthogonal matrix 𝑮 is the master segment coordinate system expressed
𝑖
as a function of the master coordinates 𝒙𝑚 with normal 𝒈3 . At 𝑡 = 0, we initialize𝑬0 = 𝑭0 =
𝑮0 but 𝑬 and 𝑭 then evolves independently based on the nodal rotational velocities 𝝎𝑠 and
4 𝑖
𝝎𝑝 = ∑ ℎ𝑖 𝝎 𝑚 . (29.41)
𝑖=1
In the numerical implementation, if ∆𝜽𝑠 = 𝝎𝑠 ∆𝑡 and ∆𝜽𝑝 = 𝝎𝑝 ∆𝑡 are the incremental
rotations of s and p at a given time step, then the coordinate systems are updated
𝑬𝑛+1 = 𝑹(∆𝜽𝑠 )𝑬𝑛 (29.42)

27-26 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

𝑭𝑛+1 = 𝑹(∆𝜽𝑝 )𝑭𝑛 (29.43)


where 𝑹(∆𝜽) denotes the finite rotation matrix corresponding to the rotational increment
∆𝜽. Of course, this only makes sense if the slave node and/or master segment have
rotational degrees of freedom. If the slave node lacks rotational degrees of freedom, then
neither bending nor torsion is constrained and 𝑬 is of no interest. If on the other hand the
master segment lacks rotational degrees of freedom, then we use 𝑭 = 𝑮 which is still of
interest to deal with the offset, 𝑮 is always well-defined. LS-DYNA automatically detects
rotational degrees of freedom and takes the proper measures.

29.10.2 Translational constraint


Given the notation in the previous section and referring to Figure 2927-22,
𝒅 = {𝒙𝑠 − 𝛼𝜆𝒆3 } − {𝒙𝑝 + (1 − 𝛼)𝜆𝒇3 } (29.44)
represents the vector we want to be zero in the translational part of the contact. Here 𝜆 is
the offset distance between 𝑝 and 𝑠 and is constant throughout the simulation.
Furthermore, 𝛼 is a constant between 0 and 1 that determines the actual tying point
according to the following simple rules. First, if the slave node is connected to a solid or
beam element or if the contact definition does not take rotational degrees of freedom into
account, then 𝛼 = 0. Otherwise, if the slave node is connected to a shell element and the
master segment is connected to a solid element, then 𝛼 = 1. If neither of those situations
apply then both slave and master sides must connect to shell elements, for which 𝛼 = 0.5,
so note that the value of 𝛼 is not accounting for the relative difference in shell thicknesses
but assumes equal shell thickness on both master and slave sides. For
*CONTACT_TIED_NODES_TO_SURFACE_CONSTRAINED_OFFSET and
*CONTACT_TIED_SHELL_EDGE_TO_SURFACE_CONSTRAINED_OFFSET, this
condition is imposed as a constraint
𝒅=𝟎 (29.45)
whereas for *CONTACT_TIED_NODES_TO_SURFACE_OFFSET and
*CONTACT_TIED_SHELL_EDGE_TO_SURFACE_BEAM_OFFSET a penalty formulation
is used. To this end we use a constitutive relation between force and displacement
𝒇 = 𝐾𝑓 𝒅 (29.46)
with 𝐾𝑓 as the penalty stiffness. Then an energy principle is employed to identify the nodal
forces and moments,
𝒇 𝑇 𝛿𝒅 = 𝑷𝑓𝑇 𝛿𝑿, (29.47)
where 𝑿 is nodal coordinate array of the slave master pair, and 𝛿 is the variation operator.
Identifying 𝑩𝑓 by
𝛿𝒅 = 𝑩𝑓 𝛿𝑿 (29.48)
the nodal force array is given by
𝑷𝑓 = 𝑩𝑇𝑓 𝒇 . (29.49)
Worth noticing is that 𝑩𝑓 is the constraint matrix corresponding to the constraint variant of
the contact.

LS-DYNA DEV 10/27/16 (r:8004) 27-27 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

29.10.3 Bending and torsion constraint


If the slave node has rotational degrees of freedom and the contact formulation is said to
treat those, bending and torsion is constrained. The bending strain is calculated as
𝜑𝑗 = 𝒈𝑗𝑇 (𝒆3 × 𝒇3 ), 𝑗 = 1,2 (29.50)
which essentially is a measure of the amount 𝒆3 and 𝒇3 deviates from being parallel. The
torsional strain is a scalar given by
1
𝜑3 = (𝒆𝑇1 𝒈2 − 𝒆𝑇2 𝒈1 ) (29.51)
2
and is a measure of the relative rotation between 𝑬 and 𝑮 with respect to the normal 𝒈3 .
The constraint to enforce is
𝜑1
𝝋 = ⎜𝜑2 ⎞
⎛ ⎟=𝟎 (29.52)
⎝𝜑3 ⎠
which for a penalty formulation leads to a constitutive relation between moment and
rotation vector
𝒎 = 𝐾𝑚 𝝋 (29.53)
with 𝐾𝑚 being a stiffness parameter. Following the approach for translational treatment,
we get the nodal force contribution
𝑷𝑚 = 𝑩𝑇𝑚 𝒎 (29.54)
where
𝛿𝝋 = 𝑩𝑚 𝛿𝑿. (29.55)
Also here 𝑩𝑚 is the constraint matrix in case a constraint formulation is used.

The formulae for 𝐾𝑓 and 𝐾𝑚 are found in earlier sections on tied interfaces while the
expressions for the matrices 𝑩𝑓 and 𝑩𝑚 are quite involved and omitted for the sake of
clarity, the nodal forces and moments are implemented by using (manual) algorithmic
differentiation.

29.11 Sliding-Only Interfaces


This option is seldom useful in structural calculations. Its chief usefulness is for treating
interfaces where the gaseous detonation products of a high explosive act on a solid
material. The present algorithm, though simple, has performed satisfactorily on a number
of problems of this latter type. We briefly outline the approach here since the algorithm is
still experimental and subject to change.

The method consists of five steps. In the first step, the mass per unit area (mass/area) and
pressure are found at each node on the slave surface. Next, the contact point for each
master node is found, and the slave mass/area and slave pressure at each master node is
interpolated from the slave surface. In the third step, this pressure distribution is applied
to the master surface to update its acceleration. In the fourth step, the normal component
of the acceleration at each node on the master surface is scaled by its z-factor defined as the
27-28 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Contact-Impact Algorithm

tied interface

Figure 29.23. Incremental searching may fail on surfaces that are not simply
connected. The new contact algorithm in LS-DYNA avoids incremental searching
for nodal points that are not in contact and all these cases are considered.

mass/area of the master surface at the master node divided by the sum of the mass/area of
the slave surface at the master node. The last step consists of resetting the normal
acceleration and velocity components of all slave nodes to ensure compatibility.

29.12 Bucket Sorting


Bucket sorting is now used extensively in both the surface to surface and single surface
contact algorithms. Version 920 of LS-DYNA no longer contains one-dimensional sorting.
Presently two separate but similar bucket sorts are in LS-DYNA. In the first and older
method we attempt to find for each node the three nearest nodes. In the newer method
which is systematically replacing the older method we locate the nearest segment.

The reasons for eliminating slave node tracking by incremental searching is illustrated in
Figure 29.23 where surfaces are shown which cause the incremental searches to fail. In
LS-DYNA tied interfaces are used extensively in many models creating what appears to the
contact algorithms to be topologically disjoint regions. For robustness, our new algorithms
account for such mesh transitions with only minor cost penalties. With bucket sorting
incremental searches may still be used but for reliability they are used after contact is
achieved. As contact is lost, the bucket sorting for the affected nodal points must resume.

In a direct search of a set of 𝑁 nodes to determine the nearest node, the number of distance
comparisons required is 𝑁 − 1. Since this comparison needs to be made for each node, the
LS-DYNA DEV 10/27/16 (r:8004) 27-29 (Contact-Impact Algorithm)
Contact-Impact Algorithm LS-DYNA Theory Manual

Bucket

1 8 7
X strips

2 6

3 4 5

Y Strips
Figure 29.24. One- and two-dimensional bucket sorting.

total number of comparisons is 𝑁(𝑁 − 1), with each of these comparisons requiring a
distance calculation
12 = (𝑥𝑖 − 𝑥𝑗 )2 + (𝑦𝑖 − 𝑦𝑗 )2 + (𝑧𝑖 − 𝑧𝑗 )2 , (29.56)
that uses eight mathematical operations. The cumulative effect of these mathematical
operations for 𝑁(𝑁 − 1) compares can dominate the solution cost at less than 100 elements.

The idea behind a bucket sort is to perform some grouping of the nodes so that the sort
operation need only calculate the distance of the nodes in the nearest groups. Consider the
partitioning of the one-dimensional domain shown in Figure 29.24. With this partitioning
the nearest node will either reside in the same bucket or in one of the two adjoining
buckets. The number of distance calculations is now given by
3𝑁
− 1, (29.57)
𝑎
where 𝑎 is the number of buckets. The total number of distance comparisons for the entire
one-dimensional surface is
3𝑁
𝑁( − 1). (29.58)
𝑎

27-30 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

Thus, if the number of buckets is greater than 3, then the bucket sort will require fewer
distance comparisons than a direct sort. It is easy to show that the corresponding number
of distance comparisons for two-dimensional and three-dimensional bucket sorts are given
by
9𝑁
𝑁( − 1) for 2D (29.59)
𝑎𝑏
27𝑁
𝑁( − 1) for 3D (29.60)
𝑎𝑏𝑐
where 𝑏 and 𝑐 are the number of partitions along the additional dimension.

The cost of the grouping operations, needed to form the buckets, is nearly linear with the
number of nodes 𝑁. For typical LS-DYNA applications, the bucket sort is 100 to 1000 times
faster than the corresponding direct sort. However, the sort is still an expensive part of the
contact algorithm, so that, to further minimize this cost, the sort is performed every ten or
fifteen cycles and the nearest three nodes are stored. Typically, three to five percent of the
calculational costs will be absorbed in the bucket sorting when most surface segments are
included in the contact definition.

29.12.1 Bucket Sorting in TYPE 4 Single Surface Contact


We set the number of buckets in the 𝑥, 𝑦, and 𝑧 coordinate directions to 𝑁𝑋, 𝑁𝑌, and 𝑁𝑍,
respectively. Letting LMAX represent the longest characteristic length (found by checking
the length of the segment diagonals and taking a fraction thereof) over all segments in the
contact definition, the number of buckets in each direction is given by
𝑥 − 𝑥min
𝑁𝑋 = max , (29.61)
LMAX
𝑦max − 𝑦min
𝑁𝑌 = , (29.62)
LMAX
𝑧max − 𝑧min
𝑁𝑍 = , (29.63)
LMAX
where the coordinate pairs (𝑥min , 𝑥max ), (𝑦min , 𝑦max ), and (𝑧min , 𝑧max ) define the extent of
the contact surface and are updated each time the bucket searching is performed. In order
to dynamically allocate memory effectively with FORTRAN, we further restrict the number
of buckets such that the total number of buckets does not exceed the number of nodes in
the contact surface, NSN or 5000:
𝑁𝑋 ⋅ 𝑁𝑌 ⋅ 𝑁𝑍 ≤ MIN (NSN, 5000). (29.64)

If the characteristic length, LMAX, is large due to an oversized contact segment or an


instability leading to a node flying off into space, the bucket sorting can be slowed down

LS-DYNA DEV 10/27/16 (r:8004) 27-31 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

considerably since the number of buckets will be reduced. In older versions of DYNA3D
this led to the error termination message “More than 1000 nodes in bucket.”

The formulas given by Belytschko and Lin [1985] are used to find the bucket containing a
node with coordinates (𝑥, 𝑦, 𝑧). The bucket pointers are given by
(𝑥 − 𝑥min )
𝑃𝑋 = 𝑁𝑋 ⋅ + 1, (29.65)
(xmax − xmin )

(𝑦 − 𝑦min )
PY = NY ⋅ + 1, (29.66)
(𝑦max − 𝑦min )

(𝑧 − 𝑧min )
PZ = NZ ⋅ + 1, (29.67)
(𝑧max − 𝑧min )
and are used to compute the bucket number given by
NB = PX + (PY − 1) ⋅ PX + (PZ − 1) ⋅ PX ⋅ PY. (29.68)

For each nodal point, 𝑘, in the contact surface we locate the three nearest neighboring nodes
by searching all nodes in buckets from
MAX(1, PX1), MIN(NX, PX + 1), (29.69)

MAX(1, PY1), MIN(NY, PY + 1), (29.70)

MAX(1, PZ1), MIN(NZ, PZ + 1). (29.71)

A maximum of twenty-seven buckets are searched. Nodes that share a contact segment
with k are not considered in this nodal search. By storing the three nearest nodes and
rechecking these stored nodes every cycle to see if the nearest node has changed, we avoid
performing the bucket sorting every cycle. Typically, sorting every five to fifteen cycles is
adequate. Implicit in this approach is the assumption that a node will contact just one
surface. For this reason the single surface contact (TYPE 4 in LS-DYNA) is not applicable to
all problems. For example, in metal forming applications both surfaces of the workpiece
are often in contact.

The nearest contact segment to a given node, 𝑘, is defined to be the first segment
encountered when moving in a direction normal to the surface away from 𝑘. A major
deficiency with the nearest node search is depicted in Figure 29.25 where the nearest nodes
are not even members of the nearest contact segment. Obviously, this would not be a
problem for a more uniform mesh. To overcome this problem we have adopted segment
based searching in both surface to surface and single surface contact.

29.12.2 Bucket Sorting in Surface to Surface and TYPE 13 Single Surface Contact

27-32 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

The procedure is roughly the same as before except we no longer base the bucket size on
𝐿𝑀𝐴𝑋 which can result in as few as one bucket being generated. Rather, the product of the
number of buckets in each direction always approaches 𝑁𝑆𝑁 or 5000 whichever is smaller,
NX ⋅ NY ⋅ NZ ≤ MIN(NSN, 5000), (29.72)
where the coordinate pairs (𝑥min , 𝑥max ), (𝑦min , 𝑦max ), and (𝑧min , 𝑧max ) span the entire
contact surface. In the new procedure we loop over the segments rather than the nodal
points. For each segment we use a nested DO LOOP to loop through a subset of buckets
from IMIN to IMAX, JMIN to JMAX, and to KMAX where
IMIN = MIN(PX1, PX2, PX3, PX4), (29.73)

IMAX = MAX(PX1, PX2, PX3, PX4), (29.74)

JMIN = MIN(PY1, PY2, PY3, PY4), (29.75)

KMIN = MIN(PZ1, PZ2, PZ3, PZ4), (29.76)

KMAX = MAX(PZ1, PZ2, PZ3, PZ4), (29.77)


and PX𝑘, PY𝑘, PZ𝑘 are the bucket pointers for the kth node. Figure 29.26 shows a segment
passing through a volume that has been partitioned into buckets.

We check the orthogonal distance of all nodes in the bucket subset from the segment. As
each segment is processed, the minimum distance to a segment is determined for every
node in the surface and the two nearest segments are stored. Therefore the required
storage allocation is still deterministic. This would not be the case if we stored for each
segment a list of nodes that could possibly contact the segment.

We have now determined for each node, 𝑘, in the contact surface the two nearest segments
for contact. Having located these segments we permanently store the node on these
segments which is nearest to node 𝑘. When checking for interpenetrating nodes we check
the segments surrounding the node including the nearest segment since during the steps

1 2 3 4 5

Normal vector at
node 3

Figure 29.25. Nodes 2 and 4 share segments with node 3 and therefore the two
nearest nodes are1 and 5. The nearest contact segment is not considered since its
nodes are not members of the nearest node set.

LS-DYNA DEV 10/27/16 (r:8004) 27-33 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

Nodes in buckets shown are checked for


contact with the segment
z

Figure 29.26. The orthogonal distance of each slave node contained in the box
from the segment is determined. The box is subdivided into sixty buckets.

between bucket searches it is likely that the nearest segment may change. It is possible to
bypass nodes that are already in contact and save some computer time; however, if
multiple contacts per node are admissible then bypassing the search may lead to
unacceptable errors.

29.13 Single Surface Contact Algorithms in LS-DYNA


The single surface contact algorithms evolved from the surface to surface contact
algorithms and the post contact searching follows the procedures employed for the surface
to surface contact. Type 4 contact in LS-DYNA uses the following steps where NSEG is the
number of contact segments and NSN is the number of nodes in the interface:

• Loop through the contact segments from 1 to NSEG

◦ Compute the normal segment vectors and accumulate an area weighted aver-
age at the nodal points to determine the normal vectors at the nodal points.

• Loop through the slave nodes from 1 to NSN

◦ Check all nearest nodes, stored from the bucket sort, and locate the node
which is nearest.

27-34 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

◦ Check to see if nearest node is within a penetration tolerance determined dur-


ing the bucket sort, if not, proceed to the end of the loop.
◦ For shell elements, determine if the nearest node is approaching the segment
from the positive or negative side based on the right hand rule. Project both
the node and the contact segment along the nodal normal vectors to account
for the shell thickness.
◦ Check for interpenetrating nodes and if a node has penetrated apply a nodal
point force that is proportional to the penetration depth.

End of Loop

Of course, several obvious limitations of the above procedure exists. The normal vectors
that are used to project the contact surface are meaningless for nodes along an intersection
of two or more shell surfaces (Please see the sketch at the bottom of Figure 29.27). In this
case the normal vector will be arbitrarily skewed depending on the choice of the
numbering of the connectivities of the shells in the intersecting surfaces. Secondly, by
considering the possibility of just one contact segment per node, metal forming problems
cannot be handled within one contact definition. For example, if a workpiece is
constrained between a die and a blankholder then at least some nodal points in the
workpiece must necessarily be in contact with two segments-one in the die and the other in
the workpiece. These two important limitations have motivated the development of the
new bucket sorting procedure described above and the modified single surface contact
procedure, type 13.

A major change in type 13 contact from type 4 is the elimination of the normal nodal vector
projection by using the segment normal vector as shown in Figure 29.27.

Segment numbering within the contact surface is arbitrary when the segment normal is
used greatly simplifying the model input generation. However, additional complexity is
introduced since special handling of the nodal points is required at segment intersections
where nodes may approach undetected as depicted in Figure 29.28a.

To overcome this limitation an additional logic that put cylindrical cap at segment
intersections has been introduced in contact type 13 (and a3). See Figure 29.28b.

Assuming the segment based bucket sort has been completed and closest segments are
known for all slave nodes then the procedure for processing the type 13 contact simplifies
to:

LS-DYNA DEV 10/27/16 (r:8004) 27-35 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

Type 4
v

Contact surface is based


on a nodal point normal
vector projection
Type 13
v

Contact surface is based


on segment normal
projection
Figure 29.27. Projection of the contact surface for a node approaching from
above is shown for types 4 and 13 contact.

• Loop through the slave nodes from 1 to NSN

◦ If node is in contact, check to see if the contact segment has changed and if so,
then update the closest segment information and the orientation flag which
remembers the side in contact. Since no segment orientation information is
stored this flag may change as the node moves from segment to segment.
◦ Check the closest segment to see if the node is in contact if not then proceed to
the end of the loop. If the slave node or contact segment connectivity is a
member of a shell element, project both the node and the contact segment
along the segment normal vector to account for the shell thickness. A nodal
thickness is stored for each node and a segment thickness is stored for each
segment. A zero thickness is stored for solid elements. The thickness can be
optionally updated to account for membrane thinning.
◦ Check for interpenetrating nodes and if a node has penetrated apply a nodal
point force that is proportional to the penetration depth.

End of Loop
Note that type 13 contact does not require the calculation of nodal normal vectors.

27-36 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

a b

Figure 29.28.

29.14 Surface to Surface Constraint Algorithm


The constraint algorithm that we implemented is based on the algorithm developed by
Taylor and Flanagan [1989]. This involves a two-pass symmetric approach with a
partitioning parameter, 𝛽, that is set between negative and positive unity where 𝛽 = 1 and
𝛽 = −1 correspond to one way treatments with the master surface accumulating the mass
and forces from the slave surface (for 𝛽 = 1) and visa versa (for 𝛽 = −1). The searching
algorithms are those used in the other contact algorithms for the surface to surface contact.

In this constraint approach the accelerations, velocities, and displacements are first updated
to a trial configuration without accounting for interface interactions. After the update, a
penetration force is computed for the slave node as a function of the penetration distance
Δ𝐿:
𝑚𝑠 Δ𝐿
𝐟𝑝 =
𝐧, (29.78)
Δ𝑡2
where 𝐧 is the normal vector to the master surface.

We desire that the response of the normal component of the slave node acceleration vector,
𝐚s , of a slave node residing on master segment 𝑘 be consistent with the motion of the
master segment at its contact segment (𝑠c , 𝑡c ), i.e.,
as = 𝜙1 (𝑠c , 𝑡c )𝑎1𝑛𝑘 + 𝜙2 (𝑠c , 𝑡c )𝑎2𝑛𝑘 + 𝜙3 (𝑠c , 𝑡c )𝑎3𝑛𝑘 + 𝜙4 (𝑠c , 𝑡c )𝑎4𝑛𝑘 . (29.79)

LS-DYNA DEV 10/27/16 (r:8004) 27-37 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

For each slave node in contact with and penetrating through the master surface in its trial
configuration, its nodal mass and its penetration force given by Equation (29.72) is
accumulated to a global master surface mass and force vector:

(𝑚𝑘 + ∑ 𝑚𝑘𝑠 ) 𝐚𝑛𝑘 = ∑ 𝐟𝑘𝑠 , (29.80)


𝑠 𝑠

where
𝑚𝑘𝑠 = 𝜙𝑘 𝑚𝑠 , (29.81)

𝐟𝑘𝑠 = 𝜙𝑘 𝐟𝑠 . (29.82)

After solving Equation (29.78) for the acceleration vector, 𝐚nk , we can obtain the
acceleration correction for the slave node as
𝐟p
𝐚ns = 𝐚s − . (29.83)
𝑚s

The above process is repeated after reversing the master and slave definitions. In the final
step the averaged final correction to the acceleration vector is found
1 1st pass 1 2nd pass
𝐚𝑛final = (1 − 𝛽)𝐚𝑛 + (1 + 𝛽)𝐚𝑛 , (29.84)
2 2
and used to compute the final acceleration at time 𝑛 + 1
𝐚𝑛+1 = 𝐚trial + 𝐚𝑛final , (29.85)

Friction, as described by Taylor and Flanagan [1989], is included in our implementation.


Friction resists the relative tangential velocity of the slave node with respect to the master
surface. This relative velocity if found by subtracting from the relative velocity:
𝐯r = 𝐯s − (𝜙1 𝐯𝑘1 + 𝜙2 𝐯𝑘2 + 𝜙3 𝐯𝑘3 + 𝜙4 𝐯𝑘4 ), (29.86)
the velocity component normal to the master segment:
𝐯t = 𝐯r − (𝐧 ⋅ 𝐯r )𝐧. (29.87)

A trial tangential force is computed that will cancel the tangential velocity
𝑚𝜐
𝐟∗ = 𝑠 𝑡 , (29.88)
Δ𝑡
where υt is the magnitude of the tangential velocity vector
𝜐𝑡 = √𝐯𝑡 ⋅ 𝐯𝑡 . (29.89)

The magnitude of the tangential force is limited by the magnitude of the product of the
Coulomb friction constant with the normal force defined as

27-38 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

fn = ms 𝐚ns ⋅ 𝐧, (29.90)

The limiting force is, therefore,


Fy = m|𝐟n |, (29.91)
And

𝐟𝑛+1 = 𝐟∗ if ∣𝐟∗ ∣ = 𝐹𝑦, (29.92)

𝐹𝑦 𝐟 ∗
𝐟𝑛+1 = |𝐟∗ |
if ∣𝐟∗ ∣ > 𝐹𝑦 . (29.93)

Therefore, using the above equations the modification to the tangential acceleration
component of the slave node is given by
∣𝐯s ∣
𝐚t = min (𝜇𝐚nt ⋅ 𝐧, ), (29.94)
Δ𝑡
which must act in the direction of the tangential vector defined as
𝐯
𝐧t = t . (29.95)
υt
The corrections to both the slave and master node acceleration components are:
ats = at 𝐧t , (29.96)
as ms
𝐚tk = −𝜙k 𝐧, (29.97)
mk t
The above process is again repeated after reversing the master and slave definitions. In the
final step the averaged final correction to the acceleration vector is found
1 1st pass 1 2nd pass
𝐚tfinal = (1 − 𝛽)𝐚t + (1 + 𝛽)𝐚t , (29.98)
2 2
and is used to compute the final acceleration at time 𝑛 + 1
𝐚𝑛+1 = 𝐚trial + 𝐚𝑛final + 𝐚tfinal . (29.99)

A significant disadvantage of the constraint method relative to the penalty method appears
if an interface node is subjected to additional constraints such as spot welds, constraint
equations, tied interfaces, and rigid bodies. Rigid bodies can often be used with this
contact algorithm if their motions are prescribed as is the case in metal forming. For the
more general cases involving rigid bodies, the above equations are not directly applicable
since the local nodal masses of rigid body nodes are usually meaningless. Subjecting the
two sides of a shell surface to this constraint algorithm will also lead to erroneous results
since an interface node cannot be constrained to move simultaneously on two mutually
independent surfaces. In the latter case the constraint technique could be used on one side
and the penalty method on the other.

LS-DYNA DEV 10/27/16 (r:8004) 27-39 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

The biggest advantage of the constraint algorithm is that interface nodes remain on or very
close to the surfaces they are in contact with. Furthermore, elastic vibrations that can occur
in penalty formulations are insignificant with the constraint technique. The problem
related to finding good penalty constants for the contact are totally avoided by the latter
approach. Having both methods available is possibly the best option of all.

29.15 Planar Rigid Boundaries


The rigid boundary represents the simplest contact problem and is therefore treated
separately. As shown in Figure 29.29 the boundary is flat, finite or infinite in extent and is
defined by an outward normal unit vector n with the origin of n at a corner point on the
wall if the wall is finite or at an arbitrary point on the wall if the wall extends to infinity.
The finite wall is rectangular with edges of length L and M. Unit vectors l and m lie along
these edges. A subset of nodes is defined, usually boundary nodes of the calculational
model, that are not allowed to penetrate. Let k represent one such boundary node and let
rkn+1 be the position vector from the origin of n to k after locally updating the coordinates.
Each time step prior to globally updating the velocities and accelerations we check k to
ensure that the nodes lies within the wall by checking that both inequalities are satisfied:
𝐫𝑘𝑛+1 ⋅ 𝐥 ≤ 𝐿,
(29.100)
𝐫𝑘𝑛+1 ⋅ 𝐦 ≤ 𝑀.
This test is skipped for the infinite rigid wall. Assuming that the inequality is satisfied, we
then check the penetration condition to see if k is penetrating through the wall,
𝐫𝑘𝑛+1 ⋅ 𝐧 < 0, (29.101)
and if so, the velocity and acceleration components normal to the wall are set to zero:
𝐚𝑘𝑛new = 𝐚𝑘𝑛old − (𝐚𝑘𝑛old ⋅ 𝐧)𝐧,
(29.102)
𝐯𝑘𝑛new = 𝐯𝑘𝑛old − (𝐯𝑘𝑛old ⋅ 𝐧)𝐧.
Here 𝐚𝑘 and 𝐯𝑘 are the nodal acceleration and velocity of node k, respectively. This
procedure for stopping nodes represents a perfectly plastic impact resulting in an
irreversible energy loss. The total energy dissipated is found by taking the difference
between the total kinetic energy of all the nodal points slaved to the rigid wall before and
after impact with the wall. This energy is computed and accumulated in LS-DYNA and is
printed in the GLSTAT (global statistics) file.
The tangential motion of the boundary node may be unconstrained, fully constrained, or
subjected to Coulomb friction while it is in contact with the rigid boundary.

27-40 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

n
l

Origin, if extent of stonewall is finite


Figure 29.29. Vector n is normal to the stonewall. An optional vector l can be
defined such that 𝐦 = 𝐧 × 𝟏. The extent of the stonewall is limited by defining L
and M. A zero value for either of these lengths indicates that the stonewall is
infinite in that direction.

Coulomb friction acts along a vector defined as:


𝐯𝑘𝑛new
𝐧𝑡 = , (29.103)
√𝐯𝑘𝑛new ⋅ 𝐯𝑘𝑛new
The magnitude of the tangential force which is applied to oppose the motion is given as
𝑛 𝑛
⎛𝑚𝑠 √𝐯𝑘new ⋅ 𝐯𝑘new ⎞
𝑓𝑡 = min ⎜⎜
⎜ , 𝜇|𝐟𝑛 |⎟

⎟, (29.104)
Δ𝑡
⎝ ⎠
i.e., the maximum value required to hold the node in the same relative position on the
stonewall or the product of the coefficient of friction and the magnitude of the normal force
whichever is less. In Equation (29.104), ms is the mass of the slave node and f𝑛 is the
normal force.

29.16 Geometric Rigid Boundaries


The numerical treatment of geometric rigid walls is somewhat similar to that for the finite
planar rigid walls. The geometric rigid walls can be subjected to a prescribed translational
motion along an arbitrarily oriented vector; however, rotational motion is not permitted.
As the geometric surface moves and contacts the structure, external work is generated
which is integrated and added to the overall energy balance. In addition to the external
work, plastic work also is generated as nodes contact the wall and assume the walls normal
LS-DYNA DEV 10/27/16 (r:8004) 27-41 (Contact-Impact Algorithm)
Contact-Impact Algorithm LS-DYNA Theory Manual

n
V
n v
m

l
L
L

regular prism cylinder


n
m v

R
n
v

l
flat surface sphere
Figure 29.30. Vector n determines the orientation of the generalized stonewalls.
For the prescribed motion options the wall can be moved in the direction V as
shown.

velocity at the point of contact. Contact can occur with any of the surfaces which enclose
the volume. Currently four geometric shapes are available including the rectangular
prism, the cylinder, flat surface, and sphere. These are shown in Figure 29.30.

29.17 VDA/IGES Contact


This cabability allows the user to read VDA/IGES surfaces directly into LS-DYNA for
analysis as contact surfaces. No mesh generation is required, and the contact is performed
against the analytic surface. LS-DYNA supports the VDA standard and an important
subset of the IGES entities including:

•#100 Circle arc


•#102 Composite Curve
•#106 Copious data
•#110 Lines
•#112 Parametric polynomial curve
•#114 Parametric polynomial surface
•#116 Points

27-42 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

•#126 NURBS Curves


•#128 NURBS Surfaces
•#142 Curve on Parametric Surface
•#144 Trimmed Parametric Surfaces
•#402 form 7-group
•#406 form 15-associate name

First, the user must specify which VDA/IGES surfaces, faces, and groups should be
attached to each material. This is done primarily through a special input file. Faces,
surfaces, and groups from several different VDA/IGES input files can be combined into
groups that later can be refered to by a user specified alias. For example, suppose a simple
sheetmetal forming problem is going to be run. The user might have an input file that
looks like this:

file punch.vda punch.bin {


alias punch { grp001 }
}
file die.vda die.bin {
alias part1 { fce001 sur002 }
alias part2 { fce003 }
}
file die2.vda die2.bin {
alias part3 { fce004 }
}
file holder.vda holder.bin {
alias holder { sur001 sur002 }
}
alias die { part1 part2 part3 }
end

In this example, the user has specified that the punch will be made up of the group
"grp001" from the file "punch.vda". The VDA file is converted to a binary file "punch.bin".
If this simulation is ever rerun, the VDA input can be read directly from the binary file
thereby significantly reducing startup time. The die in this example is made up of several
surfaces and faces from 2 different VDA files. This format of input allows the user to
combine any number of faces, surfaces, and groups from any number of VDA files to
define a single part. This single part name is then referenced within the LS-DYNA input
file.

The contact algorithm works as follows. For the sake of simplicity, we will refer to one
point as being slaved to a single part. Again, this part will in general be made up of several
VDA surfaces and faces. First, the distance from the point to each VDA surface is
computed and stored. For that surface which is nearest the point, several other parameters
are stored such as the surface coordinates of the near point on the surface. Each time step
of the calculation this information is updated. For the nearest surface the new near point is
LS-DYNA DEV 10/27/16 (r:8004) 27-43 (Contact-Impact Algorithm)
Contact-Impact Algorithm LS-DYNA Theory Manual

slave point
P (x, y, z)

s
Figure 29.31. The geometry of the patch is a function of the parametric
coordinates 𝑠 and 𝑡.

calculated. For all other surfaces the distance the point moves is subtracted from the
distance to the surface. This continually gives a lower bound on the actual distance to each
VDA surface. When this lower bound drops below the thickness of the point being
tracked, the actual distance to the surface is recalculated. Actually, if the nearest surface is
further away from the point than some distance, the near point on the surface is not tracked
at all until the point comes close to some surface. These precautions result in the distance
from the point to a surface having to be totally recomputed every few hundred timesteps,
in exchange for not having to continually track the point on each surface.

To track the point on the nearest surface, a 2D form of Newton's method is used. The
vector function to be solved specifies that the displacement vector from the surface to the
point should be parallel to the surface normal vector. The surface tangent vectors are
computed with respect to each of the two surface patch parameters, and the dot product
taken with the displacement vector. See Figure 29.31 and Equation (29.104).
∂𝐪 ∂𝐪
(𝐩 − 𝐪) ⋅ = 0 and (𝐩 − 𝐪) ⋅ = 0. (29.105)
∂s ∂t
This vector equation is then solved using Newton's method as in Equation (1.106).
𝐪𝑖+1 = 𝐪𝑖 − (𝐅′ )−1 𝐅, (29.106)
where
∂𝐪
⎛(𝐩
⎜ − 𝐪) ⋅ ⎞
𝐅(𝑠, 𝑡) = ⎜
⎜ ∂𝑠 ⎟

⎟. (29.107)

⎜ ∂𝐪⎟⎟
⎝ (𝐩 − 𝐪) ⋅
∂𝑡 ⎠
27-44 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Contact-Impact Algorithm

previous location

new location

new nearest point

previous nearest
point

Figure 29.32. Newton iteration solves for the nearest point on the analytical
surface.

The convergence is damped in the sense that the surface point is not allowed to jump
completely outside of a surface patch in one iteration. If the iteration point tries to leave a
patch, it is placed in the neighboring patch, but on the adjoining boundary. This prevents
the point from moving merely continuous (i.e., when the surface has a crease in it).
Iteration continues until the maximum number of allowed iterations is reached, or a
convergence tolerance is met. The convergence tolerance (as measured in the surface patch
parameters) varies from patch to patch, and is based on the size and shape of the patch.
The convergence criterion is set for a patch to ensure that the actual surface point has
converged (in the spatial parameters x, y, and z) to some tolerance.

29.18 Simulated Draw Beads


The implementation of draw beads is based on elastic-plastic interface springs and nodes-
to-surface contact. The area of the blank under the draw bead is taken as the master
surface. The draw bead is defined by a consecutive list of nodes that lie along the draw
bead. For straight draw beads only two nodes need to be defined, but for curved beads
sufficient nodes must be used to define the curvature. The draw bead line is discretized
into points that become the slave nodes to the master surface. The spacing of the points is
determined by LS-DYNA such that several points lie within each master segment. This is
illustrated in Figure 29.32. The dense distribution of point leads to a smooth draw bead
force distribution which helps avoid exciting the zero energy (hourglass) modes within the

LS-DYNA DEV 10/27/16 (r:8004) 27-45 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

integration points along drawbead line

3 4

1
points 1, 2, 3, and 4 define drawbeads
Figure 29.33. The drawbead contact provides a simple way of including
drawbead behavior without the necessity of defining a finite element mesh for the
drawbeads. Since the draw bead is straight, each bead is defined by only two
nodes.

shell elements in the workpiece. A three-dimensional bucket search is used for the contact
searching to locate each point within a segment of the master surface.

The nodes defining the draw beads can be attached to rigid bodies by using the extra nodes
for rigid body input option. When defining draw beads, care should be taken to limit the
number of elements that are used in the master surface definition. If the entire blank is
specified the CPU cost increases significantly and the memory requirements can become
enormous. An automated draw bead box, which is defined by specifying the part ID for
the workpiece and the node set ID for the draw bead, is available. The automated box
option allows LS-DYNA determine the box dimensions. The size of this box is based on the
extent of the blank and the largest element in the workpiece as shown if Figure 29.34.

The input for the draw beads requires a load curve giving the force due to the bending and
unbending of the blank as it moves through the draw bead. The load curve may also
include the effect of friction. However, the coulomb friction coefficients must be set to zero
if the frictional component is included in the load curve. If the sign of the load curve ID is
positive the load curve gives the retaining force per unit draw bead length as a function of
displacement, δ. If the sign is negative the load curve defines the maximum retaining force
versus the normalized position along the draw bead. This position varies from 0 (at the
origin) to 1 (at the end) along the draw bead. See Figures 29.35 and 29.36.

27-46 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

Figure 29.34. The draw bead box option automatically size the box around the
draw bead. Any segments within the box are included as master segments in the
contact definition.

When friction is active the frictional force component normal to the bead in the plane of the
work piece is computed. Frictional forces tangent to the bead are not allowed. The second
load curve gives the normal force per unit draw bead length as a function of displacement,
δ. This force is due to bending the blank into the draw bead as the binder closes on the die
and represents a limiting value. The normal force begins to develop when the distance
between the die and binder is less than the draw bead depth. As the binder and die close
on the blank this force should diminish or reach a plateau. This load curve was originally
added to stabilize the calculation.

As the elements of the blank move under the draw bead, a plastic strain distribution
develops through the shell thickness due to membrane stretching and bending. To account
for this strain profile an optional load curve can be defined that gives the plastic strain
versus the parametric coordinate through the shell thickness where the parametric
coordinate is defined in the interval from –1 to 1. The value of the plastic strain at each
through thickness integration point is interpolated from this curve. If the plastic strain at
an integration point exceeds the value of the load curve at the time initialization occurs, the
plastic strain at the point will remain unchanged. A scale factor that multiplies the shell
thickness as the shell element moves under the draw bead can also be defined as a way of
accounting for any thinning that may occur.

LS-DYNA DEV 10/27/16 (r:8004) 27-47 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

Force
positive load curve ID

Penetration distande, δ

b
Force

negative load curve ID

0 1
Normalized draw bead length
Figure 29.35. Draw bead contact model defines a resisting force as a function of
draw bead displacement.

29.19 Edge to Edge Contact


Edge to edge contact can be important in some simulations. For example, if a fan blade
breaks away from the hub in a jet turbine contact with the trailing blade will likely be along
the edges of the blades. Edge to edge contact requires a special treatment since the nodal
points do not make contact with the master segment which is the basis of the conventional
contact treatments. Currently all automatic type contact possess edge-to-edge capabilities
and therefore contact type 22 is only useful with those contact that do not possess this
capability. All contact using the segment-based formulation have edge to edge capabilities.

27-48 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

D, depth of draw bead

δ F = Ffriction +Fbending

Figure 29.36. Draw bead contact model defines a resisting force as a function of
draw bead displacement.

The basis of single edge contact is the proven single surface formulation and the input is
identical. The definition is by material ID. Edge determination is automatic. It is also
possible to use a manual definition by listing line segments. The single edge contact is type
22 in the structured input or *CONTACT_SINGLE_EDGE in the keyword input.

LS-DYNA DEV 10/27/16 (r:8004) 27-49 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

Figure 29.37. Contact between edges requires a special treatment since the nodes
do not make contact.

This contact only considers edge to edge contact of the type illustrated in Figure 29.38.
Here the tangent vectors to the plane of the shell and normal to the edge must point to each
other for contact to be considered.

29.20 Beam to Beam Contact


In the beam to beam contact the contact surface is assumed to be the surface of a cylinder as
shown in Figure 29.39. The diameter of the contact cylinder is set equal to the square root

tangent vectors in
plane of shell

Figure 29.38. Single edge contact considers contact between two edges whose
normals point towards each other.

of the area of the smallest rectangle that contains the cross section to avoid tracking the
27-50 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Contact-Impact Algorithm

orientation of the beam within the contact algorithm. Contact is found by finding the
intersection point between nearby beam elements and checking to see if their outer surfaces
overlap as seen in Figure 29.40. If the surfaces overlap the contact force is computed and is
applied to the nodal points of the interacting beam elements.

Actual beam cross section

Contact surface

Figure 29.39. Beam contact surface approximation.

intersection point where forces are applied

Figure 29.40. The forces are applied at the intersection point.

LS-DYNA DEV 10/27/16 (r:8004) 27-51 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

29.21 Mortar contact


The Mortar contact was originally implemented as a forming contact intended for stamping
analysis but has since then evolved to become a general purpose contact algorithm for
implicit time integration. The Mortar option is today available for automatic single- and
surface-to-surface contacts with proper edge treatment, beam contact, and optional features
include tie, tiebreak and interference. Contact is often the one feature that overturns the
implicit performance, so to facilitate debugging of the Mortar contacts there is substantial
information on penetrations written to the LS-DYNA message files. The Mortar contact is a
penalty based segment-to-segment contact with finite element consistent coupling between
the non-matching discretization of the two sliding surfaces and the implementation is
based on Puso and Laursen [2004a,b]. This consistency, together with a differentiable
penalty function for penetrating and sliding segments, assert the continuity and (relative)
smoothness in contact forces that is appealing when running implicit analyses. The
algorithm is primarily focusing on accuracy and robustness, and the involved calculations
associated with this aim make it expensive enough to be first and foremost recommended
for implicit analysis. For one thing, the bucket sort is performed in every cycle even in
explicit analysis, in contrast to all other contact algorithms. There are numerous details in
the implementation that simply cannot be explained without making the presentation
incomprehensible, the intention here is to summarize the general concepts of the theory
behind the implementation and draw upon this to make some general recommendations
on usage.

29.21.1 Kinematics
The Mortar contact is theoretically treated as a generalized finite element where each

ns

4
Slave segment 3
Ts

1 Π
X3 2
X2

O Master Segment
X1
Figure 29.41. Illustration of Mortar segment to segment contact

element in this context consists of a pair of contact segments. The friction model in the
Mortar contact is a standard Coulomb friction law. Each of the two segments has its iso-
27-52 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Contact-Impact Algorithm

parametric representation inherited from the underlying finite element formulation, so the
coordinates for the slave and master segments can be written
𝐱s = 𝑁s𝑖 (𝜉 , 𝜂)𝐱𝑖
𝑗
(29.21.108)
𝐱m = 𝑁m (𝜉 , 𝜂)𝐱𝑗 ,
where summation over repeated indices is implicitly understood, i.e., over the nodes. The
kinematics for the contact element can be written as the penetration
𝑑 = 𝐧sT (𝐱s − 𝐱̅m ), (29.21.109)
where 𝐧s is the slave segment normal and 𝐱̅m is the projected point on the master segment
along the slave segment normal. The element is only defined for the intersection between
the slave and master segment and for points where 𝑑 > 0, this domain is denoted Π and is
illustrated by gray in the Figure above. The sliding rate 𝐬̇ is similarly defined as
𝐬̇ = 𝐓sT (𝐱̇s − 𝐱̅m
̇ ), (29.21.110)
where 𝐓s are two co-rotational basis vectors pertaining to the slave segment.

29.21.2 Constitutive relation


The contact pressure is given by the constitutive law
𝑑
𝜎n = 𝛼𝛽𝑠 𝛽𝑚 𝜀𝐾s 𝑓 ( ), (29.21.111)
𝜀𝑑𝑐𝑠
where
𝛼 = stiffness scaling factor (SFS*SLSFAC)
𝐾s = stiffness modulus of slave segment
𝜀 = 0.03
𝑑𝑐𝑠 = characteristic length of slave segment
𝛽𝑠 = stiffness scale factor of slave segment (=1 unless specifically stated)
𝛽𝑚 = stiffness scale factor of master segment (=1 unless specifically stated)

and
⎧ 1 2 𝑑max
{ 𝑥 𝑥<
{ 4 2𝜀𝑑𝑐𝑠
𝑓 (𝑥) = ⎨ . (29.21.112)
{cubic function that depends on IGAP 𝑑max
{ ≤𝑥
⎩ 2𝜀𝑑𝑐𝑠
where 𝑑max is the maximum penetration to be given below. The Coulomb friction law is
expressed in terms of the tangential contact stress
𝐬 |𝐬|
𝛔t = 𝜇𝜎n 𝑔 ( ), (29.21.113)
|𝐬| 𝜇𝑑
where 𝜇 is the friction coefficient and

LS-DYNA DEV 10/27/16 (r:8004) 27-53 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

⎧ 𝑥 𝑥 ≤1−𝜀
{
{ 2
𝑔(𝑥) = ⎨1 − 𝜀 (1 + 𝜀 − 𝑥) 1 − 𝜀 < 𝑥 ≤ 1 + 𝜀. (29.21.114)
{
{ 4 𝜀
⎩ 1 1+𝜀<𝑥
The update of 𝐬 is done incrementally and is at the end of the step modified so that
|𝐬| ≤ 𝜇𝑑(1 + 𝜀) (29.21.115)
after the contact update.

29.21.3 Contact nodal forces


From the contact stress, the contact nodal forces are determined by the principle of virtual
work
𝐟s𝑖 = 𝐧s ∫ 𝑁s𝑖 𝜎n 𝑑Π + 𝐓s ∫ 𝑁s𝑖 𝛔t 𝑑Π
Π Π (29.21.116)
𝑗 𝑗 𝑗
𝐟m = −𝐧s ∫ 𝑁m 𝜎n 𝑑Π − 𝐓s ∫ 𝑁m 𝛔t 𝑑Π,
Π Π
where the subscript s and m stands for the slave and master nodal forces, respectively.

29.21.4 Treatment of beams, sharp solid and shell edges


The automatic Mortar contacts support contact with the lateral surface and end tips of
beams as well as edges of shell elements and sharp edges of solids. The concept of sharp
solid edges will be defined below. The theory presented above is in this case applied to
dummy segments corresponding to a faceted representation of the beam lateral surface and
the edges of the solid and shell elements, respectively, as indicated in Figure 27-42. For a
beam element the contact surface is represented by 14 faceted segments encapsulating a
cylinder with the same length and volume as the beam element itself. This implies that all
beam elements are assumed to have a circular cross section for the contact. The edges of
the shell element surface are identified assuming the user contact definition (slave or
master) consists of the entire physical component (metal sheet) in question. It is therefore
recommended to
Define the contacts using part or part sets or otherwise false edges may be created in
the interior of the component.
An edge contact element is created by extruding the shell edge in the direction of the shell
normal by a distance corresponding to the shell thickness with appopriate adjustments for
irregular geometries. To this end it is important to state that the kinematics in creating
these dummy segments do not involve rotational degrees of freedom and whence the
Mortar contact does not possess a proper offset feature, i.e., one cannot create a moment on
a shell due to sliding friction in a finite element consistent manner. The shell edge
treatment is to be seen as a simplified treatment just to incorporate a contact resistance for
these geometries. For beam elements however, rotational degrees of freedom are included
in a consistent way, thus causing beam elements to rotate with respect to their axes when
tangential friction forces are applied to the lateral surface. A sharp solid edge is detected
when the angle 𝜃 between the normals of two adjacent contact segments is larger than 𝜋/3,

27-54 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

and in this case the edge is smoothed by adding 4 segments between the two nodes
common to the two contact segments, while at the same time adjusting the size of the two
main contact segments. The effect is a rounded representation of the edge and a smoother
contact response, see Figure 27-42. The size of the smoothing is at most 5% of the size of
the smallest of the two contact segments, so the effect is reduced with mesh refinement.
Since the contact area of these edge segments may be small, the stiffness of the contact is for
those scaled by the factor 𝛽 = √3cot(𝜋−𝜃) where 𝜃 is the initial angle between the main
2
segment normals. This scale factor is applied for both the slave and master side, meaning
that if two right angled edges come into contact the stiffness is scaled by 𝛽𝑠 𝛽𝑚 =
𝜋−𝜋 𝜋−𝜋
√3 cot ( 2
) √3 cot ( 2
) = 9. Whether this is sufficient to handle most common
2 2
situations is currently unknown, so this design decision is subject for change in the future.
If a contact segment has two sharp edges with a common node, then a segment is created at
the location of that node to account for contact with the corner of the solid element
geometry. The stiffness scale factor for a corner node is the average of the scale factor for
the connected edges. The motivation behind the solid edge smoothing is two-fold; for one
thing it adds the feature of resisting penetration between a solid edge/corner and other
geometries and then it also aids establishing a physical contact state when solid elements
slide off sharp geometrical objects. More specifically it eliminates the ambiguity of which
contact segments are in contact with which, and presumably prevents sudden spikes in the
contact force, this is also illustrated in Figure 27-42.
It is important to stress that the automatic Mortar contact surfaces are always located on the
outer geometry for both slave and master sides, i.e., contact does not occur on the mid-
surface of shells. A common modelling problem is illustrated in Figure 27-43 that results in
extremely large contact stress unless the ignore flag is appropriately used. The forming
contact is treated differently not only in that shell edge or beam contact is not supported.
Here any shell master surface must be rigid with its segment normals oriented towards the
slave side of the contact. Furthermore the contact occurs here on the mid-surface in
contrast to the automatic option, while contact on the slave side still occurs on the outer
geometry. The segment orientation of the (deformable) slave side is on one hand arbitrary,
but if it consists of shell elements contact can only occur on one side for a given contact
definition. In a forming application for instance, the contact between the tool and blank
and the contact between the die and blank have to be defined using two different contact
interfaces since these contacts typically occur on different sides of the blank. Forming solid
master surfaces may be rigid or deformable, thus allowing for effects of tool deformation
and/or cooling effects.

LS-DYNA DEV 10/27/16 (r:8004) 27-55 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

Figure 27-42 29 Faceted representation of beams (including 4 quads


representing a tip end), shell and sharp solid edges, respectively, the
segment geometry indicated by red. The contact surface representation of
three cubic solid elements is illustrated, with a somewhat exaggerated
smoothing for illustration purposes. The effect of edge smoothing in
contact is illustrated in a section cut bottom right, if the red objects slides to
the right and the blue objects are fixed, the sudden detection of a parasitic
contact indicated by the double arrow results in a jump in the contact force,
this is alleviated by smoothing of the edges below.

