Note to the Reader,

Although not updated for Version 68 of MSCINASTRAN, the MSCINASTRAN Handbook for Nonlinear
Analysis, Version 67, is still being offered for a couple of reasons.
The first is that the new MSCINASTRAN Nonlinear Analysis User's Guide for Version 68 IS currently in
progress and not yet available. The goals for this new guide are to present the range of nonlinear analysis
capabilities in a user-friendly, tutorial format and to update the wealth of technical inforrnatlon ava~lablein
the current handbook. It is intended to act as a single-point source of reference for all MSCINASTRAN
nonlinear analysis capabilities.
Secondly, the MSCINASTRAN Handbook for NonlinearAnalysis contains in-depth technical inforrnatlon and
a thorough review of implementation details that are simply unavailable anywhere else. Much of thls
information has been of particular interest to advanced analysts, and as such still enjoys w~deappeal.
In addition to this handbook, we recommended that you also consult the MSCINASTRAN Reference
Manual, Section 15.3, for information pertaining specifically to Version 68 nonlinear analysis capabilitles.

Greg Moore
Manager, Technical Publications
MSCINASTRAN Nonlinear Analysis

HANDBOOK Volume I

Version 67

Sang H. Lee
 Corporate Headquarters
The MacNeal-Schwendler Corporation
815 Colorado Boulevard
Los Angeles, CA 90041-1777
Tel: (213) 258-91 11 or (800) 336-4858
FAX: (213) 259-3838
Headquarters, European Operations
MacNeal-Schwendler GmbH
lnnsbrucker Ring 15
Postfach 801240
D - 81 612 Miinchen, GERMANY
Tel: (89) 431 9870
Telex: 523 784 MSG D
FAX: (89) 436 1716
Headquarters, Far East Operations
MSC Japan Ltd.
Entsuji-Gadelius Building
2-39, Akasaka 5-chome
Minato-ku, Tokyo 107, JAPAN
Tel: (03) 3505-0266
Telex: J23363 MSCWATA
FAX: (03) 3505-091 4

DISCLAIMER

The concepts, methods, and examples presented in this text are for illustrative and educational purposes
only and are not intended to be exhaustive or to apply to any particular engineering problem or design.
The MacNeal-Schwendler Corporation assumes no liability or responsibility to any person or company
I for direct or indirect damages resulting from the use of any information contained herein.
I
a1992 by The MacNeal-SchwendierCorporation
Printed in U.S.A.
All rights reserved.

MSC and MSCI are registered trademarks and service marks of The MacNeal-Schwendler Corporation. NASTRAN is a
registered trademark of the National Aeronautics and Space Administration. MSCINASTRAN is an enhanced, proprietary
version developed and maintained by The MacNeal-SchwendlerCorporation.

NA *V67 * Z * Z * Z * HB-NLN
 PREFACE
This handbook is developed as tutorial material for MSC/NASTRAN users t o solve nonlin-
ear structural problems. It is assumed that the reader have prior knowledge of MSCINASTRAN
fundamentals as well as the proficient knowledge of solid mechanics.

The text covers the entire spectrum of nonlinear structural analysis: statics, dynamics
and stability. Different aspects of nonlinearities are exemplified. The rudimentary theories
are reviewed and extended t o practical problems. Special features of nonlinear finite elements
are highlighted. Some details of the program structure are discussed in terms of algorithms,
modules and DMAP. The handbook is self-contained for nonlinear analysis, providing a full
description of the nonlinear features of MSCINASTRAN with emphasis on the user-interface.

While maintaining consistency and continuous flow throughout the book, efforts are made
to keep each chapter independent of others so that the reader need not read the entire volume
to start. This handbook will be expanded as new capabilities are introduced. The initial
edition of this handbook consists of eleven chapters. It is recommended that the reader review
the first two chapters t o start. Then he may continue to read any chapter on the subject of his
interest. The user interface is collectively described in Chapters 7 and 9 for static and dynamic
analysis, respectively.

MSCINASTRAN nonlinear capabilities have matured through the wide utilization, con-
structive critiques, friendly advice and favorable acceptance of numerous users in our esteemed
clientele community. With this document, MSC responds t o our enthusiastic users for their
patronage with an acknowledgement of their contribution t o the growth of MSCINASTRAN.
I hope that this handbook proves t o be an easy-to-follow tutorial text for novice users and a
comprehensive reference for expert users, clarifying their understanding of our code.

Dr. R. H. MacNeal and Messrs. D. N. Herting and R. L. Harder are acknowledged for
the conception and the initial design of the nonlinear capabilities. Wherever applicable, con-
tributors t o the sections of this handbook involving various authors are listed in the Table
of Contents. Contributions t o the development efforts by many engineers and programmers
whose names are not mentioned here are acknowledged. Thanks are expressed t o A. Raiten,
C. S. Kim and C. J. Curley for their assistance in preparing Figures and Tables. Appreciation
is extended t o K. Blakely and A. Raiten for proofreading the manuscript.

March 1, 1992

Sang H. Lee, Dr. Eng., P.E.

Principal Engineer
 CONTENTS
Note: Contributors of various sections denoted in italics.
Preface

List of Tables and List of F i g u r e s

1 Introduction
1.1 Evolution
1.2 Overview of Nonlinear Capabilities
1.3 Program Architecture
1.4 Rudiments of User Interface
1.4.1 Case Control Section
1.4.2 Bulk Data Section
1.4.3 Parameters
1.4.4 Nonlinear Element Output Codes for Plotting
1.5 Nonlinear Characteristics and General Recommendations

2 Basic C o n s i d e r a t i o n s for Nonlinear Analysis

2.1 Discrete System for a Nonlinear Continuum Model
2.2 Finite Element Formulation for Equilibrium Equations
2.3 Coordinate Transformations
2.4 Displacement Sets and Reduction of Equations
2.5 Nonlinear Solution Procedure

3 Solution M e t h o d s
3.1 Adaptive Solution Strategies
3.2 Newton's Method of Iteration
3.3 Stiffness Update Strategies
3.3.1 Update Principles
3.3.2 Divergence Criteria
3.3.3 Time Expiration Criteria
3.4 Line Search Method
3.4.1 Theoretical Basis
3.4.2 Search Criteria
3.4.3 Implementation of Search Procedure
3.4.4 One-Dimensional Example
3.4.5 Other Provisions for Line Search
3.4.6 Quadratic Interpolation
3.5 Quasi-Newton Method
3.5.1 Evolution of Quasi-Newton Method
3.5.2 Criteria for BFGS Update
3.5.3 Implementation of the BFGS Update Strategy
3.5.4 One-Dimensional Example
 3.5.5 BFGS Updates for Arc-length Methods
3.6 Convergence Criteria
3.6.1 Rudimentary Considerations
3.6.2 Convergence Conditions
3.6.3 Error Functions and Weighted Normalization
3.6.4 Implementation
3.6.5 Some Observations
3.7 Arc-Length Methods for Post-Buckling Analysis
3.7.1 Basic Theory for Arc-Length Methods
3.7.2 Riks' Method and Its Variations
3.7.3 Crisfield's Method
3.7.4 Adaptive Arc-length Method
3.7.5 Verification Problems

4 Finite Elements in Nonlinear Analysis

4.1 Overview
4.1.1 Line Elements
4.1.2 Surface Elements
4.1.3 Solid Elements
4.1.4 Other Elements
4.2 Formulation of Isoparametric Finite Elements
4.2.1 Isoparametric Coordinates
4.2.2 Shape Functions
4.2.3 An Example of Element Matrix
4.2.4 Volume Integration of Element Matrices
4.2.5 Element Loads and Equilibrium
4.2.6 Element Coordinates
4.2.7 Stress Data Recovery
4.3 Gap and Friction Element
4.3.1 Introduction
4.3.2 Basic Considerations for Penalty GAP Element
4.3.3 User Interface for GAP Element
4.3.4 Theoretical Basis for Frictional Behavior
4.3.5 Subincremental Algorithm for Stiffness Update and Bisection
4.3.6 Adaptive Adjustment of Penalty Values
4.3.7 Non-adaptive GAP Element with Anisotropic Friction
4.3.8 Verification and Validation: Bouncing Mass Problem
4.3.9 Verification and Validation: Vibration with Coulomb Damping
4.3.10 Hertzian Problem: Contact Between Sphere and Rigid Plane
4.3.11 Dynamic Case: Impact of a Sphere with Rigid Plane
4.3.12 Recommendations

5 Geometric Nonlinearity
5.1 Overview and User Interface S. H. Lee
 5.2 Updated Element Coordinates S. H . Lee
5.2.1 Concept of Convective Coordinates
5.2.2 Updated Coordinates and Net Deformation
5.2.3 Provisions for Global Operation
5.3 Follower Forces S. H. Lee, D. N . Herting
5.3.1 Basic Definition
5.3.2 Implementation
5.3.3 Dynamic Follower Forces
5.3.4 Verification: Elliptic Cylinder Subject to an Internal Pressure
5.4 Treatment of Large Rotation D. V. Wallerstein,R.Allahabadi
5.4.1 Gimbal Angle Approach
5.4.2 Rotation Vector Approach
5.4.3 Bisection Due to Large Rotations
5.4.4 Output Interpretation
5.5 Verification: Large Rotation of a Beam A . Raiten
5.5.1 Problem Description
5.5.2 Finite Element Model
5.5.3 Analysis Procedure
5.5.4 Analysis Results
5.6 Canonical Approach t o the Geometric Nonlinearity S. H. Lee
5.6.1 Fundamentals of Continuum Mechanics
5.6.2 Incremental Formulations for Large Displacements and Strains
5.6.3 Total Lagrangian vs. Updated Lagrangian in Finite Elements
5.6.4 Constitutive Relations for Large Deformation
6 Material Nonlinearity and Constitutive Relations
6.1 Introduction S. H. Lee
6.1.1 Overview
6.1.2 User Interface
6.2 Plasticity S.H. Lee
6.2.1 Some Preliminaries
6.2.2 Yield Criteria
6.2.3 Yield Function
6.2.4 Strain Hardening
6.2.5 Prandtl-Reuss Stress-Strain Relations
6.2.6 Associated Flow Rule
6.2.7 Generalized Effective Plastic Strain Increment
6.2.8 Yield Function Derivatives
6.2.9 Degenerate Cases: Plane Strain, Plane Stress and Uniaxial Stress
6.2.10 Solution Algorithm for Elasto-Plastic Material
6.2.11 Verification and Validation
6.3 Nonlinear Elasticity S. H. Lee
6.3.1 Theoretical Basis
6.3.2 Solution Alogorithm
 6.3.3 Adaptation of Uniaxial Compression Stress-Strain Curve
6.3.4 Computational Procedure for Bilateral Stress-Strain Relations
6.4 Creep and Viscoelasticity S. H. Lee
6.4.1 Introduction
6.4.2 Effects of Stress and Temperature
6.4.3 Equillibrium of a Kelvin-Maxwell Model for a Stress Component
6.4.4 Adaptation of Rheological Model to Multiaxial Stress State
6.4.5 Coupling of Plasticity
6.4.6 Implementation
6.4.7 Verification and Validation
6.4.8 Nozzle-to-Spherical Shell Attachment as Benchmark Problem
6.5 Thermo-Elasticity S.S. Hsieh
6.5.1 Theoretical Basis
6.5.2 User Interface
6.5.3 Validation Problem
6.6 Initial Anisotropy S.S. Hsieh
6.6.1 Theoretical Basis
6.6.2 Validation Problems

7 Nonlinear Static Analysis

7.1 User Interface S. H. Lee
7.1.1 Case Control
7.1.2 Iteration Control: NLPARM Data
7.1.3 User Interface for Arc-length Methods: NLPCI data
7.1.4 Iteration Related Output Data
7.2 Iteration Module and Solution Sequence S.H.Lee, T.L. Bock
7.2.1 DMAP Interface of NLITER Module
7.2.2 Input Data Blocks
7.2.3 Output Data Blocks
7.2.4 Parameters
7.2.5 Diagnostic Outputs
7.3 Loads Overview D. V. Wallerstein
7.3.1 Concentrated Loads
7.3.2 Distributed Loads
7.3.3 Mass Related Loads
7.3.4 Thermal Loads
7.3.5 Loads Combination
7.4 Nonproportional Loading S.H. Lee
7.4.1 Validation of Nonproportional Loading Case
7.4.2 Biaxial Loading Case For Neutral Loading
7.5 Constraints and Enforced Motion S.S. Hsieh
7.5.1 Boundary Condition
7.5.2 Multipoint Constraint
7.5.3 Enforced Motion
 7.5.4 Example
7.6 Restarts S.S. Hsieh
7.6.1 Preliminaries
7.6.2 Restart Procedure
7.6.3 Example
7.7 Verification Problem: Analysis of a Pressure Vessel

8 Direct Time Integration

8.1 Overview S. H. Lee
8.2 Three-Point Method (NLTRD Module) D. N. Herting
8.2.1 Basic Equations
8.2.2 Nonlinear Iterations
8.2.3 Stability Analysis
8.2.4 DMAP Interface of NLTRD Module
8.3 Two-Point Method (NLTRD2 Module) S.H. Lee
8.3.1 Newmark Integration
8.3.2 Adaptation of Newmark's Method
8.3.3 Start-up Algorithm
8.3.4 DMAP Interface of NLTRD2 Module
8.4 Adaptive Time Stepping S.H. Lee
8.4.1 Introduction
8.4.2 Automatic Time Stepping Algorithm
8.4.3 Bisection Algorithm
8.5 Expedient Iteration Strategies S. H. Lee
8.6 Verification Problems S.H. Lee
8.6.1 Elastic-Plastic Rod with an End Mass Subjected to a Step Loading
8.6.2 Impulsively Loaded Clamped Plate
8.6.3 Simply Supported Beam with a Restrained Motion
8.6.4 Bouncing Weight on an Elastic Platform
8.6.5 Shallow Spherical Cap under a Concentrated Apex Load

9 Nonlinear Transient Response Analysis

9.1 User Interface S.H. Lee
9.1.1 Case Control
9.1.2 Implicit Integration Control: TSTEPNL Data
9.1.3 Iteration Related Output Data
9.2 Restarts S. S. Hsieh
9.2.1 Restarting from SOL 66 into SOL 99
9.2.2 Restarting within SOL 99
9.2.3 Example
9.3 Dynamic Effects: Mass and Damping A. Raiten
9.3.1 Mass Modeling
9.3.2 Mass Matrix
9.3.3 Damping Modeling
 9.3.4 Damping Matrix
9.4 Transient Loads and Initial Conditions D.Herting, S. H. Lee
9.4.1 Transient Loads
9.4.2 Enforced Motion
9.4.3 Initial Conditions
9.5 Diagnostic Outputs T.L. Bock
9.5.1 NLTRD Module for AUTO or TSTEP Method
9.5.2 NLTRD2 Module for ADAPT Method

10 Special Applications
10.1 Nonlinear Buckling Analysis
10.1.1 Introduction
10.1.2 Formulation of Nonlinear Buckling Analysis
10.1.3 Analysis Procedure
10.1.4 Verification Problem: Euler Column Buckling
10.1.5 Example of a Nonlinear Buckling
10.2 Nonlinear Modal Analysis
10.2.1 Analysis Procedure
10.2.2 Bending Vibration of Linear Material Beam
10.2.3 Bending Vibration of Nonlinear Material Beam
10.2.4 Plate Vibration
10.2.5 Beam Bending Vibration with Superelements
10.3 Static Analysis using SOL 99 or 129
10.3.1 Input Data for SOL 99 or 129 Static Analysis
10.3.2 A Z-Shaped Beam Loaded with A Static Force
10.3.3 Rotation of a Beam with Friction

11 Example Problems
11.1 Snap-Through Analysis of a Spherical Diaphragm in SOL 66
11.1.1 Problem Description
11.1.2 Finite Element Modeling and MSCINASTRAN Input Data
11.1.3 Analysis Results
11.1.4 Solution Progression
11.1.5 Concluding Remarks
11.2 Parametric Study on BFGS Updates and Line Search Method in SOL 66
11.2.1 Introduction
11.2.2 The Cologne Challenge: a Z-Shaped Cantilever Beam
11.2.3 Snap-Through and Snap-Back: a Spherical Shell with Backing Plate
11.2.4 Pre-Buckling Behavior: an Imperfect Spherical Cap
11.2.5 Concluding Remarks
11.3 Creep Analysis of a Thick-Walled Pressure Vessel
11.3.1 Introduction
11.3.2 Problem Description
11.3.3 Model Description
 11.3.4 Solution Procedure
11.3.5 Results and Discussion
11.3.6 Input Data Listing of Pressure Vessel
11.4 Transient Response of an Impulsively Loaded Cylindrical Panel
11.4.1 Problem Description
11.4.2 Finite Element Model
11.4.3 Analysis Results and Discussion
11.4.4 Input Data Listing for the 8x16 Model
11.5 Contact Analysis for the Shaft-Journal Interface in SOL 66 R. Louwers
11.5.1 Introduction
11.5.2 Problem Description
11.5.3 Selection of Gap Stiffness
11.5.4 Superelement and ASET
11.5.5 Analysis Procedure
11.5.6 Summary of Results
11.5.7 Friction in the GAP
11.6 Impact Analysis in SOL 99: Stress Wave Propagation in an Elastic Rod
11.6.1 Problem Description
11.6.2 Theoretical Solution
11.6.3 MSCINASTRAN Solution
11.6.4 Comparison of Results
11.6.5 Input Data Listing
11.7 Effects of KGROT on Geometric Stiffening of the Thin Shell Model
11.7.1 Theoretical Basis for Parameter K6ROT
11.7.2 Difficulties in the Thin Shell Model
11.7.3 Parametric Studies
11.7.4 QUAD4 Model of Rectangular Plate with Line Load
11.7.5 TRIA3 Model of Rectangular Plate with Line Load
11.7.6 QUAD4 Model of Square Plate with Uniform Pressure
11.7.7 QUAD4 Model of Square Plate with Concentrated Load
11.7.8 Observations
11.8 Adaptation of Creep Analysis Capability to General Viscoelastic Materials
11.8.1 Introduction
11.8.2 Formulation for Least Square Fit of Rheological Parameters
11.8.3 Computational Process
11.8.4 Analysis Procedure and Results
11.8.5 Remarks

Bibliography
Appendix A. Nonlinear Bulk Data Description

Index
 LIST OF TABLES

Summary of Properties of the Nonlinear Elements

Example Input Data for Nonlinear Static Analysis
Example Input Data for Nonlinear Dynamic Analysis
Summary of Nonlinear Case Control Data
Summary of Parameters in Nonlinear Analysis
PARAMeters used in Nonlinear Solution Sequences
Input Data Listing for Bouncing Mass Problem
Input Data Listing for Coulomb Damping Problem
SOL 66 Input Data Listing for Elliptic Cylinder
SOL 99 Input Data Listing for Elliptic Cylinder
SOL 64 Input Data Listing for Beam with Axial End Force
SOL 66 Input Data Listing for Beam with Axial End Force
SOL 64 Input Data Listing for Beam with End Moment
SOL 66 Input Data Listing for Beam with End Moment
Input Data Listing for Stress Path Dependent Case
Input Data Listing for Strain Path Dependent Case
Input Data Listing for Biaxial Loading Case
Input Data Listing for Elastic Clip
Input Data Listing for Cold Start
Input Data Listing for First Restart
Input Data Listing for Second Restart
Input Data Listing for Third Restart
Input Data Listing for Pressure Vessel Model
Input Data Listing for Elastic-Plastic Rod
Input Data Listing for Impulsively Loaded Plate
Input Data Listing for Beam with a Restrained Motion
Input Data Listing for Bouncing Weight Model
Input Data Listing for Shallow Spherical Cap
Input Data Listing for Simply Supported Beam
Verification Problem Input Data
MSCINASTRAN Output Summary
Example Problem Input Data
DMAP Alters for SOL 63 (Version 66)
SOL 66 Input Data Listing for Uniform Nonlinear Beam
SOL 63 Input Data Listing for Uniform Nonlinear Beam
SOL 66 Input Data Listing for Plate
SOL 63 Input Data Listing for Plate
SOL 66 Input Data Listing for Stepped Beam
SOL 63 Input Data Listing for Stepped Beam
SOL 66 Input Data Listing for Z-Shaped Beam
SOL 99 Input Data Listing for Z-Shaped Beam
SOL 66 Input Data Listing for Beam with Friction
SOL 99 Input Data Listing for Beam with Friction
Input Data for the Clamped Case
Partial Listing of the Input Data for the Simply Supported Case
Solution Strategy and Performance for Clamped Case
Effects of NLPARM Parameters on Convergence for Subcase 2lIncrement 4 of
Clamped Case
Solution Strategy and Performance for Simply Supported Case
Effects of NLPARM Parameters on Convergence for Subcase 2lIncrement 3 of
Simply Supported Case
Effects of NLPARM Parameters on Convergence for Subcase 4lIncrement 1 of
Simply Supported Case
Solution Strategy and Performance for Sliding and Simply Supported Case
Numerical Performance as a function of MAXQN in ZPLATE
Numerical Performance as a function of MAXQN in ZBEAM
Numerical Performance as a function of MAXQN in ZPLATE
Numerical Performance as a function of MAXLS in ZBEAM
Numerical Performance as a function of MAXQN in the spherical diaphragm
with clamped boundary
Numerical Performance as a function of MAXLS in the spherical diaphragm
with clamped boundary
Numerical Performace as a function of MAXQN in the spherical diaphragm
with simply supported boundary
Numerical Performance as a function of MAXQN in the spherical diaphragm
with sliding/simply supported boundary
Numerical Performance as a function of MAXLS in the spherical diaphragm
with sliding/simply supported boundary
Numerical Performance as a function of LSTOL in the spherical diaphragm
with sliding/simply supported boundary
Numerical Performance as a functioin of MAXQN in the imperfect
spherical cap
Numerical Performance as a function of MAXLS in the imperfect
spherical cap
Numerical Performance as a function of LSTOL in the imperfect
spherical cap
Normalized values of elapsed CPU time for analyses with and without
ASETs and/or SUPERELEMENTS
Variation of Iteration Strategy
MSCINASTRAN Input Data Listing for Case 16
Duration of Impact
Input Data Listing for QUAD4 Model with Line Load
Results of Parametric Study for K6ROT:
QUAD4 Model with Line Load
Input Data Listing for TRIA3 Model with Line Load
Results of Parametric Study for KGROT:
TRIA3 Model with Line Load
Input Data Listing for QUAD4 Model with Uniform Pressure
Results of Parametric Study for K6ROT:
QUAD4 Model with Uniform Pressure
Input Data Listing for QUAD4 Model with Concentrated Load
Results of Parametric Study for K6ROT:
QUAD4 Model with Concentrated Load
Run Progression of the Program "RHEOLOGY7'
StrainJStrain-Rate vs. Time (RHEOLOGY.OUT File Content)
C,/CpJli, vs. Stress (RHEOLOGY .PAM File Content)
Run Progression of the Program "DISPLY"
Source Listing for the Program "RHEOLOGY"
Source Listing for the Program "DISPLY"
 LIST OF FIGURES

Hierarchy of Nonlinear Looping

Newton's Methods for Iteration
Hybrid Newton's Method Based on Stiffness Matrix Update Strategy
Iteration Trap Condition
Flow Diagram for K Update Strategies (to set NEWK)
Divergence and Retreat Procedure
CPU Time Expiration Logic for NOGO Condition
Line Search Parameter vs. Line Search Error
Flow Diagram for Line Search Algorithm
Flow Diagram for Line Search Iteration
Examples for the Line Search Procedure
Quadratic Interpolation Algorithm
Flow Diagram for Iteration with Quasi Newton Updates
Comparison of Iteration Methods
(Quasi-Newton method with or without line search)
Riks' Method
Modified Riks' Method
Crisfield's Methods
Spherical Diaphragm Model with Axisymmetric Boundary Condition and
Constraints by Gap Elements
Load vs. Vertical Displacement at the Center (Clamped Case)
Imperfect Spherical Shell
MSCJNASTRAN Model for Imperfect Spherical Shell
Load-Deflection Curve of Imperfect Spherical Shell
Deformed Shapes at Different Buckling Stages
BEAM Element Coordinate System and Element Forces
Plate and shell Element Coordinate System
Forces and Stresses in Plate Elements
Volume Coordinate for Tetrahedron
Gaussian Quadrature for Tetrahedron
Shape Functions for Pressure Load
Gaussian Quadrature for Pressure Load
Element Coordinate System Definition
The Gap Element
Gap Axial Force vs. Deflection
Gap Lateral Force vs. Deflection
Schema for Subincremental Process (m = 3)
Physical Model
Symbolic MSCJNASTRAN Model
Displacement Response
Adaptive Penalty Value Adjustment on GAP 22
Physical Model for Vibration with Coulomb Damping
Horizontal Displacement Response
Comparison of Responses: Under-damped, Critically-damped, Over-damped
Horizontal Displacement during Static Loading
Contact Between Elastic Sphere and Rigid Plane
MSCINASTRAN Model for Static Analysis
Underformed Shape vs. Deformed Shape at P = 2160
Contact Radius & Approach vs. Total Force in Static Analysis
MSCINASTRAN Model for Dynamic Analysis
Reaction Force vs. Time
Vertical Displacement Response a t Select Points on Center Line
Snap Shots a t t = O., 0.051, 0.072 and 0.2 sec.
Net Deformation of a Rod
Element Coordinates vs. Displaced Coordinates
Computation of Net Deformation
Elliptic Cylinder subject t o an Internal Pressure
Displacement Responses for Elliptic Cylinder
Deformed Shapes of the Elliptic Cylinder
Beam with an Axial End Force
Beam with an End Moment
Load vs, Displacement at Free End of Beam with Axial End Force
Deflection Curves for Beam with Axial End Force
Deflection Curves for Beam with End Moment
Stress-Strain Curve Definition with TABLES1 Entry
Stress-Strain Curve Definition for a Single H
Uniaxial Rheological Model
Creep Model Parameter K p as a Function of Stress a
Creep Model Parameter Cp as a Function of Stress a
Creep Model Parameter C, as a Function of Stress a
Representation of Hardening Rules
Geometric Representation of von Mises Yield Surface
Mohr Circle Representation of the Mohr-Coulomb Yield Criterion
Von Mises Yield Locus for Plane Stress
Tresca's Yield Locus for Plane Stress
Isotropic Yield Surfaces in Principal Stress Space
Isotropic Hardening
Kinematic Hardening
Flow Diagram for Elasto-Plastic Material
Uniaxial Loading and Apparent Poisson's Ratio
Strain-Controlled Cyclic Loading
Monotonic and Cyclic Stress-Strain Curves for Gray Cast Iron
TABLES1 INPUT for Stress-Strain
Rheological Model

xiii
Flow Diagrams for Creep Process
Creep Capability Verification for Variable Temperature
Isochronous Stress-Strain Curve
Comparison of Creep Relaxation Predictions
Effective Stress Distribution after 3 hours of Creep
Finite Element Model by MSCINASTRAN
Total Circumferential Strain at Gauge Location during Pressure Loading
Total Circumferential Strain at Gauge Location (1) during Moment Loading
Total Circumferential Strain at Gauge Location (2) during Moment Loading
Effective Stress Contour Plot at t=2546 hours upon Loading
Effective Stress Contour Plot at t=3400 hours upon Loading
Cube Subjected to Uniaxial Tension
Flow Diagram for Nonlinear Process in SOL 66
Flow Diagram for Appending Process in SOL 66
Hierarchy of Load Dat'a
Validation of Nonproportional Loading Case
Finite Element Model for Biaxial Loading
Material Curve for Biaxial Loading
Biaxial Loading Path
Corresponding Strain Path for Biaxial Loading
Stress-Strain Response Curve
Elastic Clip Forced into a Cavity
Model of Rod with Spring
Force-Displacement Curve
Steel Pressure Vessel
MSCINASTRAN Model (10% Sector of the Steel Pressure Vessel)
Experimental Results vs. MSCINASTRAN
Flow Diagram for Nonlinear Process in SOL 99
Flow Diagram for Appending Process in SOL 99
Illustration of the Bisection Strategy
Flow Diagram for the Automatic Time Step Adjustment
Flow Diagram for the Bisection Algorithm
Flow Diagram for the NLTRD2 Module
Elastic-Plastic Rod Subjected t o a Step Loading
Elastic-Plastic Rod: Displacement Response and Time Step Variation
Impulsively Loaded Clamped Plate
Clamped Plate: Central Displacement Response and Time Step Variation
Simply Supported Beam with a Restrained Motion
Simply Supported Beam: Loading Point Displacement Time History
Bouncing Weight on an Elastic Platform
Bouncing Weight: Displacement Time History
Shallow Spherical Cap under a Concentrated Apex Load
Spherical Cap: Apex Displacement Response

xiv
Simply Supported Beam with a Restrained Motion
Simply Supported Beam: Loading Point Response
Dynamic Loads for Transient Analysis
Example of Combining Loads
Options in TLOADi (Types 1, 2 and 3)
Euler Buckling for Verification
Elastic-Plastic Buckling of a Clamped Spherical Cap with Flat Spot
(Load vs. Central Deflection)
Deformed Shape on Buckling
Deformed Shape Upon Buckling with a Solid Model
Simply Supported Beam
Nonlinear Stress-Strain Relationship for Simply Supported Beam
Clamped Plate (Quarter Only, Due to Symmetry)
Stepped Beam
2-Shaped Cantilever Beam
Nonlinear Stress-Strain Relationship for 2-Shaped Beam
Deformed Shapes of 2-Shaped Beam
Load-Deflection Curve for 2-Shaped Beam
Uniform Beam with a Gap
Deformed Shapes of Beam During Static Loading (SOL 66 and SOL 99)
Deformed Shapes of Beam During Static Unloading (SOL 66)
Rotation (8,) at End of Beam During Dynamic Unloading
Spherical Diaphragm Subjected t o a Uniform Pressure
Spherical Diaphragm Model with Axisymmetric Boundary Condition and
Constraints by Gap Elements
Load vs. Vertical Displacement a t the Center (Clamped Case)
Deformed Shapes of the Diaphragm (Clamped Case)
Load vs. Vertical Displacement at the Center (Simply Supported Case)
Deformed Shapes of the Diaphragm (Simply Supported Case)
Vertical and Radial Displacement at the Center and Boundary Points
vs. SPC Forces at the Periphery
Load vs. Vertical Displacement a t the Center
(Sliding and Simply Supported Case)
Cantilevered 2-Shaped Beam (QUAD4 Model)
Deformed Shapes (QUAD4 model)
Load-Deflection Curve
Spherical Diaphragm Subjected t o a Uniform Pressure
Spherical Diaphragm Model with Axisymmetric Boundary Condition and
Constraints by Gap Elements
Load vs. Vertical Displacement a t the Center (Clamped Case)
Finite Element Model of Pressure Vessel
Initial Stress Distribution
Effective Stress Distribution After 3 Hours of Creep
Stress Relaxation at the Junction of the Cylinder and End Closure
Effective Creep Strain ( x 1 0 - ~ )After 3 Hours of Creep
Impulsively Loaded Cylindrical Panel
Displacement Response at Midspan (z=6.28 in)
Comparison of Gauss Integration Schemes Using 8 x 16 MSCINASTRAN Model
Finite Element Model Exposing Gap Locations
Detail of Contact Region for Determining Local Stiffness Values
Ir', vs. Axial Gap Forces for 66-element and 132-element Models
Sections from Figure 11.5.1 Indicating which GAPS Close
Displacement vs. Time for M l m = 1
Comparison of Free End Stress (Time History for M l m = 1)
Comparison of Impact Stresses
QUAD4 Model of Rectangular Plate with Line Load
Load-Deflection at Center of QUAD4 Model with Line Load
TRIA3 Model of Rectangular Plate with Line Load
Load-Deflection a t Center of TRIA3 Model with Line Load
QUAD4 Model of Square Plate with Uniform Pressure
Load-Deflection at Grid 23 of QUAD4 Model with Uniform Pressure
QUAD4 Model of Square Plate with Concentrated Load
Load-Deflection at Grid 19 of QUAD4 Model with Concentrated Load
Creep Strain vs. Time at 1000 psi
Creep Strain-Rate vs. Time
C, vs. Stress
C, vs. Stress
I<p vs. Stress
Chapter 1

INTRODUCTION

1.1 EVOLUTION

The original version of MSCINASTRAN, released in 1970, had no nonlinear capabilities to

mention [1.1], other than some limited usage of nonlinear loads and piecewise-linear analysis
with a differential stiffness. Some ground-work for nonlinear capabilities had begun in 1973 and
necessary precursors had been implemented in the element routines by 1977. During the same
period, the database technology had been introduced in MSCINASTRAN, which facilitated
nonlinear capability development.

The primary feature of nonlinear analysis is the requirement of the iterative and incremental
processes t o obtain a solution. A prototype solution sequence, SOL 64, was developed for this
purpose and released in 1978 t o provide a tool for such an iterative and incremental procedure in
the geometric nonlinear analysis. This was the first time the convective element coordinates (in
ROD, QUAD4, TRIA3, HEXA, and PENTA) and the gimbal angle (in QUAD4 and TRIA3)
were provided for large displacements in MSCINASTRAN. However, the method was not
automatic and required the user's intervention for every iteration. The automated iteration
method with convergence criteria was developed and released in 1981 as Version 61 with a new
solution sequence, SOL 66 [1.2, 1.31. Limited nonlinear material capabilities became available
for static analysis in this version.

The current nonlinear capabilities were embodied in Version 62, released in 1982. In addi-
tion t o SOL 64 and SOL 66 for nonlinear static analysis, another solution sequence, SOL 99,
was added for nonlinear dynamic analysis in Version 62 [1.4]. Nonlinear material capabilities
(plasticity and nonlinear elasticity) and the automated modified-Newton's iteration method
were included in SOL 66 and SOL 99. The GAP element and the nonlinear BEAM element
were also introduced in this version.

The nonlinear capabilities, released in MSCINASTRAN Version 62, have been improved
and extended in the versions that followed. Major enhancements in Version 63 [1.5] include
creep analysis capability, quasi-Newton update method, composite elements and the follower
forces implemented in SOL 66. The data recovery was enhanced by implementing new modules
in SOL 66 and SOL 99 to provide the output for upstream superelements and the special
output format for nonlinear element stresses and strains. Among the element enhancements,
implementation of the plane-strain element using shell elements was most notable.

Major enhancements for the static analysis in Version 64 were improved PLOT module
and SOL 66 for consolidated data recovery by combining linear and nonlinear elements. The
nonlinear buckling analysis was supplemented in SOL 66 by RF66D84. Additional capabilities
for the nonlinear dynamic analysis in Version 64 consist of options for initial conditions (TIC),
nonlinear loads (NOLINi), and the quasi-Newton updatelline search method (LSQN). A new
element, TETRA, was also added t o the nonlinear element library. In addition, numerous
minor enhancements and error corrections were incorporated in Version 64 for computational
efficiency and user-friendliness.

Enhancements in Version 65 pertain t o the iteration algorithm in static analysis, essentially

to provide better convergence. Highlights include full utilization of the quasi-Newton updates,
discriminate use of line searches, pertinent divergence criteria, timely stiffness updates, and
unit-independent convergence criteria. SOL 66 DMAP was also improved extensively, resulting
in reduced database storage space with a parameter SMALLDB, flexible restarts, and built-in
nonlinear buckling analysis capability controlled by a parameter BUCKLE. Additional capa-
bilities include extended nonlinear-elastic material capability to allow a bilateral stress-strain
curve, and the composite beam element (PBCOMP) to allow user specification of the cross-
sectional shape.

Solution methods were further improved and expanded in Version 66. SOL 66 was reen-
forced with the Crisfield's arc-length method, primarily for the snap-through and post-buckling
analyses. Also included in SOL 66 for this version were follower centrifugal force (RFORCE)
and the temperature dependent material capabilities (thermo-elastic and creep under variable
temperature). SOL 99 was overhauled with an adaptive time stepping (automatic adjustment
and bisection) and two-point direct time integration in concert with improved iteration strate-
gies (quasi-Newton updates, line search and dynamic matrix update scheme). Options for
static analysis (using a parameter TSTATIC) and follower forces were also added to SOL 99
in this version.

Adaptive solution algorithms were extended t o SOL 66 and the GAP element in Version 67.
Two additional options (RIKS and MRIKS) for arc-length method were added and elaborated
with BFGS updates, adaptive arc-length adjustment and bisection process. Adaptive bisection
and recovery processes were also implemented for the Newton's iteration, including the creep
time bisection. The basic thrust of the adaptive GAP element was t o improve the convergence
and accuracy in the contact problems using GAP elements by introducing adaptive penalty
value adjustment, stiffness matrix updates, bisection, subincrements, and staticlkinetic fric-
tions. Large rotation capability has also been improved by introducing an auxiliary angle set
(to remove the gimbal lock at 90') and a rotation vector approach (an option selectable by a
parameter LANGLE 2). Minor enhancements in this version include introduction of numerical
damping (by a parameter NDAMP) in the adaptive time stepping, linear analysis capability
in SOL 99, initial anisotropy in nonlinear materials (plastic, creep, thermoelastic materials
combined with MAT2 or MATS), and modification of internal relaxation iteration in solid ele-
ments (HEXA and PENTA). Structured solution sequences (SOL 106 and SOL 129) were also
introduced in this version.
1.2 OVERVIEW OF NONLINEAR CAPABILITIES

Nonlinear effects in structures occur mainly due to nonlinear materials and large rotations.
Contact problems exhibit nonlinear effects due to changes in boundary conditions. All of these
attributes may be represented by nonlinear elements.

The nonlinear element library of MSCINASTRAN consists of:

r RODS, CONRODs, and TUBES for unidirectional truss members;

r BEAMS for axially and laterally deforming line members;

r QUAD4s and TRIA3s for membrane, plate and shell modeling;

HEXAs, PENTAs, and TETRAS for solid modeling;

r GAPS for contact and friction simulation.
Nonlinear elements may be combined with linear elements for computational efficiency if
the nonlinear effects can be localized. Primary operations for nonlinear elements are updating
element coordinates and applied loads for large displacements, and the internal relaxation
iteration for material nonlinearity. Refer to Table 1.2.1 for a summary of the nonlinear element
properties.

The geometric nonlinearity becomes discernible when the structure is subjected t o large
displacement and rotation. Geometric nonlinear effects are prominent in two different aspects:
geometric stiffening due t o initial displacements and stresses, and follower forces due to a
change in loads as a function of displacements. These effects are included in the current
release of MSCINASTRAN, but the large deformation effect resulting in large strains is yet
t o be implemented.

Material nonlinearity is an inherent property of any engineering material. Material non-

linear effects may be classified into many catagories. Included in the current release of
MSCINASTRAN are plasticity, nonlinear elasticity, creep, and viscoelasticity. Creep and
viscoelasticity, implemented as a generalized nonlinear viscoelastic capability, may be cou-
pled with plasticity. Many sophisticated options are available for yield criteria and hardening
behavior in plasticity.

The primary solution operations are gradual load or time increments, iterations with con-
vergence tests for acceptable equilibrium error, and stiffness matrix updates. The iterative
process is based on the modified-Newton's method combined with optional expeditious met h-
ods such as the quasi-Newton (BFGS) update and the line search. The stiffness matrix updates
are performed occasionally to improve the computational efficiency, and may be overridden at
the user's discretion. A number of options of arc-length methods are also available for snap-
through or post-buckling analysis of the static problems. For the transient response analysis,
a number of options are available for implicit direct time integration, combined with adaptive
and expedient iteration strategies similar t o those implemented for static analysis.
 Solution sequences 66 (or 106) and 99 (or 129) consolidate all the nonlinear features de-
scribed above. SOL 66 (or 106) is applicable t o static, quasi-static, and nonlinear buckling
analyses. SOL 99 (or 129) is primarily applicable t o dynamic transient response analysis with
some limited static analysis capability. Both of these solution sequences can accommodate
superelements and provide easy restarts from the database. SOL 64, the forerunner of these
solution sequences, was developed only t o analyze geometric nonlinear problems. No new ca-
pabilities have been added t o SOL 64 since Version 63. SOL 64 will be eventually removed,
because SOL 66 can be substituted for any capabilities in SOL 64. Structured solution se-
quences are created in Version 67, taking advantage of the new executive system. They are
SOL 106 and SOL 129, which are equivalent t o SOL 66 and SOL 99, respectively.

The future goal for MSCINASTRAN nonlinear capabilities is t o provide a program that
is more capable, efficient, and user-friendly. Enhancements and new capabilities will be imple-
mented t o fulfill this goal in the areas of material, geometric, and boundary nonlinearities in
addition t o elements, and solution algorithms.
 Table 1.2.1 Summary of Properties of the Nonlinear Elements

3. Edge nodes are not applicable to nonlinear elements.

4. * may be changed by the user.
1.3 PROGRAM ARCHITECTURE

MSCINASTRAN has a modular structure to separate functional capabilities which are orga-
nized under an efficient executive system. The program is divided into a series of independent
subprograms, called functional modules. A functional module is capable of performing a pre-
defined subset of operations. It is the Executive System that identifies every module t o execute
by MPL (Module Properties List).

The Executive System processes the input data by IFP (Input File Processor) and the
general initialization, which are known as Preface~operations.It then establishes and controls
the sequence of module executions in the OSCAR (Operation Sequence Control Array) based
on the user-specified DMAP (Direct Matrix Abstraction Program) or solution sequence. The
Executive System allocates system files t o the data blocks in the FIAT (File Allocation Table)
and maintains a parameter table for module interface. The Executive System is also responsible
for the database management and all the input and output operations by GIN0 (General
Input /Output Routines).

The functional module consists of a number of subroutines. Modules communicate with

each other only through secondary storage files, called data blocks (matrix or table). Each
module performs a certain function with input data blocks and produces output data blocks.
A module may communicate with the Executive System and with other modules through
parameters, which may be input and/or output variables of the module. Modules utilize main
memory dynamically. If the size of the main memory is insufficient to complete an operation,
the module uses scratch files, which reside in the secondary storage as an extension of the main
memory. This is known as a spill operation.

DMAP is a kind of macro program using a data block oriented language. The solution
sequence is a collection of module statements written in the DMAP language tailored t o process
a sequential series of operations, resulting in a specific type of structural analysis. A typical
solution sequence consists of three phases of functional operations: formation, assembly, and
reduction of matrices; solution of equations; and data recovery. Solution sequences that process
superelements have built-in superelement loops in the first and the last phases.

The nonlinear solution sequences have built-in loops in the second phase for subcase
changes, load increments, and stiffness matrix updates. Nested in this DMAP loop, nonlinear
solution processes comprise a number of internal iteration loops. Confining the discussion t o
SOL 66 and SOL 99, the hierarchy of the nonlinear looping is shown in Fig. 1.3.1. Central t o
the nonlinear processes are modules NLITER and NLTRD2 for SOL 66 (or 106) and SOL 99
(or 129), respectively. These modules are self-contained t o perform iterations for converged
solutions.
 - - - - DMAP Control Loops
Module Control Loops
-.-. Subroutine Control Loops

SUBCASES
- - -(BOUNDARIES,
- - - - -TEMPERATURES,
- - - - - LOADS,
- - -OUTPUTS)
---
r
I LOAD
-- INCREMENTS
- - - -(NLITER
- - -FOR
- STATIC)
---
I r
I I TIME STEPS (NLTRD2 FOR TRANSIENT)
I l l
STIFFNESS
- - - MATRIX
- - -UPDATES
---
r
I ITERATIONS (VECTOR ARITHMETIC)
I
LINE SEARCHES

ELEMENTS
.- . - (NLEMG)
.-
r
INTERNAL RELAXATION
(FOR
. - BEAM,
. - . - .HEXA,
- . - .AND
- PENTA)
I r
VOLUME
. - . - .INTEGRATION
- . - . - . - . (GAUSS POINTS)
I I r
STRAIN
. - . -SUBINCREMENTS
. - . - . - .(NMATD)
I I I r
I I 1 I STRESS COMPONENTS

Figure 1.3.1 Hierarchy of Nonlinear Looping

1.3 - 2
1.4 RUDIMENTS of USER INTERFACE

The input data structure, as was designed originally, still has the most popular format con-
sisting of an optional NASTRAN header, Executive Control Section, Case Control Section,
and Bulk Data Section. All the features and principles for the user interface are common in
MSCINASTRAN. Needless t o say, all the features of the user interface for nonlinear analysis
are compatible with those for the linear analysis. Review Reference [I.11 for details of the gen-
eral features. Any exceptions for nonlinear analysis will be explained in the relevant chapters
that follow.

Mechanical design is dictated by the strength, dynamic, and stability characteristics of

the structure. MSCINASTRAN provides the analysis capabilities of these characteristics with
solution sequences, each of which is designed for specific applications. The type of desired
analysis is specified in the Executive Control Section by using a solution sequence identification.
SOL 66 (or 129) is designed for static, quasi-static, and buckling analyses. SOL 99 (or 129)
provides noqlinear transient response analysis.

The basic input data required for a finite element analysis may be classified as follows:
a Geometric data

a Element data

a Material data

Boundary conditions and constraints

Loads and enforced motions

Solution methods
The first three classes of data may not be changed during the course of an analysis whereas the
last three classes of data may be changed in midcourse via subcases under the Case Control
Section. Examples of input data are illustrated in Tables 1.4.1 and 1.4.2 for nonlinear static
and dynamic analyses, respectively.
 Table 1.4.1 Example Input Data for Nonlinear Static Analysis
ID TETRAC,V65 $ CJS 18-JULY-85
SOL 66 $ NONLINEAR STATIC ANALYSIS
DIAG 8,50 $ DIAGNOSTIC PRINTOUT
TIME 8 $ FOR VAX
CEND $ END OF EXECUTIVE CONTROL DATA
TITLE = TEST OF CTETRA ELEMENT (CUBE SUBJECT TO UNIAXIAL LOADING)
SEALL = ALL $ FOR ALL MATRICIES
DISP = ALL
STRESS = ALL
SPC = 100
$ RESTART PARAMETERS
PARAM SUBID
PARAM LOADINC
PARAM LOOPID
SUBCASE 1
SUBTITLE = ELASTIC -- LOAD TO 850. PSI
LABEL = LOAD TO YIELD
LOAD = 50
NLPARM = 50
SUBCASE 2
SUBTITLE = PLASTIC -- LOAD TO 1000. PSI
LABEL = LOAD BEYOND YIELD
LOAD = 100
NLPARM = 100
SUBCASE 4
SUBTITLE = ELASTIC -- UNLOAD COMPLETELY TO 0. PSI
LABEL = FULL UNLOAD
LOAD = 200
NLPARM = 200
$OUTPUT(PLOT) $ PLOT DATA ADDED FOLLOWING THIS LINE IF NEEDED
$ END OF CASE CONTROL DATA
BEGIN BULK
$ PARAMETERS FOR NONLINEAR ITERATION
NLPARM 50 1 AUTO UPW N0
NLPARM 100 8 SEMI UPW N0
NLPARM 200 2 AUTO UPW N0

ENDDATA
 Table 1.4.2 Example Input Data for Nonlinear Dynamic Analysis
ID VGAP, V65 $ SHL
SOL 99 $ NONLINEAR TRANSIENT ANALYSIS
DIAG 8,50 $ DIAGNOSTIC PRINTOUT
TIME 5 $ FOR VAX
CEND $ END OF EXECUTIVE CONTROL DATA
TITLE = GAP VERIFICATION BY NONLINEAR TRANSIENT ANALYSIS
SEALL = ALL $ FOR ALL MATRICIES
SET 1 = 201 $ ,202,200
SET 3 = 22 $ ,23
SET 4 = 20 $ CONROD
DISP = 1
STRESS = 3 $ FOR GAP FORCES
ELFORCE = 4
LOADSET = 20
'SPC = 100
$ RESTART PARAMETERS
$PARAM LOOPID
$PARAM STIME
SUBCASE 1 $ UP TO 0.5 SECONDS
DLOAD = 210
TSTEPNL = 22
SUBCASE 2 $ UP TO 1 SECOND
DLOAD = 210
TSTEPNL = 21
OUTPUT(XYPLOT)
CSCALE = 1.3
XTITLE = TIME IN SECONDS
YTITLE = DISPLACEMENT IN INCHES
YMIN = 0.
YMAX = 15.
XYPLOT DISP RESP/201 (TI) ,202(TI) ,200(TI)
$ END OF CASE CONTROL DATA
BEGIN BULK
TSTEPNL 21 200 .0025 2 AUTO

ENDDATA
1.4.1 Case Control Section

The Case Control Section may be regarded as a central control panel where select switches
for input and output are set. The primary purpose of the Case Control ~ i t isa to define
subcases. The subcase structure in MSCINASTRAN provides a unique means of changing
loads, boundary conditions, and solution methods by making selections from the Bulk Data.
Confining the discussion to SOL 66 (or 106) and SOL 99 (or 129), loads and solution methods
may change from subcase to subcase on an incremental basis. However, constraints can be
changed from subcase t o subcase only in the static solution sequence. As a result, the subcase
structure determines a sequence of loading and constraint paths in a nonlinear analysis. The
subcase structure also allows the user to select and change output requests for printout, plot,
etc., by specifying set numbers with keywords. Any selections made above the subcase specifi-
cations are applicable t o all the subcases. Selections made in an individual subcase supersede
the selections made above the subcases. Table 1.4.3 summarizes the Case Control Data for
nonlinear analysis.
 Table 1.4.3 Summary of Nonlinear Case Control Data
Load Selection
LOAD Selects static loading condition
CLOAD Selects static load combination f ~ r ' s u ~ e r e l e m e n t s
DLOAD Selects dynamic loading conditions
LOADSET Selects static load sets defined on the Bulk data LSEQ
NONLINEAR Selects nonlinear loading (NOLINi) for transient response
IC Selects initial conditions for transient response

Solution Met hod

Selection
METHOD Selects methods for eigenvalue analysis
NLPARM Selects iteration methods for nonlinear static analysis
TSTEPNL Selects iteration methods for nonlinear transient analysis

Output Requests
DISPLACEMENT Requests output for displacements of physical points
VELOCITY Rquests output for velocities of physical points
ACCELERATION Requests output for acceleration of physical points
ELFORCE Requests output for element forces
STRESS Requests output for element stresses
SPCFORCES Requests output for constraint forces of SPC points
NLLOAD Requests output for NOLINi in transient response
OUTPUT (PLOT) Requests the beginning of the plotter output

Superelement Control
SUPER Specifies the superelement identification number and the
load sequence number
SEALL Combines the functions of SEMG, SELG, SEKR, SEMR,
and SELR
SEKR Specifies the superelement identification numbers for which
stiffness matrices will be assembled and reduced
SELG Specifies the superelement identification numbers for which
load vectors will be generated
SELR Specifies the superelement identification numbers for which
the static load matrices will be assembled and reduced
SEMG Specifies the superelement identification numbers for which
stiffness, mass, and damping matrices will be generated
SEMR Specifies the superelement identification numbers for which
the mass and d a m ~ i n ematrices
: will be assembled and reduced
1.4.2 Bulk Data Section

The primary input media is the Bulk Data, where all details of the input data are specified.
Some of the Bulk Data may not be used during a program execution if they are not selected in
the Case Control Data. The nonlinear Bulk Data are summarized in Table 1.4.4. All the input
data specially designed for nonlinear analysis will be described in the relevant chapters that
follow. Descriptions of the Bulk Data pertinent t o nonlinear analysis are listed in Appendix A.

Table 1.4.4 Summary of Nonlinear Bulk Data

Element Connectivity
CBEAM Defines connection for beam element
CGAP Defines connection for gap or frictonal element
CHEXA Defines connection for six-sided solid element
CONROD Defines connection and properties for rod
CPENTA Defines connection for five-sided solid element
CQUAD4 Defines connection for quadrilateral element with bending
and membrane stiffness
CROD Defines connection for rod with axial and torsional stiffness
CTETRA Defines connection for four-sided solid element
CTRIA3 Defines connection for triangular element with bending and
membrane stiffness
CTUBE Defines connection for a tube

Element Properties
PBCOMP Defines properties for composite CBEAM
PBEAM Defines properties for CBEAM
PCOMP Defines properties for composite material laminate
PGAP Defines properties for CGAP
PROD Defines properties for CROD
PSHELL Defines properties for CTRIA3 and CQUAD4
PSOLID Defines properties for CHEXA, CPENTA and CTETRA
PTUBE Defines properties for CTUBE
Material P r o p e r t i e s
2REEP Defines creep material properties
MAT2 Defines anisotropic material properties for shell elements
MAT8 Defines orthotropic material properties for shell elements
MAT9 Defines anisotropic material properties for solid elements
MATS1 Defines properties for plastic and nonlinear elastic material
rABLESl Defines a function for stress-dependent material properties
TABLEST Combines many TABLES1 entries for
temperature dependent material properties

Constraints
SPC Defines single-point constraints and enforced displacements
SPCl Defines single-point constraints
MPC Defines a linear relationship for two or more degrees of freedom

Loads
CLOAD Defines a static load combination for superelement loads
FORCEi Defines concentrated load at grid point
LSEQ Defines static load sets for dynamic analysis
MOMENTi Defines moment at a grid point
NOLINi Defines nonlinear transient load
PLOAD Defines pressure loads on QUAD4 and TRIA3
PLOAD2 Defines pressure loads on shell elements, QUAD4 and TRIA3
PLOAD4 Defines pressure loads on surfaces of HEXA, PENTA, TETRA
TRIA3, and QUAD4 elements
RFORCE Defines load due to centrifugal force field
TEMP Defines temperature at grid points
TEMPPi Defines temperature field for surface elements
TEMPRB Defines temperature field for line elements
TIC Specifies initial values for displacement and velocity
TLOADi Defines loads as a function of time

Solution M e t h o d s
NLPARM Defines iteration methods for nonlinear static analysis
NLPCI Defines arc-length methods for nonlinear static analysis
TSTEPNL Specifies integration and iteration methods
for nonlinear transient analysis
EIGB Defines eigenvalue extraction method for buckling analysis
1.4.3 Parameters

Parameters constitute an important part of input data in their own right because they maintain
communications among the user, DMAP, and modules. Parameters are single-valued input
data, used for requesting special features or specifying miscellaneous data. Parameters are
initialized in the MPL, which can be overridden by a DMAP initialization. Modules may
change the parameter values while the program is running.

There are two different types of parameters in MSCINASTRAN: user parameters (V,Y,name
in the DMAP) and DMAP (non-user) parameters. For the user parameters, users are autho-
rized t o change the initial values by specifying PARAM data in the Bulk Data Section or
ocassionally in the Case Control Section. Many options available in MSCINASTRAN are con-
trolled by user-specified parameters. Table 1.4.5 lists the parameters used in nonlinear solution
sequences 64, 66, 99, 106 and 129. Key parameters which allow options for nonlinear analyses
are listed below:

AUTOSPCR Default = NO in SOL 66 or 106

Provides for automatic single-point constraints for the residual structure.

BETA Default = 0.333333 in SOL 99 or 129

The value of BETA is the transient integration control factor for the
Newmark-Beta method with AUTO and TSTEP methods.

BUCKLE Default = -1
If a value of 1 is specified in SOL 66 or 106, nonlinear buckling analysis is
activated. This capability should be used in the restart runs.

CNSTRT Default = 1 for SOL 64 only

This parameter is used t o control restarts but not recommended.

DBDRNL Default = 0 for SOLS 66, 99, 106, and 129

By default in SOLs 66, 99, 106, and 129, the database directory is printed
upon completion of the nonlinear analysis phase. PARAM, DBDRNL, -1
will suppress this printout.

K6ROT Default = 100.0 for SOLS 66, 99, 106, and 129
If K6ROT > 0.0, then stiffness is added t o the normal rotation for QUAD4
and TRIA3 elements. This is an alternative method t o suppress the grid
point singularities, and is intended primarily for geometric nonlinear analy-
sis. The recommended value is in the range of 10.0 t o l.OE+4.
LANGLE Default = 1
In SOL 66 (or 106) and SOL 99 (or 129), large rotations (using
PARAM,LGDISP) are treated by the gimbal angle approach through a de-
fault option. A rotation vector approach is implemented in Version 67, and
can be selected by PARAM,LANGLE with a value of 2.

LGDISP Default = -1
If LGDISP=l, all the nonlinear elements allowed in SOLs 66 and 99 will
be assumed to have large displacement effects (updated element coordinates
and follower forces). If LGDISP=-1, no large displacement effects will be
considered. If LGDISP=2, follower force effects will be ignored while large
displacement is accounted for.

LOADINC Default = 1
Defines the beginning load step of the subcase in which the restart analysis
is to be initiated in SOL 66.

LOOPID Default = 0
Defines the loop number for initial conditions in SOL 66 or SOL 99 for
restarts.

MAXLP Default = 5
Maximum number of iterations allowed internally for element relaxation and
material point subincrement processes in SOL 99 (or 129). This parameter
is set t o 10 in SOL 66 (or 106), which does not allow user's specification of
any other value.

NDAMP Default = 0.025 for SOLs 99 and 129 only

Numerical damping introduced in the two-point integration (ADAPT
method in SOL 99 or 129) for numerical stability (implemented in Version
67). The numerical damping increases as the value of NDAMP increases,
with zero being no numerical damping. Recommended range is from 0.0 to
0.1. For most cases the value of 0.01 is adequate.

NLAYERS Default = 5 (minimum = 1, maximum = 12)

This parameter defines the number of layers used to integrate through the
thickness of isotropic material nonlinear QUAD4 and TRIA3 elements. Set
NLAYERS=l for efficiency if no bending is selected (MID2 = 0 or -1 on
all PSHELL data entries). A larger value of NLAYERS will give greater
accuracy at the cost of computing time and storage requirements.
NMLOOP Default = 0
In Version 67, nonlinear modal analysis capability is built in SOL 106 with
PARAM, NMLOOP. In SOL 106, normal modes can be computed with the
updated stiffness if PARAM, NMLOOP, n is specified where n is the desired
LOOPID. This option also requires the presence of the METHOD command
in the Case Control section and EIGR or EIGRL entry in the Bulk Data
section.

OUTPUT Default = 0
In SOL 64 this parameter is used in geometric nonlinear analysis to control
restarts in a loop. If it is desired t o skip operations in the loop prior to stress
recovery, set OUTPUT=-1 (and usually set NODATA=-1).

SDATA Default = 1
In SOL 99 (or 129), SDATA=-1 with the proper value (last value) for
LOOPID allows data recovery restarts, skipping nonlinear computation loop.

SKPLOAD Default = 1
This parameter is used t o control the calculation of loads, as an efficiency
feature. If the loads of the present subcase are the same as in the previous
case, set SKPLOAD= - 1. The default causes recalculation of the loads, even
if they do not change (SOL 64 only).

SKPMTRX Default = 1
This parameter is used t o control the matrix reduction and decomposition,
an efficiency feature in geometric nonlinear analysis (SOL 64). If the matrix
of the present subcase is the same as in the previous case (sometimes called
right side iterations), set SKPMTRX=-1. Use of this parameter results in
a lower solution cost per iteration, but will usually require more iterations
for convergence.

SLOOPID Default = 0
In SOL 99 (or 129) transient analysis, this parameter identifies the initial
conditions from a previous SOL 66 (or 106) nonlinear static solution. The
printout LOOPID=x is given for each subase and load increment in SOL 66.
Setting SLOOPID=x and providing the static data base will cause SOL 99
t o start from the static deformed position.
SMALLDB Default = -1 (No longer available from Version 66)
This parameter is provided in SOL 66 t o allow a small database option in
connection with the "INTOUT" field of the NLPARM entry. With a value
of 1, the database will retain only those data blocks corresponding to the
load steps for which an output request is made by the "INTOUT" field in
the NLPARM entry. Restarts will be restricted t o those starting from these
steps.

Default = 0
This parameter is used t o control restarts in a loop for geometric nonlinear
analysis (SOL 64). If it is desired t o skip operations in the loop prior t o
FBS (for example, restart after time limit), set SSG3 = -1 (and usually set
NODATA = -1).

STIME Default = 0.0

In SOL 99 (or 129) restarts from previous SOL 99 runs, the user provides
STIME=tN, where t~ is the last time step of the subcase to be continued
with a new or changed subcase in the new run. Thus the loading and printout
will start from t~ as though the original run contained the new subcase data.

SUBID Default = 1
Defines the record number of the case control data (for SEID=O) in which
the beginning load step is defined for the restart in SOL 66 (or 106). In
SOL 64, SUBID is the subcase identification number (from a previous SOL
64 run) used for an initial guess.

SUBSKP Default = 0
Controls the skipping of subcases which are not desired in geometric non-
linear analysis. SUBSKP can be used on restart t o skip completed cases or
can be used t o delete the static (first) or differential (second) subcase. The
number of subcases t o be skipped is input, not the subcase ID number. The
default discards all previous results and is equivalent t o a cold start.

TABS Default = 0.0

This is used t o convert units of the temperature input (OF or OC) t o the
absolute temperature, i.e.,
PARAM, TABS, 273.16 when Celsius is used
PARAM, TABS, 459.69 when Fahrenheit is used
Refer t o the Bulk Data entry CREEP for a creep analysis with SOL 66 (or
106). Refer t o PARAM,SIGMA for heat transfer analysis.
TESTNEG Default = -1 for SOL 64, -2 for Newton's method in SOL 66, +1 for arc-
length method in SOL 66 and SOL 106
Controls SOL 64 and SOL 66 solution sequence when negative terms are en-
countered on the factor diagonal of matrix decomposition. Negative terms
indicate that the differential stiffness has introduced a structural instabil-
ity. The instability may be real (structural buckling) or mathematical (the
current iteration appears unstable, but a stable solution exists).
Value Result
-1 Stop if negative terms occur.
+lor 0 Continue if negative terms occur.
-2 If negative terms exist, do not use
differential stiffness.
+2 Do not use differential stiffness.

TESTSE Default = 1.0 x

Used t o control iterations within subcases for SOL 64. For subcases 1 and 2,
the output is skipped if the strain energy is greater than TESTSE. For sub-
cases 3 onward, the solution iterates within the subcase until the incremental
strain energy is less than TESTSE.

TSTATIC Default = -1 (SOL 99 or 129 only)

If TSTATIC=l, a static solution may be obtained in SOL 99 by ignoring
inertial and damping forces. This option is available only with the adaptive
method, specified by ADAPT in the TSTEPNL.

W3,W4 Default = 0.0

The damping matrix for transient analysis is assembled from the equation

The values of W3 and W4 are used for w3 and w4, respectively. The default
values of 0.0 for W3 and W4 cause the associated terms in the damping
matrix t o be set equal t o zero, regardless of the presence of the PARAMeter
G or [ K j d ]terms. The units for W 3 and W4 are radians per unit time.
 Table 1.4.5 PARAMeters used in Nonlinear Solution Sequences

NOTES:
B - Usable in t h e Bulk Data Section only
E - Usable in either t h e Bulk D a t a or Case Control D a t a Section

I PARAMETER 1 SOLUTION I 1
I NAME ( SEQUENCE I DEFAULT / DESCRIPTION

ASING B E E 0 Removes singularities by appropriate techniques

AUTOSPC B E E NO Specifies automatic single-point-constraints
I AUTOSPCR I I E I B I NO I Specifies automatic SPC7sfor residual structure 1
BAILOUT E E 0 Terminates superelement program with singularities
BETA B 0.33333 Specifies control factor for Newmark-Beta method
BUCKLE E -1 I Selects nonlinear buckling analysis for restarts
CB1,2 E 1.O Selects superelement damping matrix coefficients
CK1,2,3 E E 1.O Selects superelement stiffness matrix coefficients
CM1,2 E E 1.0 Selects superelement mass matrix coefficients
CNSTRT B 1 Controls restarts for constraint change
COUPMASS B E E -1 Generates coupled rather than lumped mass matrices
CURV E -1 Computes stress data in a material coordinate system
CURVPLOT E -1 Requests x-y (curve) plots
I DBCLEAN I B IBI - -
B I 0 I Empties database I
DBDICT B B B -1 Prints database dictionary at the end and beginning
DBDRNL E E E 0 Prints database dictionary in the nonlinear loop
DBINIT B B B 0 Initializes database with empty data blocks
DBNAME B B B DBOl Specifies database name
DBNBLKS B B B 4000 Sets maximum number of GIN0 blocks
DBSETi B B B 0 Specifies database subsets for storing and fetching
DBSORT B B B 5 Sorts printout of database dictionary
DLOAD B B B 0 Indicates no other changes except for loads
DOPT E 0 Controls x spacing of curves for CURVPLOT
 PARAMETER SOLUTION
NAME SEQUENCE DEFAULT DESCRIPTION
64 66 99
EPPRT E E E 1.OE-8 Prints singularities of stiffness matrix
EPSBIG E 1.OE+8 Turning parameter for generalized dynamic reduction
EPSMALC E 1.OE-8 Turning parameter for generalized dynamic reduction
EPSMALU E 1.OE-10 Turning parameter for generalized dynamic reduction
EPZERO B E E 1.OE-8 Determines printed singularities compared to EPPRT
ERROR E E -1 Terminates run for superelement errors
EST B E E 2 Requests element measure and volume computation
FACTOR B B B 10,000 Generates SEQUID on SEQGP entry
GPECT B E E -1 Prints all elements connected to each grid point
I GRDPNT I B I E I E I -1 1 Executes Grid Point Weight Generator 1
G E 0.0 Specifies uniform structural damping coefficient
HFREQ E 1.OEf30 Defines upper limit of frequency range
INRLM E 0 Requests appending of inertia relief modes
KDIAG E 1.0 Applies value of a spring constant to all DOF7s
KGROT E E E 100.0 Assigns stiffness to normal rotation of QUAD4

--
and TRIA3
LANGLE

LFREQ
II II E E l \ E l
1

0.0
Selects gimbal angle approach for large rotations if 1
and rotation vector approach if 2
Defines lower limit of frequency range
LGDISP
I
IE/
I
E I
1
-1 Selects large displacement effects
I
LMODES ( E 0 Defines number of lowest modes I
LOADINC E 1 Specifies load increment ID for restarts I
LOOPID E E 0 Specifies loop ID in the database for restarts
MAXLP E 5 Maximum number of iterations for internal loop
MAXRATIO I E I 1.OE+5 1 - ~ e t e r m i n ethe
s singularity of the stiffness matrix
MODEL IB IB I B ( 0 1 Stores several models in the database I
MPCX B B B 0 Controls MPC and rigid element processing on restarts
NDAMP E 0.025 Specifies numerical damping in ADAPT method
NEWDYN E 1 I Executes MTRXIN, GKAD, and GKAM modules
 91 - P'T

uo!qDnpal
3!mvubp pmua% JOOJx y s m .10$3a~ua%!aslu!~d ~g+30'1 3 ZAIHdXd
; ~ 8 , 3 ; ~m0.q
7 3 lnopur~dpuv uo!pn3axa sassaxddns T- 3 3 8 JLnO8d
qnoquy~ddlpsln8u!s sassa~ddns S~;C 3 3 8 LSd3Xd
assqsqop u! sy301q s l v p %u!ssa30~d-qsodsalops T 3 LSOd
saBvssam $old pam~ojapunsassa~ddns T 3 3 3SINXtd
I
I

s?uawala~adnsl o j s ~ o pamlojapun
~d saysm 0 1 3 131 dnSLOld
I I I I I

$old pam~ojapunop sdurnf I T 1 3 13181 LO?d

uo!pnpa.~ pas-? s1o~puo3 0 3 a3XLON

su1syuoy3am sno!~nds s ~ o q u o 3 T- 3 H33INON
spa y ~ n guy sa%usy:, ou sals3rpu1 O 8 8 8 VLVaON

-- -

sluamaia 11s l a ~ alvlodlalu!

o 03 ~ x n slsanbax
3 01 3 S;Ld;CNIN
a~npom~ o s s a 3 o ~Bupuanbas
d uy pasn uorldo spalas I 8 8 8 b 3 ~ ~ 3 ~
PARAMETER
NAME

SEMAPOPT
) SOLUTION
SEQUENCE
64 66
B
I
99
B
1 DEFAULT

42
DESCRIPTION

Places exterior superelement grids on CSUPER

SEMAPPRT B B 3 Prints SEMAP table
I SENAME I I E I E I SENAME I Identifies superelement by name I
SEQOUT B B B 0 Controls output options for NEWSEQ
SKPLOAD E 1 Skips load vector calculation
SKPMTRX E 1 Skips matrix reduction and decomposition
SLOOPID B 0 Specifies loop ID from SOL66 database for restart
I SMALLDB ( I E I I -1 I Selects small database option I
SOLID B B B 0 Stores several solutions in same database
SPCGEN B E E 0 Places automatic SPC's in S P C l format
SSG3 B 0 Skips operations prior t o FBS
START B B B 0 Defines number of grid points for NEWSEQ
STIME E 0.0 Specifies time step for restarts
I SUBID I B I E I I 1 I Specifies subcase ID for restarts I
SUBSKP B 0 Skips a number of subcases for restarts
SUPER B B B 0 Deletes ungrouped grid points
TABID E 2 Controls punch for response spectra
TABS E 0.0 Converts units of temperature input t o absolute
TESTNEG E E -1-21 Tests for negative terms on factor diagonal of matrix
TESTSE E 1.OE+36 Tests for strain energy for iteration control
TSTATIC E 0 Selects static solution with TSTATIC=l in SOL 99
using ADAPT method in TSTEPNL
USETPRT B E E -1 Selects output type for degrees of freedom
USETSEL B E E 0 Controls sets for row list output option
WTMASS B E E 1.O Multiplies terms of structural mass matrix by value
W3,W4 E 0.0 Selects frequency for conversion of damping
1.4.4 Nonlinear Element Output Codes for Plotting

Nonlinear elements (which reference nonlinear geometry and/or material) have different codes
from linear elements. The codes for linear elements can be found in the User's Manual, Vol. 11.
The following codes may be used in SOL 66 (or 106) and SOL 99 (or 129).
(All items are stresses unless otherwise denoted)
Element Name I Word or /
(Code) Component Item
CBEAM 2 External grid point ID
(94) 3 C (alpha-numeric value)

1 I Long. stress at point C

Equivalent stress
Total strain
Effective plastic strain
Effective creep strain

(Item codes 3 thru 8 are repeated

for points D, E and I?. Then the entire
record (from 2 through N) is repeated
for end B of the element.)

CGAP COMP-X
(86) SHEAR-Y
SHEAR-Z
AXIAL-U
SHEAR-V
SHEAR-W
SLIP-v
SLIP-W

CHEXA Grid/ Gauss

(93) External grid ID (0 = center)
STRESS-X
STRESS-Y
STRESS-Z
STRESS-XY
STRESS-YZ
STRESS-ZX
Element Name Word or
(Code) Component Item
CHEXA Equivalent stress
(93) Effective plastic strain
(cont .) Effective creep strain
STRAIN-X
STRAIN-Y
STRAIN-Z
STRAIN-XY
STRAIN-YZ
STRAIN-ZX
Items 3 through 17 repeated for 8 corners

CONROD Same as CROD

(92)

CPENTA Same as CHEXA but with only 6 corners

(91)

CQUAD4 Z1 = Fibre distance 1 (Plane stress only)

(90) STRESS-X (at 21, if plane stress)
STRESS-Y (at 21, if plane stress)
STRESS-Z (Plane strain only)
STRESS-XY (at 21, if plane stress)
Equivalent stress (at 21, if plane stress)
Plastic strain (at 21, if plane stress)
Creep strain (at 21, if plane stress)
STRAIN-X (at 21, if plane stress)
STRAIN-Y (at Z1, if plane stress)
STRAIN-Z (Plane strain only)
STRAIN-XY (at Z l , if plane stress)
Items 2 through 13 repeated for fibre
distance 22 (Plane stress only)
 -
Element Name Word or
(Code) Component Item
CROD 2 Axial stress
(89) 3 Equivalent stress
4 Total strain
5 Effective plastic strain
6 Effective creep strain
7 Linear torsional stress

CTETRA Same as CHEXA except with only 4 corners

(85)

CTRIA3 Same as CQUAD4

(88)

CTUBE Same as CROD

(87)
1.5 NONLINEAR CHARACTERISTICS AND
GENERAL RECOMMENDATIONS

Modeling for nonlinear analysis is not exempted from the guidelines for good modeling practice
pertaining t o linear analysis [1.1],which are summarized as follows:
r The analyst should have some insight into the behavior of the structure to be modeled;
otherwise, a simple model should be the starting point.

r Substructuring should be considered for the modularity of the model and/or synergism
between projects and agencies involved.

r The size of the model should be determined based on the purpose of the analysis, the
trade-offs between accuracy and efficiency, and the scheduled deadline.

r Prior contemplation of the geometric modeling will increase efficiency in the long run.
Factors t o be considered include selection of coordinate systems, symmetric considera-
tions for simplification, and systematic numbering of nodal points and elements for easy
classification of locality.

r Discretization should be based on the anticipated stress gradient, i.e., a finer mesh in the
area of stress concentrations.

r Element types and the mesh size should be judiciously chosen. For example, avoid highly
distorted and/or stretched elements (with high aspect ratio); use TRIA3 and TETRA
only for geometric or topological reasons.

r The model should be verified prior t o the analysis by some visual means, such as plots
and graphic displays.
Nonlinear analysis requires better insight into structural behavior. First of all, the type of
nonlinearities involved must be determined. If there is a change in constraints due to contact
during loading, the problem may be classified as a boundary nonlinear problem and would
require GAP elements or some surface contact algorithm. The material nonlinearity is char-
acterized by material properties. However, the material nonlinear effects may or may not be
significant depending on the magnitude and duration of the loading, and occasionally on envi-
ronmental conditions. The anticipated stress level would be a key t o this issue. The geometric
nonlinearity is characterized by large rotations which usually cause large displacements. Intu-
itively, geometric nonlinear effects should be significant if the deformed shape of the structure
appears distinctive from the original geometry without amplifying the displacements. There
is no distinct limit for large displacements because geometric nonlinear effects are related t o
the dimensions of the structure and the boundary conditions. The key t o this issue is t o know
where the loading point is in the load-deflection curve of the critical area.

As for MSC/NASTRAN, the model is accepted for nonlinear analysis by SOL 66 or SOL
99, if it consists of:
 r a parameter LGDISP (for geometric nonlinearity);

r any GAP element (CGAP and PGAP);

r any active nonlinear material data (specified on MATSl and/or CREEP);

r or any combination thereof.

The model may consist of superelements, but only the residual structure (superelement 0)
may consist of nonlinear elements, mixed with any type of linear elements. All the GAP
elements are always actively nonlinear if included in the residual structure. However, other
potentially nonlinear elements in the residual structure become actively nonlinear only if a
parameter, LGDISP, is used and/or if they use the nonlinear material data specified on the
MATSl and/or CREEP data entries.

The nonproportional loading introduces additional nonlinearity to the structural mechan-

ics problem. This nonlinearity is manifested in the path-dependent problems such as plastic
deformation, creep deformation, and contact problems with frictional forces. The cyclic load-
ing, which belongs t o this category of nonlinearity, makes the problem more complicated with
another nonlinearity known as the Baushinger effect. Such a nonlinearity is characterized by
a loading path, which can be controlled by multiple subcases in SOLS 66 and 99. When the
analysis involves a path-independent nonlinear problem such as a purely geometric nonlinear
problem, an incremental process may not be required. For computational efficiency, however,
the analysis of such a problem does frequently need incremental solutions.

With these points in mind, additional recommendations are imperative for nonlinear anal-
ysis:
r Identify the type of nonlinearity and localize the nonlinear region for computational
efficiency. If unsure, perform a linear analysis by SOL 61 (or 24) or SOL 69 (or 27) prior
t o the nonlinear analysis by SOL 66 or SOL 99, respectively.

r Segregate the linear region by using superelements and/or linear elements if possible.
Notice that the potentially nonlinear elements can be used as linear elements.

r The nonlinear region usually requires a finer mesh. Use a finer mesh if severe element
distortions or stress concentrations are anticipated.

r Be prepared for restarts with the database properly stored in cases of divergence and
changing constraints or loading paths via subcases.

r The subcase structure should be utilized properly t o divide the load or time history for
conveniences in restarts, data recovery, and database storage control, not t o mention
changing constraints and loading paths.

r The load or time for the subcase should then be further divided into increments, not to
exceed 20 load steps (for SOL 66) or 200 time steps (for SOL 99) in each subcase.
r Many options are available in solution methods to be specified on NLPARM (for SOL
66) or TSTEPNL (for SOL 99) data entries. The defaults should be used on all options
before gaining experience.

r Caution should be exercised in specifying GAP properties. In particular, the closed gap
stiffness should not exceed the stiffness of the adjacent degree of freedom by 1000 times.

Normal rotation for QUAD4 and TRIA3 should be restrained by a parameter KGROT
when the geometric nonlinearity is involved.

r Understand the basic theory of plasticity or creep before using these capabilities.

Caution should be exercised in preparing input data for creep, because they are unit
dependent.

r For the transient analysis by SOL 99, some damping is desirable and the massless degree
of freedom should be avoided.

r The time step size for a transient response analysis should be carefully determined based
on the highest natural frequency of interest because it has significant effects on the
efficiency as well as the accuracy.
For any anomalies, refer to errors and limitations listed in the User's Manual Volume 11.
Chapter 2

BASIC CONSIDERATIONS FOR

NONLINEAR ANALYSIS

2.1 DISCRETE SYSTEM FOR A

NONLINEAR CONTINUUM MODEL

Theories in solid mechanics are dictated by three governing relationships [2.1]:

a The state equilibrium that requires

da;j
-+b;=O
axj
where aij are stress tensor components, b; are body forces, and xj are space coordinates;

a The constitutive relations represented by stress-strain relations, e.g., for linear elasticity,

where ~ k are
l strain tensor components and Dijkl are elastic constants;

a The compatibility represented by strain-displacement relations, e.g., for a small defor-

mation,

where u; are displacements.

These systems of governing differential equations must be satisfied for every infinitesimal
element throughout the domain of the continuum. The complete set of state variables, namely
displacements, may be determined by solving these systems of equations supplemented by
boundary conditions, and in dynamic situations by initial conditions as well. For the non-
linear problems, the governing equations should be satisfied throughout the history of load
application. The material nonlinearity is manifested in the constitutive relations. The geo-
metric nonlinearity is pronounced in the strain-displacement relations, but it also affects the
equilibrium equation by changing applied loads. Changes in constraints affect the boundary
conditions, which constitute contact problems.

Most of the known solutions for the solid mechanics problems are based on ideal geometry
and linear approximations. However, the real nature is more complicated and inherently
nonlinear. The linear system is a very particular case of a general problem. Even the nonlinear
solutions that we seek deal with only a small subset of special cases in a general category
of nonlinear problems. When the nonlinear system is confronted, no general mathematical
solutions exist and superposition no longer applies. The system may even be nonconservative.

The first phase of the structural analysis is the idealization of a physical system into a
simpler and more manageable engineering problem. The idealization process involves simpli-
fications of the geometry, boundary and joint conditions, and loading conditions, etc. using
engineering intuitions, experimental data, empirical observations, and classical solutions. If the
idealized structural system renders a problem that cannot be resorted t o a classical method of
analysis, further idealization is required, namely discretization, for numerical analysis.

Finite elements represent spatial discretization of a continuum. As such, however, they do

not immediately impose nonlinearity. When nonlinearity has t o be taken into account for large
displacements and/or stresses, a numerical model poses new dimensions t o the discretization
in addition t o the n-dimensional Euclidean space. That is, the discretization is applied t o
time, load, and material properties by using piecewise linear curves. While discretization
allows approximate solutions by numerical methods, it introduces numerous mathematical
singularities which may complicate computational processes. Fortunately, the efficiency of
modern digital computers makes it feasible t o apply complicated computational procedures t o
the complex systems of engineering problems.

For the discrete system, governing differential equations are converted t o algebraic equa-
tions. The finite element model represents a structure by an assemblage of finite elements
interconnected a t nodal points. State variables are the displacements (displacement method
or stiffness approach) of the nodal points which carry fictitious forces representing distributed
stresses actually acting on the element boundaries. The equilibrium requirements are satisfied
at nodal points by the nodal force balance. The material constitutive laws are satisfied at the
integration points of the element. The compatibility is ensured by the displacement continuity
between elements. It is noted, however, that the compatibility of the non-conforming elements
is ensured by a patch test.
2.2 FINITE ELEMENT FORMULATION FOR
EQUILIBRIUM EQUATIONS

The variational principle renders the system governing equilibrium equations when applied t o
a functional IT, representing a total potential of a continuum [2.2-2.51, i.e.,

where U is the strain energy of the system and W is the potential energy of the external loads.
The equilibrium equations can be obtained by invoking the principle of virtual work or the
Ritz method, i.e.,

which implies that the total pot.entia1 of the system must be stationary with respect to the
state variables (displacement) for equilibrium to be ensured. The functional II is so called
because it involves the integral of implicit functions of the state variables, {u).

Considering a three-dimensional continuum for a nonlinear problem, the stationarity con-

dition results in

where the dots and 6 denote infinitesimal increments and arbitrary variations, respectively.
The left-hand side represents variations in the strain energy increment and the right-hand side
represents variations in the external work which consists of body forces b; (such as a gravity
load), traction forces ti a t the boundary surface (such as pressure loads), and concentrated
forces pi. Now it remains t o determine admissible functions expressing the arguments of the
functional II in terms of state variables {u), which are valid throughout the whole region and
satisfy the boundary conditions.

The finite element method can be characterized by the following features distinguished
from the conventional Ritz methods or the matrix method for frame structures:
r The whole region of the system is divided into numerous subdomains, called finite ele-
ments, which have simple geometrical shapes.

The variational process is limited t o each finite element, which aggregates into a whole
region when assembled.

r The admissible displacement field within each element, {fi), can be expressed in terms
of nodal displacements using interpolation functions known as shape functions, N, i.e.,

where {u) is a displacement vector consisting of all nodal points of the element.
 The strain displacement relations for the element can then be established in terms of nodal
displacements using the shape functions in Eq. (2.2.4), i.e.,

where
= < Px iy Pz >
jxy jyZjZx
and the element matrix [ B ]consists of derivatives of the shape functions, evaluated at the cur-
rent deformed geometry. Notice that the geometric linear problem requires that the element
matrix be evaluated only at the initial geometry. MSCINASTRAN employs an approximate
updated Lagrangian approach for geometric nonlinear problems, by which linear strains are
computed in the updated element coordinate system in order t o eliminate the effects of the rigid
body rotation but the equilibrium is established at the final position in the stationary coordi-
nate system. This method does not require reevaluation of the element matrix [ B ](constant
in the absence of large strains) while the element coordinates are reevaluated continuously.

Equilibrium equations for an element may be obtained by reducing Eq. (2.2.3) after the
substitution of Eqs. (2.2.4) and (2.2.5), based on the small deformation theory. Then the ele-
ment boundary stresses are statically equivalent t o the nodal forces which balance the applied
external loads, i.e.,
{ F I e = {Pie (2.2.6)
with
{F}' = 1 v
[BIT {o} dV

and
{PIe = 1 v
[ N ]{ b~) dV + [NSlT i t ) d s + {PI,
where [N,] is an appropriate interpolation function for the traction force. Notice that the
equilibrium equation for an incremental load may be expressed as

where {u) should be components of co-rotational stress which is independent of a rigid body
rotation.

The element stiffness matrix can be obtained by substituting the constitutive relations into
Eq. (2.2.7), i.e.,
= [Dl {El (2.2.8)
where
{oIT = < o x by 02 Txy Tyz Tzx >
and [Dl is a material tangent matrix. The nodal forces of an element can then be expressed as
where the element stiffness is

[KIe = 1v
[BIT[Dl [ B ]dV .
Notice that this expression represents an element stiffness due t o the material stiffness without
geometric nonlinear effects. As will be shown later, an additional stiffness [ K ~due ] to initial
stresses should be included for an incremental process because the initial stresses exist from
the second increment.

The equilibrium must be satisfied in the whole region throughout the complete history of
load application. Equilibrium equations for the global discrete system are obtained when all
the elements are assembled, i.e.,

where C over m denotes a summation over all elements. For the incremental process, the
equilibrium equation may be rewritten as

with

where {a0}represents an initial stress or the stress state at the preceding load step.

Because of the approximations involved in the interpolation functions, the finite element
model provides an approximate solution even if the equilibrium equation (2.2.11) is satisfied
exactly. Consequently, the differential equations of equilibrium are not satisfied exactly even
for linear problems, but the error decreases as the finite element mesh is refined. This conver-
gence condition is required and ensured by element formulations with regard to the element
convergence criteria. The convergence, however, may not be monotonic due to non-conforming
elements or reduced integration. By virtue of the non-conforming elements and the reduced
integration, the finite element model will have added flexibility and compensate the stiffening
effects by the displacement method. With a displacement approach, the finite element model
is generally known to produce a stiffer structure than in reality.
2.3 COORDINATE TRANSFORMATIONS

The coordinate transformation is one of the most frequent operations in the finite element
method [2.6]. Vectors and matrices defined in a particular coordinate system can be trans-
formed into another system. Coordinate systems involved in MSCINASTRAN are:
a Basic coordinate system: a cartesian coordinate system on which local coordinate systems
are defined

a Local coordinate system: defined by the user in the Bulk Data, which may include special
coordinates such as cylindrical and spherical coordinate systems

a Global coordinate system: a collective coordinate system which comprises all the local
coordinate systems specified for output quantities

a Element coordinate system: a cartesian coordinate system unique t o each element

a Displaced element coordinate system: similar t o element coordinate system but defined
in the displaced position

a Material coordinate system: a cartesian coordinate sytem used t o orient anisotropic

material properties

a Modal coordinate system: a generalized coordinate system defined for each eigenmode.
It is noted that the global system is a cartesian coordinate system, although non-cartesian
coordinate systems are adopted to orient the local cartesian coordinates for output quantities.
In MSCINASTRAN, all the displacements and forces, hence the system matrices, such as
the stiffness matrix, are expressed in the global coordinates. This implies that all the major
computations involved in the analysis are processed in cartesian coordinates. Element and
material coordinate systems are defined in the element connectivity description. Now we only
have t o consider linear transformations between cartesian coordinate systems.

Let us consider a coordinate transformation between the primed and unprimed systems
which are right-handed cartesian coordinates. The transformation matrix T consists of direc-
tion cosines of unit vectors of the unprimed coordinate system, i.e.,

where

Notice that T is an orthogonal matrix and thus

Since the work and energy are invariants with respect to the coordinate transformation, i.e.,

it follows that
PI = T P .
Then the equilibrium equation,
K'ul = PI
may be expressed in the unprimed coordinate system by

with
K = T~K'T.
It is noted that the modal matrix is used as a transformation matrix for a modal transfor-
mation which is not elaborated here.

In MSCINASTRAN, the forces and displacements are transformed from element to global
coordinates and vice versa, i.e.
{ue> = TL
Tbg {ug) (2.3.5)

where Tb, transforms from element t o basic coordinates;

Tbgtransforms from global t o basic coordinates.
It is noted that Tbeis identical for all the nodes of an element but Tbgmay vary from node t o
node in the same element. The element stiffness matrix is transformed into global coordinates
by
Kgg = T$ The Icee ~b: Tbg (2.3.7)
for which the building blocks of Tbe and Tbgare (3x3) matrices formed for each nodal point
and have t o be assembled for an entire element, e.g., for a three-noded triangular shell element

where the superscript is used t o associate each (3x3) matrix with the nodal point and is
repeated for the rotational degrees of freedom.
2.4 DISPLACEMENT SETS AND
REDUCTION OF SYSTEM EQUATIONS

The equilibrium equations, and thus system matrices, are reduced in size using the displacement
set, which is a unique feature in MSCINASTRAN. Mutually exclusive subsets of the global
displacement set, {u,}, are defined as follows:
urn: degrees of freedom eliminated by multipoint constraints;

us: degrees of freedom eliminated by single-point constraints;

uo: degrees of freedom omitted by static condensation;

u,: degrees of freedom eliminated by a Bulk Data SUPORT t o suppress rigid body motion;

ur: degrees of freedom which remain for solution after reductions.

For convenience, complementary sets are defined as follows:
Un = ug - urn
Uf = Un - 'Us
Ua = Uf - uo
Ul = Ua - UT

The subsets are defined by the user with a possible exception in the s-set if (PARAM,
AUTOSPC, YES) is used. Notice that the rigid elements are equivalent to the multipoint
constraints internally in the program, but they are not selectable in the subcases as for MPCs.
Since the set-reduction operations involve many basic modules and DMAP blocks, the math-
ematics for elimination of constraints and static condensation is reviewed here.

The multipoint constraint equations are formed in the module GP4 as follows:

where
[Rmg] = [Rmm 1 Rmn] .
The module MCEl partitions [Rmg]and solves for a transformation matrix [Gmn],i.e.,

where
[GmnI = -[Rrnrn]-' [Rmn]
Then the module MCE2 partitions the global stiffness matrix, [I&,], and reduce it t o the n-set,
from which the system is reduced to

where
I{,, = G,,
T
[KmmG,, + K,,] + [IT:, G, + K;,]

and
Qn = ~ 2 Q,m + Qk
The primes are used in K;,, PA, and QL to distinguish from K,,, P,, and Q,, which are
resulting matrices after the reduction.

Equations in the n-set can be further reduced by eliminating single-point constraints, i.e.,

which is reduced to
-ri,,Uf = Pj
where
Fj = P; - Icj, Y, (by SSG2)
with
Us = Ys.
Notice that the effects of constraint forces (Q, and Q,) are not visible in Eq. (2.4.6). The
~ingle~pointconstraint forces are recovered by

Further reduction of equations in the f-set is performed by an elimination of the o-set,

known as static condensation. The f-set is partitioned by a module UPARTN as follows:

from which
{uO) = {u:) + [Goal { ~ a )

where
uz = K
: Po (by SSG3)
and
Go, = -K: KO, (by DECOMP and FBS).
Then the reduced system of equations in the a-set is obtained as

where

and
Pa = Pi + GF~Po .
There are some rules t o remember regarding the displacement sets in nonlinear analysis.
They are:
The r-set is not supported in SOLS 66 (or 106) and 99 (or 129). Do not use a Bulk Data
SUPORT.

r PARAM, AUTOSPC is applicable only t o the superelements.

r PARAM, AUTOSPCR is available in SOL 66 (or 106) and applicable t o the o-set in
the residual structure. However, the o-set may only be used in the linear part of the
structure.

r In SOL 99 (or 129), MPCs and SPCs can not be changed in the subcase level. Therefore,
all the constraints should be specified above subcases.

r In SOL 66 (or 106), SPCs and MPCs can be changed from subcase to subcase, but they
will be changed on an incremental basis.

r Rigid elements are formulated with linear multipoint constraint equations and do not
have large displacement capability. Consequently, erroneous results will be obtained if
the rigid element undergoes a large rotation. To avoid this, stiff elements should be used
in place of rigid elements for large displacement analysis.
2.5 NONLINEAR SOLUTION PROCEDURE

The general-purpose program developer faces the task of providing the best workable solution
method for a wide spectrum of problems, while maintaining flexibility by allowing the user to
specify optional parameters. Based on the extensive numerical experiments, an attempt was
made to establish a general strategy suitable for most problems without requiring insight or
experience. Variations in combining theories, algorithms, criteria and parameter values with
numerous test problems resulted in a succinct implementation.

The major feature of the nonlinear analysis is the requirement for the incremental and
iterative processes to obtain a solution. The main issue is how t o choose the most efficient
method from the options available for the incremental and iterative processes in the solution
of nonlinear equilibrium equations. The increment size for loads or time steps has the most
significant effect on the efficiency and the accuracy of the computation, particularly in the
path-dependent problems. The incremental and iterative processes are complementary t o each
other because the larger the increment size the more iterations the solution requires. While an
excessively small increment reduces the computing efficiency without any significant improve-
ment in accuracy, a large increment may deteriorate the efficiency as well as the accuracy; it
may even cause divergence.