27-56 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

Correct modelling if IGNORE is used, master


edge shows excessive penetration but contact
surface is adjusted

Incorrect modelling if IGNORE is not used,


master edge shows excessive penetration,
large contact stress

Correct modelling regardless of IGNORE,


master edge is located on outer surface of slave
segment, zero contact stress

Figure 27-43 Intersected view of shells in edge-to-surface contact. Shell mid-surfaces


indicated by dashed lines, outer surfaces by solid lines. Slave shell nodes are blue,
master shell nodes are red and initial volume of penetration is shaded.

29.21.5 Characteristic length and contact release


The characteristic lengths 𝑑𝑐𝑠 and 𝑑𝑐𝑚 of the slave and master sides, respectively, are
important in Mortar contact since they affect the contact stiffness but also because it
determines the maximum allowed penetration between two arbitrary segments. To be
more specific, two segments cannot penetrate more than 95% of the average characteristic
lengths, penetrations larger than that will not be detected. In mathematical terms this can
be stated as
𝑑𝑠 + 𝑑𝑐𝑚
𝑑max = 0.95 𝑐
2

(29.117)

where 𝑑max is the maximum penetration, 𝑑𝑐𝑠 is the characteristic length of the slave segment
and 𝑑𝑐𝑚 is the characteristic length of the master segment. This is illustrated in Figure 29.44
that shows the contact stress as function of the relative penetration. To minimize the risk of
releasing contacts one could increase the stiffness parameter SFS, but this may lead to
worse convergence in implicit analysis. To this end, the IGAP parameter can be used to
stiffen the contact for large penetrations without affecting moderate penetrations, this is
also illustrated in Figure 29.44 and can be used if the contact pressure is locally very high.
This also highlights the following important fact:

The Mortar contact has no “stick” option for improving implicit convergence.

LS-DYNA DEV 10/27/16 (r:8004) 27-57 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

4
IGAP=1
3.5 IGAP=2
IGAP=5
3 IGAP=10
Contact Stress
2.5

1.5

0.5

0
0 0.2 0.4 0.6 0.8 1
Penetration

Figure 29.44. Mortar contact stress as function of penetration relative 2𝑑max .

This may on one hand lead to worse convergence characteristics but on the other rules out
sticky behavior and underreport of contact forces in the ascii database.

The characteristic length is for shells the shell thickness, whereas for solids it is the
smallest element size in the part that the segment belongs to. The latter may lead to
unrealistically high or low contact stiffness, or it may result in a too small maximum
penetration depth, all depending on the mesh. For a mesh independent behavior the user
may set SST to the characteristic length, which should correspond to some physical
member size in the model. A high value will provide a soft contact, and vice versa. It is
recommended to
Separate shells and solids into different contact definitions
Explicitly set SST to a characteristic length for the solid definitions

29.21.6 Outputs for debugging implicit models


In implicit analysis it is almost inevitable to run into convergence problems, especially
when contacts are involved. When this happens the user usually craves for information on
what’s gone wrong. For the Mortar contact, detailed information on penetration distance
and potential contact release (i.e., penetration becomes too large for the contact to be
detected in subsequent steps) can be requested through MINFO = 1 on CONTROL_OUT-
PUT. With this option information on largest penetration, both absolute and relative, is
given in the message files after each converged step, including a warning if penetration is
close to being released. It also reports the elements with largest penetrations which makes
it easy to locate critical areas of the model in LS-PrePost. Contact release should be
avoided or otherwise results may be useless.

27-58 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

Contact surface augment SST × (SFST-1)/2

Contact surface augment (SST × SFST-T)/2

Element thickness T

Figure 29.45. Illustration of contact surface location for automatic Mortar


contact, solids on top and shells below.

On the other side of the spectrum, poor convergence could be due to a too stiff
contact. Since the stiffness for a Mortar contact segment pair only depends on the slave
segment it is recommended to
Put weak parts on the slave side
If a steel part is in contact with rubber for instance, the rubber part should be put as slave
side in the contact definition. One way of find contacts that cause poor convergence in
implicit is to turn on D3ITCTL on CONTROL_IMPLICIT_SOLUTION and RESPLT on
DATABASE_EXTENT_BINARY, which allows the user to isolate areas in the model where
convergence is poor. Not rarely this is due to contacts.

29.21.7 Initial penetrations


Initial penetrations are always reported in the message files, including the maximum
penetration and how initial penetrations are to be handled. The IGNORE flag governs the
latter and the options are

IGNORE < 0 See explanation for the corresponding positive value, the only
difference is that contact between segments belonging to the same part
is not treated
IGNORE = 0 Initial penetrations will give rise to initial contact stresses, i.e., the slave
contact surface is not modified
IGNORE = 1 Initial penetrations will be tracked, i.e., the slave contact surface is
translated to the level of the initial penetrations and subsequently
follow the master contact surface on separation until the unmodified
level is reached
IGNORE = 2 Initial penetrations will be ignored, i.e., the slave contact surface is
translated to the level of the initial penetrations, optionally with an
initial contact stress governed by MPAR1
IGNORE = 3 Initial penetrations will be removed over time, i.e., the slave contact
surface is translated to the level of the initial penetrations and pushed
back to its unmodified level over a time determined by MPAR1

LS-DYNA DEV 10/27/16 (r:8004) 27-59 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

IGNORE = 4 Same as IGNORE = 3 but it allows for large penetrations by also setting
MPAR2 to at least the maximum initial penetration

The use of IGNORE depends on the problem, if no initial penetrations are present
there is no need to use this parameter at all. If penetrations are relatively small in relation
to the maximum allowed penetration, then IGNORE = 1 or IGNORE = 2 seems to be the
appropriate choice. For IGNORE = 2 the user may specify an initial contact stress small
enough to not significantly affect the physics but large enough to eliminate rigid body
modes and thus singularities in the stiffness matrix. The intention with this is to constrain
loose parts that are initially close but not in contact by pushing out the contact surface
using SFST and applying the IGNORE = 2 option. Increasing SFST to a number larger than
unity will push the contact surface outside the geometry and contact will be detected
accordingly, see Figure. It is at least good for debugging problems with many singular
rigid body modes.

IGNORE = 3 is the Mortar interference counterpart, used for instance if there is a


desire to fit a rubber component in a structure or for eliminating initial penetrations by
simulation. With this option the contact surfaces are restored linearly in time from the
beginning of the simulation to the time specified by MPAR1. If the intention is to eliminate
initial penetrations completely, and since contact penetrations are unavoidable to some
extent, it may also in this case be of importance to use SFST to reduce the possibility that
the actual geometry is penetrated. If using a single surface definition on a complicated
geometry with many parts, a negative value of IGNORE could be of interest, since the
Mortar contact may otherwise detect spurious contacts between segments belonging to the
same part.

A drawback with IGNORE = 3 is that initial penetration must be smaller than half
the characteristic length of the contact or otherwise they will not be detected in the first
place. For this reason IGNORE = 4 was introduced where initial penetrations may be of
arbitrary size, but it requires that the user provides crude information on the level of
penetration of the contact interface. This is done in MPAR2 which must be larger than the
maximum penetration or otherwise an error termination will occur. IGNORE = 4 only
applies to solid elements at the moment.

27-60 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

𝑠1 𝑠2

𝑚2

𝑚1

𝑑(𝑡) 𝐴
𝑠1 vs 𝑚2 𝒙(𝑡) 𝒏
𝒏𝑠 𝒕
𝒙2
𝒙1 𝒚2
𝒚1
𝒏𝑚
𝒚(𝑡)
𝑠1 vs 𝑚1
𝑠2 vs 𝑚2

27-4629 2D mortar contact, 2 slave segments in contact with 2 master


segments results in three separate treatments.

29.21.8 2D Mortar contact


Automatic single and surface-to-surface Mortar contact is available and here described in
detail as an attempt to illustrate it in a setting that is hopefully easier to understand. In
Figure 27-46 (top) the contact between 2 slave and 2 master segments is shown as an
example. This particular configuration results in the treatment of 3 slave vs master contact
pairs (bottom), and for the mathematical treatment we refer to this figure as an illustration.

Kinematics
First a common tangential direction is determined, this is given as
𝒙 + 𝒚2 − 𝒙1 − 𝒚1
𝒕= 2
∥𝒙2 + 𝒚2 − 𝒙1 − 𝒚1 ∥

(29.118)
)

LS-DYNA DEV 10/27/16 (r:8004) 27-61 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

and the corresponding normal 𝒏 is perpendicular to 𝒕, with the direction convention as


illustrated. We can define a coordinate 𝑡 along this tangential direction, the origin can be
chosen arbitrarily, and then 𝒙(𝑡) and 𝒚(𝑡) are the projection coordinate along 𝒏 onto the
slave and master segment, respectively. Obviously there is a finite interval 𝑡 ∈ [𝑡1̃ , 𝑡2̃ ]
where both 𝒙(𝑡) and 𝒚(𝑡) is well-defined, and on this interval we can define the penetration
as
𝑑(𝑡) = 𝒏𝑇 (𝒙(𝑡) − 𝒚(𝑡)).

(29.119)

The overlapped interval is then further reduced to 𝑡 ∈ [𝑡1 , 𝑡2 ] for which 𝑑(𝑡) ≥ 0, note that
for the 𝑠2 vs 𝑚2 situation these two intervals are not the same. The interval [𝑡1 , 𝑡2 ] is
indicated by red in the illustration. This completes the kinematics in the normal direction.
In the tangential direction we need to define the kinematics for sliding and we do this by
associating a history variable to each slave segment. To this end, 𝑆 as a weighted measure
of the distance a slave segment has slid along the master surface and is defined as
𝑇
𝑆 = 𝑆𝑛−1 + ∑ ∫ {𝑠(𝑡) − 𝒕𝑛−1 (𝒙𝑛−1 (𝑡) − 𝒚𝑛−1 (𝑡))}𝑑𝐴𝑖
𝑖 𝐴𝑖

(29.120)
)

where 𝑠(𝑡) = 𝒕 𝑇 (𝒙(𝑡) − 𝒚(𝑡)) and the subscript 𝑛 − 1 refers to the corresponding value in
the previous step. The integral is taken over the domain 𝐴𝑖 of the intersected interval
between the slave segment and master segment 𝑖, accounting for plane strain or axial
symmetry. Note that the slave segment can be in contact with several master segments,
whence the sum. Further noting that by construction the first term in the integrand, 𝑠(𝑡) =
0, we can simplify this to
𝑇
𝑆 = 𝑆𝑛−1 − ∑ ∫ 𝒕𝑛−1 (𝒙𝑛−1 (𝑡) − 𝒚𝑛−1 (𝑡))𝑑𝐴𝑖 .
𝑖 𝐴𝑖

(29.121)
)

In the illustrated situation, slave segment 1 would get sliding contributions from master
segments 1 and 2, while slave segment 2 only from master segment 2. Likewise we define a
weighted penetration

27-62 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

𝐷 = ∑ ∫ 𝑑(𝑡)𝑑𝐴𝑖 .
𝑖 𝐴𝑖

(29.122)
)

which is used together with 𝑆 to define the friction law below.

Constitutive relations
For simplicity, we drop the explicit dependence on 𝑡 from now on and define the contact
stress (pressure) as
𝜎𝑛 = 𝐾 𝑑2

(29.123)

where 𝐾 is the contact stiffness defined as


𝛼 1 2𝐾𝑠 𝐾𝑚
𝐾= 𝑓 (𝒏𝑇𝑠 𝒏𝑚 ).
0.01 𝑇𝑠 𝑇𝑚 𝐾𝑠 + 𝐾𝑚

(29.124)

Furthermore, 𝛼 = PSF ∗ SLSFAC is a stiffness scale factor and 𝑇𝑠/𝑚 and 𝐾𝑠/𝑚 are
characteristic lengths and material stiffnesses of the slave and master segments,
respectively. The function 𝑓 is used to linearly reduce stiffness for segments that are not
parallel, note that 𝒏𝑠 and 𝒏𝑚 are the normals to the slave and master segments, and is given
as
⎧ 1
{ 0 − ≤𝑥
{ 2
{
{ 1 + 2𝑥 √ 3 1
{
𝑓 (𝑥) = ⎨ − ≤𝑥<− .
{1 − √3 2 2
{
{ √3
{
{ 1 𝑥 < − (29.125)
⎩ 2

LS-DYNA DEV 10/27/16 (r:8004) 27-63 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

The friction stress is defined as


𝑆
𝜎𝑡 = 𝜇𝜎𝑛 𝑔 ( )
𝜇𝐷

(29.126)

with 𝜇 being the Coulomb friction coefficient and 𝑔 is a continuously differentiable function
defined as
⎧ 1, 𝑥 ≥ 1.03
{
{ 25
{
{ 1 − (𝑥 − 1.03)2 , 0.97 ≥ 𝑥 > 1.03
{ 3 (29.127)
𝑔(𝑥) = ⎨ 𝑥, −0.97 ≥ 𝑥 > 0.97
{ 25 )
{
{ −1 + (𝑥 + 1.03)2 , −1.03 ≥ 𝑥 > −0.97
{
{ 3
⎩ −1, −1.03 > 𝑥

The interpretation of this law is that the magnitude of friction stress 𝜎𝑡 is at most 𝜇𝜎𝑛 , and
the fraction thereof is determined by the appropriate relation between the accumulated
sliding and penetration. Upon convergence, to yield a proper friction behavior, 𝑆 is
updated for the next step according to 𝑆𝑛 = max(−1.03𝜇𝐷, min(1.03𝜇𝐷, 𝑆)).

Nodal forces
The nodal force contribution from a given segment pair is determined from the principle of
virtual work, i.e.,
𝛿𝑊 = ∫ 𝜎𝑛 𝛿𝑑 𝑑𝐴 + ∫ 𝜎𝑡 𝛿𝑠 𝑑𝐴.
𝐴 𝐴

(29.128)

Here 𝛿 is the variation operator, and the independent variables subject to this variation are
the nodal coordinates. Thus, replacing the left hand side with the expression involving
nodal forces and coordinates, and using (29.123) and (29.126) for the right hand side we get
∑ 𝒇𝐼𝑇 𝛿𝒙𝐼 = 𝐾 ∫ 𝑑2 𝛿𝑑 𝑑𝐴 + 𝜇𝐾𝑔 ∫ 𝑑2 𝛿𝑠 𝑑𝐴.
𝐼 𝐴 𝐴

(29.129)
)

27-64 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

The nodal forces are then readily identified as


𝜕𝑑 𝜕𝑠
𝒇𝐼 = 𝐾 ∫ 𝑑 2 𝑑𝐴 + 𝜇𝐾𝑔 ∫ 𝑑2 𝑑𝐴.
𝐴 𝜕𝒙𝐼 𝐴 𝜕𝒙𝐼
(29.130)
)

where the exact expressions for the integrals in terms the nodal coordinates are, although
straightforward to derive, a bit lengthy and therefore omitted.

Stiffness matrix
The stiffness matrix is the second variation of the virtual work expression (29.128) where
we neglect the variation of the overlapped area and assume the variation can be taken on
the integrand directly
𝜕𝑔 1 𝑆 (29.131)
∆𝛿𝑊𝐼 = 2𝐾 ∫𝐴 𝑑 ∆𝑑 𝛿𝑑 𝑑𝐴 + 2𝜇𝐾𝑔 ∫𝐴 𝑑 ∆𝑑 𝛿𝑠 𝑑𝐴 + 𝜇 𝜕𝑥 {𝜇𝐷 ∆𝑆 − 𝜇𝐷 2
∆𝐷} ∫𝐴 𝜎𝑛 𝛿𝑠 𝑑𝐴
)
Here ∆ is again the variation operator, different notation to distinguish it from 𝛿 but
performing the exact same thing. At this point we emphasize that the (geometric) terms
involving ∆𝛿𝑑 and ∆𝛿𝑠 have been deliberately excluded as they, albeit being symmetric,
contribute to the indefiniteness of the tangent matrix. In (29.131), the first term on the right
hand side is the normal-normal interaction and is nicely symmetric, the remaining terms
come from friction (normal-tangent and tangent-tangent interaction) and needs
symmetrization and further simplification. To this end we neglect any terms involving
∆𝑑 𝛿𝑠 and ∆𝐷𝛿𝑠 which means that it remains to deal with the term involving ∆𝑆 𝛿𝑠. The
simplifications made are to approximate
1 (29.132)
𝜎𝑛 ≈ ∫ 𝜎𝑛 𝑑𝐴
𝐴 𝐴 )
in the last integral, and then neglect all terms in ∆𝑆 that do not pertain to the present slave
master segment pair. This results in

𝜕𝑔 1 1 (29.133)
∆𝒙𝐼𝑇 𝑲𝐼𝐽𝑇 𝛿𝒙𝐽 ≈ 2𝐾 ∫ 𝑑 ∆𝑑 𝛿𝑑 𝑑𝐴 + { ∫ 𝜎𝑛 𝑑𝐴} {∫ ∆𝑠 𝑑𝐴} {∫ 𝛿𝑠 𝑑𝐴}
𝐴 𝜕𝑥 𝐷 𝐴 𝐴 𝐴 𝐴 )
and the stiffness matrix can be identified as
𝜕𝑑 𝜕𝑑 𝜕𝑔 1 1 𝜕𝑠 𝜕𝑠 (29.134)
𝑲𝐼𝐽 = 2𝐾 ∫ 𝑑 𝑑𝐴 + { ∫ 𝜎𝑛 𝑑𝐴} {∫ 𝑑𝐴} {∫ 𝑑𝐴}.
𝐴 𝜕𝒙 𝐼 𝜕𝒙 𝐽 𝜕𝑥 𝐷 𝐴 𝐴 𝐴 𝜕𝒙 𝐼 𝐴 𝜕𝒙 𝐽 )
A characteristic feature of the Mortar contact, which can be deduced from this expression,
is that not only the nodal forces are continuous but also the stiffness matrix. This follows
from the 𝐶1 continuity of 𝑔, meaning that all involved functions above are continuous, and
that 𝑲𝐼𝐽 tends to zero as either 𝑑 or 𝐴 tends to zero. A mathematical treatment yields
𝜕𝑑(𝑡) 𝜕𝑑(𝑡) 𝜕𝑠(𝑡) 𝜕𝑠(𝑡) (29.135)
∥𝑲𝐼𝐽 ∥ ≤ 𝐾𝐴 max 𝑑(𝑡) {2 max ∥ ∥∥ ∥ + max ∥ ∥∥ ∥}
𝑡 𝑡 𝜕𝒙𝐼 𝜕𝒙𝐽 𝑡 𝜕𝒙𝐼 𝜕𝒙𝐽 )
which tends to zero as 𝑑 or 𝐴 tends to zero as the terms inside the right bracket are
bounded.
LS-DYNA DEV 10/27/16 (r:8004) 27-65 (Contact-Impact Algorithm)
Contact-Impact Algorithm LS-DYNA Theory Manual

29.21.9 Tied and tiebreak option


A tied and tiebreak option with the Mortar contact is available by appending MOR-
TAR_TIED or TIEBREAK_MORTAR to the automatic surface to surface contact keyword.
In principle, the tiebreak allows for specifying the contact stress 𝜎 as a function of the
separation 𝑑 to accomodate a given interface law that includes softening and failure (loss of
interface stiffness). The law incorporates both normal and tangential directions, usually
denoted mode I and mode II in the delamination community, and the typical appearance of
such a law is shown in Figure bm_figcohlaw. The law is in general characterized by the
energy release rate 𝐸 which is the energy per unit area required to release the contact, and the
maximum interface stress 𝜎𝑒 which is the peak contact stress before softening begins.
Referring to the above mentioned figure, the energy release rate is the integral of the curve
describing the stress vs displacement relation. The law can indeed be quite complicated,
and we refer to the keyword manual for more information regarding details of the cohesive
models available. For the Mortar tiebreak option only OPTION = 7 and OPTION = 9 are
supported. The internal treatment of the Mortar tied and tiebreak options are very similar to
that of the one-sided contacts, but the following remarks are in place.
Tied Mortar contact is penalty based and exhibits a linear relationship between
normal/tangential separation and resulting contact stress. In referring to equation
(29.21.111), 𝑓 (𝑥) = 𝑥 for all 𝑥 in all separation directions. Since the nonlinearity is
significantly reduced compared to one-sided contacts, the implicit convergence
characteristics are insensitive to scaling of the contact stiffness.
Tiebreak Mortar contact is treated similarly but the normal and tangential contact
stress is given by the constitutive model. It is strongly recommended to set CN
(OPTION = 7 and OPTION = 9) and CT2CN (only OPTION = 9) on the additional
card associated with the tiebreak option, otherwise the interface stiffness is deter-
mined internally and may be inconveniently large.
Tiebreak Mortar contact is superposed by a one-sided contact to take compressive
contact stress, mainly to prevent penetrations in the post-failure regime but also to
prevent cohesive failure due to normal compression. This contact follows the theory
described above and automatically applies the IGNORE = 2 option to start with an
initial zero contact stress.
The one-sided contact associated with the tiebreak option is frictionless as long as the
tied contact exists, this to avoid spurious interactions between the laws (mode II and
friction) in the tangential directions. The friction is activated as soon as the tied
contact is released. Furthermore, the tied interface will not take compressive stress
as this is lent to the one-sided contact.
Tied and tiebreak Mortar contacts are applied if the initial normal distance between
the slave and master segment (with respect to their outer geometries) is less than a
critical distance, 𝑑𝑡 = 0.05𝛽𝑑𝑐 . In this formula, 𝛽 is synonym for SFST on the contact
card and 𝑑𝑐 is the characteristic length as described above. Note that for solid
elements the characteristic length can be specified with SST on the contact card and
SFST is supposed to be used to adjust the tolerance for when to tie two segments, by
default tying occurs when the distance is less than 5% of the characteristic length of

27-66 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Contact-Impact Algorithm

the slave segment. Whence, for tied and tiebreak contacts SFST is not used to trans-
late the contact surface as it is for the one-sided Mortar contacts.
.

Shear Contact Forces


150000
Workpiece
100000 Support
Contact Force

50000

-50000

-100000

-150000
0 10 20 30 40 50 60
Workpiece Displacement

LS-DYNA DEV 10/27/16 (r:8004) 27-67 (Contact-Impact Algorithm)


Contact-Impact Algorithm LS-DYNA Theory Manual

Figure 27-47 A rubber compression example solved in implicit with Mortar contact
(Courtesy of Dellner Couplers AB). The graph shows the contact force between the rubber
parts and the moving workpiece and between the rubber parts and the two supports,
respectively.

27-68 (Contact-Impact Algorithm) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Geometric Contact Entities

30
Geometric Contact Entities

Contact algorithms in LS-DYNA currently can treat any arbitrarily shaped surface
by representing the surface with a faceted mesh. Occupant modeling can be treated this
way by using fine meshes to represent the head or knees. The generality of the faceted
mesh contact suffers drawbacks when modeling occupants, however, due to storage
requirements, computing costs, and mesh generation times. The geometric contact entities
were added as an alternate method to model cases of curved rigid bodies impacting
deformable surfaces. Much less storage is required and the computational cost decreases
dramatically when compared to the more general contact.

Geometric contact entities are developed using a standard solids modeling


approach. The geometric entity is defined by a scalar function 𝐺(𝑥, 𝑦, 𝑧). The solid is
determined from the scalar function as follows:
𝐺(𝑥, 𝑦, 𝑧) > 0 The point (𝑥, 𝑦, 𝑧) is outside the solid (30.1)

𝐺(𝑥, 𝑦, 𝑧) = 0 The point (𝑥, 𝑦, 𝑧) is on the surface of the solid (30.2)

𝐺(𝑥, 𝑦, 𝑧) < 0 The point (𝑥, 𝑦, 𝑧) is inside the solid (30.3)


Thus, by a simple function evaluation, a node can be immediately determined to be outside
the solid or in contact. Figure 30.1 illustrates this for a cylinder.

If the node is in contact with the solid, a restoring force must be applied to eliminate
further penetration. A number of methods are available to do this such as Lagrange
multipliers or momentum based methods. The penalty method was selected because it is
the simplest and most efficient method. Also, in our applications the impact velocities are
at a level where the penalty methods provide almost the identical answer as the exact
solution.

Using the penalty method, the restoring force is proportional to the penetration
distance into the solid and acts in the direction normal to the surface of the solid. Thus, the

LS-DYNA DEV 10/27/16 (r:8004) 28-1 (Geometric Contact Entities)


Geometric Contact Entities LS-DYNA Theory Manual

G(x, y) < 0
R
G(x, y) > 0

G(x, y) = 0

Figure 30.1. Determination of whether a node is interior or exterior to the


cylindrical surface

penetration distance and the normal vector must be determined. The surface normal vector
is conveniently determined from the gradient of the scalar function.

∂𝐺 𝐢 + ∂𝐺 𝐣 + 𝜕𝐺 𝐤
∂𝑥 ∂𝑦 𝜕𝑧
𝐍(𝑥, 𝑦, 𝑧) = , (30.4)
2 2 2
√(∂𝐺) + (∂𝐺) + (∂𝐺)
∂𝑥 ∂𝑦 ∂𝑧
for all (𝑥, 𝑦, 𝑧) such that 𝐺(𝑥, 𝑦, 𝑧) = 0. The definition of 𝐺(𝑥, 𝑦, 𝑧) guarantees that this vector
faces in the outward direction. When penetration does occur, the function 𝐺(𝑥, 𝑦, 𝑧) will be
slightly less than zero. For curved surfaces this will result in some errors in calculating the
normal vector, because it is not evaluated exactly at the surface. In an implicit code, this
would be important, however, the explicit time integration scheme in DYNA3D uses such a
small time step that penetrations are negligible and the normal function can be evaluated
directly at the slave node ignoring any penetration.
𝐺(𝑥, 𝑦) = 𝑥2 + 𝑦2 − 𝑅2 , (30.5)
The penetrations distance is the last item to be calculated. In general, the penetration
distance, 𝑑, is determined by.
𝑑 = ∣𝐗𝑛 − 𝐗′𝑛 ∣, (30.6)
where 𝐗𝑛 is the location of node 𝑛 and 𝐗′𝑛 is the nearest point on the surface of the solid.

To determine 𝐗′𝑛 , a line function is defined which passes through 𝐗𝑛 and is normal
to the surface of the solid:
𝐋(𝑠) = 𝐗𝑛 + 𝑠𝐍 (𝐗𝑛 ). (30.7)
Substituting the line function into the definition of the Equation (30.2) surface of a solid
body gives:
𝐺(𝐗𝑛 + 𝑠𝐍(𝐗)) = 0. (30.8)

28-2 (Geometric Contact Entities) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Geometric Contact Entities

If Equation (30.8) has only one solution, this provides the parametric coordinates s which
locates 𝐗′𝑛 . If Equation (30.8) has more than one root, then the root which minimizes
Equation (30.6) locates the point 𝐗′𝑛 .

The penalty method defines the restoring forces as:


̅̅̅̅̅̅̅ ),
𝐟 = 𝑝𝑑𝐍(𝐗′ (30.9)
𝑛

where 𝑝 is a penalty factor and is effectively a spring constant. To minimize the


penetration of the slave node into the solid, the constant 𝑝 is set large, however, it should
not be set so large that the Courant stability criteria is violated. This criteria for the slave
node tells us that:
2 2 𝑚
Δ𝑡 ≤ = = 2√ 𝑛 ,
𝜔max 𝐾𝑛 (30.10)
√ 𝐾𝑛
𝑚𝑛
where 𝐾𝑛 is the stiffness of node 𝑛 and 𝑚𝑛 is the mass of node 𝑛.

The penalty factor, 𝑝, is determined by choosing a value which results in a


penalty/slave mass oscillator which has a characteristic time step that is ten times larger
than the Courant time step:
𝑚𝑛
10Δ𝑡 = 2√ . (30.11)
𝑝𝑛
Solving for 𝑝𝑛 gives:
4𝑚𝑛
𝑝𝑛 = . (30.12)
(100Δ𝑡2 )

Inclusion of any structural elements into the occupant model will typically result in
very large stiffnesses due to the small time step and the (1/Δ𝑡)2 term. Thus the method is
highly effective even with impact velocities on the order of 1km/sec.

The scalar function 𝐺(𝐗) is frequently more conveniently expressed as 𝑔(𝐱)where,𝑔


is the function defined in local coordinates and 𝐱 is the position in local coordinates. The
local entity is related to the global coordinates by:
𝐱 = [T](𝐗𝑗 − 𝐐𝑗 ), (30.13)
where 𝐐𝑗 is the offset and [T] is a rotation matrix. The solid scalar function and the
penetration distance can be evaluated in either local or global coordinates with no
difference to the results. When working in local coordinates, the gradient of the local scalar
function provides a normal vector which is in the local system and must be transformed
into the global by:
̅̅̅̅̅) = [T]T 𝐧(𝐱).
𝐍(𝐗 (30.14)

LS-DYNA DEV 10/27/16 (r:8004) 28-3 (Geometric Contact Entities)


Geometric Contact Entities LS-DYNA Theory Manual

An ellipsoid is defined by the function:


𝑥 2 𝑦 2 𝑧 2
𝐺(𝑥, 𝑦, 𝑧) = ( ) + ( ) + ( ) − 1. (30.15)
𝑎 𝑏 𝑐
2𝑦
The gradient of 𝐺 is 2𝑥
𝑎2
𝐢 + 𝑎2 𝐣 + 2𝑧
𝑎2
𝐤 and the normal vector is:
𝑦
( 𝑥2 𝐢 + 2 𝐣 + 𝑧2 𝐤)
𝐧(𝑥, 𝑦, 𝑧) = 𝑎 𝑏 𝑐 , (30.16)
2
√𝑥 + 𝑦 + 𝑧
2 2

𝑎4 𝑏4 𝑐4
Substituting Equations (30.7) and (30.15) into Equation (27.2) gives:
𝑛 2 𝑛𝑦 2 𝑛 2 𝑛 𝑥 𝑛𝑦 𝑦𝑛 𝑛 𝑧
[( 𝑥 ) + ( ) + ( 𝑧 ) ] 𝑠 2 + 2 [ 𝑥2 𝑛 + 2 + 𝑧2 𝑛 ] 𝑠
𝑎 𝑏 𝑐 𝑎 𝑏 𝑐 (30.17)
𝑥𝑛 2 𝑦𝑛 2 𝑧𝑛 2
+ [( ) + ( ) + ( ) − 1] = 0.
𝑎 𝑏 𝑐

Solving this quadratic equation for 𝑠 provides the intercepts for the nearest point on
the ellipsoid and the opposite point of the ellipsoid where the normal vector, 𝐗𝑛 , also
points toward.

Currently, this method has been implemented for the case of an infinite plane, a
cylinder, a sphere, and an ellipsoid with appropriate simplifications. The ellipsoid is
intended to be used with rigid body dummy models. The methods are, however, quite
general so that many more shapes could be implemented. A direct coupling to solids
modeling packages should also be possible in the future.

28-4 (Geometric Contact Entities) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Nodal Constraints

31
Nodal Constraints

In this section nodal constraints and linear constraint equations are described.

31.1 Nodal Constraint Sets


This option forces groups of nodes to move together with a common translational
acceleration in either one or more degrees of freedom. The implementation is
straightforward with the common acceleration defined by
𝑛 𝑗
∑ 𝑗 𝑀𝑗 𝑎𝑖
𝑎𝑖common = 𝑛 , (31.1)
∑ 𝑗 𝑀𝑗
𝑗
where 𝑛 is the number of nodes, 𝑎𝑖 is the acceleration of the jth constrained node in the ith
direction, and 𝑎𝑖common is the common acceleration.

Nodal constraint sets eliminate rigid body rotations in the body that contains the
node set and, therefore, must be applied very cautiously.

31.2 Linear Constraint Equations


Linear constraint equations of the form:
𝑛
∑ 𝐶𝑘 𝑢𝑘 = 𝐶0 , (31.2)
𝑘=1

can be defined where 𝑛 is the number of constrained degrees of freedom, 𝑢𝑘 is a


constrained nodal displacement, and the 𝐶𝑘 are user-defined coefficients. Unless LS-DYNA

LS-DYNA DEV 10/27/16 (r:8004) 29-1 (Nodal Constraints)


Nodal Constraints LS-DYNA Theory Manual

is initialized by linking to an implicit code to satisfy this equation at the beginning of the
calculation, the constant 𝐶0 is assumed to be zero. The first constrained degree of freedom
is eliminated from the equations of motion:
𝑛
𝐶𝑘
𝑢1 = 𝐶0 − ∑ 𝑢 . (31.3)
𝐶 𝑘
𝑘=2 1

Its velocities and accelerations are given by


𝑛
𝐶𝑘
𝑢̇1 = − ∑ 𝑢̇ ,
𝐶 𝑘
𝑘=2 1
𝑛
(31.4)
𝐶
𝑢̈1 = − ∑ 𝑘 𝑢̈𝑘
𝐶
𝑘=2 1

respectively. In the implementation a transformation matrix 𝐋 is constructed relating the


unconstrained 𝑢 and constrained 𝑢constrained degrees of freedom. The constrained
accelerations used in the above equation are given by:
−1
𝑢constrained = [𝐋T 𝐌𝐋] 𝐋T 𝐅, (31.5)
where 𝐌 is the diagonal lumped mass matrix and 𝐅 is the righthand side force vector. This
requires the inversion of the condensed mass matrix which is equal in size to the number of
constrained degrees of freedom minus one. The inverse of the condensed mass matrix is
computed in the initialization phase and stored in core.

29-2 (Nodal Constraints) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Vectorization and Parallelization

32
Vectorization and Parallelization

32.1 Vectorization
In 1978, when the author first vectorized DYNA3D on the CRAY-1, a four-fold
increase in speed was attained. This increase was realized by recoding the solution phase
to process vectors in place of scalars. It was necessary to process elements in groups rather
than individually as had been done earlier on the CDC-7600 supercomputers.

Since vector registers are generally some multiple of 64 words, vector lengths of 64
or some multiple are appropriate. In LS-DYNA, groups of 128 elements or possibly some
larger integer multiple of 64 are utilized. Larger groups give a marginally faster code, but
can reduce computer time sharing efficiency because of increased core requirements. If
elements within the group reference more than one material model, subgroups are formed
for consecutive elements that reference the same model. LS-DYNA internally sorts
elements by material to maximize vector lengths.

Conceptually, vectorization is straightforward. Each scalar operation that is


normally executed once for one element, is repeated for each element in the group. This
means that each scalar is replaced by an array, and the operation is put into a DO-loop. For
example, the nodal force calculation for the hexahedron element appeared in a scalar
version of DYNA3D as:

E11=SGV1*PX1+SGV4*PY1+SGV6*PZ1
E21=SGV2*PY1+SGV4*PX1+SGV5*PZ1
E31=SGV3*PZ1+SGV6*PX1+SGV5*PY1
E12=SGV1*PX2+SGV4*PY2+SGV6*PZ2
E22=SGV2*PY2+SGV4*PX2+SGV5*PZ2
E32=SGV3*PZ2+SGV6*PX2+SGV5*PY2
E13=SGV1*PX3+SGV4*PY3+SGV6*PZ3
E23=SGV2*PY3+SGV4*PX3+SGV5*PZ3
E33=SGV3*PZ3+SGV6*PX3+SGV5*PY3
E14=SGV1*PX4+SGV4*PY4+SGV6*PZ4

LS-DYNA DEV 10/27/16 (r:8004) 30-1 (Vectorization and Parallelization)


Vectorization and Parallelization LS-DYNA Theory Manual

E24=SGV2*PY4+SGV4*PX4+SGV5*PZ4
E34=SGV3*PZ4+SGV6*PX4+SGV5*PY4

and in the vectorized version as:

DO 110 I = LFT, LLT


E11(I)=SGV1(I)*PX1(I)+SGV4(I)*PY1(I)+SGV6(I)*PZ1(I)
E21(I)=SGV2(I)*PY1(I)+SGV4(I)*PX1(I)+SGV5(I)*PZ1(I)
E31(I)=SGV3(I)*PZ1(I)+SGV6(I)*PX1(I)+SGV5(I)*PY1(I)
E12(I)=SGV1(I)*PX2(I)+SGV4(I)*PY2(I)+SGV6(I)*PZ2(I)
E22(I)=SGV2(I)*PY2(I)+SGV4(I)*PX2(I)+SGV5(I)*PZ2(I)
E32(I)=SGV3(I)*PZ2(I)+SGV6(I)*PX2(I)+SGV5(I)*PY2(I)
E13(I)=SGV1(I)*PX3(I)+SGV4(I)*PY3(I)+SGV6(I)*PZ3(I)
E23(I)=SGV2(I)*PY3(I)+SGV4(I)*PX3(I)+SGV5(I)*PZ3(I)
E33(I)=SGV3(I)*PZ3(I)+SGV6(I)*PX3(I)+SGV5(I)*PY3(I)
E14(I)=SGV1(I)*PX4(I)+SGV4(I)*PY4(I)+SGV6(I)*PZ4(I)
E24(I)=SGV2(I)*PY4(I)+SGV4(I)*PX4(I)+SGV5(I)*PZ4(I)
110 E34(I)=SGV3(I)*PZ4(I)+SGV6(I)*PX4(I)+SGV5(I)*PY4(I)
where 1 ≤ LFT ≤ LLT ≤ n. Elements LFT to LLT inclusive use the same material model and
n is an integer multiple of 64.

Gather operations are vectorized on most supercomputers. In the gather operation,


variables needed for processing the element group are pulled from global arrays into local
vectors. For example, the gather operation:

DO 10 I = LFT, LLT
X1(I) = X(1,IX1(I))
Y1(I) = X(2,IX1(I))
Z1(I) = X(3,IX1(I))
VX1(I) = V(1,IX1(I))
VY1(I) = V(2,IX1(I))
VZ1(I) = V(3,IX1(I))
X2(I) = X(1,IX2(I))
Y2(I) = X(2,IX2(I))
Z2(I) = X(3,IX2(I))
VX2(I) = V(1,IX2(I))
VY2(I) = V(2,IX2(I))
VZ2(I) = V(3,IX2(I))
X3(I) = X(1,IX3(I))
X3(I) = X(2,IX3(I))
X3(I) = X(3,IX3(I))
X8(I) = X(1,IX8(I))
Y8(I) = X(2,IX8(I))
Z8(I) = X(3,IX8(I))
VX8(I) = V(1,IX8(I))
VY8(I) = V(2,IX8(I))

30-2 (Vectorization and Parallelization) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Vectorization and Parallelization

10 VZ8(I) = V(3,IX8(I))

initializes the nodal velocity and coordinate vector for each element in the subgroup LFT to
LLT. In the scatter operation, element nodal forces are added to the global force vector.
The force assembly does not vectorize unless special care is taken as described below.

In general, the element force assembly is given in FORTRAN by:

DO 30 I = 1,NODFRC
DO 20 N = 1,NUMNOD
DO 10 L = LFT,LLT
RHS(I,IX(N,L))=RHS(I,IX(N,L))+FORCE(I,N,L)
10 CONTINUE
20 CONTINUE
30 CONTINUE

where NODFRC is the number of force components per node (3 for solid elements, 6 for
shells), LFT and LLT span the number of elements in the vector block, NUMNOD is the
number of nodes defining the element, FORCE contains the force components of the
individual elements, and RHS is the global force vector. This loop does not vectorize since
the possibility exists that more that one element may contribute force to the same node.
FORTRAN vector compilers recognize this and will vectorize only if directives are added
to the source code. If all elements in the loop bounded by the limits LFT and LLT are
disjoint, the compiler directives can be safely added. We therefore attempt to sort the
elements as shown in Figure 32.1 to guarantee disjointness.

ELEMENT BLOCKING FOR VECTORIZATION

LS-DYNA DEV 10/27/16 (r:8004) 30-3 (Vectorization and Parallelization)


Vectorization and Parallelization LS-DYNA Theory Manual

Block 1 Block 2

Block 3 Block 3

Figure 32.1. Group of 48 elements broken into 4 disjoint blocks.

The current implementation was strongly motivated by Benson [1989] and by work
performed at General Motors [Ginsberg and Johnson 1988, Ginsberg and Katnik 1989],
where it was shown that substantial improvements in execution speed could be realized by
blocking the elements in the force assembly. Katnik implemented element sorting in a
public domain version of DYNA3D for the Belytschko-Tsay shell element and added
compiler directives to force vectorization of the scatter operations associated with the
addition of element forces into the global force vector. The sorting was performed
immediately after the elements were read in so that subsequent references to the stored
element data were sequential. Benson performed the sorting in the element loops via
indirect addressing. In LS-DYNA the published GM approach is taken.

Implementation of the vectorization of the scatter operations is implemented in for


all elements including the solid, shell, membrane, beam, and truss elements. The sorting is
completely transparent to the user.

32.2 Parallelization
In parallelization, the biggest hurdle is overcoming Amdahl’s law for multitasking
[Cray Research Inc. 1990]
1
𝑆𝑚 = ,
𝑓𝑝 (32.1)
𝑓𝑠 +
𝑁
where

30-4 (Vectorization and Parallelization) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Vectorization and Parallelization

𝑆𝑚 = maximum expected speedup from multitasking


𝑁 = number of processors available for parallel execution
𝑓𝑝 = fraction of a program that can execute in parallel
𝑓𝑠 = fraction of a program that is serial

Table 29.1 shows that to obtain a speed factor of four on eight processors it is
necessary to have eighty-six percent of the job running in parallel. Obviously, to gain the
highest speed factors the entire code must run in parallel.

LS-DYNA has been substantially written to function on all shared memory parallel
machine architectures. Generally, shared memory parallel speed-ups of 5 on 8 processors
are possible but this is affected by the machine characteristics. We have observed speeds of
5.6 on full car crash models on a machine of one manufacturer only to see a speed-up of 3.5
on a different machine of another manufacturer.

% N=2 N=4 N=8 N = 16 N = 32 N = 64 N = 128 N = 256


86.0% 1.75 2.82 4.04 5.16 5.99 6.52 6.82 6.98
90.0% 1.82 3.08 4.71 6.40 7.80 8.77 9.34 9.66
92.0% 1.85 3.23 5.13 7.27 9.20 10.60 11.47 11.96
94.0% 1.89 3.39 5.63 8.42 11.19 13.39 14.85 15.71
96.0% 1.92 3.57 6.25 10.00 14.29 18.18 21.05 22.86
98.0% 1.96 3.77 7.02 12.31 19.75 28.32 36.16 41.97
99.0% 1.98 3.88 7.48 13.91 24.43 39.26 56.39 72.11
99.2% 1.98 3.91 7.58 14.29 25.64 42.55 63.49 84.21
99.4% 1.99 39.3 7.68 14.68 26.98 46.44 72.64 101.19
99.6% 1.99 3.95 7.78 15.09 28.47 51.12 84.88 126.73
99.7% 1.99 3.96 7.84 15.31 29.28 53.83 92.69 145.04
99.8% 2.00 3.98 7.89 15.53 30.13 56.84 102.07 169.54
99.9% 2.00 3.99 7.94 15.76 31.04 60.21 113.58 203.98
100.0% 2.00 4.00 8.00 16.00 32.00 64.00 128.00 256.00

Table 29.1.Maximum theoretical speedup Sm, on N CPUs with parallelism [Cray Research
Inc. 1990].

In the element loops element blocks with vector lengths of 64 or some multiple are
assembled and sent to separate processors. All elements are processed in parallel. On the
average a speed factor of 7.8 has been attained in each element class corresponding to
99.7% parallelization.

A significant complication in parallelizing code is that a variable can sometimes be


updated simultaneously by different processors with incorrect results. To force the
processors to access and update the variable in a sequential manner, a GUARD compiler
directive must be introduced. This results in an interruption to the parallel execution and
can create a bottleneck. By sorting the data in the parallel groups or by allocating

LS-DYNA DEV 10/27/16 (r:8004) 30-5 (Vectorization and Parallelization)


Vectorization and Parallelization LS-DYNA Theory Manual

additional storage it is usually possible to eradicate the GUARDS from the coding. The
effort may not be worth the gains in execution speed.

The element blocks are defined at the highest level and each processor updates the
entire block including the right hand side force assembly. The user currently has two
options: GUARD compiler directives prevent simultaneous updates of the RHS vector
(recommended for single CPU processors or when running in a single CPU mode on a
multi-processor), or assemble the right hand side in parallel and let LS-DYNA prevent
conflicts between CPU’s. This usually provides the highest speed and is recommended,
i.e., no GUARDS.

When executing LS-DYNA in parallel, the order of operations will vary from run to
run. This variation will lead to slightly different numerical results due to round-off errors.
By the time the calculation reaches completion variations in nodal accelerations and
sometimes even velocities are observable. These variations are independent of the
precision and show up on both 32 and 64 bit machines. There is an option in LS-DYNA to
use an ordered summation of the global right hand side force vector to eliminate numerical
differences. To achieve this the element force vectors are stored. After leaving the element
loop, the global force vector is assembled in the same order that occurs on one processor.
The ordered summation option is slower and uses more memory than the default, but it
leads to nearly identical, if not identical, results run to run.

Parallelization in LS-DYNA was initially done with vector machines as the target
where the vector speed up is typically 10 times faster than scalar. On vector machines,
therefore, vectorization comes first. If the problem is large enough then parallelization is
automatic. If vector lengths are 128, for example, and if 256 beam elements are used only a
factor of 2 in speed can be anticipated while processing beam elements. Large contact
surfaces will effectively run in parallel, small surfaces having under 100 segments will not.
The speed up in the contact subroutines has only registered 7 on 8 processors due to the
presence of GUARD statements around the force assembly. Because real models often use
many special options that will not even vectorize efficiently it is unlikely that more than
95% of a given problem will run in parallel on a shared memory parallel machine.

30-6 (Vectorization and Parallelization) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Airbags

33
Airbags

Additional information on the airbag modeling and comparisons with experimental


data can be found in a report [Hallquist, Stillman, Hughes, and Tarver 1990] based on
research sponsored by the Motor Vehicles Manufacturers Association (MVMA).

33.1 Control Volume Modeling


A direct approach for modeling the contents of the airbag would be to discretize the
interior of the airbag using solid elements. The total volume and pressure-volume
relationship of the airbag would then be the sum of all the elemental contributions.
Although this direct approach could be applied in a straight forward manner to an inflated
airbag, it would become very difficult to implement during the inflation phase of the airbag
deployment. Additionally, as the model is refined, the solid elements would quickly
overwhelm all other computational costs and make the numerical simulations prohibitively
expensive.

An alternative approach for calculating the airbag volume, that is both applicable
during the inflation phase and less computationally demanding, treats the airbag as a
control volume. The control volume is defined as the volume enclosed by a surface. In the
present case, the ‘control surface’ that defines the control volume is the surface modeled by
shell or membrane elements comprising the airbag fabric material.

Because the evolution of the control surface is known, i.e., the position, orientation,
and current surface area of the airbag fabric elements are computed and stored at each time
step, we can take advantage of these properties of the control surface elements to calculate
the control volume, i.e., the airbag volume. The area of the control surface can be related to
the control volume through Green’s Theorem
∂𝜓 ∂𝜙
∭𝜙 𝑑𝑥𝑑𝑦𝑑𝑧 = − ∭ 𝜓 𝑑𝑥𝑑𝑦𝑑𝑧 + ∮ 𝜙𝜓𝑛𝑥 𝑑𝛤 , (33.1)
∂𝑥 ∂𝑥

LS-DYNA DEV 10/27/16 (r:8004) 31-1 (Airbags)


Airbags LS-DYNA Theory Manual

where the first two integrals are integrals over a closed volume, i.e., 𝑑𝑣 = 𝑑𝑥𝑑𝑦𝑑𝑧, the last
integral is an integral over the surface enclosing the volume, and 𝑛𝑥 is the direction cosine
between the surface normal and the 𝑥 direction (corresponding to the x-partial derivative);
similar forms can be written for the other two directions. The two arbitrary functions 𝜙
and 𝜓 need only be integrated over the volume and surface.

The integral form of the volume can be written as

𝑉 = ∭ 𝑑𝑥𝑑𝑦𝑑𝑧. (33.2)

Comparing the first of the volume integrals in Equation (33.1) to Equation (33.2), we can
easily obtain the volume integral from Equation (33.1) by choosing for the two arbitrary
functions
𝜙 = 1, (33.3)

𝜓 = 𝑥𝑥 , (33.4)
leading to

𝑉 = ∫ ∫ ∫ 𝑑𝑥𝑑𝑦𝑑𝑧 = ∮ 𝑥𝑛𝑥 𝑑𝛤 . (33.5)

The surface integral in Equation (33.5) can be approximated by a summation over all the
elements comprising the airbag, i.e.,
𝑁
∮ 𝑥𝑛𝑥 𝑑𝛤 ≈ ∑ 𝑥̅𝑖 𝑛𝑖𝑥 𝐴𝑖 , (33.6)
𝑖=1

where for each element i: 𝑥̅𝑖 is the average x coordinate, 𝑛𝑖𝑥 is the direction cosine between
the elements normal and the 𝑥 direction, and 𝐴𝑖 is the surface area of the element.

Although Equation (33.5) will provide the exact analytical volume for an arbitrary
direction, i.e., any 𝑛, the numerical implementation of Equation (33.5), and its approxima-
tion Equation (33.6), has been found to produce slightly different volumes, differing by a
few percent, depending on the choice of directions: if the integration direction is nearly
parallel to a surface element, i.e., the direction cosine is nearly zero, numerical precision
errors affect the volume calculation. The implementation uses as an integration direction, a
direction that is parallel to the maximum principle moment of inertia of the surface.
Numerical experiments have shown this choice of integration direction produces more
accurate volumes than the coordinate or other principle inertia directions.

Because airbag models may contain holes, e.g., holes for inflation and deflation, and
Green’s Theorem only applies to closed surfaces, a special treatment is needed for
calculating the volume of airbags with holes. This special treatment consists of the
following:

• The n-sized polygon defining the hole is identified automatically, using edge
locating algorithms in LS-DYNA.

31-2 (Airbags) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Airbags

• The n-sized polygon is projected onto a plane, i.e., it is assumed to be flat; this
is a good approximation for typical airbag hole geometries. Planar symmetry should work
with the control volume capability for one symmetry plane.
• The area of the flat n-sided polygon is calculated using Green’s Theorem in
two dimensions.
• The resulting holes are processed as another surface element in the airbag
control volume calculation.

33.2 Equation of State Model


As explained above, at each time step in the calculation the current volume of the
airbag is determined from the control volume calculation. The pressure in the airbag
corresponding to the control volume is determined from an equation of state (EOS) that
relates the pressure to the current gas density (volume) and the specific internal energy of
the gas.

The equation of state used for the airbag simulations is the usual ‘Gamma Law Gas
Equation of State’,
𝑝 = (𝑘 − 1)𝜌𝑒, (33.7)
where 𝑝 is the pressure, 𝑘 is a constant defined below, 𝜌 is the density, and 𝑒 is the specific
internal energy of the gas. The derivation of this equation of state is obtained from
thermodynamic considerations of the adiabatic expansion of an ideal gas. The incremental
change in internal energy, 𝑑𝑈, in 𝑛 moles of an ideal gas due to an incremental increase in
temperature, 𝑑𝑇, at constant volume is given by
𝑑𝑈 = 𝑛𝑐v 𝑑𝑇, (33.8)
where 𝑐v is the specific heat at constant volume. Using the ideal gas law we can relate the
change in temperature to a change in the pressure and total volume, 𝑣, as
𝑑(𝑝𝑣) = 𝑛𝑅𝑑𝑇, (33.9)
where 𝑅 is the universal gas constant. Solving the above for 𝑑𝑇 and substituting the result
into Equation (33.8) gives
𝑐v 𝑑(𝑝𝑣) 𝑑(𝑝𝑣)
𝑑𝑈 = = , (33.10)
𝑅 (𝑘 − 1)
where we have used the relationship
𝑅 = 𝑐p − 𝑐v , (33.11)
and the notation
𝑐p
𝑘= . (33.12)
𝑐v

LS-DYNA DEV 10/27/16 (r:8004) 31-3 (Airbags)


Airbags LS-DYNA Theory Manual

Equation (33.10) may be rewritten as


𝜌 0 𝑣0 𝑝
𝑑𝑈 = 𝑑 ( ), (33.13)
𝑘−1 𝜌
and integrated to yield
𝑈 𝑝
𝑒= = . (33.14)
𝜌0 𝑣0 𝜌(𝑘 − 1)
Solving for the pressure
𝑝 = (𝑘 − 1)𝜌𝑒. (33.15)

The equation of state and the control volume calculation can only be used to
determine the pressure when the specific internal energy is also known. The evolution
equation for the internal energy is obtained by assuming the change in internal energy is
given by
𝑑𝑈 = −𝑝𝑑𝑣, (33.16)
where the minus sign is introduced to emphasize that the volume increment is negative
when the gas is being compressed. This expression can be written in terms of the specific
internal energy as
𝑑𝑈 𝑝𝑑𝑣
𝑑𝑒 = =− . (33.17)
𝜌 0 𝑣0 𝜌0 𝑣
Next, we divide the above by the equation of state, Equation (4.11.144), to obtain
𝑑𝑒 𝜌(𝑘 − 1)𝑑𝑣 (𝑘 − 1)𝑑𝑣
=− =− , (33.18)
𝑒 𝜌 0 𝑣0 𝑣
which may be integrated to yield
ln𝑒 = (1 − 𝑘)ln𝑉, (33.19)
or evaluating at two states and exponentiating both sides yields
𝑣 (1−𝑘)
𝑒2 = 𝑒1 ( 2 ) . (33.20)
𝑣1

The specific internal energy evolution equation, Equation (33.20), the equation of
state, Equation (4.11.144), and the control volume calculation completely define the
pressure-volume relation for an inflated airbag.

31-4 (Airbags) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Airbags

33.3 Airbag Inflation Model


Airbag inflation models have been used for many years in occupant simulation
codes such as CAL3D [Fleck, 1981].

The inflation model we chose to implement in LS-DYNA is due to Wang and


Nefske[1988] and more recent improvements to the model in LS-DYNA were suggested by
Wang [1992]. In their development they consider the mass flow due to the vents and
leakage through the bag. We assume that the mass flow rate and the temperature of the
gas going into the bag from an inflator are provided as tabulated functions of time.

A pressure relation is defined:


𝑝e
𝑄= , (33.21)
𝑝2
where 𝑝e is the external pressure and 𝑝2 is the internal pressure in the bag. A critical
pressure relationship is defined as:
⁄ 𝑘
2 𝑘−1
(33.22)
𝑄crit =( ) ,
𝑘+1
where 𝑘 is the ratio of specific heats:
𝑐p
𝑘= . (33.23)
𝑐v
If 𝑄 ≤ 𝑄crit then 𝑄 = 𝑄crit .

Wang and Nefske define the mass flow through the vents and leakage by

𝑝2 1⁄ 𝑘𝑅 𝑘−1⁄
𝑚̇ 23 = 𝐶23 𝐴23 𝑄 𝑘 √2𝑔 (33.24)
𝑐( ) (1 − 𝑄 𝑘 ),
𝑅√𝑇2 𝑘−1

and

𝑝2 1⁄ 𝑘𝑅 𝑘−1⁄
𝑚̇ ′23 = 𝐶′23 𝐴′23 𝑄 𝑘 √2𝑔
𝑐( ) (1 − 𝑄 𝑘 ), (33.25)
𝑅√𝑇2 𝑘−1

where 𝐶23 , 𝐴23 , 𝐶′23 , 𝐴′23 , 𝑅 and 𝑔𝑐 are the vent orifice coefficient, vent orifice area, the
orifice coefficient for leakage, the area for leakage, the gas constant, and the gravitational
conversion constant, respectively. The internal temperature of the airbag gas is denoted by
𝑇2 . We note that both 𝐴23 and 𝐴′23 can be defined as a function of pressure [Wang, 1992]
or if they are input as zero they are computed within LS-DYNA. This latter option requires
detailed modeling of the airbag with all holes included.

A uniform temperature and pressure is assumed; therefore, in terms of the total


airbag volume 𝑉2 and air mass, 𝑚2 , the perfect gas law is applied:

LS-DYNA DEV 10/27/16 (r:8004) 31-5 (Airbags)


Airbags LS-DYNA Theory Manual

𝑝2 𝑉 = 𝑚2 𝑅𝑇2 . (33.26)
Solving for 𝑇2 :
𝑝2 𝑉
𝑇2 = , (33.27)
𝑚2 𝑅
and substituting Equation (33.27) into equations (33.25), we arrive at the mass transient
equation:
√ 2
√𝑘 (𝑄𝑘 − 𝑄𝑘+1⁄𝑘 )
√ (33.28)
𝑚̇ out = 𝑚̇ 23 + 𝑚̇ ′23 = 𝜇√2𝑝2 𝜌
⎷ 𝑘−1
where
𝜌 = density of airbag gas,
𝜇 = bag characterization parameter,
𝑚̇ out = total mass flow rate out of bag.

In terms of the constants used by Wang and Nefske:


𝜇 = √𝑔𝑐 (𝐶23 𝐴23 + 𝐶′23 𝐴′23 ). (33.29)
We solved these equations iteratively, via function evaluation. Convergence usually occurs
in 2 to 3 iterations.

The mass flow rate and gas temperature are defined in load curves as a function of
time. Using the mass flow rate we can easily compute the increase in internal energy:
𝐸̇in = 𝑐p 𝑚̇ in 𝑇in , (33.30)
where 𝑇in is the temperature of the gas flowing into the airbag. Initializing the variables
pressure, 𝑝, density, 𝜌, and energy, 𝐸, to their values at time 𝑛, we can begin the iterations
loop to compute the new pressure, 𝑝𝑛 + 1 , at time 𝑛 + 1.

𝑛+1⁄2 𝑝𝑛 + 𝑝𝑛+1
𝑝 =
2
𝑛+1⁄2 𝜌𝑛 + 𝜌𝑛+1
𝜌 =
2
𝑛+1⁄2 𝐸𝑛 + 𝐸𝑛+1 (33.31)
𝐸 =
2

𝑛+1⁄2

⎜ 𝑝𝑒 ⎞

𝑄 = max ⎜
⎜ , 𝑄 ⎟.
⎜ 1 crit ⎟

𝑛+
2
⎝ 𝑝2 ⎠
The mass flow rate out of the bag, 𝑚̇ out can now be computed:

31-6 (Airbags) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Airbags

√ 2 𝑘+1⁄
√ ⎛ 𝑛+1⁄2 ⁄𝑘 𝑛+1⁄2 𝑘⎞
√𝑘 ⎜ 𝑄 −𝑄 ⎟ (33.32)
𝑛+1⁄2 𝑛+1⁄2 𝑛+1⁄ √ ⎝ ⎠
𝑚̇ out = 𝜇√2𝑝2 𝜌 2 ,
⎷ 𝑘−1
where
𝑛+1⁄2 𝑛+1⁄2
𝑝2 =𝑝 + 𝑝e , (33.33)

and the total mass updated:


𝑛+1⁄2 𝑛+1⁄2
𝑚𝑛+1 = 𝑚𝑛 + Δ𝑡 (𝑚̇ in − 𝑚out )
(33.34)
𝑛+1⁄2 𝑚𝑛 + 𝑚𝑛+1
𝑚 = .
2

The energy exiting the airbag is given by:


𝑛+1⁄2
𝑛+ ⁄ 1 𝑛+ ⁄ 1 𝐸
𝐸̇out 2 = 𝑚̇ out 2 , (33.35)
𝑛+1⁄
𝑚 2

we can now compute our new energy at time 𝑛 + 1


n+ ⁄ 1
𝑛+ ⁄ 1
𝑛+ ⁄2 𝑛+ ⁄2 1 1
𝐸𝑛+1 = 𝐸𝑛 + Δ𝑡 (Ė in 2 − 𝐸̇out 2 ) − 𝑝 Δ𝑉 , (33.36)

𝑛+1⁄2
where Δ𝑉 is the change in volume from time 𝑛 to 𝑛 + 1. The new pressure can now be
computed:
𝐸𝑛+1
𝑝𝑛+1 = (𝑘 − 1)
, (33.37)
𝑉 𝑛+1
which is the gamma-law (where 𝑘 = 𝛾) gas equation. This ends the iteration loop.

33.4 Wang's Hybrid Inflation Model


Wang's proposed hybrid inflator model [1995a, 1995b] provides the basis for the
model in LS-DYNA. The first law of thermodynamics is used for an energy balance on the
airbag control volume.
𝑑
(𝑚𝑢)cv = ∑ 𝑚̇ 𝑖 ℎ𝑖 − ∑ 𝑚̇ 𝑜 ℎ𝑜 − 𝑊̇ cv − 𝑄̇cv , (33.38)
𝑑𝑡
where
𝑑
(𝑚𝑢)cv = rate of change of airbag internal energy
𝑑𝑡
∑ 𝑚𝑖 ℎ𝑖 = energy into airbag by mass flow (e. g. , inflator)

LS-DYNA DEV 10/27/16 (r:8004) 31-7 (Airbags)


Airbags LS-DYNA Theory Manual

∑ 𝑚𝑜 ℎ𝑜 = energy out of airbag by mass flow (e. g. , vents)


𝑊̇ cv = ∫ 𝑃𝑑𝑉̇ = work done by airbag expansion
𝑄̇cv = energy out by heat transfer through airbag surface.

The rate of change of internal energy, the left hand side of Equation (33.38), can be
differentiated:
𝑑 𝑑𝑚 𝑑𝑢 𝑑𝑚 𝑑 𝑑𝑇
(𝑚𝑢) = 𝑢+𝑚 = 𝑢 + 𝑚 (𝑐v 𝑇) = 𝑚̇ 𝑢 + 𝑚𝑐v̇ 𝑇 + 𝑚𝑐v , (33.39)
𝑑𝑡 𝑑𝑡 𝑑𝑡 𝑑𝑡 𝑑𝑡 𝑑𝑡
where we have used the definition
𝑢 = 𝑐v 𝑇. (33.40)

Then, the energy equation can be re-written for the rate of change in temperature for
the airbag
𝑑𝑇cv ∑ 𝑚̇ 𝑖 ℎ𝑖 − ∑ 𝑚̇ 𝑜 ℎ𝑜 − 𝑊̇ cv − 𝑄̇cv − (𝑚̇ 𝑢)cv − (𝑚𝑐v̇ 𝑇)cv
= (33.41)
𝑑𝑡 (𝑚𝑐v )cv
Temperature dependent heat capacities are used. The constant pressure molar heat
capacity is taken as:
𝑐p̅ = 𝑎 ̅ + 𝑏̅𝑇, (33.42)
and the constant volume molar heat capacity as:
𝑐v̅ = 𝑎 ̅ + 𝑏̅𝑇 − 𝑟 ̅, (33.43)
where
𝑟 ̅ = gas constant = 8.314 J/gm-mole K
𝑎 ̅ = constant [J/gm-mole K]
𝑏̅ = constant [J/gm-mole K2]

Mass based values are obtained by dividing the molar quantities by the molecular
weight, 𝑀, of the gas
𝑎̅ 𝑏̅ 𝑟̅
𝑎= , 𝑏= , 𝑟= . (33.44)
𝑀 𝑀 𝑀
The constant pressure and volume specific heats are then given by
𝑐p = 𝑎 + 𝑏𝑇 (33.45)

𝑐v = 𝑎 + 𝑏𝑇 − 𝑟. (33.46)
The specific enthalpy and internal energy becomes:
𝑇
𝑏𝑇 2
ℎ = ∫ 𝑐𝑝 𝑑𝑇 = 𝑎𝑇 + (33.47)
2
0

31-8 (Airbags) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Airbags

𝑇
𝑏𝑇 2
𝑢 = ∫ 𝑐𝑣 𝑑𝑇 = 𝑎𝑇 + − 𝑟𝑇. (33.48)
2
0

For ideal gas mixtures the molecular weight is given as:


1
𝑀= ,
𝑓 (33.49)
∑ 𝑖
𝑀𝑖
and the constant pressure and volume specific heats as:
𝑐p = ∑ 𝑓𝑖 𝑐p(𝑖) (33.50)

𝑐v = ∑ 𝑓𝑖 𝑐v(𝑖) , (33.51)
where
𝑓𝑖 = mass fraction of gas 𝑖
𝑀𝑖 = molecular weight of gas 𝑖
𝑐p(𝑖) = constant pressure specific heat of gas 𝑖
𝑐v(𝑖) = constant volume specific heat of gas 𝑖.

The specific enthalpy and internal energy for an ideal gas mixture with temperature
dependent heat capacity are
𝑇
𝑏𝑖 𝑇 2
ℎ = ∫ ∑ 𝑓𝑖 𝑐p(𝑖) 𝑑𝑇 = ∑ 𝑓𝑖 (𝑎𝑖 𝑇 + ) (33.52)
2
0

𝑇
𝑏𝑖 𝑇 2
𝑢 = ∫ ∑ 𝑓𝑖 𝑐v(𝑖) 𝑑𝑇 = ∑ 𝑓𝑖 (𝑎𝑖 𝑇 + − 𝑟𝑖 𝑇). (33.53)
2
0

The rate of change of temperature for the airbag is


𝑑𝑇cv ∑ 𝑚̇ 𝑖 ℎ𝑖 − ∑ 𝑚̇ 𝑜 ℎ𝑜 − 𝑊̇ cv − 𝑄̇cv − (𝑚̇ 𝑢)cv − (𝑚𝑐v̇ 𝑇)cv
= . (33.54)
𝑑𝑡 (𝑚𝑐v )cv

The energy in by mass flow becomes:


𝑏𝑖 𝑇𝑖2
∑ 𝑚̇ 𝑖 ℎ𝑖 = ∑ 𝑚̇ 𝑖 (𝑎𝑖 𝑇𝑖 + ), (33.55)
2
𝑚̇ 𝑖 is specified by an inflator mass inflow vs. time table
𝑇𝑖 is specified by an inflator temperature vs. time table
𝑎, 𝑏 are input constants for gas 𝑖

And the energy out by mass flow:

LS-DYNA DEV 10/27/16 (r:8004) 31-9 (Airbags)


Airbags LS-DYNA Theory Manual

2
𝑏𝑖 𝑇cv
∑ 𝑚̇ 𝑜 ℎ𝑜 = ∑ 𝑚̇ 𝑜 [ ∑ 𝑓𝑖 (𝑎𝑖 𝑇𝑐𝑣 + )]. (33.56)
gases 2

The gas leaves the airbag at the control volume temperature 𝑇𝑐𝑣 . The mass flow rate
out through vents and fabric leakage is calculated by the one dimensional isentropic flow
equations per Wang and Nefske. The work done by the airbag expansion is given by:

𝑊̇ cv = ∫ 𝑃𝑑𝑉̇ , (33.57)

𝑃 is calculated by the equation of state for a perfect gas, 𝑝 = 𝜌𝑅𝑇 and 𝑉̇ is calculated by
LS-DYNA

For the energy balance, we must compute the energy terms (𝑚̇ 𝑢)cv and (𝑚𝑐v )cv .
Conservation of mass leads to:
𝑚̇ cv = 𝑚̇ 𝑖 − 𝑚̇ 𝑜
(33.58)
𝑚cv = ∫ 𝑚̇ cv 𝑑𝑡.

The internal energy is given by


2
𝑏𝑖 𝑇cv
𝑢cv = ∑ 𝑓𝑖 (𝑎𝑖 𝑇cv + − 𝑟𝑖 𝑇cv ), (33.59)
2
and the heat capacity at contact volume is:
(𝑐v )cv = ∑ 𝑓𝑖 (𝑎𝑖 + 𝑏𝑖 𝑇cv − 𝑟𝑖 ). (33.60)

33.5 Constant Volume Tank Test


Constant volume tank tests are used to characterize inflators. The inflator is ignited
within the tank and, as the propellant burns, gas is generated. The inflator temperature is
assumed to be constant. From experimental measurements of the time history of the tank
pressure it is straightforward to derive the mass flow rate, 𝑚̇ . From energy conservation,
where 𝑇i and 𝑇t are defined to be the temperature of the inflator and tank, respectively, we
obtain:
𝑐p 𝑚̇ 𝑇i = 𝑐v 𝑚̇ 𝑇t + 𝑐v 𝑚𝑇̇t . (33.61)

For a perfect gas under constant volume, 𝑉̇ = 0, hence,


𝑝̇𝑉 = 𝑚̇ 𝑅𝑇t + 𝑚𝑅𝑇̇t , (33.62)

31-10 (Airbags) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Airbags

and, finally, we obtain the desired mass flow rate:


𝑐v 𝑝̇𝑉
𝑚̇ = . (33.63)
𝑐p 𝑅𝑇i

LS-DYNA DEV 10/27/16 (r:8004) 31-11 (Airbags)


LS-DYNA Theory Manual Dynamic Relaxation and System Damping

34
Dynamic Relaxation and System
Damping

Dynamic relaxation allows LS-DYNA to approximate solutions to linear and


nonlinear static or quasi-static processes. Control parameters must be selected with
extreme care or bad results can be obtained. The current methods are not compatible with
displacement or velocity boundary conditions, but various body loads, thermal loads,
pressures, and nodal loads are allowed. The solutions to most nonlinear problems are path
dependent, thus results obtained in the presence of dynamic oscillations may not be the
same as for a nonlinear implicit code, and they may diverge from reality.

In LS-DYNA we have two methods of damping the solution. The first named
“dynamic relaxation” is used in the beginning of the solution phase to obtain the initial
stress and displacement field prior to beginning the analysis. The second is system
damping which can be applied anytime during the solution phase either globally or on a
material basis.

34.1 Dynamic Relaxation For Initialization


In this phase only a subset of the load curves is used to apply the static load which is
flagged in the load curve section of the manual. The calculation begins and executes like a
normal LS-DYNA calculation but with damping incorporated in the update of the
displacement field.
Our development follows the work of Underwood [1986] and Papadrakakis [1981]
with the starting point being the dynamic equilibrium equation, Equation (23.1) with the
addition of a damping term, at time 𝑛:
𝐌𝐚𝑛 + 𝐂𝐯𝑛 + 𝐐𝑛 (𝐝) = 0, (34.1)

𝐐𝑛 (𝐝) = 𝐅𝑛 − 𝐏𝑛 − 𝐇𝑛 , (34.2)

LS-DYNA DEV 10/27/16 (r:8004) 32-1 (Dynamic Relaxation and System Damping)
Dynamic Relaxation and System Damping LS-DYNA Theory Manual

where we recall that 𝐌 is the mass matrix, 𝐂 is the damping matrix, 𝑛 indicates the nth
time step, 𝐚𝑛 is the acceleration, 𝐯𝑛 the velocity, and 𝐝 is the displacement vector. With Δ𝑡
as the fixed time increment we get for the central difference scheme:
𝑛+1⁄2 𝑛−1⁄2
(𝐝𝑛+1 − 𝐝𝑛 ) (𝐯 −𝐯 )
𝑛+1⁄2 (34.3)
𝐯 = ; 𝐚𝑛 = .
Δ𝑡 Δ𝑡
For 𝐯𝑛 we can assume an averaged value
1 𝑛+1⁄ 1
𝐯𝑛 = (𝐯 2 + 𝐯𝑛− ⁄2 ), (34.4)
2
and obtain
−1
𝑛+1⁄2 1 1 1 1 𝑛−1⁄2
𝐯 =( 𝐌 + 𝐂) [( 𝐌 − 𝐂) 𝐯 − 𝐐𝑛 ], (34.5)
Δ𝑡 2 Δ𝑡 2
𝑛+1⁄2 (34.6)
𝐝𝑛+1 = 𝐝𝑛 + Δ𝑡𝐯 .

In order to preserve the explicit form of the central difference integrator, 𝐌 and 𝐂
must be diagonal. For the dynamic relaxation scheme 𝐂 has the form
𝐂 = 𝑐 ⋅ 𝐌. (34.7)
If Equation (34.7) is substituted into (34.5) the following form is achieved
𝑛+1⁄2 2 − 𝑐Δ𝑡 𝑛−1⁄ 2Δ𝑡
𝐯 = 𝐯 2+ ⋅ 𝐌−1 ⋅ 𝐐𝑛 . (34.8)
2 + 𝑐Δ𝑡 2 + 𝑐Δ𝑡
Since 𝐌 is diagonal, each solution vector component may be computed individually from
𝑛+1⁄2 2 − 𝑐Δ𝑡 𝑛−1⁄2 2Δ𝑡 𝐐𝑛𝑖
𝐯𝑖 = 𝐯𝑖 + . (34.9)
2 + 𝑐Δ𝑡 2 + 𝑐Δ𝑡 𝑚𝑖
As a starting procedure it is suggested by Underwood
𝐯0 = 0
(34.10)
𝐝0 = 0.
Since the average value is used for 𝐯𝑛 , which must be zero at the beginning for a quasi-
static solution
−1⁄2 1⁄
(34.11)
𝐯 = −𝐯 2,

thus the velocity at time +1 ⁄ 2 is


Δ𝑡 −1 𝑜
1⁄
𝐯
𝐌 𝐐. 2 =− (34.12)
2
A damping coefficient must now be selected to obtain convergence to the static solution in
minimal time. The best estimate for damping values is based on the frequencies of the
structure. One choice is to focus on an optimal damping parameter as suggested by

32-2 (Dynamic Relaxation and System Damping) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Dynamic Relaxation and System Damping

Papadrakakis [1981]. Then dynamic relaxation is nothing else but a critically damped
dynamic system
𝐶 = 𝐶cr = 2𝜔min 𝑚, (34.13)
with 𝑚 as modal mass. The problem is finding the dominant eigenvalue in the structure
related to the “pseudo-dynamic” behavior of the structure. As the exact estimate would be
rather costly and not fit into the explicit algorithm, an estimate must be used.
Papadrakakis suggests
∥𝐝𝑛+1 − 𝐝𝑛 ∥
𝜆𝐷 = . (34.14)
∥ 𝐝𝑛 − 𝐝𝑛−1 ∥
When this quantity has converged to an almost constant value, the minimum eigenvalue of
the structure can be estimated:

2
(𝜆2𝐷 − 𝜆𝐷 ⋅ 𝛽 + 𝛼)
𝜔min =− , (34.15)
𝜆𝐷 ⋅ 𝛾
where
2 − 𝑐Δ𝑡
𝛼=
2 + 𝑐Δ𝑡
𝛽=𝛼+1 (34.16)
2Δ𝑡2
𝛾= .
2 + 𝑐Δ𝑡

The maximum eigenvalue determines the time step and is already known from the
model
2 4.0
𝜔max = . (34.17)
(Δ𝑡)2
Now the automatic adjustment of the damping parameter closely follows the paper of
Papadrakakis, checking the current convergence rate compared to the optimal convergence
rate. If the ratio is reasonably close, then an update of the iteration parameters is
performed.
2 2
4.0 √𝜔min ⋅ 𝜔max
(34.18)
𝑐= 2
.
Δ𝑡 (𝜔min 2
+ 𝜔max )
As is clearly visible from Equation (34.18) the value of highest frequency has always a
rather high influence on the damping ratio. This results in a non-optimal damping ratio, if
the solution is dominated by the response in a very low frequency compared to the highest
frequency of the structure. This is typically the case in shell structures, when bending
dominates the solution. It was our observation that the automatic choice following
Papadrakakis results in very slow convergence for such structures, and this is also
mentioned by Underwood for similar problems. The damping ratio should then be fully
adjusted to the lowest frequency by hand by simply choosing a rather high damping ratio.
LS-DYNA DEV 10/27/16 (r:8004) 32-3 (Dynamic Relaxation and System Damping)
Dynamic Relaxation and System Damping LS-DYNA Theory Manual

An automatic adjustment for such cases is under preparation. For structures with
dominant frequencies rather close to the highest frequency, convergence is really improved
with the automatically adjusted parameter.

If the automated approach is not used then we apply the damping as


𝑛+1⁄2 𝑛−1⁄2
𝐯 = 𝜂𝐯 + 𝐚𝑛 Δ𝑡, (34.19)

where 𝜂is an input damping factor (defaulted to .995). The factor, 𝜂, is equivalent to the
corresponding factor in Equations (31.7- 31.8).

The relaxation process continues until a convergence criterion based on the global
kinetic energy is met, i.e., convergence is assumed if
𝐸ke < CVTOL ⋅ 𝐸max
𝑘𝑒 , (34.20)
where CVTOL is the convergence tolerance (defaulted to .001). The kinetic energy excludes
any rigid body component. Initial velocities assigned in the input are stored during the
relaxation. Once convergence is attained the velocity field is initialized to the input values.
A termination time for the dynamic relaxation phase may be included in the input and is
recommended since if convergence fails, LS-DYNA will continue to execute indefinitely.

34.2 Mass Weighted Damping


With mass weighted damping, the Equation (23.2) is modified as:
𝑛
𝐚𝑛 = 𝐌−1 (𝐏𝑛 − 𝐅𝑛 + 𝐇𝑛 − 𝐅damp ), (34.21)
where
𝑛
𝐹damp = 𝐷𝑠 𝑚𝑣. (34.22)

As seen from Figure 34.1 and as discussed above the best damping constant for the
system is usually the critical damping constant: Therefore,
𝐷𝑠 = 2𝜔min (34.23)
is recommended.

34.3 Dynamic Relaxation—How Fast Does it Converge?

32-4 (Dynamic Relaxation and System Damping) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Dynamic Relaxation and System Damping

The number of cycles required to reduce the amplitude of the dynamic response by a factor
of 10 can be approximated by [see Stone, Krieg, and Beisinger 1985]
𝜔
ncycle = 1.15 max . (34.24)
𝜔min
Structural problems which involve shell and beam elements can have a very large ratio and
consequently very slow convergence.

Figure 34.1. Displacement versus time curves with a variety of damping


coefficients applied to a one degree-of-freedom oscillator.

LS-DYNA DEV 10/27/16 (r:8004) 32-5 (Dynamic Relaxation and System Damping)
LS-DYNA Theory Manual Heat Transfer

35
Heat Transfer

LS-DYNA can be used to solve for the steady state or transient temperature field on
three-dimensional geometries. Material properties may be temperature dependent and
either isotropic or orthotropic. A variety of time and temperature dependent boundary
conditions can be specified including temperature, flux, convection, and radiation. The
implementation of heat conduction into LS-DYNA is based on the work of Shapiro [1985].

35.1 Conduction of Heat in an Orthotropic Solid


The differential equations of conduction of heat in a three-dimensional continuum is
given by
∂𝜃
𝜌𝑐𝑝 = (𝑘𝑖𝑗 𝜃,𝑗 ) + 𝑄 (35.1)
∂𝑡 ,𝑖

subject to the boundary conditions, 𝜃 = 𝜃𝑠 on Γ1, 𝑘𝑖𝑗 𝜃,𝑗 𝑛𝑖 + 𝛽𝜃 = 𝛾 on Γ2, and initial
conditions at 𝑡0 :
𝜃Γ = 𝜃0 (𝑥𝑖 ) at 𝑡 = 𝑡0 . (35.2)
where
𝜃 = 𝜃(𝑥𝑖 , 𝑡) temperature
𝑥𝑖 = 𝑥𝑖 (𝑡) coordinates as a function of time
𝜌 = 𝜌(𝑥𝑖 ) density
𝑐𝑝 = 𝑐𝑝 (𝑥𝑖 , 𝜃) specific heat
𝑘𝑖𝑗 = 𝑘𝑖𝑗 (𝑥𝑖 , 𝜃) specific heat
𝑄 = 𝑄(𝑥𝑖 , 𝜃) internal heat generation rate per unit volume Ω
𝜃Γ = prescribed temperature on Γ1
𝑛𝑖 = normal vector to Γ2
Equations (35.1)-(35.2) represent the strong form of a boundary value problem to be solved
for the temperature field within the solid.

32.1
Heat Transfer LS-DYNA Theory Manual

DYNA3D employs essentially the same theory as TOPAZ [Shapiro 1985] in solving
Equation (35.1) by the finite element method. Those interested in a more detailed
description of the theory are referred to the TOPAZ User’s Manual. Brick elements are
integrated with a 2 × 2 × 2 Gauss quadrature rule, with temperature dependence of the
properties accounted for at the Gauss points. Time integration is performed using a
generalized trapezoidal method shown by Hughes to be unconditionally stable for
nonlinear problems. Newton’s method is used to satisfy equilibrium in nonlinear
problems.

The finite element method provides the following equations for the numerical
solution of Equations (35.1)-(35.2)
𝐶𝑛+𝛼
[ + 𝛼𝐻𝑛+𝛼 ] {𝜃𝑛+1 − 𝜃𝑛 } = {𝐹𝑛+𝛼 − 𝐻𝑛+𝛼 𝜃𝑛 } (35.3)
Δ𝑡
where

[𝐶] = ∑[𝐶𝑖𝑗𝑒 ] = ∑ ∫ 𝑁𝑖 𝜌𝑐𝑁𝑗 𝑑Ω (35.4)


𝑒 𝑒 Ω𝑒

[𝐻] = ∑[𝐻𝑖𝑗𝑒 ] = ∑ ⎡
⎢ ∫ ∇ 𝑁𝑖 𝐾∇𝑁𝑗 𝑑Ω + ∫ 𝑁𝑖 𝛽𝑁𝑗 𝑑Γ⎤
𝑇
⎥ (35.5)
𝑒 𝑒 ⎣Ω𝑒 Γ𝑒 ⎦

[𝐹] = ∑[𝐹𝑖𝑒 ] = ∑ ⎡
⎢ ∫ 𝑁𝑖 𝑞𝑔 𝑑Ω + ∫ 𝑁𝑖 𝛾𝑑Γ⎤
⎥ (35.6)
𝑒 𝑒 ⎣Ω𝑒 Γ 𝑒 ⎦

The parameter 𝛼 is taken to be in the interval [0,1]. Some well-known members of


this 𝛼-family are
•𝛼 Method
• 0 forward difference; forward Euler
• 1⁄2 midpoint rule; Crank-Nicolson
• 2⁄3 Galerkin
• 1 backward difference, fully implicit

35.2 Thermal Boundary Conditions


Boundary conditions are represented by
∂𝜃 ∂𝜃 ∂𝜃
𝑘𝑥 𝑛𝑥 + 𝑘𝑦 𝑛𝑦 + 𝑘𝑧 𝑛𝑧 = 𝛾 − 𝛽𝜃 = 𝑞′′̇ . (35.7)
∂𝑥 ∂𝑦 ∂𝑧

32.2
LS-DYNA Theory Manual Heat Transfer

By convention, heat flow is positive in the direction of the surface outward normal vector.
Surface definition is in accordance with the right hand rule. The outward normal vector
points to the right as one progresses from node N1 to N2 to N3 and finally to N4. See
Figure 35.1.

Boundary conditions can be functions of temperature or time. More than one


boundary condition can be specified over the same surface such as in a case of combined
convection and radiation. For situations where it is desired to specify adiabatic (i.e., 𝑞′′̇ =
0) conditions, such as at an insulated surface or on a line of symmetry, no boundary
condition need be specified. This is the default boundary condition in LS-DYNA.

Temperature boundary condition can be specified on any node whether on the


physical boundary or not.

Flux, convection, and radiation boundary conditions are specified on element


surface segments defined by 3 (triangular surface) or 4 nodes (quadrilateral surface). These
boundary conditions can be specified on any finite element surface whether on the physical
boundary or not.
• Flux: Set 𝑞′′̇ = 𝑞𝑓 , where 𝑞𝑓 is defined at the node points comprising the flux
boundary condition surface.
• Convection: A convection boundary condition is calculated using 𝑞′′̇ = ℎ(𝑇 − 𝑇∞ ),
where ℎ is heat transfer coefficient, (𝑇 − 𝑇∞ ) is temperature potential. LS-DYNA
evaluates ℎ at the film temperature

1
𝑇 = (𝑇surf + 𝑇∞ ) (35.8)
2

• Radiation: A radiation boundary condition is calculated using 𝑞′′̇ = ℎ 𝑟 (𝑇 4 − 𝑇∞


4
),
where ℎ 𝑟 = 𝜎𝜀𝐹 is a radiant-heat-transfer coefficient.

N4

N1

N3
N2
Figure 35.1. Definition of the outward normal vector

32.3
Heat Transfer LS-DYNA Theory Manual

35.3 Thermal Energy Balances


Various energy terms are printed and written into the plot file for post processing
using the code LS-PREPOST. The energy terms are:

• change in material internal energy for time step,


• change in material internal energy from initial time,
• heat transfer rates on boundary condition surfaces,
• heat transfer rates on enclosure radiation surfaces,
• 𝑥, 𝑦, and 𝑧 fluxes at all nodes.

35.4 Heat Generation


Volumetric heat generation rates may be specified by element, by material, or both
(in which case the effect is additive). Volumetric heat generation rates can be a function of
time or temperature.

35.5 Initial Conditions


Initial temperature conditions can be specified on the nodal data input cards or on
the nodal temperature initial condition cards. If no temperatures are specified, the default
is 0. For nonlinear steady state problems the temperature initial condition serves as a first
guess for the equilibrium iterations.

35.6 Material Properties


Heat capacity and thermal conductivity may be functions of temperature. Since the
density and heat capacity appear only as a product in the governing equations, the
temperature dependence of the density may be included in the temperature dependence of
the heat capacity. Material properties are evaluated at the element Gauss point
temperature or average element temperature.

The thermal conductivity may be either isotropic or orthotropic. For an orthotropic


material, the three material axes (𝑥′1 , 𝑥′2 , 𝑥′3 ) are orthogonal and the thermal conductivity
tensor 𝐊 is diagonal.

The thermal conductivity tensor 𝐊 in the global coordinate system is related by


32.4
LS-DYNA Theory Manual Heat Transfer

𝐾𝑖𝑗 = 𝐾′𝑖𝑗 𝛽𝑚𝑖 𝛽𝑛𝑗 , (35.9)


where
𝛽𝑖𝑗 = cos(𝑥′𝑖 , 𝑥𝑗 ). (35.10)

35.7 Nonlinear Analysis


In a nonlinear problem, 𝐶, 𝐻, and 𝐹 are functions of temperature. Newton’s method
is used to transform equation 32.4 into an alternate form which contains temperature
derivatives of 𝐶, 𝐻, and 𝐹 (i.e., the tangent matrix). Iterations are required to solve this
alternate form.

In a steady state nonlinear problem, an initial guess should be made of the final
temperature distribution and included in the input file as an initial condition. If your guess
is good, a considerable savings in computation time is achieved.

35.8 Units
Any consistent set of units with the governing equation may be used. Examples are:

Quantity Units
temperature K C F
space m cm ft
time s s hr
density kg/m3 g/cm3 Lbm/ft3
heat capacity J/kg k cal/g c Btu/LbmF
thermal conductivity W/m K cal/s cm C Btu/hr ft F
thermal generation W/M3 cal/s cm3 Btu/hr ft3
heat flux W/m2 cal/s cm2 Btu/hr ft2

32.5
LS-DYNA Theory Manual Adaptivity

36
Adaptivity

LS-DYNA includes an h-adaptive method for the shell elements. In an h-adaptive


method, the elements are subdivided into smaller elements wherever an error indicator
shows that subdivision of the elements will provide improved accuracy. An example of an
adaptive calculation on a thin wall square cross section beam is shown in Figure 36.1. In
Figures 36.2 through 36.4 a simple metal stamping simulation is shown [also see Galbraith,
Finn, et. al., 1991]. In the following, the methodologies used in the h-adaptive method in
LS-DYNA are described. The objective of the adaptive process used in LS-DYNA is to
obtain the greatest accuracy for a given set of computational resources. The user sets the
initial mesh and the maximum level of adaptivity, and the program subdivides those
elements in which the error indicator is the largest. Although this does not provide control
on the error of the solution, it makes it possible to obtain a solution of comparable accuracy
with fewer elements, and, hence, less computational resources, than with a fixed mesh.

LS-DYNA uses an h-adaptive process, where parts of the mesh are selectively
refined during the course of the solution procedure. The methodology used is based on
Belytschko, Wong, and Plaskacz [1989]. In the former, elements were also fused or
combined when it was felt that they were no longer needed. It was found that the
implementation of fusing procedures for general meshes, such as occur in typical
applications of commercial programs, is too complex, so only fission is included.
Adaptivity in LS-DYNA can be restricted to specific groups of shell elements. Elements
that fall in this group are said to be in the active adaptivity domain.

LS-DYNA DEV 10/27/16 (r:8004) 34-1 (Adaptivity)


Adaptivity LS-DYNA Theory Manual

Figure 36.1. One level adaptive calculation on a square cross section beam.

Figure 36.2. Aluminum blank with 400 shells in blank and four rigid tools.

34-2 (Adaptivity) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Adaptivity

Figure 36.3. Adaptive calculations using two adaptive levels.

In the h-adaptive process, elements are subdivided into smaller elements where
more accuracy is needed; this process is called fission. The elements involved in the fission
process are subdivided into elements with sides ℎ/2, where ℎ is the characteristic size of the
original elements. This is illustrated in Figure 36.5 for a quadrilateral element. In fission,
each quadrilateral is subdivided into four quadrilaterals (as indicated in Figure 36.2) by
using the mid-points of the sides and the centroid of the element to generate four new
quadrilaterals.

Figure 36.4. Final shape of formed part with 4315 shell elements per quarter.

LS-DYNA DEV 10/27/16 (r:8004) 34-3 (Adaptivity)


Adaptivity LS-DYNA Theory Manual

Figure 36.5. Fissioning of a Quadrilateral Element

The fission process for a triangular element is shown in Figure 36.6 where the
element is subdivided into four triangles by using the mid-points of the three sides. The
adaptive process can consist of several levels of fission. Figure 36.5 shows one subdivision,
which is called the second refinement level. In subsequent steps, the fissioned elements can
again be fissioned in a third refinement level, and these elements can again, in turn, be
fissioned in a fourth level, as shown in Figure 36.7. The levels of adaptivity that occur in a
mesh are restricted by three rules:
• The number of levels is restricted by the maximum level of adaptivity that is
allowed in the mesh, which is generally set at 3 or 4. At the fourth level up to 64
elements will be generated for each element in the initial mesh.
• The levels of adaptivity implemented in a mesh must be such that the levels of
adaptivity implemented in adjacent elements differ by, at most, one level.
• The total number of elements can be restricted by available memory. Once the
specified memory usage is reached, adaptivity ceases.

The second rule is used to enforce a 2-to-1 rule given by Oden, Devloo and
Strouboulis [1986], which restricts the number of elements along the side of any element in
the mesh to two. The enforcement of this rule is necessary to accommodate limitations in
the data structure.

The original mesh provided by the user is known as the parent mesh, the elements
of this mesh are called the parent elements, and the nodes are called parent nodes. Any

Figure 36.6. Fissioning of a Triangular Element

elements that are generated by the adaptive process are called descendant elements, and
34-4 (Adaptivity) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Adaptivity

Figure 36.7. Quadrilateral Element Fissioned to the fourth level

any nodes that are generated by the adaptive process are called descendant nodes.
Elements generated by the second level of adaptivity are called first-generation elements,
those generated by third level of adaptivity are called second-generation elements, etc.

The coordinates of the descendant nodes are generated by using linear interpolation.
Thus, the coordinates of any node generated during fission of an element are given by
1
𝑥𝑁 = (𝑥𝐼 + 𝑥𝐽 ), (36.1)
2
where 𝑥𝑁 is the position of the generated node and 𝑥𝐼 and 𝑥𝐽 are the nodes along the side
on which 𝑥𝑁 was generated for a typical element as shown in Fig. 33.1. The coordinate of
the mid-point node, which is generated by fission of a quadrilateral element, is given by
1
𝑥𝑀 = (𝑥𝐼 + 𝑥𝐽 + 𝑥𝐾 + 𝑥𝐿 ), (36.2)
4
where 𝑥𝑀 is the new midpoint node of the fissioned quadrilateral and 𝑥𝐼 , 𝑥𝐽 , 𝑥𝐾 and 𝑥𝐿 are
the nodes of the original quadrilateral. The velocities of the nodes are also given by linear
interpolation. The velocities of edge nodes are given by
1
𝑣𝑁 = (𝑣𝐼 + 𝑣𝐽 ), (36.3)
2
and the angular velocities are given by
1
𝜔𝑁 = (𝜔𝐼 + 𝜔𝐽 ). (36.4)
2
The velocities of a mid-point node of a fissioned quadrilateral element are given by
1
𝑣𝑀 = (𝑣𝐼 + 𝑣𝐽 + 𝑣𝐾 + 𝑣𝐿 ), (36.5)
4
1
𝜔𝑀 = (𝜔𝐼 + 𝜔𝐽 + 𝜔𝐾 + 𝜔𝐿 ). (36.6)
4

The stresses in the descendant element are obtained from the parent element by
setting the stresses in the descendant elements equal to the stresses in the parent element at
the corresponding through-the-thickness quadrature points.

LS-DYNA DEV 10/27/16 (r:8004) 34-5 (Adaptivity)


Adaptivity LS-DYNA Theory Manual

out-of-plane

undeformed deformed
Figure 36.8. Refinement indicator based on angle change.

In subsequent steps, nodes which are not corner nodes of an all attached elements
are treated as slave nodes. They are handled by the simple constraint equation.

Refinement indicators are used to decide the locations of mesh refinement. One
deformation based approach checks for a change in angles between contiguous elements as
shown in Figure 36.8. If 𝜁 > 𝜁tol then refinement is indicated, where 𝜁tol is user defined.

34-6 (Adaptivity) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Adaptivity

Figure 36.9. The input parameter, ADPASS, controls whether LS-DYNA backs
up and repeats the calculation after adaptive refinement.

After the mesh refinement is determined, we can refine the mesh and continue the
calculation or back up to an earlier time and repeat part of the calculation with the new
mesh. For accuracy and stability reasons the latter method is generally preferred; however,
the former method is preferred for speed. Whether LS-DYNA backs up and repeats the
calculation or continues after remeshing is determined by an input parameter, ADPASS.
This is illustrated in Figure 36.9.

LS-DYNA DEV 10/27/16 (r:8004) 34-7 (Adaptivity)


LS-DYNA Theory Manual Implicit

37
Implicit

37.1 Introduction
Implicit solvers are properly applied to static, quasi-static, and dynamic problems with a
low frequency content. Such applications include but are not limited to
• Static and quasi-static structural design and analysis
• Metal forming, especially, the binderwrap and springback
• Gravitational loading of automotive structures
• Linear buckling and vibration analysis
An advantage of the implicit solver on explicit integration is that the number of load or
time steps is typically 100 to 10000 times fewer. The major disadvantage is that the cost per
step is unknown since the speed depends mostly on the convergence behavior of the
equilibrium iterations which can vary widely from problem to problem.

An incremental-iterative numerical algorithm is implemented in LS-DYNA. The method is


stable for wide range of nonlinear problems that involve finite strain and arbitrarily large
rotations. Accuracy consideration usually limits the load increment or time step size. An
inaccurate solution will often not converge. Nine iterative schemes are available including
the full Newton method and eight quasi-Newton methods. These are:
• Full Newton,
• BFGS (default),
• Broyden,
• Davidon-Fletcher-Powell (DFP) [Schweizerhof 1986],
• Davidon symmetric, [Schweizerhof 1986],
• modified constant arc length with BFGS,
• modified constant arc length with Broyden’s,

LS-DYNA DEV 10/27/16 (r:8004) 35-1 (Implicit)


Implicit LS-DYNA Theory Manual

• modified constant arc length with DFP,


• modified constant arc length with Davidon.
A line search is combined with each of these schemes along with automatic stiffness
reformations, as needed, to avoid non-convergence. LS-DYNA defaults to the BFGS quasi-
Newton method which is the most robust although the other methods are sometimes
superior. Generally, the quasi-Newton methods require fewer iterations than the modified
Newton method since they exhibit superlinear local convergence due to the rank one or
rank two updates of the stiffness matrix as the iterations proceed. In this chapter,
important aspects of the static and dynamic algorithm are explained, hopefully, in a way
that will be understandable to all users. The arc length methods are generally used in
solving snap through buckling problems and details on one specific implementation is
taken up in the next chapter.

37.2 Equations
37.2.1 Discretization
Neglecting constraints, discretization formally leads to the matrix equations of motion
𝑹 = 𝑴𝒙̈ + 𝑭𝑖 − 𝑭𝑒 = 𝟎 (37.1)
where
𝒙̈ = acceleration vector of length 𝑛
𝑴 = 𝑛 × 𝑛 mass matrix
𝑭𝑒 = body force and external load vector of length 𝑛
𝑭𝑖 = internal force vector of length 𝑛.

It is implicitly assumed that the involved vectors 𝑭𝑖 and 𝑭𝑒 depend on6

𝒙̇, 𝒙 = velocity and coordinate vectors of length 𝑛


𝑡 = simulation time
as well as on some history of the deformation, while 𝑴 is constant. In practice, the only
independent variable is 𝒙 since the velocity 𝒙̇ and acceleration 𝒙̈ are typically expressed in
terms of this coordinate vector by the time integration scheme used, and 𝑡 is given.
Deformation history is typically accounted for in internal variables associated with features
in the model, such as plastic strains in materials and frictional sliding in contacts.

6 This is true prior to time discretization, vectorsand depend on and on only through the exact time
differentiation. Once time discretization is done according to some scheme, the dependence is on and not
necessarily on the discretized but rather ∆ /∆ , see Section 1.4.1.
35-2 (Implicit) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Implicit

A diagonal lumped mass matrix is obtained by row summing according to

𝑀𝑖𝑖 = ∫ 𝜌𝜙𝑖 ∑ 𝜙𝑗 𝑑𝑉 = ∫ 𝜌𝜙𝑖 𝑑𝑉 (37.2)


𝑗
𝑉 𝑉

where 𝜌 is material density and 𝑉 denotes the body over which integration occurs.

The primary nonlinearities, which are due to geometric effects and inelastic material
behavior, are accounted for in 𝑭𝑖 ,

𝑭𝑖 = ∫ 𝑩𝑇 𝝈𝑑𝑉 , (37.3)
𝑉

where 𝑩 is the strain-displacement matrix and 𝝈 is the stress.

Additional nonlinearities arise in 𝑭𝑒 due to geometry dependent applied loads, such as


contacts and loads on segments.

Explicit integration trivially satisfies (37.1) since the calculation of the acceleration
guarantees equilibrium, i.e., from time step 𝑗 to 𝑗 + 1 we use
𝑗 𝑗
𝒙̈𝑗 = 𝑴 −1 [𝑭𝑒 − 𝑭𝑖 ]. (37.4)
The explicit update of the velocities and coordinates is given by
𝒙̇ 𝑗+1/2 = 𝒙̇ 𝑗−1/2 + 𝛥𝑡𝑗 𝒙̈𝑗 , (37.5)

𝒙 𝑗+1 = 𝒙 𝑗 + 𝛥𝑡𝑗+1/2 𝒙̇𝑗+1/2 . (37.6)


Stability places a limit on the time size. This step size may be very small and, consequent-
ly, a large number of steps may be required. Implicit analysis employs schemes that are
unconditionally stable, thus allowing for larger steps at the cost of a more expensive update
of the geometry.

37.2.2 Constraints
Constraints is obviously an important ingredient in nonlinear finite elements, and may
include for instance simple point or motion constraints, slave nodes constrained to rigid
bodies, joints and tied contacts. The principle behind (37.1) is that of virtual work, stating
that
𝛿𝒙 𝑇 𝑹 = 0 (37.7)
for any admissible virtual displacement field 𝛿𝒙. If there are no constraints, any
displacement field is admissible and hence 𝑹 = 𝟎. The presence of constraints put

LS-DYNA DEV 10/27/16 (r:8004) 35-3 (Implicit)


Implicit LS-DYNA Theory Manual

restrictions on 𝛿𝒙 and the following is an attempt to derive the proper nonlinear equations
in this context.