It is impossible to optimize the incremental step size in the absence of prior knowledge of
the structural response. The best engineering judgement should be exercised t o determine the
increment size based on the severity of the nonlinearity. Needless t o say, no incremental load
steps are required when the response is linear. In principle, the size of the load increment
(or time increment for creep analysis) should be so chosen t o yield a uniform rate of change
in strains or stresses for the material nonlinear problems and a uniform rate of change in
displacements for geometric nonlinear problems. Some adaptive methods are available, such
as an automatic time step adjustment and bisection of loads upon divergence.

User specifications for solution methods in nonlinear analyses are allowed via:
Bulk Data NLPARM for the static analysis in SOL 66 or 106 (to be selected by a Case
Control Data NLPARM)

Bulk Data TSTEPNL for the transient response analysis in SOL 99 or 129 (to be selected
by a Case Control Data TSTEPNL)

Bulk Data EIGB for the nonlinear buckling analysis in SOL 66 or 106 (to be selected by
a Case Control Data METHOD)
The increment size can be varied from subcase t o subcase by specifying different NLPARM or
TSTEPNL. It is recommended to define separate NLPARM or TSTEPNL for every subcase
even if the same values are specified, so that changes can be accommodated in the subcase
level as needed. Determination of the proper incremental time step will be treated in chapter
8, because it is dictated by dynamic characteristics as well as the loading history.
 The strain displacement relations for the element can then be established in terms of nodal
displacements using the shape functions in Eq. (2.2.4), i.e.,

where
{elT = < Px iy e, jXyjyz jzx>
and the element matrix [B] consists of derivatives of the shape functions, evaluated at the cur-
rent deformed geometry. Notice that the geometric linear problem requires that the element
matrix be evaluated only at the initial geometry. MSCJNASTRAN employs an approximate
updated Lagrangian approach for geometric nonlinear problems, by which linear strains are
computed in the updated element coordinate system in order to eliminate the effects of the rigid
body rotation but the equilibrium is established at the final position in the stationary coordi-
nate system. This method does not require reevaluation of the element matrix [B] (constant
in the absence of large strains) while the element coordinates are reevaluated continuously.

Equilibrium equations for an element may be obtained by reducing Eq. (2.2.3) after the
substitution of Eqs. (2.2.4) and (2.2.5), based on the small deformation theory. Then the ele-
ment boundary stresses are statically equivalent to the nodal forces which balance the applied
external loads, i.e.,
{F)e = {PIe (2.2.6)
with
{PIe = Jy [BIT dV
and
{PIe = /v [ N ]{ ~b ) d V t [NSlT{ t ) d s t {PI,
where [N,] is an appropriate interpolation function for the traction force. Notice that the
equilibrium equation for an incremental load may be expressed as

{F) = /v [BIT {ti) dV = {P)

where {ti) should be components of co-rotational stress which is independent of a rigid body
rotation.

The element stiffness matrix can be obtained by substituting the constitutive relations into
Eq. (2.2.7), i.e.,
(4 = [Dl (2) (2.2.8)
where
{ulT = <Ox fly 02 Try Tyz Trr >
and [Dl is a material tangent matrix. The nodal forces of an element can then be expressed as
Chapter 3

SOLUTION METHODS

3.1 ADAPTIVE SOLUTION STRATEGIES

Nonlinear finite element computations comprise material processes, element force computa-
tions, and various global solution strategies. The computational procedure involves incre-
mental and iterative processes ranging from local subincrements to global solution processes.
Performance of the finite element program can be scrutinized from three different perspec-
tives: computational efficiency, solution accuracy and effectiveness. All of these attributes of
the nonlinear program can be improved by adaptive algorithms.

There is a broad range of processes for which adaptive algorithms may be adopted in the
computational procedure of nonlinear finite element analysis. The size of the load or time
increment has the most profound effect on the efficiency as well as accuracy. However, it is
difficult t o determine optimal load or time increment size. The adaptive algorithm alleviates
this difficulty. The most CPU consuming processes in nonlinear analysis are the stiffness
matrix update operation and element force calculation. From the efficiency point of view, the
number of stiffness matrix updates and the number of iterations should be minimized, which
may be conflicting requirements. The adaptive algorithms can help t o reduce these processes
or t o choose a more desirable iteration path. Effectiveness may be interpreted as reliability
or user-friendliness. If the program provides a plausible solution or direction despite the poor
estimates or selection of control parameters, the engineering time can be reduced considerably.
The effectiveness of the program is significantly improved by the adaptive algorithms.

The adaptive algorithm is related t o cybernetics or artificial intelligence in a broad sense.

In engineering applications, the adaptive system is defined t o be an automatic control system
which has the capability of automatic measurement of process dynamics and of automatic
readjustment (or redesign) of the control algorithm. Favorable performance of the automatic
control system is obtained when control parameters are tuned t o process dynamics. The
control criterion is most critical and tuning the control parameters is the major design task.
The adaptive algorithm is analogous t o the automatic control system with the process to be
controlled being numerical rather than physical.

Major efforts have been made to implement adaptive solution algorithms for nonlinear
analysis in MSCINASTRAN. The groundwork was carried out during Version 65 development.
The adaptive capabilities for the transient response analysis were implemented in Version 66,
which included the automatic time stepping, adaptive update and bisection, and plasticity
induced bisection strategy. Further development of the adaptive capabilities continued for
Version 67, which included the load step bisection strategy for static analysis, creep time step
bisection, automatic arc-length adjustment, adaptive GAPIfriction element, and the adaptive
bisection strategies based on the large rotation and material processes. As a result, a cascade of
computational feedback control loops have been implemented in the incremental and iterative
processes in conjunction with the adaptive subincremental processes in the local level processes.
The design objectives are:
a Relieve users from uncertainties (guessing, trial-and-error).

a Maintain accuracy and efficiency without user intervention throughout the analysis.
a Promote effectiveness in terms of engineering time.
Adaptive features implemented in MSCINASTRAN nonlinear capabilities as of Version 67
are as follows:
a Newton's iteration for static and implicit dynamic analysis

- Convergence test: error functions and convergence criteria.

- Divergence criteria.
- Adaptive stiffness matrix update strategies.
- Selective BFGS updates.
- Selective line search processes.
- Adaptive bisection and recovery of load increment.
- Reiteration process.
a Arc-length methods for static post-buckling and snap-through problems

- Crisfield's arc-length method.

- Riks and modified Riks methods.
- Adaptive arc-length adjustment.
- Selective BFGS Updates.

- Adaptive switching algorithm for limiting cases.

- Adaptive correction in case of path reversal.
a Direct time integration for transient response analysis

- Two-point Newmark integration.

 - Automatic time step adjustment.
- Adaptive bisection and recovery.
Large rotations for Geometric Nonlinearity

- Gimbal angle approach with auxiliary angle.

- Rotation vector approach.
- Adaptive bisection based on rotations.
Processes for Material Nonlinearity

- Internal relaxation iteration for select elements.

- Adaptive subincremental scheme in plasticity and creep routines.
- Adaptive bisections in plasticity.
- Adaptive bisection on creep time increment.
- Error control and stiffness updates for unloading.

Processes in Gap/Friction Element for contact problems

- Adaptive subincremental process.

- Adaptive stiffness matrix update strategy.
- Adaptive bisection strategy.
- Adaptive penalty value adjustment.
- Adaptive friction t o simulate stick-slip forces.

The main goal of the self-adaptive method is t o relieve users from the chores and uncer-
tainties of guessing or trial-and-error searches. By achieving this goal, an effective algorithm
design is automatically attained for efficiency as well as for solution accuracy. The primary
virtue of the self-adaptive method, however, is not the computational efficiency but the overall
engineering efficiency. This is because the adaptability makes it possible to produce a solu-
tion at the very first trial without requiring advanced knowledge of structural behavior. As
a result, the goal of improving user-friendliness, solution efficiency and the accuracy has been
accomplished.

It is desirable to have a fully automated program which efficiently renders accurate solutions
t o nonlinear problems. Adaptive algorithms in nonlinear analysis have shone some light toward
this goal. The adaptive algorithms in the lower level processes (element or material) in concert
with adaptive solution methods in global incremental/iterative processes, make it possible
t o maintain an acceptable accuracy while maintaining the solution efficiency without user
intervention throughout the dynamic environment of nonlinear computation. Although the
optimal combination of all parameters is yet to be found, the success of the adaptive algorithms
in various levels presents the future direction of nonlinear capabilities development.
3.2 NEWTON'S METHOD OF ITERATION

The equilibrium equations in the g-set may be written as

where {P,),{Q,), and {F,) represent vectors of applied loads, constraint forces, and element
nodal forces, respectively. Element nodal forces are nonlinear functions of displacements for
nonlinear elements. Since the equilibrium condition is not immediately attained in the presence
of nonlinear elements, an iterative scheme such as the Newton-Raphson method is required.
Since the error vanishes at constrained points and the constraint forces vanish at free points,
the unbalanced forces acting at nodal points at any iteration step are conveniently defined as
an error vector by
{Ra} = {Pa} - {Fa} - (3.2.2)
Notice that the a-set is equivalent t o the 1-set in the nonlinear analysis because the r-set does
not exist. The subscript a will be dropped for simplicity in the following discussion.

Based on Newton's method [3.1, 3.21, a linearized system of equations is solved for incre-
mental displacements by Gaussian elimination in succession. The Jacobian of the error vector
emerges as the tangential stiffness matrix. The equation t o solve at the i-th iteration is

where

{ n u i > = {ui) - {ui-1)

and
{ R ~ )= {P) - { ~ ( u " } .
The iteration continues until the residual error {R) and the incremental displacements {Au)
become negligible, which is signified by the convergence criteria.

The tangential stiffness consists of the geometric stiffness in addition t o the material stiff-
ness, i.e., without regard t o the coordinate transformation,

where [ K m ]and [ K d ]refer t o the material and the differential stiffnesses, respectively. The
material stiffness is given in Eq. (2.2.10) with a material tangential matrix for [Dl. The
differential stiffness, which is caused by the initial stress, is defined as follows:
where [ B N ]represents the second order effects in the strain-displacement relations, [GI consists
of derivatives of shape functions and [MI is a function of stresses. Notice that the initial
displacement stiffness is not included in [I<T]because its effects are already eliminated in the
element formulation.
Newton's procedure is implemented using a corrective force in MSCINASTRAN until Ver-
sion 65. Recalling that the element forces for linear elements are expressed as

a corrective force vector may be defined as

{C) = {F} - [I<]{u}.

The iteration starts with initial values

{uO) = last converged displacement;

{RO} = P - ~ ( u ' ) = {AP} +
{Re};

where {AP) is an incremental load vector and {Re} is a residual load error carried over from
the last converged solution. Then the successive error vectors can be evaluated by

It is noted that the corrective force vector vanishes for linear elements. The corrective force
calculation was removed in Version 66 and the error vector is computed directly from the
internal forces, i.e.,

Then the residual load error is automatically carried over to the next incremental process.

The merit of the Newton-Raphson method is the quadratic rate of convergence, i.e.,

where u* is a true value of {u}, q is a constant, and 11 11 represents a vector norm. From a
practical standpoint, however, determination of the tangential stiffness and its inverse at each
iteration entails a considerable amount of computation. As Fig. (3.2.1) suggests, one may
resort to the modified Newton's method which requires the tangential stiffness to be evaluated
just once at the initial position, {uO), and used thereafter t o solve for {Aui}. However, more
iterations are required for a given accuracy by the modified Newton's method. The Gaussian
elimination method is better suited for this approach than the iterative descent method because
the decomposition is performed only once.
 (a) Newton-Raphston Method

LOAD

ITERATL-\
COUNT

k
DEFLECTIGN
(b) Modified Newton's Method

Figure 3.2.1 Newton's Methods for Iteration

3.3 STIFFNESS UPDATE STRATEGIES

Among other features of the solution algorithm, the stiffness matrix update has probably the
most profound effect on the success of the nonlinear solution. In spite of its significance,
however, it is very difficult to implement a robust algorithm for update strategy due to the
lack of a priori information regarding the right timing for an update. In this section, stiffness
update strategies are reviewed with respect t o the static analysis.

Variations of the modified Newton's method are adopted in MSCJNASTRAN. A variation

of the modified Newton's methods is t o update the stiffness matrix at every few iterations.
However, the modified Newton's method could lead t o divergence when the stiffness changes
drastically, as demonstrated in Fig. 3.3.1, unless the tangential stiffness is reevaluated at the
critical point. To this end, an adaptive matrix update method is indispensible. Stiffness update
strategies are established t o update the stiffness matrix on an as needed basis [3.3] such as
probable divergence.

3.3.1 Update Principles

Newton's method could be trapped in an infinite loop, oscillating about the local maximum
as illustrated in Fig. 3.3.2(a). This difficulty is overcome during the Newton's iteration by
when the tangential stiffness is not positive definite
discarding the differential stiffness, [lid],
as shown in Fig. 3.3.2(b). However, the non-positive definite stiffness matrix will be retained
and used for iteration if the arc-length method is used for the static solution. This problem is
controlled by PARAM,TESTNEG.

From the user's point of view, the stiffness update process is primarily controlled by the
input data in KMETHOD (AUTO, SEMI or ITER) and KSTEP (integer) fields of the NL-
PARM Bulk Data entry. The stiffness matrix is updated by a DMAP loop in SOL 66 (or 106)
controlled by a non-user parameter NEWK which has a value of -1, 1,or 2. The new stiffness
is formed whenever the iteration module NLITER returns a positive value for NEWK.

The value of NEWK will be set t o 1 under the following conditions, most of which have
been determined empirically:
a The ITER method is specified for KMETHOD and the iteration count (i) is an integer
multiple of KSTEP.

a The first iteration is finished with the SEMI method.

a The solution has converged with the AUTO or SEMI method and

ik 2. KSTEP

where ik is the number of iterations t o convergence after the stiffness update.

 The solution has converged with the AUTO or SEMI method and
NQNV > MAX(MAXQN - 6 , 5 )
where NQNV is the number of quasi-Newton vectors accumulated and MAXQN (integer
parameter t o be specified in NLPARM) represents the maximum number of quasi-Newton
vectors allowed.
The KMETHOD is AUTO or SEMI and if ik > 2 and
tk < t; x (N; - 7) (3.3.1)
where tk and ti are the CPU time required for a K update and an iteration, respectively,
and Ni is the expected number of iterations for convergence without a K update. The
value of N; (printed under the heading ENIC at every iteration) is estimated by
log (EPSPIE;)
N; =
log Xf
with E: being the relative error in terms of loads at the i-th iteration, EPSP the specified
tolerance for E; and
+
Xf = min (A;, 0.7 O.lX;, 0.99)
where the convergence rate Xi (printed under the heading LAMBDA at every iteration)
is computed by

with an initial value XI = 0.1.

The KMETHOD is AUTO or SEMI and if ik > 2 and
(N; - 7) > (MAXITER - i)
where i is an iteration count and MAXITER (integer parameter in NLPARM) is the
maximum number of iterations allowed for an increment.

3.3.2 Divergence Criteria

The value of NEWK will be set to 2 if the solution tends t o diverge. The MAXDIV field in the
NLPARM entry requires an integer to specify a limit on the probable divergence conditions
allowed for each iteration t o continue. There are two classifications in the divergence condition:
probable and absolute. The absolute divergence is treated as two occurrences of the probable
divergence. Divergence criteria are defined based on the ratio of energy errors evaluated
before and after the iteration, i.e., for the i-th iteration,

Then, the divergence criteria are established as follows:

 Absolute divergence if E i> 1 + or E i< -1012
Probable divergence if -1012 < E i< -1 or excessive line search,
where the excessive line search condition is defined in Section 3.4. Another absolute divergence
condition is established based on the heuristic observations t o prevent thrashing with excessive
K updates, i.e.,

KMATUP > (I + j ) and KMETHOD=AUTO

4
where KMATUP is a DMAP parameter representing the number of stiffness updates within
an incremental step.

Recalling those criteria, the solution is considered diverging if

NDIV 2 MAXDIV

where NDIV is the number of probable divergences encountered during the iteration. It
is noted that NDIV is incremented by 2 if the absolute divergence is encountered. When
the solution is diverging, or NEWK=2, and if the value of MAXDIV is positive, the current
solution base (displacement) is retracted so that the stiffness matrix is evaluated in reference
t o the preceding iteration step. If the divergence is detected a t the first iteration ( i k = l ) after a
new [K] is formed, however, the current solution base is not retracted t o prevent a repetition of
the same diverging iteration. If the absolute divergence condition (NEWK=2) is encountered
in two successive iterations in spite of the new stiffness, the computation will be terminated
with a fatal message (UFM 4551) t o that effect.

3.3.3 Time Expiration Criteria

If the remaining CPU time is not sufficient to continue the computation, the NLITER module
will set the NOGO flag (DMAP parameter NSKIP=-2) and exit with a UFM 4311 to that
effect. The purpose is to allow adequate time for the data recovery with solutions previously
obtained. Sufficiency of the remaining time (t,) is determined based on the K update time
(tk) and the iteration time (t;), depending on the succeeding process anticipated. To reserve
a minimum of 5% of the user-specified CPU time (t,) for data recovery, the remaining time is
defined as
t, = (t, - elapsed CPU time) - 0.05tS.
Then the NOGO flag will be set under the following conditions:
tT < 1.2tk
when the solution has not converged and the stiffness matrix is scheduled t o be updated.

t, < lot;
when the solution needs more iterations without a new stiffness.
 t, < 1.8tk
when the solution has converged and the stiffness is to be updated.

t, < 0.8tk
when the solution has converged and the stiffness update is not required. This condition
includes pseudo convergence, where the number of iterations has reached MAXITER
(>O). If the number of iterations reaches MAXITER while MAXITER is negative, the
NOGO flag will be set regardless of the remaining time.
Flow diagrams for the stiffness update strategies, the divergence a n d r e t r e a t proce-
d u r e , and the time expiration logic are shown for clarity in Figures 3.3.3, 3.3.4, and 3.3.5,
respectively.
 (Path taken with I

stiffness matrix update)

F
A

y (Path taken without

Y
P--
stiffness matrix update)

- - - - -

/'
I'
1 I I I
,u
u" u* UI 1

Figure 3.3.1 Hybrid Newton's Method based on

Stiffness Matrix Update Strategy
 b
DEFLECTION
(a) Iteration stops at 3 with PARAM,TESTNEG=-1 if the stiffness becomes negative.
Iteration will be trapped if continued with TESTNEG=+l or 0.

LOAD ?

~ c - Omitted for these

I two iterations

I b-
DEFLECTION
(b) Iteration sequence drops the differential stiffness with PARAM,TESTNEG=-2
when the negative diagonal terms appear in the stiffness matrix.

Figure 3.3.2 Iteration Trap Condition

 Update t i m e t t

1
I t e r a t i o n tim ti
NEYK * -1

-0
MXITER > 0

In, I
4
NSKIP

I
9 -2

Figure 3.3.3 Flow Diagram for K Update Strategies (to set NEWK)
 e Hasage for Diwrgrtce

6 w k lip a step
no
i-1 i-1 4
u .,. R
9 1

. Figure 3.3.4 Divergence and Retreat Procedure

 'I
Remaining t i m e t r
K update time tk
I t e r a t i o n t i m e ti

EXIT1: No k update
EXIT2: K update
EX I T 3 : .NOGO

Figure 3.3.5 CPU Time Expiration Logic for NOGO Condition

3.3 - 9
3.4 LINE SEARCH METHOD

The line search method is well established as a basic descent method in nonlinear program-
ming [3.4]. Although the method has been used for many years to improve the rate of conver-
gence in the nonlinear iterations, the implementation scheme was not well elaborated in the
finite element analysis [3.5-3.91. While the line search method was implemented and used in
MSCJNASTRAN, some insights were gained for making the option more viable and effective.
The line search method was implemented and modified t o optimize its usefulness for a general
class of problems in conjunction with quasi-Newton updates using MSCJNASTRAN [3.10].

3.4.1 Theoretical Basis

The functional t o minimize in the finite element analysis is the total potential of a continuum,
which is reduced t o a quadratic function of displacements {u) in linear problems as follows:

where U is the strain energy of the system and W is the potential energy of the external
loads. The basic idea behind Newton's method for the nonlinear iteration is that the objective
function (II) is approximated by a quadratic function in the vicinity of the trial solution point;

The stationarity condition of Eq. (3.4.2) results in an iteration scheme for Newton's method
with
+
{ui+') = {ui) I~'-'{R~}
where the unbalanced force {Ri) represents the negative of the gradient vector at {ui), and the
tangential stiffness matrix, [K] at {ui), represents the Hessian matrix of the total potential.

The function II must be convex in the vicinity of a solution t o have a minimum point. The
function is convex if and only if the Hessian matrix is positive semi-definite, i.e.,

where {d) denotes a direction vector, corresponding to {u - ui). Then, a feasible direction {d)
must satisfy the necessary and sufficient condition:

where the equality should be satisfied at the solution. This is the theoretical basis of the line
search method.
 The process of determining the local minimum point in a given direction is called the line
search. Considering the i-th iteration, the new solution set is determined by

with
di = K - l ~ i - l
where a positive search parameter ( a ) is determined such that

The search parameter is introduced to expedite the convergence when Newton's method is
used. It could be effectively used at a distance from the solution, where nonquadratic terms
may not be negligible. By a line search procedure satisfying Eq. (3.4.6), the solution set {u)
is moved t o a relative minimum point in the feasible direction { d ) . Thus, the solution set is
placed at a superior position for the next iteration.

The line search should be particularly effective when the modified Newton's method or
the adaptive update method is used, because the approximation employed in Eq. (3.4.2) is
not appropriate. The supplementary approximation for a quadratic function can be made by
utilizing the gradients known at two successive iterations to estimate the Hessian matrix [K],

which is known as the method of false position or the secant method [3.2]. The local minimum
point can be obtained by making the derivative of Eq. (3.4.7) vanish, which results in

This equation represents a linear interpolation in terms of { u ) and {R), which is the basis
of the line search method. Along the direction of descent {d), the objective function to be
minimized can be regarded as a function of a single variable, the search parameter. The total
potential may be expressed in terms of the search parameter by substituting Eq. (3.4.5) into
Eq. (3.4.7); i.e.,

When the derivative of Eq. (3.4.9) with respect to a disappears, the search parameter for a
local minimum is obtained by
3.4.2 Search Criteria

It is impractical in general to obtain the exact minimum point for the line search procedure
because it does not render a final solution. Furthermore, each line search costs a considerable
amount of computing time, comparable to that of a modified Newton's iteration, to evaluate
element forces. Therefore, it is often desirable to sacrifice accuracy in the line search and to
conserve the overall computing time by terminating the search procedure before it has actually
converged t o the local minimum.

One method of restricting the line search is t o limit the maximum number of consecutive
searches at each iteration cycle. Another criterion for limiting line searches is based on the
divergence rate defined by a ratio of energy errors before and after the iteration, i.e.,

with
E,, = { d i ) T { ~ i - l )
where i and k are indices for the iteration and the line search (within the i-th iteration),
respectively. Then, a criterion is established to end the line search procedure if

- LSTOL 5 Ek 5 LSTOL (3.4.12)

where LSTOL is a tolerance specified in the NLPARM entry.

In the neighborhood of the solution point, the search parameter should have a value close
to unity. The line search will result in a dilemma if the selected search parameter is too small
or too large. If it is too small (i.e., close t o zero), the iteration is not effective because the
solution vector retracts t o the old value. To circumvent such a situation, the bisection process
can be adopted when a approaches zero; i.e.,
1
anew = -(sold t
2
anew) if anew < ATOL (3.4.13)

where ATOL is a tolerance. An excessively large value of a (compared to unity) does not
validate the underlying assumption of close proximity t o the solution point and may cause a
numerical instability.

3.4.3 Implementation of Search Procedure

The search method was varied depending on the value of the relative error Ek in Eq. (3.4.11).
If Ek is initially (i.e., a t a = 1) greater than unity, the solution is diverging and the stiffness
matrix should be updated without the line search. If Ek is initially less than unity but outside
the tolerance, the search procedure is initiated by utilizing one of the following two schemes:
 Linear interpolation with Illinois algorithm if Ek < -LSTOL

Doubling scheme if LSTOL < Ek < 1

where LSTOL has a default value of 0.5. Note that the line search would not be meaningful
if the denominator (Eo) in Eq. (3.4.11) approaches zero. If it is close t o zero, in fact, the
solution should have almost converged and the line search would not be required. Under such
circumstances, the line search is blocked by assigning zero t o E l .

The domains for different operations are shown schematically in Fig. 3.4.1, where Ek is
expressed as a function of a. The relative error Ek is normalized such that Ek is unity at
a = 0. If E l is negative, there should be an a, between 0 and 1, that makes the error Ek
disappear. This value is predicted by a linear interpolation, i.e.,

where k is a line search index. With k = 1, corresponding to a = 1, Eq. (3.4.14) is equivalent

t o Eq. (3.4.10). A linear extrapolation can also be conceived by Eq. (3.4.14). However, two
points (corresponding t o indices k and k - 1) involved in Eq. (3.4.14) are not necessarily the
most recent data points acquired from the preceding line searches. These two points were so
chosen t o maintain two most recent points with the errors in opposite sign and the extrapo-
lation was avoided. This scheme is employed because the error function ( E k ) may not be a
monotonic function of a , as exemplified in Fig. 3.4.1. Thus, Eq. (3.4.14) should be rewritten as:

anew - sold
ak+l = a n e w - E n e w
Enew - Eold

The Illinois algorithm was introduced for the subsequent searches t o achieve faster con-
vergence in the succeeding search process. If two successive line searches result in the error
of the same sign, the next line search uses one-half the value of the error in opposite sign for
interpolation, i.e.,

before using Eq.(3.4.14a). An example of the Illinois algorithm is illustrated in Fig. 3.4.1. As
shown in the example, the error Ek is allowed t o increase in magnitude if it is negative because
the root (Ek= 0) is guaranteed t o exist in this case.

For a positive value of E l (for a = I ) , the zero-crossing could be predicted by a linear

extrapolation based on the same mathematical reasoning as interpolation. This extrapolation
may result in a very large value of a, if El is close to 1. For this reason, a doubling scheme
was employed as a conservative approach by simply doubling the old value of a. The search
procedure is terminated if the doubling scheme results in an increasing relative error Ek. If Ek
becomes negative (< -LSTOL), the doubling scheme is switched to the interpolation method
in the subsequent searches as shown in Fig. 3.4.2. The initial and the final values of Ek are
printed for every iteration under the headings E-FIRST and E-FINAL, respectively.
 The number of consecutive line searches allowed for each iteration is limited by a user-
specified parameter, MAXLS. This parameter, MAXLS, is defaulted to 5 for static analysis
and 2 for dynamic analysis. Upper and lower bounds of a are effectively set by combined
effects of MAXLS and the doubling scheme or the bisection. The bisection process is initiated
based on the criterion:
a < ATOL
in which
1
ATOL =
10 t MAXLS
During the line search iteration, the data relevant to the computed minimum error (Emin,amin,
and R,;,) are resumed in case the line search is terminated due to reaching MAXLS or
increasing error ( E k ) .The line search algorithm and its iteration scheme are shown succinctly
by flow diagrams in Figs. 3.4.2 and 3.4.3.

The method of false position is not globally convergent, or rather the process converges
only if it is started in close proximity to the solution point. The order of its convergence
approaches 1.618, the golden mean [3.4]. The convergence rate of the current line search process
is somewhat reduced by the doubling scheme, but expedited by virtue of the Illinois algorithm.
However, the line search process is terminated before accurate convergence is achieved, and
the convergence rate as such is rather immaterial.

3.4.4 One-Dimensional Example

It is instructive to examine the line search procedure in a single degree of freedom system. Con-
sider a fictitious structure that responds to the displacement with the internal force described
by a parabolic function:
+
F(u) = -U 2 6u.
Suppose that the structure is in equilibrium at u = 1 with an internal force (F) of 5. Starting
from this point, the external load (P) is increased from 5 to 8, as shown by a dashed line
in Fig. 3.4.4. The first iteration, ending at point 1, represents a Newton's iteration with the
initial stiffness (KO= 6); i.e.,
R0=p-FO=8-5=3

which results in

and
To perform the line search at point 1, the relative error in Eq. (3.4.11) is estimated:

Since this error is smaller than the line search error tolerance by default (LSTOL= 0.5), the
line search will normally be skipped. If the line search is enforced (e.g., LSTOL= 0.4), the
doubling scheme will be employed since El is positive. The doubling scheme will result in:

which is an exact solution and the iteration is finished. For the sake of comparison, however,
let us employ an extrapolation scheme for the line search, i.e., based on Eq. (3.4.14):

which results in:

u; = uO+ 1.714 * dl = 1.86
and
R2 = P - F 2 = 8 - 7.7 = 0.3
This point corresponds t o point 2, which can be obtained by connecting points 0 and 1 on the
dashed line.
The line search seems to be more effective in the stiffening structure, which is illustrated
with another parabolic function representing an internal force:

Suppose that the structure is in equilibrium at u = 1 with F = 2 and an external force (P) of
6 is applied as shown by a solid line in Fig. 3.4.4. The first iteration is a modified Newton's
iteration with a stiffness evaluated at certain point previously, e.g., KO= 2:

The first line search will result in:

The second line search will result in:

The second line search would not have been performed if LSTOL= 0.5. Let us assume that
LSTOL= 0.02 to continue discussion. The third line search would result in:

Now that E3 is positive again, the Illinois algorithm is invoked:

As the line search continues, the solution converges to u = 2 with a = 0.5, without requiring
additional iterations. Such a convergence by using the line search alone is a peculiar behavior
in the single degree of freedom problem. The line search has a single parameter, which is not
sufficient to render a solution to a multi-variable system unless the direction vector happened
to be oriented in the right direction.

3.4.5 Other Provisions for Line Search

Notice that the displacement vector in the g-set is required to compute internal element forces
(F) at every iteration, while the residual error vector (R), to be minimized in the iterative
process, comprises only the 1-set. For this reason, conversions from the I-set to the g-set and
vice versa should be performed at every iteration. In this context, there is another conversion
of displacement increments to treat the cumulative gimbal angle for large rotations. The line
search uses the direction vector in the g-set to avoid additional set conversions in the search
process, i.e.,
ug = ug + adg,
where dg can be obtained by merging displacement increments in the m-set (Au,) and s-set
(AY,) to the I-set displacement increment from the Newton's iteration. It is noted that the
line search cannot be performed at the first iteration if the enforced displacements (SPCD or
SPC) are specified. This is because dg includes AY, at the first iteration, but is set t o zero at
the second iteration and on.

Considering Eq. (3.4.11), the line search would not be meaningful if the denominator (Eo)
approaches zero, not to mention the numerical difficulty . In fact, if the denominator is close
to zero, the solution should have almost converged and the line search should not be required.
There are provisions to block the line search under such circumstances as follows:
Reset E1=O if ABS(Eo) < EPSW x E* for the first iteration;
Reset E1=O if ABS(Eo) < EPSWxE* for subsequent iterations;

Reset E1=LSTOL if ABS(Ek) > LSTOL and ABS(Eo) < 0.01 EPSWXE*;
where EPSW is a convergence tolerance in terms of work from NLPARM and E* is introduced
for dimensionless criteria (defined in Section 3.6). The first condition is employed for the creep
analysis because the line search or the quasi-Newton update is not appropriate for the first
iteration of the creep which always causes a small value of Eo, As will become clear in Section
3.5, the quasi-Newton update is not used if E1=O. The third condition is, therefore, introduced
to allow quasi-Newton updates while suppressing the line search.

It was discovered that the solution algorithm tends to thrash line searches when the struc-
ture becomes physically unstable, such as at the collapse load. To alleviate exhausting CPU
resources in such cases, excessive line searches are treated like a probable divergence condition
by incrementing NDIV by 1. The excessive line search is defined to satisfy

TLSCT > (2 x MAXLS - 1)

where TLSCT is the cumulative number of line searches after the new stiffness matrix is formed
within an increment.

Upon termination of the line search procedure, the results will be printed with other iter-
ation information under the following heading:
FACTOR : cr (final value for the search parameter)
EFIRST : El (relative error at a=l, divergence rate)
EFINAL : E; (Ek at the end of line search)
N-LS : LSCT (line search count k at the end)
 A Before Line Search (k=l)
C] First Line Search (k=2)
EK (El= Divergence Rate) 0 Second Line Search (k=3)

NOTES:

obtained by a linear interpolation

0 obtained by a linear interpolation with Illinois algorithm

Figure 3.4.1 Line Search Parameter vs. Line Search Error

Figure 3.4.2 Flow Diagram for the Line Search Algorithm
 T
START

EmL - 1 d 2 'ABS (u'll

I
a -
Initialire LSCl 0
1, ROOT NQ a,,
-- -0
-
Emin 10'2, En, 1 -
t
-
.-~i u i - I + , di
R' P - F Iu9

Save R ~ * if' LSC~ o -

Update R,
if flag is set
E ' - Oil IE; / < m~

A . ~a
E' - (di (Ri

Exln

Line Search Algorithm

Resume amin Emin Rmin

- NDlV - NDlV + 1

Compute enon
Print iteration information
NDlV - NDlV + 3 Convergence test

Figure 3.4.3 Flow Diagram for the Line Search Iteration

Load (F)

Displacement (u)

Figure 3.4.4 Examples for the Line Search Procedure

3.4.6 Quadratic Interpolation

A quadratic interpolation scheme is added to the line search method in Version 67. The
quadratic interpolation is implemented as an optional feature t o be used instead of the linear
interpolation. A quadratic line search on the first iteration is the default option. In order to
suppress the quadratic interpolation, SYSTEM(141)=1 must be included in the NASTRAN
definition at the beginning of the input data, i.e.,

NASTRAN SYSTEM(141)=1$

which will revive the linear interpolation on the first iteration as in Version 66.

What follows is the theoretical basis for the quadratic interpolation. Let the error (gener-
alized residual force vector) be defined (in terms of a ) by

E ( a ) = dT R ( a ) for a >0. (3.4.16)

Known values of this function are

Eo = E(0) = dT RO for a = 0

and
El = E ( l ) = dT R1 for a = 1 .
The derivative of E can be expressed as

where KO is a stiffness matrix formed at some preceeding step or iteration. In reference to

Fig. 3.4.5, the coefficients for the quadratic curve fitting can be obtained as

where cz = El from E ( l ) = El. Solving for E ( a ) = 0,

Real roots can exist for _< 0.25. Considering the previous search criteria, the doubling
scheme is used for 0.5 < 3 < 1 and the linear interpolation with Illinois algorithm is used for
3 < -0.5. Since the doubling scheme is safe and conservative, it is necessary t o introduce
the quadratic interpolation only for < -0.5. In this domain the determinant should be
Since a > 0, the positive sign in Eq. (3.4.19) should be discarded, i.e.,

which is valid for 95 0. Notice that the geometric stiffening case belongs to this domain.
The quadratic interpolation is deemed effective for the geometric stiffening case.

Figure 3.4.5 Quadratic Interpolation Algorithm

3.5 QUASI-NEWTON METHOD

The quasi-Newton update method is employed as an option to accelerate the convergence

and improve the overall effectiveness of the modified Newton's iteration. When the BFGS
method was implemented and tested in MSC/NASTRAN, our experience showed that the im-
plementation method and adjustment of parameters had a significant impact on the program's
performance [3.11].

Quasi-Newton update methods are considered the most sophisticated methods of the New-
ton's iterative solution schema and represent the culmination of extensive algorithm develop-
ment for Newton's iterative process [3.4]. They have been developed for efficient nonlinear
iterations by approximating the inverse Hessian matrix in place of the true inverse that is
required in Newton's method. Quasi-Newton methods have been widely and successfully used
in nonlinear optimization applications. One of the quasi-Newton techniques, known as BFGS
(Broyden-Fletcher-Goldfarb-Shanno) update, was introduced into the finite element method
by Matthies and Strang [3.8].

By the BFGS update, the information acquired during the iteration is used to modify the
inverse stiffness matrix. This approximate update to the inverse stiffness matrix results in
a secant modulus in the search direction. As these updates accumulate, the BFGS method
renders a stiffness matrix resembling tangential stiffness in the limit. When combined with
the line search [3.10], the performance of the BFGS update with respect to effectiveness and
efficiency depends largely on the implementation. It is extremely difficult to tune the tolerances
and parameters to the optimal condition, which may be problem dependent, for the general
class of problems.

3.5.1 Evolution of Quasi-Newton Method

The basic concept of quasi-Newton methodology is to construct an approximate inverse Hes-

sian (stiffness) matrix using information gathered during the descent process. The current
approximation is used at the next iteration to define the next feasible direction in the modified
Newton's method. Upon the first iteration, a feasible direction of descent is given by

where {R) represents an error vector to be minimized and KO is a Hessian matrix. In the
absence of line searches or quasi-Newton updates, the second iteration would lead t o the next
feasible direction:
d2 = K;' {R(ul))
with
u1 = u0 + d1
Consider a Taylor series expansion of the load error {R) about {ui), expressed as
where
R(u) = P - F ( u )
with F ( u ) being the aggregate vector of element nodal forces. Assuming that the load stiffness
due to follower forces is negligible, the stiffness matrix is formed by

In view of Eq. (3.5.1), the data from two points, {ui-') and {ui), should provide some infor-
mation about [K], because they should satisfy:

and
= R"-' - R;.

It is natural t o attempt constructing successive approximations t o [K]-' based on the data

obtained during the iterative procedure.

The earliest quasi-Newton scheme was proposed by Davidon and later elaborated by Fletcher
and Powell. By this scheme which is referred to as the DFP method [3.4], the inverse Hessian is
updated by adding two symmetric rank one matrices a t each iteration. Therefore, the scheme
is a rank two correction procedure, i.e.,

Notice that the formula simply satisfies Eq. (3.5.2) while preserving positive definiteness and
symmetry of [K]-'.

It is also possible t o update approximations t o the stiffness matrix itself, rather than its
inverse. Recalling the complementary roles of [K] and [K]-' with respect to the quasi-Newton
vectors in Eq. (3.5.2), the formula for [K] is found by interchanging y and 6, i.e.,

Another way of finding the formula for the inverse stiffness matrix is to invert Eq. (3.5.6).
This can be done by applying the general inversion identity known as the Sherman-Morrison
formula. The resulting formula is referred to as the Broyden-Fletcher-Goldfarb-Shanno(BFGS)
update for the inverse Hessian, i.e.,
Numerical experiments have indicated that the performance of the BFGS method is superior
t o that of the DFP method [3.4].

An equivalent form of Eq. (3.5.7) can be expressed for the j-th BFGS update as:

[~j]-'= [K~-I]-' [Cj]+ ~j {6j) {6jjT

[cjlT (3.5.8)

where
[cjI = [ I ] - zj {yj) {6jIT
and
1
Z'=
V 6 j I T {yj) '
This is a recurrence formula employed in MSCINASTRAN, which is applicable t o every pair of
quasi-Newton(QN) vectors. Notice that the index j for the BFGS update may be different from
the iteration index i. The stiffness matrix (therefore its inverse) is assumed t o be symmetric
and positive definite throughout the derivation.

3.5.2 Criteria for BFGS Update

There are cases where the BFGS update is not appropriate. These cases include the following:
Case 1. When the solution diverges at the current iteration.
Case 2. When the BFGS update makes [K]-I near singular.
Case 3. When the BFGS update makes [ K ]near singular.
Case 4. When two successive data points involved in QN vector generation are too closely
spaced.
Case 5. When the change in [K]-* due to BFGS update is negligible.
Characteristics of such cases should be identified so that undesirable updates can be avoided.

The BFGS update formula in Eq. (3.5.8) can be expressed in an equivalent form as follows:

where

with

and
The condition number (ratio of the largest to the smallest eigenvalue) of the matrix [A] can
be used t o determine the characteristics of [K]-l. Eigenvalues of [A] consist of multiple roots
of unity and a single non-unity eigenvalue, i.e.,

This non-unity eigenvalue itself or its reciprocal becomes the condition number.

In MSCINASTRAN, the criteria for BFGS updates are established based on the squared
value of Eq. (3.5.11), which can be simplified as:

where Ei is the divergence rate defined as

with
{di) = [K]-' {R~-')
and a is the line search parameter [3.10]. Based on Eq. (3.5.12), the following observations
can be made:
Case 1. In case of divergence ( E i > I), X has an imaginary value.
Case 2. As the condition number approaches zero, [Kj]'l will become nearly singular.
Case 3. If Ei approaches 1, the condition number approaches infinity, which makes [Kj] tend
t o be singular.
Case 4. A small change in energy (i.e., {6j)T{yj) approaches zero) makes the condition num-
ber approach infinity, which causes [Kj] t o become singular.
Case 5. The BFGS should not be needed near the convergence or in the linear case where X
is close to unity.
Hence the undesired BFGS updates can be filtered based on the squared value of the condition
number in Eq. (3.5.12).

3.5.3 Implementation of the BFGS Update Strategy

The overall picture of the iteration module, in which the BFGS update method is implemented,
is shown by a flow diagram in Fig. 3.5.1. Undesirable BFGS updates can be avoided by
establishing two admissible domains for the expression in Eq. (3.5.12), in which the updates
are performed:
T O L l < X~ < 1 - TOL2 for E% 0
and
+
1 TOL2 < X~ < TOL3 for E~ > 0.
where the tolerances are determined based on the numerical experiments [3.14]. The algorithm
begins with an initial stiffness [KO](evaluated at some previous step), which has already been
decomposed, i.e.,
K ~ L= D L ~ (3.5.14)
where L is a lower triangular matrix and D is a diagonal matrix with the pivot elements in the
diagonal. Then, the direction vector is obtained by the forward and backward substitution,
expressed symbolically as
d; = K-1 ~ i - 1
0

The quasi-Newton vectors are determined based on the data from two iteration steps, i.e.,

and
{7j} = {Ri-'} - {Ri),
where a is unity if the line search is not performed. Notice that vectors {di) and {Ri-l)
are generated before the line search is initiated and no QN vectors are created during the
line search process. If the candidate quasi-Newton vectors satisfy the criterion based on the
condition number, these vectors will be stored for the subsequent iterations by appending them
in a file (named QNV). However, the number of QN vector pairs to be stored is limited by a
user-specified parameter, MAXQN, which is defaulted t o the maximum number of iterations
allowed (defaulted t o 20). If MAXQN has been exceeded, the QNV file is not altered until it
is purged upon a stiffness matrix update. It is noted that the stiffness matrix is updated upon
convergence if MAXQN has been exceeded in order to stop dragging too many QN vectors.
The BFGS update can be suppressed by specifying MAXQN=O.

The BFGS update is a sequence of contiguous processes of dot products (4n multiplications
for each pair of QN vectors with n components), combined with a forward and backward sub-
stitution (FBS). Using Eq. (3.5.8) for the BFGS update, multiple QN vectors can be processed
as follows:

Letting
{qj-l) = [CjI {qjI
and

with initially
the recurrence formula is obtained as:

where
{qj-1) = [Cj]{ q j ) = { q j ) - Pj {^lj).
This formula is processed recursively for pj and { q j - l ) , sweeping the index backward from j
through 1. Then, the FBS operation applied to {qo) t o obtain { s o ) , i.e.,

The computational procedure can be completed by applying the following recurrence formula:

with
Tj = ~j {~j-~}.
For this procedure, the recursion is processed forward for rj and s j , sweeping the index from
1 to j. The end result { s j ) represents the direction vector { d Z ) .

3.5.4 One-Dimensional Example

It is instructive to examine the effects of the BFGS update in the analysis of a single degree-
of-freedom system. Consider a fictitious structure that responds t o the displacement with the
internal force described by a parabolic function:

Suppose that the structure is in equilibrium at u=l with an internal force ( F ) of 5. Starting
from this point, the external load ( P ) is increased from 5 to 8 , as shown in Fig. 3.5.2. The
iteration process is illustrated with data points labeled 0 through 4. The first process, ending
at point 1, represents a modified Newton's iteration with the initial stiffness (Ko=6), i.e.,

which results in
u1=u0+d1=1.5 and ~'=6.75.

If the BFGS update is performed at this point, the QN vectors will reduce to:

= u 1 - uO = 1.5 - 1 = 0.5
and
71 = R0 - R1 = 3 - (8 - 6.75) = 1.75
Introducing the BFGS update formula in Eq. (3.5.8),

from which the secant stiffness at 1 (connecting points 0 and 1) is found to be 3.5. This is a
significant improvement compared to the old stiffness (Ko=6) and the tangential stiffness (of
a value 3) at point 1. The second iteration using li;: will end at point 2 resulting in

and
p2 = 7.70.
The second pair of QN vectors can now be computed:

and
72 = R' - R2 = 1.25 - (8 - 7.7) = 0.95
which results in a modified [K] with BFGS update as:

Notice that the stiffness value of 2.639 (equivalent t o connecting points 1and 2) is arrived at by
recursive application of Eq. (3.5.8) with two sets of QN vectors, although the first QN vector
set has no effect on the end result in this case of a single degree-of-freedom problem. From this
observation, it can be deduced that the order of QN vector application is not commutative.
Invoking the BFGS update criteria using Eqs. (3.5.12) and (3.5.13), values of X2 for the first
and the second QN vectors are 1.714 and 1.316, respectively. Both of the QN vectors satisfy
the criteria t o be used for BFGS updates in the subsequent iterations.

Let us now examine the effects of combining the line search [3.10] with the BFGS update.
To perform the line search at point 1, the relative error is estimated:

If the line search is enforced by using a linear extrapolation with Eo=l,ao=O,and a l = l , the
line search factor is found to be:
which results in:
ul* = u0 $1.714 x d l = 1.86 .
This point corresponds to point 2, which is the very point that was predicted by the sec-
ond iteration using the BFGS update. This result occurs because the concept of line search
methodology is developed based on the secant method using a single parameter, and the BFGS
update is equivalent t o the line search in a single degree-of-freedom model. If the BFGS update
is applied at this point, the quasi-Newton vectors will be reduced to:

and

which indicates a secant stiffness (connecting points 0 and 2) having a value of 3.14. The next
iteration will result in point 3 in Fig. 3.5.2, i.e.,

with
R3 = P - F3 = 0.09.
Compared to point 2 obtained without the line search, point 3 shows a significant improvement.
However, the third iteration using BFGS updates without the line search would have given a
more accurate result than point 3.
 t l NQNV = NQNV +1
Initialize: l [bj]= ~ [ d ' ]
u O ,RO = P-F(IJ~),K;' l [rj]= [Ri-'1 - [R']
TOL1 = 1
TOL2 = Z j =1-
T O L 3 = 100
NQNV=O, i=O l Append [bj],[ y j ] ,Zj to Q N V
A

T YES
.i=i+l

.
l ui = ui-l + a d '

R' = P - F ( u ' )
A-

NO
di+l = K - 1 ~f

-
4
I
If Q N V exists,
~,-l= K;J1 C, - Z, 6,bT

.
l
Compute E' =
v

Line Search Iteration;

~2 =a
l-E'

NO
T

Compute errors YES

l Print iteration information
Convergence Test ( C O N V )
l Set K update flag ( N E W K )

FIGURE 3.5.1.Flow Diagram for Iteration with Quasi Newton Updates.