Constraints augment the 𝑛 equations of motion by imposing 𝑚 additional equations


𝑯(𝒙, 𝑡) = 𝟎, (37.8)
and consequently (37.1) is to be reduced to 𝑛 − 𝑚 equations. To this end, we introduce the
𝑚 × 𝑛 constraint matrix
𝜕𝑯
𝑪= , (37.9)
𝜕𝒙
and conveniently partition the global vector 𝒙 into an independent (solution) part 𝒙𝐼 of
length 𝑛 − 𝑚 and dependent part 𝒙𝐷 of length 𝑚. This partitioning is represented by
projection matrices 𝑷𝐼 and 𝑷𝐷 such that
𝒙𝐼 = 𝑷𝐼 𝒙, 𝒙𝐷 = 𝑷𝐷 𝒙 (37.10)
and
𝒙 = 𝑷𝐼𝑇 𝒙𝐼 + 𝑷𝐷
𝑇
𝒙𝐷 , (37.11)
and the space of admissible virtual displacements is that of any 𝛿𝒙𝐼 . The criterion for a valid
partitioning is that the 𝑚 × 𝑚 matrix given by
𝑇 (37.12)
𝑪𝐷 = 𝑪𝑷𝐷 ,
is non-singular, which is always possible unless there are conflicting or redundant
constraints. The Linear Constraint PACKage (LCPACK) in LS-DYNA performs this
partitioning and invalid constraints will result in error termination. Introducing the 𝑚 ×
(𝑛 − 𝑚) matrix
𝑪𝐼 = 𝑪𝑷𝐼𝑇 (37.13)
we can combine the variations of (37.8) and (37.11) and use (37.12) and (37.13) to obtain
𝛿𝒙 = 𝑷 𝑇 𝛿𝒙𝐼 , (37.14)
where
−𝑇
𝑷 = 𝑷𝐼 − 𝑪𝐼𝑇 𝑪𝐷 𝑷𝐷 (37.15)
is a projection-like matrix from the global variable space to the independent solution space.
Using this expression for 𝛿𝒙 in the principle of virtual work (37.7), the arbitrariness of 𝛿𝒙𝐼
results in
𝑸 = 𝟎, (37.16)
with
𝑸 = 𝑷𝑹. (37.17)
This can be interpreted as to solve for zero force in the direction of deformation that is not
constrained, in other directions the constraints induce reaction forces that are typically
monitored in the LS-DYNA ascii and binout databases associated with the constraint type.

35-4 (Implicit) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Implicit

37.3 Implicit Statics


37.3.1 Linearization
For the implicit static solution 𝑴 = 𝟎 in (37.1) and the residual and constraint vector at a
given time 𝑡 becomes an implicit function of 𝒙 only. We seek the vector 𝒙 such that (37.8)
holds together with
𝑸(𝒙) = 𝟎. (37.18)
Assume an approximation 𝒙 𝑘 to 𝒙 for 𝑘 = 1, 2, 3...etc. In the neighborhood of 𝒙 𝑘 we use a
linear approximation to (37.8) and (37.18) given by
𝑲(𝒙 𝑘 )𝛥𝒙𝐼𝑘 = 𝑭(𝒙 𝑘 ), (37.19)
and iterate for the solution
𝒙𝐼𝑘+1 = 𝒙𝐼𝑘 + 𝑠𝛥𝒙𝐼𝑘 . (37.20)
Here 𝑠 is a convenient step size to be discussed below, as will be the update of the
𝑘+1
dependent part 𝒙𝐷 . The linear system (37.19) is derived using the assumption that 𝑷 is
independent of 𝒙, which is generally not true. Many constraints in LS-DYNA are nonlinear
and a strict linearization would have to take the second variation of constraints into
account. But this would be inherently complex, thus linearizing (37.18) using (37.17) just
becomes
𝜕𝑹 𝑘
𝑸(𝒙 𝑘+1 ) ≈ 𝑸(𝒙 𝑘 ) + 𝑷 (𝒙 )∆𝒙 𝑘 = 𝟎. (37.21)
𝜕𝒙
To be able to solve for ∆𝒙 𝑘 this needs to be combined with a linearized equation of the
constraint (37.8)
𝑯(𝒙 𝑘+1 ) ≈ 𝑯(𝒙 𝑘 ) + 𝑪(𝒙 𝑘 )∆𝒙 𝑘 = 𝟎, (37.22)
and an incremental correspondent to (37.11)
∆𝒙 𝑘 = 𝑷𝐼𝑇 ∆𝐱𝐼𝑘 + 𝑷𝐷
𝑇 𝑘
∆𝐱𝐷 . (37.23)
These last two equations together with (37.12) and (37.13) can be used to yield an
expression for the dependent part of the displacement increment
𝑘 −1 𝑘
∆𝐱𝐷 = −𝑪𝐷 (𝒙 )[𝑯(𝒙 𝑘 ) + 𝑪𝐼 (𝒙 𝑘 )∆𝐱𝐼𝑘 ]. (37.24)
Using (19.22) in (37.23) in (37.21) results in expressions of the vector 𝑭 and jacobian 𝑲 in
(37.19) as
𝜕𝑹 𝑘 𝑇 −1 𝑘
𝑭(𝒙 𝑘 ) = −𝑸(𝒙 𝑘 ) + 𝑷 (𝒙 )𝑷𝐷 𝑪𝐷 (𝒙 )𝑯(𝒙 𝑘 ) (37.25)
𝜕𝒙
and
𝜕𝑹 𝑘 𝑇
𝑲(𝒙 𝑘 ) = 𝑷 (𝒙 )𝑷 . (37.26)
𝜕𝒙

LS-DYNA DEV 10/27/16 (r:8004) 35-5 (Implicit)


Implicit LS-DYNA Theory Manual

The dependent part of the solution vector is in general analogous to (37.20) of the
indendent part
𝑘+1 𝑘 𝑘
𝒙𝐷 = 𝒙𝐷 + 𝑠∆𝒙𝐷 (37.27)
except for those constraint equations where there is an explicit expression 𝒙𝐷 = 𝒙𝐷 (𝒙𝐼 ),
then of course
𝑘+1
𝒙𝐷 = 𝒙𝐷 (𝒙𝐼𝑘+1 ). (37.28)
The jacobian matrix 𝑲 is the assembly of the tangent moduli of materials, external loads,
contacts, etc., and is in practice only an approximation due to the complexity of taking all
dependencies into account. In particular it does not include the geometric stiffness
contribution from internal forces by default, see IGS on CONTROL_IMPLICIT_GENERAL.
A speculative reason is that this has a smoothing effect and eliminates negative eigenvalues
due to compressive stresses. If the deformation mode is known to be mainly in tension or
if the material is hyper-elastic, including the geometric stiffness could improve
convergence however. Often the stiffness matrix is assumed symmetric and positive
definite, but is not limited to those characteristics. Note also that (37.25) indicates that both
𝑸 and 𝑯 must vanish to render a zero 𝑭, thus a zero displacement increment ∆𝒙 in the
iterative scheme.

Figure 3735-1 Linear axial buckling of an aluminium beverage can.

37.3.2 Linear theory


For a linear solution, (37.19) is solved once to obtain the linear displacement vector 𝒖 = ∆𝒙,
and often the following assumptions apply. The configuration 𝒙 to which the linear
approximation apply is stress free and all constraints are fulfilled, so the right hand side 𝑭
in (37.25) is essentially the external applied load 𝑭𝑒 , 𝑭 = 𝑷𝑭𝑒 , which is constant.

35-6 (Implicit) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Implicit

Furthermore, 𝑷 is constant due to trivial constraints and the stiffness matrix 𝑲 in (37.26)
thus evaluates from the linearization of internal forces
𝑲 = 𝑷 ∫ 𝑩𝑇 𝑬𝑩𝑑𝑉 𝑷 𝑇 , (37.29)
𝑉
where we used 𝑬 = 𝜕𝝈 𝜕𝜺
to denote the constitutive matrix. So in more conventional terms,
the linear equation is written
𝑲𝒖 = 𝑭. (37.30)
Once solved, the stress
𝝈 = 𝑬𝑩𝒖 (37.31)
can be evaluated from the constitutive law for the resulting deformation. The stiffness
matrix 𝑲 in (37.30) is symmetric and positive definite if 𝑬 is and 𝑷 eliminates all rigid body
modes, whence the solution 𝒖 is unique. Furthermore, linearity implies that substituting
the right hand side for 𝑭𝜆 = 𝜆𝑭, the solution changes to 𝒖𝜆 = 𝜆𝒖 and the resulting stress
changes to
𝝈𝜆 = 𝜆𝝈. (37.32)
A linear solution is obtained by putting NSOLVR = 1 on CON-
TROL_IMPLICIT_SOLUTION.

Now go back to the nonlinear static equation and scale a constant external load 𝑭𝑒 with 𝜆
𝑷[𝑭𝑖 − 𝜆𝑭𝑒 ] = 𝟎 (37.33)
and assume an updated configuration 𝒙𝜆 with resulting stress 𝝈𝜆 to be the solution. To see
how the solution changes with 𝜆 we can perturb (37.33) to obtain
𝜕𝑭
𝑷 [ 𝑖 𝑷 𝑇 𝛿𝒙𝜆 − 𝛿𝜆𝑭𝑒 ] = 𝟎 (37.34)
𝜕𝒙
where we applied (37.14). This is conveniently rewritten as
𝑲𝜆 𝛿𝒙𝜆 = 𝛿𝜆𝑭, (37.35)
where the presence of stress will change the evaluation of the stiffness matrix from that in
(37.29) to
𝑲𝜆 = 𝑲 + 𝑷 ∫ 𝓑 𝑇 𝝈𝜆 𝓑𝑑𝑉 𝑷 𝑇 . (37.36)
𝑉
Here the second term is the geometric contribution to the tangent stiffness matrix where𝓑
is a matrix consisting of shape function derivatives for the finite element. We are interested
in the question of uniqueness of 𝛿𝒙𝜆 as the solution to (37.35), and this uniqueness is lost
when
det(𝑲𝜆 ) = 0. (37.37)
If one assumes that deformations are small enough to keep 𝑉, 𝑩, 𝑬 and 𝓑 independent of 𝜆,
and that (37.32) holds with (37.31) and (37.30), then (37.37) can be stated as the following
eigenvalue problem
𝑲𝛿𝒖 = −𝜆𝑲𝜎 𝛿𝒖 (37.38)
with the geometric (nonlinear) stiffness matrix given by
𝑲𝜎 = 𝑷 ∫ 𝓑 𝑇 𝝈𝓑𝑑𝑉 𝑷 𝑇 (37.39)
𝑉
and 𝑲 is the material (linear) stiffness matrix again given by (37.29). This is the theory
behind linear buckling, see CONTROL_IMPLICIT_BUCKLE, with 𝜆 being the buckling

LS-DYNA DEV 10/27/16 (r:8004) 35-7 (Implicit)


Implicit LS-DYNA Theory Manual

load parameter and 𝛿𝒖 the associated buckling mode. Usually solutions with 𝜆 > 0 are of
interest, and from (37.38) and (37.39) it then follows that the principal stresses cannot be
positive throughout the domain of integration. In other words, the model must somehow
be in a compressed state. The full procedure is
1.Assemble 𝑲 by (37.29).
2.Solve (37.30) for a constant reference load 𝑭.
3.Evaluate resulting stress 𝝈 by (37.31).
4.Assemble 𝑲𝜎 by (37.39).
5.Solve (37.38) for 𝜆 and 𝛿𝒖.

An example of linear buckling is shown in Figure 3735-1, simulating the stepping on an


aluminium beverage can.

𝐹0 𝐹0
𝐹 𝐹

𝐹1 𝐹1
𝐹2
𝐹2
𝑥 𝑥
𝑥2 𝑥1 𝑥0 𝑥2 𝑥1 𝑥0

Figure 3735-2 Full Newton (left) compared to quasi-Newton secant updates (right),
the linear approximation to obtain 𝑥2 is for full Newton the exact tangent in
(𝑥1 , 𝐹1 ) while it is the linear extension between the points (𝑥0 , 𝐹0 ) and (𝑥1 , 𝐹1 ) for

37.3.3 Quasi-Newton iterations


Now back to the nonlinear theory. To obtain the solution at load increment 𝑗 + 1 given the
solution at load increment 𝑗, linearized equations of the form
𝑲𝑘 𝛥𝒙𝑘 = 𝑭𝑘 , (37.40)
are assembled and solved where 𝑘 is the iteration number and
𝑲𝑘 = Tangent stiffness matrix
Δ𝒙𝑘 = Desired increment in displacements
𝑭𝑘 = Residual load vector.
This should be seen a generalization of (37.19) in the sense that the tangent stiffness matrix
𝑲 need not be calculated as (37.26) but can be based on other approximations to be
presented. We do however assume, without loss of generality, that (37.40) is in the

35-8 (Implicit) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Implicit

independent system of variables so the residual vector 𝑭 is given by (37.25). We here just
substitute the sub-index 𝐼 in the incremental displacement ∆𝒙 for the iteration counter 𝑘 to
simplify the notation in the following. The coordinate vector is updated
𝒙𝑘+1 = 𝒙𝑘 + sΔ𝒙𝑘 , (37.41)
where 𝑠 is a parameter between 0 and 1 found from a line search.

If the tangent stiffness matrix is calculated as (37.26) for each 𝑘 then this is termed a full
Newton method, but it may be beneficial to use so called quasi-Newton updates of 𝑲 −1 to
avoid the cost of solving a linear system of equations in each iteration. Four such methods
for updating the stiffness matrix are available
• Broyden’s first method
• Davidon
• DFP
• BFGS
and these involve rank 1 or rank 2 stiffness updates. Quasi-Newton methods are less
expensive than the full Newton method but often still result in robust program. In one
dimension, quasi-Newton corresponds to secant iterations and this method compared to
full Newton is illustrated in Figure 3735-2. Henceforth we assume that 𝑲0 represents the
last assembled matrix according to (37.26) and then 𝑲𝑘 , 𝑘 = 1,2,3, … are quasi-Newton
updates to be given in the following.

The secant matrices 𝑲𝑘 are found via the quasi-Newton equations


𝑲𝑘 Δ𝒙𝑘−1 = Δ𝑭𝑘 , (37.42)
where
Δ𝑭𝑘 = 𝑭𝑘−1 − 𝑭𝑘 . (37.43)
Two classes of matrix updates that satisfy the quasi-Newton equations are of interest; rank
1 updates
𝑲𝑘 = 𝑲𝑘−1 + 𝛼𝒛𝒚𝑇 , (37.44)
and rank 2 updates
𝑲𝑘 = 𝑲𝑘−1 + 𝛼𝒛𝒚𝑇 + 𝛽𝒗𝒖𝑇 . (37.45)
Substituting (37.44) into (37.42) gives
𝑲𝑘−1 𝛥𝒙𝑘−1 + 𝛼𝒛𝒚𝑇 𝛥𝒙𝑘−1 = 𝛥𝑭𝑘 . (37.46)
1
By choosing 𝛼 = 𝒚𝑇 𝛥𝒙 and 𝒛 = −𝑭𝑘 , equation (37.42) is satisfied. Note that 𝒚 is an
𝑘−1
arbitrary vector but is restricted such that
𝒚𝑇 𝛥𝒙𝑘−1 ≠ 0. (37.47)
The tangent stiffness update is

LS-DYNA DEV 10/27/16 (r:8004) 35-9 (Implicit)


Implicit LS-DYNA Theory Manual

𝑭𝑘 𝒚 𝑇
𝑲𝑘 = 𝑲𝑘−1 − , (37.48)
𝒚𝑇 𝛥𝒙𝑘−1
resulting generally in non-symmetric secant matrices. The inverse forms are found by the
Sherman-Morrison formula
−1 𝑨−1 𝒂 𝒃𝑇 𝑨−1
(𝑨 + 𝒂𝒃𝑇 ) = 𝑨−1 − . (37.49)
1 + 𝒃𝑇 𝑨−1 𝒂
where 𝑨 is a nonsingular matrix and 𝒂 and 𝒃 are arbitrary vectors such that 1 + 𝒃𝑇 𝑨−1 𝒂 ≠
0. The inverse form for (37.48) can be found by letting
−𝑭𝑘
𝑨 = 𝑲𝑘−1 , 𝒂= , 𝒃 = 𝒚, (37.50)
𝒚𝑇 𝛥𝒙 𝑘−1

in the Sherman-Morrison formula. Therefore,


−1
⎡ [𝑲𝑘−1 𝑭𝑘 ]𝒚𝑇 ⎤ −1
𝑲𝑘−1
= ⎢𝑰 + 𝑇 𝛥𝑭 ⎥ 𝑲𝑘−1 . (37.51)
⎣ 𝒚 𝑘 ⎦
Broyden´s method use 𝒚 = ∆𝒙𝑘−1 resulting in non-symmetric stiffness updates, and an
algorithm for this exploits that
𝒅 ∆𝒙 𝑇 𝒅 ∆𝒙 𝑇 𝒅 ∆𝒙 𝑇
𝑲𝑘−1 = [𝑰 + 𝑘 𝑘−1 ] [𝑰 + 𝑘−1 𝑘−2 ] … [𝑰 + 1 0 ] 𝑲0−1 (37.52)
𝛾𝑘 𝛾𝑘−1 𝛾1
where
−1 𝑇 (37.53)
𝒅𝑘 = 𝑲𝑘−1 𝑭𝑘 , 𝛾𝑘 = ∆𝒙𝑘−1 ∆𝑭𝑘 .
To illustrate the efficiency of a quasi-Newton method we outline the steps to obtain ∆𝒙𝑘 .
1.Assume ∆𝒙0 , 𝑭𝑘 and 𝑭𝑘−1 (recalling (37.43)) are known, as well as ∆𝒙𝑖 , 𝒅𝑖 and 𝛾𝑖 for 𝑖 =
1, … , 𝑘 − 1.
2.Solve 𝒆0 = 𝑲0−1 𝑭𝑘 .
𝑇
∆𝒙𝑖−1 𝒆𝑖−1
3.Recursively compute 𝒆𝑖 = 𝒆𝑖−1 + 𝛾𝒊 𝒅𝑖 for 𝑖 = 1, … , 𝑘 − 1.
𝑇
4.Set 𝒅𝑘 = 𝒆𝑘−1 and 𝛾𝑘 = ∆𝒙𝑘−1 ∆𝑭𝑘 .
𝑇
∆𝒙𝑘−1 𝒅𝑘
5.Compute ∆𝒙𝑘 = [1 + 𝛾 𝒌 ] 𝒅𝑘 .
Noteworthy here is that an iteration consists of a forward and backward substitution of an
already factorized matrix (step 2) plus a sequence of vector operations (steps 3-5), which
altogether presumably is much less expensive than solving an entire system of linear
equations. The algorithm requires an in-core storage consisting of 2𝑘 vectors and 𝑘 scalars
to complete iteration 𝑘.

Again, recalling the quasi-Newton equation (37.42), and substituting (37.45) gives
𝑲𝑘−1 𝛥𝒙𝑘−1 + 𝛼𝒛𝒚𝑇 𝛥𝒙𝑘−1 + 𝛽𝒗𝒖𝑇 𝛥𝒙𝑘−1 = 𝛥𝑭𝑘 , (37.54)
1 1
and set 𝛼 = 𝒚𝑇 𝛥𝒙 , 𝒛 = −𝑭𝑘−1 , 𝛽 = 𝒖𝑇 𝛥𝒙 and 𝒗 = 𝛥𝑭𝑘 . Here 𝒚 and 𝒖 are arbitrary vectors
𝑘−1 𝑘−1
that are non-orthogonal to 𝛥𝒙𝑘−1 , i.e.,

35-10 (Implicit) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Implicit

𝒚𝑇 𝛥𝒙𝑘−1 ≠ 0, (37.55)
and
𝒖𝑇 𝛥𝒙𝑘−1 ≠ 0. (37.56)
In the BFGS method
𝒚 = 𝑭𝑘−1 𝒖 = 𝛥𝑭𝑘 , (37.57)
which leads to the following update formula
𝛥𝑭𝑘 𝛥𝑭𝑘𝑇 𝑇
𝑭𝑘−1 𝑭𝑘−1
𝑲𝑘 = 𝑲𝑘−1 + 𝑇
− 𝑇
, (37.58)
𝛥𝒙𝑘−1 𝛥𝑭𝑘 𝛥𝒙𝑘−1 𝑭𝑘−1
that preserves symmetry of the secant matrix. A double application of the Sherman-
Morrison formula then leads to the inverse form.

Special product forms have been derived for the DFP and BFGS updates and exploited by
Matthies and Strang [1979],
𝑲𝑘−1 = [𝑰 + 𝒒𝑘 𝒑𝑇𝑘 ] 𝑲𝑘−1
−1
[𝑰 + 𝒑𝑘 𝒒𝑘𝑇 ]. (37.59)
The primary advantage of the product form is that the determinant of 𝑲𝑘 and therefore, the
change in condition number can be easily computed to control updates. The updates
vectors are defined as
𝑇
𝛥𝒙𝑘−1 𝛥𝑭𝑘
𝒑𝑘 = −𝛥𝑭𝑘 − 𝑭𝑘−1 √ , (37.60)
𝑇
𝛥𝒙𝑘−1 𝑭𝑘−1

𝛥𝒙𝑘−1
𝒒𝑘 = 𝑇
. (37.61)
𝛥𝒙𝑘−1 𝛥𝑭𝑘
Noting that the determinant of 𝑲𝑘 is given by
2
det(𝑲𝑘 ) = det(𝑲𝑘−1 )[1 + 𝒒𝑘𝑇 𝒑𝑘 ] , (37.62)
it can be shown that the change in condition number, 𝑐, is
4
[√𝒑𝑇𝑘 𝒑𝑘 𝒒𝑘𝑇 𝒒𝑘 + √𝒑𝑇𝑘 𝒑𝑘 𝒒𝑘𝑇 𝒒𝑘 + 4 (1 + 𝒑𝑇𝑘 𝒒𝑘 )]
𝑐= . (37.63)
2
[4 (1 + 𝒑𝑇𝑘 𝒒𝑘 )]
Following the approach of Matthies and Strang [1979] this condition number is used to
decide whether or not to do a given update. The quasi-Newton condition (37.42) is easily
verified using (37.59) for a real non-singular tangent matrix 𝑲𝑘−1 and it follows that BFGS
preserves not only symmetry but also positive definiteness. Interestingly enough the
converse is not true; if 𝑲𝑘−1 is indefinite then 𝑲𝑘 may still be positive definite.

An algorithm would utilize that


𝑲𝑘−1 = [𝑰 + 𝒒𝑘 𝒑𝑇𝑘 ] … [𝑰 + 𝒒1 𝒑𝑇1 ]𝑲0−1 [𝑰 + 𝒑1 𝒒1𝑇 ] … [𝑰 + 𝒑𝑘 𝒒𝑘𝑇 ], (37.64)
LS-DYNA DEV 10/27/16 (r:8004) 35-11 (Implicit)
Implicit LS-DYNA Theory Manual

so the computation of ∆𝒙𝑘 = 𝑲𝑘−1 𝑭𝑘 requires a sequence of vector operations, followed by a


forward and backward substitution of a factorized matrix, followed by yet another
sequence of vector operations. The BFGS algorithm requires an in-core storage of the
vectors 𝒑𝑖 and 𝒒𝑖 , 𝑖 = 1, … , 𝑘 − 1, and ∆𝒙𝑘−1 and 𝑭𝑘−1 to complete iteration 𝑘. It is the default
and most robust quasi-Newton algorithm for the nonlinear implicit solver and there is no
reason to switch. The only practical issue is how to control when to reform the tangent
stiffness matrix, which typically depends on the degree of nonlinearity for the problem at
hand. In fact, if LCPACK = 2 on CONTROL_IMPLICIT_SOLVER, i.e., working with
independent variables only, BFGS is mandatory and comes in two flavors, NSOLVR = 2 or
NSOLVR = 12 on CONTROL_IMPLICIT_SOLUTION.

There is also the option LCPACK = 3, for which LS-DYNA uses a non-symmetric matrix
assembly when solving the linear system of equations. This adds to the computational cost
for each iteration but my improve convergence because of better tangents of certain
features. The non-symmetric solver is illustrated in Figure 3735-3 for simulating a clamped
cantilever beam subject to a follower load. In general the stiffness matrix is symmetric
when the force is derived from an energy potential, or in other words is conservative. This is
for instance the case for a hyperelastic material response or a physically admissible
pressure load, see Schweizerhof and Ramm [1984] for a discussion. Non-symmetry arise
e.g. from non-conservative forces, such as frictional contact, or physically inadmissible
design dependent loads. The example in Figure 3735-3 serves as one of the latter since the
load is not coming from a physical source, such as a water pressure, and it should be seen
as an academic example to prove a point.

Follower load 𝐹

𝐹 = 1/16 𝐹 = 1/4 𝐹 = 1/2 𝐹=1

Figure 3735-3 Nonlinear implicit solution of an elastic cantilever beam with a follower
force, von Mises stress is fringed..

35-12 (Implicit) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Implicit

37.3.4 Tangent stiffness reformations


In the previous section we have used 𝑲0 in (37.52) and (37.64) to represent the last
assembled tangent stiffness matrix according to (37.26). While the quasi-Newton updates
𝑲𝑘 , 𝑘 = 1,2, … are justified in the context of moderate nonlinearities and/or for a few
iterations, it will eventually break down convergence for more complex problems. Whence
there are criteria for reforming the tangent stiffness matrix and start over with the quasi-
Newton (BFGS) updates.

The first criterion is simply a user defined upper limit of 𝑘, governed by ILIMIT on
CONTROL_IMPLICIT_SOLUTION. Using ILIMIT = 1 means that no quasi-Newton
updates are performed and should be used for highly nonlinear problems, larger values of
ILIMIT can be used if the nonlinearities are considered less severe. ILIMIT = 11 is the
default and is a reasonable value to use for starters.

A second criterion is that of increased residual norm, i.e., if any of the quantities
𝑇𝑘 = √𝑭𝑘𝑇 𝑱𝑡 𝑭𝑘 (37.65)
or
𝑅𝑘 = √𝑭𝑘𝑇 𝑱𝑟 𝑭𝑘 (37.66)
for some 𝑘 is their respective largest attained since start iterating, the stiffness matrix is
reformed. Here 𝑱𝑡 and 𝑱𝑟 are diagonal matrices with ones or zeros on the diagonal that
extract the translational and rotational degrees of freedom, respectively, and one speaks of
translational or rotational divergence.

A third criterion is called energy explosion, if


𝑇 𝑇
log(1 + ∣∆𝒙𝑘−1 𝑱𝑡 𝑭𝑘−1 ∣) + log(1 + ∣∆𝒙𝑘−1 𝑱𝑡 𝑭𝑘 ∣) > 36 (37.67)
or
𝑇 𝑇
log(1 + ∣∆𝒙𝑘−1 𝑱𝑟 𝑭𝑘−1 ∣) + log(1 + ∣∆𝒙𝑘−1 𝑱𝑟 𝑭𝑘 ∣) > 36 (37.68)
the stiffness matrix is reformed. One can speak also here of translational and rotational
energy explosion and this simply indicates that something really bad has happened with
the last search direction.

Finally, a necessary criterion for continuing with BFGS updates is deemed


∆𝒙𝑘𝑇 𝑭𝑘 > 0 (37.69)
since this would otherwise indicate a search direction corresponding to a zero or negative
eigenvalue in the tangent matrix. If this criterion is violated the tangent stiffness matrix is
reformed with a warning message issued that negative energy is detected during quasi-
Newton updates. This can only happen if there are negative eigenvalues in the tangent
stiffness matrix in the first place.

37.3.5 Line search


Line search is critical for decent convergence characteristics, and there are several criteria to
choose from, see LSMTD and LSTOL on CONTROL_IMPLICIT_SOLUTION. One of the

LS-DYNA DEV 10/27/16 (r:8004) 35-13 (Implicit)


Implicit LS-DYNA Theory Manual

criteria available is that the norm of residual 𝑭 must somehow decrease, LSMTD = 2, but
this will inevitably lead to ridiculously small steps and is not recommended or presented
further here. A more useful criterion is instead that an iterate 𝑘 + 1 is accepted if
∣∆𝒙𝑘𝑇 𝑭𝑘+1 ∣ ≤ 𝜀𝑠 ∆𝒙𝑘𝑇 𝑭𝑘 (37.70)
where 𝜀𝑠 is the line search convergence tolerance LSTOL. See Figure 3735-4 for an
illustration. This criterion is derived from a hypothetical assumption of the existence of an
energy potential 𝑊(𝒙𝑘 + 𝑠∆𝒙𝑘 ) for the residual force 𝑭 and that the potential is minimized
along the search direction
𝜕𝑊 𝜕𝑊 𝜕𝒙
= = 𝑭 𝑇 ∆𝒙𝑘 = 0. (37.71)
𝜕𝑠 𝜕𝒙 𝜕𝑠
𝑭 𝑇 ∆𝒙𝑘 (1.60) not solvable
Acceptable interval
according to (37.70)
𝑭𝑘𝑇 ∆𝒙𝑘
𝜀𝑠 𝑭𝑘𝑇 ∆𝒙𝑘

𝑠=1
Negative starting values
indicate negative
eigenvalues in tangent

Figure 3735-4 Illustration of line search based on energy, may be


complemented with norm of 𝑭. Dashed line indicates a line search that
So in a sense, (37.70) is the solution of (37.71) to within a certain tolerance and a Ridder
algorithm is used to narrowing in on a candidate. It is not unusual however that (37.70) is
not well-posed, an indefinite stiffness matrix may result in a negative right hand side or the
search direction may be bad enough to not render a solution, see Figure 3735-4. In those
special cases some conservative approach is taken to provide a reasonable iterate 𝑘 + 1.
This summarizes in brief LSMTD = 4.

A closer examination (and numerical experiments) reveals that 𝑭𝑘+1 is not necessarily
bounded by (37.70). Therefore this criterion can be complemented with
𝑇
√𝑭𝑘+1 𝑭𝑘+1 ≤ [1 + 𝜀𝑠 ]√𝑭𝑘𝑇 𝑭𝑘 . (37.72)
This option is invoked with LSMTD = 5 and means that both (37.70) and (37.72) are to be
satisfied for accepting an iterate 𝑘 + 1. While this is a more robust approach, it also
requires more residual force evaluations and is not recommended as default. It has proved
to work well for implicit rubber simulations and complicated contact problems.

35-14 (Implicit) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Implicit

37.3.6 Convergence check


Convergence criteria are based on displacement, energy and residual force quantities. The
exact definitions of norms and scalar products used depend to a great extent on parameter
settings on CONTROL_IMPLICIT_SOLUTION, in particular on the parameter NLNORM.
Historically it has been assumed that rotational degrees of freedom are not appropriate for
checking convergence and therefore only translational degrees of freedom have been
considered. With advancements in the development of the implicit solver, this is today
considered an old fashioned way of thinking, but backward compatibility is maintained
and the user is referred to the keyword manual for information regarding the choices in
this context. Here the principles behind convergence checks are presented from a
pragmatic standpoint assuming the following generic displacement and force norms and
(energy) scalar products
‖∆𝒙‖ = √∆𝒙𝑇 𝑱𝑱∆𝒙, ‖𝑭‖ = √𝑭 𝑇 𝑱 † 𝑱 † 𝑭, 〈∆𝒙, 𝑭〉 = ∆𝒙 𝑇 𝑱𝑱 † 𝑭. (37.73)
Here 𝑱 is a diagonal matrix that appropriately scales the various degrees of freedom to
account for units, i.e., scaling radians with some characteristic length for consistency. It
also accounts for user input in the sense that 𝑱 may contain zeros on the diagonal to
indicate that rotational degrees of freedom should not be accounted for. 𝑱 † is the pseudo-
inverse of 𝑱, a notation introduced since 𝑱 actually may be singular due to the zeros on the
diagonal. For some convergence option check is performed on translational and rotational
degrees of freedom separately, we don’t elaborate on such details here but instead refer to
the keyword manual.

Convergence for accepting an iterate 𝑗 + 1 is assumed if the three conditions


‖𝛥𝒙𝑘 ‖ < max(𝜀𝑑 𝑢max , √max(𝜀𝑎 , 0)) (37.74)

〈∆𝒙𝑘 , 𝑭𝑘 〉 < max(𝜀𝑒 𝑒0 , 10000max(𝜀𝑎 , 0)) (37.75)

‖𝑭𝑘 ‖ < max(𝜀𝑟 𝑓0 , 10000max(𝜀𝑎 , 0)) (37.76)


are satisfied simultaneously. Here 𝜀𝑑 , 𝜀𝑒 , 𝜀𝑟 and 𝜀𝑎 are the displacement, energy, residual
and absolute tolerances corresponding to DCTOL, ECTOL, RCTOL and ABSTOL,
respectively. In the right hand side of (37.74), 𝑢max = ∥𝒖max ∥ is the maximum attained
displacement in any iteration 𝑘 measured from the position at the start of this implicit step𝑗
if DNORM = 1 and from the start of the simulation if DNORM = 2. In the right hand side
of (37.75), 𝑒0 = 〈∆𝒙0 , 𝑭0 〉 where ∆𝒙0 and 𝑭0 are the first incremental displacement and
residual vectors for this implicit step 𝑗. Likewise 𝑓0 = ∥𝑭0 ∥ in the right hand side of (37.76) is
the norm of the first residual vector for this implicit step 𝑗. If 𝑱 has full rank and the tangent
stiffness matrix is symmetric and positive definite then these conditions make intuitive
sense, small changes in displacements and/or small residual forces indicate that an iterate
is “close-enough” to the solution. Under these conditions the left hand side of (37.75)
define norms in both the incremental displacement ∆𝒙𝑘 and residual force 𝑭𝑘 , and 𝑒0 > 0.

LS-DYNA DEV 10/27/16 (r:8004) 35-15 (Implicit)


Implicit LS-DYNA Theory Manual

A backdoor out is an absolute convergence test on the maximum norm on translational


displacement. That is, convergence is always detected if the maximum value of any
translational degree of freedom in the incremental displacement array ∆𝒙𝑘 is smaller than
𝑑√max(𝜀𝑎 , 0), where 𝑑 is a characteristic size calculated as the diagonal of the smallest box
aligned along the global coordinate system that encapsulates the model. Setting ABSTOL
to a negative number is an alternative, then all absolute convergence checks above become
inactive and instead the backdoor out is to assume convergence when ‖𝑭𝑘 ‖ < max(0, −𝜀𝑎 ).
This latter option requires some a priori knowledge of the force level and may require
monitoring the residual norm for a few trial iterations to pick a decent 𝜀𝑎 , but it could be a
sensible choice if the problem shows erratic behavior in displacement and energy due to
severe nonlinearities.

Another criterion that must be met for convergence is that simple prescribed motion
constraints on nodes and rigid bodies, if they exist, must be “almost” satisfied. The
background is that only partial line searches (𝑠 < 1) will not satisfy simple motion
constraints, and at least one full line search step (𝑠 = 1) must be accomplished during the
iterations. For difficult problems, this sometimes never happens and therefore convergence
is prevented due to unfulfilled boundary conditions until all prescribed motion is satisfied
to within 1%.

37.3.7 Automatic time stepping


If convergence is not attained after reforming the stiffness matrix a given amount of times,
see MAXREF on CONTROL_IMPLICIT_SOLUTION, one of two things will happen. By
default, LS-DYNA terminates with an error message, but if automatic time stepping is
turned on, see IAUTO on CONTROL_IMPLICIT_AUTO, then LS-DYNA will try to resolve
the problem with a smaller step size. More specifically, if convergence is not attained with
a time step ∆𝑡old , then LS-DYNA backs up to beginning of step and retries with a time step
∆𝑡new = max(∆𝑡min , 10−0.3 ∆𝑡old ) (37.77)
where ∆𝑡min is a user defined minimum step. If ∆𝑡old = ∆𝑡min when attempting this then
LS-DYNA will terminate with an error.

With the automatic time stepper turned on, LS-DYNA will not only cut the time step for
convergence failure but also adjust the time step when converging. The adjustment is
based on a user defined iteration window, i.e., a range of iteration numbers that are
deemed acceptable for convergence, and if the number of iterations for convergence falls
outside this window the next time step will be adjusted as follows. If convergence is
attained for more iterations than acceptable then the time step for the next implicit step is
given by (37.77). If instead convergence is attained for fewer iterations then the next time
step is given by
∆𝑡new = min(∆𝑡max , 100.2 ∆𝑡old ) (37.78)
where ∆𝑡max is a maximum defined step. In this way LS-DYNA narrows in on an optimum
time step for which the number of iterations to converge falls within the specified window.

35-16 (Implicit) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Implicit

There is much more information regarding this option in the keyword manual and the user
is referred thereto for practical issues.

37.4 Implicit Dynamics


37.4.1 Newmark time integration
Nonlinear implicit dynamics principally follows the exact same solution algorithm as for
nonlinear implicit statics, the only difference is that dynamic terms are included in the
residual force. That is, (37.1) now reads
𝑹 = 𝑴𝒙̈ + 𝑫𝒙̇ + 𝑭𝑖 − 𝑭𝑒 = 𝟎, (37.79)
where global damping is introduced through the matrix 𝑫, so the residual is now
essentially a function of 𝒙, 𝒙̇ and 𝒙̈. The dependence on the latter two is eliminated by
introducing the Newmark time integration scheme
𝛥𝒙 𝒙̇𝑗 1 1
𝒙̈ = 2
− − ( − 𝛽) 𝒙̈𝑗 , (37.80)
𝛽𝛥𝑡 𝛽𝛥𝑡 𝛽 2

𝒙̇ = 𝒙̇𝑗 + 𝛥𝑡(1 − 𝛾)𝒙̈ 𝑗 + 𝛾𝛥𝑡𝒙̈, (37.81)

𝒙 = 𝒙 𝑗 + 𝛥𝒙. (37.82)
Here, Δ𝑡 is the time step size, 𝛽 and 𝛾 are the free parameters of integration and we have
used ∆𝒙 to denote the total displacement from step 𝑗 to step 𝑗 + 1. For 𝛾 = 1 ⁄ 2 and 𝛽 = 1 ⁄
4 the method becomes the trapezoidal rule and is energy conserving. If
1
𝛾> , (37.83)
2
2
1 1
𝛽 > ( + 𝛾) , (37.84)
4 2
numerical damping is induced into the solution leading to a loss of energy and momentum.
By inserting (37.80) and (37.81) into (37.79) and using (37.82) to eliminate ∆𝒙, we are back to
an equation in 𝒙 only and can apply the algorithm starting with (37.18) and everything
thereafter holds.

37.4.2 Practical considerations


Linearizing (37.79) using (37.80), (37.81) and (37.82) results in a tangent matrix
𝜕𝑹 𝑴 𝛾𝑫 𝜕[𝑭𝑖 − 𝑭𝑒 ]
= + + , (37.85)
𝜕𝒙 𝛽∆𝑡2 𝛽∆𝑡 𝜕𝒙
and leads to an interesting observation. Since 𝑴 is symmetric and positive definite the first
term assures that the resulting tangent is positive definite as long as the time step ∆𝑡 is
sufficiently small, so including dynamics will enhance the robustness of the numerical

LS-DYNA DEV 10/27/16 (r:8004) 35-17 (Implicit)


Implicit LS-DYNA Theory Manual

procedure. In fact, the eigenvalues of the resulting tangent can be made arbitrarily large by
decreasing ∆𝑡 at the cost of requiring more steps to obtain the solution.

Not only robustness but also the accuracy in dynamic implicit depends on the size of the
time step ∆𝑡, roughly speaking only frequencies up to ~∆𝑡−1 can be resolved. The method
is therefore not perfectly suitable for contact-impact and restitution problems, for contact a
crude rule of thumb is that the time a node or body is in contact should span at least a few
time steps to appropriately resolve the resulting impulse. Having said this, these problems
are still difficult to solve reasonably well.

For implicit static contact problems, a common usage of dynamics is to temporarily


suppress rigid body modes while a desired contact state is being established. In such a case
it is recommended to use numerical damping by for instance choosing 𝛾 = 0.6 and𝛽 = 0.38
to render a smooth response. This technique has proved successful and it is often sufficient
to have dynamics initially turned on and then turned off at a time when all rigid body
modes have been eliminated by contact. But it is also possible to arbitrarily switch
dynamics on and off during a simulation, all this is explained on CON-
TROL_IMPLICIT_DYNAMICS in the keyword manual.

37.4.3 Linear theory


Linearization of (37.79) with respect to a configuration 𝒙 that is in static equilibrium, i.e.,
𝑷[𝑭𝑖 − 𝑭𝑒 ] = 𝟎, can be written as
𝑴𝒖̈ + 𝑫𝒖̇ + 𝑲𝒖 = 𝑭, (37.86)
where we use 𝒖 = ∆𝒙 to denote the displacement measured from the reference point 𝒙. For
the sake of clarity we also abused some notation, using 𝑴 and 𝑫 to actually mean 𝑷𝑴𝑷 𝑇
and 𝑷𝑫𝑷 𝑇 , respectively. Furthermore, 𝑲 denotes the sum of the material (37.29) and
geometric (37.39) contributions to the static tangent stiffness matrix as well as any
contributions from the external force, e.g., contacts,
𝜕𝑭
𝑲 = 𝑷 ∫[𝑩𝑇 𝑬𝑩 + 𝓑 𝑇 𝝈𝓑]𝑑𝑉 𝑷 𝑇 − 𝑷 𝑒 𝑷 𝑇 . (37.87)
𝜕𝒙
𝑉
The right hand side 𝑭 in (37.86) should be interpreted as an external (small) load that only
depends on time 𝑡 and serves to dynamically perturb the static equilibrium. This equation
can be discretized in time using the Newmark time integration scheme above and then
efficiently integrated with no update of the tangent stiffness matrix between steps. This of
course assumes that the displacements 𝒖 are small enough to not affect 𝑲 to great extent.

Consider undamped free vibration in (37.86), i.e, 𝑫 = 𝟎 and 𝑭 = 𝟎, and transform this
equation from time to frequency plane assuming constant 𝑴 and 𝑲. This results in an
eigenvalue problem
𝑲𝒖 = 𝜔2 𝑴𝒖 (37.88)
that can be solved in LS-DYNA for the angular frequency 𝜔 and mode shape 𝒖. The stress
𝝈 in (37.87) affects the frequency 𝜔 in the following way. If the model is in a tensile state,
i.e., the principal stresses are positive, then the eigenvalues of 𝑲 increase compared to a

35-18 (Implicit) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Implicit

stress free state and from (37.88) the frequencies will increase. Conversely, the frequencies
will decrease if the principal stresses decrease, this is the effect of tuning a guitar string by
increasing or decreasing its tension. If linearizing with respect to a contact state, the last
term in (37.87) will in effect constrain relative motion between parts or nodes in contact.
More specifically, the relative normal displacement in 𝒖 will be zero, and the relative
tangential motion will be governed by the present stick/slip condition. If in stick mode the
relative tangential displacement in 𝒖 will be zero, and if in slip mode it is unconstrained.
See CONTROL_IMPLICIT_EIGENVALUE for the available options to solve (37.88).

An example using modal analysis in this respect is shown in Figure 3735-5.

LS-DYNA DEV 10/27/16 (r:8004) 35-19 (Implicit)


Implicit LS-DYNA Theory Manual

𝑓 = 15 𝐻𝑧

𝑓 40 𝐻 𝑓 43 𝐻
Figure 3735-5 Intermittent eigenvalue analysis of tire. Model shown in top
left, followed by lowest frequency modes for unpressurized tire (top right),
inflated tire (bottom left) and inflated tire and frictional contact (bottom
right). Resultant mode displacements are fringed.

35-20 (Implicit) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Implicit

In general 𝑫 ≠ 𝟎 and 𝑫 and/or 𝑲 may be non-symmetric as discussed in Section 37.3.3, but


we assume 𝑴 is symmetric and positive definite and still 𝑭 = 𝟎. The characteristic
equation approach for solving (37.86) makes use of the harmonic ansatz
𝒖(𝑡) = ∑ exp(𝜇𝑗 𝑡)𝚽𝑗 (37.89)
𝑗
which is inserted into (37.86) to yield
∑ exp(𝜇𝑗 𝑡){𝜇2𝑗 𝑴 + 𝜇𝑗 𝑫 + 𝑲}𝚽𝑗 = 𝟎 (37.90)
𝑗
to which the quadratic eigenvalue problem is associated7
{𝜇2 𝑴 + 𝜇𝑫 + 𝑲}𝚽 = 𝟎. (37.91)
Because of the non-symmetry of the involved matrices, the eigenvalues to (37.91) may be
complex but come in conjugate pairs. That is, if 𝜇 = 𝑟 + 𝑖𝑠 is an eigenvalue then 𝜇̅̅̅ = 𝑟 − 𝑖𝑠 is
also an eigenvalue, and if 𝚽 = 𝚼 + 𝑖𝚿 is the eigenvector associated with 𝜇 then𝚽 ̅̅̅̅̅̅ = 𝚼 − 𝑖𝚿
is the eigenvector for 𝜇̅̅̅. However, examining the sum of two such terms in (37.90) yields
exp(𝜇𝑡){𝜇2 𝑴 + 𝜇𝑫 + 𝑲}𝚽 + exp(𝜇̅̅̅𝑡){𝜇̅̅̅2 𝑴 + 𝜇̅̅̅𝑫 + 𝑲}𝚽̅̅̅̅̅̅
= 2exp(𝑟𝑡){((𝑟2 − 𝑠2 )𝑴 + 𝑟𝑫 + 𝑲)(cos(𝑠𝑡) 𝚼 − 𝑠𝑖𝑛(𝑠𝑡)𝚿) (37.92)
− 𝑠(2𝑟𝑴 + 𝑫)(sin(𝑠𝑡) 𝚼 + cos(𝑠𝑡)𝚿)},
meaning that the solution in time domain is real as expected. If all eigenvalues are purely
complex, i.e., 𝜇𝑗 = 𝑖𝑠𝑗 and the real parts 𝑟𝑗 = 0 vanish, then the solution to (37.86) is
harmonic with angular frequencies 𝜔𝑗 = ∣𝑠𝑗 ∣. This is apparent from (37.92) and corresponds
to the traditional solution to (37.88), with the exception of a non-zero damping matrix 𝑫.
But if an eigenvalue has a real part 𝑟𝑗 > 0 then from (37.92) the solution is exponentially
growing and thus unstable. Solving (37.91) can therefore be used to detect instabilities in a
system, eigenvalues with positive real parts correspond to unstable modes. A common
indicator used for this is the damp ratio defined as
𝑟
𝜗 = −2 , (37.93)
|𝑠|
so basically a negative damp ratio indicates an unstable mode. An application of this is
brake squeal, see Figure 3735-6, for which friction instabilities may result in unwanted

noise and erratic behavior. Noticable is that if 𝑫 = 𝟎 and 𝜇 is an eigenvalue then – 𝜇 is also

an eigenvalue, so then complex eigenvalues come in clusters of four, {𝜇, 𝜇̅̅̅, −𝜇, −𝜇̅̅̅}. Then it
suffice that eigenvalues have nonzero real parts 𝑟𝑗 ≠ 0 for a system to be unstable, which
corresponds to negative eigenvalues in (37.88).
To solve the system (37.91), it is transformed to a first order eigenvalue problem using𝚽 ̃=
𝜇𝚽, resulting in
[ 𝟎 𝑰 ][𝚽] = 𝜇[ 𝑰 𝟎 ][𝚽], (37.94)
−𝑲 −𝑫 𝚽 ̃ 𝟎 𝑴 𝚽 ̃

7 The eigenvalue problem (1.91) is readily obtained by Laplace transforming (1.86).


34.21
Implicit LS-DYNA Theory Manual

which can be solved by well-established eigenvalue algorithms. In the output files, LS-
DYNA reports eigenvalues with positive imaginary parts only, i.e., 𝑠𝑗 > 0, and the real and
imaginary parts (when non-zero) of the associated eigenvector 𝚽.

0.02

0.015

0.01

0.005
Damp ratio

-0.005

-0.01

-0.015

-0.02
0 2 4 6 8 10 12 14 16
Mode

Figure 3735-6 Brake squeal application. Pressure pads are applied to a rotating
disc and a nonsymmetric eigenvalue solution reveals friction instabilities. The
damp ratio 𝜗 is plotted at the top as function of mode number, see (37.93). Mode
#6 is the unstable mode and is depicted bottom right with displacements fringed.

34.22
LS-DYNA Theory Manual Arc-length

38
Arc-length

Arc-length methods are available in LS-DYNA for NSOLVR specified between 6 and 9, this
and all other parameters in this Section are located on the *CONTROL_IMPLICIT_SOLU-
TION keyword. These solvers use the Riks/Crisfield methods but unfortunately go under
the old LSDIR.EQ.1 option which makes them somewhat limited in terms of applicability.
For LSDIR.EQ.2 the arc-length method described in Ritto-Corrêa and Camotim [2008] is
implemented for the combination of NSOLVR.EQ.12 and ARCMTH.EQ.3. For this method
the parameters ARCPSI (𝜓), ARCALF (𝛼) and ARCTIM apply, out of which the last simply
tells at what time arc-length is initiated and the first two are to be described in more detail
below. We begin by an explanatory overview of the arc-length method in general for
which we will constantly be referring to Figure 38.7 below. After that the mathematical
details are revealed.

38.1 Overview
An implicit static problem is driven by a parameter 𝑡 referred to as the time, and assuming
that a solution is obtained at 𝑡 = 𝑡𝑛 the objective is to determine the solution given the
constraint 𝑡 = 𝑡𝑛+1 , where 𝑡𝑛+1 is given. We assume that the problem can be associated
with representative force and displacement parameters 𝑓 and 𝑑 and we distinguish between
load- and displacement-driven problems. An example of a load-driven problem is when
𝑓 = 𝑓 (𝑡) is an external load and 𝑑 is the resulting displacement of the associated nodes, and
likewise a displacement-driven problem is when 𝑑 = 𝑑(𝑡) is a prescribed displacement and
𝑓 is the corresponding reaction force. A solution is likely obtained if the problem is stable,
i.e., the force-displacement curve is monotonically increasing, but this method is not
designed to handle limit or turning points. A limit point is illustrated in figure (a) and a
turning point in figure (b), the solution in the next step may not be the one desired or not
even exist, in both cases we want to find the solution that continuously follow the path
corresponding to the force-displacement curve.

LS-DYNA DEV 10/27/16 (r:8004) 35-23 (Arc-length)


Arc-length LS-DYNA Theory Manual

The solution for this is to set the stepping parameter 𝑡 free, i.e., replace the constraint 𝑡 =
𝑡𝑛+1 with an arc-length constraint 𝑔(𝑑, 𝑓 ) = 0. In words, this basically means that a
multidimensional sphere (arc) is put around the last converged solution and the next
solution is to be found on that given sphere, the stepping parameter 𝑡 is now a solution
variable. This makes the problem well-posed but unfortunately there are multiple
solutions to the problem, and it may turn out that the wrong solution is found. In figures
(c) and (d), the effect of the arc-length constraints is illustrated and there are two possible
solutions, one feasible that allows us to continue in the right direction and one infeasible
that takes us in the wrong direction. The latter phenomenon is termed doubling back, and
is in practice not easily avoided. Two additional parameters are available that have shown
to improve the robustness in this respect.

35-24 (Arc-length) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Arc-length

a) f b) f

d d

c) f d) f

d d

e) f f) f

d d

g) f h) f

d d

Converged solution at step n


Infeasible solution at step n + 1
Feasible solution at step n + 1
Center of arc for α < 0, makes infeasible solution less probable
Figure 38.7. Representative force-displacement curves for illustrating arc-length
and accompanying parameters

In figure (d), two of the infeasible solutions can in practice be avoided by including the
stepping parameter in the arc-length constraint, thus converting a cylinder to a sphere in
space-time. This is adjusted by the paremeter 0 ≤ 𝜓 < 1, and the constraint reads
(1 − 𝜓)𝑔(𝑑, 𝑓 ) + 𝜓(𝑡 − 𝑡𝑛 )2 = 0, the effect of this constraint is illustrated in figures (e) and
(f), and should be compared with figures (c) and (d). Note that two infeasible solutions are
avoided when comparing figures (d) and (f), it may sometimes be worth using a non-zero
value for 𝜓, e.g., 𝜓 = 0.1.

LS-DYNA DEV 10/27/16 (r:8004) 35-25 (Arc-length)


Arc-length LS-DYNA Theory Manual

Another problem is that the feasible and infeasible solution may be too close to the last
converged solution, making the result from the simulation very unpredictable. For this a
parameter 𝛼 is introduced that translates the center of the spatial sphere in the direction of
the linear prediction (i.e., the first Newton iterate of the implicit solution procedure).
Assuming that this prediction is in the direction we want, using 𝛼 < 0 will move the center,
and consequently the infeasible solution, away from where the iterates are taking place. In
addition, the radius of the arc will increase making it less probable to find the incorrect
solution. This option has shown effective in solving snap-through problems when using
small steps to resolve maximum load values, and is illustrated in figures (g) and (h). For
snap-back problems, using 𝛼 = 1 could be an interesting choice since this centers the arc
right between the previously converged point and the first predictor in the arc length
method, thus encouraging the next solution to be found in the reversed direction. An
example of a snap-back problem is shown in Figure 35-9.

38.2 Nonlinear equations


The following, except for a change in notation, is very similar to the nonlinear theory
presented in the previous chapter. The generalization is that 𝑡 will here be treated as an
independent variable that is constrained by arc-length instead of given as a constant. The
nonlinear variables are denoted
𝒙
𝒙 = [𝒙 𝐼 ], (38.95)
𝐷

that we assume can be divided into a set of independent and dependent variables.
Furthermore we have the time parameter 𝑡 which may serve as the actual time (for
dynamic problems) or just a stepping parameter (for quasi-static problems). The division
into independent and dependent variables is motivated by the constraint equation that
must be fulfilled, i.e.,
𝒉(𝒙, 𝑡) = 𝒉(𝒙𝐼 , 𝒙𝐷 , 𝑡) = 𝟎. (38.96)
From the constraint, the constraint matrix is evaluated as
𝜕𝒉 𝜕𝒉 𝜕𝒉
=[ ] = [𝑪𝐷𝐼 𝑪𝐷𝐷 ], (38.97)
𝜕𝒙 𝜕𝒙𝐼 𝜕𝒙𝐷
which in turn determines the space of trial functions used to establish the nonlinear finite
element equation,
−1 𝑇 𝒓𝐼
[𝑰𝐼𝐼 −(𝑪𝐷𝐷 𝑪𝐷𝐼 ) ][𝒓𝐷 ] = 𝟎, (38.98)

where
𝒓
𝒓(𝒙, 𝑡) = 𝒓(𝒙𝐼 , 𝒙𝐷 , 𝑡) = [𝒓 𝐼 ] (38.99)
𝐷

is the full residual divided into the set of independent and dependent variables. See
previous chapter for further details leading up to (38.98).

35-26 (Arc-length) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Arc-length

38.3 Newton iterations


Here we assume that we are in a given configuration given by
𝒙 (𝑛,𝑖) = 𝒙 (𝑛) + ∆𝒙 (𝑛,𝑖) ,
(38.100)
𝑡(𝑛,𝑖) = 𝑡(𝑛) + ∆𝑡(𝑛,𝑖) .
where superscript 𝑛 = 0, 1, 2, … represents converged implicit states and 𝑖 = 0, 1, 2, …
represents non-converged Newton iterates. We implicitly assume that 𝒙 (0) = 𝒙̅, ∆𝒙 (𝑛,0) = 𝟎,
𝑡(0) = 0 and ∆𝑡(𝑛,0) = 0 are given. In this configuration LS-DYNA computes the full
residual given as
𝒓
𝒓 (𝑛,𝑖) = [𝒓 𝐼 ], (38.101)
𝐷

as well as its dependence on the time parameter


𝜕𝒓
𝜕𝒓
(𝑛,𝑖) ⎡ 𝐼⎤
( ) =⎢ 𝜕𝑡 ⎥, (38.102)

⎢𝜕𝒓 ⎥⎥
𝜕𝑡 𝐷
⎣ 𝜕𝑡 ⎦
together with the stiffness matrix given as
(𝑛,𝑖)
𝜕𝒓 𝑲𝐼𝐼 𝑲𝐼𝐷
( ) =[ ]. (38.103)
𝜕𝒙 𝑲𝐷𝐼 𝑲𝐷𝐷
Likewise the constraint residual, its dependence on the time parameter and constraint
matrix are evaluated and given by
𝒉(𝑛,𝑖) = 𝒉,
(𝑛,𝑖)
𝜕𝒉 𝜕𝒉
( ) = , (38.104)
𝜕𝑡 𝜕𝑡
(𝑛,𝑖)
𝜕𝒉
( ) = [𝑪𝐷𝐼 𝑪𝐷𝐷 ].
𝜕𝒙
The reduced residual and stiffness matrix are then formed as
𝑇 𝑲𝐼𝐼 𝑲𝐼𝐷 𝑰
𝑲̂ 𝐼𝐼 = [𝑰𝐼𝐼 −1
−(𝑪𝐷𝐷 𝑪𝐷𝐼 ) ] [𝑲 ] [ −1𝐼𝐼 ],
𝐷𝐼 𝑲𝐷𝐷 −𝑪𝐷𝐷 𝑪𝐷𝐼

−1 𝑇 𝒓𝐼 𝑲𝐼𝐷 −1
𝒓𝐼̂ = [𝑰𝐼𝐼 −(𝑪𝐷𝐷 𝑪𝐷𝐼 ) ] {[𝒓𝐷 ] − [𝑲 ] 𝑪𝐷𝐷 𝒉} ,
𝐷𝐷 (38.105)
⎧ 𝜕𝒓𝐼 ⎫
𝜕𝒓𝐼̂ {
{ ⎡ ⎤ 𝜕𝒉 }
}
= [𝑰𝐼𝐼 −1 𝑇
] ⎢ 𝜕𝑡 ⎥ − [ 𝑲𝐼𝐷 ] 𝑪 −1 ,
−(𝑪𝐷𝐷 𝑪𝐷𝐼 ) ⎨⎢ ⎥
𝜕𝑡 { ⎢𝜕𝒓𝐷 ⎥ 𝑲𝐷𝐷 𝐷𝐷
𝜕𝑡 ⎬
}
{ }
⎩⎣ 𝜕𝑡 ⎦ ⎭
and the independent search direction is given by

LS-DYNA DEV 10/27/16 (r:8004) 35-27 (Arc-length)


Arc-length LS-DYNA Theory Manual

𝜕𝒙𝐼𝑠
𝛿𝒙𝐼 = 𝑠𝛿𝒙𝐼𝑠 + 𝛿𝑡 . (38.106)
𝜕𝑡
Here 𝑠 is the line search parameter and
𝛿𝒙𝐼𝑠 = −𝑲̂ 𝐼𝐼−1 𝒓𝐼̂ ,
𝜕𝒙𝐼𝑠 𝜕𝒓 ̂ (38.107)
= −𝑲̂ 𝐼𝐼−1 𝐼 .
𝜕𝑡 𝜕𝑡
The full search direction is completed by computing the dependent part as
𝑠
𝑠 𝜕𝒙𝐷
𝛿𝒙𝐷 = 𝑠𝛿𝒙𝐷 + 𝛿𝑡 , (38.108)
𝜕𝑡
where
𝑠
𝛿𝒙𝐷 −1
= −𝑪𝐷𝐷 𝑪𝐷𝐼 𝛿𝒙𝐼𝑠 − 𝑪𝐷𝐷
−1
𝒉,
𝜕𝒙𝐷𝑠
𝜕𝒙 𝑠 (38.109)
−1 𝜕𝒉
= −𝑪𝐷𝐷−1
𝑪𝐷𝐼 𝐼 − 𝑪𝐷𝐷 .
𝜕𝑡 𝜕𝑡 𝜕𝑡
Finally the new configuration is updated by means of
∆𝒙𝐼(𝑛,𝑖+1) = ∆𝒙𝐼(𝑛,𝑖) + 𝛿𝒙𝐼 ,
(𝑛,𝑖+1) (𝑛,𝑖) (38.110)
∆𝒙𝐷 = ∆𝒙𝐷 + 𝛿𝒙𝐷 ,
∆𝑡(𝑛,𝑖+1) = ∆𝑡(𝑛,𝑖) + 𝛿𝑡.
Upon convergence we set
∆𝒙𝐼(𝑛) = ∆𝒙𝐼(𝑛,𝑖+1) ,
(𝑛) (𝑛,𝑖+1) (38.111)
∆𝒙𝐷 = ∆𝒙𝐷 ,
∆𝑡(𝑛) = ∆𝑡(𝑛,𝑖+1) ,
and hence
𝒙𝐼(𝑛+1) = 𝒙𝐼(𝑛) + ∆𝒙𝐼(𝑛) ,
(𝑛+1) (𝑛) (𝑛) (38.112)
𝒙𝐷 = 𝒙𝐷 + ∆𝒙𝐷 ,
𝑡(𝑛+1) = 𝑡(𝑛) + ∆𝑡(𝑛) .

35-28 (Arc-length) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Arc-length

Previously converged solution


Infeasible solution
Feasible solution
Predictore solution, found in the outward
Normal direction of previous arc
Infeasible predictor solution
Corrector steps occur along the arc
Figure 38.8. Predictor and corrector steps

38.4 Arc-length constraint – predictor step


For the predictor step, 𝑖 = 0, the following constraint is imposed
𝑇
(∆𝒙𝐼(𝑛,1) ) ∆𝒙𝐼(𝑛,1) ∆𝑡(𝑛,1) ∆𝑡(𝑛,1)
(1 − 𝜓) 𝑇
+𝜓 − 1 = 0, (38.113)
(∆𝒙𝐼(0,1) ) ∆𝒙𝐼(0,1) ∆𝑡(0,1) ∆𝑡(0,1)

where for 𝑛 = 0 we use


𝜕𝒙𝐼𝑠
∆𝒙𝐼(0,1) = ∆𝑡 ̅
𝜕𝑡 (38.114)
(0,1)
∆𝑡 = ∆𝑡.̅

LS-DYNA DEV 10/27/16 (r:8004) 35-29 (Arc-length)


Arc-length LS-DYNA Theory Manual

Referring to Figure 38.8 this constraint is geometrically interpreted as to find a predictor


solution on the arc with the previously converged solution as its center. Writing out the
above in terms of known quantities results in the following two possible values of the
increment in time step parameter
𝑇 −1/2
𝑠 𝑠
⎛𝜓 (𝜕𝒙𝐼 ) 𝜕𝒙𝐼 + 𝜓 ⎟
𝛿𝑡± = ± ⎜ ⎞ , (38.115)
𝑥 𝑡
⎝ 𝜕𝑡 𝜕𝑡 ⎠
corresponding to the two possible predictor solutions on the arc, where
1−𝜓
𝜓𝑥 = 𝑇
,
(∆𝒙𝐼(0,1) ) ∆𝒙𝐼(0,1)
(38.116)
𝜓
𝜓𝑡 = .
∆𝑡(0,1) ∆𝑡(0,1)
The actual value used is detemined from the sign of
𝜓𝑥 (𝑛−1) 𝛼 (𝑛−1,1) 𝑇 𝜕𝒙𝐼𝑠
2
(∆𝒙 𝐼 − ∆𝒙𝐼 ) + 𝜓𝑡 ∆𝑡(𝑛−1) , (38.117)
(1 − )𝛼 2 𝜕𝑡
2
if positive 𝛿𝑡 = 𝛿𝑡+ , otherwise 𝛿𝑡 = 𝛿𝑡− . This condition is to say that the solution
continues in the direction of the previously converged state, for the initial step 𝑛 = 0,
𝛿𝑡 = 𝛿𝑡+ . Again referring to Figure 38.8 we simply want to avoid going backwards to
the infeasible predictor solution.

38.5 Arc-length constraint – corrector steps


For the corrector steps, 𝑖 > 0, the following constraint is imposed
𝑇
(∆𝒙𝐼(𝑛,𝑖+1) − 𝛼 ∆𝒙𝐼(𝑛,1) ) (∆𝒙𝐼(𝑛,𝑖+1) − 𝛼 ∆𝒙𝐼(𝑛,1) ) ∆𝑡(𝑛,𝑖+1) ∆𝑡(𝑛,𝑖+1)
(1 − 𝜓) 2 2 + 𝜓 − 1 = 0, (38.118)
𝛼 2 (0,1) 𝑇 (0,1) ∆𝑡 (0,1) ∆𝑡(0,1)
(1 − ) (∆𝒙𝐼 ) ∆𝒙𝐼
2
which geometrically says that we should find the next iterate on the arc. Expanding, this
amounts to
𝑇
𝜕𝒙𝐼𝑠 𝛼 (𝑛,1) 𝜕𝒙𝐼𝑠 𝛼 (𝑛,1)
𝛼𝑥 (∆𝒙𝐼(𝑛,𝑖) + 𝑠𝛿𝒙𝐼𝑠 + 𝛿𝑡 (𝑛,𝑖)
− ∆𝒙𝐼 ) (∆𝒙𝐼 𝑠
+ 𝑠𝛿𝒙𝐼 + 𝛿𝑡 − ∆𝒙𝐼 ) +
𝜕𝑡 2 𝜕𝑡 2 (38.119)
𝛼𝑡 (∆𝑡(𝑛,𝑖) + 𝛿𝑡)(∆𝑡(𝑛,𝑖) + 𝛿𝑡) − 1 = 0,

35-30 (Arc-length) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Arc-length

where
1−𝜓
𝛼𝑥 = 2 𝑇
(1 − 𝛼) (∆𝒙𝐼(0,1) ) ∆𝒙𝐼(0,1)
2 (38.120)
𝜓
𝛼𝑡 = (0,1) (0,1) .
∆𝑡 ∆𝑡
This can be written in terms of a polynomial in 𝑠 and 𝛿𝑡 as
𝑎𝑠𝑠 𝑠2 + 𝑎𝑡𝑡 𝛿𝑡2 + 2𝑎𝑠𝑡 𝑠𝛿𝑡 + 2𝑎𝑠 𝑠 + 2𝑎𝑡 𝛿𝑡 (38.121)
where
𝑎𝑠𝑠 = 𝛼𝑥 (𝛿𝒙𝐼𝑠 )𝑇 𝛿𝒙𝐼𝑠
𝑇
𝜕𝒙 𝑠 𝜕𝒙𝐼𝑠
𝑎𝑡𝑡 = 𝛼𝑥 ( 𝐼 ) + 𝛼𝑡
𝜕𝑡 𝜕𝑡
𝜕𝒙𝐼𝑠
𝑎𝑠𝑡 = 𝛼𝑥 (𝛿𝒙𝐼𝑠 )𝑇
𝜕𝑡
(38.122)
𝜕𝒙 𝑠
𝑎𝑠𝑡 = 𝛼𝑥 (𝛿𝒙𝐼𝑠 )𝑇 𝐼
𝜕𝑡
𝛼 𝑇
𝑎𝑠 = 𝛼𝑥 (∆𝒙𝐼(𝑛,𝑖) − ∆𝒙𝐼(𝑛,1) ) 𝛿𝒙𝐼𝑠
2
𝛼 𝑇 𝜕𝒙 𝑠
𝐼
𝑎𝑡 = 𝛼𝑥 (∆𝒙𝐼(𝑛,𝑖) − ∆𝒙𝐼(𝑛,1) ) + 𝛼𝑡 ∆𝑡(𝑛,𝑖) .
2 𝜕𝑡
For a given line search parameter value, the time increment can have two possible values

−𝑎𝑠𝑡 𝑠 − 𝑎𝑡 ± √(𝑎2𝑠𝑡 − 𝑎𝑡𝑡 𝑎𝑠𝑠 )𝑠2 + 2(𝑎𝑠𝑡 𝑎𝑡 − 𝑎𝑠 𝑎𝑡𝑡 )𝑠 + 𝑎2𝑡


𝛿𝑡± = . (38.123)
𝑎𝑡𝑡
and the value we use for the update is given by 𝛿𝑡 = 𝛿𝑡+ if 𝑎𝑡 ≥ 0, otherwise 𝛿𝑡 = 𝛿𝑡− . This
decision is based on the requirement of having 𝛿𝑡 → 0 when 𝑠 → 0.

LS-DYNA DEV 10/27/16 (r:8004) 35-31 (Arc-length)


Arc-length LS-DYNA Theory Manual

Panel Response
1

0.8

0.6
Force

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Displacement

Figure 35-9 Snap-back buckling of panel, normalized force displacement


curve shown

35-32 (Arc-length) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Sparse Direct Linear Equation Solvers

39
Sparse Direct Linear Equation Solvers

LS-DYNA has 5 options for direct solution of the sparse systems of linear equations
that arise in LS-DYNA. All 5 options are based on the multifrontal algorithm [Duff and
Reid, 1983]. Multifrontal is a member of the current generation of sparsity preserving
factorization algorithms that also have very fast computational rates. That is multifrontal
works with a sparsity preserving ordering to reduce the overall size of the direct
factorization and the amount of work it takes to compute that factorization.

39.1 Sparsity Preserving Orderings


In LS-DYNA there are two ordering algorithms for preserving the sparsity of the
direct factorization. The algorithms are Multiple Minimum Degree (MMD) and METIS
[Karypis and Kumar, 1998]. MMD computes the ordering using locally based decisions
and a bottom-up approach. It is inexpensive and very effective for small problems that are
problems with fewer than 100,000 rows. METIS computes the ordering from a top down
approach. While METIS usually takes more time than MMD to compute the ordering, the
METIS ordering reduces the work for the factorization enough to recover the additional
ordering cost. METIS is especially effective for large problems, especially those that are
modeling three-dimensional solids.

The user can specify either algorithm using keyword *CONTROL_IMPLICIT_LIN-


EAR. The default is to use MMD for problems with fewer than 100,000 rows and METIS
for problems with more than 100,000 rows. We recommend that the user try both
orderings as sometimes MMD is better than METIS on large problems that are not three-
dimensional solids.

LS-DYNA DEV 10/27/16 (r:8004) 36-1 (Sparse Direct Linear Equation Solvers)
Sparse Direct Linear Equation Solvers LS-DYNA Theory Manual

39.2 Multifrontal Algorithm


The multifrontal algorithm factors a sparse matrix in a way that vastly reduces the
amount of work required to compute the factorization compared to methods such as the
frontal, profile, skyline, and variable band. These older methods counted on clustering the
nonzero entries of the factorization close to the diagonal to keep the size of the factorization
and the amount of work required to compute the factorization to a minimum. The
factorization was then computed in a serial, left-to-right fashion, essentially following a
chain of computations.

The multifrontal algorithm instead follows a tree of computations where the tree
structure is established by the sparsity preserving orderings, See Figure 39.1. It is this tree
structure that greatly reduces the work required to compute a factorization and the size of
the resulting factorization. At the bottom of the tree, a frontal matrix is assembled with the
original matrix data and those columns that are fully assembled are eliminated. The
remainder of the frontal matrix is updated from the factored columns and passed up the
tree to the parent front in what is called an update matrix. As the computation works its
way up the tree, a frontal matrix is formed by assembling the original matrix data and the
update matrices from its children in the tree. The fully assembled columns are factored
and the remaining columns updated and passed up the tree. At the root (top) of the tree,
the remaining columns are factored.

By organizing the factorization as a sequence of partial factorization of dense frontal


matrices, the multifrontal algorithm can be very fast in performing the required
computations. It can use all of the modern technology for dense linear algebra to get high
performance computational kernels that should achieve near peak computational
performance for a given processor. Only 1% to 5% of the work of the factorization is
performed with slower operations such as scatter/gather.

36-2 (Sparse Direct Linear Equation Solvers) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Sparse Direct Linear Equation Solvers

original matrix data

To parent front

From children fronts

computations in single front

Figure 39.2. Single front algorithm.

39.3 The Five Solver Options


In LS-DYNA three new direct solution options were added. For backward
compatibility, the two older options were kept. The five options are:

Multifrontal elimination tree


Figure 39.1. Multifrontal algorithm.

Solver Method
LS-DYNA DEV 10/27/16 (r:8004) 36-3 (Sparse Direct Linear Equation Solvers)
Sparse Direct Linear Equation Solvers LS-DYNA Theory Manual

No.
1 Older implementaion of Solver No. 4. Uses Real*4 arithmetic, has out of
memory capabilities as well as distributed memory parallelism. Only uses
MMD ordering. Was former default method. Retained for backward
compatibility. We recommend switching to Solver No. 4 for improved
performance.
3 Same as 1 except uses Real*8 arithmetic. We recommend switching to Solver
No. 5 or 6 for improved performance.
4 Real*4 implementation of multifrontal which includes automatic out-of-memory
capabilities as well as distributed memory parallelism. Can use either MMD or
METIS orderings. Default method.
5 Real*8 implementation of Solver No. 4.
6 Multifrontal solver from BCSLIB-EXT [Boeing Company, 1999]. Uses Real*8
arithmetic with extensive capabilities for large problems and some Shared
Memory Parallelism.
Can use either MMD or METIS orderings. If the other solvers cannot factor the
problem in the allocated memory, try using this solver.

We strongly recommend using Solvers 4 through 6. Solvers 1 and 3 are included for
backward compatibility with older versions of LS-DYNA but are slower the Solvers 4
through 6. Solvers 4 and 5 are 2 to 6 times faster than the older versions, respectively.
Solver 6 on a single processor computer should be comparable to Solver 5 but has more
extensive capabilities for solving very large problems with limited memory. Solvers 4 and
5 should be used for distributed memory parallel implementations of LS-DYNA. Solver 6
can be used in shared memory parallel.

In an installation of LS-DYNA where both integer and real numbers are stored in 8
byte quantities, then Solvers 1 and 3 are equivalent and Solvers 4 and 5 are equivalent.

39.4 Treating Matrix Singularities


LS-DYNA has two different techniques for preventing singularities in the stiffness
matrix, K. The most common type of matrix singularity arises from the use of certain types
of shell elements. These shell elements generate no matrix contribution in the normal
direction for each node. Depending on the geometry around the node and what other
types of elements are connected to the node, there may or may not be a matrix singularity
associated with the rotation around the normal direction at one or more nodes. This is
commonly called the drilling rotation singularity.

The first way LS-DYNA has for preventing such matrix singularities is to add a
small amount of stiffness in the normal direction at each node of every shell element that
has the drilling rotation problem. This “drilling” stiffness matrix is orthogonal to rigid
36-4 (Sparse Direct Linear Equation Solvers) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Sparse Direct Linear Equation Solvers

body motions. The user can control whether this approach is used and how much stiffness
is added via the *CONTROL_IMPLICIT_SOLUTION keyword card. DRLMTH and DRL-
PARM are set in fields 5 and 6. If DRLMTH = 1 then this approach is used. The amount of
stiffness added is controlled via DRLPARM. The default for DRLPARM is 1.0 for linear
problems and 100.0 for nonlinear problems. DRLPARM ∗ .0001 is added in the normal
direction at each node to the diagonal terms associated with the rotational degrees of
freedom for certain types of elemental matrices. For eigenvalue problems the amount of
stiffness added is 1.E-12.

Adding stiffness to handle the drilling rotation problem has been used extensively.
While a robust and reliable approach, its drawback is that the added stiffness may affect
the quality of the computed results. The user can also select not to use this approach and
depend solely on AUTOSPC, the other method for preventing matrix singularities.

AUTOSPC stands for AUTOmatic Single Point Constraints. AUTOSPC examines K


after all of the elemental matrices have been assembled and all of the constraints have been
applied for columns that are singular. The user controls AUTOSPC using ASPCMTH and
ASPCTOL, fields 7 and 8 of the CONTROL_IMPLICIT_SOLUTION keyword card. If
ASPCMTH = 1, AUTOSPC is used. For every set of columns of K that correspond to the
translational or rotational degrees of freedom for a node or rigid body those columns are
examined. The singular values of the diagonal block of the columns are computed. If the
ratio of the smallest and largest singular values is less than ASPCTOL then the set of the
columns is declared singular and a constraint is imposed to remove the singularity. The
defaults for ASPCTOL is 1.E-6 when the matrix is assembled in REAL*4 precision and 1.E-8
when REAL*8 is used. The imposed constraint sets the degree of freedom to zero that is
associated with the column that has the largest component in the null space of the columns.
If all of the singular values are less than ASPCTOL all of the degrees of freedom in the
block are constrained to zero.

LS-DYNA DEV 10/27/16 (r:8004) 36-5 (Sparse Direct Linear Equation Solvers)
LS-DYNA Theory Manual Sparse Eigensolver

40
Sparse Eigensolver

LS-DYNA now includes the Block Shift and Invert Lanczos eigensolver from
BCSLIB-EXT. This eigensolver is used in LS-DYNA to compute the normal modes and
mode shapes for the vibration analysis problem
𝐊𝚽 = 𝐌𝚽𝚲, (40.1)
where 𝐊 and 𝐌are the assembled stiffness and mass matrices, 𝚽 are the eigenvectors
(normal mode shapes) and 𝚲 are the eigenvalues (normal modes).

The Lanczos algorithm iteratively computes a better and better approximation to the
extreme eigenvalues and the corresponding eigenvectors of the ordinary eigenvalue
problem 𝐀𝚽 = 𝚽𝚲where 𝐀 is a real symmetric matrix using only matrix-vector multiplies.
To use Lanczos on the vibration analysis problem it must be changed to
(𝐊 − 𝛔𝐌)−1 𝐌𝚽 = 𝚽𝚯, (40.2)
where each shifted and inverted eigenvalue 𝜃𝑖 = 1/(𝜆𝑖 − 𝜎). This change to an ordinary
eigenvalue problem makes the eigenvalues of the original problem near  become the
extreme eigenvalues of the ordinary eigenvalue problem. This helps the Lanczos algorithm
compute those eigenvalues quickly.

BCSLIB-EXT uses a sophisticated logic to choose a sequence of shifts, 𝜎𝑖 , to enable


the computation of a large number of eigenvalues and eigenvectors. At each shift the
factorization of 𝐊 − 𝛔𝐌 is computed. The factorization provides the matrix inertia that
tells the algorithm how many eigenvalues are to the left of any given 𝜎𝑖 . Given the inertia
information, BCSLIB-EXT can tell how many eigenvalues are in a given interval and
determine if all of the eigenvalues in that interval have been computed. As a result,
BCSLIB-EXT is a very robust eigensolver.

The implementation of BCSLIB-EXT in LS-DYNA includes a shared memory


implementation. However only limited parallel speed-up is available for most problems.
This is because the eigensolution requires a vast amount of data that for most problems this
data has to be stored on I/O files. The wall clock time for the eigensolver is as much a
LS-DYNA DEV 10/27/16 (r:8004) 37-1 (Sparse Eigensolver)
Sparse Eigensolver LS-DYNA Theory Manual

function of the speed of the I/O subsystem on the computer as the CPU time. Parallelism
can only speed up the CPU time and does nothing to speed-up the I/O time.

The user can request how many and which eigenvalues to compute using the
keyword *CONTROL_IMPLICIT_EIGENVALUE. Via the parameters on this keyword, the
user can request any of the following problems:
• Compute the lowest 50 modes (that is nearest to zero)
• Compute the 20 modes nearest to 30 Hz.
• Compute the lowest 20 modes between 10 Hz and 50 Hz.
• Compute all of the modes between 10 Hz and 50 Hz.
• Compute all of the modes below 50 Hz.
• Compute the 30 modes nearest to 30 Hz between 10 Hz and 50 Hz.

40.1 The Eigenvalue Problem for Rotating Systems


Rotating systems, such as the compressor and turbine assembly of a jet engine, have
large inertial forces that are functions of the distance from the axis of rotation. These forces
are naturally generated in LS-DYNA if the system is modeled as rotating at the proper
angular velocity. However, this is often inconvenient for postprocessing because the
solution has an oscillatory character imposed it due to the rotation. A commonly used
approach to bypass this difficulty is to impose body forces that are equivalent to the inertial
forces due to rotation. In LS-DYNA, these forces are imposed through
*LOAD_BODY_GENERALIZED and related keywords.

For a system with a constant angular velocity 𝛚 = {𝜔𝑥 , 𝜔𝑦 , 𝜔𝑧 }T , the body force
added to the applied load is
𝐅B = −𝐌{2𝛚 × 𝐮̇ + 𝛚 × (𝛚 × (𝐫 + 𝐮))}. (40.3)

In this equation, 𝐫 is the initial coordinate at a point and 𝐮 is the displacement.


Because the body force is a function of both the velocity and displacement, it contributes
both damping and stiffness matrices to the eigenvalue problem. Furthermore, since the
term involving the initial coordinate creates an initial stress in the structure, the initial
stress matrix 𝐊𝜎 (also called the nonlinear stiffness) is also added to the eigenvalue
problem.

The damping and stiffness terms are easily derived in matrix form once the cross
product is expressed in matrix form.

37-2 (Sparse Eigensolver) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Sparse Eigensolver

0 −𝜔𝑧 𝜔𝑦
⎡ ⎤ 𝑥
⎢ 𝜔𝑧
𝛚 × 𝐫 = 𝛀𝐫 = ⎢ 0 −𝜔𝑥 ⎥
⎥ {𝑦}. (40.4)
⎣−𝜔𝑦 𝜔𝑥 0 ⎦ 𝑧
The linearized equation for vibration is
𝐌𝐮̈ + 𝐂𝐮̇ + [𝐊 + 𝐊σ ]𝐮 = −𝐌{𝛀𝐮̇ + 𝛀2 𝐮}. (40.5)

Rewriting this equation into the traditional form for eigenvalue analysis produces:
𝐌𝐮̈ + 𝐂R 𝐮̇ + 𝐊R 𝐮 = 0
𝐂R = 𝐂 + 𝐌𝛀 (40.6)
𝐌𝐮̈ + 𝐂R 𝐮̇ + 𝐊R 𝐮 = 0.

The inertial contribution to the damping matrix is not symmetric, nor does it fulfill
the requirements for Rayleigh damping, and therefore the resulting eigenvectors and
eigenvalues are complex. The inertial term to the stiffness matrix is, however, symmetric
and it softens the structure, thereby reducing its natural frequencies.

If the damping term is omitted, the matrices are real and symmetric, and the
resulting eigenvalue problem may be solved with the standard eigenvalue methods. The
natural frequencies won’t be correct, but they are typically close enough to the complex
solution that they can be used for initial design calculations.

LS-DYNA DEV 10/27/16 (r:8004) 37-3 (Sparse Eigensolver)


LS-DYNA Theory Manual Boundary Element Method

41
Boundary Element Method

LS-DYNA can be used to solve for the steady state or transient fluid flow about a
body using a boundary element method. The method is based on the work of Maskew
[1987], with the extension to unsteady flow with arbitrary body motion following the work
of Katz and Maskew [1988]. The theory which underlies the method is restricted to
inviscid, incompressible, attached fluid flow. The method should not be used to analyze
flows where shocks or cavitation are present.

In practice the method can be successfully applied to a wider class of fluid flow
problems than the assumption of inviscid, incompressible, attached flow would imply.
Many flows of practical engineering significance have large Reynolds numbers (above 1
million). For these flows the effects of fluid viscosity are small if the flow remains attached,
and the assumption of zero viscosity may not be a significant limitation. Flow separation
does not necessarily invalidate the analysis. If well-defined separation lines exist on the
body, then wakes can be attached to these separation lines and reasonable results can be
obtained. The Prandtl-Glauert rule can be used to correct for non-zero Mach numbers in
air, so the effects of aerodynamic compressibility can be correctly modeled (as long as no
shocks are present).

41.1 Governing Equations


The partial differential equation governing inviscid, incompressible fluid flow is
given by LaPlace’s equation
∇2 𝛷 = 0, (41.1)
where 𝛷 is the velocity potential (a scalar function). The fluid velocity anywhere in the
flow field is equal to the gradient of 𝛷. The boundary condition on this partial differntial
equation is provided by the condition that there must be no flow in the direction normal to
the surface of the body. Note that time does not appear in Equation (41.1). This is because

LS-DYNA DEV 10/27/16 (r:8004) 38-1 (Boundary Element Method)


Boundary Element Method LS-DYNA Theory Manual

the assumption of incompressibility implies an infinite sound speed; any disturbance is felt
everywhere in the fluid instantaneously. Although this is not true for real fluids, it is a
valid approximation for a wide class of low-speed flow problems.

Equation (41.1) is solved by discretizing the surface of the body with a set of
quadrilateral or triangular surface segments (boundary elements). Each segment has an
associated source and doublet strength. The source strengths are computed from the free-
stream velocity, and the doublet strengths are determined from the boundary condition.
By requiring that the normal component of the fluid velocity be zero at the center of each
surface segment, a linear system of equations is formed with the number of equations equal
to the number of unknown doublet strengths. When this system is solved, the doublet
strengths are known. The source and doublet distributions on the surface of the body then
completely determine the flow everywhere in the fluid.

The linear system for the unknown doublet strengths is shown in Equation (1.2).
[mic]{𝜇} = {rhs}. (41.2)

In this equation  are the doublet strengths, [mic is the matrix of influence
coefficients which relate the doublet strength of a given segment to the normal velocity at
another segment’s mid-point, and rhs is a right-hand-side vector computed from the
known source strengths. Note that mic is a fully-populated matrix. Thus, the cost to
compute and store the matrix increases with the square of the number of segments used to
discretize the surface of the body, while the cost to factor this matrix increases with the
cube of the number of segments. Users should keep these relations in mind when defining
the surface segments. A surface of 1000 segments can be easily handled on most any
computer, but a 10,000 segment representation would not be feasible on any but the most
powerful supercomputers.

41.2 Surface Representation


The surface of the body is discretized by a set of triagular or quadrilateral surface
segments. The best fluid-structure interaction results will be obtained if the boundary
element segments coincide with, and use identical nodes as, the structural segments used
to define the body. An input format has been implemented to make this easy if thin shell
elements are used to define the structure (see the User’s Manual). Using the same nodes to
define the boundary elements and the structure guarantees that the boundary elements
follow the structure as it deforms, and provides a means for the fluid pressure to load the
structure.

The nodes used to define the corners of the boundary element segments must be
ordered to provide a normal vector which points into the fluid (Figure 41.1).

38-2 (Boundary Element Method) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Boundary Element Method

normal

node 4 node 3

node 1 node 2
Figure 41.1. Counter-clockwise ordering of nodes when viewed from fluid
looking towards solid provides unit normal vector pointing into the fluid.

Triangular segments are specified by using the same node for the 3rd and 4th corner
of the segment (the same convention used for shell elements in LS-DYNA). Very large
segments can be used with no loss of accuracy in regions of the flow where the velocity
gradients are small. The size of the elements should be reduced in areas where large
velocity gradients are present. Finite-precision arithmetic on the computer will cause
problems if the segment aspect ratios are extremely large (greater than 1000). The most
accurate results will be obtained if the segments are rectangular, and triangular segments
should be avoided except for cases where they are absolutely required.

41.3 The Neighbor Array The fluid velocities (and, therefore, the fluid
pressures) are determined by the gradient of the velocity potential. On the surface of the
body, this can be most easily computed by taking derivatives of the doublet distribution on
the surface. These derivatives are computed using the doublet strengths on the boundary
element segments. The “Neighbor Array” is used to specify how the gradient is computed
for each boundary element segment. Thus, accurate results will not be obtained unless the
neighbor array is correctly specified by the user.

Each boundary element segment has 4 sides (see Figure 41.2). Side 1 connects the 1st
and 2nd nodes, side 2 connects the 2nd and 3rd nodes, etc. The 4th side is null for
triangular segments.

LS-DYNA DEV 10/27/16 (r:8004) 38-3 (Boundary Element Method)


Boundary Element Method LS-DYNA Theory Manual

node 4 node 3
side 3
side 4
side 2

side 1
node 1 node 2
Figure 41.2. Each segment has 4 sides.

For most segments the specification of neighbors is straightforward. For the typical
case a rectangular segment is surrounded by 4 other segments, and the neighbor array is as
shown in Figure 41.3. A biquadratic curve fit is computed, and the gradient is computed as
the analytical derivative of this biquadratic curve fit evaluated at the center of segment j.

There are several situations which call for a different specification of the neighbor
array. For example, boundary element wakes result in discontinuous doublet distributions,
and the biquadratic curve fit should not be computed across a wake. Figure 41.4 illustrates
a situation where a wake is attached to side 2 of segment 𝑗. For this situation two options
exist. If neighbor (2, 𝑗) is set to zero, then a linear computation of the gradient in the side 2
to side 4 direction will be made using the difference between the doublet strengths on
segment 𝑗 and segment neighbor (4, 𝑗). By specifying neighbor (2, 𝑗) as a negative number
the biquadratic curve fit will be retained. The curve fit will use segment 𝑗, segment
neighbor (4, 𝑗), and segment –neighbor (2, 𝑗); which is located on the opposite side of
segment neighbor (4, 𝑗) as segment 𝑗. The derivative in the side 2 to side 4 direction is then
analytically evaluated at the center of segement j using the quadratic curve fit of the
doublet strengths on the three segments shown.

neighbor (3, j)

neighbor(4, j) side 3
segment j neighbor (2, j)
side 4 side 2
side 1

neighbor (1, j)

Figure 41.3. Typical neighbor specification.

38-4 (Boundary Element Method) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Boundary Element Method

-neighbor (2, j) neighbor (4, j) segment j


side 4 side 2

Figure 41.4. If neighbor (2, 𝑗) is a negative number it is assumed to lie on the


opposite side of neighbor (4, 𝑗) as segment 𝑗.

A final possibility is that no neighbors at all are available in the side 2 to side 4
direction. In this case both neighbor (2, 𝑗) and neighbor (4, 𝑗) can be set to zero, and the
gradient in that direction will be assumed to be zero. This option should be used with
caution, as the resulting fluid pressures will not be accurate for three-dimensional flows.
However, this option is occaisionally useful where quasi-two dimensional results are
desired. All of the above options apply to the side 1 to side 3 direction in the obvious ways.

For triangular boundary element segments side 4 is null. Gradients in the side 2 to
side 4 direction can be computed as described above by setting neighbor (4, 𝑗) to zero (for a
linear derivative computation) or to a negative number (to use the segment on the other
side of neighbor (2, 𝑗) and a quadratic curve fit). There may also be another triangular
segment which can be used as neighbor (4, 𝑗) (see Figure 41.5).

41.4 Wakes
Wakes should be attached to the boundary element segments at the trailing edge of
a lifting surface (such as a wing, propeller blade, rudder, or diving plane). Wakes should
also be attached to known separation lines (such as the sharp leading edge of a delata wing

neighbor(4, j)
segment j
side 2

Figure 41.5. Sometimes another triangular boundary element segment can be


used as neighbor(4,j).

at high angles of attack). Wakes are required for the correct computation of surface
LS-DYNA DEV 10/27/16 (r:8004) 38-5 (Boundary Element Method)
Boundary Element Method LS-DYNA Theory Manual

pressures for these situations. As described above, two segments on opposite sides of a
wake should never be used as neighbors. Correct specification of the wakes is required for
accurate results.

Wakes convect with the free-stream velocity. The number of segments in the wake
is controlled by the user, and should be set to provide a total wake length equal to 5-10
times the characteristic streamwise dimension of the surface to which the wake is attached.
For example, if the wake is attached to the trailing edge of a wing whose chord is 1, then
the total length of the wake should at least 5, and there is little point in making it longer
than 10. Note that each wake segment has a streamwise length equal to the magnitude of
the free stream velocity times the time increment between calls to the Boundary Element
Method routine. This time increment is the maximum of the LS-DYNA time step and
DTBEM specified on Card 1 of the BEM input. The influence coefficients for the wake
segments must be recomputed for each call to the Boundary Element Method, but these
influence coefficients do not enter into the matrix of influence coefficients which must be
factored.

41.5 Execution Time Control


The Boundary Element Method will dominate the total execution time of a LS-
DYNA calculation unless the parameters provided on Card 1 of the BEM input are used to
reduce the number of calls to the BEM. This can usually be done with no loss in accuracy
since the characteristic time of the structural dynamics and the fluid flow are so different.
For example, the characteristic time in LS-DYNA is given by the characteristic length of the
smallest structural element divided by the speed of sound of the material. For typical
problems this characteristic time might be on the order of microseconds. Since the fluid is
assumed to be incompressible (infinite speed of sound), the characteristic time of the fluid
flow is given by the streamwise length of the smallest surface (e.g. a rudder) divided by
the fluid velocity. For typical problems this characteristic time might be on the order of
milliseconds. Thus, for this example, the boundary element method could be called only
once for every 1000 LS-DYNA iterations, saving an enormous amount of computer time.

The parameter DTBEM on Card 1 of the BEM input is used to control the time
increment between calls to the boundary element method. The fluid pressures computed
during the last call to the BEM will continue to be used for subsequent LS-DYNA iterations
until DTBEM has elapsed.

A further reduction in execution time may be obtained for some applications using
the input parameter IUPBEM. This parameter controls the number of calls to the BEM
routine between computation (and factorization) of the matrix of influence coefficients
(these are time-consuming procedures). If the motion of the body is entirely rigid body
motion there is no need to recompute and factor the matrix of influence coefficients, and

38-6 (Boundary Element Method) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Boundary Element Method

the execution time of the BEM can be significantly reduced by setting IUPBEM to a very
large number. For situations where the motion of the body is largely rigid body motion
with some structural deformation an intermediate value (e.g. 10) for IUPBEM can be used.
It is the user’s responsibility to verify the accuracy of calculations obtained with IUPBEM
greater than 1.

The final parameter for controlling the execution time of the boundary element
method is FARBEM. The routine which calculates the influence coefficients switches
between an expensive near-field and an inexpensive far-field calculation depending on the
distance from the boundary element segment to the point of interest. FARBEM is a
nondimensional parameter which determines where the far-field boundary lies. Values of
FARBEM of 5 and greater will provide the most accurate results, while values as low as 2
will provide slightly reduced accuracy with a 50% reduction in the time required to
compute the matrix of influence coefficients.

41.6 Free-Stream Flow


The free-stream flow is specified in the second card of input. The free-stream
velocity is assumed to be uniform. The free-stream static pressure is assumed to be
uniform, and can be used to load the structure for hydrostatic pressure. If the structure has
an internal pressure, the free-stream static pressure should be set to the difference between
the external and internal static pressures. The Mach number can be used to correct for the
effect of compressibility in air (as long as no shocks are present). Following the Prandtl-
Glauert correction, the pressures due to fluid flow are increased as follows
𝑑𝑝uncorrected
𝑑𝑝corrected = (41.3)
√1 − 𝑀 2
where M is the free-stream Mach number. Note that this correction is only valid for flows
in a gas (it is not valid for flows in water).

LS-DYNA DEV 10/27/16 (r:8004) 38-7 (Boundary Element Method)


LS-DYNA Theory Manual SPH

42
SPH

Smoothed Particle Hydrodynamics (SPH) is an N-body integration scheme


developed by Lucy, Gingold and Monaghan [1977]. The method was developed to avoid
the limitations of mesh tangling encountered in extreme deformation problems with the
finite element method. The main difference between classical methods and SPH is the
absence of a grid. Therefore, the particles are the computational framework on which the
governing equations are resolved. This new model requires a new calculation method,
which is briefly explained in the following.

42.1 SPH Formulation


42.1.1 Definitions
The particle approximation of a function is:

Πℎ 𝑓 (𝑥) = ∫ 𝑓 (𝑦)𝑊(𝑥 − 𝑦, ℎ)𝑑𝑦, (42.1)

where 𝑊 is the kernel function.

The Kernel function 𝑊 is defined using the function 𝜃 by the relation:


1
𝑊(𝐱, ℎ) = 𝜃(𝐱). (42.2)
ℎ(𝐱)𝑑
where 𝑑 is the number of space dimensions and ℎ is the so-called smoothing length which
varies in time and in space.

𝑊(𝐱, ℎ) should be a centrally peaked function. The most common smoothing kernel
used by the SPH community is the cubic B-spline which is defined by choosing 𝜃 as:

LS-DYNA DEV 10/27/16 (r:8004) 39-1 (SPH)


SPH LS-DYNA Theory Manual

⎧ 3 3
{
{ 1 − 𝑢2 + 𝑢3 for |𝑢| ≤ 1
{ 2 4
𝜃(𝑢) = 𝐶 × ⎨1 3 (42.3)
{ (2 − 𝑢) for 1 ≤ |𝑢| ≤ 2
{
{4
⎩0 for 2 < |𝑢|
where C is a constant of normalization that depends on the number of space dimensions.

The SPH method is based on a quadrature formula for moving particles ((𝐱𝑖 (𝑡)) 𝑖 ∈
{1. . 𝑁}, where 𝐱𝑖 (𝑡) is the location of particle 𝑖, which moves along the velocity field v.

The particle approximation of a function can now be defined by:


𝑁

Π 𝑓 (𝐱𝑖 ) = ∑ 𝑤𝑗 𝑓 (𝐱𝑖 )𝑊(𝐱𝑖 − 𝐱𝑗 , ℎ), (42.4)
𝑗=1
𝑚𝑗
where 𝑤𝑗 = 𝜌𝑗 is the “weight” of the particle. The weight of a particle varies proportionally
to the divergence of the flow.

The SPH formalism implies a derivative operator. A particle approximation for the
derivative operator must be defined. Before giving the definition of this approximation,
we define the gradient of a function as:
∇𝑓 (𝑥) = ∇𝑓 (𝑥) − 𝑓 (𝑥)∇1(𝑥), (42.5)
where 1 is the unit function.

Starting from this relation, we can define the particle approximation to the gradient
of a function:
𝑁 𝑚𝑗
Πℎ ∇𝑓 (𝐱𝑖 ) = ∑ [𝑓 (𝐱𝑗 )𝐴𝑖𝑗 − 𝑓 (𝐱𝑖 )𝐴𝑖𝑗 ], (42.6)
𝑗=1
𝜌𝑗

1 ||𝐱𝑖 −𝐱𝑗 ||
where 𝐴𝑖𝑗 = 𝜃′( ℎ ).
ℎ𝑑+1

We can also define the particle approximation of the partial derivative ∂𝑥∂𝛼:
𝑁
∂𝑓
Π ( 𝛼 )(𝐱𝑖 ) = ∑ 𝑤𝑗 𝑓 (𝐱𝑗 𝐴𝛼 (𝐱𝑖 , 𝐱𝑗 ),

(42.7)
∂𝑥 𝑗=1

1 (𝐱𝑖 −𝐱𝑗 ) |∣𝐱𝑖 −𝐱𝑗 ∣|


where 𝐀 is the operator defined by: 𝐀(𝐱𝑖 , 𝐱𝑗 ) = 𝜃′ (ℎ(𝐱 ,𝐱 )), 𝐴𝛼 is the
ℎ𝑑+1 (𝐱 𝑖 ,𝐱𝑗 ) |∣𝐱 −𝐱 ∣|
𝑖 𝑗 𝑖 𝑗

component 𝛼 of the 𝐀 vector.

42.1.2 Discrete Form of Conservation Equations


We are looking for the solution of the equation:

39-2 (SPH) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual SPH

𝐿𝑣 (𝜙) + div𝐅(𝐱, 𝑡, 𝜙) = 𝑆, (42.8)


where 𝜙 ∈ 𝑅𝑑 is the unknown, 𝐅𝛽 with 𝛽 ∈ {1. . 𝑑} represents the conservation law and 𝐿𝑣
is the transport operator defined by:
𝑑 ∂(𝐯𝑙 𝜙)
∂𝜙
𝐿𝑣 : 𝜙 → 𝐿𝑣 (𝜙) = +∑ . (42.9)
∂𝑡 𝑙=1 ∂𝑥𝑙

The strong formulation approximation:


In the search of the strong solution, the equation is kept at its initial formulation.
The discrete form of this problem implies the definition of the operator of derivation 𝐷
defined by:
𝐷: 𝜙 → 𝐷𝜙(𝑥) = ∇𝜙(𝑥) − 𝜙(𝑥)∇1(𝑥). (42.10)
The particle approximation of this operator is:
𝑁
𝐷ℎ 𝜙(𝐱𝑖 ) = ∑ 𝑤𝑗 (𝜙(𝐱𝑗 ) − 𝜙(𝐱𝑖 ))𝐴𝑖𝑗 , (42.11)
𝑗=1

where 𝐴𝑖𝑗 is defined previously.