305 - 9
 Point u F K Derived from

0 1.0 5.0 4.0 Starting Point

1 1.5 6.75 3.0 First modified Newton iteration

2 l.8G 7.70 2.29 After line search or BFGS update

3 1.956 7.91 2.088 Second iteration with LS and BFGS

4 2.0 8.0 2.0 Solution Point

Load (F)
10

-- I

Applied load (P = 8 )

--
--

--

Displacement ( u )

FIGURE 3.5.2. Comparison of Iteration Methods

(Quasi-Newton method with or without line search)
3.5.5 BFGS Updates for Arc-length Methods

The BFGS update is required to make the arc-length methods more effective. The BFGS
updates have been implemented for the arc-length methods in Version 67. It has been proven
that the BFGS update is effective and efficient for convergence, especially when the stiffness
matrix is deficient. The BFGS updates compensate for the deficiency and frequently provide
converged solutions in difficult problems.

The iteration equation for the arc-length method (Section 3.7.1) is shown to be:

where the follower matrix may be omitted. The iterative process can be established by decom-
posing Eq. (3.5.20) into two parts:

and
KAup = A P
Then the trial solution is obtained by

with
AU = AuR + ApAup
where A p is obtained from the constraint equation.

The residual error, R in Eq. (3.5.21), is evaluated at every iteration and Eq. (3.5.21) is
solved again by FBS. However, Eq. (3.5.22) needs to be solved only once upon stiffness matrix
update in the absence of the BFGS update. Two different approaches for adopting BFGS
updates can be contemplated. The BFGS update may be applied to Eqs. (3.5.21) and (3.5.22).
Then, FBS has to be processed on both equations at every iteration. It is also possible to
adopt the BFGS update only on Eq. (3.5.21), i.e.,

where k implies that the BFGS update is incoporated into K . The BFGS does not have to
be implemented in Eq. (3.5.22), or Aup does not have to be reevaluated unless a new stiffness
matrix is formulated. The solution should not be altered by introducing the BFGS, but the
convergence will be expedited. This is because the effect of Aup diminishes as the solution
converges (Ap approaches zero).

It is necessary to modify the quasi-Newton vector y for the arc-length method. For a
constant external applied load, the quasi-Newton vector is defined as in Eq. (3.5.4). The
applied load varies as a function of the load factor (p) at every iteration in the arc-length
methods. The changes in p during the iteration can be accommodated in the quasi-Newton
vector (y) by

where NINC is the number of increments for the subcase.

Various strategies of the BFGS updates for the arc-length methods are implemented with a
control by SYSTEM(145). The NASTRAN definition at the top of the input data may include

NASTRAN SYSTEM(145) = n $

where
n = 0 (default) t o update AuR and Aup at every iteration with y*
n = l t o update AuR only with y
n =2 to update AuR only with 7'
It is noted that these options are available in Version 67.
3.6 CONVERGENCE CRITERIA

The convergence test is an important factor that affects accuracy and overall efficiency in
nonlinear finite element analysis. Out-of-balance forces and changes in displacements should
vanish upon convergence in an iterative process. The energy error accommodates both quanti-
ties and is usually adequate for most problems. However, the displacements could be in gross
error while the residual load error is negligible, or vice versa.

In order t o ensure accurate and consistent convergence, multiple criteria with errors mea-
sured in terms of displacements, loads, and energy should be combined. It is the error function
and the convergence tolerance that characterize the criteria. Error functions are formulated
using the weighted normalization so that the error measures are dimensionless. Tolerances
should be realistic for the solution scheme to be efficient. In this context, variations are con-
sidered in search of the best workable combinations of error functions and tolerances for a wide
class of structural problems.

3.6.1 Rudimentary Considerations

The convergence test is a decision-making process, on which termination of the iterative process
is based, while the true solution is not known. The convergence criteria are extremely important
for the incrementalliterative solution strategy to be effective and efficient, because improper
criteria could cause inefficiency as well as inaccuracy. It is rather astounding to find a scarcity
of publications on this subject, considering the significant impact of the convergence criteria
on the accuracy and the efficiency of the computation. Two distinct aspects are involved in
the convergence criteria:
a Error functions t o be minimized by the iteration
a Tolerances of error functions within which errors are acceptable.
Both aspects must be defined properly for the criteria t o be effective, for the solution scheme
to be efficient, and for the solution t o be accurate.

A literature search [3.7, 3.8, 3.12,3.13] has revealed that there are no universally accepted
convergence criteria t o date in the field of finite element analysis. Conditions t o be met by
ideal convergence criteria for a general-purpose finite element analysis have been contemplated.
The convergence criteria should:
a be satisfied for linear cases a t all times;
a be independent of structural units;

a be reliable (cancellation of errors are not acceptable);

a render consistent accuracy;
a be independent of structural characteristics (stiffening or softening);
 be able to handle all the loading cases including constant loading, unloading, and no
external loading (applicable to creep analysis);

r have smooth transitions after the stiffness updates and loading changes.
These conditions dictate the formulation of error functions to be discussed [3.14].

3.6.2 Convergence Conditions

The iteration continues until the convergence is attained by satisfying the convergence criteria
and the residual error vector at convergence is carried over to the next incremental step. When
the convergence criteria are satisfied, the out-of-balance forces and the changes in displacements
should be sufficiently small so that the remaining error is not physically significant nor will it
cause any detrimental effects, numerically or physically, on the succeeding incremental steps.
Convergence tolerances have the following effects:
r excessively tight tolerances cause a waste of computing resources for unnecessary accu-
racy;

r excessively loose tolerances cause not only inaccuracy but convergence difficulties in the
subsequent steps due t o cumulative errors.
The fundamental difficulty of the convergence tests for a structural analysis lies in the fact that
the base vectors (forces and displacements) involve inconsistent units, namely, combinations
of forces and moments or translations and rotations. Indiscriminate use of these vectors will
cause unit-dependent convergence criteria. For example, while an error in forces is dominant
when the model is expressed in newton-meter, the error would be dominated by moments if
the same model is described in newton-millimeter.

The most natural and reasonable criterion for the convergence test is formulated in terms of
an energy error. The energy error is the logical choice because both the out-of-balance forces
{ R ) and the change in displacements {Au) should be minimized by the iteration process.
Furthermore, energy quantities do not pose problems of inconsistent units due t o mixed units
associated with translations and rotations.

Although the convergence test in terms of energy errors is usually adequate, some distinct
errors are not detected with this criterion; i.e., displacements are in gross error while the
residual load error is negligible, or vice versa. This would be the case if the degrees-of-freedom
in error have a very small or a very large stiffness. Such cases compel the need for criteria in
terms of loads and displacements. Nominally, by visualizing the load-deflection curve for a one-
dimensional case, it can be noticed that the convergence criterion in terms of loads governs the
stiffening structure and the criterion in terms of displacements governs the softening structure.
Scalar error functions for these criteria are formulated t o be dimensionless by introducing the
weighted normalization.
3.6.3 Error Functions and Weighted Normalization

All the error functions are defined as relative errors in terms of vector norms. The vector
norm is defined by a sum of absolute values of the components (II norm or 11 [I1), where
subscript 1's are omitted for simplicity in the following development. The vectors used in
these norms, however, are not raw vectors but weighted vectors using other vector quantities
as weighting functions. The weighting functions are chosen so that the products have common
and consistent structural units.

To define a relative error in terms of energy (or work) which is printed at every iteration
under the heading EWI, the energy error is normalized by another energy quantity ( E * ) that
is relatively stable throughout the analysis, i.e.,

with

and

where { A P ) represents the incremental load (including increments in reaction forces due t o
enforced displacements) and { P ) pertains t o the total load at the previous loading step; and {u)
and { A u ) represent the total and incremental displacements at the current iteration. Norms
as defined in Eqs. (3.6.2) and (3.6.3) are not dot products and should be viewed as weighted
vector norms. They are devised t o prevent cancellation of errors, based on the concept of the
Cauchy-Schwarz in equality. Notice that smooth transitions between incremental steps upon
load changes are attempted by defining an absolute sum for { P I ) as in Eq. (3.6.4), utilizing
the triangle inequality. Thus the normalization factor does not vanish unless the incremental
load as well as the total load disappear simultaneously.

The error function in terms of loads (printed a t every iteration under the heading EPI) is
made dimensionless by a weighted normalization using the total displacement vector, i.e.,

in which the numerator and denominator are defined respectively as in Eqs. (3.6.2) and (3.6.3).
Notice that the current displacement vector is used as a weighting function; thus, the energy
units resulting from the products are common and cancel each other. Cancellation of errors
due to orthogonality is thereby prevented effectively.

In order t o formulate an error function in terms of displacements, the contraction factor

(q) of { A u ) should be introduced because the error in displacements is not known. Assuming
a geometric progression of the displacement increments, the contraction factor is defined as a
ratio of displacement increments between two successive steps, i.e.,

where i is an iteration index. If the contraction factor is assumed to remain constant with a
value less than unity, the upper bound of the absolute error in displacements can be estimated.
For all positive integers n, we have

where qn denotes the n-th power of q. Taking the limit as n approaches infinity, the above
expression is reduced to:
4
11 u - ui11 2 11 Allz 11 -
1-q
(3.6.7)

The convergence criterion in terms of displacements is not accurate, as is obvious from the
derivation of Eq. (3.6.7) with many assumptions. Contrary t o the assumption, the contraction
factor may be ill-conditioned and does indeed fluctuate erratically. An averaging scheme is
employed to smooth a fluctuating q. After trial and error, we adopted:

with an initial value q1 = 0.99.

The error function in displacements (printed at every iteration under the heading EUI) is
made dimensionless by a weighted normalization using the stiffness matrix, i.e.,

in which the weighting function {w) is formed by collecting the square root of the diagonal
terms of [I<],i.e.,
wj = 6 (3.6.10)
and the norms of the weighted displacements are defined similarly t o Eq. (3.6.2). All the
product terms have the same unit of the square root of the energy, which implies more equitable
participation of all terms in the error estimation. Note that the contraction factor is not used
for the divergence check because of its erratic and unreliable behavior.

An alternative approach has been considered t o formulate error functions in loads and
displacements using a characteristic length specified by the user. The weighting functions in
this case will be formed using a mapping vector { L ) consisting of 1's for the translational
degree-of-freedom and the specified characteristic length for the rotational degree-of-freedom.
The vector {L) and its reciprocal, denoted by { L ' ) are used as weighting functions for errors
in displacements and loads, respectively. Then the error functions in terms of loads and
displacements are defined similarly to Eqs. (3.6.5) and (3.6.9) using {L') and {L) , respectively.
With this approach, the units are unified and errors become dimensionless. This method is
intuitively appealing from an engineering perspective, because the weighting functions are
known functions and the errors are measured in terms of physical units of length and force.
However, it may be difficult or cumbersome to determine what the characteristic length should
be.

3.6.4 Implementation

The convergence tolerance determines the efficiency of the solution scheme as well as the
accuracy of the solution. The tolerance should be realistic, not too toght nor too loose. It
is difficult to choose optimal default values for the convergence tolerances. However, efforts
have been made t o set the default values to provide reliable solutions to the general class
of problems. Thus, default tolerances should be adhered t o until good reasons are found to
change them.

The convergence test is performed at every iteration after the line search process (if appli-
cable). Three error functions (in terms of displacements, loads, and energy) are computed and
compared to tolerances, i.e.,

E, < EPSU (= l o w 3 by default)

Ep < EPSP (= by default)

E, < EPSW (= by default)
where EPSU, EPSP, and EPSW are tolerances specified in the NLPARM entry. However, only
those criteria chosen by the user (combinations of U,P, and/or W ) are designed t o be satisfied
for convergence. The default values have been determined as a result of numerous tests on 47
problems with a wide variety of characteristics and options. The goal of the default tolerances
is t o consistently provide sufficiently accurate solutions to a wide spectrum of problems without
sacrificing efficiency.

The solution should converge in a single iteration if the structural response is linear. The
default option is designed to comply with this requirement. However, if the user inadvertently
specifies an undesirable condition such as an extremely small tolerance, the solution t o the
linear case should still converge regardless of the convergence criteria. Thus, the absolute
convergence criterion is designed for this purpose using the error function in terms of the load
error, i.e.,
Ep < * EPSP (= 10-lo by default).
It is noted that divergence conditions are established independent of convergence criteria.
3.6.5 Some Observations

Some objectives of the ideal convergence criteria have not been fully met by the criteria as
currently implemented. In view of Eq. (3.6.3)) errors E, and E p are amplified when the load
increment is decreased. Therefore, the reduced load increment effectively results in a tightened
convergence test. Without { A P } in Eq. (3.6.4), however, the convergence tests will fail in the
unloading cases. Without { P ) in Eq. (3.6.4), the convergence tests will fail in the constant
loading cases as in a creep analysis. Convergence tests on E, and Ep will fail if no external
loads are applied in more than two consecutive loading cases, resulting in a null vector for
{P') in Eq. (3.6.4). This is the case when the creep relaxation is analyzed under residual
stresses. Such difficulties can be avoided by introducing a fictitious load t o a disjoint nodal
point in the model. Convergence tests can be, in effect, bypassed by assigning large values to
the tolerances.

The error in displacements is not an appropriate measure for the first iteration because of
an arbitrarily assigned initial value for the contraction factor in Eq. (3.6.8). Another potential
problem exists in the displacement error for using the stiffness matrix as a weighting func-
tion. With the modified Newton method, this approach uses an old stiffness matrix while the
tangential stiffness may change significantly, which can cause improper weight distribution.
Numerical experimentation proves, however, that most problems do not require convergence
criterion in Eq. (3.6.9) if criteria in Eqs. (3.6.1) and (3.6.5) are satisfied. Introducing diver-
gence criteria independent of the convergence rate is proven effective because of the erratic
nature of convergence behavior. The divergence criteria play an important role in trapping the
run-away solution. Two categories of divergence (absolute and probable) help t o handle cases
with different structural characteristics. The probable divergence is designed to expedite the
convergence of the stiffening structure, for instance, as a complement t o the usual divergence
classified as absolute divergence.

Due t o the nature of the subject, the performance measure in terms of convergence test
is rather qualitative and heuristic than quantitative. Admitting that the test data are not
available in a systematic and statistical form, the results presented here are derived from
testing and tuning of parameters and tolerances with fifty nonlinear problems as described in
Tables 3.6.1 and 3.6.2. The work was carried out during the development of MSCINASTRAN
Version 65, and tests were conducted in comparison with Version 64. As illustrated in Table
3.6.2, the current convergence criteria combined with line search [3.10] and BFGS updates
[3.11] improved the convergence characteristics remarkably. Notice that 11problems out of 50,
which failed t o converge in Version 64, have been analyzed t o completion successfully in Version
65. The significant contribution of the current convergence criteria to the overall performance
of the program is demonstrated in providing reliable solutions with consistent accuracy for all
the converged solutions, covering a wide spectrum of problems. This effect is pronounced in
three problems which gave erroneous solutions in Version 64 and ran correctly in Version 65,
as indicated in Table 3.6.2.
 Table 3.6.1 Test Problem Description
I . BBUCKL - Beam - for Euler buckling
2. BCTEST - Boundary condition test
3. BEAMZ - 10-BEAM model to test SMALLDB - reduced from ZBEAM for short run
4. BMLIFT - Beam with Gap/Lift-off
5. CBEAM - Cantilever beam with boundary condition changes
6. CLUTCH - Electro-mechanical clutch
7. CRDEMOS - Static analysis of nozzle to sphere attachment
8. CSR3065 - Gap error with friction
9. CTESHP - 1 HEXA and 2 PENTA elements to test plastic material
10. CTHEXA - 1 HEXA element to test plastic material
11. CTPENTA - 2 PENTA elements to test plastic material
12. CTSHELL - A single QUAD4 to test plastic material, loading and unloading
13. DRUM - Drum head model with thermal load
14. FOLFO - Follower force test problem with CRDEMO
15. GAPS - Gap elements - opening and closing in sequence
16. GAPTST - Gap test with electro-mechanical clutch
17. GOODRICH - 2 layers of HEXA's to model bi-metal
18. HAMMER - Cantilevered rod hit by a mass with initial velocity specified
- mass connected to rod by gap
19. HDRULE - Kinematic hardening rule test - tension,unloadiong, compression
20. LGDROD - 2 DOE model with thermal load
21. LGDTEST - 3 shell elements to test LGDISP
22. MNEWTON - Modified Newton method with line search on RODTEST
23. NLCTIRE - Tire model with composite element and large displacement effects
24. NLELAS - Nonlinear-elastic material with differnet tension- compression properties
25. NLELAS2 - Shell elements with bilateral nonlinear-elastic material
26. NLELAS3 - Nonlinear-elastic HEXA with hydrostatic load
27. NLPATHl - Non-proportional loading specified by strain path
28. NLPATH2 - Non-proportional loading specified by stress path
29. NLROD - 1-D plasticity in rod
30. NLSHELL - Imperfect spherical shell with shell element
31. NLSOLID - Imperfect spherical shell with solid model, includes buckling analysis
32. RODTEST - 2 DOE test problem for BEGS method with line search
33. SBUCKL - NLSHELL - for buckling analysis
34. SE66S1 - Superelement model of two tips plus a residual(TRIA3) in sort 1
35. SE66S2 - Superelement model of two tips plus a residual(QUAD4) in sort 2
36. SFRAME - Curved box-beam test for large displacements
37. SHLBEAM - Cantilevered beam with shell elements
38. SNAPS - Snap-thru of a spherical dome
39. SUPLCOP - Non-superelement model of two tips plus a residual
40. SWAVE - Cantilevered rod with a step forcing function at the end, testing stress wave
41. T C R P - Creep of plate under thermal load
42. TRIPOD - Basic nonlinear test with rod elements
43. UNIAXlP - Creep at constant stress
44. UNIAX2P - Elastoplastic-creep, reproducing isochronous curve
45. UNIAX2R - Code test for creep - relaxation prediction
46. UNIAX2T - Creep under variable temperature
47. UNIAX4 - Effects of stress reversal
48. VGAP - Gap verification with nonlinear transient analysis
49. ZBEAM - Z-shaped beam analysis with 10 BEAM elements
50. APLATE - Z-shaped beam analysis with 10 QUAD4 elements
 Table 3.6.2 Test Results Cornparision : V64 vs. V65
Problem No.Iters No.K Updates No. L.S. I Remarks
Name V64 V65 V64 V65
BBUCKL 5 5 0 0
BBUCKL(R) 2 2 0 0 0
BCTEST 26 22 0 0 0
BEAMZ - 30 - 6 0
BEAMZ(R) - 10 - 2 0
BMLIFT 2 2 2 2 0
CBEAM 6 6 0 0 0
CLUTCH 9+ 54 3 12 56 1
CLUTCH@) - 141 - 6 8
CRDEMOS 14 16 0 0 1 2
CSR3065 8+ 5 8 5 0 3
CTESHP 5t 15 2 2 3
CTHEX A 5+ 15 3 3 0
CTPENTA 12+ 9 3 1 4
CTSHELL 15 23 2 0 0
CTSHELL(R) 3 3 0 0 0
DRUM - 36+ - 3 31
DRUM(R) - 15 - 2 10
FOLFO 12 9 0 0 0
GAPS 39 26 3 1 44 4
GAPTST 29 24 3 1 34 5
GOODRICH 11+ 155 3 14 20
HAMMER - - - - - 6
HDRULE 21+ 15 9 2 6
HDRULE(R) 13 5 6 1 1 7
LGDROD 51 67 17 3 31 8
LGDTEST 17+ 40+ 3 5 36 9
LGDTEST 4+ 15+ 1 1 5
MNEWTON 4+ 79 1 15 18
NLCTIRE 13 39 6 3 5 10
NLELAS - 35 - 1 8
NLELAS2 16+ 40 0 1 7
NLELAS3 - 11 - 0 1
NLPATHl 25 20 4 4 0
NLPATH2 35 28 9 4 3
NLROD 10 9 3 3 0
NLSHELL 78 68 13 4 0
NLSOLD 77+ 74+ 21 11 20 11
NLSOLID(R) - 15 - 3 4 12
RODTEST 35(+) 47(46) 14 12(9) 29(30) 13
SBUCKL 51 60 13 13 6
SBUCKL(R) 10 11 2 2 0
SE66S1 12 13 2 2 0
SE66S2 11 13 0 0 0
SFRAME 54+ 71+ 19 15 13 14
SFRAME(R) -
7+ - 3 5
SHLBEAM 55+ 82 2 0 4
SHLBEAM(R) 60 41 6 4 0
SNAPS 36 40 5 2 15
SUPLCOP 9 10 0 0 0
SWAVE - - - - - 6
TCRP 14 13 1 0 0
TRIPOD 14 22 2 1 14 15
UNIAXlP 20 13 4 2 3
UNIAXZP 75 89 5 3 21
UNIAX2R 31 46 0 0 0 15
UNIAX2T 146 179 3 2 0
UNIAX4 31 37 5 4 0
. VGAP 22 22 4 4 0 6
ZBEAM 267+ 1907 54 200 310 16
ZPLATE 822 2621 292 335 291 16
+
(R) : Restart run, : Diverging solution, - : V64 ble for this analysis
3.7 ARC-LENGTH METHODS FOR
POST-BUCKLING ANALYSIS

The static nonlinear response of a structure can not be analyzed beyond the critical limit load
by the conventional Newton's method. Although the post-buckling state is not usually allowed
in the structural design, the prediction of such response could be of interest in some cases. In
the design process, for instance, it may be desirable t o trace the response of the snap-through
or post-buckling behavior. The arc-length methods allow solutions in the unstable regime for
such class of problems. It is noted that they are not efficient solution methods, but effective
methods t o render solutions in the unstable regime which is not necessarily a physically viable
state. An arc-length method, known as Crisfield's method, was implemented in Version 66
of MSCINASTRAN. However, some convergence difficulties have been identified and resolved
in Version 66A. In addition, a number of options for arc-length methods and the adaptive
bisection algorithm have been implemented in Version 67.

3.7.1 Basic Theory for Arc-Length Method

The concept of the arc-length method is to modulate the applied loads in order t o produce
solutions with displacement increments of manageable size for a given load step. In order t o
modulate the applied load, an additional variable (the load factor) and a constraint equation
are introduced. There are various approaches t o providing a constraint equation [3.15-3.181.

Consider a residual load {R}

where F represents the internal forces, and the total external load P is expressed as

where Po denotes the applied load at the end of the preceding subcase, A P represents the load
increment in the current subcase, and p is the load factor varying from 0 to 1 (but not t o be
limited to this range) within the subcase. Linearizing { R } about ( u , p ) , Eq. (3.7.1) can be
expressed as

Based on Eqs. (3.7.1) and (3.7.31, the equilibrium condition at ( u + Au, p + Ap) dictates that

where [Elis the follower matrix, [g]is the stiffness matrix [I<],and = AP.
 The iteration equation could be derived by rearranging Eq. (3.7.4):

where the follower matrix may be omitted. The iterative process can be established by decom-
posing Eq. (3.7.5) into two parts:

Ir'AuR = R(u,p ) and Ir'Aup = A P (3.7.6)

Then the trial solution is obtained by

with
Au = AuR + ApAup
where A p can be obtained from the constraint equation.

3.7.2 Riks' Method and Its Variations

The displacement increment is limited by a constraint equation:

where w is a user-specified scaling factor (to be specified in NLPCI) and A t is defined by

if the initial value of A p is given. The constraint of Eq. (3.7.9) has a disparity in the dimension
by mixing the displacements with the load factor. For this reason, the scaling factor (w) is
introduced so that the user can scale p to the appropriate dimension or delete the A p term.
The default value of w is zero (Fig. 3.7.3b). The iteration follows the path on the plane normal
t o the initial tangent as shown in Fig. 3.7.1. Therefore the subsequent iterations (i > 1) must
satisfy
{AU')~{AU~) + * api= 0 (3.7.10)
Recalling that the first iteration should result in

Eq. (3.7.10) may be reduced t o

from which the load factors for the subsequent iterations are determined by
and
pi = pi-l +

Notice that the normal plane does not change during the iteration by Riks' method. In addition,
{nup} remains constant if the iteration process is the modified Newton's method.

Alternatively, the normal plane may be updated at every iteration. If the normal plane
is t o be normal t o the cumulative incremental displacements for the preceding iterations as
shown in Fig. 3.7.2, the orthogonality condition in Eq. (3.7.10) should be modified to:

The increment in the load factor for i > 1 is obtained by solving Eq. (3.7.10b), i.e.,

This variation of Riks' method has an advantage over Crisfield's method as it avoids the
solution of a quadratic equation.

3.7.3 Crisfield's Method

Instead of iterating on the normal plane, the solution is sought on the surface defined by
Eq. (3.7.9) with an arc-length of A! as depicted in Fig. 3.7.3a, i.e.,

This constraint can be interpreted as keeping the incremental displacement constant, if w=0,
as shown in Fig. 3.7.3b. Substituting Eq. (3.7.8) into the preceding equation, we obtain a
quadratic equation in terms of Ap:

where

Since Crisfield's method leads t o a quadratic equation, the selection of the proper root of
this equation becomes the most critical process for the success of this method. There are two
The root is chosen so that the angle between two vectors {ui-I- uO)and {ui- uO)is less than
90 degrees, i.e.,

There are cases where no roots can be found. Such is the case when the trial solution is far
from the true solution and stays outside the region covered by the arc-length. In this case, the
trial solution vector is scaled so that the direction vector intersects with the surface defined by
Eq. (3.7.12).

The wrong choice of the root could cause an unintentional loading path reversal, by which
the solution returns t o the previous state. Such cases can be detected by checking the orthogo-
nality of the incremental displacements of the two successive solutions. If this case is detected,
the root is chosen so that the angle between {ul- uO)and {ui- uO)is an acute angle.

3.7.4 Adaptive Arc-Length Method

It would be difficult for a user to estimate a proper arc-length for multi-degree-of-freedom

problems. The initial arc-length for the Crisfield method can be determined by

with
Ap 1 = p1= 1
-
NINC
Notice that NINC is a user-specified value for the number of increments in the subcase.

The arc-length should be continuously updated at every increment using the information
gathered during the preceding increment. One method is t o reduce the arc-length if it requires
an excessive number of iterations to attain a converged solution, i.e.,

where Id is the desired number of iterations for convergence and I,,, is the number of iterations
required for convergence at the preceding step. It is also desirable to decrease the arc-length
if the stiffness ratio increases, i.e.,

with
where the subscripts ( j , j - 1, j - 2) denote the converged solution steps. In terms of the
arc-length, the computation is simplified as

* A!,-,
Ir',at;o = ABS
Apn
I
The adaptive process should be based on the arc-length ratio, i.e.,

Atnew
ALRATIO = -
Atold
Combining two criteria, the new arc-length ratio may be adapted to the nonlinearity by

If (iz
Imax
1
and -
Ii7Tatio
then ALRATIO = MIN

If (dZ
Imaz
and - then ALRATIO = MAX

else ALRATIO = 1.
In order t o maintain the stability for the adaptive process, ALRATIO should also be bounded,
lee.,
MINALR < ALRATIO < MAXALR (3.7.19)
where MINALR and MAXALR can be user-specified parameters, defaulted to 0.25 and 4.,
respectively. If the adjusted ALRATIO falls outside the bounds, ALRATIO should be reset
t o the limit. Then the arc-length can be updated at the beginning of the next step based on
ALRATIO as follows:
A,! = ALRATIO * Atold

If the iteration produces pi which exceeds the value of p for output points, Api should be
adjusted t o the specified value of p for the output, i.e.,

Ap* = MIN(Ap, p* - pi-1) (3.7.20)

and
Au* = Auk +A~*AU$
where p* is the specified value for output (end of subcase or intermediate points). In the
stiffening case this limit may be exceeded at the first increment of the subcase. This limiting
process may be implemented as follows:

Process Eq. (3.7.20) at every iteration with

1
p* = C-
NINC
 if INTOUT=ALL, otherwise p* = 1.
Then if
+
1) ui nu*- UO ( ( <ne
Api = Ap* and Aui = Au*
otherwise ignore Ap* and Au* and proceed t o update ui and pi.

The arc-length should be updated upon convergence whenever this limiting process takes
place. Notice that this limiting process in effect reduces the arc-length method to the conven-
tional Newton's iteration.

In the unstable regime where the stiffness is negative, the load factor decreases with a
forward step. When this happens, the sign of Apl should be reversed. This possibility should
be examined at the beginning of each increment. The sign can be determined by the sign of a
dot product, i.e.,

Ap'3 = - ( p .3 - p 3-1
. )

An adaptive bisection algorithm is also incorporated to cope with divergent cases. If the
iterative process using the arc-length method tends to diverge, the arc-length is bisected. The
bisection is combined in concert with the stiffness matrix update strategy. The bisection
procedure will continue until the iterative process is stabilized and a converged solution is
found. However, the number of contiguous bisections is limited by a user-specified parameter,
MAXBIS in the NLPARM Bulk Data entry. The variable arc-length at every increment invokes
the recovery from the bisection process once the difficulties in convergence are overcome.
 U

Figure 3.7.1. Riks' Method

Figure 3.7.2. Modified Riks' Method

(a) Arc-length in terms of Combined Variables

(b) Arc-length in terms of Displacements

Figure 3.7.3. Crisfield's Methods.

3.7.5 Verification Problems

The most notable applications of the arc-length method are analyses of the snap-through
behavior and the post-buckling behavior. Such examples are illustrated below.

SNAP-THROUGH AND SNAP-BACK: A SPHERICAL SHELL WITH BACKING PLATE

A spherical diaphragm resting on a rigid backing plate is subjected to a uniform pressure.

The analysis objective is to find the structural response for snap-through and snap-back under
the clamped boundary condition. Additional constraints are applied after the snap-through
by the backing plate. The material is linear elastic, and the deformation pattern is assumed
t o be axisymmetric because the diaphragm is connected t o a plunger which prevents rotation
of the apex.

A 10-degree sector of the diaphragm was modeled using shell elements with axisymmetric
boundary conditions. The contact process between the diaphragm and the backing plate was
simulated by gap elements as shown in Fig. 3.7.4. The main features of these problems were
geometric nonlinearities due t o large displacements, follower forces, and changes in constraints
due to the backing plate. This model was analyzed by Crisfield's method and the results
are compared to those obtained by the conventional Newton's method [3.10]. In the latter,
the convergence behavior was rather erratic and unpredictable around the critical loads. In
addition, since Newton's method is unable t o trace the structural response throughout the snap-
through process, the obtained solutions were discontinuous. The missing solution portions were
conjectured and represented by dashed lines.

The load-deflection curve at the apex point, constructed by combining loading and un-
loading paths, is shown in Fig. 3.7.5. The snap-through occurs at around 0.97 psi, and the
corresponding displacement is 0.18 in. at the apex point. During the snap-through process
(between points A and B), the displacement increased with a decreasing load until the stiffness
became positive again when a tensile stress started t o build up in the membrane direction of
the shell. This turning point (point B) corresponds t o the snap-back point with a pressure
of 0.41 psi and a displacement of 0.61 in. Beyond point B, the pressure rose rapidly as the
solution proceeded. Since the apex point came into contact with the backing plate at 0.98 psi,
the point did not translate beyond the gap distance (0.757 in.) despite the increasing pressure.

This problem does not involve the material nonlinearity and, therefore, the unloading
curve should follow the reversed loading curve. This proved the validity of Crisfield's method
in MSC/NASTRAN. Note that the unloading steps did not coincide with the loading steps
since the load increment varied adaptively during the analyses.
 BOUNDARY CONDITIONS

G
I

A SECTION OF A
7SPHERICAL SHELL

Figure 3.7.4. Spherical Diaphragm Model with Axisymmetric

Boundary Condition and Constraints by Gap Elements.
 x Loading (Newton)
Unloading (Newton) 0
X

A Loading (Crisfield)
Unloading (Crisfield) !
- - - Conjectured
0x
0

Load
\
v
(psi) /
rt ;
&
A
' ,$
g
\A\ 3
M - B ~ ~

- .j - ;. ;
.
i..
!
-.-----..---?--.-7--..-..-.r-
: ; : I ;
. .- , .?----..-

0.5 0.75
Displacement (in)

Figure 3.7.5. Load vs. Vertical Displacement at the Center

(Clamped Case).
POST-BUCKLING BEHAVIOR: AN IMPERFECT SPHERICAL CAP

A spherical shell with an initial imperfection was analyzed [3.19]. The shell was subjected t o
an external uniform pressure, while the periphery was clamped. The problem was assumed to
remain axisymmetric in geometry and loading throughout the deformation. The details of the
geometry and the material are shown in Fig. 3.7.6. The initial imperfection was introduced by
making the radius of curvature near the apex greater than the shell radius. The material was
elasto-plastic with von Mises yield criterion and kinematic hardening. The large displacement
effect was also included in the analysis. The finite element model shown in Fig. 3.7.7 represents
a 10-degree sector with axisymmetric boundary conditions.

In the analysis, the external pressure was gradually increased: 2000 psi, 3000 psi, and 4000
psi at the end of SUBCASE 1, SUBCASE 2, and SUBCASE 3, respectively. Since Newton's
method is usually more efficient than the arc-length methods if the stiffness matrix is positive-
definite and not close to being singular, it was decided to use Newton's method in the first two
subcases and Crisfield's in the third. The solution proceeded successfully and passed beyond
the peak point, unlike Newton's method. The run automatically stopped when the number
of load increments reached the maximum limit MXINC. Fig. 3.7.8 shows the load-deflection
curve at the apex. The peak pressure and its corresponding displacement are identified as 3574
psi and 0.0139 in., respectively. Since a finite number of load steps was used, the predicted
solution would not have passed through the exact peak point, but the error in the buckling
pressure should be small because the curve is plateaued in the vicinity of the peak. On the other
hand, the maximum displacement at buckling lies between 0.0139 in. and 0.0159 in., where the
DECOMP module encountered negative terms in the factor diagonal of the stiffness matrix.
By using the arc-length method, the buckling load as well as the post-buckling behavior can
be predicted in a single run. The deformed shapes are shown in Fig. 3.7.9.

When the same model was analyzed by using Newton's iteration method in all three sub-
cases, the solution diverged in the third subcase. A restart run was made in SOL 66 for the
eigenvalue analysis [3.20], starting from the solution at P=3300 psi. Solutions a t P=3400 psi
and 3500 psi were repeated and a buckling analysis was performed using the stiffnesses be-
tween those two steps. The buckling pressure and displacement estimated by this method are
3547 psi and 0.0122 in., respectively. This solution is not as accurate as that predicted by the
arc-length method.
Geometw: Material:

-
h = 0.0251 in.
R 0.8251 in.
R, = 1.1506 in.
7075-T6 Aluminum
E = 10.8 x 1O6 psi
u = 0.3
a = 0.267 in. q = 7.8 x 104 psi
y = 20'

-
b = 0.14328 in.
a 14.3065'
c = 0.32908 in.
Strain Hardening:
Slope E, = 1.1 x 1O6 psi
P = 37.7612' Kinematic

Boundary Condition:
Periphery Clamped

Figure 3.7.6. Imperfect Spherical Shell.

Figure 3.7.7. MSCJNASTRAN Model for Imperfect Spherical Shell.
 4000.
A
- * B
A
a
3000. -
a C
A,

- t
?
Pressure
(psi)
2000. -
$ri O Newton's method
$ Buckling point (Newton)
- i A Crisfield's method
4 Peak point (Crisfield)
- Kao
1000. -

0. I I I 1 I I I
0.00 o.bl o. b2 0.b3 0.04 0.05
Central Deflection (in)

Figure 3.7.8. Load-Deflection Curve of Imperfect Spherical Shell.

 0: Undeformed Shape
A: at q=3574 psi, u=0.0139 in.
B: a t q=3243 psi, u=0.0286 in.
C : at q=2863 psi, u=0.0495 in.

Figure 3.7.9. Deformed Shapes at Different Buckling Stages.

Chapter 4

FINITE ELEMENTS IN
NONLINEAR ANALYSIS

4.1 OVERVIEW

In nonlinear finite element analysis, lower-order elements are often preferred over higher-order
ones because of their robustness and reasonable accuracy at reduced costs. MSCINASTRAN
employs linear elements, rather than quadratic or cubic elements, t o process nonlinearity.
When using lower-order elements, quadrilateral and hexahedral elements are generally pre-
ferred over triangular, pentahedral or tetrahedral elements. Triangular and tetrahedral ele-
ments can exhibit excessively stiff behavior, and caution is needed when using these elements
in bulk.

Caution is also needed when different element types are combined in a model, and if these
elements are incompatible (e.g., HEXA and QUAD$). In such cases, some provision (e.g.,
appropriate constraints) may be necessary at the interface boundary. Modeling the joints
(such as bolted, riveted, or welded) is particularly difficult. For lack of better information, the
joints are usually modeled as rigid or free in certain degrees-of-freedom. If improved accuracy is
required at such joints, the characteristics of the joint (stiffness and/or damping) may have to
be identified from experiments or the local analysis of a detailed model at the joint. Modeling
of the boundary conditions at the supports poses similar difficulties. Ideal boundaries are
represented as free, clamped, pinned, roller or ball joints. The reality tends to be in smeared
condition.

Another common modeling problem involves offsets a t the interface boundary. Offset op-
tions in beam, plate and shell elements are provided by MSCINASTRAN. These offset options
are not fully validated for nonlinear analysis, especially for material nonlinearity. Alternatives
to these offset options are
 using very stiff rod elements for the offset distance

using rigid elements or MPC relations if the large rotation is not involved.
The MSCINASTRAN element library includes about forty elements, of which some are
considered obsolete and some are not applicable t o the nonlinear analysis. Elements which
may be used and are recommended for nonlinear analysis are tabulated below:

Line element Surface element Solid element 0t her elements

* CROD * CQUAD4 * CHEXA CGAP
* CONROD CQUAD8 * CPENTA CELASi
* CTUBE * CTRIA3 * CTETRA CTRIAX6
CBAR CTRIA6 +* CQUAD4,CTRIA3
* CBEAM CSHEAR CONM2
CBEND RTRPLT GENEL
RROD,RBAR RBE2,RBES
Notes:
* denotes elements capable of geometric and material nonlinear processes.
+ denotes a plane-strain element.
All the elements listed above may be used in the nonlinear analysis (SOL 66 and SOL
99) if they are combined properly in the model. Elements that are potentially capable of
nonlinear processes (those denoted by *) become actively nonlinear if the parameter LGDISP
is turned on or nonlinear material is assigned t o those via MATS1 or CREEP bulk data. As
for geometric nonlinearity, MSC /NASTRAN does not currently support large strain capability.
However, large displacement is treated effectively by computing element stresses and strains in
the updated element coordinates. Active nonlinear elements may only be used in the residual
superelement. Linear elements [4.1], however, may be used in the residual superelement for
the region that remains linear throughout the analysis. All the upstream superelements are
assumed t o remain linear throughout the analysis.

4.1.1 Line Elements

The truss type elements (CROD, CONROD and CTUBE) differ only in input convenience.
The ROD element is a straight element connecting two GRID points, with axial and torsional
stiffnesses in displacements u and 8, respectively. The ROD element can sustain axial force
P and torque T, producing axial and shear stresses as follows:
 and

with
ox = E dcx and Toz =GYB~

where E is the Young's modulus or an appropriate modulus for the nonlinear material and G
is the shear modulus.

Geometric nonlinear effects are reflected in the stiffness via the differential stiffness and the
internal forces using the method of updated element coordinates. However, nonlinear material
behavior will only be exhibited in the axial deformation. The torsional stress is assumed linear
and not coupled t o the axial stress, i.e., the linear shear stress is simply computed by

where T is the torsional moment, J the torsional rigidity, and c the user-specified value to
determine the stress. The stress is determined based on a single Gauss integration; therefore
the stress and strain are assumed constant throughout the length. The strain is computed
based on the small strain theory, i.e.,

where L and Lo denote the current and original length of the element, respectively. The
stiffness matrix for a ROD element is the (8x8) matrix shown below:

with

where
k,= -,E A GJ
ke = - and
Aax
kd = -
L L L

The BAR element is derived from classical beam theory; the plane cross-section remains
planar under deformation. It provides exact linear solutions for end loads of any kind. Its
major limitations are that it is prismatic (i.e., properties do not vary along the length), that
the shear center and the neutral axis coincide, and that the torsional stiffening due t o warping
is neglected. These limitations are relaxed with the BEAM element in addition to the nonlinear
capabilities to be described in the following section. Geometrical properties and the directions
of internal forces of the BEAM are shown in Fig. (4.1.1).
 The BEND element forms a circular arc, connecting two nodal points. Its primary use is
to model a piping elbow. RROD and RBAR are rigid elements, corresponding to ROD and
BAR, respectively.

4.1.2 Surface Elements

The plate and shell elements (QUAD4, QUAD8, TRIA3, TRIAG) can be used to model mem-
branes, plates and shells. The quadratic elements (QUAD8 and TRIAG) are applicable only
for the linear part of the structure and provide mid-side nodes that may be deleted. These
quadratic elements can represent the singly-curved shells or linear curvature shells more ac-
curately than the linear shell elements at the same cost. However, this is not the case in the
doubly-curved shells.

Linear shell elements (QUAD4 and TRIA3) may be used for nonlinear analysis with ge-
ometric and material nonlinearities. In general, the quadrilateral elements are more reliable
than the triangular elements, because the TRIA3 element exhibits behavior too stiff in mem-
brane action. Properties of shell elements are specified in the PSHELL data, but PCOMP can
be alternatively used to model the composite layers. If the PCOMP is selected, however, no
nonlinear material may be referenced. In other words, no material nonlinear capability exists
in the composite elements.

The element coordinate systems are implicitly defined by positions and the order of the
nodal connectivity. The different material coordinate system with respect to the element
coordinate system may be defined in the shell element connection card as shown in Fig. (4.1.2).
Components of forces and stresses, as shown in Fig. (4.1.3), are always recorded and output
in the element coordinate system.

The SHEAR element can be used t o model the shear panel in the linear part of the structure.
The RTRPLT element may be used to model a rigid plate in the linear structure.

4.1.3 Solid Elements

Solid elements (HEXA, PENTA and TETRA) are quadratic isoparametric elements that may
be degenerated to linear elements. Various schemes of Gaussian quadrature may be selected
for the volume integration in the element property data specified in the PSOLID entry. The
element coordinate systems are implicitly defined by positions and the order of the nodal con-
nectivity (see description of HEXA, PENTA and TETRA for details). The material coordinate
system (defaulted to the element coordinates) can be defined in the PSOLID entry. Stresses
and strains for these elements are recorded in the material coordinate system.

The quadratic element with mid-side nodes may not be used in the nonlinear region. Linear
solid elements applicable t o the nonlinear region will be discussed in the following section.
4.1.4 Other Elements

The GAP element simulates a point-to-point contact including the friction effect, but without
effects of large rotation. The formulation is based on the penalty method. The GAP element
is a nonlinear element and loses its significance if included in a linear analysis. Plane strain
elements are available for nonlinear analysis using QUAD4 and TRIA3 elements by specifying
-1 in the MID2 field of the associated PSHELL entry. Plane strain elements were implemented
by suppressing bending and enforcing the plane strain condition instead of the plane stress
condition.

Other elements, which are linear but available for nonlinear analysis, include a lumped
elastic spring (ELASi), an axisymmetric element (TRIAXG), a lumped mass (CONM2), a
general element (GENEL), and rigid elements (RBE2, RBE3). The GENEL element is used
t o represent the stiffness of a substructure which has an arbitrary number of connected grid
points. The input d a t a may be obtained either from another computer run or from test data.
Figure 4.1.1 BEAM Element Coordinate System and Element Forces

4.1 - 6
 (a) TRIA3

(b) QUAD4

Figure 4.1.2 Plate and Shell Element Coordinate Systems

4.1 - 7
 (a) Forces and moments

(b) Stresses

Figure 4.1.3 Forces and Stresses in Plate Elements

4.1 - 8
4.2 FORMULATION OF ISOPARAMETRIC ELEMENTS

In the finite element method, parametric mapping is frequently used t o map an irregular
region into a regular one. The coordinate system used in the parametric mapping is a natural
coordinate system suitable for the geometry. For instance, a natural coordinate system (I,7 )
is used for a quadrilateral surface in which each corner node has an extremum value of $1 or
-1 in and 7 . Then, the geometry of the internal points of the finite element can be described
in terms of the nodal coordinates by the parametric mapping. The mathematical expression
for this parametric mapping or interpolation is called a shape function. The displacement field
inside the element should also be interpolated in terms of nodal displacements. The mapping
is isoparametric if the same shape function is used t o interpolate the displacement field as well
as the geometry. The merit of isoparametric mapping is that the displacement field is invariant
t o the orientation of the cartesian coordinate system x and y.

Most of the MSCINASTRAN elements are isoparametric elements, in which the shape
functions are expressed in terms of isoparametric coordinates. In what follows, derivations
are shown for the isoparametric coordinates, shape functions, element matrix describing the
strain-displacement relations, volume integration for stiffness and mass matrices by Gauss
quadrature, and element loads are derived for a tetrahedron element t o illustrate element
related operations.