Finally, the discrete form of the strong formulation is written:


𝑑
(𝑤 𝜙(𝐱𝑖 )) + 𝑤𝑖 𝐷ℎ 𝐹(𝐱𝑖 ) = 𝑤𝑖 𝑆(𝐱𝑖 ), (42.12)
𝑑𝑡 𝑖
But this form is not conservative; therefore the strong formulation is not numerically
acceptable. Thus, we are compelled to use the weak form.

The weak formulation approximation:


In the weak formulation, the adjoint of the 𝐿𝑣 operator is used:
𝑑
∂𝜙 ∂𝜙
𝐿∗𝑣 : 𝜙 → 𝐿∗𝑣 (𝜙) = + ∑ 𝑣𝑙 𝑙 . (42.13)
∂𝑡 𝑙=1 ∂𝑥
The discrete form of this operator corresponds to the discrete formulation of the adjoint of
𝐷ℎ,𝑠 :
𝑁
𝐷∗ℎ,𝑠 𝜙(𝐱𝑖 ) = ∑ 𝑤𝑗 (𝜙(𝐱𝑖 )𝐴𝑖𝑗 − 𝜙(𝐱𝑗 )𝐴𝑗𝑖 ). (42.14)
𝑗=1

A discrete adjoint operator for the partial derivative is also necessary, and is taken to
be the 𝛼 − 𝑡ℎ component of the operator:
𝑁
𝐷∗𝛼 𝜙(𝐱𝑖 ) = ∑ 𝑤𝑗 𝜙(𝐱𝑗 )𝐴𝛼 (𝐱𝑖 , 𝐱𝑗 ) − 𝑤𝑗 𝜙(𝐱𝑖 )𝐴𝛼 (𝐱𝑗 , 𝐱𝑖 ) (42.15)
𝑗=1

LS-DYNA DEV 10/27/16 (r:8004) 39-3 (SPH)


SPH LS-DYNA Theory Manual

These definitions are leading to a conservative method. Hence, all the conservative
equations encountered in the SPH method will be solved using the weak form.

42.1.3 Applications to Conservation Equations


With the definitions explained above, the conservation equations can now be written
in their discrete form.

Momentum conservation equation:


𝑑𝐯𝛼 1 ∂(𝜎 𝛼𝛽 )
(𝐱𝑖 (𝑡)) = (𝐱𝑖 (𝑡)), (42.16)
𝑑𝑡 𝜌𝑖 ∂𝑥𝑖
where 𝛼, 𝛽 are the space indices.

The particle approximation of the weak form of this equation is:

𝑑𝐯𝛼 𝑁
⎛ 𝜎 𝛼,𝛽 (𝐱𝑖 ) 𝜎 𝛼,𝛽 (𝐱𝑗 ) ⎞
(𝐱 ) = ∑ 𝑚𝑗 ⎜ 𝐴𝑖𝑗 − 𝐴𝑗𝑖 ⎟. (42.17)
𝑑𝑡 𝑖 𝑗=1 ⎝ 𝜌𝑖
2 𝜌2𝑗 ⎠

Energy conservation equation:


𝑑𝐸 𝑃
= − ∇𝐯. (42.18)
𝑑𝑡 𝜌

The particle approximation of the weak form of this equation is:


𝑑𝐸 𝑃𝑖 𝑁
(𝐱 ) = − 2 ∑ 𝑚𝑗 (𝑣(𝐱𝑗 ) − 𝑣(𝐱𝑖 ))𝐴𝑖𝑗 . (42.19)
𝑑𝑡 𝑖 𝜌𝑖 𝑗=1

42.1.4 Formulation Available in LS-DYNA


It is easy from the general formulation displayed in Equation (42.14) to extend the
SPH formalism to a set of equations of discretization for the momentum equation.

For example, if we choose the smoothing function to be symmetric, this can lead to
the following equation:
𝛼,𝛽
𝑑𝐯𝛼 𝑁
⎛ 𝜎 𝛼,𝛽 (𝐱𝑖 ) 𝜎 (𝐱𝑗 )⎞
(𝐱 ) = ∑ 𝑚𝑗 ⎜ + ⎟ 𝐴𝑖𝑗 . (42.20)
𝑑𝑡 𝑖 𝑗=1 ⎝ 𝜌𝑖
2 𝜌2𝑗 ⎠
This is what we call the “symmetric formulation”, which is chosen in the *CONTROL_SPH
card (IFORM = 2).

Another possible choice is to define the momentum equation by:


39-4 (SPH) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual SPH

𝛼,𝛽
𝑑𝐯𝛼 𝑁
⎛ 𝜎 𝛼,𝛽 (𝐱𝑖 ) 𝜎 (𝐱𝑗 ) ⎞
(𝐱𝑖 ) = ∑ 𝑚𝑗 ⎜ 𝐴𝑖𝑗 − 𝐴𝑗𝑖 ⎟. (42.21)
𝑑𝑡 𝑗=1 ⎝ 𝜌𝑖 𝜌𝑗 𝜌𝑖 𝜌𝑗 ⎠

This is the “fluid formulation” invoked with IFORM = 5 which gives better results
than other SPH formulations when fluid material are present, or when material with very
different stiffness are used.

42.2 Sorting
In the SPH method, the location of neighboring particles is important. The sorting
consists of finding which particles interact with which others at a given time. A bucket sort
is used that consists of partitioning the domain into boxes where the sort is performed.
With this partitioning the closest neighbors will reside in the same box or in the closest
boxes. This method reduces the number of distance calculations and therefore the CPU
time.

42.3 Artificial Viscosity


The artificial viscosity is introduced when a shock is present. Shocks introduce
discontinuities in functions. The role of the artificial viscosity is to smooth the shock over
several particles. To take into account the artificial viscosity, an artificial viscous pressure
term Π𝑖𝑗 [Monaghan & Gingold 1983] is added such that:
𝑝𝑖 → 𝑝𝑖 + Π𝑖𝑗 , (42.22)
where Π𝑖𝑗 = 𝜌̅1 (−𝛼𝜇𝑖𝑗 𝑐𝑖𝑗̅ + 𝛽𝜇2𝑖𝑗 ).
𝑖𝑗

The notation 𝑋 ̅̅̅ ̅𝑖𝑗 = 1 (𝑋𝑖 + 𝑋𝑗 ) has been used for median between 𝑋𝑖 and 𝑋𝑗 , 𝑐 is the
2
adiabatic sound speed, and
⎧ 𝑣𝑖𝑗 𝑟𝑖𝑗
{ℎ̅𝑖𝑗 2 if 𝑣𝑖𝑗 𝑟𝑖𝑗 < 0
𝜇𝑖𝑗 = ⎨ 𝑟𝑖𝑗 + 𝜂2 (42.23)
{
⎩ 0 otherwise
Here, 𝑣𝑖𝑗 = (𝑣𝑖 − 𝑣𝑗 ), and 𝜂 = 0.01ℎ̅𝑖𝑗 which prevents the denominator from vanishing.
2 2

LS-DYNA DEV 10/27/16 (r:8004) 39-5 (SPH)


SPH LS-DYNA Theory Manual

42.4 Time Integration


We use a simple and classical first-order scheme for integration. The time step is
determined by the expression:
ℎ𝑖
𝛿𝑡 = 𝐶CFL 𝑀𝑖𝑛𝑖 ( ), (42.24)
𝑐𝑖 + 𝑣𝑖
where the factor 𝐶CFL is a numerical constant.

The calculation cycle is:


Velocity/positions
Start LS-DYNA

Accelerations Smoothing length


LS-DYNA SPH

contact, boundary conditions Sorting


LS-DYNA SPH

Particles forces Density, strain rates


SPH SPH

Pressure, thermal energy, stresses


LS-DYNA

42.5 Initial Setup


Initially, we have a set of particles with two kinds of properties: physical and
geometrical properties.

Physical Properties:
The mass, density, constitutive laws are defined in the ELEMENT_SPH and the
PART cards.

Geometrical Properties:
39-6 (SPH) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual SPH

The geometrical properties of the model concern the way particles are initially
placed. Two different parameters are to be fixed: Δ𝑥𝑖 lengths and the CSLH coefficient.
These parameters are defined in the SECTION_SPH card.

A proper SPH mesh must satisfy the following conditions: it must be as regular as
possible and must not contain too large variations.

For instance, if we consider a cylinder SPH mesh, we have at least two possibilities:

The mesh number 2 includes too many inter-particle distance discrepancies.


Therefore, the first mesh, more uniform, is better.

Finite element coupling


Coupling finite elements and SPH elements is realized by using contact algorithms.
Users can choose any “nodes_to_surface” contact type where the slave part is defined with
SPH elements and the master part is defined with finite elements.

LS-DYNA DEV 10/27/16 (r:8004) 39-7 (SPH)


LS-DYNA Theory Manual Element-Free Galerkin

43
Element-Free Galerkin

Mesh-free methods, which construct the approximation entirely in terms of nodes,


permit reduced restriction in the discretization of the problem domain and are less
susceptible to distortion difficulties than finite elements. For a variety of engineering
problems with extremely large deformation, moving boundaries or discontinuities, mesh-
free methods are very attractive. The two most commonly used approximation theories in
mesh-free methods are the moving least-squares (MLS) approximation in the Element-free
Galerkin (EFG) method [Belytschko et al. 1994], and the reproducing kernel (RK)
approximation in the reproducing kernel particle method (RKPM) [Liu et al. 1995]. Since
these two methods lead to an identical approximation when monomial basis functions are
used, the MLS approximation is used as a basis to formulate the mesh-free discrete
equations in this section.

43.1 Moving least-squares


The Element-free Galerkin method uses the moving least-squares approximation to
construct the numerical discretization. The discrete MLS approximation of a function 𝑢(𝐱),
denoted by 𝑢ℎ (𝐱), is constructed by a combination of the monomials as
𝑛
𝑢ℎ (𝐱) = ∑ 𝐻𝑖 (𝐱)𝑏𝑖 (𝐱) ≡ 𝐇T (𝐱)𝐛(𝐱), (43.1)
𝑖=1

where 𝑛 is the order of completeness in this approximation, the monomial 𝐻𝑖 (𝐱) are basis
functions, and 𝑏𝑖 (𝐱) are the coefficients of the approximation.

The coefficients 𝑏𝑖 (𝐱) at any point 𝐱 are depending on the sampling points 𝐱𝐼 that are
collected by a weighting function 𝑤𝑎 (𝐱 − 𝐱𝐼 ). This weighting function is defined to have a
compact support measured by ‘a’, i.e., the sub-domain over which it is nonzero is small
relative to the rest of the domain. Each sub-domain ΔΩ𝐼 is associated with a node 𝐼. The
most commonly used sub-domains are disks or balls. A typical numerical model is shown
in Figure 43.1.
LS-DYNA DEV 10/27/16 (r:8004) 40-1 (Element-Free Galerkin)
Element-Free Galerkin LS-DYNA Theory Manual

ΩI

Figure 43.1. Graphical representation of mesh-free discretization

In this development, we employ the cubic B-spline kernel function as the weighting
function:

⎧2 2 3
{ − 4 (‖𝐱 − 𝐱𝐼 ‖) + 4 (‖𝐱 − 𝐱𝐼 ‖) ‖𝐱 − 𝐱𝐼 ‖ 1 ⎫
{3 for 0 ≤ ≤ }
{ 𝑎 𝑎 𝑎 2}}
{ }
{4 ‖𝐱 − 𝐱𝐼 ‖ ‖𝐱 − 𝐱𝐼 ‖ 2 4 ‖𝐱 − 𝐱𝐼 ‖ 2 1 ‖𝐱 − 𝐱𝐼 ‖ }
𝑤𝑎 (𝐱 − 𝐱𝐼 ) = ⎨ − 4 ( ) + 4( ) − ( ) for < ≤ 1⎬ (43.2)
{3 𝑎 𝑎 3 𝑎 2 𝑎 }
{ }
{ }
{0 otherwise }
{ }
⎩ ⎭

The moving least-squares technique consists in minimizing the weighted L2 -Norm


NP 𝑛 2
𝐽 = ∑ Wa (𝐱) (𝐱 − 𝐱𝐼 ) [∑ 𝐻𝑖 (𝐱)𝑏𝑖 (𝐱) − 𝑢(𝐱𝐼 )] , (43.3)
𝐼=1 𝑖=1

where NP is the number of nodes within the support of 𝐱 for which the weighting function
𝑤𝑎 (𝐱 − 𝐱𝐼 ) ≠ 0.

Equation (39.3) can be written in the form


𝐽 = (𝐇𝐛 − 𝐮)T Wa (𝐱)(𝐇𝐛 − 𝐮), (43.4)
where
𝐮T = (𝑢1 , 𝑢2 , ⋯ 𝑢NP ), (43.5)

40-2 (Element-Free Galerkin) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Element-Free Galerkin

{𝐇(𝐱1 )}T
⎡ ⎤
𝐇 = ⎢⋯ ⎥, (43.6)
T
⎣{𝐇(𝐱NP )} ⎦

{𝐇(x𝑖 )}T = {𝐻1 (𝐱𝑖 ), … 𝐻𝑛 (𝐱𝑖 )}, (43.7)

𝐖a = diag[𝑤𝑎 (𝐱 − 𝐱1 ), ⋯ , 𝑤𝑎 (𝐱 − 𝐱NP )]. (43.8)

To find the coefficients 𝐛 we obtain the extremum of 𝐽 by


∂𝐽
= 𝐌[n] (𝐱)𝐛(𝐱) − 𝐁(𝐱)𝐮 = 0, (43.9)
∂b
where 𝐌[𝑛] (𝐱) is called the moment matrix of 𝑤𝑎 (𝐱 − 𝐱𝐼 ) and is given by
𝐌[n] (𝐱) = 𝐇T 𝐖a (𝐱)𝐇, (43.10)

𝐁(𝐱) = 𝐇T 𝐖a (𝐱). (43.11)


So we have
−1
𝐛(𝐱) = 𝐌[n] (𝐱)𝐁(𝐱)𝐮. (43.12)

For 𝐌[𝑛] (𝐱) to be invertible, the support of 𝑤𝑎 (𝐱 − 𝐱) needs to be greater than a


minimum size that is related to the order of basis functions. Using the solution of
Equations (43.1), (43.10), (43.11) and (43.12), the EFG approximation is obtained by
NP
𝑢ℎ (𝐱) = ∑ Ψ𝐼 (𝐱)𝑢𝐼 , (43.13)
𝐼=1

where the EFG shape functions Ψ𝐼 (𝐱) are given by


−1
Ψ𝐼 (𝐱) = 𝐇T (𝐱)𝐌[n] (𝐱)𝐁(𝐱), (43.14)
and 𝚿𝐼 (𝐱) are nth-order complete, i.e.
𝑁𝑃
𝑝 𝑞 𝑝 𝑞
∑ Ψ𝐼 (𝐱)𝑥1𝐼 𝑥2𝐼 = 𝑥1 𝑥2 for 𝑝 + 𝑞 = 0, ⋯ 𝑛. (43.15)
𝐼=1

43.2 Integration constraint and strain smoothing


The convergence of the Galerkin method for a partial differential equation is
determined by approximation for the unknowns and the numerical integration of the weak
form. EFG shape functions with linear consistency can be obtained from MLS
approximation with linear basis functions. The employment of linearly consistent mesh-

LS-DYNA DEV 10/27/16 (r:8004) 40-3 (Element-Free Galerkin)


Element-Free Galerkin LS-DYNA Theory Manual

free shape functions in the Galerkin approximation, however, does not guarantee a linear
exactness in the solution of the Galerkin method. It has been shown by Chen et al. [2001]
that two integration constraints are required for the linear exactness solution in the
Galerkin approximation.
NIT
∑ ∇ΨI (x̂L )AL = 0 for {𝐼: supp(Ψ𝐼 ) ∩ Γ = 0}, (43.16)
𝐿=1

NIT NITh
∑ ∇ΨI (x̂L )AL = ∑ nΨ𝐼 (x̃𝐿 )𝑠𝐿 for {𝐼: supp(Ψ𝐼 ) ∩ Γℎ ≠ 0}. (43.17)
𝐿=1 𝐿=1

where Γℎ is the natural boundary, Γ is the total boundary, 𝐧 is the surface normal on Γℎ , x̂𝐿
and 𝐴𝐿 are the spatial co-ordinate and weight of the domain integration point, respectively,
x̃𝐿 and 𝑠𝐿 are the spatial co-ordinate and weight of the domain of natural boundary
integration point, respectively, NIT is the number of integration points for domain
integration and NITh is the number of integration points for natural boundary integration.

A strain smoothing method proposed by Chen and Wu [1998] as a regularization for


material instabilities in strain localization was extended in their nodal integration method
[Chen et al, 2001] to meet the integration constraints. Here, we adopt the similar concept
for the domain integration. If starts with a strain smoothing at the representative domain
of a Gauss point by
1
̃ 𝑢ℎ𝑖 (x𝐿 ) =
∇ ∫ ∇𝑢ℎ (x ) 𝑑Ω, 𝐴𝐿 = ∫ 𝑑Ω, (43.18)
𝐴𝐿 Ω𝐿 𝑖 𝐿 Ω𝐿

̃ is the smoothed gradient


where Ω𝐿 is a representative domain at each Guass point and ∇
operator. By applying divergence theorem to Equation (43.18) to yield
1
̃ 𝑢ℎ𝑖 (x𝐿 ) =
∇ ∫ n𝑢ℎ (x ) 𝑑Γ, (43.19)
𝐴𝐿 Γ𝐿 𝑖 𝐿
where Γ𝐿 is the boundary of the representative domain of Guass point L. Introducing EFG
shape functions into Equation (25.22) yields
1
̃ 𝑢ℎ𝑖 (x𝐿 ) = ∑
∇ ̃ Ψ𝐼 (x𝐿 ) ⋅ 𝑑𝑖𝐼 .
∫ Ψ (x)n 𝑑Γ ⋅ 𝑑𝑖𝐼 ≡ ∑ ∇ (43.20)
𝐼
𝐴𝐿 Γ𝐿 𝐼 𝐼

It can be shown that the smoothed EFG shape function gradient ∇ ̃ Ψ𝐼 (xL ) meets the
integration constraints in Equations (43.16) and (43.17) regardless of the numerical
integration employed.

40-4 (Element-Free Galerkin) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Element-Free Galerkin

43.3 Lagrangian strain smoothing for path-dependent


problems
To avoid the tensile instability caused by the Eulerian kernel functions, the
Lagrangian kernel functions are implemented in the current LS-DYNA.

To introduce the Lagrangian EFG shape function into the approximation of a path-
dependent problem, the strain increment Δ𝑢𝑖,𝑗 is computed by
∂Δ𝑢𝑖 ∂Δ𝑢𝑖 −1 −1
Δ𝑢𝑖,𝑗 = = 𝐹 = Δ𝐹𝑖𝑘 𝐹𝑘𝑗 . (43.21)
∂𝑥𝑗 ∂𝑋𝑘 𝑘𝑗

The strain smoothing of Δ𝑢𝑖,𝑗 at a material pointx𝐿 is computed by

Δ𝑢̃𝑖,𝑗 (x𝐿 ) = Δ𝐹𝑖𝑘 ̃−1 (x𝐿 ),


̃ (x𝐿 )𝐹𝑘𝑗 (43.22)
where 𝐹𝑖𝑗̃ (x𝐿 ) is the Langrangian strain smoothing of deformation gradient and is given by
1
𝐹𝑖𝑗̃ (x𝐿 ) = ∫ 𝑢ℎ 𝑁 𝑑Γ + δ𝑖𝑗 . (43.23)
𝐴𝐿 Γ𝐿 𝑖 𝑗

43.4 Galerkin approximation for explicit dynamic


computation
The strong form of the initial/boundary value problem for elasto-dynamics is as
follows:
ρ𝐮̈ = ∇ ⋅ 𝛔 + 𝐟b in Ω, (43.24)
with the divergence operator ∇, the body force 𝐟𝑏 , mass density ρ ,and with the boundary
conditions:
𝐮 = 𝐮0 on Γ𝑢
(43.25)
𝛔 ⋅ 𝐧 = 𝐡 on Γℎ ,
and initial conditions
𝐮(𝐗, 0) = 𝐮0 (𝐗)
(43.26)
𝐮̇(X, 0) = 𝐮̇0 (𝐗).

To introduce the Lagrangian strain smoothing formulation into the Galerkin


approximation, an assumed strain method is employed. The corresponding weak form
becomes:

LS-DYNA DEV 10/27/16 (r:8004) 40-5 (Element-Free Galerkin)


Element-Free Galerkin LS-DYNA Theory Manual

∫ ρδ𝐮 ⋅ 𝐮̈dΩ + ∫ δ𝛆̃ : 𝛔dΩ = ∫ δ𝐮 ⋅ 𝐟b dΩ + ∫ δ𝐮 ⋅ 𝐡dΓ. (43.27)


Ω𝑥 Ω𝑥 Ω𝑥 Γℎ

Following the derivation for explicit time integration, the equations to be solved
have the form
δ𝐮T 𝐌𝐮̈ = δ𝐮T 𝐑, (43.28)
where
̈ 𝑑2𝐼̈ , 𝑑3𝐼̈ ]𝑇
𝐮̈𝐼 = [𝑑1𝐼,

𝑀𝐼𝐽 = ∫ ρΨ𝐼 (𝐱)Ψ𝐽 (𝐱)𝑑Ω = ∫ ρ0 Ψ𝐼 (𝐗)Ψ𝐽 (𝐗)𝑑Ω


Ω𝑥 Ω𝑋 (43.29)
̃ 𝑇𝐼 (𝐱) ⋅ 𝛔(𝐅̃ )𝑑Ω − [Ψ𝐼 (x)𝐡]∣ Γ − ∫ Ψ𝐼 (x)𝐟b 𝑑Ω,
R𝐼 = ∫ 𝐁 ℎ
Ω𝑥 Ω𝑥

where B̃𝑇𝐼 (x) is the smoothed gradient matrix obtained from Equation (43.22), 𝑑𝑖𝐼 is the
coefficient of the approximation or the “generalized” displacement.

43.5 Imposition of essential boundary condition


In general, mesh-free shape functions Ψ𝐼 do not possess Kronecker delta properties
of the standard FEM shape functions, i.e.
Ψ𝐼 (xJ ) ≠ δ𝐼𝐽 . (43.30)

This is because, in general, the mesh-free shape functions are not interpolation
functions. As a result, a special treatment is required to enforce essential boundary
conditions. There are many techniques for mesh-free methods to impose the essential
boundary condition. Here, we adopt the transformation method as originally proposed for
the RKPM method by Chen et al. [1996].

Therefore, to impose the essential boundary conditions using kinematically


admissible mesh-free shape functions by the transformation method, Equation (43.28) can
be written as
δ𝐮̂T 𝐌
̂ 𝐮̈ = δ𝐮̂T 𝐅̂ int , (43.31)
where
𝐮̂ = 𝐀𝐮; 𝐴𝐼𝐽 = Ψ𝐽 (𝑋𝐼 ). (43.32)
or
𝐮 = 𝐀−1 𝐮̂, (43.33)

40-6 (Element-Free Galerkin) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Element-Free Galerkin

and
̂ = 𝐀−T 𝐌𝐀−1 ; 𝐅̂ int = 𝐀−T 𝐅int .
𝐌 (43.34)

A mixed transformation method [Chen et al. 2000] is also considered as an


alternative to impose the essential boundary conditions. The mixed transformation method
is an improved transformation method that the coordinate transformation is only applied
for the degrees of freedom associated with the essential and contact boundaries.

The nodes are partitioned into three groups: a boundary group 𝐺𝐵1 which contains
all the nodes subjected to kinematic constraints; group 𝐺𝐵2 which contains all the nodes
whose kernel supports cover nodes in group 𝐺𝐵1 ; and internal group 𝐺𝐼 which contains the
rest of nodes. Nodes numbers are re-arranged in the following order in the generalized
displacement vector:
u𝐵1
𝐮 = ⎢u𝐵2 ⎤

⎥ (43.35)
⎣u𝐼 ⎦
where 𝐮𝐵1 , 𝐮𝐵2 and 𝐮𝐼 are the generalized displacement vectors associated with groups 𝐺𝐵1 ,
𝐺𝐵2 and 𝐺𝐼 respectively. The transformation in Equation (43.32) is also re-arranged as
𝐵 BB
𝐮̂ = [𝐮̂𝐼 ] [Λ IB Λ BI ] [𝐮𝐵 ] ≡ 𝚲
̂ 𝐮, (43.36)
𝐮̂ Λ Λ II 𝐮𝐼
where
𝐵1 𝐵1 B1 B1
𝐮̂B = [û 𝐵 ] ; 𝐮B = [u𝐵 ] ; 𝚲BB = [Λ B 𝐵 Λ B1B2 ] ; 𝚲BI = [0 ] ; 𝚲IB
û 2 u 2 Λ 2 1 Λ B2B2 Λ𝐵2𝐼 (43.37)
= [Λ𝐼𝐵1 Λ𝐼𝐵2 ].

Here, we introduce a mixed displacement vector 𝐮∗ ,


𝐵
𝐮∗ = [𝐮̂𝐼 ] [Λ
BB
Λ BI ] [𝐮𝐵 ] ≡ 𝚲∗ 𝐮, (43.38)
𝐮 0 I 𝐮𝐼
and Λ ∗ and its inverse are:
BB −1 −1
𝚲 ∗ = [Λ Λ BI ] ; 𝚲∗−1 = [Λ BB −Λ BB ΛBI ]. (43.39)
0 I 0 I
Only the inversion of Λ 𝐵𝐵 is required in Equation (22.83.15).

Using the mixed coordinates in Equation (43.38), the transformed discrete Equation
(43.31) becomes
T T
δ𝐮∗ 𝐌∗ 𝐮̈∗ = δ𝐮∗ 𝐑∗ , (43.40)
where

LS-DYNA DEV 10/27/16 (r:8004) 40-7 (Element-Free Galerkin)


Element-Free Galerkin LS-DYNA Theory Manual

−T −1 −T
𝐌∗ = 𝐀∗ 𝐌𝐀∗ ; 𝐑 ∗ = 𝐀∗ 𝐑. (43.41)

The computation in Equations (43.41) is much less intensive than that in Equation
(43.31), especially when the number of boundary and contact nodes is much smaller than
the number of interior nodes.

43.6 Mesh-free Shell


The extension of explicit mesh-free solid analysis to shell analysis is described in this
section. Two projection methods are developed to generate the shell mid-surface using the
moving-least-squares approximations. A co-rotational, updated Lagrangian procedure is
adopted to handle arbitrarily large rotations with moderate strain responses of the shell
structures. A local boundary integration method in conjunction with the selective reduced
integration method is introduced to enforce the linear exactness and relieve shear locking.

43.6.1 Mesh-free Shell Surface Representation


Surface reconstruction from disorganized nodes is very challenging in three
dimensions. The problem is ill posed, i.e., there is no unique solution. Lancaster et al.
[1981] first proposed a fast surface reconstruction using moving least squares method.
Their approach was then applied to the computational mechanics under the name ‘mesh-
free method’. Implicitly, the mesh-free method uses a combination of smooth basis
functions (primitives) to find a scalar function such that all data nodes are close to an iso-
contour of that scalar function in a global sense. In reality, the shell surface construction
using the 3D mesh-free method is inadequate. This is because the topology of the real
surface can be very complicated in three dimensions. Without the information on the
ordering or connectivity of nodes, the reconstructed surface will not be able to represent
shell intersections, exterior boundaries and shape corners.

In our development of mesh-free shells, we assume that a shell surface is described


by a finite element mesh. This can be easily accomplished by converting a part of shell
finite elements into mesh-free zone. With the connectivity of nodes provided by the finite
element mesh, a shell surface can be reconstructed with mesh-free interpolation from the
nodal positions
̃𝐼 (𝐗)𝐱𝐼 ,
𝐱̅ = Ψ (43.42)
̃𝐼 (𝐗) is
where 𝐱𝐼 is the position vector of the finite element node on the shell surface and Ψ
the mesh-free shape function. In the above surface representation, a 3D arbitrary shell
surface needs to be projected to a 2D plane. Two approaches for the projection of mesh-
free shell surface are used:

40-8 (Element-Free Galerkin) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Element-Free Galerkin

Projection
ξ

Figure 43.2. Mesh-free shell global approach

• Global parametric representation: The whole shell surface is projected to a paramet-


ric plane and the global parametric coordinates are obtained with a parameteriza-
tion algorithm from the patch of finite elements.
• Local projection representation: A local area of the shell is projected to a plane based
on the existing element where the evaluated point is located.

Global parametric approach


In the global approach, a mesh-free zone with a patch of finite elements is mapped
onto a parametric plane with an angle-based triangular flattening algorithm [Sheffer and de
Sturler 2001], (see Figure 43.2). The idea of this algorithm is to compute a projection that
minimizes the distortion of the FE mesh angles. The mesh-free shape functions are defined
in this parametric domain and given by
̃𝐼 (𝐗) = Ψ
Ψ ̃𝐼 (𝜉 , 𝜂), (43.43)
where (𝜉 , 𝜂) is the parametric coordinates corresponding to a point X.

Local projection approach


Different from the parameterization algorithm that constructs the surface globally,
we reconstruct the surface locally by projecting the surrounding nodes onto one element.
In the local projection method, nodes in elements neighboring the element where the
evaluated point is located (for example, the element i in Figure 43.3) are projected onto the
plane which the element defines (the “M-plane” in Figure 43.3). In this figure, (𝑥̂, 𝑦̂, 𝑧̂)𝑖 is a
local system defined for each projected plane and (𝑥̅, 𝑦̅, 𝑧̅)𝐼 is a nodal coordinate system
defined for each node where 𝑧̅ is the initial averaged normal direction.

The mesh-free shape functions are then defined with those locally projected
coordinates of the nodes
Ψ𝐼 (𝐗) = Ψ𝐼 (𝑥̂, 𝑦̂). (43.44)
However, the shape functions obtained directly above are non-conforming, i.e.

LS-DYNA DEV 10/27/16 (r:8004) 40-9 (Element-Free Galerkin)


Element-Free Galerkin LS-DYNA Theory Manual

^
zi
z¯ I
y¯ I ^i
y

i M-plane
M-plane
I x¯ I ^xi
J J K

Figure 43.3. Mesh-free shell local projection

Ψ𝐼 (𝐗𝐽 )∣M−plane ≠ Ψ𝐼 (𝐗𝐽 )∣N−plane . (43.45)

When the shell structure degenerates to a plate, the constant stress condition cannot
be recovered. To remedy this problem, an area-weighed smoothing across different
projected planes is used to obtain the conforming shape functions that are given by
NIE
∑𝑖=1 Ψ𝐼 (𝑥̂𝑖 , 𝑦̂𝑖 )𝐴𝑖
̃𝐼 (𝐗) = Ψ
Ψ ̃𝐼 (𝑥̂, 𝑦̂) = . (43.46)
𝑁𝐼𝐸
∑𝑖=1 𝐴𝑖
where NIE is the number of surrounding projected planes that can be evaluated at point X,
𝑨𝒊 is the area of the element 𝑖, and (𝑥̂𝑖 , 𝑦̂𝑖 ) is the local coordinates of point X in the projected
plane 𝑖.
With this smoothing technique, we can prove that the modified shape functions satisfy at
least the partition of unity property in the general shell problems. This property is
important for the shell formulation to preserve the rigid-body translation.

When the shell degenerates to a plate, we can also prove that the shape functions
obtained from this smoothing technique will meet the n-th order completeness condition as
NP
𝑖 𝑘𝑗 𝑗
̃𝐼 (𝐗)𝑋1𝐼
∑Ψ 𝑋2𝐼 𝑋3𝐼 = 𝑋1𝑖 𝑋2 𝑋3𝑘 , 𝑖 + 𝑗 + 𝑘 = 𝑛. (43.47)
𝐼=1

This is a necessary condition for the plate to pass the constant bending patch test.

43.6.2 Updated Lagrangian Formulation and Co-rotational Procedure


The mesh-free shell formulation is based on the Mindlin-Reissner plate theory, thus
the geometry and kinematical fields of the shell can be described with the reference surface
and fiber direction. The modified Mindlin-Reissner assumption requires that the motion
and displacement of the shell are linear in the fiber direction. Assume that the reference

40-10 (Element-Free Galerkin) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Element-Free Galerkin

V3 ζ
^
z
y^

X x^

z

y
x
Figure 43.4. Geometry of a shell.

surface is the mid-surface of the shell, the global coordinates and displacements at an
arbitrary point within the shell body are given by

𝐱 = 𝐱̅ + ζ 𝐕3 , (43.48)
2

𝐮=𝐮 ̅̅̅̅ + ζ 𝐔. (43.49)
2
where 𝐱̅ and 𝐮 ̅̅̅̅ are the position vector and displacement of the reference surface,
respectively. 𝐕3 is the fiber director and 𝐔 is the displacement resulting from the fiber
rotation (see Figures 43.4 and 43.5). ℎ is the length of the fiber.

With the mesh-free approximation, the motion and displacements are given by
𝑁𝑃 𝑁𝑃
𝜁 ℎ𝐼
̃𝐼 (𝜉 , 𝜂)𝐱𝐼 + ∑ Ψ
𝐱(𝜉 , 𝜂, 𝜁 ) = 𝐱̅(𝜉 , 𝜂) + 𝐕(𝜉 , 𝜂, 𝜁 ) ≈ ∑ Ψ ̃𝐼 (𝜉 , 𝜂) 𝐕 , (43.50)
𝐼=1 𝐼=1
2 3𝐼

Initial configuration
u U
V03
X
x V3
X̄ ū


Deformed Configuration
Figure 43.5. Deformation of a shell.

LS-DYNA DEV 10/27/16 (r:8004) 40-11 (Element-Free Galerkin)


Element-Free Galerkin LS-DYNA Theory Manual

𝑁𝑃 𝑁𝑃 𝛼
𝜁 ℎ𝐼
𝐮(𝜉 , 𝜂, 𝜁 ) = 𝐮 ̃𝐼 (𝜉 , 𝜂)𝐮𝐼 + ∑ Ψ
̅̅̅̅(𝜉 , 𝜂) + 𝐔(𝜉 , 𝜂, 𝜁 ) ≈ ∑ Ψ ̃𝐼 (𝜉 , 𝜂) [−𝐕2𝐼 𝐕1𝐼 ] {𝛽𝐼 },(43.51)
𝐼=1 𝐼=1
2 𝐼

where 𝐱𝐼 and 𝐮𝐼 are the global coordinates and displacements at mesh-free node 𝐼,
respectively. 𝐕3𝐼 is the unit vector of the fiber director and 𝐕1𝐼 , 𝐕2𝐼 are the base vectors of
the nodal coordinate system at node 𝐼. 𝛼𝐼 and 𝛽𝐼 are the rotations of the director vector 𝐕3𝐼
about the 𝐕1𝐼 and 𝐕2𝐼 axes. ℎ𝐼 is the thickness. The variables with a superscripted bar refer
̃𝐼 is the 2D mesh-free shape functions constructed based on one
to the shell mid-surface. Ψ
of the two mesh-free surface representations described in the previous section, with (𝜉 , 𝜂)
either the parametric coordinates or local coordinates of the evaluated point.

The local co-rotational coordinate system (𝑥̂, 𝑦̂, 𝑧̂) is defined at each integration point
on the shell reference surface, with 𝑥̂ and 𝑦̂ tangent to the reference surface and 𝑧̂ in the
thickness direction (see Figure 43.6). The base vectors are given as
𝐱,ξ 𝐱,ξ × 𝐱,η
𝐞̂1 = , 𝐞̂3 = , 𝐞̂2 = 𝐞̂3 × 𝐞̂1 . (43.52)
∥𝐱,ξ ∥ ∥𝐱,ξ × 𝐱,η ∥

In order to describe the fiber rotations of a mesh-free node in a shell, we introduce a


nodal coordinate system whose three base vectors are 𝐕1 , 𝐕2 and 𝐕3 , see Figure 43.6,
where 𝐕3 is the fiber director at the node and 𝐕1 , 𝐕2 are defined as follows
𝐱̂ × 𝐕3
𝐕1 = , 𝐕 = 𝐕3 × 𝐕1 . (43.53)
∣𝐱̂ × 𝐕3 ∣ 2

The rotation of the fiber director is then obtained from the global rotations:
𝛼 𝐕T
{𝛽} = [ 1T ] Δθ, Δθ = [Δ𝜃1 Δ𝜃2 Δ𝜃3 ]T . (43.54)
𝐕 2

40-12 (Element-Free Galerkin) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Element-Free Galerkin

z^ z^s y^s
V3 y^
V2

x^ s
V1
x^

z
y

x
Figure 43.6. Local co-rotational and nodal coordinate systems.

In the local co-rotational coordinate system, the motion and displacements are
approximated by the mesh-free shape functions
NP NP
hI
̃I x̂iI + ζ ∑ Ψ
x̂i = ∑ Ψ ̃I 𝑉̂ , (43.55)
I=1 I=1
2 3𝑖𝐼

NP NP
hI α
û i = ∑ Ψ ̃I
̃I ûiI + ζ ∑ Ψ ̂ 2iI
[−V ̂ 1iI ] { I }.
V (43.56)
2 β I
I=1 I=1

The Lagrangian smoothed strains [Chen et al. 2001b] are given by


̃m
ε̃m = ∑ 𝐁 ̂ b ̃b ̂ s ̃s ̂
I 𝐝I , ε̃ = ζ ∑ 𝐁I 𝐝I , ε̃ = ∑ 𝐁I 𝐝I , (43.57)
I I I

where the smoothed strain operators are calculated by averaging the consistent strain
operators over an area around the evaluated point
1 1 1
̃m
𝐁 𝐼 (𝐱𝑙 ) = ∫ 𝐁 ̂m ̃b
𝐼 𝑑𝐴 , 𝐁𝐼 (𝐱𝑙 ) = ∫ 𝐁 ̂ b𝐼 𝑑𝐴 , 𝐁
̃ s𝐼 (𝐱𝐿 ) = ∫ 𝐁 ̂ s𝐼 𝑑𝐴, (43.58)
𝐴𝑙 Ω𝑙 𝐴𝑙 Ω𝑙 𝐴𝐿 Ω𝐿
with
̂m
𝐁 𝐼

−1 ̃ ℎ𝐼 −1 ̃ ℎ𝐼
⎡Ψ̃𝐼,𝑥 0 0 −𝐽13 Ψ𝐼 𝑉̂ 𝐽13 Ψ𝐼 𝑉̂ ⎤
⎢ 2 2𝑥𝐼 2 1𝑥𝐼 ⎥
⎢ ⎥ (43.59)
=⎢ ̃𝐼,𝑦 −1 ̃ ℎ𝐼 ̂ −1 ̃ ℎ𝐼 ̂ ⎥,
⎢0 Ψ 0 −𝐽23 Ψ𝐼 𝑉2𝑦𝐼
2
𝐽23 Ψ𝐼 𝑉1𝑦𝐼
2 ⎥
⎢ ⎥
⎢ −1 ̃ ℎ𝐼 ̂ −1 ̃ ℎ𝐼 ̂ −1 ̃ ℎ𝐼 ̂ −1 ̃ ℎ𝐼 ̂ ⎥
̃ ̃𝐼,𝑥
⎣Ψ𝐼,𝑦 Ψ 0 −𝐽23 Ψ𝐼 𝑉2𝑥𝐼 − 𝐽13
2
Ψ𝐼 𝑉2𝑦𝐼
2
𝐽23 Ψ𝐼 𝑉1𝑥𝐼 + 𝐽13 Ψ𝐼 𝑉1𝑦𝐼 ⎦
2 2

LS-DYNA DEV 10/27/16 (r:8004) 40-13 (Element-Free Galerkin)


Element-Free Galerkin LS-DYNA Theory Manual

Γl

Ωl ζ
ξ
xl η
ΓL

xL

Figure 43.7. Integration scheme for mesh-free shells.

ℎ𝐼 ℎ𝐼
⎡0 0 0 ̃𝐼,𝑥
−Ψ 𝑉̂ ̃𝐼,𝑥
Ψ 𝑉̂ ⎤
⎢ 2 2𝑥𝐼 2 1𝑥𝐼 ⎥
b
⎢ ℎ ℎ𝐼 ⎥
̂ 𝐼 = ⎢0
𝐁 0 0 ̃𝐼,𝑦 𝐼 𝑉̂2𝑦𝐼
−Ψ ̃𝐼,𝑦 𝑉̂1𝑦𝐼
Ψ ⎥, (43.60)
⎢ 2 2 ⎥
⎢ ⎥
⎢ ℎ𝐼 ℎ ℎ𝐼 ℎ𝐼 ⎥
̃𝐼,𝑥 𝐼 𝑉̂2𝑦𝐼
̃𝐼,𝑦 𝑉̂2𝑥𝐼 − Ψ ̃ ̂ ̃ ̂
⎣0 0 0 −Ψ
2 2
Ψ𝐼,𝑦 𝑉1𝑥𝐼 + Ψ𝐼,𝑥 𝑉1𝑦𝐼 ⎦
2 2

−1 ̃ ℎ𝐼 −1 ̃ ℎ𝐼 ̂ −1 ̃ ℎ𝐼 −1 ̃ ℎ𝐼 ̂
⎡0 0 ̃𝐼,𝑦
Ψ −𝐽33 Ψ𝐼 𝑉̂2𝑦𝐼 − 𝐽23 Ψ𝐼 𝑉2𝑧𝐼 𝐽33 Ψ𝐼 𝑉̂1𝑦𝐼 + 𝐽23 Ψ𝐼 𝑉1𝑧𝐼 ⎤
̂ s𝐼 = ⎢
𝐁 2 2 2 2 ⎥, (43.61)

⎢ ⎥

̃𝐼,𝑥 −1 ̃ ℎ𝐼 ̂ −1 ̃ ℎ𝐼 ̂ −1 ̃ ℎ𝐼 ̂ −1 ̃ ℎ𝐼 ̂
⎣0 0 Ψ −𝐽33 Ψ𝐼 𝑉2𝑥𝐼 − 𝐽13 Ψ𝐼 𝑉2𝑧𝐼
2 2
𝐽33 Ψ𝐼 𝑉1𝑥𝐼 + 𝐽13 Ψ𝐼 𝑉1𝑧𝐼 ⎦
2 2
and 𝐉−1 is the inverse of the Jacobian matrix at the integration point. The local degrees-of-
freedom are
𝐝̂ 𝐼 = [𝑢̂𝑥𝐼 𝑢̂𝑦𝐼 𝑢̂𝑧𝐼 𝛼𝐼 𝛽𝐼 ]T . (43.62)
The internal nodal force vector is
T T
𝐅̂Iint = ∫ 𝐁
̃m ̃b ̃sT
I σ̂ dΩ + ∫ ζ𝐁I σ̂ dΩ + ∫ 𝐁I σ̂ dΩ. (43.63)
Ω Ω Ω

The above integrals are calculated with the local boundary integration method.
Each background finite element is divided into four integration zones, shown as Ω𝑙 in
Figure 43.7. In order to avoid shear locking in the analysis of thin shells, the shear term
(third term in Eq. (43.63)), should be under-integrated by using one integration zone in
each background element (Ω𝐿 in Figure 43.7). Accordingly, the co-rotational coordinate
systems are defined separately at the center of each integration zone, as shown in Figure
43.6.

The use of the updated Lagrangian formulation implies that the reference coordinate
system is defined by the co-rotational system in the configuration at time t. Therefore, the
40-14 (Element-Free Galerkin) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Element-Free Galerkin

local nodal force and displacement vectors referred to this coordinate system must be
transformed to the global coordinate system prior to assemblage.

LS-DYNA DEV 10/27/16 (r:8004) 40-15 (Element-Free Galerkin)


LS-DYNA Theory Manual Linear shells

44
Linear shells

44.1 Shells for Linear Analysis


It is common to construct elements for linear analysis by the superimposition of a
plate and a membrane element. If the base plates and membrane elements involve only
three translational degrees-of-freedom and two in-plane rotational degrees-of-freedom, the
resulting element then contains 5 degrees-of-freedom per node since there is an
unconstrained rotational degree-of-freedom normal to the mid surface of the shell. This
unconstrained mode can cause problems when linking the shell to other elements such as
beam elements in three-dimensional space. For this reason, the linear elements in LS-
DYNA are based on published formulations that include a drilling degree-of-freedom,
which is added to the membrane part of the element to form a 24 degree-of-freedom shell
element. These elements pass all patch tests, have 6 rigid body modes, and have no
spurious mechanisms.

44.2 Wilson’s Shell (element #20)


This quadrilateral element is constructed as described above and is discussed in
more detail by Wilson [2000]. The triangular element, which is an 18 degree-of-freedom
complement to the quadrilateral elements, follows the same procedure. In a linear analysis
in LS-DYNA, automatic sorting is invoked if a mesh has both quadrilateral and triangular
elements within a single part ID. This sorting ensures the proper treatment of triangles.

44.2.1 Plate Element


The 4 node quadrilateral plate element is based on the 8 node, quadratic
quadrilateral plate element, which has 16 rotational degrees-of-freedom, i.e., two per nodal
point. The implementation in LS-DYNA directly follows the textbook by Wilson [2000]

LS-DYNA DEV 10/27/16 (r:8004) 41-1 (Linear shells)


Linear shells LS-DYNA Theory Manual

Plate

Membrane

Figure 44.1. Shell as assembly of plate and membrane elements

where the complete details of the element are provided. A condensed overview is given
here. The shell theory makes the following assumptions:
The fiber remains straight and inextensible

The normal stress in the thickness direction is zero

The local x and y rotations of the shell are interpolated from the equations:
4 8
𝜃𝑥 (𝑟, 𝑠) = ∑ 𝑁𝑖 (𝑟, 𝑠)𝜃𝑥𝑖 + ∑ 𝑁𝑖 (𝑟, 𝑠)Δ𝜃𝑥𝑖
𝑖=1 𝑖=5
(44.1)
4 8
𝜃𝑦 (𝑟, 𝑠) = ∑ 𝑁𝑖 (𝑟, 𝑠)𝜃𝑦𝑖 + ∑ 𝑁𝑖 (𝑟, 𝑠)Δ𝜃𝑦𝑖 ,
𝑖=1 𝑖=5

where nodes 5-8 are at the mid side of the element. The interpolation functions are given
by
1 1
𝑁1 = (1 − 𝑟)(1 − 𝑠) 𝑁5 = (1 − 𝑟2 )(1 − 𝑠)
4 2
1 1
𝑁2 = (1 + 𝑟)(1 − 𝑠) 𝑁6 = (1 + 𝑟)(1 − 𝑠2 )
4 2
(44.2)
1 1
𝑁3 = (1 + 𝑟)(1 + 𝑠) 𝑁7 = (1 − 𝑟2 )(1 + 𝑠)
4 2
1 1
𝑁4 = (1 − 𝑟)(1 + 𝑠) 𝑁8 = (1 − 𝑟)(1 − 𝑠2 ).
4 2
In his formulation, Wilson resolves the rotation of the mid side node into tangential and
normal components relative to the shell edges. The tangential component is set to zero
leaving the normal component as the unknown, which reduces the rotational degrees-of-
freedom from 16 to 12, see Figure 44.2.
Δ𝜃𝑥 = sin𝛼𝑖𝑗 Δ𝜃𝑖𝑗
(44.3)
Δ𝜃𝑦 = −cos𝛼𝑖𝑗 Δ𝜃𝑖𝑗

41-2 (Linear shells) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Linear shells

Θj
ΔΘy

ΔΘx
i = 1, 2, 3, 4
ΔΘij j = 2, 3, 4, 1
αij m =5, 6, 7, 8

Θi
Figure 44.2. Element edge [Wilson, 2000]

4 8
𝜃𝑥 (𝑟, 𝑠) = ∑ 𝑁𝑖 (𝑟, 𝑠)𝜃𝑥𝑖 + ∑ 𝑀𝑥𝑖 (𝑟, 𝑠)Δ𝜃𝑖
𝑖=1 𝑖=5
(44.4)
4 8
𝜃𝑦 (𝑟, 𝑠) = ∑ 𝑁𝑖 (𝑟, 𝑠)𝜃𝑦𝑖 + ∑ 𝑀𝑦𝑖 (𝑟, 𝑠)Δ𝜃𝑖 .
𝑖=1 𝑖=5

Ultimately, the 4 mid side rotations are eliminated by using static condensation, a
procedure that makes this shell very costly if used in explicit calculations.

The local 𝑥 and 𝑦 displacements relative to the mid surface are functions of the 𝑧-
coordinate and rotations:
𝑢𝑥 (𝑟, 𝑠) = 𝑧𝜃𝑦 (𝑟, 𝑠)
(44.5)
𝑢𝑦 (𝑟, 𝑠) = −𝑧𝜃𝑥 (𝑟, 𝑠).