4.2.1 Isoparametric Coordinates

Cartesian coordinates are not convenient t o describe the geometry or the displacement field of
a tetrahedron element. Let us introduce a set of volume coordinates (L1 L2 L3 L4), such that

Volume (P234)
L1 = ,etc.
Volume (1234)

as shown in Fig. 4.2.1. The relation between volume and cartesian coordinates can be estab-
lished for a linear tetrahedron (without midside nodes) as follows:

It is obvious from Eq. (4.2.1) that the shape functions are simply the volume coordinates, i.e.,

N1 = L1 , N2 = Lz , ... ,e t c . (4.2.2)

because x = C Nix;.
 Shape functions for the quadratic tetrahedron can be derived using Lagrangian interpola-
tion. These are

N1 = (2L1 - l)L1, etc. for corner nodes

and Ns = 4L1L2, etc. for midside nodes

Notice that these shape functions, Eqs. (4.2.2) and (4.2.3), satisfy element convergence criteria:
integrability (en-'continuity for n-th derivative) and completeness (no straining by a rigid
body mode, constant strain condition, and continuous displacement field).

With the isoparametric element, the same shape functions are used to describe the dis-
placement field as well as the geometry, i.e.,

x= C N ~ x ; ,etc. and u= C N ; U ~ etc.

,

The strains are calculated as

= [Bl{ul
where strain vector
{&lT = < EX ~y EZ Y X ~y y z Yrx >
element matrix
[ B ] = [BlBS . . . Bi . . .BlO]
nodal displacement vector

with

Since the shape functions are defined in terms of local coordinates, a coordinate transfor-
mation is required t o obtain global derivatives. Volume coordinates involve four coordinates
(one of which is dependent), and the Jacobian matrix will become rectangular. To avoid this
difficulty, let us introduce local coordinates (J, 7, <) as follows:

Derivatives with respect to the local coordinates can be expressed in terms of global deriva-
tives using the chain rule, i.e.,

where the Jacobian matrix is expressed in terms of shape functions as

This Jacobian matrix must be inverted t o obtain global derivatives, from which the element
matrix is formed. Notice that the determinant of the Jacobian matrix is called Jacobian which
represents a volume change, i.e.,
d x d y d z = det J d t d q d c (4.2.10)

4.2.2 Shape Functions

To make tetrahedron element compatible with other solid elements (HEXA and PENTA),
deletion of any or all of.the midside nodes is permitted [4.2]. The shape functions are modified
with Kroneker deltas (65 - blO) where
Si = 0 if the midside node i is deleted
= 1 if the midside node i is not deleted
The goal is t o construct functions which are unity at the associated node and zero at all other
nodes, regardless of any combination of deleted midside nodes, by the following scheme:
at the corner nodes

at the midside nodes

N5 = 4t7S5,etc.

Shape functions are identified as follows:

Derivatives of the shape functions with respect to the local coordinates are obtained as
follows:
 These shape functions and derivatives may be reduced to those for 4-noded and 10-noded
tetrahedron elements. It can be verified that, for any combination of deleted midside nodes,
there exist a condition

which satisfies the constant strain requirement.

4.2.3 An Example of Element Matrix

To illustrate the computational procedure, an element matrix for a linear tetrahedron (4-noded)
element is explicitly derived here. The shape functions and the derivatives are tabulated below:

Node arv,arv,m
Ni at aq

1 E 1 0 0
2 rl 0 1 0
3 c 0 0 1
4 1 - -1 -1 -1

]1:; ;: j
Then the Jacobian matrix may be found as

[J]= [ 1 0 0 - 1
0 1 0 -1
0 0 1 - 1
Xl

24
Yl

Y4
Zl
214
= [x2.
234
Y14
y24
Y34
I:
214

where (x;j = x; - xj) is used for convenience. Upon inverting the Jacobian matrix, we have

[ J ]= -
det J [ :; pi ): ]
a1 a2 a3

where
The global derivatives of the shape functions are
- - 7-
-
arv,arv,%aiv,
ax ax ax ax a2 a3 -(a1 + +
a2 a3)

1
arv,arv,arv,arJ,=-
ay ay ay ay det J bl b2 b3 -(bl + f
b2 b3)

arv,arv,arv,*
az az az az - el c2 c3 -(el + +
c2 ~ 3 -)

Hence the element matrix will be

Notice that the rank of [B] matrix is 6.

4.2.4 Volume Integration of Element Matrices

By virtue of variational principles, the element stiffness matrix is derived as follows:

where D is a (6x6) material tangential matrix.

While analytical integration is possible, there are some advantages in using numerical
integration. The Gaussian quadrature is used for the tetrahedron as usual. The minimum
number of intergration points required for non-singular stiffness matrix may be determined
based on
Total dof - rigid body dof
minimum n =
No. of strain components for each point
These numbers are found t o be 1 and 4 for linear and quadratic tetrahedron, respectively.
Using the integration formulas, shown in Fig. 4.2.2, the element stiffness matrices for linear
and quadratic tetrahedron may be computed as
1
[Ice]= - JdetJI B ~ D B for 4-noded TETRA
6

1
[ K e ]= - x
6
-.
1
- x
a-d

gauss
ldet J ( B ~ D B for 5-10 noded TETRA

where the scaling factor 116 is introduced t o compensate I JI = 6 x (Volume). One-point scheme
should be used for 4-noded tetrahedron (all the midside nodes deleted) and four-point scheme
otherwise.

It is difficult to determine lumped masses when some of the midside nodes are deleted.
However, the consistent mass matrix may be obtained using the same shape functions, i.e.,

where [ N ] = [NIT N a I . . . NloI] with I being (3x3) identity matrix. Again the Gaussian
quadrature 4-point formula is used to find

[MIe=p,
1
x
gauss
ldet Jl [ N ] ~ [ N ]

Notice that 4-point scheme is to be used even for 4-noded tetrahedron. For computational
convenience, the consistent mass may be converted to the lumped mass. One way t o achieve
this is t o take the diagonal terms and scale them so that the total mass is preserved.

4.2.5 Element Loads and Equilibrium

Nodal forces are computed from the element stresses using element matrix, i.e.,

{ F J e= /v B~{u)~v (4.2.16)

The Gaussian integration is performed as

{F)' = 61 ldet JI [BIT{u)

12 x 1 12x6
for 4-noded tetrahedron and

{ F J e= -
1
x
24 g a u s s
ldet JI [ B T ] { a )
 for (5-10)-noded tetrahedron.

Thermal load is like an initial strain. Therefore, nodal forces due to thermal load are
obtained by

where { E ~ = ) aAT
~ < 1110 0 0 >. With anisotropic thermal properties, { E ~ becomes
) ~
AT < crl, a2 . . .a6 > in general.

The pressure load applied t o any surface of the tetrahedron may be distributed to the grid
points using the shape functions applicable to the 6-noded triangular element with removable
midside nodes, i.e.,

where { n ) is a unit direction vector associated with a pressure (p) and

[Nt]= [nyrN;I . . . N;I]
with I being ( 3 x 3) identity matrix. Shape functions ( N i ) are shown in Fig. 4.2.3. Area
integration should be performed using Gaussian quadrature formulas shown in Fig. 4.2.4, i.e.,
a-c

V
gauss

During the nonlinear iteration process, the equilibrium is not reached until convergence is
achieved. The equilibrium is sought in the global level when the residual load R approaches
zero. The residual load vector is defined as
linear nonlinear
{ R ) = {P}- { l i e u e }- (4.2.21)
e

where {P) is the applied load vector including the thermal load and C implies assemblage in
the global coordinate system (coordinate transformation required).

4.2.6 Element Coordinates

The element coordinate system for the tetrahedron element is shown in Fig. 4.2.5. It is defined
with the initial element geometry such that
e The origin is at the first grid point in the connectivity, GI.
The x-axis is determined by connecting the origin G1 to node G2, i.e.,
 The y-axis is determined by orthonormalization (Gram-Schmidt process) of the edge
direction (FI3)with respect to x-axis,

a The z-axis is orthogonal to x and y according to the right-hand rule, i.e.?

Then, the transformation from element coordinates to the basic coordinates is simply

{:I basic = I T b e j { : ]

element
+{:;I basic

where

[Tbel = [ i x j x kz
i! .
2,
I: ky
Jz kz
] (4.2.25)

and < z e ,ye, ze>Fasicis the position vector of the element coordinate system with respect t o
the basic coordinate system. Notice that [The] is an orthogonal matrix, i.e.,

[Tbe]-l = [TbelT

4.2.7 Stress Data Recovery

In case of linear analysis, the stresses at the integration point are recovered as

with
AT = {N;}*{T~ - T,;)
where {a), [B], {u)and{a) are defined in the element coordinate system, and the shape function
{N;} interpolates Gauss point temperatures from the nodal temperatures. In case of nonlinear
analysis, stresses are computed again upon convergence, starting from the last converged st ate
(last converged solution of a and u), i.e.,

Given {sold) and

{A&}= { E ) -
with
{&I= [Bl{u) - { 4 A T
the nonlinear material routine computes updated stresses {o,,,}, which are stored in ESTNL.

To output grid point stresses, the stresses at the integration points must be extrapolated t o
the nodal points. A linear extrapolation can be applied using stresses at the element c.g. and
the corner Gauss point associated with a grid point. For a tetrahedron, referring to Fig. 4.2.2,

'Jo 1
01 =
'Ja -
CU - 114
(1 - 4) 'Jo +
where
1
00 = - ( ~ a
4
+ + +
' ~ b 'JC ' ~ d )

This yields
'J =
1- P
-0, -
P ( ' J b + 'Jc + ' J d )
-
1-4p 1-4p
In matrix form for all the grid points in the element

A -B -B -B

where
(i] grid

1-
A=-
-

- P-
- Sym.

and B = -
P
A -B
A
-B
-B
A - (i]
gauss

1 - 4p 1 - 4p
This process must be operated on every component of stress vector. No extrapolation is
required for 1-point integration.
Figure 4.2.1. Volume Coordinates for Tetrahedron

4.2 - 11
 Linear Quadratic

1 point integration 4 point integration

Figure 4.2.2 Gaussian Quadrature for Tetrahedron

* Area coordinate system (t 7 C) such that

* Shape Functions:

Figure 4.2.3 Shape Functions for Pressure Load

4.2 - 13
 Linear Quadratic

1 point integration 3 point integration

Figure 4.2.4 Gaussian Quadrature for Pressure Load

 Origin at G1

x-axis: in the direction of the edge from G1 to G2

y-axis: orthogonalized from edge (Gl-G3), K3

z-axis: orthogonal t o x and y according t o the right hand rule

Figure 4.2.5 Element Coordinate System Definition

4.3 GAP AND FRICTION ELEMENT

One of the major areas in nonlinear analysis that poses difficulties in convergence is the applica-
tion of GAP elements. This has been particularly true when friction is involved. Numerical dif-
ficulty is inherent in nature for the penalty method which is employed in the MSCJNASTRAN
GAP element. Furthermore, the GAP element simulates a point-to-point contact, and is often
*
used to simulate surface contact problems. Deficiencies of the penalty GAP element are coped
with by implementing an adaptive GAP element in Version 67 of MSCINASTRAN. The adapt-
ability is based on the subincremental scheme with the capability t o update the stiffness at
proper timing, t o bisect the increment when it is excessive, and t o adjust the penalty stiffnesses
to the proper values in concert with the nonlinear and dynamic environment. In addition, the
kinetic friction effect (different from the static friction) is introduced to simulate the physics
more accurately by distinguishing stick-slip forces.

4.3.1 Introduction

Contact problems are common in engineering practice. The GAP element simulates a unidirec-
tional point-to-point contact using the Bulk Data CGAP and PGAP. The GAP element is one
of the major areas that pose difficulties to the convergence of the iterative process, particularly
when the friction is involved. The penalty method is used t o simulate the rigidity between two
degrees of freedom by adding a large value t o the stiffness so that the two have approximately
the same displacements. Penalty values are introduced t o avoid penetration and t o enforce the
sticking condition (static friction) between two contact points. Difficulties (or inaccuracies)
arise when the penalty values are not properly chosen. The success of this method is highly
dependent on the user's choice of penalty values and requires a compromise between accuracy
and numerical performance.

The Lagrange multiplier method has been used by researchers [4.3-4.51 in contact problems.
The method enforces the constraints exactly. However, it also poses some difficulties. The main
drawbacks of the Lagrange multiplier method are: (1) the problem size increases by Lagrange
multiplier variables and (2) the stiffness matrix has zero diagonal terms, and requires pivoting
and resequencing. Some variations of the Lagrange multiplier method have been proposed
[4.6-4.81 t o mitigate the deficiencies. Another school of thought for the dilemma is t o alleviate
the difficulties by implementing the adaptive penalty method t o choose a proper penalty value
and adjust it to control the numerical stability and the accuracy [4.9-4.101. Although the
constraint is not enforced exactly, the effectiveness of the penalty method lies in preserving
the size and the bandwidth of the stiffness matrix. MSCINASTRAN employs this approach
in the adaptive GAP element implemented in Version 67.
4.3.2 Basic Considerations for Penalty GAP Element

The GAP element changes its status when the load is applied. Consider the internal forces
of a GAP element with an isotropic friction. Force components in the GAP element are the
axial compressive force (F,) and the friction forces in lateral directions (F, and F,) in terms
of displacements (u, v, w) in the element coordinate system. These internal forces can be
computed based on the GAP status as follows:
When the GAP is open (no contact, no lateral stiffness),

F, = ICbu 5 0. and Fy = F, = 0. (4.3.1)

where Kb is an arbitrary open stiffness.

When the GAP is closed and sliding (no friction),

F, = ICau > 0. and Fy= F, = 0 (4.3.2)

where Iiais the penalty value for the closed stiffness.

When the GAP is closed and sticking (static friction),

where Fx = K,u, F, = Ktv, and F, = Ktw. It is noted that the sticking condition is
dictated by the static coefficient of friction (p,) and the transverse shear stiffness ( K t ) .

When the GAP is closed and slipping (kinetic friction),

where - -

with the friction dictated by the kinetic coefficient of friction pk.

The element stiffness matrix can be formed as

with - =-
I{ a{F>
a@>
where the stiffness components are derived for each GAP status as follows:

When the GAP is open,

Kb 0.

sym. 0.
 When the GAP is closed and sliding,

-
I{ =
["a R]
sym. 0.

When the GAP is closed and sticking,

Ir', 0.

sym.

When the GAP is closed and slipping,

It is noted that the slipping due to friction introduces an unsymmetric stiffness matrix.

There are some inherent difficulties involved in the penalty GAP element. Since the New-
ton's method is sensitive t o abrupt changes in stiffness during iterations, it could cause a
divergent or oscillatory solution when the GAP changes its status. The timing of the stiffness
update is crucial to the convergence, which necessitates an adaptive stiffness update criterion
based on the GAP status. Furthermore, it the GAP status undergoes two or more consecutive
changes, adaptive stiffness update alone cannot stabilize the solution and the increment size
should be reduced.

The stick-slip behavior of the frictional GAP element is path-dependent in a manner similar
t o the plasticity. T h e lateral forces due to friction should be considered for equilibrium. If the
internal forces in the GAP elements are computed in a single step, the accuracy deteriorates
when the increment produces large changes in displacements with friction. In order t o trace
the path-dependent solution, the subincremental scheme within an incremental load step is
desirable [4.11].

Optimal penalty values ( K , and I<,) are difficult to assess. Furthermore, they have t o
be updated continuously during the deformation process, because the stiffness of the adjacent
structure changes during the incremental process. It becomes necessary t o adjust penalty
values adaptively by the program throughout the analysis.
4.3.3 User Interface for G A P Element

Connectivity is specified in the Bulk Data entries CGAP as shown below:

1 2 3 4 5 6 7 8 9 10
-
CGAP EID PID GA GB X1,GO X2 X3 CID

CG AP 17 2 110 112 3

Line GA-GB defines the GAP element x-axis as shown in Fig. 4.3.1. If GA and GB are
coincident in space, CID is used t o define the element coordinate system. The orientation of
the y and z axes is determined by GO (or XI, X2, X3) defining the x-y plane. The element coor-
dinate system does not rotate as a result of deflections. In other words, the element coordinate
system is not updated even when the geometric nonlinearity is considered (LGDISP=l).

I Go (orientation vector)

Figure 4.3.1 The G a p Element

The axial and lateral force-deflection curves of the GAP element are shown in Figs. 4.3.2
and 4.3.3. Properties that characterize these curves are specified in the Bulk Data PGAP.
PGAP PID uo Fo Ka Kb Kt PI 112

PGAP 2 0. 0. 136 l.E-2 1.E5 0. 0. +PG

TMAX MAR TRMIN +PG

+PG 0. loo. 0.001

F, (compression)
t

tension I - -I + (compression)

1 UA - UB
Figure 4.3.2 Gap Axial Force vs. Deflection
( slope K , is used when UA - UB 2 Uo )

Nonlinear Shear

Figure 4.3.3 Gap Lateral Force vs. Deflection

 The initial GAP opening is defined by Uo (defaulted t o 0)) not by the separation dis-
tance. The preload is defined by Fo (defaulted t o 0). Notice that positive values of force and
displacement represent compression in the GAP element coordinates. If the GAP is closed
(UA - UB 2 Uo), the axial stiffness ( K , ) has a large value relative t o the adjacent structure.
When the GAP is open, there is a small open stiffness ( K b ,which is defaulted t o 10-*I<,) in
the axial direction. For most contact problems, Ii, should be chosen t o be three orders of mag-
nitude higher than the stiffness of the neighboring grid points. The factor may be reduced t o
facilitate convergence a t the expense of the accuracy. A much larger K , value may slow down
convergence or cause divergence, while a much smaller K , value may result in an inaccuracy.

The transverse shear stiffness Iit (defaulted t o p l K a ) is used when the friction is active
upon contact. When the GAP is open, there is no transverse stiffness. When the GAP is
closed and if there is friction, the GAP has the elastic stiffness ( K t ) in the transverse direction
until the lateral force exceeds the friction force and the slip starts t o occur. If K t = 0, the
friction effect is ignored (equivalent t o p1 = p2 = 0). The recommended range of the transverse
<
stiffness value is O.OIIia 5 lit I<,.

When the adaptive GAP (default option) is used, p1 and p;! represent static and kinetic
coefficients of friction p, and pk, respectively. If the nonadaptive GAP element is chosen, the
anisotropic coefficients of friction ( p y and p,) are the input for p1 and pg. Directions y and
z are defined by the orientation vector in the CGAP entry, which corresponds t o the GAP
element coordinate system.

Penalty values (I<, and l i t ) are adjusted adaptively if TMAX > 0.0. There are two kinds
of GAP elements: adaptive GAP and nonadaptive GAP. If TMAX >0.0, the adaptive GAP
element is chosen. When TMAX = 0.0 (default option), penalty values will not be adjusted
but other adaptive features will be active (the gap induced stiffness update, gap induced
bisection, and subincremental process). The value of TMAX = -1.0 selects the nonadaptive
(old) GAP element. The recommended allowable penetration TMAX is about 10% of the
element thickness for plates and shells (or equivalent thickness for other elements) which are
connected t o the GAP element. If the structure is a massive solid, TMAX should be determined
such that the penetration is two or three orders less than the elastic deformation of the solid
model. The initial estimate of TMAX could be made such that the penetration is inconspicuous
relative t o the overall dimension of the structure, e.g., & of the overall length. The penalty
value adjustment will thrash if TMAX is too small, and no adjustment will occur if TMAX is
too large.

The maximum adjustment ratio MAR ( 1 < MAR < lo6, defaulted t o 100) is used only for
penalty value adjustment of the adaptive GAP element. Upper and lower bounds of the ad-
justed penalty value are MAR * I i Oand I i O / M A ~respectively,
, where IiOis the user-specified
value for li, or lit. The parameter TRMIN is a fractional number (0.0 5 TRMIN 5 1.0,
defaulted t o 0.001), representing a lower bound for the allowable penetration. The lower
bound for the allowable penetration is computed by TRMIN * TMAX. The penalty values are
decreased if the penetration is below the lower bound.
 GAP element output may be obtained by the STRESS ouput request in the Case Control
Section. An example of the GAP output format is shown below:
ELECTRO-MECHANICAL CLUTCH SEPTEMBER 5, 1990 MSCINASTRAN 8/30/90 PAGE 66
DISPLACEMENT IN MM., FORCE IN NEWTON

SUBCASE 1
LOAD S T E P = 2.00000E-01
S T R E S S E S ( F O R C E S ) IN G A P E L E M E N T S ( C G A P )

ELEMENT - F O R C E S IN E L E M SYST- - D I S P L A C E M E N T S IN E L E M SYST-

ID COMP-X SHEAR-Y SHEAR-Z AXIAL-U TOTAL-V TOTAL-W SLIP-V SLIP-W STATUS
31 1.2051E-06 0.0 0.0 1.2977E-02 -2.38913-01 0.0 -2.38913-01 0.0 SLIDE

The element forces and displacements are computed in the element coordinate system. Positive
values of the axial force (Fx) and'the axial displacement ( U ) represent compression. Lateral
forces are computed for the frictional case, and printed as SHEAR-Y and SHEAR-Z. Lateral
displacements in the y and z directions are printed under TOTAL-V, TOTAL-W (representing
total lateral displacements) and SLIP-V, SLIP-W (representing translation of the slip cen-
ter). When the GAP is open (regardless of friction) or sliding without friction, the total
displacements (TOTAL-V, TOTAL-W) and the slip center displacements (SLIP-V, SLIP-W)
have the same value. For the sticking or slipping cases with friction, the magnitude of the
slip center translation (SLIP-V, SLIP-W) should be smaller than the magnitude of the total
displacement. The last column of the output shows the GAP status (available in V67): OPEN,
SLIDE, STICK or SLIP.

4.3.4 Theoretical Basis for Frictional Behavior

According to the Coulomb friction law, the reaction force due t o friction is proportional to the
normal force (FN). Slipping does not occur as long as the external force exerted in the lateral
direction (FT) is smaller than the friction force, i.e.,

where p, is the static coefficient of friction. Slipping starts when the external lateral force
exceeds the static friction force. The friction force is reduced t o the kinetic friction force
( p k F N ) during the slip, since pk < p, in general. In order t o simulate the stick-slip behavior
of the current GAP element, the frictional yield function is formed as follows:

with
where Kt is a penalty value (representing an elastic stiffness) during sticking, and (v,, w,)
represents a translation of the slip center in the y-z plane. The GAP is sticking, if f < 0.

When the GAP starts to slip (f 2 O), the lateral displacement can be decomposed into
two parts: recoverable sticking part (denoted by superscript st) and irrecoverable slipping part
(denoted by superscript s l ) ; i.e.,

Slipping is similar to the plastic deformation [4.11]. Adopting the associated flow rule, the slip
displacements can be defined in terms of the effective slip increment (AyS') by

where

and

Notice that slipping occurs along the direction normal to the slip locus, where (Fy,F,) represent
the forces in the normal direction t o the slip locus.

The slip yield surface, which can be represented by a circular cone in the element force
space, is kept up-to-date by updating slip forces and the slip center. Substituting Eqs. (4.3.11)
and (4.3.12) into (4.3.10), we obtain

where

Substituting Eq. (4.3.13) into Eq. (4.3.9) with f = 0, we can express nyS'as below:

Combining Eq. (4.3.13) and (4.3.14)) the final form of Eq. (4.3.13) becomes

The slip center must be updated accordingly using Eq. (4.3.12)) (4.3.14) and (4.3.15):
4.3.5 Subincremental Algorithm for Stiffness Update and Bisection

In the absence of friction, the GAP element may change its status from open to sliding and
vice versa. When the friction is present, the status of the GAP element may change from open
t o stick and further to slip, and vice versa. General principles adopted in the adaptive stiffness
matrix update and bisection strategies induced by GAP elements are:

Allow only one state change in any GAP element within each increment. Bisect the load
or time step if the GAP undergoes two state changes (e.g., open t o slip) in a single step.
Update the stiffness matrix when the GAP changes its status (e.g., open to stick, or slip
to stick).
Subdivide the increment for the slip process [4.11].

In order to achieve the adaptability, it is necessary t o use the subincremental approach within
a global increment. Furthermore, slipping with friction is a path-dependent process and the
subincremental scheme allows a more accurate solution t o the path-dependent process.

Two major steps are required to implement the subincrement method. Let the displacement
increment for the i-th iteration be defined as follows:

where n represents the previously converged step. The first step is t o find the open/close
boundary. Assume that the GAP closes from the opening state at the i-th iteration. The
scale factor a (0 5 cr 5 1) is used t o determine the point at which the GAP starts t o close as
follows: ,
210 -u=
<
if U, 5 uo uZ (open + closed)
0 if Un L UO (closed + closed)

where uo represents the initial opening of the GAP element.

The second step is t o find the stick/slip boundary. Assume that the GAP changes status
from stick to slip a t the i-th iteration. The scale factor ,B (0 5 ,B 5 1)is used t o determine the
point where the GAP begins to slip. The frictional yield function should be satisfied (f = 0)
at the stick/slip boundary, i.e.,

+ ) J(F;
p s ( ~ z P K ~ A U= + P K ~ A V+) ~(F: + ~ l i ~ ~ w ) (4.3.18)
~

which can be rearranged to solve for P, resulting in a quadratic function of P:

ap2 +2b/?+ c = 0 (4.3.19)
where Fz, F; and FF represent GAP internal forces at the last converged solution. It can be
shown that the quadratic equation is guaranteed to produce a real and positive root. Since
the GAP is slipping at the i-th iteration,

Expanding Eq. (4.3.20) using the relation in Eq. (4.3.19), the following inequalities can be
obtained:
a-c 2
a+2b+c<0 and b2-ac>(l) 2 0

These inequalities lead to a proper root

and for a - 0

Once the initial slip point is found, the slip region is subdivided into m subincrements
(1 < m 5 10). The number of subdivisions are determined adaptively by

The displacement increment for each subincrement becomes

The GAP normal force (F,) is gradually increased, and the the slip force ( F , and F,) is
updated in m subincrements. After the subincremental process, the element stiffness and the
slip center are updated. The subincremental process is schematically described in Fig. 4.3.4.

It is noted that the "slip-to-slip" case is catagorized into two groups. One is the slip in the
same direction and the other is the slip in the opposite direction. The condition for the slip in
the same direction can be determined by the dot product of the last converged friction force
and the friction force increment, i.e.,
Slipping in the same direction is also signified by a double root of /? at zero. If the slip occurs
in the opposite direction, /? has two roots with one being zero. In this case, the root at zero is
discarded and the subincrement is processed from the slip locus a t the opposite side signified
by nonzero P.

for final position

Figure 4.3.4 Schema for Subincremental Process (m = 3)

4.3.6 Adaptive Adjustment of Penalty Values

Penalty values are adaptively adjusted upon convergence for the next increment. If the penalty
value is too small and produce an appreciable penetration, the accuracy of the solution may
not be acceptable. Then the penalty value (K,) should be increased to reduce the penetration.
On the other hand, numerical difficultiesarise if the penalty values are set too high. It is highly
desirable to reduce the penalty values in such cases to improve the numerical performance.

The adaptive penalty value adjustment is designed to adjust penalty values automatically,
if necessary, using the user-specified initial values and the maximum allowable penetration
(TMAX). If the GAP penetration exceeds the allowable value supplied by the user, penalty
values (Ii, and Iit)are increased, and vice versa if the penetration is smaller than the minimum
allowable value, a fraction of TMAX. The adjustments are made a t converged points. It is
necessary t o introduce a stepping function in order t o avoid thrashing [4.12]. The stepping
function is designed such that the adjustment ratio is in powers of 10, i.e.,

where f ( R ) is ..., 0.1, I., 10, ... and n denotes the incremental step with
ug
U -
R=- if u - uo > TMAX
TMAX

Uo U -
R = if u - uo < TMAX * TRMIN
TMAX * TRMIN
where TRMIN represents the user-specified parameter t o define a lower bound for allowable
penetration.

It is noted that two penalty values (Ii, and K t ) are adjusted by the same ratio, starting
from the user-specified initial values. Since the ratio of Ii, t o K t constitutes the slip criterion,
an incompatibility will occur in the constitutive relations between two consecutive solutions if
I<,and K t are adjusted independently. The proper ratio between two penalty values can be
established logically. Assume that the lateral displacement under the sticking condition is E
when the GAP starts t o slip. The slip condition requires that

If it is desired t o limit the lateral displacement under the sticking condition (6) by the same
magnitude as the normal penetration (u - uo), the magnitude of shear stiffness K t has t o be
related t o Ii, by
K t 2: p k K a . (4.3.23)

If the user inputs TMAX = 0, the penalty value is fixed and the adjustment algorithm is
not invoked. The penalty value adjustment is limited by an upper and lower bound defined by
the user-specified maximum adjustment ratio (MAR), i.e.,

IiO
< I( < MAR * K O
MAR -
where KO is the user-specified value for Ii, or Ii,. The global stiffness matrix is updated
whenever the penalty values are adjusted.
4.3.7 Non-adaptive GAP Element with Anisotropic Friction

The original GAP element can have an anisotropic friction by specifying direction-dependent
friction coefficients associated with the GAP orientation. When the GAP is in contact (closed),
the compressive force in the axial direction is defined by

where Fo is the preload and uo is the initial GAP opening. The slip yield function may be
modified for anisotropic friction as follows:

where p, and p, are coefficients of friction in the y and z direction. If the GAP is sticking,
the friction forces become

with the associated stiffness as follows:

In the case of slipping, the lateral forces are given by

The associated GAP stiffness is given by

where the symmetry is enforced.

4.3.8 Verification and Validation: Bouncing Mass Problem

A 4-lb weight is resting on a spring-supported platform which is initially compressed 4 inches

in the position shown in Fig. 4.3.5. The platform is released, and after moving 2 inches, the
platform is halted by a stopper. The response of the weight is t o be determined [4.13]. The
spring modulus of the platform support is 10 lb/in.

Sprl ng k-1OlblIn.

Figure 4.3.5 Physical Model

From the conservation of energy

AV, + AV, + A T = 0.
T h e change in energy from the initial state when the platform is stopped is
1
AV, = -k(x; - xi) = -60 lb-in
2
AV, = mgh = 8 lb-in

from which the velocity of the mass a t the time of departure from the platform can be obtained

Up t o this point, the motion of the mass is governed by the equation

With the initial conditions xo = -4 in. and xo = 0 in/sec, the response of the mass is
expressed as
The elapsed time at the moment the platform hits the stopper may be obtained by solving the
above equation for t with xl = 2 in, i.e.,

tl = -
Wn 0 ( + )Tip
xo
+ = 0.035'7 ser

where w, = 6 = 31.0644 radlsec. As the mars departs from the platform, the motion of
the mass is described by
h = xot - 1 2 (4.3.34)

relative t o the stopper. Elasped time when the mass reaches the peak is obtained as

t2 = tl +-
" = 0.0357 + 0.2595 = 0.2953 sec.
9
The height of the mass at this moment is

'X
h,, = 2 = 13.0 in
29
from the stopper. The theory predicts that one full cycle will take 0.5905 sec.

I
GAP (22) Load t o s l m l a t e
I f n l t l a l preload
Plrtfotn

COWRR) (20)

S P ng~
Gnvlty

0 ~ r P o~i n t sd

Figure 4.3.6 Symbolic MSCINASTRAN Model

 The MSCINASTRAN model is depicted schematically in Fig. 4.3.6. The spring is modeled
with a CONROD element, for which the stiffness is given as k = 7, with A (area) = 1.0 in2,
E (Young's modulus) = 100 psi, and 1 (length) = 10 in. The mass is modeled as CONM2 21,
which is connected t o the platform (GRID 201) by G A P 22. The stopper is represented by
a fixed point (GRID 203), t o which the platform is connected by GAP 23. The base (GRID
200) is fixed t o the ground. An initial displacement (4 inches) is simmulated by a step force
(40 lbs) applied a t GRID 201. The gravity field is applied by a GRAV entry. Static loads are
converted to the dynamic loads using Bulk Data LSEQ and TLOAD2 with a Case Control
command LOADSET. A small mass CONM2 24 (0.1 % of the mass block) is provided at the
platform t o suppress extraneous jittering. Small viscous damping (0.5 % a t 31 radlsec.) is
introduced and adaptive time stepping is used. T h e input d a t a listing is given in Table 4.3.1.

The solution t o the MSCINASTRAN model is obtained by the nonlinear transient analysis
method (SOL 99). T h e analysis started with initial At = 0.0025 sec. for the duration of 1 sec.
G A P elements had the initial stiffness of l o 4 lb/in for the closed position with TMAX of 0.005
in. for adaptive adjustment. At the onset, the penalty value (K,) of GAP 22 was adjusted to
lo5. T h e bisection was also activated by the adaptive time stepping during the initial phase.
After the initial settlements, the automatic time stepping algorithm quadrupled the time step
size. The stiffness matrix was updated three times due t o G A P status changes in the vicinity
of t = 0.035 sec., t = 0.545 sec. and t = 0.614 sec. T h e automatic time stepping algorithm
caused the reduction of At t o a half of the specified value (0.00125 sec.) while the platform
was in motion around t = 0.58 sec.

t i m e in sec.
Figure 4.3.7 Displacement Response
 The displacement response in Fig. 4.3.7 shows a peak height of approximately 14.56 inches
and the period of approximately 0.583 sec. Discrepancies between the MSC/NASTRAN solu-
tion and the analytical solution may be attributed to the following:
a Some momentum is dissipated due to the dimping and the small mass introduced at the
platform.
a The artificial stiffness introduced to the GAP elements changes the dynamic character-
istics slightly.
a Due t o the finite interval of the time increment, the time stepping scheme can not be
synchronized exactly with the GAP closing/opening.

In order t o verify the adaptive penalty value adjustment algorithm, a parametric study is
conducted with the penalty value (closed stiffness) of GAP 22. T h e value was varied from lo3
t o lo7 with TMAX = 0.005. The effectiveness of adaptive penalty value adjustment is shown
in Fig. 4.3.8. Notice that penalty values were accepted in a wide range ( l o 4 t o lo6) by the
adaptive adjustment scheme. A tighter tolerance (TMAX = 0.001) was also tested t o examine
its effect. Indeed, the penalty value was adjusted to the narrower range ( l o 5 t o lo6).

rum
l ' ' ' ' l ~ r ~ . ~ ~ ~ . ~ , . v 9 # f
-0 0 . '; 0.4 0.6 0 .E 1
Time [sec

Figure 4.3.8 Adaptive Penalty Value Adjustment o n G A P 22

4.3 - 17
 Table 4.3.1 Input Data Listing for Bouncing Mass Problem
ID VGAPM ,V66 $ SHL 12-11-90
TIME 5 $ CPU MINUTES FOR VAX
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
SOL 99
CEND
TITLE = GAP VERIFICATION
SUBTITLE = NONLINEAR TRANSIENT ANALYSIS
SEALL = ALL
ECHO = UNSORT
SET 1 = 201,202
SET 3 = 22,23
SET 4 = 20 $ CONROD
DISP = I
STRESS = 3 $ FOR GAP FORCES
ELFORCE = 4
LOADSET = 20
SUBCASE I
DLOAD = 210
TSTEPNL = 21
SUBCASE 2
DLOAD = 210
TSTEPNL = 21
OUTPUT(XYPLOT)
CSCALE = 1.3
PLOTTER = NAST
XTITLE = TIME IN SEC
YTITLE = DISP IN INCH
XYPLOT DISP RESP/201(TI) ,202(TI)
BEGIN BULK
$ SOLUTION STRATEGY
TSTEPNL 21 200 .0025 I ADAPT
1
+TSI
+TS2 0
$ PROPERTIES
PARAM W4 31.
MAT1 I 100. .3
PGAP 3 1.E+4
+PG3 5.-3
PGAP 4 2. I.E+4
+PG4 5.-3
$ LOADING
FORCE 201 201 40. I. 0. 0.
GRAV 200 386. -1. 0. 0.
LOAD 202 I. 1. 200 I. 201
LSEQ 20 301 202
TLOADI 210 301 0 220
TABLED1 220
+TAB1 0. 0. .001 1. 100. 1. ENDT
$ BOUNDARY CONDITIONS
GRDSET
$ GEOMETRY
GRID 200 -6.
GRID 201 4.
GRID 202 4.1
GRID 203 6.
$ CONNECTIVITY
CONROD 20 200 201
CONM2 24 20 1
CONM2 21 202
CGAP 22 3 201
CGAP 23 4 201
ENDDATA '

4.3.9 Verification and Validation: Vibration with Coulomb Damping

A mass block (m) connected to a spring (k) is resting on a frictional surface as shown in
Fig. 4.3.9. The motion of the mass block is governed by

mu+ku=-F for u > 0. (4.3.35)

mii+ku= F for u < 0.

where F is the friction force 14.141.

Figure 4.3.9 Physical Model for Vibration with Coulomb Damping [4.14]

Suppose the block is displaced by uo >> Flk and released. The solution while u < 0 is
where w = m. The mass block comes t o a stop ( u = 0) at w t = n. For T 5 wt < 27r while
u > 0. the solution is

It is noted that the acceleration ( u )is discontinuous at t = 5 , %, %, . . while the velocity is

a,

continuous. From the observation that u = -(uo - 2 F I k ) at t = 2 , and u = -(uo - 4 F l k ) a t

t = %, the decay is a linear decrement at the rate of 4 F l k per cycle until I u,, 151 F 1.
The MSCINASTRAN model uses a GAP element t o simulate friction with k, = lo6,
kt = l o 5 , p, = 0.3 and pk = 0.2. Let m (CONM2) be 100 kg in a gravity field of lm/sec2,
which gives rise t o a normal force (FN) of 100 N. A rod element is used t o simulate an elastic
spring with k = lo3 N/m. The input data listing is given in Table 4.3.2. The analysis is
performed in three subcases using SOL 99. The first two subcases use PARAM, TSTATIC,
1 for static analyses, which apply the preloads (gravity and a horizontal force of P=220 N),
resulting in an initial displacement

The third subcase uses PARAM, TSTATIC, 0 for dynamic response upon release of the load
P. The natural frequency (w) of the system is 3.1623 rad/sec and the period is 1.987 sec. After
the first cycle, the amplitude decreases t o

After two and a half cycles, the mass block is stagnated. Ideally, the mass block should come
t o a stop with a residual displacement due t o friction. However, because of the shear stiffness
( k t ) in the GAP element, the mass block oscillates a t w, = , / v m = 31.623 radlsec. This
effect is shown as ripples after two and a half cycles in Fig. 4.3.10.
 Figure 4.3.10 Horizontal Displacement Response

In order t o examine the effects of the frictional force, the analysis is performed with some
variations in the model. First, the mass (m) is increased to 366.7 kg and, in turn, the normal
force ( F N )is increased to 366.7 N. Then the initial displacement (uO)is decreased to 0.1467
and the natural frequency (a)is reduced to 1.6514 rad/sec. In this case, the decay rate is such
that the amplitude is reduced to zero in a half cycle (wt = T ) as shown in Fig. 4.3.11. This
is analogous to the critically-damped case in viscous damping. If the friction coefficients are
also increased ( p , = 0.45 and pk = 0.3) and the horizontal applied force (P) is increased to
300 N, the initial displacement (uo) becomes 0.19 m. The mass will come to rest when u = 0,
which occurs at t = $ = 1.9 sec. The residual displacement should have an asymptotic value
of 0.03 from Eq. (4.3.36). As shown in Fig. 4.3.11, this is analogous to the over-damped case
in viscous damping.

The effect of distinctive friction coefficients (static and kinetic) during the static loading is
shown in Fig. 4.3.12. For the over-damped case, the GAP will be sticking up to the friction
force of 165 N, governed by p,. As the block starts to slip, the friction force is reduced to 110
N, complying with pk. It is noted that an excessive number of load steps was taken during the
static loading in order to show the reduction in the frictional force at the stick/slip transition.
Despite the excessive steps, the peak at 165 N is not shown due to the discrete load increment
size.
 Table 4.3.2 Input Data Listing for Coulomb Damping Problem
ID COULOMB, V67 $ SHL 12-21-90
TIME 5 $ CPU MINUTES FOR VAX
SOL 99 $ NONLINEAR TRANSIENT ANALYSIS
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
CEND
TITLE = COULOMB FRICTION VIBRATION PROBLEM
SUBTITLE = NONLINEAR STATIC AND TRANSIENT ANALYSIS
LABEL = TIMOSHENKO, YOUNG, AND WEAVER; P. 198
ECHO = BOTH
SEALL = ALL
SET I = 2, 12, 22
SET 2 = 10, 110, 210
SET 3 = 20, 120, 220
DISP = 1
VELO = I
ACCE = I
STRES = 2
FORCE = 3
LOADSET = 1000
SUBCASE 10 $ COMPRESS GAP
PARAM TSTATIC 1
DLOAD = 100
TSTEPNL = 10
SUBCASE 20 $ SHEAR GAP
PARAM TSTATIC I
DLOAD = 200
TSTEPNL = 20
SUBCASE 30 $ DYNAMIC ANALYSIS
PARAM TSTATIC 0
DLOAD = I00
TSTEPNL = 30
OUTPUT(XYPLOT)
CSCALE = 1.5
XTITLE = TIME (SEC)
XGRID LINES = YES
YGRID LINES = YES
XMIN=2.
XMAX=8.
YTITLE = DISPLACEMENT AT THE MASS BLOCK
XYPLOT DISP RESP /2(TI), 12(Ti) ,22(TI)
YTITLE = VELOCITY AT THE MASS BLOCK
XYPLOT VELO RESP /2(TI) ,12(Ti)
YTITLE = ACCELERATION AT THE MASS BLOCK
XYPLOT ACCE RESP /2(Ti)
YTITLE = STRESS IN ROD
XYPLOT STRESS RESP 30/20(2), 120(2) ,220(2)
YTITLE = SHEAR FORCE IN GAP
XYPLOT STRESS RESP 30/10(3)
 YTITLE = TOTAL-V DISP IN GAP
XYPLOT STRESS RESP 30/10(6)
YTITLE = SLIP-V DISP IN GAP
XYPLOT STRESS RESP 30/10(8), IlO(8) ,210(8)
BEGIN BULK
$ SOLUTION CONTROL
TSTEPNL 10 4 .25 I ADAPT 20
TSTEPNL 20 25 .04 I ADAPT 20
TSTEPNL 30 30 .2 I ADAPT
$------------------------------------------------------------------------- $
$ GEOMETRY AND CONNECTIVITY FOR OSCILLATORY SYSTEM
GRID I 0. 0. 0. 123456
GRID 2 100. 0. 0. 3456
GRID 3 100. -1. 0. 123456
CGAP 10 10 2 3 1.
CONROD 20 I 2 20 1.
CONM2 30 2 100.
$ PROPERTIES
PGAP 10 0. 0. 1.+6 1.+5 .3 .2
MAT1 20 1.+5
$------------------------------------------------------------------------- $
$ GEOMETRY AND CONNECTIVITY FOR CRITICALLY DAMPED CASE
GRID I1 0. 0. 0. 123456
GRID 12 100. 0. 0. 3456
GRID 13 100. -1. 0. 123456
CGAP 110 110 12 13 I.
CONROD 120 11 12 120 I.
CONM2 130 12 366.7
$ PROPERTIES
PGAP 110 0. 0. I.+6 1.+5 .3 .2
MAT1 120 1.+5
$------------------------------------------------------------------------- $
$ GEOMETRY AND CONNECTIVITY FOR OVER-DAMPED CASE
GRID 21 0. 0. 0. 123456
GRID 22 100. 0. 0. 3456
GRID 23 100. -1. 0. 123456
CGAP 210 210 22 23 I.
CONROD 220 21 22 220 I.
CONM2 230 22 366.7
$ PROPERTIES
PGAP 210 0. 0. 1.+6 1.+5 .45 .3
MAT1 220 1.+5
$------------------------------------------------------------------------- $
$ LOADING
LSEQ 1000 110 111
GRAV Ill I. 0. -1. 0.
TLOAD2 100 110 0 0. 100.
$
LSEQ 1000 210 211
FORCE 211 2 220. I. 0. 0.
FORCE 211 12 220. 1. 0. 0.
FORCE 211 22 300. 1. 0. 0.
DLOAD 200 1. I. 100 I. 2000
TLOADI 2000 210 0 20 1
TABLED1 201
tTBD2I 0. 0. 1. 0. 2. I. ENDT
ENDDATA

4.3.10 Hertzian Problem: Contact Between Sphere and Rigid Plane

Mechanics of the frictionless contact between two elastic bodies were first solved by H. Hertz
and published in 1881. Hertz presented solutions to a family of two-body contact problems
with general curved surfaces for the loading normal to the surface. Stresses at the contact
area depend on the deformation of the bodies in contact. Typical examples are ball and
roller bearings, gears, cams, and rolling wheels. Hertz's analysis revealed that the pressure
distribution on the contact surface between two curved bodies is represented by a semiellipsoid
constructed on the surface of contact.
z

1
Figure 4.3.13 Contact Between Elastic Sphere and Rigid Plane
Let us confine our discussion t o the contact of an elastic sphere with a rigid plane as shown
in Fig. 4.3.13. The hemispherical pressure distribution on the surface of contact with a radius
a gives the maximum pressure (qo) at the center by
where P is the total compressive force. The solution for two balls in contact with radii R1 and
R2 is given [4.15] with the radius of contact area as follows:

and for the rigid body motion (or approach distance)

with
1 - v; 1 - v;
k1 = --- and k2 = ---
xE1 7rE2
for two balls (1 and 2) with different Young's moduli and Poisson's ratios, respectively. For
the contact of an elastic sphere with a rigid flat plane, Eqs. (4.3.39) and (4.3.40) are reduced
with R2 = cc and E2 = cc to

and

where cr is identical to the displacement of points on the sphere remote from the deformed
area.

Stresses at the center of the contact surface are given by

1
aT= = -(1
2
+ 2v)o, and = -40

The maximum shear stress occurs on the z-axis at a distance of 0 . 4 7 ~below the surface of
contact, given as
TTz = ~0~ = 0.31q0

The maximum tensile stress occurs at the circular contact boundary, given as

At this boundary, the circumferential stress (as) is compressive with the same magnitude as
a,, which represents a pure shear stress state.

For the static analysis, a symmetric half of a 10 degree sector of the sphere is modeled by
HEXA and PENTA elements with axisymmetric boundary conditions as shown in Fig. 4.3.14.
Contact with a rigid plane is simulated by adaptive GAP elements. Uniform vertical displace-
ment of the symmetric plane is maintained by tying the vertical displacements of all the nodes
on the top plane (loading plane) by MPC relations to the one at the centerline (z-axis),on
which 1/36 of the total compressive force is applied.

R = 8.0 in.
E = 1000 psi.
v = 0.3

Figure 4.3.14 MSC/NASTRAN Model for Static Analysis

The static analysis was performed in 20 increments up to the maximum loading (P = 2160).
The analysis ran t o completion in 27 iterations with four GAP induced stiffness updates.
The deformed shape at the maximum loading is shown in Fig. 4.3.15 in comparison with the
undeformed model. Two additional cases were analyzed: one with geometric nonlinearity
(PARAM,LGDISP) and the other with friction (p, = 0.3 and pk = 0.2). Deformed shapes for
other cases were not discernible from the one in Fig. 4.3.15. All cases ran to completion without
requiring any user intervention. In the frictional case, however, bisections were activated three
times during the initial phase, and 30 incremental steps were required.
The radius of contact surface and the approach distance (vertical displacement at the
loading plane) are plotted as functions of jhe total force (P) in Fig. 4.3.16, which shows the
MSC/NASTRAN solution in comparison with the Hertz solution. Notice that the effects of
friction and geometric nonlinearity are gradually manifested as the load increases. It is noted
that the frictional case appears more accurate at the initial stage. This is because the frictional
case took much smaller increments due to bisections during the initial phase, and the point of
contact could be more closely predicted.
 Figure 4.3.15 Undeformed Shape vs. Deformed Shape at P = 2160

CU
-

TCI-

~ n ' - c o n t a c t r o d jus (Hprt7)

..-..-.approach dlstance ( H e r t z '
0 MSC/NRSTRQN so l u t i o n
X MSC/NQSTRQN ( w / f r 1c t i o n '
+ MSC/NQSTRQN Iw/LG@ISP)
-...............il
.,m...."' ..................81.-"*."'
C
.............
......
m 1 ' ' ' 1 ' ' ' ' ~ * I 1 t ~I I ~
0 588 1000 1500 2008 2508
Load [ l b u

Figure 4.3.16 Contact Radius & Approach vs. Total Force in Static Analysis

4.3 - 28
4.3.11 Dynamic Case: Impact of a Sphere with Rigid Plane

The Hertzian problem shown in Fig. 4.3.13 is considered for dynamic response. In the impact
of an elastic sphere with the rigid plane, the duration of contact is very long compared t o the
period of lowest mode of vibration in the sphere. Vibrations and the stress wave can therefore
be neglected, and the statical solutions in Eqs. (4.3.39a) and (4.3.40a) are assumed to be valid
during impact [4.15]. The applied force P can be obtained from

where m and v are the mass and the velocity of the sphere, respectively. Introducing a notation

the acceleration of approach becomes

from which

where vo is the velocity of the sphere at the beginning of impact.

The value of the maximum compressive force is obtained by

where a,,, at the maximum compression can be found by substituting di = 0 into Eq. (4.3.41),
resulting in

where p is the mass density of the sphere. The duration of impact is given by

For the impact analysis, a 10 degree sector of the sphere is modeled by HEXA and PENTA
elements with axisymmetric boundary conditions as shown in Fig. 4.3.17. Contact with a rigid
plane is simulated by adaptive GAP elements. The initial velocity of 10 injsec. is specified for
the sphere a t the beginning of the impact. The analysis started with the initial time step size
of 1 msec. and progressed to completion without any difficulty. The adaptive time stepping
algorithm adjusted At t o 2 and 4 msec. during the initial stage. The bisection was activated
during the maximum impact to adjust At to 0.25 msec. In the meantime the GAP induced
stiffness updates were activated (a total of 15 times) when GAPs were closing. After the peak
impact, the time step size fluctuated between 0.25 msec. and 4.0 msec. due to thrashing of
bisection and automatic adjustment, which seems t o have been caused by opening GAPs. This
thrashing has been remedied by introducing the numerical damping (PARAM,NDAMP).

R = 5.0 in.
E = 1000 psi.
v = 0.3
p = 0.01 lb-sec2/in4

Figure 4.3.17 MSCINASTRAN Model for Dynamic Analysis

According t o Eqs. (4.3.42) and (4.3.43), the maximum compression force (P,,,) is 1246.5
lbs, which occurs when the approach distance (a,,,) reaches 0.5251. MSC/NASTRAN anal-
ysis shows P,,, = 1440 and a,,, = 0.53 as shown in Figs. 4.3.18 and 4.3.19, respectively.
The maximum indentation occurs in the vicinity of t = 0.072 sec. with the contact radius of
approximately 1.71 while Eq. (4.3.39a) predicts 1.62. The duration of impact is 0.1544 sec.
from Eq. (4.3.44). MSC/NASTRAN analysis as shown in Fig. 4.3.18 indicates that the dura-
tion is around 0.143 sec. Small discrepancies are attributed to the discrete contact rather than
continuous contact due t o the discreteness in space and time. Fig. 4.3.20 shows snap shots of
the deformed configuration at time 0.051 sec., 0.072 sec. and 0.2 sec.
 NONLINEAR I R A N S I E ~ Il M P A C l Of A SPHERt ON A R I S I 0 P L A I P L I N E
V E L O C l l V I N I H E -2 1 s 1 0 .

Figure 4.3.18 Reaction Force vs. Time

 Curves in the order of
Near contact point
Center point on mid-plane
Point at the top

NONLINEAR I R A I S I E N I I M P A C l OF A SPHERE ON A R I G 1 8 F L A l P L A I E
V E L O C L I V IN INE -Z I S l a .

Figure 4.3.19 Vertical Displacement Response a t Select Points o n Center Line

 Clockwise from
Upperleft corner:
t = O.,0.051
t = 0.2,0.072

Figure 4.3.20 Snap Shots at t=O., 0.051, 0.072 and 0.2 sec.
4.3.12 Recommendations

The most crucial task in using the GAP element is t o determine proper penalty values, which
dictates the accuracy, efficiency and the effectiveness of the analysis. Penalty values should be
chosen as large as possible for solution accuracy but as small as possible for solution efficiency.
The recommended value for the penalty stiffness is 1000 times the stiffness of the adjacent
structure, which will produce an error of 0.1%. Even with the advanced knowledge of the
structural characteristics of the contact region, user specification of the penalty stiffness is
merely an educated guess. This guess has t o be a conservative one a t the expense of the
computational efficiency, because the proper value varies throughout the analysis.

The main goal of the self-adaptive method is t o relieve users from the chores and uncer-
tainties of guessing or trial-and-error searches. By achieving this goal, an optimum algorithm
design is automatically attained for efficiency as well as for solution accuracy. A self-adaptive
algorithm has been proven effective and efficient for nonlinar analysis in MSCJNASTRAN.
The primary virtue of the adaptive method, however, is not the computing efficiency but the
overall engineering efficiency. This is because the adaptability makes it possible t o obtain the
solution a t the very first trial.

The adaptive GAP element implemented in Version 67, in concert with adaptive solution
methods [4.12,4.16], makes it possible t o maintain an acceptable accuracy while maintaining the
solution efficiency without user intervention throughout the dynamic environment of nonlinear
computation. The applicability of the present algorithm is deemed viable t o simulation of
surface contacts with friction. A limitation exists, however, that one of the contact surfaces
should not rotate by a large angle because the GAP element orientation is not updated for
large rotations. It is noted that there is an advantage in maintaining the original orientation
of GAP elements when a contact with a fixed flat surface is simulated.
Chapter 5

GEOMETRIC NONLINEARITY

5.1 OVERVIEW AND USER INTERFACE

Geometric nonlinearities are manifested in problems involving large rotations and large defor-
mation. The characteristics are follower forces due to large rotations, geometric stiEening due
to initial stress effect (as s result of large rotations), and large strains due to large defarmation.
Geometric nonlinear effects should be significant if the deformed shape of the strccture
appears distinctive from the original geometry by a visual inspection. A more rigorous and
quantitative definition for the large displacements can be derived from the plate theory of
Kirchhoff and Love: the small deflection theory is valid for a maximum deflection of less :$an
20% of the plate thickness or 2% of the small span length [5.1]. However, this definition
seems to be a little co~servativefor numerical analysis, and there is no distinct limit for large
displacements because geometric nonlinear effects are related t o the boundary conditions as
well as the dimensions of the structure. If the load-deflection curve of the critical point can be
estimated, the loading point should be in the nonlinear portion of the curve.

Geometric nonlinear effects [5.2] in the structure involving large rotations, whether rigid
body rotations or deformation inc'.uced rotations, are self-evident. Stiffening of a membrare,
stiffness in a pendulum or snap-through of an arch belong to this category. The inotion of
a pendulum under gravity is caused by geometric (differential) stiffness. Follower forces are
manifested when the applied loads are displacement dependent, such as pressure load and
thermal load applied on the surface that rotates. Centrifugal force is another example of
follower forces. Large strain effects are pronounced in metal forming problems which could
have strains exceeding 100%. Finite strain formulation is required to treat the problems in this
category. MSCINASTRAN does not currently support the large strain capability. In inost
structural applications, however, moderately large strains (20 to 30 %) appear in local areas if
there is any large deformation. .MSC/NASTRAN can be used for that category of problems.
Other geometric nonlinear effects are treated by updated element coordinates, gimbal angles
(or rotation vector), and the differential stiffness [ K ~ ] .

The geometric nonlinearity is controlled by only one parameter (PARAM, LGDISP) with
the following values in SOL 66 and SOL 99:
a 0 for geometrically linear analysis

1 for geometrically nonlinear analysis

2 for geometrically nonlinear, but without follower force effect

With values of 1 or 2 for LGDISP, all the potentially nonlinear elements become actively
nonlinear elements unlike the material nonlinear model.

MSC/NASTRAN has a distinct approach t o the large rotation, for which the element
coordinates are continuously updated to the current configuration during the iteration. The
equilibrium is sought in the deformed position. Consider the internal force computation as
follows [5.3]:
F = pTo dV (5.1.1)

The element matrix (B)is defined from the strain definition as

in which [B]could be divided into two parts (linear and nonlinear), i.e.,

Upon differentiation of Eq. (5.1.1), we have

where {o) represents stresses with reference to the original coordinates. Substituting Eq.
(5.1.3) and
do = D d~ = D B du,

-
(5.1.5)
Eq. (5.1.4) becomes
+
d F = [ K L KR + K,] du
tangential matrix
with
KL=JVB,T D BLdv

and
in which K L represents the usual linear stiffness matrix, K R a stiffness due t o large rotation,
and K , a geometric stiffness dependent on the initial stress level.

Now it remains t o define the nonlinear part of the element matrix (BN). The definition of
finite strains based on the Lagrangian formulation (referred t o the initial configuration) is as
follows:

with other components obtained similarly. In matrix notation

where { E L ) is the usual infinitesimal strain vector and { E N } is the nonlinear strain vector
consisting of the second order terms, i.e.,

where -
a: 0 0
0 a: 0
[A] = 0 0 a:
6 x 9 a: a: o
0 a$ a:
a,T 0 a:
b

and

Introducing shape functions (N;) and nodal displacements { u ) (using an example of a 10-noded
tetrahedron), displacement derivatives are expressed by

and
where

and

Fkom the properties of matrices A and 0, it can be shown that

from which

The initial stress stiffness [I<o]can be derived as follows:

K , da = /V d ~ u; d V = dV

in which
where

with I being the (3x3) identity matrix. Finally the geometric stiffness is

It has been found that stiffness matrices caused by geometric nonlinearity (I<R and Ii,)
can be computed from the matrices [A], [GI, and [MI with the following observations:
[GI is dependent upon the initial geometry, hence stays constant unless the geometry is
updated. This matrix is used in forming [KR] and [I',].

[A] is used in forming [I<R].

[A] is dependent on the rotations and should be updated
continuously.

[MI is used in forming [I{,]. [MI is dependent on the stresses and should be updated
continuously.
The primary functions of nonlinear stiffness matrices can be interpreted as follows:
The matrix takes into account the effects of large rotations. The large displacement
effects, due t o rigid body translation and rotation, are treated effectively in the absence
of large strains by updating element coordinates in MSCINASTRAN.

Geometric stiffness matrix [Ii,]takes into account the effects of the initial stresses. This
effect becomes important with geometric stiffening, and is used for instability analysis.
The geometric stiffness matrix [Ii,] is equivalent t o the differential stiffness [Kd] in
MSC/NASTRAN.
5.2 UPDATED ELEMENT COORDINATES

When the large displacement effect is included in the nonlinear analysis, MSC/NASTRAN
employs a method of displaced element coordinate system. This method allows large rotations
by updating element coordinates to the deformed geometry, and the equilibrium is computed
in the deformed configuration.

5.2.1 Concept of Convective Coordinates

The concept is based on the fact that the rigid body motion does not contribute to the strain
energy and is eliminated from the internal force computation. Consider a rod which underwent
rigid body motion as well as deformation as shown below:

Figure 5.2.1 Net Deformation of a Rod

The net displacement ud is measured in the displaced element coordinate system by over-
laying the original element on top of the deformed element. The element force can simply be
computed by
{FdIe = [KIe{ud) (5.2.1)
where the superscript e denotes an elemental operation and the subscript d denotes the vectors
in the displaced element coordinate system. Then the element forces should be transformed
into the common coordinate system (namely global coordinate system denoted by a subscript
g) before assembly for global operations, i.e.,

where the summation sign implies an assembly operation, and Tbdand Tbgare transformation
matrices from displaced to basic and from global t o basic coordinate systems, respectively.
5.2.2 Updated Coordinates and Net Deformation

Referring to Fig. 5.2.2, a quadrilateral element is shown in its original and deformed positions
(denoted by subscripts e and d, respectively) with reference to the basic coordinate system
(denoted by a subscript b).

Figure 5.2.2 Element Coordinates vs. Displaced Coordinates

The element coordinate system is established by bisecting the diagonals of the quadrilateral.
Transformation from the element coordinate sytem t o basic coordinates is simply

basic element basic

where the position vector (Xiin Fig. 5.2.2) of the element coordinate system with respect to
the basic coordinate system is denoted by < ze, ye, ze >casi,,
and transformation matrix [The]
is composed of direction cosines of unit vectors of the element coordinate system with respect
to the basic coordinate system, i.e.,
As the element deforms or displaces, the element coordinate system moves and this is defined
as a displaced coordinate system. The displaced coordinate system is established in the same
manner as the element coordinate system. Again the transformation should be performed
similarly, i.e.,

{ ) basic =lTbdI{}
displaced
+{::} basic

where < sd,yd, zd is the position vector of the displaced element coordinate system with
respect t o the basic coordinate system ( ~in fFig. 5.2.2) and [Tbd] is formed simliarly to [T*,].

In order to isolate the deformation from the rigid body displacements, nodal displacements
are computed in the displaced element coordinate system by overlaying the original element
as shown in Fig. 5.2.3.

element
defor~ation

Figure 5.2.3 Computation of Net Deformation

The net displacements can be computed by substracting the original nodal coordinates in
the element coordinate system from the displaced nodal coordinates in the displaced element
coordinate system, i.e.,

{}{I}~{)~ (5.2.6)
in which the nodal coordinates in the element and displaced element coordinate systems can
be computed by the following transformations:

{ i Ie = [TbelT [{ i } { i! I b aSi ]
basic -

and

bnsic global basic

Substitution of Eqs. (5.2.7) and (5.2.8) into Eq. (5.2.6) results in

where { u g ) is a total displacement (translational components only) in the global coordinates.

In the absence of the large displacement effect, the net displacement u d in Eq. (5.2.9) is reduced
to:
{.:I = d s { u ; l

5.2.3 Provisions for Global Operation

It is noted that the net rotations (8,, By and 8, of each node associated with the shell and
beam elements) are computed by a gimbal angle approach (or rotation vector approach) be-
fore computing element forces. Subsequently, the element forces have t o be transformed to
the global coordinate system before assembly for equilibrium check. The internal forces are
computed using net displacements and rotations, ud,i.e.,

if the material is linear or

if nonlinear material is involved. Consequently, the tangent stiffness matrix is formed in

the global system by assembling the element stiffness matrices transformed into the global
coordinate system from the displaced coordinate system, i.e.,

The update process is performed at every iteration and the updated nodal displacements
ud are used whenever strains and stresses are computed. Effectively, the second order effect
due to large rigid body motion is eliminated. However, the displacement output shows the
total displacements in the global coordinates, i.e.,

where the superscript i denotes operations on each nodal point. The transformation matrix
[Tbd]is computed for each element after each iteration and stored in the ESTNL data block
for stiffness matrix update when required by the stiffness matrix update strategy. On the
other hand, the transformation [Tbg]is computed for each nodal point and it is not stored but
recomputed whenever it is needed. The nodal coordinates in the undeformed geometry, Xb,
are available from the data block BGPDT.
This approach can be interpreted as approximate updated Lagrangian method, since the
motion of the body follows Lagrangian description. Stresses are computed in the deformed
geometry just like Cauchy stress. However, this method of displaced coordinate system is
a unique and salient feature in MSCINASTRAN. The referential geometry in the updated
Lagrangian method is brought up-to-date at every incremental step upon convergence but
fixed during the iterative process, which is inherently different from the current method of
updating the coordinate system.
5.3 FOLLOWER FORCES

The term "follower force" usually refers to the applied loads that change direction and magni-
tude with structural displacements and rotations, e.g.,

where p is the magnitude of the pressure on the surface A , interpolated by a shape function
N , and iidA changes as a function of u. They generally occur with fluid pressures such as
the pressurized balloon, inflated tire or the lift load on the airplane wing. Other physical
applications involve kinematics such as the classical fire hose instability problem or inertia
loads on spinning bodies. In MSCJNASTRAN, the term applies to specific load inputs as
defined below.

5.3.1 Basic Definition

For geometrical nonlinear analysis, the MSCJNASTRAN static loads belong to one of two
catagories, namely:
Forces and moments defined by fixed vector inputs, which may be calculated once per
run and cannot change direction or magnitude.
Forces and moments defined by the location of one or more GRID points, which include
pressures, thermal loads, and centrifugal forces.
The first catagory includes simple forces, moments and enforced displacements. The second,
follower force catagory, includes the following Bulk Data inputs:
FORCEl,FORCE2,MOMENTl,MOMENT2:The direction changes with displacements
of the referenced GRID points. The magnitudes of these concentrated loads are constant.

PLOAD,PLOADB,PLOAD4: The pressure loads follow the surface of the solid elements
(HEXA, PENTA and TETRA) or shell elements (QUAD4 and TRIA3) defined by the
GRID points and will change in magnitude if the area changes.
RFORCE: Centrifugal loads change in magnitude and direction with motion of the masses
attached t o the GRID points. The effect may be destabilizing if large motions occur. It
is recommended that lumped masses be used with these loads.

TEMP, TEMPI, etc: Thermal loads are built into the element force calculations and will
automatically follow the elements.
Also note that upstream superelements are assumed t o be linear and therefore the upstream
loads will remain fixed in magnitude and direction. In addition, forces on omitted degrees bf
freedom (when ASET or OMIT data are present) should not be follower forces.
5.3.2 Implementation

The follower forces depend on the GRID displacements and therefore must be recalculated
for each nonlinear iteration and line search. The basic equation for residual error, defined by
Eq. (3.2.2), becomes:
{R:} = {pa(ui)} - { ~ a ( u ~ ) } (5.3.2)
where the applied load vector {Pa)is now a variable. Corrective Loads are computed based
on the updated geometry and added to the initially applied loads to account for the follower
forces, i.e.,
P(.) = P(O) f (4 +
where
f (u)= P ( u ) - P(0).
Note that thermal effects are included in the vector { F } .

In turn, the tangent matrix could be calculated from Eq. (3.2.3) using derivatives of the
loads, which is termed follower matrix [5.4, 5.51. However, the nonlinear solution process in
MSCINASTRAN ignores the stiffness effects of the changing loads and use the approximation:

The effect of the approximation is minor in most cases. However, it could become a major
concern in thin shell models with pressure loads causing large rotations, where the converged
solutions will be correct but the rate of convergence may be slow or cause divergence. Also
the buckling solutions or modal analysis on preloaded structure with pressure load may not be
correct due to the approximate tangent matrix if the effect of the follower matrix is significant.

The follower force effects in the analysis can be controlled by the parameter LGDISP. Three
options are available in PARAM LGDISP:
0 for no geometric nonlinearity
1 for full geometric nonlinearity (including follower forces)

2 for deactivating follower forces while maintaining other geometric effects (rotation of
element coordinates and differential stiffness).
The third option was made available from Version 66, primarily t o verify the effects of the
follower forces.

5.3.3 Dynamic Follower Forces

All of the follower forces listed above, which are available in static nonlinear analysis (SOL 66
and SOL 106), are also available in nonlinear transient analysis (SOL 99 and SOL 129) except
for the following restrictions:
 Centrifugal forces (RFORCE) are not recalculated in SOL 99 or SOL 129.

Thermal effects are not supported in the nonlinear transient analysis.

The applied load is a function of displacement (follower forces) as well as time, i.e.,

where pi(u) represents the follower forces, g(t) is a forcing function in terms of time, and
N ( u ) ~is the applied load via NOLINi options. The Load Set Identification number is de-
fined on LSEQ Bulk Data. The TLOADl and TLOAD2 data define the time dependency
and refer t o the DAREA IDS of the LSEQ data.

Dynamic loading with DAREA Bulk Data and initial conditions do not follow the struc-
ture and are treated as linear loads.

5.3.4 Verification: Elliptic Cylinder Subject to an Internal Pressure

A thin-walled cylinder having an elliptical shape [5.2] is subjected to an increasing internal

pressure, up to a maximum of 5 psi. Fig. 5.3.1 shows one quarter of the cross-section of the
elliptic cylinder consisting of a row of 12 QUAD4 elements. The material of the cylinder is
elastic-plastic obeying the von Mises yield criterion and the isotropic hardening rule, with the
plasticity modulus being 1%of the Young's modulus. The material properties are specified
by the MAT1 and MATS1 Bulk Data entries. All the translational degrees-of-freedom in the
longitudinal direction are constrained t o represent a typical segment of the long cylinder. The
pressure loading is applied both statically (SOL 66) and dynamically (SOL 99) by PLOAD4
Bulk Data entries and the displacement responses are observed. Geometric nonlinearity effects
are included in the analysis by specifying PARAM, LGDISP, 1. The solution strategy is
provided by default methods in the NLPARM Bulk Data entry in SOL 66 and by the TSTEPNL
Bulk Data entry in SOL 99. Table 5.3.1 and 5.3.2 list the input data for SOL 66 and SOL 99,
respectively. The effect of follower forces is a conspicuous feature in this problem.

In the static analysis, the pressure load of 5 psi is applied in 5 subcases (total of 51
increments) with gradually increasing incremental pressure. The bisection is activated seven
times while the geometric effects are predominent during loading up t o 0.015 psi (in the first
subcase). The loading history and the numerical performance of each subcase are tabulated
below.
 Numerical Performance During- Static Analysis
subcase load no. of no. of no. of no. of no. of
ID (psi) increments iterations I< updates bisections line searches
1 0.05 20 656 174 7 40 1
2 0.1 10 143 26 0 81
3 0.5 8 126 26 0 57
4 1.O 5 53 8 0 16
5 5.0 8 60 7 0 13

The dynamic analysis is performed in five subcases of equal duration (1 sec) with an initial
time step size of 10 msec. The internal pressure of the cylinder is increasing linearly at the rate
of 1.0 psi/sec. The bisection process is activated 37 times during the first second of the time
history. In the beginning of the analysis, bisections take place due to rapid changes in geometric
effects. The bisection is again activated when the plastic deformation starts while approaching
the peak (maximum and minimum) displacements. The plastic deformation occurs in the
region near the small radius on the major axis of the ellipse (point B in Fig. 5.3.1). It is
noted that the plastic reloading occurs after the unloading while approaching the minimum
peak displacements. The automatic time step adjustment is activated 35 times during the first
second of the analysis, mainly to recover from the bisection process by stretching the previously
reduced time step size.

Fig. 5.3.2 shows the displacement responses at two points (denoted by A and B) on the
minor and major axes of the ellipse. The vibration frequency increases as the internal pressure
increases, but the amplitude diminishes due to the geometric stiffening. The static response is
plotted in the same figure, which shows good agreement between the static response and the
mean dynamic amplitude.

The static and dynamic deformed shapes (snap shots at the extreme magnitudes of the
vibration) in the vicinity of 5 psi are shown in Fig. 5.3.3. The static deformation turns the
elliptical into a nearly-circular cylinder, and the dynamic deformation is dominated by the
ovaling mode. The effect of follower forces is verified and validated by the fact that the static
deformed shape is approaching a circular shape. It is noted that the circular deformation pat-
tern is formed at less than 0.5 psi in static analysis, and the geometric nonlinearity disappears
beyond that point.
 t = 0.1 in.; thickness
E = 30 x lo6 psi; Young's modulus
Y = 0.3 ; Poisson's ratio
a, = 50000 psi; yield stress
H = 3 x 10' psi; plasticity modulus
p = 0.00283 lb-sec2/in4; mass density

to =5 sec

Figure 5.3.1 Elliptic Cylinder Subject to an Internal Pressure

I displacement of point A

-
--- dynamic response
static response

displacement of point B

0 1

Pressure (psi)

Figure 5.3.2 Displacement Responses for Elliptic Cylinder

 Major Axis

Figure 5.3.3 Deformed Shapes of the Elliptic Cylinder

5.3 - 7
 Table 5.3.1 SOL 66 Input Data Listing for Elliptic Cylinder
ID ELLIP66, V67 $ APR 6-25-91, SSH 4/20/88, SHL 3/9/88, KIM 2/87
SOL 66 $ NONLINEAR STATIC ANALYSIS
TIME 30 $ CPU MINUTES
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
CEND
TITLE = ELLIPTIC CYLINDER UNDER INTERNAL PRESSURE
SEALL = ALL
ECHO = BOTH
SET I = 1000,1120
DISP = I
SPC = 200
SUBCASE I
LOAD = 10
NLPARM = 100
SUBCASE 2
LOAD = 20
NLPARM = 200
SUBCASE 3
LOAD = 30
NLPARM = 300
SUBCASE 4
LOAD = 40
NLPARM = 400
SUBCASE 5
LOAD = 50
NLPARM = 500
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
SET I PLOTEL
VIEW O., O., 0.
AXES Z , X , Y
MAXI DEFO .I
FIND SCALE ORIGIN I SET I
PLOT STATIC 0 I RANGE 1.0 MAXI DEFO .I SET I
PLOT STATIC 0 2 RANGE 2.0 MAXI DEFO .I SET I
PLOT STATIC 0 3 RANGE 3.0 MAXI DEFO .I SET 1
PLOT STATIC 0 4 RANGE 4.0 MAXI DEFO .I SET I
PLOT STATIC 0 5 RANGE 5.0 MAXI DEFO .I SET I
OUTPUT(XYPLOT)
CSCALE 1.3
PLOTTER NAST
YMIN = -60.
YMAX = 80.
XTITLE = TIME IN SEC
YTITLE = DISP OF GRID 1000 AND 1120
XYPLOT DISP RESP/l000(Tl),ll20(T2)
BEGIN BULK
PARAM LGDISP 1
$$ SOLUTION CONTROL
NLPARM 100 20 AUTO YES
NLPARM 200 10 AUTO YES
NLPARM 300 8 AUTO YES
NLPARM 400 5 AUTO YES
NLPARM 500 8 AUTO YES
$$ LOADING
PLOAD4 10 101 THRU 112
PLOAD4 20 101 THRU 112
PLOAD4 30 101 THRU 112
PLOAD4 40 101 THRU 112
PLOAD4 50 101 THRU 112
$$ GEOMETRY
GRID 1000
GRID 1001
GRID 1010
GRID 1011
GRID 1020
GRID 1021
GRID 1030
GRID 1031
GRID 1040
GRID 1041
GRID 1050
GRID 1051
GRID 1060
GRID 1061
GRID 1070
GRID 1071
GRID 1080
GRID 1081
GRID 1090
GRID 1091
GRID 1100
GRID 1101
GRID 1110
GRID 1111
GRID 1120
GRID 1121
GRID 2000
GRID 3000
GRID 4000
$$ CONNECTIVITY
CQUAD4 101 100
CQUAD4 102 100
CQUAD4 103 100
CQUAD4 104 100
CQUAD4 105 100
CQUAD4 106 100
CQUAD4 107 100
CQUAD4 108 100
CQUAD4 109 100
CQUAD4 110 100
CQUAD4 Ill 100
CQUAD4 112 100
$$ PROPERTIES
PSHELL 100 110 .I 110
MAT1 110 3.+7 .3
MATS1 110 PLASTIC 3 .+5
$$ CONSTRAINTS
SPCl 200 16
SPCI 200 26
$$ FOR PLOT
PLOTEL 1000 1000
PLOTEL 1001 1010
PLOTEL 1002 1020
PLOTEL 1003 1030
PLOTEL 1004 1040
PLOTEL 1005 1050
PLOTEL 1006 1060
PLOTEL 1007 1070
PLOTEL 1008 1080
PLOTEL 1009 1090
PLOTEL 1010 1100
PLOTEL 1011 ill0
PLOTEL 1012 2000
PLOTEL 1013 2000
$
ENDDATA
 Table 5.3.2 SOL 99 Input Data Listing for Elliptic Cylinder
ID ELLIP99, V67 $ APR 6-25-91, SSH 4/28/88, SHL 3/9/88, KI-OOK KIM 2/87
SOL 99 $ NONLINEAR TRANSIENT ANALYSIS
TIME 200 $ CPU MINUTES
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
CEND
TITLE = ELLIPTIC CYLINDER UNDER INTERNAL PRESSURE
SUBTITLE = PROPORTIONALLY LOADED UP TO 5 PSI IN 5 SECONDS
SEALL = ALL
ECHO = BOTH
SET I = 1000,1120
DISP = I
SPC = 200
LOADSET = 1000
SUBCASE I
DLOAD = I 0 0
TSTEPNL = 100
SUBCASE 2
DLOAD = I00
TSTEPNL = 200
SUBCASE 3
DLOAD = 100
TSTEPNL = 300
SUBCASE 4
DLOAD = 100
TSTEPNL = 400
SUBCASE 5
DLOAD = 100
TSTEPNL = 500
OUTPUT(PL0T)
CSCALE 1.3
PLOTTER NAST
SET 1 PLOTEL
VIEW O., O., 0.
AXES Z , X , Y
MAXI DEFO .I
FIND SCALE ORIGIN 1 SET I
PLOT TRANS 0 TIME 4.8 5.0 MAXI DEFO .I SET I
OUTPUT(XYPLOT)
CSCALE 1.3
PLOTTER NAST
XTITLE = TIME IN SEC
YTITLE = DISP OF GRID 1000 AND 1120
XYPLOT DISP RESP/IOOO(Tl), 1120(T2)
BEGIN BULK
PARAM LGDISP 1
$$ SOLUTION CONTROL
TSTEPNL 100 100 .Ol ADAPT

5.3 - 11
TSTEPNL 200 100 ADAPT
TSTEPNL 300 100 ADAPT
TSTEPNL 400 100 ADAPT
TSTEPNL 500 100 ADAPT
$$ LOADING
LSEq 1000 900
PLOAD4 110 101 THRU 112
TLOADI 100 900
TABLED1 120 +TBDI
+TBDI 0. 0. 5. I. ENDT
$$ GEOMETRY
GRID 1000
GRID 1001
GRID 1010
GRID 1011
GRID 1020
GRID 1021
GRID 1030
GRID 1031
GRID 1040
GRID 1041
GRID 1050
GRID 1051
GRID 1060
GRID 1061
GRID 1070
GRID 1071
GRID 1080
GRID 1081
GRID 1090
GRID 1091
GRID 1100
GRID 1101
GRID I110
GRID 1111
GRID 1120
GRID 1121
GRID 2000
GRID 3000
GRID 4000
$$ CONNECTIVITY
CqUAD4 101 100
CQUAD4 102 100
CqUAD4 103 100
CqUAD4 104 100
CQUAD4 105 100
CqUAD4 106 100
CQUAD4 107 100
CqUAD4 108 100
CqUAD4 109 100
CqUAD4 110 100
CqUAD4 111 100
CqUAD4 112 100
$$ PROPERTIES
PSHELL 100 110 .1 110 110
MAT1 110 3.+7 .3 .283-2
MATS1 110 PLASTIC 3 .+5 50000
$$ CONSTRAINTS
SPCI 200 16
SPCi 200 26
$$ FOR PLOT
PLOTEL 1000 1000
PLOTEL 1001 1010
PLOTEL 1002 1020
PLOTEL 1003 1030
PLOTEL 1004 1040
PLOTEL 1005 1050
PLOTEL 1006 1060
PLOTEL 1007 1070
PLOTEL 1008 1080
PLOTEL 1009 1090
PLOTEL 1010 I100
PLOTEL 1011 I110
PLOTEL 1012 2000
PLOTEL 1013 2000
$
ENDDATA
5.4 'TREATMENT OF LARGE ROTATIONS

Geometric nonlinearities may produce large displacements and rotations. Unlike displacements,
finite rotations (components 4 , 5 and 6 of the MSCINASTRAN displacement vector) can not be
updated by simple vector addition. In MSCINASTRAN, there are two different approaches t o
update rotations: namely, gimbal angle approach and rotation vector approach (implemented
in Version 67).

The user has the option t o select either of the two approaches. This is done by specifying
PARAM,LANGLE,l (default option) for the gimbal angle approach and PARAM,LANGLE,2
for the rotation vector approach in either the Bulk Data or the Case Control. The value of
LANGLE can not be changed between subcases or at the restart, and is determined at the
time of the cold start.

5.4.1 Gimbal Angle Approach

The gimbal angle approach is selected by specifying PARAM,LANGLE,l. In the gimbal angle
approach, the three components of rotations at a grid point (O,, O,, 0,) are interpreted as finite
rotations of a rigid body attached to the grid point subjected t o three successive rotations as
follows: first the body is rotated about the grid point's global z-axis by an amount 0,; then
it is rotated about the reoriented y-axis by an amount 0,; and finally it is rotated about the
twice reoriented x-axis by an amount 0,. The three angles Ox, 0, and 0, are called gimbal
angles in MSCINASTRAN [5.6].

It can be shown that the same final orientation of the rigid body can also be obtained by
first rotating the body by an amount of Ox about the global x-axis, then by an amount of 8,
about the global y-axis, and finally by an amount of 0, about the global z-axis. The output is
in terms of Ox, 8, and O,, the three gimbal angles.

The rotation matrix R corresponding t o the gimbal angles is given by

where the matrices Rx,Ry, and RZ are

1 0 0
with c; = cos 6; and s; = sin 6; for i=x,y,z and 6; are the three gimbal angles. Multiplying Rz,
Ry, and Rx one obtains

Three tasks are performed for every iteration in the gimbal angle approach as follows:

Step 1. Update the Gimbal Angles

The gimbal angles are updated using

The expressions for the incremental gimbal angles are derived by transforming gimbal angles
to global angles and equating them with those obtained from the equilibrium solution. If we
assume the global incremental angles (SOi, i=x,y,z) are small, we have

It should be noted that matrices Rz and Ry are based on the gimbal angles at the beginning
of the iteration. This is reasonable because of the small angle assumption. Solving for the
incremental gimbal angles, the following expressions are obtained

where c; and s;, i=x,y,z correspond t o the gimbal angles a t the beginning of iteration.

It can be seen from the above equations that there is a singularity at Oy = 90'. This is
called gimbal lock. The gimbal lock in MSCINASTRAN is avoided by switching temporarily
t o a different set of angles, called auxiliary angles, and then reverting back to the gimbal
angles. This is possible because the rotation matrix is unique and there exist several ways t o
obtain the final orientation of a rigid body attached t o a grid point. The interpretation of the
auxiliary angles is that the rotation of a rigid body attached to a grid point is computed by
first rotating the body by an amount of 4, about the global x-axis, then by an amount of 4,
about the global y-axis, and finally by an amount of 4, about the global x-axis. The algorithm
t o avoid gimbal lock is as follows:
1. Form rotation matrix R(6) based on the three gimbal angles at the beginning of the
iteration.
 2. Obtain the auxiliary angles q5 from R(9).

3 . Update the auxiliary angles, i.e., +if l = + $ A4.

4. Form updated rotation matrix from the updated auxiliary angles, i.e., R(q5 + A+).
5 . Recover the updated gimbal angles, i.e., Oi+'.
This algorithm may sometimes lead to sudden changes in the gimbal angles. However, it does
not affect the solution because the element routines use the rotation matrix which is unique.

Step 2. Eliminate Rigid Body Rotation

The rigid body rotation is removed from the rotation matrix as follows:

where R is the rotation matrix corresponding to angular deformations. The matrices T b d ,

Tbg, and Tbe transform the deformed element coordinates, the global coordinates, and the
undeformed element coordinates, respectively, to basic coordinates. The matrices Tdg and
Teg transform the global coordinates t o deformed and undeformed element coordinates, re-
spectively.

Step 3. Calculate Angular Deformation

The gimbal angle approach assumes that the angular deformations are small. Hence, the
angular deformations are recovered as follows:

69, = arcsin i ( ~ 3 - 2 ~ 2 3 )
6ey = arcsin i ( 8 1 3 - ~ 3 1 )
b& = arcsin i ( ~ 2 -l ~ 1 2 )

These angular deformations are then used t o calculate the element forces.

5.4.2 Rotation Vector Approach

The rotation vector approach is selected by specifying PARAM,LANGLE,B. In the rotation

vector approach the three components of rotation at a grid point, denoted as $, $,, $, t o
distinguish from the gimbal angles, are interpreted as three components of a rotation vector
(+). The orientation of a rigid body attached t o a grid point is obtained by rotating the body
by an amount !l!,called the magnitude of rotation, about an axis of rotation p:

where
and

(5.4.13)

The output from MSCINASTRAN is in terms of $, $,, and $, the three components of
rotation vector +.
Let p; be a component of the unit vector parallel to the axis of rotation and Q be the
magnitude of rotation. Then the rotation matrix is as follows [5.3]:
- -
+
cos Q p:(l - cos Q ) +
-p, sin Q p,py(l - cos !P) +
p, sin Q p,p,(l - cos Q)

R= p, sin Q + p,p,(l - cos Q) cos Q + p t ( l - cos Q ) -p, sin Q + p,p,(l - cos Q)

- -py sin Q + p,p,(l - cos Q) p, sin Q + pyp,(l - cos Q) cos XQ + p:(l - cos Q) -
(5.4.14)

Like the gimbal angle approach, three tasks are also performed in the rotation vector
approach. The details of the three tasks are:

Step 1. Update the Rotation Vector

The rotation vector can be updated by successive rotations. First update the rotation
matrix, i.e.,
~ i + -l SR R~ (5.4.15)
where S R is the rotation matrix for the current iteration, and then recover the updated rotation
vector from the updated rotation matrix R~+'.MSCINASTRAN employs an algorithm which
assumes that the incremental rotations (S$; i=x,y,z) in any iteration are small. That is SR
can be approximated by
1 -S$z 6$,

- S$x 1

In this algorithm the updated rotation vector zl,i+l is split into a scalar and a vector (no
longer unit vector):
+i+i = ~ i + Pi+l
1 (5.4.17)
with
Qi+l = Qi + AQ
and
pi+' = pi +Ap (5.4.19)
The scalar increment, AQ, and the vector increment, A p , are recovered from given Qi, pi and
Szl,, where Szl, is obtained from equilibrium iteration. Comparing the elements of the matrix
Eq. (5.4.15) and after some transformations the following expressions are obtained:
 APX =
(1 + cos Q)(b$, +
- A Q pX) (W, p, - S$, py) sin \.I!
(5.4.21)
2 (sin llj + A P cos !J)
(1 + cosg)(S$y - AQ py) + (S$, px - S$, pZ) sin tlj
APY = 2 (sin Q + A@ cos Q)

APZ =
+
(1 cos Q)(S$z - A g pz) +
(SgXpy - S$, p,) sin
2 (sin D
! +
AQ cos @ )
Special care has t o be taken at the singularity points Q = 0 and J! = n. The angles +i as well
as the updated angles l have t o be checked for the singularity points. Therefore, one needs

an estimate for the updated angle +i+l through vector addition of $i and S$:
+i+l = +i+ S$ (5.4.24)
The singularity points are checked for the angle $i and Whenever $ior $i+l are close
to a singularity with a tolerance of kc,the angles are updated through simple vector addition
according t o Eq. (5.4.24). The vector addition is consistent with the first order update, and
does not affect the converged solution. The increment in @ is limited t o n per iteration.

Step 2. Eliminate Rigid Body Rotation

The procedure to eliminate rigid body rotation is the same as that used in the gimbal angle
approach.

Step 3. Calculate Angular Deformation

The angular deformations for the current iteration are calculated from the rotation matrix
R, which is of the same form as given in Eq. (5.4.14), as follows: the magnitude $ is obtained
from the first invariant of R. That is:

The components of the rotation vector p are obtained from the off-diagonal components of R:

1 ~ 1 -3 ~ 3 1
p ~ = 2s i n g
1 ~ 2 -1 I212
Pz = -
2 sin P!

It becomes obvious that the recovery procedure outlined above has singularities at 6! = 0, n,
and therefore is only valid in the range 0 < Q < n. For the singularity at !4 = 0 the small angle
I = n is avoided by restricting
rotation matrix in Eq, (5.4.16) is used, and the singularity a t D
@! t o
O < . < ? (5.4.29)
2
This is not only reasonable but necessary t o avoid unrealistic element distortions.
5.4.3 Bisection Due to Large Rotations

Both the rotation vector and gimbal angle approaches assume that the incremental rotations
from the equilibrium iterations satisfy the small angle theory. However, in actual computations
this may not be true. Therefore, in MSCINASTRAN the user has an option (default) t o
bisect the load if the incremental rotations are large. The user can choose t o bisect the load or
time step if bei, i=x,y,z exceeds RTOLB specified in degrees in the eighth field of the second
continuation entry for the NLPARM and the TSTEPNL bulk data entries. The current default
value for RTOLB is 20'.

In nonlinear problems the solution is obtained by iterations, and since the small angle
rotation matrix in Eq. (5.4.16) is a first order approximation, in many problems it may be a
computing advantage t o use a larger value for RTOLB. Using a larger value of RTOLB may
result in more iterations. However, no accuracy is lost if the solution converges.

The gimbal angle approach also assumes small angular deformations (in performing Step
3). Some accuracy in the results may be lost if the angular deformations are large. Therefore,
it is more appropriate t o use the rotation vector approach if the angular deformations are
expected to be large.

5.4.4 Output Interpretation

In MSCINASTRAN the user can specify enforced non-zero angular displacements. These
angular displacements are always enforced about the global axes. However, the printed results
from the gimbal angle approach are the three gimbal angles. The gimbal angles in general for
3D rotations are not the same as the global angles. Therefore, in the gimbal angle approach,
the printed results may not be equal t o the specified value. This should not cause any alarm t o
the user, as the results represent gimbal angles while the specified rotation is about the global
axes.

Consider a rigid body attached t o a grid point. Let the rigid body be subjected to two
successive rotations of 20' (0.349 radians) about the (0, 5,
$) axis. The total rotation is 40'
(0.698 radians) about the (0, A, A) axis. The output from MSCINASTRAN for the three
rotations, i.e., components 4, 5 and 6 of the displacement vector for the grid point at the end
of each rotation is as follows:

1. After First Rotation:

Qz 0, Qz
Gimble angle approach 0.0 2.4683 x 10-I 2.4683 x 10-I
Rotation vector approach 0.0 2.4683 x 10-I 2.4683 x 10-I
 2. After Second Rotation:
Ox 0, 82
Gimble angle approach 6.2192 x 4.8617 x 10-I 5.0885 x 10-I
Rotation vector approach 0.0 4.9365 x 10-I 4.9365 x 10-I

The details of the calculations for the two approaches are as follows:

Gimbal Angle Approach

1. First Rotation Increment:

Gimbal angles at the beginning of increment

Incremental angles from global iteration

Incremental gimbal angles from Eq. (5.4.8)

Updated gimbal angles a t the end of increment

2. Second Rotation Increment:

Gimbal angles at the beginning of increment

Incremental angles from global iteration

 r Incremental gimbal angles from Eq. (5.4.8)

ne, = 6.2192 x 10-2

A@, = 2.3941 x 10-I
Ae, = 2.6202 x 10-I

r Updated gimbal angles at the end of increment

Rotation Vector Approach

1. First Rotation Increment:

r At the beginning of increment

r Incremental angles from global iteraiion

r Updated angles at the end of increment from Eq. (5.4.24)

2. Second Rotation Increment:

r At the beginning of increment

for which
Incremental angles from global iteration

for which

At the end of increment

$ = $ + All, = 3.4907 x 10-I + 3.4907 x 10-I = 6.9814 x 10-I

updated angles = ll,p

As a result
5.5 VERIFICATION: LARGE ROTATION OF A BEAM

This example compares geometric nonlinear analysis using SOL 64 and SOL 66. SOL 64 is an
older trial DMAP developed for geometric nonlinear analysis only, whereas SOL 66 (or 106)
is more recently developed and contains many features not available in SOL 64. SOL 66 is
also simpler to use because the solution is controlled by the NLPARM Bulk Data instead of
subcases and parameters as in SOL 64. It is therefore preferable t o use SOL 66 (or 106) for the
geometric nonlinear analysis. Two problems are studied: a beam with an axial end force and
a beam with an end moment. SOL 64 and SOL 66 in Version 66A are used for the analysis of
the two problems. The main feature of these problems is geometric nonlinearity due to large
rotations.

5.5.1 Problem Description

A uniform beam is clamped at one end and free at the other end. The beam has a length of
12 inches and a cross-sectional area of 0.15 in2. The Young's modulus is 20 x lo6 lb/in2 and
the moment of inertia I,, is 2.813 x in4. Two cases are analyzed according t o the loading
conditions at the free end of the beam: axial end force and end moment.