Wilson shows, where it is assumed that the normal displacement along each side is
cubic, that the transverse shear strain along each side is given by,
1 1 2
𝛾𝑖𝑗 = (𝑢𝑧𝑗 − 𝑢𝑧𝑖 ) − (𝜃𝑖 + 𝜃𝑗 ) − Δ𝜃𝑖𝑗 , (44.6)
𝐿 2 3
which can be rewritten, referring to Figure 44.3 as:
1 sin𝛼𝑖𝑗 cos𝛼𝑖𝑗 2
𝛾𝑖𝑗 = (𝑢𝑧𝑗 − 𝑢𝑧𝑖 ) − (𝜃𝑥𝑖 + 𝜃𝑥𝑗 ) + (𝜃𝑦𝑖 + 𝜃𝑦𝑗 ) − Δ𝜃𝑖𝑗 , (44.7)
𝐿 2 2 3

The nodal shears are then written in terms of the side shears as
𝛾 cos𝛼𝑖𝑗 sin𝛼𝑖𝑗 𝛾𝑥𝑧
[𝛾𝑖𝑗 ] = [ ] [ ], (44.8)
𝑘𝑖 cos𝛼𝑘𝑖 sin𝛼𝑘𝑖 𝛾𝑦𝑧
which can be inverted to obtain the nodal shears:

LS-DYNA DEV 10/27/16 (r:8004) 41-3 (Linear shells)


Linear shells LS-DYNA Theory Manual

y
αki Θy
Θx
k
j

γki γij
ΔΘki
i = 1, 2, 3, 4
ΔΘij j = 2, 3, 4, 1
γxz αij m =5, 6, 7, 8

i γyz x

Figure 44.3. Nodal and edge shear strains [Wilson 2000].

𝛾𝑥𝑧 1 sin𝛼𝑘𝑖 −cos𝛼𝑘𝑖 𝛾𝑖𝑗


[𝛾 ] = [−sin𝛼 cos𝛼𝑖𝑗 ] [𝛾𝑘𝑖 ]. (44.9)
𝑦𝑧 cos𝛼𝑖𝑗 sin𝛼𝑘𝑖 − cos𝛼𝑘𝑖 sin𝛼𝑖𝑗 𝑖𝑗

The standard bilinear basis functions are used to interpolate the nodal shears to the
integration points.

44.2.2 Membrane Element


The membrane element, which is also coded from Wilson’s textbook [2000], is based
on the eight node isoparametric element, see Figure 44.4.

The inplane displacement field for the 8 node membrane is interpolated, using the
serendipity shape functions with the mid-side relative displacements, from:
4 8
𝑢𝑥 (𝑟, 𝑠) = ∑ 𝑁𝑖 (𝑟, 𝑠)𝑢𝑥𝑖 + ∑ 𝑁𝑖 (𝑟, 𝑠)Δ𝑢𝑥𝑖
𝑖=1 𝑖=5
(44.10)
4 8
𝑢𝑦 (𝑟, 𝑠) = ∑ 𝑁𝑖 (𝑟, 𝑠)𝑢𝑦𝑖 + ∑ 𝑁𝑖 (𝑟, 𝑠)Δ𝑢𝑦𝑖 .
𝑖=1 𝑖=5

It is desired to replace the mid side relative displacment by drilling rotations at the
corner nodes. Consider Figure 44.5: the mid-side normal displacements along the edge are
parabolic, i.e.,
𝐿𝑖𝑗
Δ𝑢𝑖𝑗 = (Δ𝜃𝑗 − Δ𝜃𝑖 ), (44.11)
8

41-4 (Linear shells) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Linear shells

while the mid-side tangential displacements are interpolated linearly from the end node
displacements, thus,
𝐿𝑖𝑗
Δ𝑢𝑥 (𝑟, 𝑠) = cos𝛼𝑖𝑗 Δ𝑢𝑖𝑗 = cos𝛼𝑖𝑗 (Δ𝜃𝑗 − Δ𝜃𝑖 )
8
(44.12)
𝐿𝑖𝑗
Δ𝑢𝑦 (𝑟, 𝑠) = −sin𝛼𝑖𝑗 Δ𝑢𝑖𝑗 = −sin𝛼𝑖𝑗 (Δ𝜃𝑗 − Δ𝜃𝑖 ),
8
4 8
𝑢𝑥 (𝑟, 𝑠) = ∑ 𝑁𝑖 (𝑟, 𝑠)𝑢𝑥𝑖 + ∑ 𝑀𝑥𝑖 (𝑟, 𝑠)Δ𝜃𝑖
𝑖=1 𝑖=5
(44.13)
4 8
𝑢𝑦 (𝑟, 𝑠) = ∑ 𝑁𝑖 (𝑟, 𝑠)𝑢𝑦𝑖 + ∑ 𝑀𝑦𝑖 (𝑟, 𝑠)Δ𝜃𝑖 .
𝑖=1 𝑖=5

This element has one singularity in the drilling mode of equal corner rotations, see
Figure 44.6.

Ibrahimbegovic and Wilson [1991] added a penalty formulation to the potential


energy of the element to eliminate the singularity. The following penalty term connects the
averaged nodal rotation to the continuum mechanics rotation
1 ∂𝑢𝑥 ∂𝑢𝑦
( − )−𝜔 (44.14)
2 ∂𝑦 ∂𝑥

s
6 3
4

7 r
5

1 8 2

s
4 3
r

1 2
Figure 44.4. Eight node membrane element.

LS-DYNA DEV 10/27/16 (r:8004) 41-5 (Linear shells)


Linear shells LS-DYNA Theory Manual

Δuy i = 1, 2, 3, 4
ΔΘj j = 2, 3, 4, 1
Lij m = 5, 6, 7, 8
Δux

Δuij
αij
ΔΘi

Figure 44.5. Corner node drilling rotations and mid side edge normal
displacement [Wilson, 2000].

Figure 44.6. Zero energy mode

at the center of the element. The element performance is highly insensitive to the chosen
value of the penalty factor and some fraction of the elastic modulii, G or E, is frequently
used.
• 5, 8 or 9 point quadrature can be applied. The 5 and 8 point schemes induce a ‘soft’
first deformational mode, whereas the 9 point Gaussian quadrature results in a
stiffer mode.
• A membrane locking correction (Taylor) is applied to (i) alleviate a membrane-
bending interaction associated with the drilling degrees of freedom and (ii) allow

41-6 (Linear shells) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Linear shells

the standard application of the consistent nodal load at the edge. The correction has
a slight stiffening effect (see e.g. Cook Cantilever).
• A warping correction is applied using the rigid link correction (see Figure 44.7).

44.3 Assumed Strain/Membrane with Drilling Degree-of-


freedom (element #18)
44.3.1 Membrane Element
Formulation is the same as above for element type 20.

44.3.2 Plate Element


The Discrete Kirchhoff Quadrilateral element is an excellent thin shell element based
on
• Rotational field is interpolated using the 8-node isoparametric parent element.
• Transverse displacement w assumed as cubic along the sides and collocated along
the sides and at the nodes using the Kirchhoff condition that equates the fiber
rotation to the slope. The Kirchhoff assumptions are satisfied along the entire
boundary of the element.
• The rotational field about an axis parallel to the side is constrained linearly along
the sides.

The warping correction is applied as above.

Flat Element

Figure 44.7. Flat element

LS-DYNA DEV 10/27/16 (r:8004) 41-7 (Linear shells)


Linear shells LS-DYNA Theory Manual

44.4 Differences between Element Types 18 and 20.


The DKQ does not account for transverse shear because it locally enforces the
Kirchhoff condition. Hence, element type 20 is better for layered composites and
thick plates.

41-8 (Linear shells) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Random Geometrical Imperfections

45
Random Geometrical Imperfections

45.1 Introduction to Random Geometrical Imperfections


Using Karhunen-Loève Expansions
There are different methods of incorporating imperfections, depending on the availability
of accurate imperfection data. The method implemented into LS-DYNA v971 uses a
spectral decomposition of geometrical or thickness uncertainty, more specifically through
using Karhunen-Loève expansions (see *PERTURBATION keyword). To specify the
covariance of the random field of the geometrical imperfections or thickness variation, two
methods are available. The first is to use available experimentally-measured imperfection
fields as input for a principal component analysis based on pattern (face) recognition
literature. This method reduces the cost of the resulting eigen-analysis. The second is to
specify the covariance function analytically and to solve the resulting Friedholm integral
equation of the second kind using a wavelet-Galerkin approach, also obtained from
literature. Six different analytical covariance kernels (e.g., exponential and triangular) are
available for selection.

45.2 Methodology
45.2.1 Generation of random fields using Karhunen-Loève expansion
The Karhunen-Loève expansion (e.g., Ghanem and Spanos [2003]) provides an
attractive way of representing a random (stochastic) field (process) through a spectral
decomposition, , as a function of x (e.g., two spatial variables):

𝜛(x, 𝜃) = 𝜛(̅ x) + ∑ √𝜆i 𝜉𝑖 (𝜃)𝑓𝑖 (x), (45.1)
𝑖=1

where the 𝜉𝑖 are uncorrelated zero-mean random variables with unit variance, and 𝜛(̅ x)is
the average random field or mean of the process. The functions 𝑓𝑖 are the eigenfunctions of

LS-DYNA DEV 10/27/16 (r:8004) 42-1 (Random Geometrical Imperfections)


Random Geometrical Imperfections LS-DYNA Theory Manual

the covariance kernel, C, with 𝜆𝑖 the associated eigenvalues, obtained from the spectral
decomposition of the covariance function via the solution on a domain 𝐷 of the Fredholm
integral equation of the second kind,

∫ 𝐶(x1 , x2 )𝑓𝑖 (x1 )𝑑x1 = 𝜆𝑖 𝑓𝑖 (x2 ). (45.2)


𝐷

The eigenfunctions form an orthogonal set

∫ 𝑓𝑖 (x)𝑓𝑗 (x)𝑑x = 𝛿𝑖𝑗 . (45.3)


𝐷

Normally, a finite 𝑀 number of terms are kept in the series expansion:


𝑀
𝜛(x, 𝜃) = 𝜛(̅ x) + ∑ √𝜆i 𝜉𝑖 (𝜃)𝑓𝑖 (x). (45.4)
𝑖=1

If 𝜛 is Gaussian, then 𝜉𝑖 are also Gaussian. For non-Gaussian processes (with arbitrary but
specified marginal distributions), 𝜉𝑖 are unknown. Phoon et al ([2002a], [2005]) give an
iterative procedure for obtaining 𝜉𝑖 given a target marginal distribution.

45.2.2 Solution of Fredholm integral of the second kind for analytical covariance
functions

The Wavelet-Galerkin method (Phoon et al [2002]) is used to perform the solution,


and can be described as follows.

By defining a set of basis functions: 𝜑1 (𝑥), 𝜑2 (𝑥),…,𝜑𝑁 (𝑥), each eigenfunction 𝑓𝑖 (𝑥)
can be approximated by the linear combination:
𝑁
𝑓𝑖 (𝑥) = ∑ 𝑑𝑖𝑘 𝜑𝑘 (𝑥), (45.5)
𝑘=1

where the 𝑑𝑖𝑘 are constant coefficients. By substituting (45.5) into Fredholm equation (45.2)
(using a scalar 𝑥 (one-dimensional random process) as an example) and writing as an error:
𝑁 𝑁
∫ 𝐶(x1 , x2 ) ∑ 𝑑𝑖𝑘 𝜑𝑘 (𝑥1 )𝑑x1 − 𝜆𝑖 ∑ 𝑑𝑖𝑘 𝜑𝑘 (𝑥1 ) = 0. (45.6)
𝐷 𝑘=1 𝑘=1

By making the error orthogonal to the basis functions:


𝑁 𝑁
∫⎡⎢ ∫ 𝐶(x1 , x2 ) ∑ 𝑑𝑖𝑘 𝜑𝑘 (𝑥1 )𝑑x1 − 𝜆𝑖 ∑ 𝑑𝑖𝑘 𝜑𝑘 (𝑥1 )⎤
⎥ 𝜑𝑗 (𝑥2 )𝑑𝑥2 = 0, (45.7)
𝐷 ⎣𝐷 𝑘=1 𝑘=1 ⎦
we get
𝑁 𝑁
∑ 𝑑𝑖𝑘 ⎡
⎢∬ 𝐶(x1 , x2 ) 𝜑𝑘 (𝑥1 )𝜑𝑗 (𝑥2 )𝑑x1 𝑑x2 ⎤
⎥ − 𝜆𝑖 ∑ 𝑑𝑖𝑘 ⎡
⎢ ∫ 𝜑𝑘 (𝑥2 )𝜑𝑗 (𝑥2 )𝑑x2 ⎤
⎥ = 0, (45.8)
𝑘=1 ⎣𝐷 ⎦ 𝑘=1 ⎣𝐷 ⎦

42-2 (Random Geometrical Imperfections) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Random Geometrical Imperfections

or the eigensystem
𝐀𝐃 = 𝚲𝐁𝐃 (45.9)
1
with Λ𝑖𝑗 = 𝛿𝑖𝑗 𝜆𝑗 . Orthogonal wavelets 𝜓(𝑥) are used (∫0 𝜓𝑗 (x)𝜓𝑘 (x)𝑑x = ℎ𝑗 𝛿𝑗𝑘 ) as basis
functions,
𝑁
𝐟𝑖 (𝑥) = ∑ 𝑑𝑖𝑘 𝜓𝑘 (𝑥) = 𝛙T (𝑥)𝐃(𝑖) , (45.10)
𝑘=1

so that the covariance function can be expressed as:


𝑁 𝑁
̅ 𝜓𝑗 (𝐱1 )𝜓𝑘 (𝐱2 ) = 𝛙T (𝐱1 )𝐀
𝐶(𝐱1 , 𝐱2 ) = ∑ ∑ 𝐴𝑗𝑘 ̅̅̅̅̅̅ 𝛙(𝐱2 ), (45.11)
𝑗=1 𝑘=1

̅̅̅̅̅̅ is the 2D wavelet transform of 𝐶(𝐱1 , 𝐱2 ) given by


where 𝐀
1 1 1
̅ =
𝐴𝑗𝑘 ∫ ∫ 𝐶(𝐱1 , 𝐱2 )𝜓𝑗 (𝐱1 )𝜓𝑘 (𝑥2 )𝑑x1 𝑑x2 . (45.12)
ℎ𝑗 ℎ𝑘 0 0
Substituting (45.10) and (45.11) into (45.2), we again get an eigenvalue problem
𝛙T (𝑥)𝐀
̅̅̅̅̅̅𝐇𝐃(𝑖) = 𝜆𝑖 𝛙T (𝑥)𝐃(𝑖) . (45.13)
⁄ 1
Or, equating coefficients of 𝛙T and using the transformation 𝐃̂ (𝑖) = 𝐇 2 𝐃(𝑖) and 𝐀
̂=
1⁄ 1⁄
𝐇 2𝐀̅̅̅̅̅̅𝐇 2 , the eigensystem
̂ 𝐃̂ (𝑖) = 𝜆𝑖 𝐃̂ (𝑖) .
𝐀 (45.14)
The eigenvectors from (45.14) are transformed to the eigen functions (of (45.9)) by the
equation:
−1⁄2
𝐟𝑖 (𝑥) = 𝛙T (𝑥)𝐇 𝐃̂ (𝑖) . (45.15)

The double integral in (45.12) is constructed using two successive 1D discrete


wavelet transforms in the form of Mallat’s tree algorithm (Phoon et al [2002a]). Haar
wavelets are used because of their simplicity and ability to capture the field characteristics.
The eigenfunctions (45.15) and associated eigenvalues 𝜆𝑖 can be used to construct random
fields (using (45.4)) with the same second-order statistics as the covariance model used.

The available covariance functions (Ghanem and Spanos [2003]), are:


• Exponential covariance function (First-order Markov process (autoregressive)),

−|𝑥1 − 𝑥2 |
𝐶(x1 , x2 ) = exp ( ) (45.16)
𝐿𝑐

• Triangular covariance function

LS-DYNA DEV 10/27/16 (r:8004) 42-3 (Random Geometrical Imperfections)


Random Geometrical Imperfections LS-DYNA Theory Manual

|𝑥1 − 𝑥2 |
𝐶(x1 , x2 ) = 1 − (45.17)
𝐿𝑐

• Sine covariance function

sin𝐿𝑐 (𝑥1 − 𝑥2 )
𝐶(x1 , x2 ) = (45.18)
𝐿𝑐 (𝑥1 − 𝑥2 )

• Squared exponential covariance function

−|𝑥1 − 𝑥2 |2
𝐶(x1 , x2 ) = exp ( ) (45.19)
𝐿𝑐

• Wiener-Levy covariance function

𝐶(x1 , x2 ) = min(x1 , x2 ) (45.20)

• Uniformly modulated nonstationary covariance function

−|𝑥1 − 𝑥2 |
𝐶(x1 , x2 ) = exp(−(𝑥1 − 𝑥2 ))exp (45.21)
𝐿𝑐
with 𝐿𝑐 the correlation length in the respective direction.

45.2.3 Generating eigenfunctions from experimentally measured fields


If experimentally measured random fields are available, then the eigenfunctions and
eigenvalues in (45.1) or (45.4) can be determined from the second-order statistics of the
measurements. Following the pattern recognition method proposed by Turk and Pentland
[1991], the procedure described next is used.

Given a set of 𝑀 field measurements, e.g., geometrical imperfections on an 𝑁1 × 𝑁2


mesh, we can represent the measurements as 1-D vectors of length 𝑁1 × 𝑁2, i.e., Γ1 , …, Γ𝑀 .
The average vector is defined by
1 𝑀
𝚿= ∑𝚪 , (45.22)
𝑀 𝑛=1 𝑛
allowing us to define the deviation of each measured field from the average, also as a
vector: 𝚽𝑖 = 𝚪𝑖 − 𝚿. Combining the deviation vectors into a covariance matrix, 𝐂, we get
1 𝑀
𝐂= ∑ 𝚽 𝚽T = 𝐀𝐀T , (45.23)
𝑀 𝑛=1 𝑛 𝑛
where 𝐀 = [𝚽1 𝚽2 … 𝚽𝑀 ]. The covariance matrix has a set of orthonormal eigenvectors,
𝐟𝑖 , and associated eigenvalues, 𝜆𝑖 obtained through a principal component analysis, i.e.,
𝐀𝐀T 𝐟𝑖 = 𝜆𝑖 𝐟𝑖 . (45.24)

42-4 (Random Geometrical Imperfections) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Random Geometrical Imperfections

The eigenpairs are chosen such that


1 𝑀 2
𝜆𝑖 = ∑(𝐟𝑖 T 𝚽𝑛 ) , (45.25)
𝑀 𝑛=1
is a maximum, subject to
𝐟𝑖 T 𝐟𝑘 = 𝛿𝑙𝑘 , (45.26)
with 𝛿 the Kronecker delta. As the size of the covariance matrix is (𝑁1 × 𝑁2)2 , determining
the eigenvectors and eigenvalues in (41.24) can be a time-consuming and memory-intensive
task for large measurement meshes. This computation can be simplified if the number of
measurement samples is less than the mesh count (𝑀 < 𝑁1 × 𝑁2), as there are then only

𝑀– 1, rather than 𝑁1 × 𝑁2, meaningful eigenvectors. This is done by considering the

eigenvectors of another matrix 𝐋 = 𝐀T 𝐀, as embodied in the eigensystem


𝐀T 𝐀v𝑖 = 𝜇𝑖 𝐯𝑖 . (45.27)
Premultiplying both sides in (20.1.6) by 𝐀, we get
𝐀𝐀T 𝐀𝐯𝑖 = 𝜇𝑖 𝐀𝐯𝑖 , (45.28)

which implies that the 𝑀 – 1 eigenvectors, 𝐀𝐯𝑖 , are also eigenvectors of 𝐀𝐀T or 𝐂. The

eigensystem in (20.1.6), however, is only size (𝑀 × 𝑀), as 𝐋 = 𝐀T 𝐀, where 𝐿𝑛𝑚 = 𝚽𝑇𝑚 𝚽𝑛 .


Once the eigenvectors of 𝐋 are obtained, the required eigenfunctions, 𝐟𝑖 , are recovered
through the linear combination
𝑀
𝐟𝑖 = ∑ v𝑖𝑘 𝚽𝑘 , 𝑖 = 1, … , 𝑀 (45.29)
𝑘=1

The 𝑀 – 1 eigenvalues of 𝐋 and 𝐂 are identical, i.e., 𝜆𝑖 = 𝜇𝑖 . Finally 𝐟𝑖 and 𝜆𝑖 are used in

(45.4) to construct the required Karhunen-Loève expansion of the random fields.

LS-DYNA DEV 10/27/16 (r:8004) 42-5 (Random Geometrical Imperfections)


Frequency Domain LS-DYNA Theory Manual

46
Frequency Domain

46.1 Frequency Response Functions


Frequency response function (FRF) is a characteristic of a system that has a
measured or computed response resulting from a known applied input. Mathematically it
is a transfer function and expresses the structural response to an applied force as a function
of frequency. The response can be given in terms of displacement, velocity, or acceleration.
Frequency response functions are complex functions, with real and imaginary components.
They can also be written in terms of magnitude and phase pairs.

46.1.1 FRF Computations


FRF is computed using mode superposition method, in frequency domain. When
damping is included, the dynamic response of a system is governed by
𝐦𝐮̈ + 𝐜𝐮̇ + 𝐤𝐮 = 𝐩(𝑡), (46.1.30)
where 𝐦, 𝐜 and 𝐤 are the mass, damping and stiffness matrices, 𝐩(𝑡) is the external force.
Using the mode superposition method, the displacement response can be expressed
by
𝑁
𝐮 = ∑ 𝜙𝑛 𝑞𝑛 (𝑡) = Φq, (46.1.2)
𝑛=1

where 𝜙𝑛 is the n-th mode shape and 𝑞𝑛 (𝑡) is the n-th modal coordinates.
With the substitution of Equation (46.1.2) into Equation (46.1.30), the governing
equation can be rewritten as
𝐦𝚽𝐪̈ + 𝐜𝚽𝐪̇ + 𝐤𝚽𝐪 = 𝐩(𝑡). (46.1.3)
Pre-multiplying by ΦT gives
𝐌𝐪̈ + 𝐂𝐪̇ + 𝐊𝐪 = 𝐩(𝑡), (46.1.4)
The orthogonality of natural modes implies that the following square matrices are
diagonal:

42-6 (Frequency Domain) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Frequency Domain

𝐌 ≡ 𝚽𝐓 𝐦𝚽, 𝐊 ≡ 𝚽𝐓 𝐤𝚽, (46.1.5)


where the diagonal elements are
𝐌𝒏 = 𝛟𝐓𝒏 𝐦𝛟𝒏 , 𝐊𝒏 = 𝛟𝐓𝒏 𝐤𝛟𝒏 (46.1.6)
Since 𝐦 and 𝐤 are positive definite, the diagonal elements of 𝐌 and 𝐊 are positive. They
are related by
𝐊𝒏 = 𝛚𝟐𝒏 𝐌𝒏 . (46.1.7)
The square matrix 𝐂 is obtained similarly as follows
𝐂 = 𝚽𝐓 𝐜𝚽. (46.1.8)
𝐂 may or may not be diagonal, depending on the distribution of damping in the
system. If 𝐂 is diagonal (the diagonal elements are 𝐶𝑛 = 𝜙𝑇𝑛 𝑐𝜙𝑛 ), Equation (46.1.4)
represents N uncoupled differential equations in modal coordinates 𝑞𝑛 , and the system
is said to have classical damping and the systems possess the same natural modes as
those of the undamped system. Only the classical damping is considered in this
approach.

The right hand side vector (generalized force) 𝐏(𝑡) is


𝐏(𝑡) = ΦT 𝑝(𝑡) (46.1.9)
For an N-DOF system with classical damping, each of the N differential equations in
modal coordinates is

𝑀𝑛 𝑞𝑛̈ + 𝐶𝑛 𝑞𝑛̇ + 𝐾𝑛 𝑞𝑛 = 𝑃𝑛 (𝑡) (46.1.10)


or,

𝑃𝑛 (𝑡)
𝑞𝑛̈ + 2𝜁𝑛 𝜔𝑛 𝑞𝑛̇ + 𝜔𝑛2 𝑞𝑛 = (46.1.11)
𝑀𝑛
where the modal damping coefficient  n is defined as

𝐶𝑛
𝜁𝑛 = (46.1.12)
2𝑀𝑛 𝜔𝑛
Applying Fourier transform to both sides of Equation (46.1.11), one obtains

𝑃𝑛 (𝜔)
(−𝜔2 + 2𝑖𝜁𝑛 𝜔𝑛 𝜔 + 𝜔𝑛2 )𝑞𝑛 (𝜔) = (46.1.13)
𝑀𝑛
The structural displacement response in frequency domain can be represented as
𝑁
𝜙𝑛 𝑃𝑛 (𝜔)
𝐮(𝜔) = ∑ (46.1.14)
𝑛=1
(−𝜔2 2
+ 2𝑖𝜁𝑛 𝜔𝑛 𝜔 + 𝜔𝑛 ) 𝑀𝑛

LS-DYNA DEV 10/27/16 (r:8004) 42-7 (Frequency Domain)


Frequency Domain LS-DYNA Theory Manual

Thus the displacement frequency response function (Compliance) can be expressed


as (suppose that the excitation is applied at node j and the response is evaluated for node k)

𝑁
𝜙𝑛 (𝑥𝑘 ) 𝑃̃ 𝑛 (𝑥𝑗 )
𝐅𝐑𝐅𝒖 (𝑥𝑗 , 𝑥𝑘 , 𝜔) = ∑ (46.1.15)
𝑛=1
(−𝜔2 + 2𝑖𝜁𝑛 𝜔𝑛 𝜔 + 𝜔𝑛2 ) 𝑀𝑛
The velocity frequency response function (Mobility) can be expressed as

𝑁
𝜙𝑛 (𝑥𝑘 ) 𝑃̃ 𝑛 (𝑥𝑗 )
𝐅𝐑𝐅𝒗 (𝑥𝑗 , 𝑥𝑘 , 𝜔) = 𝜔𝑖 ∑ (46.1.16)
𝑛=1
(−𝜔2 + 2𝑖𝜁𝑛 𝜔𝑛 𝜔 + 𝜔𝑛2 ) 𝑀𝑛
The acceleration frequency response function (Accelerance) can be expressed as

𝑁
𝜙𝑛 (𝑥𝑘 ) 𝑃̃ 𝑛 (𝑥𝑗 )
𝐅𝐑𝐅𝒂 (𝑥𝑗 , 𝑥𝑘 , 𝜔) = −𝜔2 ∑ (46.1.17)
𝑛=1
(−𝜔2 + 2𝑖𝜁𝑛 𝜔𝑛 𝜔 + 𝜔𝑛2 ) 𝑀𝑛
~
where Pn ( x j ) is obtained as

𝑃̃ 𝑛 (𝑥𝑗 ) = 𝜙𝑇𝑛 𝑝̃(𝑥𝑗 ) (46.1.18)


and ~p ( x j ) is the space distribution of the harmonic force excitation (in the case of point
force excitation, ~ p ( x )  1 at node j in specified direction of excitation and 0 elsewhere).
j

46.1.2 About the damping


Damping can be given in several forms (see Keyword user manual in Appendix). A
very common type of damping used in the nonlinear analysis of structure is to assume that
the damping matrix is proportional to the mass and stiffness matrices, or

𝐜 = 𝛼𝐦 + 𝛽𝐤 (46.1.19)
This type of damping is normally referred to as Rayleigh damping. For classically
damped system,

2𝝎𝒏 𝜁𝑛 = 𝜙𝑇𝑛 𝐜𝜙𝑛 (46.1.20)


Due to the orthogonality of the mass and stiffness matrices, it can be rewritten as

2𝝎𝒏 𝜁𝑛 = 𝛼 + 𝛽𝜔𝑛2 (46.1.21)


or,
𝛼 𝜔
𝜁𝑛 = + 𝑛𝛽 (46.1.22)
2𝜔𝑛 2

42-8 (Frequency Domain) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Frequency Domain
46.2 ACOUSTIC FEM
A frequency domain acoustic finite element method has been implemented in LS-
DYNA, to model the acoustic behavior of a confined acoustic fluid volume. This method is
based on nodal velocity/pressure formulation. Three types of elements are available. They
are hexahedron, pentahedron, and tetrahedron elements.

46.2.1 Theory basis


The governing equation for the acoustic problem is the Helmholtz equation.
∇𝟐 𝑝 + 𝑘 𝟐 𝑝 = 0 (46.2.1)
where 𝑝 is the acoustic pressure; 𝑘 = 𝜔/𝑐 is called the wave number; 𝜔 = 2𝜋𝑓 is the circular
frequency of the acoustic wave; and 𝑐 is the wave speed.
For vibro-acoustic problems, the boundary condition is given as follows,
𝜕𝑝
= −𝑖𝜌𝜔𝑣𝑛 (46.2.2)
𝜕𝑛
where 𝑛 is the normal vector pointing outside from the acoustic volume; 𝑖 = √−1 is the
imaginary unit; 𝜌 is the acoustic fluid density and 𝑣𝑛 is the normal velocity.
Using the weighted residue technique and taking the shape function 𝑁𝑖 as the
weighting function, the governing equation can be written as

∫ ∇2 𝑝𝑁𝑖 𝑑𝑉 + ∫ 𝑘 2 𝑝𝑁𝑖 𝑑𝑉 = 0 (46.2.3)


𝑉 𝑉

Using the Green’s theorem, Equation (2.3) can be written as

𝜕𝑝
− ∫ ∇𝑝∇𝑁𝑖 𝑑𝑉 + 𝑘 2 ∫ 𝑝𝑁𝑖 𝑑𝑉 = − ∫ 𝑁 𝑑Γ (46.2.4)
𝜕𝑛 𝑖
𝑉 𝑉 Γ

With the substitution of the boundary condition (46.2.2) into Equation (46.2.4), and
taking the nodal pressure as the unknown variables, a linear equation system can be
established and solved in frequency domain. Since there is only one variable on each node,
this method is very fast.

LS-DYNA DEV 10/27/16 (r:8004) 42-9 (Frequency Domain)


Rotor Dynamics LS-DYNA Theory Manual

47
Rotor Dynamics

47.1 Introduction
Rotor dynamics is a specialized branch of engineering science concerned with the behavior
and diagnosis of rotating structures. It is a study of vibration of rotating parts found in a
wide range of equipment including engine, turbine, aircraft, hard disk drive and more.
The analysis of the rotator dynamics involves two coordinate systems: rotating and fixed
coordinate systems. The equations of motion in the two coordinate systems are both
introduced.

47.2 Two Coordinate systems


The interpretation of rotational phenomena requires the introduction of a rotating
coordinate system in relation to the fixed coordinate system. Figure 1.1 depicts the
relationship between the two coordinate systems. OXYZ is the fixed coordinate system and
oxyz is the rotating coordinate system. R is the location vector of the disk center; r is the
location vector of a point P with respect to the rotating coordinate system. Let the velocity
of rotation be defined as .

42-10 (Rotor Dynamics) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Rotor Dynamics

Figure 1.1 A rotating disk in two coordinate systems.

The position of the point P with respect to OXYZ is 𝒓 ,̅ so


𝒓 ̅ = 𝑹 + 𝒓. (47.1)
The velocity of point P is:
𝑑𝒓 ̅ 𝑑𝑹 𝑑𝒓
𝒗
̅̅ ̅ = = + = 𝑽 + 𝒗 + 𝜴 × 𝒓, (47.2)
𝑑𝑡 𝑑𝑡 𝑑𝑡
where 𝑽 (=𝑑𝑹𝑑𝑡
) is the velocity of the origin o; 𝒗 is the velocity of point P in the rotating
coordinate system.

The acceleration of point P can be calculated as:


̅̅ ̅ 𝑑𝑽 𝑑𝒗 𝑑(𝜴 × 𝒓 )
𝑑𝒗 𝒅𝜴
𝒂̅ = = + + = 𝑨 + 𝒂 + 𝟐𝜴 × 𝒗 + 𝜴 × (𝜴 × 𝒓) + × 𝒓, (47.31)
𝑑𝑡 𝑑𝑡 𝑑𝑡 𝑑𝑡 𝑑𝑡
where, 𝑨 is the acceleration of the origin o, 𝒂 is the acceleration of point P in rotating
coordinate system.

We assume that the origin of the rotating coordinate system is fixed, so that:
𝑽 =𝑨=𝟎. (47.32)
By substituting (1.4) to (1.3):
𝑑𝜴
𝒂̅ = 𝒂 + 𝟐𝜴 × 𝒗 + 𝜴 × (𝜴 × 𝒓) + ×𝒓 . (47.33)
𝑑𝑡

47.3 Forces in the Rotating Coordinate System


We now place a particle with mass m into the position of the point P we were following.
From (1.5), we can express the force of the particle in the rotating system as:

LS-DYNA DEV 10/27/16 (r:8004) 42-11 (Rotor Dynamics)


Rotor Dynamics LS-DYNA Theory Manual

𝐝𝛀
𝑭 = 𝑚𝒂 = 𝑚𝒂̅ − 2𝑚𝜴 × 𝒗 − 𝑚𝜴 × (𝜴 × 𝒓) − 𝑚 × 𝐫. (47.34)
𝐝t
The first term 𝑚𝒂̅ is the force in the fixed coordinate system. All other terms on the right
hand side are inertia forces arising in the rotating system. The Coriolis force is the
following quantity:
𝑭𝑪 = −2𝑚𝜴 × 𝒗. (47.35)
The third term produces the familiar centrifugal force:
𝑭𝑪𝒇 = −𝑚𝜴 × (𝜴 × 𝒓). (47.36)
The last term introduces the Euler fore when there is a nonzero rate of change in the
magnitude of the rotation vector:
𝒅𝜴
𝑭𝑬 = −𝑚 × 𝒓. (47.37)
𝒅𝑡

47.4 Transformation between Coordinate Systems


Let’s assume that the rotation axis coincides with one of the axis of the rotating coordinate
system, specifically to the z axis. An arbitrary rotation axis will be discussed later. In this
case, the rotational velocity becomes:
𝜴 = 0 ∙ 𝒊 + 0 ∙ 𝒋 + Ω ∙ 𝒌. (47.38)
We further restrict our computations to constant rotational velocity; hence the Euler force
will not appear in the formulations (Euler force is easy to add to our equations if the
rotational velocity is not constant though).

We write the location of a particle in the rotating coordinate system as:


𝑥
𝒓 = {𝑦}. (47.39)
𝑧
It is transformed to the location in the fixed system as follows:

{𝑥̅⎫
⎧ } 𝑐𝑜𝑠Ωt −𝑠𝑖𝑛Ωt 0 𝑥 𝑥
𝐫 ̅ = ⎨𝑦̅⎬ = ⎡
⎢𝑠𝑖𝑛Ωt 𝑐𝑜𝑠Ωt ⎤
0⎥ {𝑦} = 𝑯 {𝑦}. (47.40)
{
⎩ 𝑧̅}
⎭ ⎣ 0 0 1⎦ 𝑧 𝑧
𝑯 is the transformation matrix. It is easy to get that 𝑯 𝑇 𝑯 = 𝑰, where 𝑰 is he identity
matrix. Other matrices that may be used later are also given here:
−𝑠𝑖𝑛Ωt −𝑐𝑜𝑠Ωt 0
𝑯̇ = Ω ⎡
⎢ 𝑐𝑜𝑠Ωt −𝑠𝑖𝑛Ωt 0⎤ ̅̅̅̅̅̅,
⎥ = Ω𝐇 (47.41)
⎣ 0 0 0⎦

42-12 (Rotor Dynamics) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Rotor Dynamics

−𝑐𝑜𝑠Ωt sinΩt 0
𝑯̈ = Ω2 ⎡
⎢−𝑠𝑖𝑛Ωt −𝑐𝑜𝑠Ωt 0⎤
2 ̿̿̿̿̿
⎥ = Ω ̿𝐇 , (47.42)
⎣ 0 0 0⎦
0 1 0
̅̅̅ ̅ 𝑇 𝑯 = ⎡
𝑯 ⎢−1 0 0⎤⎥ = 𝑷, (47.43)
⎣0 0 0⎦
0 −1 0
̅̅̅̅̅ = ⎡
𝑯𝑇𝑯 ⎢1 0 0⎤
T
⎥ = 𝐏 = −𝑷, (47.44)
⎣0 0 0⎦
1 0 0
̅̅̅̅̅ 𝑇 = ⎡
̅̅̅̅̅ 𝑇 𝑯
𝑯 ⎢0 1 0⎤⎥ = 𝐉. (47.45)
⎣0 0 0 ⎦

47.5 Equation of Motion in Rotating Coordinate System


When the particle undergoes a nodal translation in the rotating coordinate system as
shown in Figure 1.2, it can be defined as:
𝑢
𝒖 = { 𝑣 }. (47.46)
𝑤
The location vector in the fixed coordinate is:
𝒓 ̅ = 𝑯(𝒓 + 𝒖) (47.47)
After calculate the velocity in the fixed coordinate system, we can get the kinetic energy
due to translation displacement as:
𝑚Ω2 𝑇 𝑚 (47.48)
𝑇= (𝒓 𝑱𝒓 + 2𝒓 𝑇 𝑱𝒖 + 𝒖𝑇 𝑱𝒖) + (2Ω𝒖̇𝑇 𝑷 𝑇 𝒓 + 2Ω𝒖𝑇 𝑷𝒖̇ + 𝒖̇𝑇 𝒖̇).
2 2

Figure 1.2 Translation nodal displacement.


LS-DYNA DEV 10/27/16 (r:8004) 42-13 (Rotor Dynamics)
Rotor Dynamics LS-DYNA Theory Manual

Figure 1.3 Rotational nodal displacement.


The nodal displacement of the point can also be rotational as shown in Figure 1.3. We
assume small rotation of the rotating point:
⎧ 𝜑
{ ⎫ }
𝛉 = ⎨𝜙 ⎬. (47.49)
{
⎩𝜃 }⎭
The nodal rotation requires the consideration of mass and inertia. The center of the mass
point is coincident with the node. The inertia moment can be given with the concentrated
mass input of commercial finite element codes, or they can be defined in connection with
the surrounding mass point, or even with a simplified model by attaching six submasses to
the node, as shown in figure 1.4.

Figure 1.4 Node with six masses located at offset x’, y’ and z’ from the node center.

With this, the location vector in the fixed coordinate is of the form if only considers
rotational displacement:
𝒓 ̅ = 𝑯(𝒓 + 𝒖) = 𝑯(𝒓 + 𝑨𝛉), (47.50)
where
0 𝑧 −𝑦

𝑨 = ⎢−𝑧 𝑥 ⎤
0 ⎥. (47.51)
⎣𝑦 −𝑥 0 ⎦
Then we can get the kinetic energy due to rotational displacement as:
𝑚Ω2 𝑇 𝑚
𝑇= (𝒓 𝑱𝒓 + 𝟐𝒓 𝑇 𝑱𝑨𝛉 + 𝛉𝑻 𝑨𝑇 𝑱𝑨𝛉) + (2Ω𝒓 𝑇 𝑷𝑨𝛉̇ + 2Ω𝛉𝑇 𝑨𝑇 𝑷𝑨𝛉̇ + 𝛉̇𝑇 𝑨𝑇 𝑨𝛉̇). (47.52)
2 2
42-14 (Rotor Dynamics) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual Rotor Dynamics

Applying Lagrange’s equation, then we can obtain the final equation of motion as follows:
𝑴𝒖 𝟎 𝑪 𝟎 𝒁 𝟎 𝒖 𝑭
[ ] {𝒖̈̈} + 2Ω [ 𝒖 ] {𝒖̇̇ } − Ω2 [ 𝒖 ] { } = { 𝒄𝒖 }, (47.53)
𝟎 𝑴𝛉 𝛉 𝟎 𝑪𝛉 𝛉 𝟎 𝒁𝛉 𝛉 𝑭𝒄𝛉
where 𝑴𝒖 and 𝑴𝛉 are the mass and inertia matrices; 𝑪𝒖 and 𝑪𝛉 are the gyroscopic
matrices; 𝒁𝒖 and 𝒁𝛉 are the centrifugal softening matrices; 𝑭𝒄𝒖 and 𝑭𝒄𝛉 are centrifugal force.
Note that we don’t consider the other system damping and external force terms here, but
they can be added to (1.25) accordingly.

47.6 Equation of Motion in Fixed Coordinate System


The location vector in the fixed coordinate is:
𝒓 ̅ = 𝑯(𝒓 + 𝒖) = 𝑯𝒓 + 𝒖̅. (47.54)
Here the 𝒖̅ represents the displacement of the point in the fixed coordinate system due to
the nodal translation displacement. Similar analysis as in section 1.5 can be done for the
equation of motion in fixed coordinate system. We only give the final equation here:
𝑴𝒖̅ 𝟎 ̈ 0 0 𝑭𝒄𝒖̅
[ ] {𝒖̅̈ } + Ω [0 𝒖̇
𝑪𝛉̅̅̅̅ ] {𝛉̇} = { 𝟎 }, (47.55)
𝟎 𝑴𝛉̅̅̅̅ 𝛉 ̅̅̅̅
where 𝑴𝒖̅ and 𝑴̅𝛉̅̅̅ are the mass and inertia matrices; 𝑪̅𝛉̅̅̅ is the gyroscopic matrix; 𝑭𝒄𝒖̅ is the
centrifugal force. All of them are written in the fixed coordinate system. Note that only
nodal rotations contributed to the gyroscopic matrix. Same as before, we don’t consider the
other system damping and external force terms here, but they can be added to (1.27)
accordingly.

47.7 Arbitrary Rotation Axis


All the above analysis is based on the assumption that the rotation axis is coincide with the
z-axis. Here we will give a way to transform all variables back to global if the rotation axis
is not coincide with the z-axis (Figure 1.5). The transformation matrix from z axis to
rotation axis is denoted as T and it is easy to get 𝑻 𝑇 𝑻 = 𝑰 .

LS-DYNA DEV 10/27/16 (r:8004) 42-15 (Rotor Dynamics)


Rotor Dynamics LS-DYNA Theory Manual

Figure 1.5 Rotation axis not coincide with z axis.

The location vector in the fixed coordinate becomes:


𝒓 ̅ = 𝑻 𝑇 𝑯(𝒓 + 𝒖 ). (47.56)
where 𝒓 and 𝒖 are the location vector and displacement in the coordinate system Ox’y’z’,
in which the z axis is transformed to the rotation axis z’ by rotation matrix 𝑻.At the same
time:
𝒓′ = 𝑻𝒓 , (47.57)

𝒖 = 𝑻𝒖. (47.58)
where, 𝒓 and 𝒖 are the location vector and displacement in the rotating coordinate system
Oxyz.

The equation of motion in (1.25) and (1.27) can be simplified written as:
𝑴𝟎 𝒂 + 𝑪𝟎 𝒗 + 𝑲𝟎 𝒖 = 𝑭𝟎 , (47.59)
And the force term can be written as
𝑭𝟎 = 𝒇𝟎 𝒓, (47.60)
By substituting (1.28), (1.29) and (1.30) to the equation of motion, we can get new mass,
damping, stiffness matrices and force vector as follows:
𝑴 = 𝑻 𝑻 𝑴𝟎 𝑻 , (47.61)

𝑪 = 𝑻 𝑻 𝑪𝟎 𝑻, (47.62)

𝑲 = 𝑻 𝑻 𝑲𝟎 𝑻, (47.63)

𝑭 = 𝑻 𝑻 𝒇𝟎 𝑻𝒓. (47.64)
So the equation of motion becomes:

42-16 (Rotor Dynamics) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual Rotor Dynamics

𝑴𝒂 + 𝑪𝒗 + 𝑲𝒖 = 𝑭 . (47.65)
After the equation of motion is obtained, it can then be solved using the implicit solver.
Especially, the damping and stiffness matrices are related to the rotational velocity, so the
eigen-frequencies might change with the change of rotational velocity. A diagram to
represent this relationship is called Campbell diagram. An example is given in Figure 1.6.

LS-DYNA DEV 10/27/16 (r:8004) 42-17 (Rotor Dynamics)


Rotor Dynamics LS-DYNA Theory Manual

Figure 1.6 A disk is spinning with the center axis, the mode frequencies change with the
increase of rotating speed.

42-18 (Rotor Dynamics) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual References

48
References

Abbo A.J., S.W. Sloan, “A Smooth Hyperbolic Approximation to the Mohr-Coulomb


Yield Criterion,” Computers and Structures, Vol 54, No 1, 1995.

Addessio, F.L., D.E. Carroll, J.K. Dukowicz, F.H. Harlow, J.N. Johnson, B.A. Kashiwa,
M.E. Maltrud, H.M. Ruppel, “CAVEAT: A Computer Code for Fluid Dynamics
Problems with Large Distortion and Internal Slip,” Report LA-10613-MS, UC-32, Los
Alamos National Laboratory (1986).

Ahmad, S., Irons, B.M. and Zienkiewicz, O.C., “Analysis of Thick and Thin Shell
Structures by Curved Finite Elements,” Int. J. Numer. Meths. Eng., 2 (1970).

Amsden, A. A., and Hirt, C. W., “YAQUI: An Arbitrary Lagrangian-Eulerian


Computer Program for Fluid Flow at All Speeds,” Los Alamos Scientific Laboratory,
LA-5100 (1973).

Amdsden, A., A., Ruppel, H. M., and Hirt, C. W., “SALE: A Simplified ALE Computer
Program for Fluid Flow at All Speeds,” Los Alamos Scientific Laboratory (1980).

Argyris, J.H., Kelsey, S., and Kamel, H., “Matrix Methods of Structural Analysis: A
Precis of recent Developments,” Matrix Methods of Structural Analysis, Pergamon
Press (1964).

Arruda, E., and M. Boyce, "A Three-Dimensional Constitutive Model for the Large
Stretch Behavior of Rubber Elastic Materials," published in the Journal of the Mechanics
and Physics of Solids, Vol. 41, No. 2, 389-412 (1993).

Auricchio, F., Taylor, R.L. and Lubliner J., “Shape-memory alloys: macromodelling and
numerical simulations of the superelastic behavior”, Computer Methods in Applied
Mechanics and Engineering 146, 281-312 (1997).

LS-DYNA DEV 10/27/16 (r:8004) 43-1 (References)


References LS-DYNA Theory Manual

Auricchio, F. and Taylor, R.L., “Shape-memory alloys: modeling and numerical


simulations of the finite-strain superelastic behavior”, Computer Methods in Applie
Mechanics and Engineering 143, 175-194 (1997).

Back, S.Y., and Will, K.M., “A Shear-flexible Element with Warping for Thin-Walled
Open Sections,” International Journal for Numerical Methods in Engineering, 43, 1173-1191
(1998).

Bahler AS: The series elastic element of mammalian skeletal muscle. Am J Physiol
213:1560-1564, 1967.

Bammann, D. J., and E.C. Aifantis, “A Model for Finite-Deformation Plasticity,” Acta
Mechanica, 70, 1-13 (1987).

Bammann, D. J., “Modeling the Temperature and Strain Rate Dependent Large
Deformation of Metals,” Proceedings of the 11th US National Congress of Applied
Mechanics, Tucson, AZ (1989).

Bammann, D. J., and Johnson, G., “On the Kinematics of Finite-Deformation Plasticity,”
Acta Mechanica 69, 97-117 (1987).

Bammann, D.J., Chiesa, M.L., McDonald, A., Kawahara, W.A., Dike, J.J. and Revelli,
V.D., “Predictions of Ductile Failure in Metal Structures,” in AMD-Vol. 107, Failure
Criteria and Analysis in Dynamic Response, edited by. H.E. Lindberg, 7-12 (1990).

Bandak, F.A., private communications, U.S. Dept. of Trans., Division of Biomechanics


Research, 400 7th St., S.W. Washington, DC 20590 (1991).

Barlat and Lian, J., “Plastic Behavior and Stretchability of Sheet Metals, Part I: A Yield
Function for Orthotropic Sheets Under Plane Stress Conditions,” International Journal of
Plasticity, 5, 51-66 (1989).

Barlat, F., Lege, D.J., and Brem, J.C., “A Six-Component Yield Function for Anisotropic
Materials,” International Journal of Plasticity, 7, 693-712 (1991).

Bathe, K. J., Finite Element Procedures in Engineering Analysis, Prentice-Hall (1982).

Bathe, K. J., and Wilson, E.L., Numerical Methods in Finite Element Analysis, Prentice-
Hall (1976).

Bathe, K.J., and Dvorkin, E.N., “A Continuum Mechanics Based Four Node Shell
Element for General Nonlinear Analysis,” Int. J. Computer-Aided Eng. and Software,
Vol. 1, 77-88 (1984).

43-2 (References) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual References

Bathe, K.-J. and Dvorkin, E.N. A four node plate bending element based on Mindlin-
Reissner plate theory and a mixed interpolation, Int. J. Num. Meth. Eng., 21, 367-383,
1985.

Batoz, J.L. and Ben Tahar, M. Evaluation of a new quadrilateral thin plate bending
element, Int. J. Num. Meth. Eng., 18, 1644-1677, 1982.

Battini, J., and Pacoste C., “Co-rotational Beam Elements with Warping Effects in
Instability Problems,” Computational Methods in Applied Mechanical Engineering, 191,
1755-1789, (2002).

Bazeley, G.P., Cheung, W.K., Irons, B.M. and Zienkiewicz, O.C., “Triangular Elements
in Plate Bending—Conforming and Nonconforming Solutions in Matrix Methods and
Structural Mechanics,” Proc. Conf. on Matrix Methods in Structural Analysis, Rept.
AFFDL-R-66-80, Wright Patterson AFB, 547-576 (1965).

Belytschko, T., “Transient Analysis,” Structural Mechanics Computer Programs, edited


by W. Pilkey, et. al., University Press of Virginia, 255-276 (1974).

Belytschko, T., and Bindeman, L. P., "Assumed Strain Stabilization of the Eight Node
Hexahedral Element," Comp. Meth. Appl. Mech. Eng. 105, 225-260 (1993).

Belytschko, T., and Hsieh, B.J., ”Nonlinear Transient Finite Element Analysis with
Convected Coordinates,” Int. J. Num. Meths. Engrg., 7, 255–271 (1973).

Belytschko, T., and Leviathan, I., “Projection schemes for one-point quadrature shell
elements,” Comp. Meth. Appl. Mech. Eng., 115, 277-286 (1994).

Belytschko, T., and Lin, J., “A New Interaction Algorithm with Erosion for EPIC-3”,
Contract Report BRL-CR-540, U.S. Army Ballistic Research Laboratory, Aberdeen
Proving Ground, Maryland (1985).

Belytschko, T., Lin, J., and Tsay, C.S., “Explicit Algorithms for Nonlinear Dynamics of
Shells,” Comp. Meth. Appl. Mech. Eng. 42, 225-251 (1984) [a].

Belytschko, T., Ong, J. S.-J., Liu, W.K. and Kennedy, J.M., “Hourglass Control in Linear
and Nonlinear Problems”, Comput. Meths. Appl. Mech. Engrg., 43, 251–276 (1984b).

Belytschko, T., Stolarski, H., and Carpenter, N., “A C0 Triangular Plate Element with
One-Point Quadrature,” International Journal for Numerical Methods in Engineering,
20, 787-802 (1984) [b].

LS-DYNA DEV 10/27/16 (r:8004) 43-3 (References)


References LS-DYNA Theory Manual

Belytschko, T., Schwer, L., and Klein, M. J., “Large Displacement Transient Analysis of
Space Frames,” International Journal for Numerical and Analytical Methods in
Engineering, 11, 65-84 (1977).

Belytschko, T., and Tsay, C.S., “Explicit Algorithms for Nonlinear Dynamics of Shells,”
AMD, 48, ASME, 209-231 (1981).

Belytschko, T., and Tsay, C.S., “WHAMSE: A Program for Three-Dimensional


Nonlinear Structural Dynamics,” Report EPRI NP-2250, Project 1065-3, Electric Power
Research Institute, Palo Alto, CA (1982).

Belytschko, T., and Tsay, C. S., “A Stabilization Procedure for the Quadrilateral Plate
Element with One-Point Quadrature,” Int. J. Num. Method. Eng. 19, 405-419 (1983).

Belytschko, T., Yen, H. R., and Mullen R., “Mixed Methods for Time Integration,”
Computer Methods in Applied Mechanics and Engineering, 17, 259-175 (1979).

Belytschko, T., “Partitioned and Adaptive Algorithms for Explicit Time Integration,” in
Nonlinear Finite Element Analysis in Structural Mechanics, ed. by Wunderlich, W.
Stein, E, and Bathe, J. J., 572-584 (1980).

Belytschko, T., Wong, B.L., and Chiang, H.Y., “Improvements in Low-Order Shell
Elements for Explicit Transient Analysis,” Analytical and Computational Models of
Shells, A.K. Noor, T. Belytschko, and J. Simo, editors, ASME, CED, 3, 383-398 (1989).

Belytschko, T., Wong, B.L., and Chiang, H.Y., “Advances in One-Point Quadrature Shell
Elements,” Comp. Meths. Appl. Mech. Eng., 96, 93-107 (1992).

Belytschko, T., Wong, B. L., Plaskacz, E. J., "Fission - Fusion Adaptivity in Finite
Elements for Nonlinear Dynamics of Shells," Computers and Structures, Vol. 33, 1307-
1323 (1989).

Belytschko, T., Lu, Y.Y. and Gu, L., “Element-free Galerkin Methods,” Int. J. Numer.
Methods Engrg. 37, 229-256 (1994).

Benson, D.J., “Vectorizing the Right-Hand Side Assembly in an Explicit Finite Element
Program,” Comp. Meths. Appl. Mech. Eng., 73, 147-152 (1989).

Benson, D. J., and Hallquist, J.O., “A Simple Rigid Body Algorithm for Structural
Dynamics Program,” Int. J. Numer. Meth. Eng., 22 (1986).

Benson, D.J., and Hallquist J.O., “A Single Surface Contact Algorithm for the
Postbuckling Analysis of Shell Structures,” Comp. Meths. Appl. Mech. Eng., 78, 141-
163 (1990).
43-4 (References) LS-DYNA DEV 10/27/16 (r:8004)
LS-DYNA Theory Manual References

Benson, D. J., “Momentum Advection on a Staggered Mesh,” Journal of Computational


Physics, 100, No. 1, (1992).

Benson, D. J., “Vectorization Techniques for Explicit Arbitrary Lagrangian Eulerian


Calculations,” Computer Methods in Applied Mechanics and Engineering (1992).

Berstad, T., "Material Modelling of Aluminium for Crashworthiness Analysis", Dr.Ing.


Dissertation, Department of Structural Engineering, Norwegian University of Science
and Technology, Trondheim, Norway (1996).

Berstad T., Hopperstad, O.S., and Langseth, M., “Elasto-Viscoplastic Consitiutive


Models in the Explicit Finite Element Code LS-DYNA,” Proceedings of the Second
International LS-DYNA Conference, San Francisco, CA (1994).

Bischoff M. and Ramm E., “Shear deformable shell elements for large strains and
rotations,”
. Int. J. Numer. Methods, 40, 4427-4449 (1997).

Blatz, P.J., and Ko, W.L., “Application of Finite Element Theory to the Deformation of
Rubbery Materials,” Trans. Soc. of Rheology, 6, 223-251 (1962).

Bodig, Jozsef and Benjamin A. Jayne, Mechanics of Wood and Wood Composites,
Krieger Publishing Company, Malabar, FL (1993).

Boeing, Boeing Extreme Mathematical Library BCSLIB-EXT User's Guide, The Boeing
Company, Document Number 20462-0520-R4 (2000).

Borrvall, T., Development and implementation of material tangent stiffnesses for


material model 76 in LS-DYNA, ERAB-02:46, Engineering Research Nordic AB,
Linköping (2002).

Borrvall, T., Revision of the implementation of material 36 for shell elements in LS-
DYNA, ERAB Report E0307, Engineering Research Nordic AB, Linköping (2003).

Björklund O., “Ductile Failure in High Strength Steel Sheets,” Linköping Studies in
Science and Technology. Dissertations No. 1579, ISSN 0345-7524 (2014).

Brekelmans, W.A.M., Scheurs, P.J.G., and de Vree, J.H.P., “Continuum damage


mechanics for softening of brittle materials”, Acta Mechanica, 93, 133-143, (1991).

Brooks, A. N., and Hughes, T. J. R., “Streamline Upwind/Petrov-Galerkin


Formulations for Convection Dominated Flows with Particular Emphasis on the
Incompressible Navier-Stokes Equations,” Computer Methods in Applied Mechanics
and Engineering, 32, 199-259, (1982).

LS-DYNA DEV 10/27/16 (r:8004) 43-5 (References)


References LS-DYNA Theory Manual

Burton, D.E., et. al., “Physics and Numerics of the TENSOR Code,” Lawrence
Livermore National Laboratory, Internal Document UCID-19428 (July 1982).

Cardoso, R.P.R. and Yoon J-W., “One point quadrature shell element with through-
thickness stretch, Computer Methods in Applied Mechanics and Engineering,” 194(9),
1161-1199 (2005).

Chang, F.K., and Chang, K.Y., “Post-Failure Analysis of Bolted Composite Joints in
Tension or Shear-Out Mode Failure,” J. of Composite Materials, 21, 809-833 (1987).[a]

Chang, F.K., and Chang, K.Y., “A Progressive Damage Model for Laminated
Composites Containing Stress Concentration,” J. of Composite Materials, 21, 834-855
(1987).[b]

Chang, F.S., “Constitutive Equation Development of Foam Materials,” Ph.D.


Dissertation, submitted to the Graduate School, Wayne State University, Detroit,
Michigan (1995).

Chang, F.S., J.O. Hallquist, D. X. Lu, B. K. Shahidi, C. M. Kudelko, and J.P. Tekelly,

“Finite Element Analysis of Low Density High-Hystersis Foam Materials and the
Application in the Automotive Industry,” SAE Technical Paper 940908, in Saftey
Technology (SP-1041), International Congress and Exposition, Detroit, Michigan (1994).

Chen, J.S., Pan, C., Wu, C.T. and Liu, W.K., “Reproducing Kernel Particle Methods for
Large Deformation Analysis of Nonlinear Structures,” Comput. Methods Appl. Mech.
Engrg. 139, 195-227 (1996).

Chen, J.S. and Wu, C.T., “Generalized Nonlocal Meshfree Method in Strain
Localization,” Proceeding of International Conference on Computational Engineering Science,
Atlanta, Georiga, 6-9 October (1998).

Chen, J.S. and Wang, H.P. “New Boundary Condition Treatmens in Meshfree
Computation of Contact Problems,” Computer Methods in Applied Mechanics and
Engineering, 187, 441-468, (2001a).

Chen, J.S., Wu, C.T., Yoon, S. and You, Y. “A Stabilized Conforming Nodal Integration
for Galerkin Meshfree Methods,” Int. J. Numer. Methods Engrg. 50, 435-466 (2001b).

Chen, W.F., and Baladi, G.Y., Soil Plasticity: Theory and Implementation, Elesvier, New
York, (1985).

43-6 (References) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual References

Christensen, R.M. “A Nonlinear Theory of Viscoelasticity for Application to


Elastomers,” Journal of Applied Mechanics, 47, American Society of Mechanical Engineers,
762-768 (December 1980).

Chung, K., and K. Shah, “Finite Element Simulation of Sheet Metal Forming for Planar
Anisotropic Metals,” Int. J. of Plasticity, 8, 453-476 (1992).

Cochran, S.G., and Chan, J., “Shock Initiation and Detonation Models in One and Two
Dimensions,” University of California, Lawrence Livermore National Laboratory, Rept.
UCID-18024 (1979).

Cohen, M., and Jennings, P.C., “Silent Boundary Methods for Transient Analysis”, in
Computational Methods for Transient Analysis, T. Belytschko and T.J.R. Hughes,
editors, North-Holland, New York, 301-360 (1983).

Cook, R. D., Concepts and Applications of Finite Element Analysis, John Wiley and
Sons, Inc. (1974).

Couch, R., Albright, E. and Alexander, “The JOY Computer Code,” University of
California, Lawrence Livermore National Laboratory, Rept. UCID-19688 (1983).

Cray Research Inc., “CF77 Compiling System, Volume 4: Parallel Processing Guide,”
SG-3074 4.0, Mendota Heights, MN (1990).

Crisfield M.A., Non-linear Finite Element Analysis of Solids and Structures, Volume 2,
Advanced Topics, John Wiley, New York (1997).

DeBar, R. B., “Fundamentals of the KRAKEN Code,” Lawrence Livermore Laboratory,


UCIR-760 (1974).

Deshpande, V.S. and N.A. Fleck, “Isotropic Models for Metallic Foams,” Journal of the
Mechanics and Physics of Solids, Vol. 48, 1253-1283, (2000).

Dobratz, B.M., “LLNL Explosives Handbook, Properties of Chemical Explosives and


Explosive Simulants,” University of California, Lawrence Livermore National
Laboratory, Rept. UCRL-52997 (1981).

Duff, I. S., and Reid, J. K., "The Multifrontal Solution of Indefinite Sparse Symmetric
Linear Equations," ACM Transactions of Mathematical Software, 9, 302-325 (1983).

Dvorkin, E.N. and Bathe, K.J. “A continuum mechanics based four-node shell element
for general nonlinear analysis,” International Journal for Computer-Aided Engineering
and Software, 1, 77-88 (1984).

LS-DYNA DEV 10/27/16 (r:8004) 43-7 (References)


References LS-DYNA Theory Manual

Englemann, B.E. and Whirley, R.G., “A New Explicit Shell Element Formulation for
Impact Analysis,” In Computational Aspects of Contact Impact and Penetration, Kulak,
R.F., and Schwer, L.E., Editors, Elmepress International, Lausanne, Switzerland, 51-90
(1991).

Englemann, B.E., Whirley, R.G., and Goudreau, G.L., “A Simple Shell Element
Formulation for Large-Scale Elastoplastic Analysis,” In Analytical and Computational
Models of Shells, Noor, A.K., Belytschko, T., and Simo, J.C., Eds., CED-Vol. 3, ASME,
New York, New York (1989).

Farhoomand, I., and Wilson, E.L., “A Nonlinear Finite Element Code for Analyzing the
Blast Response of Underground Structures,” U.S. Army Waterways Experiment Station,
Contract Rept. N-70-1 (1970).

Feng, W.W., Private communication, Livermore, CA (1993).

Flanagan, D.P. and Belytschko, T., “A Uniform Strain Hexahedron and Quadrilateral
and Orthogonal Hourglass Control,” Int. J. Numer. Meths. Eng. 17, 679-706 (1981)

Fleck, J.T., “Validation of the Crash Victim Simulator,” I - IV, Report No. DOT-HS-806
279 (1981).

Fleischer M., Borrvall T., and Bletzinger K-U., “Experience from using recently
implemented enhancements for Material 36 in LS-DYNA 971 performing a virtual
tensile test”, 6th European LS-DYNA Users Conference, Gothenburg, Sweden, 2007.

“Fundamental Study of Crack Initiation and Propagation,” Author unknown, LLNL


report, document received from LSTC (2003).

Galbraith, P.C., M.J. Finn, S.R. MacEwen, et.al., "Evaluation of an LS-DYNA3D Model
for Deep-Drawing of Aluminum Sheet", FE-Simulation of 3-D Sheet Metal Forming
Processes in Automotive Industry, VDI Berichte, 894, 441-466 (1991).

Ghanem RG, Spanos PD. Stochastic Finite Elements – A Spectral Approach, Springer-
Verlag, 1991, Revised Edition, Dover, 2003.

Ginsberg, M. and Johnson, J.P., “Benchmarking the Performance of Physical Impact


Simulation Software on Vector and Parallel Computers,” Proc. of the Supercomputing
88: Vol. II, Science and Applications, Computer Society Press (1988).

Gingold, R.A., and Monaghan, J.J., “Smoothed Particle Hydrodynamics: Theory and
Application to Non-Spherical Stars,” Mon. Not. R. Astron. Soc. 181, 375-389 (1977).

43-8 (References) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual References

Ginsberg, M., and Katnik, R.B., “Improving Vectorization of a Crashworthiness Code,”


SAE Technical Paper No. 891985, Passenger Car Meeting and Exposition, Dearborn, MI
(1989).

Giroux, E.D., “HEMP User's Manual,” University of California, Lawrence Livermore


National Laboratory, Rept. UCRL-51079 (1973).

Govindjee, S., Kay G.J., and Simo, J.C., “Anisotropic Modeling and Numerical
Simulation of Brittle Damage in Concrete,” Report Number UCB/SEM M-94/18,
University of California at Berkeley, Department of Civil Engineering (1994).

Govindjee, S., Kay G.J., and Simo, J.C., “Anisotropic Modeling and Numerical
Simulation of Brittle Damage in Concrete,” International Journal for Numerical Methods in
Engineering, Volume 38, pages 3611-3633 (1995).

Green, A.E. and Naghdi, P.M., “A General Theory of Elastic-Plastic Continuum,”


Archive for Rational Mechanics and Analysis, 18, 251 (1965).

Groenwold, A.A. and Stander, N. An efficient 4-node 24 d.o.f. thick shell finite
element with 5-point quadrature. Eng. Comput. 12(8), 723-747, 1995.

Grimes, Roger, Lewis, John G., and Simon, Horst D., "A Shifted Block Lanczos
Algorithm for Solving Sparse Symmetric Generalized Eigenproblems," SIAM Journal of
Matrix Analyis and Applications, 15, 228-272 (1994).

Guccione, J., A. McCulloch, and L. Waldman, "Passive Material Properties of Intact


Ventricular Myocardium Determined from a Cylindrical Model," published in the
ASME Journal of Biomechanical Engineering, 113, 42-55 (1991).

Hallquist, J.O., “Preliminary User’s Manuals for DYNA3D and DYNAP (Nonlinear
Dynamic Analysis of Solids in Three Dimension),” University of California, Lawrence
Livermore National Laboratory, Rept. UCID-17268 (1976) and Rev. 1 (1979).[a]

Hallquist, J.O., “A Procedure for the Solution of Finite Deformation Contact-Impact


Problems by the Finite Element Method,” University of California, Lawrence Livermore
National Laboratory, Rept. UCRL-52066 (1976).

Hallquist, J.O., “A Numerical Procedure for Three-Dimensional Impact Problems,”


American Society of Civil Engineering, Preprint 2956 (1977).

Hallquist, J.O., “A Numerical Treatment of Sliding Interfaces and Impact,” in: K.C. Park
and D.K. Gartling (eds.) Computational Techniques for Interface Problems, AMD, 30,
ASME, New York (1978).

LS-DYNA DEV 10/27/16 (r:8004) 43-9 (References)


References LS-DYNA Theory Manual

Hallquist, J.O., “NIKE2D: An Implicit, Finite-Element Code for Analyzing the Static and
Dynamic Response of Two-Dimensional Solids,” University of California, Lawrence
Livermore National Laboratory, Rept. UCRL-52678 (1979).[b]

Hallquist, J.O. (1998), LS-DYNA Theoretical Manual, May (1998).

Hallquist, J.O., “User's Manual for DYNA2D— An Explicit Two-Dimensional


Hydrodynamic Finite Element Code with Interactive Rezoning,” University of
California, Lawrence Livermore National Laboratory, Rept. UCID-18756 (1980).

Hallquist, J.O., “User's Manual for DYNA3D and DYNAP (Nonlinear Dynamic
Analysis of Solids in Three Dimensions),” University of California, Lawrence Livermore
National Laboratory, Rept. UCID-19156 (1981).[a]

Hallquist, J. O., “NIKE3D: An Implicit, Finite-Deformation, Finite-Element Code for


Analyzing the Static and Dynamic Response of Three-Dimensional Solids,” University
of California, Lawrence Livermore National Laboratory, Rept. UCID-18822 (1981).[b]

Hallquist, J.O., “DYNA3D User's Manual (Nonlinear Dynamic Analysis of Solids in


Three Dimensions),” University of California, Lawrence Livermore National
Laboratory, Rept. UCID-19156 (1982; Rev. 1: 1984; Rev. 2: 1986).

Hallquist, J.O., “Theoretical Manual for DYNA3D,” University of California, Lawrence


Livermore National Laboratory, Rept. UCID-19401 (March, 1983).

Hallquist, J.O., “DYNA3D User's Manual (Nonlinear Dynamic Analysis of Solids in


Three Dimensions),” University of California, Lawrence Livermore National
Laboratory, Rept. UCID-19156 (1988, Rev. 4).

Hallquist, J.O., “DYNA3D User's Manual (Nonlinear Dynamic Analysis of Solids in


Three Dimensions),” Livermore Software Technology Corporation, Rept. 1007 (1990).

Hallquist, J.O., LS-DYNA Keyword User’s Manual, version 970, Livermore Software
Technology Corporation (April, 2003).

Hallquist, J.O., Benson, D.J., and Goudreau, G.L., “Implementation of a Modified


Hughes-Liu Shell into a Fully Vectorized Explicit Finite Element Code,” Proceedings of
the International Symposium on Finite Element Methods for Nonlinear Problems,
University of Trondheim, Trondheim, Norway (1985).

Hallquist, J.O., and Benson, D.J., “A Comparison of an Implicit and Explicit


Implementation of the Hughes-Liu Shell,” Finite Element Methods for Plate and Shell

43-10 (References) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual References

Structures, T.J.R. Hughes and E. Hinton, Editors, 394-431, Pineridge Press Int., Swanea,
U.K. (1986).

Hallquist, J.O., and Benson, D.J., “DYNA3D User’s Manual (Nonlinear Dynamic
Analysis of Solids in Three Dimensions),” University of California, Lawrence Livermore
National Laboratory, Rept. UCID-19156 (1986, Rev. 2).

Hallquist, J.O. and Benson, D.J., “DYNA3D User’s Manual (Nonlinear Dynamic
Analysis of Solids in Three Dimensions),” University of California, Lawrence Livermore
National Laboratory, Rept. UCID-19156 (1987, Rev. 3).

Hallquist, J.O., Stillman, D.W., Hughes, T.J.R., and Tarver, C.,”Modeling of Airbags
Using MVMA/DYNA3D,” LSTC Report (1990).

Harten, A., “ENO Schemes with Subcell Resolution,” J. of Computational Physics, 83,
148-184, (1989).

Hashin, Z, “Failure Criteria for Unidirectional Fiber Composites,” Journal of Applied


Mechanics, 47, 329 (1980).

Herrmann, L.R. and Peterson, F.E., “A Numerical Procedure for Viscoelastic Stress
Analysis,” Seventh Meeting of ICRPG Mechanical Behavior Working Group, Orlando,
FL, CPIA Publication No. 177 (1968).

Hill A.V., "The heat of shortening and the dynamic constants of muscle," Proc Roy Soc
B126:136-195, (1938).

Hill, R., “A Theory of the Yielding and Plastic Flow of Anisotropic Metals,” Proceedings
of the Royal Society of London, Series A., 193, 281 (1948).

Hill, R. “Consitiutive Modeling of Orthotropic Plasticity in Sheet Metals,” Journal of


Mechanics and Physics of Solids, 38, Number 3, 405-417 (1999).

Holmquist, T.J., G.R. Johnson, and W.H. Cook, "A Computational Constitutive Model
for Concrete Subjected to Large Strains, High Strain Rates, and High Pressures, pp. 591-
600, (1993).

Holtz R.D., W. D. Kovacs, “ An Introduction to Geotechnical Engineering”, Prentice


Hall, Inc. New Jersey, 1981

Hopperstad, O.S. and Remseth, S., “A Return Mapping Algorithm for a Class of Cyclic
Plasticity Models,” International Journal for Numerical Methods in Engineering, 38, 549-564
(1995).

LS-DYNA DEV 10/27/16 (r:8004) 43-11 (References)


References LS-DYNA Theory Manual

Hughes, T.J.R., The Finite Element Method, Linear Static and Dynamic Finite Element
Analysis, Prentice-Hall Inc., Englewood Cliffs, NJ (1987).

Hughes, T.J.R., “Generalization of Selective Integration Procedures to Anisotropic and


Nonlinear Media,” Int. J. Numer. Meth. Eng. 15, 9 (1980).

Hughes, T.J.R. and Carnoy, E., “Nonlinear Finite Element Shell Formulation
Accounting for Large Membrane Strains,” AMD-Vol.48, ASME, 193-208 (1981).

Hughes, T.J.R. and Liu, W.K., “Nonlinear Finite Element Analysis of Shells: Part I.
Two-Dimensional Shells.” Comp. Meths. Appl. Mechs. 27, 167-181 (1981).

Hughes, T.J.R. and Liu, W.K., “Nonlinear Finite Element Analysis of Shells: Part II.
Three-Dimensional Shells.” Comp. Meths. Appl. Mechs. 27, 331-362 (1981).

Hughes, T.J.R., Liu,W.K., and Levit, I., “Nonlinear Dynamics Finite Element Analysis of
Shells.” Nonlinear Finite Element Analysis in Struct. Mech., Eds. W. Wunderlich, E.
Stein, and K.J. Bathe, Springer-Verlag, Berlin, 151–168 (1981).

Hughes, T.J.R., Taylor, R. L., Sackman, J. L., Curnier, A.C., and Kanoknukulchai, W.,
“A Finite Element Method for a Class of Contact-Impact Problems,” J. Comp. Meths.
Appl. Mechs. Eng. 8, 249-276 (1976).

Hughes, T.J.R., and Winget, J., “Finite Rotation Effects in Numerical Integration of Rate
Constitutive Equations Arising in Large-Deformation Analysis,” Int. J. Numer. Meths.
Eng., 15, 1862-1867, (1980).

Hulbert, G.M. and Hughes, T.J.R., “Numerical Evaluation and Comparison of


Subcycling Algorithms for Structural Dynamics,” Technical Report, DNA-TR-88-8,
Defense Nuclear Agency, Washington, DC (1988).

Ibrahimbegovic, A. and Wilson, E.L. A unified formulation for triangular and


quadrilateral flat shell finite elements with six nodal degrees of freedom, Comm. Applied
Num. Meth, 7, 1-9, 1991.

Isenberg, J., D.K. Vaughn, and I. Sandler, "Nonlinear Soil-Structure Interaction," EPRI
Report MP-945, Weidlinger Associates, December (1978).

Jabareen, M., and Rubin, M.B., A Generalized Cosserat Point Element (CPE) for Isotropic
Nonlinear Elastic Materials including Irregular 3-D Brick and Thin Structures, J. Mech.
Mat. And Struct., Vol 3-8, 1465-1498 (2008).

43-12 (References) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual References

Jabareen, M., Hanukah, E. and Rubin, M.B., A Ten Node Tetrahedral Cosserat Point
Element (CPE) for Nonlinear Isotropic Elastic Materials, J. Comput. Mech. 52, 257-285
(2013).

Jaumann, G. “Geschlossenes System Physikalischer und Chemischer Differentialdesefz,


Sitz Zer. Akad. Wiss. Wein, (IIa), 120, 385 (1911).

Johnson, G.C. and Bammann D.J., “A Discussion of Stress Rates in Finite Deformation
Problems" International Journal of Solids and Structures, 20, 725-737 (1984).

Johnson, G.R. and Cook,W. H., “A Constitutive Model and Data for Metals Subjected
to Large Strains, High Strain Rates and High Temperatures,” presented at the Seventh
International Symposium on Ballistics, The Hague, The Netherlands, April (1983).

Johnson, G.R. and T.J. Holmquist, "An Improved Computational Model for Brittle
Materials" in High-Pressure Science and Technology - 1993 American Institute of
Physics Conference Proceedings 309 (c 1994) pp.981-984 ISBN 1-56396-219-5.

Johnson, C., Navert, U., and Pitkaranta, J., “Finite Element Methods for Linear
Hyperbolic Problems,” Computer Methods in Applied Mechanics and Engineering, 45,
285-312, (1984).

Jones, N., “Structural Aspects of Ship Collisions,” Chapter 11, in Structural


Crashworthiness, Eds. N. Jones and T Wierzbicki, Butterworths, London, 308-337
(1983).

Jones, R.M., Mechanics of Composite Materials, Hemisphere Publishing Co., New York
(1975).

Ju J.W., “Energy-Based Coupled Elastoplastic Damage Models at Finite Strains”, J. of


Eng Mech., Vol 115, No 11, Nov 1989.

Karypis, G., and Kumar V., "METIS: A Software Package for Partitioning Unstructured
Graphs, Partitioning Meshes, and Computing Fill-Reducing Orderings of Sparse
Matrices," Department of Computer Science, University of Minnesota (1998).

Katz, J., and Maskew, B., “Unsteady Low-Speed Aerodynamic Model for Complete
Aircraft Configurations,” Journal of Aircraft 25, 4 (1988).

Kenchington, G. J., “A Non-Linear Elastic Material Model for DYNA3D,” Proceedings


of the DYNA3D Users Group Conference, September 1988, published by Boeing
Computer Services (Europe) Limited.

LS-DYNA DEV 10/27/16 (r:8004) 43-13 (References)


References LS-DYNA Theory Manual

Kennedy, J. M., Belytschko,T., and Lin, J. I., “Recent Developments in Explicit Finite
Element Techniques and their Applications to Reactor Structures,” Nuclear Engineering
and Design 97, 1-24 (1986).

Kim, N., Seo, K., and Kim, M., “Free Vibration and Spatial Stability of Non-symmetric
Thin-Walled Curved Beams with Variable Curvatures,” International Journal of Solids and
Structures, 40, 3107-3128, (2003).

Klisinski M., “Degradation and Plastic Deformation of Concrete”, Ph.D. thesis, Polish
Academy of Sciences, 1985, IFTR report 38.

Kosloff, D. and Frazier, G.A.,“Treatment of Hourglass Patterns in Low Order Finite


Element

Codes,” Int. J. Numer. Anal. Meth. Geomech. 2, 57-72 (1978)

Kretschmann, D. E., and David W. Green, “Modeling Moisture Content-Mechanical


Property Relationships for Clear Southern Pine,” Wood and Fiber Science, 28(3), pp.
320-337 (1996).

Kreyszig, E., Advanced Engineering Mathematics, John Wiley and Sons, New York,
New York (1972).

Krieg, R.D.,”A Simple Constitutive Description for Cellular Concrete,” Sandia National
Laboratories, Albuquerque, NM, Rept. SC-DR-72-0883 (1972).

Krieg, R.D. and Key, S.W., Implementation of a Time Dependent Plasticity Theory into
Structural Computer Programs, Vol. 20 of Constitutive Equations in Viscoplasticity:
Computational and Engineering Aspects (American Society of Mechanical Engineers,
New York, N.Y., 1976), 125-137.

Kumar, P., Nukala, V. V., and White, D. W., “A Mixed Finite Element for Three-
dimensional Nonlinear Analysis of Steel Frames,” Computational Methods in Applied
Mechanical Engineering, 193, 2507-2545, (2004).

Lancaster, P. and Salkauskas, K., “Surfaces generated by moving least squares


methods”, Math. Comput. 37, 141-158, (1981).

Landshoff, R., “A Numerical Method for Treating Fluid Flow in the Presence of
Shocks,” Los Alamos Scientific Laboratory, Rept. LA-1930 (1955).

Lee, E.L. and Tarver, C.M., “Phenomenological Model of Shock Initiation in


Heterogeneous Explosives,” Phys. of Fluids, 23, 2362 (1980).

43-14 (References) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual References

Lemaitre, J., A Course on Damage Mechanics. Springer-Verlag, (1992).

Lemmen, P.P.M and Meijer, G.J., “Failure Prediction Tool Theory and User Manual,”
TNO Building and Construction Research, The Netherlands, Rept. 2000-CMC-R0018,
(2001).

Lewis B.A., “Developing and Implementing a Road Side Safety Soil Model into LS-
DYNA”, FHWA Research and Development Turner-Fairbank Highway Research
Center. Oct, 1999.

Librescu, L. Qin, Z., and Ambur D.R., “Implications of Warping Restraint on Statics
and Dynamics of Elastically Tailored Thin-Walled Composite Beams,” International
Journal of Mechanical Sciences, 45, 1247-1267 (2003).

Liu, Joseph W. H., "Modification of the Minimum-Degree Algorithm by Multiple


Elimination," ACM Transactions on Mathematical Software, 11, 141-153 (1985).

Liu, W.K. and Belytschko, T., “Efficient Linear and Nonlinear Heat Conduction with a
Quadrilateral Element,” Int. J. Num. Meths. Engrg., 20, 931–948 (1984).

Liu, W.K., Belytschko, T., Ong, J.S.J. and Law, E., “Use of Stabilization Matrices in
Nonlinear Finite Element Analysis,” Engineering Computations, 2, 47–55 (1985).

Liu, W.K., Lecture Notes on Nonlinear Finite Element Methods, Northwestern


University, (1983, revised in 1992).

Liu, W.K., Guo, Y., Tang, S. and Belytschko T., “A Multiple-Quadrature Eight-node
Hexahedral Finite Element for Large Deformation Elastoplastic Analysis,” Comput.
Meths. Appl. Mech. Engrg., 154, 69–132 (1998).

Liu, W.K., Hu, Y.K., and Belytschko, T., “Multiple Quadrature Underintegrated Finite
Elements,” Int. J. Num. Meths. Engrg., 37, 3263–3289 (1994).

Liu, W. K., Chang, H., and Belytschko, T., “Arbitrary Lagrangian-Eulerian Petrov-
Galerkin Finite Elements for Nonlinear Continua,” Computer Methods in Applied
Mechanics and Engineering, to be published.

Liu, W.K., Jun, S. and Zhang, Y.F., “Reproducing Kernel Particle Method,” Int. J.
Numer. Methods Fluids 20, 1081-1106 (1995).

Lucy, L.B., “Numerical Approach to Testing the Fission Hyphothesis,” Astron. J. 82,
1013-1024 (1977).

LS-DYNA DEV 10/27/16 (r:8004) 43-15 (References)


References LS-DYNA Theory Manual

Lysmer, J. and Kuhlemeyer, R.L., “Finite Dynamic Model for Infinite Media”, J. Eng.
Mech. Div. ASCE, 859-877 (1969).

MacNeal R.H. and Harder R.L., “A Proposed Standard Set of Problems to Test Finite
Element Accuracy,” Finite Elements Anal. Des., (1985, 3–20 (1985).

Maenchen, G. and Sack, S., “The Tensor Code,” Meth. Comp. Phys. 3, (Academic
Press), 181-263 (1964).

Maffeo, R., “TRANCITS Program User’s Manual,” General Electric Company,


Cincinnati, OH, NASA Report CR-174891, (1985).

Maffeo, R., “Burner Liner Thermal/Structural Load Modeling,” General Electric


Company, Cincinnati, OH, NASA Report CR-174892, (1984).

Maker, B. N., Private communication, Lawrence Livermore National Laboratory. Dr.


Maker programmed and implemented the compressible Mooney-Rivlin rubber model.

Marchertas, A. H., and Belytschko,T. B., “Nonlinear Finite Element Formulation for
Transient Analysis of Three Dimensional Thin Structures,” Report ANL-8104, LMFBR
Safety, Argonne National Laboratory, Argonne, IL (1974).

Margolin, L. G., personal communication to D. Benson (1989).

Maskew, B., “Program VSAERO Theory Document,” NASA CR-4023 (1987).

Matthies, H., and G. Strang, The Solution of Nonlinear Finite Element Equations, Int.,
Journal for Numerical Methods in Engineering, 14, No. 11, 1613-1626.

Matzenmiller, A., and Schweizerhof, K., “Crashworthiness Simulatios of Composite


Structures - A First Step with Explicit Time Integration,” Nonlinear Computational
mechanics - A State of the Art, edited by P.W. Wriggers, et. al., Springer-Verlag, (1991).

Mauldin, P.J., R.F. Davidson, and R.J. Henninger, “Implementation and Assessment of
the Mechanical-Threshold-Stress Model Using the EPIC2 and PINON Computer
Codes,” Report LA-11895-MS, Los Alamos National Laboratory (1990).

McGlaun, personal communication to D. Benson, Sandia National Laboratories, (1990).

McGlaun, J. M., Thompson, S. L., and Elrick, M. G., ”CTH: A Three-Dimensional


Shock Wave Physics Code,” Proc. of the 1989 Hypervelocity Impact Symposium (1989).

Mindlin, R.D., “Influence of Rotary Inertia and Shear on Flexural Motions of Isotropic,
Elastic Plates,” J. Appl. Mech. 18, 31-38 (1951).

43-16 (References) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual References

Mizukami, A., and Hughes, T. J. R., “A Petrov-Galerkin Finite Element Method for
Convection Dominated Flows: An Accurate Upwinding Technique for Satisfying the
Maximum Principle,” Computer Methods in Applied Mechanics and Engineering, Vol.
50, pp. 181-193 (1985).

Monaghan, J.J., and Gingold, R.A., “Shock Simulation by the Particle Method of SPH,”
Journal of Computational Physics, 52, 374-381 (1983).

Murray Y.D., “Modeling Rate Effects in Rock and Concrete”, Proceedings of the 8th
International Symposium on Interaction of the Effects of Munitions with Structures,
Defense Special Weapons Agency, McLean VA, USA, (1997).

Murray, Y. D., “Modeling Rate Effects in Rock and Concrete,” Proceedings of the 8th
International Symposium on Interaction of the Effects of Munitions with Structures,
Defense Special Weapons Agency, McLean, VA, USA, (1997).

Murray, Y.D., Users Manual for Transversely Isotropic Wood Model, aptek technical
report to fhwa (to be published), (2002).

Murray, Y.D., and J. Reid, Evaluation of Wood Model for Roadside Safety Applications,
aptek technical report to fhwa (to be puslished), (2002).

Nagtegaal, J.C., Parks, D.M., and Rice J.R., “On Numerically Accurate Finite Element
Solution in the Fully Plastic Range”, Computer Methods in Applied Mechanics and
Engineering, 4, 153 (1974).

Newman, W.M., and Sproull, R.F., Principles of Interactive Computer Graphics,


McGraw-Hill, New York (1979).

Newmark, N., “A Method of Computation for Structural Dynamics,” Journal of the


Engineering Mechanics Division, Proceedings of the American Society of Civil
Engineers, 67-94 (1959).

Neilsen, M.K., Morgan, H.S., and Krieg, R.D., “A Phenomenological Constitutive Model
for Low Density Polyurethane Foams,” Rept. SAND86-2927, Sandia National
Laboratories, Albuquerque, NM (1987).

Noh, W.F., Meth. Comp. Phys. 3, (Academic Press) (1964).

Noh, W.F., “Numerical Methods in Hydrodynamic Calculations,” University of


California, Lawrence Livermore National Laboratory, Rept. UCRL-52112 (1976).

Oden, J.T, P. Devloo, and T. Strouboulis, "Adaptive Finite Element Methods for the
Analysis of Inviscid Compressible Flow: Part I. Fast Refinement/Unrefinement and
LS-DYNA DEV 10/27/16 (r:8004) 43-17 (References)
References LS-DYNA Theory Manual

Moving Mesh Methods for Unstructured Meshes," Computer Methods in Applied


Mechanics and Engineering, 59, 327-362 (1986).

Ogden, R.W., Non-Linear Elastic Deformations, Ellis Horwood Ltd., Chichester, Great
Britian (1984).

Okuda T., Yamamae Y. and Yasuki T., Request for MAT126 Modification, Microsoft
Power Point presentation, Toyota Communication Systems and Toyota Motor
Corporation, 2003.

Oliver, J., “A Consistent Characteristic Length of Smeared Cracking Models,”


International Journal for Numerical Methods in Engineering, 28, 461-474 (1989).

Rajendran, A.M.,"Material Failure Models", in Material Behavior at High Strain Rates,


Course Notes, Computational Mechanics Associates, Monterey, CA, 424 (1989).

Papadrakakis, M., “A Method for the Automated Evaluation of the Dynamic Relaxation
Parameters,” Comp. Meth. Appl. Mech. Eng. 25, 35-48 (1981).

Phoon KK, Huang SP, Quek ST. Implementation of Karhunen–Loeve expansion for
simulation using a wavelet-Galerkin scheme. Probabilist Eng Mech 2002;17(3):293–303.
(2002a)

Phoon KK, Huang SP, Quek ST. Simulation of second-order processes using
Karhunen–Loeve expansion. Comput Struct 2002; 80(12):1049–60. (2002b)

Phoon KK, Huang HW, Quek ST. Simulation of strongly non-Gaussian processes using
Karhunen–Loeve expansion, Probabilistic Engineering Mechanics 20 (2005) 188–198.

Pian, T.H.H., and Sumihara, K., “Rational Approach for Assumed Stress Elements,” Int.
J. Num. Meth. Eng., 20, 1685-1695 (1985).

Prokic, A., “New Warping Function for Thin-Walled Beams. I: Theory,” Journal of
Structural Engineering, 122, 1437-1442, (1994).

Przemieniecki, J. S., Theory of Matrix Structural Analysis, McGraw-Hill Book


Company, New York, NY (1986).

Reid, S.R. and C. Peng, “Dynamic Uniaxial Crushing of Wood,” Int. J. Impact
Engineering, Vol. 19, No. 5-6, pp. 531-570, (1997).

Reyes, A., O.S. Hopperstad, T. Berstad, and M. Langseth, “Implementation of a


Material Model for Aluminium Foam in LS-DYNA”, Report R-01-02, Restricted,

43-18 (References) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual References

Department of Structural Engineering, Norwegian University of Science and


Technology, (2002).

Richards, G.T., “Derivation of a Generalized von Neumann Pseudo-Viscosity with


Directional Properties,” University of California, Lawrence Livermore National
Laboratory, Rept. UCRL-14244 (1965).

Richtmyer, R.D., and Morton, K.W., Difference Equations for Initial-Value Problems,
Interscience Publishers, New York (1967).

Sandler, I.S., and D. Rubin, “An Algorithm and a modular subroutine for the cap
model,” Int. J. Numer. Analy. Meth. Geomech., 3, 173-186 (1979).

Schwer, L.E., Cheva, W., and Hallquist, J.O., “A Simple Viscoelastic Model for Energy
Absorbers Used in Vehicle-Barrier Impacts,” In Computational Aspects of Contact
Impact and Penetration, Kulak, R.F., and Schwer, L.E., Editors, Elmepress International,
Lausanne, Switzerland, 99-117 (1991).

Shapiro, A.B., “TOPAZ3D - A Three Dimensional Finite Element Heat Transfer Code,”
University of California, Lawrence Livermore National Laboratory, Report UCID-20481
(1985).

Shapiro, A.B., “REMAP: a Computer Code That Transfers Node Information Between
Dissimilar Grids,” Lawrence Livermore National Laboratory, UCRL-ID-104090, (1990).

Shkolnikov, M.B. Private Communication (1990-1991).

Simo, J.C. and Govindjee, S., ``Exact Closed-Form Solution of the Return Mapping
Algorithm in Plane Stress Elasto-Viscoplasticity," Engineering Computations, 5, 254-258
(1988).

Simo, J.C. and Hughes, T.J.R. “On the variational foundations of assumed strain
methods,” Journal of Applied Mechanics, 53, 1685-1695 (1986).

Simo, J.C., Ju, J.W., “Stress and Strain Based Continuum Damage Models”, Parts I & II,
Int. J. of Solids and Structures, Vol 23, No 7, (1987).

Simo, J.C., Ju, J.W., Pister, K.S., and Taylor, R.L. “An Assessment of the Cap Model:
Consistent Return Algorithms and Rate-Dependent Extension,” J. Eng. Mech., 114,
No. 2, 191-218 (1988a).

Simo, J.C., Ju, J.W., Pister, K.S., and Taylor, R.L. “Softening Response, Completeness
Condition, and Numerical Algorithms for the Cap Model,” Int. J. Numer. Analy.
Meth. Eng. (1990).
LS-DYNA DEV 10/27/16 (r:8004) 43-19 (References)
References LS-DYNA Theory Manual

Simo, J.C. and Taylor, R.L., “A Return Mapping Algorithm for Plane Stress
Elastoplasticity,” International Journal for Numerical Methods in Engineering, 22, 649-670
(1986).

Steinberg, D.J. and Guinan, M.W., “A High-Strain-Rate Constitutive Model for Metals,”
University of California, Lawrence Livermore National Laboratory, Rept. UCRL-80465
(1978).

Stillman, D. W., and Hallquist, J. O., “LS-INGRID: A Pre-Processor and Three-


Dimensional Mesh Generator for the Programs LS-DYNA, LS-NIKE3D, and
TOPAZ3D,” Version 3.1, Livermore Software Technology Corporation (1992).

Stojko, S., privated communication, NNC Limited, Engineering Development Center


(1990).

Stolarski, H., and Belytschko, T., “Shear and Membrane Locking in Curved Elements,”
Comput. Meths. Appl. Mech. Engrg., 41, 279–296 (1983).

Stone, C.M., Krieg, R.D., and Beisinger, Z.E., “Sancho, A Finite Element Computer
Program for the Quasistatic, Large Deformation, Inelastic Response of Two-
Dimensional Solids,” Sandia Report, SAND 84-2618, UC-32, Albuquerque, NM (1985).

Storakers, B., “On Material Representation and Constitutive Branching in Finite


Compressible Elasticity”, Royal Institute of Technology, Stockholm Sweden, (1985).

Sturt, R.M.V., and B.D. Walker, J.C. Miles, A. Giles, and N. Grew, “Modelling the
Occupant in a Vehicle Context-An Integrated Approach,” 13th International ESV
Conference, Paris, November 4-7 (1991).

Sussman, T. and Bathe, K.J., “A Finite Element Formulation for Nonlinear


Incompressible Elastic and Inelastic Analysis,” Computers & Structures, 26, Number 1/2,
357-409 (1987).

Tarver, C.M., and Hallquist, J.O., “Modeling of Two-Dimensional Shock Initiation and
Detonation Wave Phenomena in PBX 9404 and LX-17,” University of California,
Lawrence Livermore National Laboratory, Rept. UCID84990 (1981).

Taylor, R.L. and Simo, J.C. Bending and membrane elements for the analysis of thick
and thin shells, Proc. of NUMETA Conference, Swansea, 1985.

Taylor, R.L., Finite element anlysis of linear shell problems, in Whiteman, J.R. (ed.),
Proc. of the Mathematics in Finite Elements and Applications, Academic Press, New York,
191-203 (1987).

43-20 (References) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual References

Taylor, L.M., and Flanagan, D.P., “PRONTO3D A Three-Dimensional Transient Solid


Dynamics Program,” Sandia Report: SAND87-1912, UC-32 (1989).

Thompson, S. L., “CSQ -- A Two Dimensional Hydrodynamic Program with Energy


Flow and Material Strength,” Sandia Laboratories, SAND74-0122 (1975).

Thompson, R., L., and Maffeo, R. L., “A Computer Analysis Program for Interfacing
Thermal and Structural Codes,” NASA Lewis Research Center, Report NASA-TM-
87021 (1985).

Trefethen, L.N., “Group Velocity in Finite Difference Schemes,” SIAM Review, 24, No.
2 (1982).

Tsai, S.W., and Wu, E.M., “A General Theory of Strength for Anisotropic Materials,”
Journal of Composite Materials, 58-80 (1971).

Tuler, F.R. and B.M. Butcher, "A Criterion for the Time Dependence of Dynamic
Fracture," The International Journal of Fracture Mechanics, 4, No. 4 (1968).

Turk M, Pentland A. Eigenfaces for Recognition, Journal of Cognitive Neuroscience,


1991; 3(1):71-86.

Turkalj, G. Brnic, J., and Prpic-Orsic J., “Large Rotation Analysis of Elastic Thin-Walled
Beam-Type Structures Using ESA Approach,” Computers & Structures, 81, 1851-1864,
(2003).

Underwood, P., “Dynamic Relaxation,” Computational Method for Transient Analysis,


Belytschko, T., and Hughes, T.J.R., Eds., 1, 245-263, (1986).

Van Leer, B., “Towards the Ultimate Conservative Difference Scheme. IV. A New
Approach to Numerical Convection,” Journal of Computational Physics, 23, 276-299
(1977).

Vawter, D., "A Finite Element Model for Macroscopic Deformation of the Lung,"
published in the Journal of Biomechanical Engineering, 102, 1-7, (1980).

von Neumann, J., and Richtmyer, R.D., “A Method for the Numerical Calculation of
Hydrodynamical Shocks,” J. Appl. Phys., 21, 232 (1950).

Walker, B.D., and P.R.B. Dallard, “An integrated Approach to Vehicle Crashworthiness
and Occupant Protection Systems”, SAE International Congress and Exposition, Detroit,
Michigan, (910148), February 25-March 1 (1991).

LS-DYNA DEV 10/27/16 (r:8004) 43-21 (References)


References LS-DYNA Theory Manual

Walker, H.F., Numerical Solution of Nonlinear Equations, University of California,


Lawrence Livermore National Laboratory, Rept. UCID-18285 (1979).

Wang, J. T., and O. J. Nefske, “A New CAL3D Airbag Inflation Model,” SAE paper
880654, (1988).

Wang, J. T., private communication (1992).

Wang, J.T., "An Analytical Model for an Airbag with a Hybrid Inflator", Publication
R&D 8332, General Motors Development Center, Warren, MI (1995).

Wang, J.T., "An Analytical Model for an Airbag with a Hybrid Inflator", AMD-Vol. 210,
BED 30, ASME, 467-497 (1995).

Warsi, Z. U. A., “Basic Differential Models for Coordinate Generation,” in Symposium


on the Numerical Generation of Curvilinear Coordinate Systems, Nashville, Tenn.
(1982).

Whirley, R. G., Hallquist, J. O., and Goudreau, G. L., “An Assessment of Numerical
Algorithms for Plane Stress and Shell Elastoplasticity on Supercomputers,” Engineering
Computations, 6, 116-126 (June, 1989).

Wilkins, M.L., “Calculations of Elastic Plastic Flow,” Meth. Comp. Phys., 3, (Academic
Press), 211-263 (1964).

Wilkins, M. L., “Calculation of Elastic-Plastic Flow,” University of California, Lawrence


Livermore National Laboratory, Rept. UCRL-7322, Rev. I (1969).

Wilkins, M.L., “Use of Artificial Viscosity in Multidimensional Fluid Dynamic


Calculations,” J. Comp. Phys. 36, 281 (1980).

Wilkins, M.L., Blum, R.E., Cronshagen, E., and Grantham, P., “A Method for Computer
Simulation of Problems in Solid Mechanics and Gas Dynamics in Three Dimensions and
Time,” University of California, Lawrence Livermore National Laboratory, Rept.
UCRL-51574 (1974).

Wilson, E.L., Three Dimensional Static and Dynamic Analysis of Structures, A


publication of Computers and Structures, Inc., Berkeley, California, [1996-2000].

Winslow, A. M., “Equipotential Zoning of Two-Dimensional Meshes,” Lawrence


Radiation Laboratory, UCRL-7312 (1963).

Winslow, A. M., “Equipotential Zoning of The Interior of a Three-Dimensional Mesh,”


Report to LSTC (1990).

43-22 (References) LS-DYNA DEV 10/27/16 (r:8004)


LS-DYNA Theory Manual References

Winters, J.M., "Hill-based muscle models: A systems engineering perspective," In


Multiple Muscle Systems: Biomechanics and Movement Organization, JM Winters and
SL-Y Woo eds, Springer-Verlag (1990).

Winters J.M. and Stark L., "Estimated mechanical properties of synergistic muscles
involved in movements of a variety of human joints,": J Biomechanics 21:1027-1042,
(1988).

Woodruff, J.P., “KOVEC User’s Manual,” University of California, Lawrence Livermore


National Laboratory, Rept. UCRL-51079 (1973).

Yen, C.F. and Caiazzo, A, “Innovative Processing of Multifunctional Composite Armor


for Ground Vehicles,” ARL-CR-484, U.S. Army Research Laboratory, Aberdeen Proving
Ground, MD, (2001).

Yunus, S.M., Pawlak, T.P., and Cook, R.D., “ Solid Elements with Rotational Degrees of
Freedom, Part I-Hexahedron Elements,” To be published, (1989).

Zajac F.E., "Muscle and tendon: Properties, models, scaling, and application to
biomechanics and motor control, "CRC Critical Reviews in Biomedical Engineering
17(4):359-411, (1989).

LS-DYNA DEV 10/27/16 (r:8004) 43-23 (References)


References LS-DYNA Theory Manual

43-24 (References) LS-DYNA DEV 10/27/16 (r:8004)

You might also like