Beam with Axial End Force

Initially, the beam is subjected t o an axial force P=0.9Pc, and a small lateral force Pi=O.lPcr
(initial disturbance) at the free end as shown in Fig. 5.5.1. The lateral force is subsequently
removed and large rotations of the beam are produced when the axial force is increased above
the critical value PC,. For this problem, the critical load which is calculated by Timoshenko
and Gere [5.7] is

The beam rotation is determined for the ratios of P/PcT of 1.152,1.518,2.541,4.029, and 9.116
in order t o compare the results with those of Timoshenko and Gere.

Figure 5.5.1 Beam with an Axial End Force.

Beam with End Moment
A moment with a magnitude of 3000 in-lb is applied in the +z direction at the free end of the
beam, as shown in Fig. 5.5.2, and large rotations are produced.

I I I ~ I I I I I I IM = 3000 in-lb

Figure 5.5.2 Beam with an End Moment.

5.5.2 Finite Element Model

The beam is modeled by twelve BEAM elements with linear elastic material. The clamped
boundary condition at one end of the beam is imposed by constraining all six components of
GRID 100. The applied load at the free end of the beam (GRID 112) is specified using the
FORCE Bulk Data entry for the axial end force case and the MOMENT Bulk Data entry for
the end moment case. Geometric nonlinearity effects are included in the SOL 66 analysis by
specifying PARAM,LGDISP,l. The solution strategy is provided by the NLPARM Bulk Data
entry in SOL 66. In the SOL 64 analysis, the solution is controlled by subcases. The input
data for SOL 64 and SOL 66 are given in Tables 5.5.1 and 5.5.2, respectively, for the axial end
force case and in Tables 5.5.3 and 5.5.4, respectively, for the end moment case.

5.5.3 Analysis Procedure

SOL 64 Analysis

Subcases and parameters control the number of iterations, load vector updates, and stiffness
matrix updates in SOL 64. The stiffness matrix is updated after every iteration and there is
one load increment per subcase. The first two subcases correspond t o the linear elastic and the
differential stiffness solutions, respectively. PARAM,TESTSE,l.-10 is included in the Bulk
Data for an automatic convergence test based on strain energy.

The analysis is performed in seven subcases for the beam with axial end force. The initial
loading conditions (axial force of 0.9Pc, and lateral force of O.lPc,) are contained in the first
two subcases. Five additional subcases are included t o specify the five axial force ratios of
PIP,, with no lateral force.
 For the beam with end moment, the analysis is performed in twelve subcases. The first two
subcases correspond to the linear elastic and differential solutions, and the last ten subcases
correspond t o the ten moment increments.

SOL 66 Analysis

In SOL 66, the NLPARM Bulk Data controls the load increment size and the iteration strategy.
Automatic convergence tests are provided and no special parameters need t o be included to
control the solution strategy.

The analysis is performed in six subcases for the beam with axial end force. The first
subcase, which contains the initial loading, triggers large bending beyond the critical buckling
load. The lateral force is removed and each of the remaining subcases correspond t o an
increasing axial force ratio of P/P,,. Each subcase has one increment and the ITER option of
KMETHOD is selected with KSTEP=l, specifying the full Newton's method.

For the beam with end moment, the analysis is performed in one subcase which has ten
increments. Runs are made with different KMETHOD options (ITER, SEMI, or AUTO),
which specify the stiffness matrix update strategy, while KSTEP=l. The analysis executes
successfully with each of the KMETHOD options when the line search and BFGS updates
are suppressed (MAXLS=O and MAXQN=O). However, when the AUTO option is selected
with line search and BFGS updates, the solution diverges at the first load increment. The
performance in terms of the number of iterations and stiffness updates is tabulated below.
Performance for Beam with E n d Moment
No. of No. of
KMETHOD K updates iterations

AUTO

5.5.4 Analysis Results

The MSCINASTRAN analysis results for the beam with axial end force and the values calcu-
lated by Timoshenko and Gere [5.7] are tabulated below for the ratios of PIP,,. The distance
yend is the displacement of the free end of the beam in the vertical direction and xend is the
horizontal distance from the free end of the beam t o its clamped end. The tangent at the free
end of the beam has an angle Bend with the horizontal. The MSCINASTRAN results are in
good agreement with the values of Timoshenko and Gere. The relation between the axial and
lateral displacement of the free end of the beam (ul12 and vllz) and the axial force is shown
in Fig. 5.5.3. The displaced shapes of the beam are shown in Fig. 5.5.4.
 Load-Deflection Data of Beam with Axial End Force

Notes: values in ( ) denote solution in Timoshenko and Gere [5.7]

Figure 5.5.3 Load vs. Displacement at Free End of Beam with Axial End Force.
Figure 5.5.4 Deflection Curves for Beam with Axial End Force.
 The displaced shapes of the beam with end moment are shown in Fig. 5.5.5. The deflection
curves are labeled with the increment number (from 1 t o 10). The moment magnitude and
Oend, the angle that the tangent at the free end of the beam has with the horizontal, are
listed for each incremental step. Note that after ten increments, the beam has deformed into
a complete circle.

inc moment (in-lb) Bend

1 300 36.7"
2 600 73.3"
3 900 110.0"
4 1200 146.7"
5 1500 183.3"
6 1800 2'20.0"
7 2100 256.6"
8 2400 293.3"
9 2700 330 .OO
10 3000 366.6"

Figure 5.5.5 Deflection Curves for Beam with End Moment.

5.5 - 6
 Table 5.5.1 SOL 64 Input Data Listing for Beam with Axial End Force
ID PSTBUCK64, V66A $ APR 20-MAR-90
TIME 5 $ CPU MINUTES
SOL 64 $ GEOMETRIC NONLINEAR ANALYSIS
DIAG 8
CEND
TITLE = LARGE DEFLECTIONS OF A BEAM --- V6402S
SUBTITLE = TIMOSHENKO AND GERE, THEORY OF ELASTIC STABILITY; P.76
LABEL = P(CR1T) = EI*(.5*PI/L)**2 = 96.4
ECHO = BOTH
SEALL = ALL
DISP = ALL
SPCF = ALL
ELFOR = ALL
SPC = 100
SUBCASE 1 $ P/P(CR)=0.900,PLUS LATERAL
LOAD = 900
SUBCASE 2 $ P/P(CR)=0.900 PLUS LATERAL
LOAD = 900
SUBCASE 3 $ P/P(CR)=l. 152
LOAD = 1152
SUBCASE 4 $ P/P(CR)=l.518
LOAD = 1518
SUBCASE 5 $ P/P (CR) =2.541
LOAD = 2541
SUBCASE 6 $ P/P (CR) =4.029
LOAD = 4029
SUBCASE 7 $ P/P(CR)=9.116
LOAD = 9116
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
SET 1 = ALL
VIEW 0. , 0. , 0.
AXES Z , X , Y
PTITLE = TOP VIEW
FIND SCALE ORIGIN 1 SET 1
PLOT STATIC 0 SET 1 ORIGIN 1 SYMBOL 1
BEGIN BULK
$ PARAMETERS
PARAM TESTSE 1.-10
$ GEOMETRY
GRID 100 0. 0. 0.
GRID 101 1. 0. 0.
GRID 102 2. 0. 0.
GRID 103 3. 0. 0.
GRID 104 4. 0. 0.
GRID 105 5. 0. 0.
GRID 106 6. 0. 0.
GRID 107 7.
GRID 108 8.
GRID 109 9.
GRID 110 10.
GRID Ill 11.
GRID 112 12.
$ CONNECTIVITY
CBEAM 1 10 100 I01
CBEAM 2 10 101 102
CBEAM 3 10 102 103
CBEAM 4 10 103 104
CBEAM 5 10 104 105
CBEAM 6 10 105 106
CBEAM 7 10 106 107
CBEAM 8 10 107 108
CBEAM 9 10 108 109
CBEAM 10 10 109 110
CBEAM 11 10 110 Ill
CBEAM 12 10 Ill 112
$ PROPERTIES
PBEAM 10 11 .I5
MAT1 I1 2.+7 1.+7
$ CONSTRAINTS
SPCI 100 126 100
$ LOADING -- P(CRIz96.4
FORCE 900 112
FORCE 1152 112
FORCE 1518 112
FORCE 2541 112
FORCE 4029 112
FORCE 9116 112
$
ENDDATA
 Table 5.5.2 SOL 66 Input Data Listing for Beam with Axial End Force
ID PSTBUCK66, V66A $ APR 20-MAR-90
TIME 5 $ CPU MINUTES
SOL 66 $ NONLINEAR STATIC ANALYSIS
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
CEND
TITLE = LARGE DEFLECTIONS OF A BEAM --- V6402S
SUBTITLE = TIMOSHENKO AND GERE, THEORY OF ELASTIC STABILITY; P.76
LABEL = P(cRIT) = EI*(.5*PI/L)**2 = 96.4
ECHO = BOTH
SEALL = ALL
DISP = ALL
SPCF = ALL
ELFOR = ALL
SPC = 100
NLPARM = 1000
SUBCASE I $ P/P(CR)=0.900 PLUS LATERAL
LOAD = 900
SUBCASE 2 $ P/P (CR) =I.152
LOAD = 1152
SUBCASE 3 $ P/P(CR)=I.518
LOAD = 1518
SUBCASE 4 $ P/P (CR) =2.541
LOAD = 2541
SUBCASE 5 $ P/P(CR)=4.029
LOAD = 4029
SUBCASE 6 $ ~/~(CR)=9.116
LOAD = 9116
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
SET I = ALL
VIEW 0. , 0. , 0.
AXES Z , X , Y
PTITLE = TOP VIEW
FIND SCALE ORIGIN I SET I
PLOT STATIC 0 SET I ORIGIN I SYMBOL I
BEGIN BULK
$ PARAMETERS
PARAM LGDISP I
$
NLPARM 1000 I ITER I YES +NLP
+NLP 0
$ GEOMETRY
GRID 100 0. 0. 0.
GRID 101 I. 0. 0.
GRID 102 2. 0. 0.
GRID 103 3. 0. 0.
GRID 104 4. 0. 0.
GRID 105 5.
GRID 106 6.
GRID 107 7.
GRID 108 8.
GRID 109 9.
GRID 110 10.
GRID 111 11.
GRID 112 12.
$ CONNECTIVITY
CBEAM 1 10 100
CBEAM 2 10 101
CBEAM 3 10 102
CBEAM 4 10 103
CBEAM 5 10 104
CBEAM 6 10 105
CBEAM 7 10 106
CBEAM 8 10 107
CBEAM 9 10 108
CBEAM 10 10 109
CBEAM 11 10 110
CBEAM 12 10 11 1
$ PROPERTIES
PBEAM 10 11 .15
MAT1 11 2.+7 1.+7
$ CONSTRAINTS
SPC1 100 126 100
$ LOADING -- P(CR)=96.38286
FORCE 900 112
FORCE 1152 112
FORCE 1518 112
FORCE 2541 112
FORCE 4029 112
FORCE 9116 112
$
ENDDATA
 Table 5.5.3 SOL 64 Input Data Listing for Beam with End Moment
ID BMMOM64, V66A $ APR 20-MAR-90
TIME 5 $ CPU MINUTES
SOL 64 $ GEOMETRIC NONLINEAR ANALYSIS
DIAG 8
CEND
TITLE = LARGE DEFLECTION DEMO
ECHO = BOTH
SEALL = ALL
DISP = ALL
SPCF = ALL
ELFOR = ALL
SPC = 100
SUBCASE I
LOAD = 103
SUBCASE 2
LOAD = 103
SUBCASE 3
LOAD = 103
SUBCASE 4
LOAD = 106
SUBCASE 5
LOAD = 109
SUBCASE 6
LOAD = 112
SUBCASE 7
LOAD = 115
SUBCASE 8
LOAD = I 1 8
SUBCASE 9
LOAD = 121
SUBCASE 10
LOAD = 124
SUBCASE 11
LOAD = 127
SUBCASE 12
LOAD = 130
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
SET I = ALL
AXES Z, X, Y
VIEW O., O., 0.
PTITLE = TOP VIEW
FIND SCALE ORIGIN I SET 1
PLOT STATIC 0 SET 1 ORIGIN I SYMBOL 1
BEGIN BULK
$ PARAMETERS
PARAM TESTSE 1.-10
$ GEOMETRY
GRID 100
GRID 101
GRID 102
GRID 103
GRID 104
GRID 105
GRID 106
GRID 107
GRID 108
GRID 109
GRID 110
GRID Ill
GRID 112
$ CONNECTIVITY
CBEAM I 100 101 0.
CBEAM 2 I01 102 0.
CBEAM 3 102 103 0.
CBEAM 4 103 104 0.
CBEAM 5 104 105 0.
CBEAM 6 105 106 0.
CBEAM 7 106 107 0.
CBEAM 8 107 108 0.
CBEAM 9 108 109 0.
CBEAM 10 109 110 0.
CBEAM I1 110 Ill 0.
CBEAM 12 Ill 112 0.
$ PROPERTIES
PBEAM 10
MAT1 11
$ CONSTRAINTS
SPCI 100
$ LOADING
MOMENT 103
MOMENT 106
MOMENT 109
MOMENT 112
MOMENT 115
MOMENT 118
MOMENT 121
MOMENT 124
MOMENT 127
MOMENT 130
$
ENDDATA
 Table 5.5.4 SOL 66 Input Data Listing for Beam with End Moment
ID BMMOM66, V66A $ APR 20-MAR-90
TIME 5 $ CPU MINUTES
SOL 66 $ NONLINEAR STATIC ANALYSIS
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
CEND
TITLE = VERY LARGE DEFLECTION DEMO
ECHO = BOTH
SEALL = ALL
DISP = ALL
SPCF = ALL
ELFOR = ALL
SPC = 100
SUBCASE 10
LOAD = 120
NLPARM = 1000
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
SET I = ALL
AXES Z, X, Y
VIEW O., O., 0.
PTITLE = TOP VIEW
FIND SCALE ORIGIN I SET I
PLOT STATIC 0 SET I ORIGIN I SYMBOL I
BEGIN BULK
$ PARAMETERS
PARAM LGDISP I
$
NLPARM 1000 10 ITER I YES +NLI
+NLI 0
$ GEOMETRY
GRID 100
GRID 101
GRID 102
GRID 103
GRID 104
GRID 105
GRID 106
GRID 107
GRID 108
GRID 109
GRID 110
GRID Ill
GRID 112
$ CONNECTIVITY
CBEAM I 10
CBEAM 2 10
CBEAM 3 10
CBEAM 4 103 104 0.
CBEAM 5 104 105 0.
CBEAM 6 105 106 0.
CBEAM 7 106 107 0.
CBEAM 8 107 108 0.
CBEAM 9 108 109 0.
CBEAM 10 109 I10 0.
CBEAM 11 110 Ill 0.
CBEAM 12 Ill 112 0.
$ PROPERTIES
PBEAM 10
MAT1 11
$ CONSTRAINTS
SPCI 100
$ LOADING
MOMENT 120
$
ENDDATA
5.6 CANONICAL APPROACH TO
GEOMETRIC NONLINEARITY

In geometric linear analysis where the displacements are assumed small regardless of the mate-
rial nonlinearity, it is implied that the strain-displacement relations [B]stay constant. However,
the element matrix, [B], as well as the applied external loads vary with large displacements.
Conventional methods [5.8-5.121 such as the total Lagrangian, updated Lagrangian, or Eulerian
method treat these effects of large displacements by starting with the finite strain, in contrast
t o an approximate updated Lagrangian approach employed by MSCJNASTRAN based on the
infinitesimal strain. These conventional methods which are based on the classical continuum
mechanics theory [5.13-5.161, referred t o as canonical approach, are reviewed in this section.

5.6.1 Fundamentals of Continuum Mechanics

The kinematics of continua may be described from two different points of view. If the motion
of a body is described in terms of the undeformed configuration using a referential coordinate
system (denoted by X ) attached t o the object material, the method is known as the Lagrangian
(or material) formulation. In contrast, the description of the motion in terms of the deformed
configuration using spatial coordinates fixed in space (denoted by x) is called the Eulerian
(or spatial) formulation. The Lagrangian description may be interpreted as a mapping of the
initial configuration into the current configuration, i.e.

Conversely, the Eulerian formulation gives the initial position of the particle at time t in terms
of the current position x, i.e.
X = X(x,t). (5.6.2)
The most fundamental quantity of these descriptions is the material deformation gradient
defined as

F = [$I [ =
& & az,
axl ax, ax,
axl ax2 &&
&& ax3
ax, axz ax3
(5.6.3)

Then, the spatial deformation gradient can be defined by F-l.

Consider a squared length of an infinitesimal line element in terms of the deformation

gradient, i.e.
dxT dx = (F d ~ ) (*F d X ) = d x T C d X (5.6.4)
with
c = F ~ F
where Green's deformation tensor is defined by C. The deformation tensor similarly defined
for the Eulerian formulation is called Cauchy's deformation tensor and is denoted by c. As is
apparent from the derivation, these deformation tensors are symmetric and positive-definite.
The change in the squared length is used as a measure of deformation, i.e.

in which the Lagrangian (also called Green's) finite strain tensor is defined t o be

The Eulerian formulation can be derived likewise, i.e.

in which the Eulerian (also called Almansi's) finite strain tensor is defined t o be

Noting that x = X + u, the finite strain tensors can be expressed in terms of the displace-
ment gradient, i.e.

and

where the repeated indices represent summation. When components of the displacement gra-
dient are sufficiently small compared t o unity, the product terms are negligible and the finite
strain tensors reduce t o the infinitesimal strain definition. Then, the distinction between the
Lagrangian description and the Eulerian description disappears; which is the basis of the small
deformation theory. Notice that the expression for the finite strain tensor is exact and not a
second order approximation.

The displacement gradient can be decomposed into a symmetric and a skew-symmetric

part, i.e.

Within the framework of the infinitesimal theory, the symmetric part is the linear strain tensor
and the skew-symmetric part is the linear rotation tensor. When the displacement gradient
is finite, it can no longer be reduced t o a strain and a rotation by an additive decomposition.
According t o the polar decomposition theorem, however, the multiplicative decomposition is
possible if the deformation gradient is not a singular tensor, i.e.
where R is the orthogonal rotation tensor, and M and N are positive definite symmetric
tensors, known as the right and left stretch tensors, respectively. It follows that

by virtue of RTR = I. A significant interpretation can be given to Eq. (5.6.14); e.g., each
principal value of M, which is known as the stretch ratio (A = dx/dX), is the square root
of the corresponding principal value of C, and the principal directions of C and M coincide.
Based on the polar decomposition, it can be shown that the finite strain tensors are invariants
with respect to the rigid-body motion.

Stresses are usually measured in the current configuration, because the equilibrium must
be established in the deformed configuration. The stress in the deformed state is known as the
Cauchy stress, consistent with the Eulerian strain tensor. In the course of analysis, however,
if the strains are referred t o the original position of particles as in the Lagrangian description,
stresses should be defined similarly. Two Piola-Kirchhoff stress tensors are defined in the
undeformed (or referential) configuration. The first Piola-Kirchhoff stress tensor (also known
as Lagrangian stress), denoted by Tij, is defined to give the actual force acting on the deformed
surface element with an area dS, but the force is assigned t o the unit area of the corresponding
undeformed surface dSo. This stress tensor is easy t o define but has the disadvantage of being
non-symmetric. The second Piola-Kirchhoff stress tensor (also called Kirchhoff stress), denoted
by Sij, is defined to give the corresponding force acting on and assigned t o the undeformed
surface element dSo, but the force is related to the actual force in the deformed surface by the
deformation gradient. These Piola-Kirchhoff stress tensors are related t o the Cauchy stress
tensor as follows:
( det F ) loij] = F [Sij]F~ = F [Tij] (5.6.15)
or inversely,
[Sij] = ( det F ) F-I [aG]( F - ' ) ~ = [Ti,] ( F - ' ) ~ (5.6.16)
where the determinant of the deformation gradient F represents the change in volume due t o
the deformation, i.e.
- v
dxl dx2 dx3 - -
det F =
dXldX2dX3 Vo'
The second Piola-Kirchhoff stress tensor has little physical meaning; however, it has advantages
of being symmetric and independent of rigid-body rotation. It is the second Piola-Kirchhoff
stress tensor that is energetically conjugate t o the Lagrangian strain tensor, i.e., the strain
energy is preserved by
Sij SEij dV = (5.6.18)
5.6.2 Incremental Formulations for Large Displacements and Strains

The kinematic tensors in the incremental process of flow problems such as in plasticity, creep or
viscoplasticity are frequently expressed in terms of time rates in the process of linearization t o
follow the deformation paths. For instance, the velocity gradient rather than the displacement
gradient is used, i.e. - -

with a symmetric part:

1
D = - (L
2
+ LT)
and a skew symmetric part:
1
w = - ( L - LT)
2
where D and W are called the rate of deformation (or stretching) tensor and the vorticity
(or spin) tensor, respectively. While the rate of deformation tensor is independent of rigid-
body motion, the vorticity tensor represents the angular velocity of the rigid-body rotation.
Recalling that the natural-strain increment is defined as

t o give the following logarithmic strain by an integration over the length, i.e.

The rate of deformation tensor can be used for a 3-dimensional generalization of the natural-
strain increment in time dt, i.e.
da = D dt . (5.6.21)

The rate of deformation and the vorticity tensors can be interpreted as the material deriva-
tives of the Eulerian linear strain and rotation tensors, respectively. The material derivative
is the time rate of change of any property of a moving continuum measured by an observer
traveling with a specific particle under consideration. In contrast, the time rate of change
measured at a fixed point in space is called the local rate of change or the spatial derivative.
Consider an acceleration which is a material derivative of the velocity. When the velocity is
described only in terms of spatial variables x which are implicit functions of time, i.e.

21 = "(2, t),

then the acceleration can be calculated by

where the subscripts identify the variables held constant. The first term gives the local rate
of change and the second term gives the convective rate of change. The material derivative of
the deformation gradient can be obtained by
av- -
F. = - = av ax = LF.
ax ax ax
The principle of material frame-indifference states that constitutive equations must be
invariant under changes of reference frame. Functions (vectors or tensors) are called frame-
indifferent if their values are unchanged by the superposed rigid-body motion when the frame of
reference is changed from {x, t) to {x*, t*). Specifically, when the change of frame is described
by
x*(X, t*) = c(t) + &(t) x ( X , t ) , (5.6.24)
the vectors and tensors transform according to

v* = Q(t) v (for vectors) (5.6.25)

and
S* = Q(t) s ~ ( t ) ~ (for tensors) (5.6.26)
where Q(t) is a rotation tensor and c(t) represents a translation of the coordinate origin. It
can be shown, however, that the deformation gradient F transforms like a vector. It is noted
that the spin tensor is not frame-indifferent and transforms by

with
R = QQT = - Q Q' T = -QT

where Q is the material derivative of Q, and R is the angular velocity tensor of the unstarred
frame relative to the starred frame.

Of our particular interest is the time rate of change of the stress among constitutive variables
in the incremental formulation because the material derivative of the stress (irij) is not frame-
indifferent even though Uij is. However, it can be demonstrated that the frame-indifferent
stress rate is defined as
;=&-WO+OW (5.6.29)
which is known as the Jaumann stress rate or the corotational stress rate tensor and is denoted
by v over stress. The first term on the right side represents the material rate of change with
respect t o a fixed coordinate system and the last two terms represent the convective rate of
change. Finite strains have a similar problem because the material derivative of the finite strain
is not frame-indifferent. However, it can be shown that the rate of deformation tensor [D)
is frame-indifferent, despite the fact that the velocity is not. As pointed out earlier, the rate
of deformation tensor is a material derivative of the linear strain. Thus, the frame-indifferent
constitutive relations for the incremental formulation can be obtained by using the Jaumann
stress rate and the infinitesimal strain rate, i.e.

where D represents the constitutive relations.

Now we are ready to derive the incremental formulations for solving general nonlinear prob-
lems with large displacements and large strains. From the principle of virtual displacements,
the basic equation to solve in accordance with the Eulerian approach is

which is to be satisfied in the currently deformed configuration. The fundamental difficulty

with the Eulerian approach is that the current geometry of the body, which is the referential
configuration, is yet to be determined. Two popular approaches to the incremental formulations
are considered here: the total Lagrangian and the updated Lagrangian approaches, both based
on the Lagrangian description of kinematics. In the total Lagrangian formulation, all the
variables are referred t o the initial configuration. Without loss of generality, however, such a
natural procedure of the Lagrangian formulation may be applied with an arbitrary choice of
the referential configuration. When the choice of the referential configuration is the preceding
incremental load (or time) step, the method is known as the updated Lagrangian formulation.

The basic equation can be rewritten for the total Lagrangian formulation as

and for the updated Lagrangian formulation as

where the left superscripts refer to the current configuration of the body and the left sub-
scripts refer t o the referential coordinate axes, with 0, t, and t' denoting the initial, the t-th
incremental, and the (t+l)th incremental steps, respectively. It is noted that the expression
for the updated Lagrangian formulation is obtained by substituting t for the left subscripts
C in the total Lagrangian formulation. In the total Lagrangian formulation, the incremental
process of multiple load steps is applied to the Lagrangian formulation. On the other hand, the
updated Lagrangian formulation may be regarded as an application of the Lagrangian method
t o a single load step with the referential configuration brought up-to-date continuously, treat-
ing the previously converged configuration as an initial geometry. Now it remains to linearize
Eqs. (5.6.31a) and (5.6.3110) to find the tangential stiffness. It will be assumed for simplicity
that the applied load is independent of deformation.
5 -6.3 Total Lagrangian vs. Updated Lagrangian in Finite Elements

Recalling that t'u = tu+u for the total Lagrangian formulation, the stress and strain (omitting
indices) may be decomposed as follows:

and
t' = ;E + o E 1
,E + oE2 + oE3
with

Noting that
S ~ =
E S(oE1 + oE2 + oE3)
and
St'U = Su ,
the substitution of Eqs. (5.6.32) and (5.6.33) into Eq. (5.6.31a) gives the following equation
with some rearrangement of terms:

Notice that 0El is the linear strain increment causing a linear stiffness [I<'],0E2 is the nonlinear
strain increment causing a stiffness [ K ~due ] to the initial stress, and 0E3 is a cross-product
term which causes a stiffness [ K g ] due t o the initial displacement. If the strain-displacement
relations are linearized by
[%It = ;B = b~~ + ~ B N ,
the equilibrium equation can be rewritten in matrices as
where (omitting the left-superscript t and left-subscript 0 of B),

with a constitutive relation

is = OD i E .
During the iteration, the displacements are updated continuously, i.e., for the i-th iteration

with which 6 BN and 6s are reevaluated to yield $ BN and $5'. Thus, the unbalanced force
vanishes as the solution approaches convergence.

The total Lagrangian approach is not appropriate for the path-dependent inelastic problems
such as a plasticity problem. The reason is that the constitutive relations OD in Eq. (5.6.37)
must be defined with reference t o the initial geometry, which is meaningful only for elastic
materials such as a hyperelastic material. In fact, many researchers reported difficulties with
the total Lagrangian method in those problems. Thus, we have to resort t o the updated
Lagrangian approach for general geometric nonlinear problems. The concept of the updated
Lagrangian method is realized by updating the state variables at the end of each incremental
step with respect to the converged configuration or convected coordinates, i.e.

For the updated Lagrangian process, stresses and strains may be decomposed (omitting
the left-subscript t ) as follows:
= tS S + (5.6.40)
and
t'
E = El + E2
with

and
Noting that t S = a, the substitution of Eqs. (5.6.40) and (5.6.41) into Eq. (5.6.31b) gives, with
some rearrangement of terms,

Introducing the constitutive relations at t by

s +
= t ~ ( ~E l~ )

and linearizing Eq. (5.6.42) by neglecting E 2 in the first term, the equilibrium equation can be
written in a matrix form (omitting the left-subscript t of B and D ) as

[I(' + g d ]{u) = t' P - lVBE {a) dV

with
I(' = ly[BE tD BL]dV

and

where the current strain-displacement relation is obtained by

and the volume integration is performed over the current volume.

5.6.4 Constitutive Relations for Large Deformat ion

Difficulties arise in using the Lagrangian constitutive relations in Eq. (5.6.43). First of all,
the Lagrangian strain increments are not additive while the elasto-plastic deformation process
requires
&total = d,se +d ~ .~ (5.6.46)
However, since the rate of deformation tensor is additive, we can establish the rate constitutive
v
relations in terms of the Jaumann stress rate ( a ) and the rate of deformation tensor as presented
in Eq. (5.6.30). The rate of deformation tensor (D) is related to the Lagrangian strain rate by

By allowing the coordinate axes of the referential configuration to be momentarily coincident

with that of the current geometry, the distinction disappears and the Lagrangian strain rate
may be used as a constitutive variable, i.e.

Recalling the generalization of the natural-strain increment defined in Eq. (5.6.21), an integra-
tion of Eq. (5.6.48) may be interpreted t o give the logarithmic strain which is denoted by E*,
i.e.
E* = J ~ d t . (5.6.49)

This is another advantage of using the rate equations since we can now formulate the consti-
tutive relations [tD]based on the true stress-strain curve obtained from the uniaxial test. It
is noted that the generalized logarithmic strain, which has a physical meaning in the updated
Lagrangian coordinates, is different from either the Lagrangian or the Eulerian strain.

For every iteration, displacements are updated by

and the stresses on the right-hand side of Eq. (5.6.44) by

However, the frame-indifferent constitutive relations involve the Jaumann stress rate, which
is related t o 6 by Eq. (5.6.29), in order t o account for the rotation of the stress during the
increment. Since the Cauchy stress and the Kirchhoff stress are related by Eq. (5.6.15), the
material derivative of the Cauchy stress can be obtained by taking a material derivative of
Eq. (5.6.15), i.e.
FSF~ = (detF)(divva + ir - L a - O L ~ ) (5.6.52)
using Eq. (5.6.23) and the material derivative of (det F ) from
d
- (det F ) = J = J
dt
vi,i = J div v . (5.6.53)
By using Eq. (5.6.29) and the momentarily coincident referential configuration, Eq. (5.6.52)
reduces t o
s = av - D o - a D + d i v v a . (5.6.54)
For incompressible materials, as in plasticity, Eq. (5.6.54) further reduces to

using Eq. (5.6.19) with a substitution of E for D. Then the stress increment is calculated by

It is noted that the volume integration is performed over the configuration at t in Eq. (5.6.42).
As the solution converges, the stress becomes tS, and should be transformed to the Cauchy
stress a t t' according t o Eq. (5.6.15).

For the hyperelastic material, the Eulerian constitutive relations may be obtained from the
Lagrangian constitutive relations by

= -F F ~ D F T F T . (5.6.57)
det F
The Lagrangian strain can be transformed t o the Eulerian strain at t' by

where [ t F ]is the deformation gradient at t' with reference t o the configuration at t.
Chapter 6

MATERIAL NONLINEARITY
AND CONSTITUTIVE
RELATIONS

6.1 INTRODUCTION

6.1.1 Overview

Material nonlinearity becomes an issue in the analysis if the stress-strain relationship or the
constitutive equations are nonlinear. Constitutive equations characterize the macroscopic be-
havior of the material, which stems from the microscopic constitution of the material. All the
engineering materials are inherently nonlinear. From the mathematical point of view, it is not
feasible t o characterize a nonlinear material by a single set of equations for the entire range
of environmental conditions, such as the loading, temperature, and the rate of deformation.
Even in the restricted region of these conditions for specific needs, the mathematical model of
the material is still an idealization (e.g., incompressibility) of more general nonlinearity.

Depending on the application or the situation, the material behavior may be idealized or
simplified t o account for certain effects which are important in the analysis. The linear elastic
(Hookean) material is the simplest case of all in solid mechanics. As the strain exceeds the in-
finitesimal range, the material exhibits nonlinear behavior: nonlinear elastic if it is recoverable
and plastic if it is irrecoverable. If the temperature effects on the material properties become
an important issue, coupling between thermal and mechanical behavior should be properly
taken into consideration, which is the realm of thermo-elasticity and thermo-plasticity. If the
strain-rate has significant effects on the material response, we have to resort t o the theories of
visco-elasticity and visco-plasticity. Anisotropic effects add complexity t o the material models
of all kinds in solid mechanics.
 Development of material constitutive models has been prolific in recent years. Detailed
discussion in this book will be confined to the capabilities provided in MSCINASTRAN.

6.1.2 User Interface

The data for nonlinear elastic or plastic material applications are contained in the Bulk Data
entry MATSl, in connection with the MATl entry with the same MID. The MATSl entry
specifies the stress-dependent material properties according to the format shown below. A brief
description of each of the fields is provided, along with default values and further elaborating
remarks as applicable.

MATS1 MID TID TYPE H YF HR. LIMIT1 LIMIT2

MATS1 17 28 PLASTIC 0.0 1 1 2.+4

MID Identification number of a MATl entry (Integer > 0).

TID Identification number of a TABLES1 or TABLEST entry (Integer 2 0). If H is given,
then this field should be blank. See Remark 1.
TYPE Specifies the type of material nonlinearity: NLELAST (Nonlinear elastic) or PLASTIC
(Elastoplastic). See Remarks 2 and 3.
H Work hardening slope (slope of stress vs. plastic strain) in units of stress (Real). For elastic-
perfectly plastic cases, H=O.O (default). For more than a single slope in the plastic range,
the stress-strain data must be supplied on a TABLES1 entry referenced by TID and this
field must be blank. See Remark 3.
YF Yield function (Integer) is selected by one of the following values:
1 von Mises (Default)
2 Tresca
3 Mohr-Coulomb
4 Drucker-Prager
HR Hardening Rule (Integer) is selected by one of the following values:
1 Isotropic hardening (Default)
2 Kinematic hardening
3 Combined hardening
LIMIT1 Parameter representing an initial yield point (Real). See Remark 4.
LIMIT2 Parameter representing the internal friction angle for the Mohr-Coulomb and Drucker-
Prager yield criteria. (0 5 Real < 45'). See Remark 4.
Remarks:

1. If TID is given, TABLES1 entries (xi, y;) of stress-strain data ( E ~Yk)

, must conform t o
the following rules (see Figure 6.1.1):

(a) If TYPE is PLASTIC, the curve must be defined in the first quadrant. The first
point must be at the origin (xl = 0, yl = 0) and the second point (x2, y2) must
be at the initial yield point (Y1 or 2c) specified on the MATS1 entry. The slope of
the line joining the origin to the yield stress must be equal to the value of E on the
MAT1 entry.
(b) If TYPE is NLELAST, the full stress-strain curve (-co < x < co) may be defined
in the first and the third quadrant to accommodate different uniaxial compression
data. If the curve is defined only in the first quadrant, then the curve must start at
the origin (xl = 0, yl = 0) and the compression properties will be assumed identical
t o tension properties.
(c) If TYPE is PLASTIC and TID is given, it may not reference a TABLEST entry.

2. If TYPE is NLELAST, the stress-strain data given in the TABLES1 entry will be used
t o determine the stress for a given value of strain. The values H, YF, HR, LIMIT1, and
LIMIT2 will not be used in this case.

For nonlinear elastic temperature-dependent material properties, TID must reference

a TABLEST entry and TABLEST subsequently references TABLES1 entries for differ-
ent temperatures. Also a MATT1 entry should be supplied to specify the temperature
dependence of material properties.

3. If T Y P E is PLASTIC, either the table identification, TID, or the work hardening slope,
H, may be specified but not both. If the table ID is omitted, the work hardening slope,
H, should be specified in field 5 unless the material is perfectly-plastic. The plasticity
modulus (H) is related t o the tangential modulus (ET) by

where E is the elastic modulus and ET (= g)is the slope of the uniaxial stress-strain
curve in the plastic region. See Figure 6.1.2.

4. LIMIT1 and LIMIT2 are parameters used in the yield function and must be defined as
follows:
 Yield Function LIMIT1 LIMIT2
von Mises (1) Yield stress in tension, Y1 Not used
or Tresca (2)
Mohr-Coulomb (3) 2cCohesion, 2c (in stress units) Angle of internal
or Drucker-Prager (4) friction 4 (in degrees)

Y (or a)

1
H3
Y3 .-; /I
Hz I
k=3
y2- 1
,,<

/
/
I
H1 k=2

,'
AT; I I
II
I

y1 .-' ,
,'
! / k=l I
I I
1 ~f TYPE is PLASTIC:
I I
I I F: = Effective Plastic Strain
I I
j

E II II II I I H k = & Y= kq I - y k

I I

I I
I
I
1

I I
I I
1. . - . . . ? . . - - - - - - . . - - - -
I i
0 El z; E2 F; E3 E

Figure 6.1.l. Stress-Strain Curve Definition with TABLES1 Entry.

 Y (or F)

Figure 6.1.2. Stress-Strain Curve Definition for a Single H.

The parameters in the MATS1 Bulk Data entry pertain only t o nonlinear elastic and plastic
material nonlinearities. The required data for creep analysis are contained in the Bulk Data
entry CREEP, which defines the creep characteristics based on experimental data or known
empirical creep laws. In order t o activate the creep analysis, the CREEP entry must contain the
same MID as an associated MAT1 Bulk Data entry and the NLPARM Bulk Data entry must
have a positive real value in the DT field. The DT value specifies the incremental time interval
for creep analysis. The time unit in the CREEP entry used t o define the creep characteristics
with coefficients a through g must be consistent with the DT time unit. The CREEP entry
format is shown below with default values, followed by a brief description of each field and
applicable remarks.

CREEP MID RT EXP FORM TIDKP TIDCP TIDCS THRESH

CREEP 8 0. 1.E-9 CRLAW

TYPE a b c d e f g

121 6.9853-6 2.444 7.0323-4 0.1072 6.733-9 0.1479 3.

i
MID Identification number of a MAT1 entry (Integer > 0).
RT Reference temperature (To)at which creep characteristics are defined (Real). Default =
0.0. See Remark 1.
AH
EXP Temperature-dependent term, e-m, in the creep rate expression (0.0 < Real 5 1.0).
Default = 1.OE-9. See Remark 1.
FORM Specifies the form of the input data defining creep characteristics: CRLAW (empirical
creep law) or TABLE (tabular input data of creep model parameters). See Remark 2.
TIDKP Identification number of a TABLES1 entry for specification of the creep model pa-
rameter K p ( a ) (Integer > 0). See Remark 3.
TIDCP Identification number of a TABLES1 entry for specification of the creep model pa-
rameter C,(a) (Integer > 0). See Remark 3.
TIDCS Identification number of a TABLES1 entry for specification of the creep model pa-
rameter C,(a) (Integer > 0). See Remark 3.
THRESH Threshold limit for creep process corresponding t o the threshold strain (0.0 < Real
< 1.OE-3). Default = 1.OE-5. Threshold stress under which creep does not occur is
computed as THRESH*E, where E is the Young's Modulus.
TYPE Specifies the identification number (three-digit integer) of the empirical creep law type
(111 5 Integer 5 300). See Remark 4.
a-g Coefficients of the empirical creep law specified in the TYPE field (Real). See Remark 4.

Remarks:

1. Creep law coefficients (a-g) are usually determined by least squares fit of experimental
data obtained under a constant temperature. The reference temperature (To)at which
creep behavior is characterized must be specified in the RT field if the temperature of
the structure is different from this reference temperature. The unit of the temperature
input (OF or OC) must be indicated in the PARAM entry TABS as follows:

PARAM, TABS, 273.16 (if Celsius is used)

PARAM, TABS, 459.69 (if Fahrenheit is used)
When the correction for the temperature effect is required, the temperature distribution
must be defined in the Bulk Data entries (TEMP, TEMPP1, and/or TEMPRB) which
are selected by the Case Control command, TEMP(L0AD) = SID, within the subcase.

From the thermodynamic consideration, the creep rate is expressed as

 where A H is the free energy of activation
R is the gas constant (= 1.98 cal/mole OK)
T is the absolute temperature
A is strainlsec per activation

If the creep characteristics are defined at the temperature To (RT field), the creep rate
at the temperature T is corrected by a factor

AH
where ( e - m ) is t o be specified in the EXP field.
2. If FORM = CRLAW, the creep law type and the coefficients (a-g) must be specified in
the CREEP continuation fields. If FORM = TABLE, the creep model parameters (Kp,
Cp, and C,) are to be specified in the TABLES1 entries whose identification numbers
appear in the TIDKP, TIDCP, and TIDCS fields, respectively. In this case, the CREEP
continuation field should not be used.
3. The creep model parameters (K,, Cp, and C,) represent parameters in the uniaxial rhe-
ological model shown in Figure 6.1.3. Tabular entries (xi,y;) in the TABLES1 entry will
consist of (a;, Ii,;), (a;, Cpi), or (a;, C,;) for the input of 14, Cp, and C,, respectively.
In the case of a linear viscoelastic material, the parameter (hrp, C,, and C,) values are
constant and two data points with an identical value should be specified for each param-
eter.

The creep model parameters must have positive values as shown in Figures 6.1.4-6.1.6. If
the table look-up results in a negative value, the value will be reset t o zero and a warning
message (TABLE LOOK-UP RESULTS IN NEGATIVE VALUE OF CREEP MODEL
PARAMETER IN ELEMENT ID = ****) will be detected.
4. Two different classes of empirical creep laws are available at present. The first class of
creep law is
+
~ " ( at), = A(u)[l - e-R(u)t] h'(u)t
Oak Ridge National Laboratory recommends the parameters A(u), R(u), and K ( u ) in
the form given in the following table:

Parameter Type 1 Type 2

40) aub aExp(ba)

R(a) cExp(du) cud

K(u) e[sinh(f u)]g eExp( f a )

Each of the three digits (e.g., 111, 121, etc.) in the TYPE field selects type 1 or 2 for
the parameters A(u), R(u), and K ( u ) , respectively.
 • he second class of creep law (TYPE = 300) is expressed as

where the values b and d are limited t o the ranges

and
0.2 < d < 1.0
T h e coefficient g should be blank for creep law type xx2 material, and coefficients c, e,
f, and g should be blank for creep law type 300. T h e coefficients (a-g) are dependent on
the structural units. Caution must be exercised t o make these units consistent with the
rest of the input data.
5. Creep analysis requires an initial static solution a t t=O, which can be obtained by a
subcase with an NLPARM on which DT=O.

Primary
Elastic

Figure 6.1.3. Uniaxial Rheological Model.

 Figure 6.1.4. Creep Model Parameter I(, as a Function of Stress o.

Figure 6.1.5. Creep Model Parameter C, as a Function of Stress u.

Kips-hours
cs
in3

a(ksi)
Figure 6.1.6. Creep Model Parameter C, as a Function of Stress 0.
6.2 PLASTICITY

Two broad categories of material nonlinearity are nonlinear elastic analysis and plastic analysis.
In both cases, the stress-strain relation is nonlinear; however, the unloading follows the stress-
strain curve in the case of nonlinear elastic analysis, whereas elastic unloading takes place in
plasticity.

The plasticity is characterized by its path dependency. The strain is not a function of the
current state of stress alone, but depends on the previous loading or stress history. This is
exemplified by the simple case of zero stress, when permanent sets of differing magnitude can
be established by varying histories in which the stress starts and finishes at zero.

6.2.1 Some Preliminaries

Some of the commonly used terms in plasticity [6.1, 6.21 are given below.
Yield Stress: Yield stress is usually measured as the value of stress which produces the
smallest measurable permanent strain.

Yield Criteria: In simple tension, there exists a yield point at which the material will begin
to deform plastically. However, if the stress state a t a point is not uniaxial but consists
of stress components in different directions, a criterion is required t o define which com-
bination of multiaxial stresses will cause yielding. Such criteria are called yield criteria.
The first step of any plasticity analysis is to decide a yield criterion.

Yield Surface: Any yield criterion is expressible in the form

where f(a)is a function of the stress state and Y ( K is

) a function of the strain hardening.
Geometric representation of the yield function is a surface in the stress space (al-a;!-as
coordinates). This surface is called the yield surface.

Loading and Unloading: If the stress is changed so that the stress tends t o move out of the
elastic region, the process is called loading. If the stress value is changed such that the
stress value moves into the elastic region, the process is called unloading.

Proportional Loading: Loading condition under which all the stress components increase
in the same proportion.

Strain Hardening: When the material undergoes plastic deformation or change in strain, the
yield surface expands and/or translates with increasing strain. This is known as strain
hardening or work hardening and the material which does not exhibit strain hardening
is called perfectly-plastic material.
Bauschinger Effect: When a plastically deformed specimen is unloaded, residual stresses on
a microscopic scale remain and influence the plastic yielding for the different loading.
If the previous strain was a uniform extension and the specimen is then reloaded in
compression in the opposite direction, it is observed that yielding occurs a t a reduced
stress. This is known as the Bauschinger effect.
The phenomenon of the Bauschinger effect can be described with reference t o Fig. 6.2.1,
which shows an idealized stress-strain curve of a ductile material: first deformed by uniform
tension, the load removed, and reloaded in compression. According t o the kinematic hardening
model (which is one extreme viewpoint), it is assumed that the elastic unloading range will
be double the initial yield stress. If the initial yield stress in tension is ay and the specimen
is loaded up t o a stress a1 and unloaded, the plastic yielding will begin in compression at
stress a;!given by a2 = a1 - 20,. At the other extreme is the model that shows the hardening
mechanism acting equally in tension and compression. Thus compressive yielding will occur
when a2 = - 0 1 . This model shows no Bauschinger effect. This is the simplest of the theories
to apply and is most frequently used. This behavior is called the isotropic hardening rule.
Between these theories there are models that compromise the attributes of the kinematic and
isotropic hardening for the onset of compressive yield after tensile hardening. This behavior
can be modeled in MSCINASTRAN using the combined hardening option.

Figure 6.2.1. Representation of Hardening Rules

6.2.2 Yield Criteria

MSCINASTRAN provides four different yield criteria [6.3 - 6.61: von Mises, Tresca, Mohr-
Coulomb and Drucker-Prager. The von Mises criterion is most commonly used in plastic
analysis of ductile materials. The Tresca criterion is more suitable for brittle materials. The
criteria by Mohr-Coulomb and Drucker-Prager are suitable for frictional materials such as soil
and concrete.

The von Mises yield criterion states that yielding begins when the elastic distortion
energy (or second invarient of deviatoric stress, Jl) exceeds a critical value. In simple tension,
this critical value is related to the yield stress (Y), i.e.,

where
1 1
J'
2
= -O!.O!.
2 23 23
= -[( + (02 -
6 01 - 0 2 ) ~ +
03)~ (03 - 01)~]. (6.2.2)

Geometrically, this equation represents a right circular cylinder perpendicular to the oc-
tahedral ( n ) plane. This cylinder is the von Mises yield surface, as shown in Fig. 6.2.2. The
left-hand side of Eq. (6.2.1) may be expressed in terms of the general stress state, i.e.,

where a is termed the effective stress. For pure shear of the plane stress case,

and, from Eqs. (6.2.1) and (6.2.3),

V

where Y is the yield stress of the material, i.e., yield stress in pure shear is 1
.times the yield
6
stress in simple tension.
Figure 6.2.2. Geometric Representation of von Mises Yield Surface.
 The T r e s c a yield c r i t e r i o n (sometimes referred t o as the maximum shear theory) assumes
that yielding will occur when the maximum shear stress exceeds the value of the maximum
shear stress occurring under simple tension, i.e.,

where a1 and a3 are principal stresses such that a s < a 2 < a l . Geometric representation of
+ +
this equation is a hexagonal surface perpendicular t o the .rr plane (al 0 2 a3 = 0) in the
stress space.

For frictional materials, the cohesion c and the internal friction angle 4 are the governing
properties. The yield criterion for such materials is given by M o h r - C o u l o m b as

where r is the magnitude of the shear stress,

a, is the normal stress,
c is the cohesion, and
4 the angle of internal friction. (0 <
4 < 45')

Graphically Eq. (6.2.7) represents a straight line tangent t o the largest Mohr stress circle as
shown in Fig. 6.2.3. In terms of principal stresses (0 > a1 2 a 2 _> as), Eq. (6.2.7) becomes
1
-(a1 - as) cos 4 2 c - (-
a1 + + --
a3 - a1 03
- sin 4 ) tan 4
2 2 2
which may be reduced t o
a1 - a3 + (al + as) sin q5 = 2c cos 4 (6.2.9)
This may be reduced t o Tresca's criterion with 4 = 0 and Y = 2c.
An approximation t o the Mohr-Coulomb yield criterion was presented by Drucker and
Prager, known as the D r u c k e r - P r a g e r yield criterion, which is a modification of the von
Mises yield criterion, i.e.,
3a0, + ( J : ) ~ ' ~= K (6.2.10)
where
2 sin 4
a=
&(3 - sin 4)
6c cos 4
K =
&(3 - sin 4 )
This yield surface has the form of a circular cone coinciding with the outer apices of the Mohr-
Coulomb hexagonal cone at any section. Rearranging terms of Eq. (6.2.10), the Drucker-Prager
yield function is given as
6 sin 4 6c cos 4
3 - sin 4
+
am &( J ; ) ~ ' ~>
- 3 - sin4
 which is reduced to von Mises criterion for 4 = 0.

I - (a, + a312
- 01
-
7

Figure 6.2.3. Mohr Circle Representation of t h e Mohr-Coulomb Yield Criterion.

6.2.3 Yield Function

Associated with the yield criteria, there exist yield functions in the form of

where f (a) is an effective stress, a function of the stress state

and Y ( P ) is a tensile yield stress, a function of the strain hardening.

Geometric representation of the yield function is a surface in the stress space (al - a2 -
. a3 coordinates), which is called a yield surface. The von Mises yield surface is a circular
cylinder and Tresca's is a hexagonal cylinder. The yield surface is reduced to the yield locus
in the case of plane stress, as shown in Fig. 6.2.4 and 6.2.5 for von Mises and Tresca criteria.
respectively. Both the Mohr-Coulomb and Drucker-Prager yield functions represent conical
surfaces as shown in Fig. 6.2.6.
Figure 6.2.4 Von Mises Yield Locus for Plane Stress.

02

1 02 = fly
0 , ,,,, ...... 4 ... .-.;

.. - ,.,'
,/,' 1
/

u2 - 01 = uy ,.'
,.'
,.' ;-u1 = uy
,.'
/,"
,/'
.......... . . . . . . . .-. ... . . . .. ......
. ... . . . . . . . . . . . . . . . . . . . . .
-uy ,,,,"by
,,/',
,,/'
-u1 = uy +-/ ,.' +--- 61 - 02 = uy
j : ,,"/'
I ,
/

7.....L
. . +! ,.'
-uy
m2 = uy
!

Figure 6.2.5 Tresca's Yield Locus for Plane Stress.

6.2 - 7
 Drucker-Prager @ > 0

u r = q = q

J- at (a) von Mises and Drucker-Prager

Mohr-Coulomb @ > 0
(1882) (1733)

Space
 The stress function f ( a ) represents the effective stress or equivalent stress, which is equiv-
alent to the uniaxial stress in tension. The value of f ( a ) can be computed as follows:

i) For von Mises criterion

1 112
f(4 = fi(~2)
ii) For Tresca's criterion
f(o) = 01 - 03

iii) For Mohr-Coulomb criterion

iv) For Drucker-Prager criterion

6 sin 4
f ( ~ .=
) 3 - sin 4 a m + A(J;)~'~
The function Y ( P ) is a measure of the tensile yield stress. The value of Y ( P ) can be found
from the user-specified stress-strain curve by an interpolation, i .e.,

with
c; = &k
Yk
- -
E

, is a data point such that Z: 5 P < z:+~, Yl is the initial yield point, and ,G' = 0.)
where ( E ~Yk)
0.5 or 1. (for kinematic, combined, or isotropic hardening, respectively). Since Yl = 2c for
frictional materials, it is obvious from Eqs. (6.2.9) and (6.2.11) that Eq. (6.2.16) must be
modified by multiplying correction factors as follows:

Y ( P ) = cosq5 * Y(P) for Mohr-Coulomb (6.2.18)

3 cos 4
Y(P) =
3 - sin 4
* Y(P) for Drucker-Prager

Notice that Y > 0.0, otherwise the program issues a user error message: "Implausible stress-
strain curve in EID=xxxn.
6.2.4 Strain Hardening

The yield surfaces discussed previously correspond t o the yield surface of a material when
yielding starts. For a material that strain hardens, the yield surface must change for continued
straining. Such a change of the yield surface is governed by the hardening rules.

The isotropic hardening theory, due to Hill, assumes that the loading surface expands
uniformly about the origin in the stress space during plastic flow, maintaining the same shape,
center and orientation as the original yield surface. Fig. 6.2.7 shows the initial and subsequent
yield surfaces on a two-dimensional plot when the stress state shifts from point 1 to point 2.
Unloading and subsequent reloading in the reverse direction will result in yielding at the stress
state represented by point 3. The path 2-3 will be elastic and 0-2 is equal to 0-3. The main
drawback t o this theory is its inability to account for the Bauschinger effect exhibited by most
materials. The equation for subsequent yield surfaces is given by

where Y ( P )is a measure of the expansion of the yield surface in the stress space, and the
total effective plastic strain (3')is a monotonically increasing scalar function during plastic
deformation.

The kinematic hardening theory, due t o Prager, assumes that the loading surface trans-
lates as a rigid body in stress space during plastic deformation, maintaining the size, shape and
orientation of the yield surface (see Fig. 6.2.8). Thus, this theory accounts for the Bauschinger
effect. The yield function for kinematic hardening is expressed as

where a;j represents the translation of the center of the yield surface, and Y is assumed
constant. The theory as set forth by Prager postulates that the incremental translation of the
loading surface in the stress space occurs in the direction of the exterior normal to the surface
at the instantaneous stress state. This relation is expressed analytically by

where c is a scalar parameter called the hardening modulus, and d&Yj is the increment of plastic
strain.

However, inconsistencies arise when Prager's theory is applied in various subspaces of stress.
In other words, the loading surface will not, in general, translate in the direction of the exte-
rior normal in the stress subspace with the full nine dimensional stress space. To avoid this
difficulty, Ziegler has proposed a modification of Prager's rule as

The scalar d p in Eq. (6.2.23) is determined from the condition that the stress state must remain
on the translated loading surface during plastic deformation. Substituting Eq. (6.2.23) into
the total differential of Eq. (6.2.21),

and solving for d p ,

Z d c ,
dp =
( o . .- a . . 2 i
'3 a31

A theory that combines kinematic and isotropic work hardening states that

for which d p is found t o be

-do -4HdP
dp = a.f(~-~.)
(0- a ) aa
where H is the instantaneous slope (g)
of the uniaxial stress-strain curve and = 0.5.
Eq. (6.2.25) reduces t o kinematic hardening when Y ( P )is a constant and t o isotropic hardening
when a;j = 0 and Y ( P )is monotonically increasing.

The specification of the initial yield surface (by means of a choice of the yield criterion)
defines the stress state at which plastic flow will begin. The specification of a hardening rule
defines the stress state for subsequent yielding.
 Figure 6.2.7. Isotropic Hardening.

Loadi ng Curve

Y i e l d Curve

Figure 6.2.8. Kinematic Hardening.

6.2.5 Prandtl-Reuss Stress-Strain Relations

One of the major differences between stress-strain relations in the elastic range and those in
the plastic range is that strains in the plastic range are, in general, not uniquely determined by
the stresses but are dependent on the past history as well. Due to the path-dependent nature
of the plastic strains, increments of plastic strain are calculated by an incremental procedure
and the total strains obtained by integration or summation.

After Saint Venant proposed that the principal axes of the strain increment coincided with
the axes of principal stress, a general relationship between components of the strain increment
and stress was established by Levy and von Mises, independently [6.1]. Levy-Mises equations,
however, use the total strain increments and are applicable only t o rigid-plastic materials.
Levy-Mises equations were modified t o allow applications for elasto-plastic materials by Prandtl
and Reuss. These equations are known as the Prandtl-Reuss stress-strain relations, i.e.,

where
1
+
- [(ox- uy)2 ( u y - uz)2
2
+ (0, - + 3 ( ~ ,+2 ~ +
T ; ~ T,"~)

and

These stress-strain relations are consistent with the von Mises yield criterion. For other
yield criteria, a general derivation follows.

6.2.6 Associated Flow Rule

Based on the condition that the principal axes of the plastic strain-increment must coincide
with the principal stress axes, and defining a function of stress components ( a i j )as the plastic
potential function (Q), the plastic strain-increments can be expressed as

where dX is a Lagrange multiplier. For a so-called stable plastic material which has a stationary
value of the plastic work done in a given increment, such a function (Q) exists and is assumed
identical to the yield function (F),i.e.,

where F is given by Eq. (6.2.25). This is called the associated flow rule. The geometric
implication of this relationship is that the plastic strain-increment vector is normal t o the
yield surface, known as normality principle.

During the incremental loading, the incremental strain is assumed t o be equal t o the sum
of the incremental elastic and incremental plastic strains, i.e.,

But
{d&Ie= [D,]-'{do)
with [D,]
being the elasticity matrix;

and upon substituting Eqs. (6.2.30) and (6.2.32) into (6.2.31), it is found that

{ d ~ =) [D,]-' { d a ) + dX{ -}dd Fo

When plastic yield is in progress, the stresses are on the yield surface given by the yield
function, which, when differentiated, provides the expression

where K (work-hardening parameter) can be a;j or P , whichever case it may be. Then
Eq. (6.2.34) may be conveniently expressed as
dF
- d o ) - H*dX = 0
{ao}
 8F 1
H* = --drc-
arc dX
where H* has a physical implication which will be shown later.

After a few matrix manipulations t o eliminate {do) from Eqs. (6.2.33) and (6.2.35), the
scalar multiplier dX can be obtained as

where {s) is reduced to { g } and

of dX from Eq. (6.2.33) gives
, the terms H* and 1%)are functions of {a}. Elimination
{do) = [DepI{d~) (6.2.38)
with the elasto-plastic matrix defined as [6.5]

6.2.7 Generalized Effective Plastic Strain Increment

The effective plastic strain increment can be defined using the work-hardening hypothesis.
Work-hardening is postulated as the amount of work done during plastic deformation, i.e.,

According t o the Euler's theorem, if the function f(aij) is homogeneous and of degree one,
e.g.7
f (Xaij) = Xf (aij) (6.2.41)
then the following equality exists:

All four yield criteria, provided in MSCINASTRAN, satisfy Eq. (6.2.41).

By virtue of Euler's theorem, Eq. (6.2.40) may be restated as

The stress-strain data from the uniaxial tension test (in x-direction) may be related t o Eq. (6.2.43)
by considering the work done by the plastic deformation during the test, i.e.,
When the effective stress ( a ) corresponds to the stress under uniaxial tension (Y), the corre-
sponding effective strain increment must be equivalent t o dX by equating corresponding terms
of Eqs. (6.2.43) and (6.2.44). Hence,

and dX is defined by Eq. (6.2.37). In fact, it can be shown that dX in Eq. (6.2.37) is identical
t o d P as defined in Eq. (6.2.28) with the von Mises yield function. It is necessary to clarify H*
in Eq. (6.2.36) using Eq. (6.2.45). Combining Eqs. (6.2.30), (6.2.35) and (6.2.45), Eq. (6.2.36)
can be restated as

It is revealed that H * is identical t o H which is the slope of the uniaxial stress-strain curve, F.
dY

This is a natural consequence of the work-hardening, which should follow the simple tension
test data regardless of the hardening rule.

6.2.8 Yield Function Derivatives

The derivatives of the yield function with respect t o stress components are frequently required
in the computation of plastic strains. This frequent computation is conveniently simplified by
adopting a unified approach to the various yield criteria. All the yield functions are expressed
in terms of stress invariants [6.6], and the gradient vector is obtained by differentiating the
yield function with respect t o these invariants first. This unified approach also facilitates t o
avoid the difficulty of singularities in the Tresca and the Mohr-Coulomb yield criteria.

The principal stresses can be expressed in terms of stress invariants as follows:

where

and

with
and
J; = aLaha: + ~ T ~ -~ T + a & ~ , +2 ~
~ ~ T
1
o~ ~
2
~ ~
T ~ )

Substitution of this expression into Eqs. (6.2.13) and (6.2.14) results in the Tresca and Mohr-
Coulomb yield criteria in terms of stress invariants:

f ( a ) = 2 (J;)'I2 cos 0 (6.2.13~)

and

f ( a ) = 2 a , sin 4 + 2 ( J 2 )112[cos 0 -
1 1 (6.2.14~)

The gradient vector is computed by differentiating the stress function f ( a ) representing

effective stress for each yield criterion as follows:

with
a f ae
-a f- - -- af af af ae
a ~ ; aea~, and
aJ;
-
a ~ +; --.
aeaJ;
The gradient vectors of the stress invariants are found independent of yield criteria as follows:

1
= - < 1 1 1 0 0 0 >

= < a Ix a& o: 2rXy 2rYz 2~~~ >

= 2
< ( a & d - rYz + J;
(&: -
2
T~~ + 3), J::

The derivatives of the effective stress with respect t o the stress invariants are found as
below:
i) Tresca's criterion:

af
- -
& sin 0
a J; J; cos 30
ii) von Mises' criterion:

iii) Mohr-Coulomb criterion:

-'f
darn
-
- 2 sin 4

a
-f -
cos 6'
-[(I + tan 6' tan 36') + 1
- sin $ (tan 36' - tan 6')]
aJ: @ fi
af
- -
-
sin 6' + sin $ cos 6'
a J; J; cos 39
iv) Drucker-Prager criterion:

Notice that when 0 = 30°, representing corners of the Tresca and Mohr-Coulomb yield
surfaces, it is not possible t o compute

af and a
f
-
a J; a J;
This difficulty arises because the gradient vector cannot be uniquely defined at these corners.
To prevent such difficulties, a provision is made t o use the expressions for von Mises and
Drucker-Prager yield criteria when 101 2 29'.

6.2.9 Degenerate Cases: Plane Strain, Plane Stress and Uniaxial Stress

In the case of plane strain, the general expressions derived so far must be modified with

and
E~ = = 0
but
Ae; + A&: thermal -
= A&, - A&, - -OAT
The deviatoric stress invariants can be reduced t o

J2r = -1 (ai2
2
+ oL2 + 0:') + rXy
2

2
Jh = at1 (a,a,
1 1
- r,,).

The gradient vector ( 2 )will consist of four components < 2.L

aux 2.L
auy 2.L 2.f-
au, aTxy >, which can
be computed with

{*jT = 3 < 1 1 1 0 >

1
{$jT = < a: ah a: 2r,, >
{%jT = < (aha: + $),(060: + $), (oLo; - r2, + J'
+),-2o:r,, >
The elasticity matrix for the plane strain case is

In the case of plane stress, the equations can be reduced with

but
Ae, = AE: + Aef: # 0
and
Aef: = -A&: - Aei.
Notice that E; does not have to be stored as in the plane strain case. The deviatoric stress
invariants can be computed with a, = 0. The elasticity matrix is reduced to:

In the case of uniaxial loading, all the stress components become zero except a,. The
deviatoric stress invariants become simply

J21 = 1- gz
2 and J31 = 2-a3: 3
3 27
The effective stresses become

f ( a ) = ABS lax1 for von Mises and Tresca

f ( a ) = ABS lax) + a, sin 4 for Mohr-Coulomb

2 sin 4
f (0) = ABS lox1 + 3 - sin 4 a x for Drucker-Prager

The gradient vector can be reduced to a scalar, i.e.,

for von Mises and Tresca

for Mohr-Coulomb

for Drucker-Prager

The elasticity matrix is reduced t o a scalar:

6.2.10 Solution Algorithm for Elasto-Plastic Material

The material routine interfaces with element routines with the following 110 data:
r Input: P, {a},{a}, {A&}

Output: Gew,{anew}, {anew}, [De,], a

The governing equation for the incremental process of elasto-plastic deformation is as follows:

with

where H and {g)are functions of {a}.

The computational process in the plasticity routine is depicted in the flow diagram in
Fig. 6.2.9. The procedure can be summarized in steps as follows:
Step 1. Initialize
Zero = * Yl, E? = p,
{ a } = {a} and Dep = De
Step 2. Estimate the stress increment using the elasticity matrix.

Then the trial stress is

{at} = {(. + {Aae}
Step 3. Translate the yield surface for kinematic or combined hardening:

Step 4. Find the current yield stress (Y) from the input stress-strain curve as follows:

i) Increment lc until C: 5 9' < where E: = - %.

ii) Compute the current plasticity modulus:

iii) Compute the current yield stress:

Y = Yl + p (Yk - Yl) + /3 Hk ( P - E;)

where Yl is the initial yield point and P = 0.) 0.5, or 1. for kinematic, combined, or
isotropic hardening, respectively.
iv) Make corrections for the case of Mohr-Coulomb or Drucker-Prager criterion.
Step 5. Estimate the yield function:

Step 6. Determine whether the plastic deformation is involved.

If Fl 5 0, the process is elastic. Return with:

Step 7. Determine whether the material is yielding.

If lFol 5 0, the material is on the yield surface. Go t o Step 10 with

{ali = {ao) and y = 0
Step 8. Check the error.

If Fo > 0, exit with a message "Effective stress is greater than yield stress in element
ID =XXY" .

Step 9. Bring the stress state t o the first yield point, if the material is elastic initially.

If Fo < 0, material goes to plastic from the elastic state. Update the stress compo-
nents t o place the stress state on the yield point.

Fo
{a}; = (00) + 7 {AaeI where y = -
Fl - Fo

Step 10. Find the number of incremental steps required as follows:

m = Int. ( F s T R EF1
ss * Y) + I

and
2 < m < LOOPCT
where FSTRESS is a user parameter in the NLPARM data entry and is defaulted t o 0.2.
Step 11. If m indicates excessive increment size, set the flag t o activate the bisection in order
t o reduce the load or time increment, which is based on

where bisection is activated if M > 10. Prior t o Version 66, a warning message was
issued instead of bisection: "Excessive incremental load is applied in element ID=xxx.
Reduce the load increment by a factor of M for better solution."

Step 12. The governing equation will be integrated by the Euler method of step by step in-
tegration in m subincrements.

The incremental stress (equivalent t o the equally divided subincrement of Ae) is

where {Aa,} = De {A&).

Step 13. Repeat the following loop m times.

i) Compute { g J i a n d Hi based on {a); and q

ii) Find
 iii) Compute

iv) If AX; < -5 x set the unloading flag up (notice that AX = 0 by the end of
this loop) and eliminate negative AX, i.e.,

AX; = MAX(AX;, 0.0)

v) Compute the stress subincrement

{Au), = {Au) - AX; {d);

vi) If the hardening rule is not isotropic

Ap =
{%IT {Au); - p H AX;
(2): (0);

and
A p = MAX(&, 0.0)
{Aa); = Ap{a);
{Au}; = {Aa); - {Act};
{a);+1 = {a); +' {Aa);
vii) Update the stress and the plastic strain

and
{ E ~ )= { E ~ )+ AX;
($1;
Step 14. Error Correction step

i) Compute the error: 6 = F3 = f (a;+,) - Y(Z:+,)

ii) If 161 FSTRESS * Y in the converging state, EXIT with a fatal error message, "Error
exceeded xx% of current yield stress in element ID=xxx".
iii) If 161 < 0, ignore the error and jump t o Step 15.
 iv) If S 2 0, make a correction as follows [6.7]:

S
{a>;+, = { ~ } i + m- -{a>;+m
Y

Step 1 5 . Offset the center of the yield surface.

Step 16. If the unloading flag is up, the material is assumed to be unloaded. Since the
material becomes elas tic during unloading, return with the elastic tangential matrix, i.e.,

Step 17. Otherwise, return with the elasto-plastic tangential matrix:

 Initialize
Estimated stress increment
Zero = i~-'* Y i
-p -P {ME)= DE{&€>
€, = E
Trial stress {a?) = {a) + {ME>
{<>I = {<Il DIP = DE
r....'.....'---"----.----..--.-----....-----..----..--. ......................

Find current yield stress Y

ICompute
Fe = f(o,)
Fl = f(ai)
-Y
-Y I
Return

I
I

Haterial on Yield Surface Yes Fatal Error I

{@I = {a,}, 7= 0

L
i
Number of incremental steps I

Elastic to Plastic Iransition I

1
F1 4
fi = Int(-
FSTAESSW
1' <a11 = <a,> + r{hr> I
I

Fe
-
I

with7 = - I

Fi - Fe I
I
I
I
I

No '
t 2 5 m ! Min(i0, LOOPCI) I
I

Incremental Stress I

1 I

{ha> = 4 1 - r){&r> I

m I

Uarning ! B
I

Message I
I
-----.---..--...-...-..--....---.....---..----..--#\....--..--...-.-.-.---..--
1

Figure 6.2.9a Flow Diagram for Elasto-Plastic Material (NMEP)

 Based on {a} and €,

Cofipute {a), = (-3, and HI

w Find {dl,

Compute AXI :
:DE{~)I

{a):{&)
HI + {a):{d)I
-

Set Unloading
Yes Flag
+ AX, = 0

I Stress Incre~ent I I

h Update

Figure 6.2.9b Flow Diagram for Elasto-Plastic Material (NMEP)

 Compute error
4 = -P
F3 = f(gI+,,) - Y(EI+,)

Fatal Error
lbl FSlRESSwY
)

and CONU=l

Offset yield center

{0II+"= {o}~,,,+

[ Yes
I

Figure 6 . 2 . 9 ~ Flow Diagram for Elasto-Plastic Material (NMEP)

6.2 - 27
6.2.11 Verification and Validation

The elasto-plastic material capabilities were verified with respect t o the algorithm, accuracy,
and programming errors with a wide variety of problems in MSC/NASTRAN. All types of
elements (1-D, 2-D and 3-D) have been tested and known solutions have been reproduced [6.3,
6.41. Some of the example problems of interest are illustrated below:

Uniaxial Loading and Apparent Poisson's Ratio

Experiments have shown that the volume of material does not change due to plastic strain.
This incompressibility is signified by a Poisson's ratio of 0.5 in the Prandtl-Reuss stress-strain
relations. The lateral dimension of the specimen shrinks during the simple tension test as
shown in Fig. 6.2.10(a). The Poisson's ratio is defined as v = -E,/E, in the elastic domain.
If the Poisson's ratio is to be measured from the elastic-plastic deformation (termed apparent
Poisson's ratio), a transition would be observed from an initially elastic (v) t o a smeared value
with 0.5 due to plastic strain. The strain components in simple tension for the elastic-plastic
deformation can be decomposed into elastic and plastic parts, i.e.,

&a = E: + E$ ; in tension direction

and
E, = -VE; - 0.5~: ; in width direction
The apparent Poisson's ratio then becomes

with

The variation of v* with respect to P is plotted in Fig. 6.2.10(b). It is shown that the
apparent Poisson's ratio v* is a continuous and smooth function of ,f3 with a value of v for
,f3 = 0 and asymptotic t o 0.5 as ,f3 approaches infinity. However, P has a finite value limited by
material properties (E and H ) , i.e.,

lim -
L l

Pma, = *+m H(P)

A finite element simulation of a simple tension test by MSCINASTRAN resulted in data points
circled in Fig. 6.2.10(b).
 (a) Uniaxial Tension Specimen

U
~LD-----------------------
n
u

Analyticd
0 MSC/NASTRAN Data Points
+'
C
: -
i
iu
.

(3 FJ
CL
, ,

I I I I I I I 1
'9
0 2 4 6 8 10 12 14 .I 6
Ratio of P l a s t . i v E last i c Strain

(b) Apparent Poisson's Ratio

Figure 6.2.10 Uniaxial Loading and Apparent Poisson's Ratio

Cyclic Loading and Hysteresis

If the stress state is changed such that the stress value moves into the elastic region, the process
is called unloading. The cyclic loading, involving successive loading, unloading and reloading,
makes the problem more complicated due t o a phenomenon known as the Bauschinger effect.
A finite element simulation of the cyclic loading has been performed with a thin plate specimen
subjected t o repetitive uniaxial tension and compression. The loading cycle was controlled by
strain using the enforced displacements. The stress-strain responses of the material obeying
the von Mises yield criterion are shown in Fig. 6.2.11(a) using the isotropic, kinematic and
combined hardening rules. During the first tensile loading phase shown in the first quadrant
(0 to A), the responses for the three hardening rules are identical t o the original strain-strain
curve specified in the input data. The kinematic hardening exhibits the Bauschinger effect upon
reloading in the reverse direction, while the isotropic hardening shows a monotonic increase in
the yield stress. The combined hardening curve falls in between the isotropic and kinematic
hardening curves, as expected. The plasticity modulus is kept continuous upon reloading for all
three hardening rules, i.e., an identical modulus is resumed upon reloading from the previously
loaded point. Note that not all the flexure points are represented in the plot because of discrete
solution points. The solution process progressed without any difficulty caused by unloading.

Kinematic hardening has an advantage over isotropic hardening for the Bauschinger effect.
However, kinematic hardening only describes a perfect kinematic hardening material for which
the size of the yield surface does not change. In reality, such a material does not exist. Suppose
a material has a low initial yield stress and a high rate of hardening, as some materials may
possess. If kinematic hardening is used t o model such a material exactly, the reloading in
compression could occur while the applied stress is still in tension as shown in Fig. 6.2.11(b).
This behavior is physically unrealizable. In order t o avoid such an anomaly, the initial yield
point may have t o be artificially elevated. The difficulty of such problems lies in the fact
that the Bauschinger effect is known as a qualitative and not a quantitative description. The
combined hardening rule is complementary for such cases.
 u, ( k s i )
+ I ~ J L ~ U lc
-*- t i o r d e r ll,y
17ernat li Harder l n g
- - + - ~ - ~ ~ i n= el a
l H a r d e r incj
m

( a ) Stress Strain Hysteresis

(b) Anomaly in Kinematic Hardening

Figure 6.2.11 Strain-Controlled Cyclic Loading

6.3 NONLINEAR ELASTICITY

The nonlinear elastic capability was developed for Version 62 to predict the multi-axial stress-
strain behavior for the nonlinear elastic material, for which only the simple tension stress-
strain data are available. The theory and algorithm are adequate to trace the stress-strain
curve accurately for the uniaxial loading cases. However, the theory is not based on the
classical theory of finite elasticity [6.10]. Consequently, some of the constitutive relations may
be violated in the multiaxial stress cases. Nevertheless, this capability can be justified by
stating that the design is not intended for the large-strain deformation.

The current nonlinear elastic capability, modified in Version 65, can accommodate bilateral
properties (uniaxial tension-compression) accurately for the uniaxial deformations. This design
is considered adequate for the multiaxial deformation if the strain remains in the small range.
Intrinsic deficiency due t o oversimplification of the theory will introduce errors, but the error
may not be noticeable unless large strains are applied in the multiaxial deformation. For large
strain applications the generalized hyperelastic material based on the classical theory of finite
elasticity is being planned to be implemented.

6.3.1 Theoretical Basis

The nonlinear elastic capability in MSCJNASTRAN was designed t o satisfy the equivalence
of the deformation work per unit volume in the simple tension t o the strain energy per unit
volume (conservation of energy), while the work done for deformation may be defined by a
stress-strain curve in simple tension, i.e.,

It was further assumed that the effective strain (E) may be defined by

From the total differential of Eq. (6.3.2), we obtain

Substituting Eq. (6.3.3) into Eq. (6.3.1), stresses may be expressed in terms of total strains,
i.e., -
0
( 0 ) = EIDeI{~) (6.3.4)
The tangential matrix for such material may be obtained by differentiating Eq. (6.3.4), i.e.,,
6.3.2 Solution Algorithm

Users specify nonlinear stress-strain curve, a(<), in the TABLES1 entry along with MATl and
MATS1 entries. Young's modulus and the Poisson's ratio are available from MATl entry. The
element routine calls the material routine driver NMATD, which calls the subroutine NMEVD
for the nonlinear material. The material routine interfaces with element routines with the
following data:

The computational procedure is described below:

Step 1. Upon entry t o the subroutine NMEVD, the new strain state is computed by

Step 2. The effective strain ( 2 ) is computed based on {E),,, by

where

or
1
g2 - - [E:
1 - v2 + E: + + 1-v
~ V & X E ~-I-~xy]
2
for plane stress

+ for 3-D Solid

2

Step 3. The effective stress (a) is determined by looking-up the user-specified stress-strain
curve for E.

Step 4. The new stress state is determined by

Step 5. The tangential matrix is determined by

 for which $$ is the slope at E, i.e.,

where (xk,yk) is the k-th data point in the TABLES1 entry and k is determined such
<
that xk E < xk+1. It is noted that k = l for E < xl and k = k,, - 1 for E 2 x,,,.

6.3.3 Adaptation of Uniaxial Compression Stress-Strain Curve

Until Version 64, the effective stress-strain curve is determined only by the uniaxial tension data
in two and three dimensional cases. The uniaxial compression data will be ignored if provided
for two and three dimensional elements. Some materials, however, exhibit appreciably different
behavior in compression from that in tension even in the small strain range. A typical case of
such material is shown is Fig. 6.3.1.

For uniaxial loading, the magnitude of the strain in that direction becomes the effective
strain, i.e.,

E= E, for uniaxial tension in x

E= -E, for uniaxial compression in x

We need to find the effective stress ( a ) corresponding t o Z. There are two known data points
as shown in Fig. 6.3.2; namely the effective stress for uniaxial tension (at) and the effective
stress for uniaxial compression (a,). Some method of interpolation or extrapolation is required
t o predict the effective stress for the general stress state using two known data points.

The first stress invariant ( I l ) is adopted for interpolation/extrapolation,i.e.,

Considering that the pure shear is in the midway between simple tension and simple compres-
sion, it seems appropriate t o use the first stress invariant. Hydrostatic tension and compression
cases will impose lower and upper bounds for extrapolation, i.e.,

Il = a, for uniaxial tension/compression

Il = 0 for pure shear
Il = 3p for hydrostatic pressure

The instantaneous modulus ( g )should be interpolated or extrapolated in the same man-

ner.
 o Monotonic-Compression
o Cyclic-Compression
40 ....i

STRESS
(ksi) 32 ....;
. ..

24 ....( o Monotonic - Tension

o Cyclic - Tension
D Fracture

0 ... .............................................................................. !-;

...............-.
"

0 0.002 0.004 0.006 0.008

STRAIN

Figure 6.3.1. Stress-Strain Curves for Gray Cast Iron

6.3.4 Computational Procedure for Bilateral Stress-Strain Relations

The new stress state is proportional in magnitude to the effective stress (a),which should be
determined as follows:
1. Compute the effective stress (a,)based on {o,), i.e.,
 for plane stress

for plane strain

2. Compute the first invariant of (0,):

1 = ox
I + + oz
+y

where a, = 0 for plane stress.

3. Determine the ratio (r) by normalizing Il by a,, i.e.,

where r signifies the relative distance from the midpoint of a, and at at B, as shown in
Fig 6.3.2. It would be implausible to process a large value of r (such is the case with a
hydrostatic load). Therefore, r will be confined t o a plausible range, -1 5 r 1. The <
value will be reset to the limit ( r = &1 ) if r lies outside the range.
4. Look up the user-specified stress-strain curve in the TABLES1 entry and determine at
and a,, i.e.,

5. Determine a based on at,a, and r , i.e.,

For the tangent matrix, the instantaneous modulus (g)

should be determined using the
same ratio ( r ) as follows:
1. Compute the instantaneous slope at E for tension, i.e.,
6% Yi+l - Yi
(,It = for xi 5 B < x;+1
x;+1 - 2;
where (x;, y;) is the i-th data point in the TABLES1 entry.
2. Compute the instantaneous solpe at (-E) for compression, i.e.,
aa yj+l - yj
for x j < -E < xj+l
( z ) c =
xj+l - x j

3. Determine ($$) based on ($$)t, (E),,and T, i.e.,

Figure 6.3.2. TABLES1 INPUT for Stress-Strain

6.3 - 6
6.4 CREEP A N D VISCOELASTICITY

A creep analysis capability, using a generalized viscoelastic model [6.11], is introduced to

process the creep behavior coupled with elastoplastic deformation. The formulation is based
on the step-by-step time integration of the Kelvin-Maxwell rheological model with non-constant
parameters.

The concept of a rheological model is extended t o the multiaxial stresses by the Prandtl-
Reuss stress-strain relationship, from which the tangential stiffness matrix is formed for New-
ton's iteration. If the plastic deformation is coupled with creep, the algorithm will seek a
solution in two distinct steps. Various choices of empirical creep laws are available and small
variations in temperature are allowed as implemented in MSCINASTRAN.

6.4.1 Introduction

At the macroscopic level, the creep phenomenon is best observed in the uniaxial test under
constant load and the relaxation test under constant strain a t constant temperature. A speci-
men subjected t o a constant uniaxial tension at an elevated temperature exhibits three distinct
phases in a time frame: primary creep, secondary creep and the tertiary creep to rupture.
If the specimen is unloaded after some creep deformation, the elastic strain is immediately
recovered and a portion of the creep strain is gradually recovered. The recoverable portion
of the creep deformation is called primary creep and the non-recoverable portion, secondary
creep. The tertiary creep, similar to necking in plasticity, is considered as a localized instability
phenomenon, which is beyond the scope of this handbook.

Elaborating on the Kelvin-Maxwell rheological creep model developed by Badani [6.12],

the creep behavior is coupled with plastic deformation and the variable temperature effect is
included. The Kelvin element represents the primary creep behavior and the Maxwell element
describes the secondary creep. For a generalization of the viscoelastic material behavior, the
rheological model parameters are treated as nonlinear functions of the effective stress and the
temperature. The step-by-step integration is performed assuming that the strain-rate varies
linearly during each time increment while the rheological model parameters remain constant
for a short time interval.

A number of empirical creep laws, recommended by the Oak Ridge National Laboratory
(ORNL), are provided along with options for general tabular input of the rheological model
parameters as functions of effective stress. When the creep characteristics are specified in
terms of empirical creep laws, the program converts the empirical formula t o the corresponding
rheological model.
6.4.2 Effects of Stress and Temperature

An analytical solution t o the Kelvin-Maxwell model subjected to a constant stress (a)is given
by
c - u (6.4.1)
Etotal - - t t EC .
c, przmary

with

where t denotes time and parameters Kp, Cp, and C, are identified in Fig. 6.4.1. For a
varying stress case, however, an instantaneous strain-rate should be employed to have the creep
hardening effects properly accounted. The creep hardening (and softening) effects, assumed
t o be exhibited by the primary creep, are accounted for by using the creep hardening law,
wherein creep strain rates are expressed in terms of primary creep strain rather than time, i.e.

where

Notice that the creep strain rate reverses the sign with proper effects of the cummulative
primary creep when the sign of the stress reverses. Due t o this virtue, the stress reversal effects
are properly taken into account without any special provision in the algorithm.

The creep strain rate is, in general, a function of absolute temperature. Microscopically,
the creep deformation is an integrated effect of dislocations of the crystal structure primarily
due t o thermal activation and stress, The effects of temperature at the microscopic level can
be quantified as
i c = A,-AH/RT (6.4.3)

where AH is the energy of activation, R is the gas constant, T is the absolute temperature
and A is in strainlunit time. On this basis, the creep model parameters measured at a ref-
erence temperature (To) can be used t o compute the creep strain rate a t the temperature T
in the vicinity of To, allowing small variations in the ambient temperature. The creep model
parameters are corrected as

Cs(T) =
cs(To),
F and CP (To)
Cp(T) = -
Fc

where
6.4.3 Equillibrium of a Kelvin-Maxwell Model for a Stress Component

Suppose that the Kelvin-Maxwell model is applied t o a typical component of a deviatoric

stress-strain pair, denoted by (Asl, el). Referring t o Fig. 6.4.1, the state equilibrium equation
of the model a t any instant is expressed as:

[C]{A&} + [Ii]{Ae} = {As} (6.4.5)

where

and
AS}^ = < Asl 0 >
Introducing a linearly varying strain-rate during the time interval, the strain-rate increment
can be expressed as n

by which Eq. (6.4.5) is reduced t o

Defining the stiffness of the primary and secondary creep elements by

2cs
kl = K p + -
2Cp
At
and k2 = -
At
the equivalent creep stiffness (LC)for a deviatoric stress-strain pair can be determined by

Solving Eq. (6.4.7) for Ael gives

kc Ael = As' + As1

where As' is a pseudo-incremental stress, determined by

The pseudo-incremental stress represents the change in deviatoric stress component due
t o creep relaxation. This value of As' can be converted t o the equivalent psuedo-incremental
strain (Ae'), which represents the change in deviatoric strain component due to creep. By
virtue of creep volume constancy, deviatoric strains (Ae') and the strain rates (11 and i 2 ) in
Eq. (6.4.10) are identical to the ordinary strain components. Hence the pseudo-incremental
strain can be expressed as
As'
AE' = - = 2 - i2) + s i z ]
kc bl
in which the first and the second terms are contributions of the Maxwell and the Kelvin
elements, respectively. Notice that as At approaches zero, Eq. (6.4.11) reduces to

Ae' cl At il = At iio,,,

which makes it easier to comprehend Eq. (6.4.11).

Primary Secondary
Creep (K,) Creep (K2)

r=-Je=rel *
-
b o(t)

or AS,

Figure 6.4.1. Rheological model

6.4.4 Adaptation of Rheological Model to Multiaxial Stress State

The concept of the rheological model for a stress-strain pair is extended to the general multi-
axial stresses with the aid of effective stress and effective strain-rate by adopting the Prandtl-
Reuss stress-strain relationship, i.e.,

2
{iC)= -32 -
a
{s)

where creep strain rates

deviatoric stresses
{ s l T = < 0; 0 u 2 r x y 2 ~ y Z 2rzx >,
and the effective stress (a) and the effective creep strain rate ( p )are defined analogously t o
those in plasticity with von Mises yield criterion.
 Implied by this relationship, there is a unique set of rheological parameters (Kp, cp,
and c , )
based on the effective stress, which can be related t o all the deviatoric stress components and
the corresponding creep strain rates. In terms of effective creep model parameters, Eqs. (6.4.8)
and (6.4.11) may be rewritten as follows:

and

The creep strain rates in Eq. (6.4.2) may be rewritten likewise, i.e., in terms of deviatoric
stresses

where

Then the pseudo-incremental strain in Eq. (6.4.11a) is reduced to

In the absence of plastic deformation, the total strain increment (elastic and creep) may
be expressed as
{Ace +
Aec) = [DL' +
D;'] {ACT} (6.4.14)
where De and Dc are material matrices for elasticity and creep, respectively. However, the
total strain increment must be corrected with a pseudo-incremental strain {A&'}, i.e.,

Combining Eqs. (6.4.14) and (6.4.15), we obtain

{ACT}= [Be,]{Ae - A&'} (6.4.16)
where the elastic-creep tangent matrix may be conveniently obtained for an isotropic material
by - -
K + $kec K - '3k ec K - itec0 0 0
K+ikec K-ikec o o o
I o o o
[Dec] =
l2 k ec 0 0
SYM c 0

L ikec -
with li being a bulk modulus (K = -), and kec the stiffness of the elastic-creep component
defined by
1
-
kec
- -2G1+ - 1
kc
where G denotes the shear modulus.

6.4.5 Coupling of Plasticity

When the plastic deformation is involved in the creep analysis, the plastic strain increment
should be included in the total strain increment. The plastic strain increment may be obtained
by
{ A E ~ }= D,' {Aa) (6.4.18)
where

da
with plasticity modulus H = - and the function ( f ) defining effective stress (a).Introducing
da
Eq. (6.4.18) into Eq. (6.4.14), the elasto-plastic-creep stress-strain relations are established as

[DL' + D,' + D;'] {Ao) = {AE - AE') (6.4.19)

from which the stress increment {Aa) can be obtained. However, Eq. (6.4.19) may be rear-
ranged as
(6.4.20)
a
with

Ax =
{gjTDeC{A&- AE') = A P > O
H + {gjTDec { g )
where the effective plastic strain increment is denoted by A P

The creep deformation tends to relax the stress gradient in the absence of further increments
in external loads. In the creep-dominant process, therefore, plastic deformation can be induced
only by creep deformation to alleviate stresses in the neighboring material. For this reason,
the material routine employs a solution scheme which seeks a solution in two distinct steps
when the plastic deformation is coupled with creep. First, it solves for the incremental stress
components with an elastic-creep material as in Eq. (6.4.16). Then, if the new stress state
+
( a = a,ld Aa) exceeds the current yield stress, i.e.,
with F being a yield function, a correction is made on a previously obtained incremental stress
based on Eq. (6.4.20) and the plastic strains are computed. The algorithm for this process is
depicted by a flow diagram in Fig. 6.4.2.

- Comp

I Interpolate Fp. C,. c.

based on ri
(Process Bulk Data cards CREEP TABLES I I I

Form [ D , , ] matrix
-
i No
1 1
Solve elastic-creep
problem for j Ao)
Compute A&-'(A&')
Compute Ad
Compute (A&., j irst Yes
Compute A&-'
Update E-' = E - ' + AE" Store [Dec]
Compute 1
Update ( 4 ;= [ G )+ { A G ;

Plast~c~ty
Routine I
I
I
I
I
I
I
I
I

:{-1
I
Compute a = I
I
Compute ( d ) = D,,{a} I
I
I aTAa
I
I
Compute b.= -H + aTd
I
I
I
I Compute Aa = Aa - A l d I
I
I
Comctlon for klnematlc hardening I
I Update a = a + Aa I
I 6-P = 6-9 + M I
I Error Correct~onlf requrrcd I
I I
I I
I_____--_---_---------------------------I

Figure 6.4.2. Flow Diagram for Creep Process

6.4.6 Implementation

The quasi-static nonlinear analysis capability for the generalized viscoelastic model coupled
with the plastic deformation is implemented into the general-purpose nonlinear static analysis
method in MSCINASTRAN. The solution method employs an adaptive Newton's iteration
with accelerated convergence schemes such as the BFGS updates and the line search process.
A nonlinear static analysis can readily be converted t o the quasi-static analysis and vice versa
by adding or subtracting a few related data. All the potentially nonlinear elements (ROD,
BEAM, QUAD4, TRIA3, HEXA, PENTA and TETRA) are applicable t o creep analysis, but
not all the nonlinear elements in the model need to be composed of creep material.

The material routines are implemented to adopt an adaptive subincremental scheme by

which the computations in material routines are carried out in small increments within a global
incremental step. The creep routine subdivides an incremental time step such that the effective
creep strain increments and effective stress increments are bounded by a fraction of the total
creep strain and effective stress, respectively. If the plastic deformation occurs during the creep
process, the creep routine calls the plasticity routine. The plasticity routine again subdivides
the given increment into smaller increments if the increment in the yield stress exceeds a
certain fraction of the current yield stress. The fraction which defines the bounds is a user
specified parameter with a default value of 0.2. By virtue of adaptive subincremental scheme
within the material routines, the non-proportional loading or unloading does not present any
difficulty. The non-proportional loading cases including neutral loading have been examined
and validated by controlling either the stress or strain path in two-dimensional problems with
the elasto-plastic material.

Users are allowed t o either specify an empirical creep law or provide direct input of rheo-
logical parameter values as functions of the effective stress. If the creep behavior is prescribed
by a creep law, creep law parameters are converted t o rheological parameters whenever a
new stress state is computed for the creep analysis. Rheological parameters are corrected for
changes in temperature based on Eq. (6.4.4) if the operating temperature is different from the
temperature at which creep behavior is characterized. The empirical creep laws provided in
MSCINASTRAN are in the following equation forms with user specified coefficients (a-g):

and
c C ( a , t ) = A(a) [1 - e-R(u)t] + Ir'(0)t
for all combinations of types 1 and 2 for A, R, and K , which are expressed as

Parameter Type 1 Type 2

-
40) aab a Exp (ba)
R(a) c Exp (do) cad
WJ) e [sinh (f a)]g e EXP(f0 )
 Conversion of creep laws in the form of Eq. (6.4.22) t o the rheological model is performed
by simply matching equivalent terms in Eq. (6.4.2). However, conversion of the popular and
classical creep law in Eq. (6.4.21) to the rheological model is not so simple due to lack of
similarity. In this case, equivalent rheological parameters are computed by matching the total
creep strain and their first and second derivatives of Eqs. (6.4.1) and (6.4.21).

The material routine computes the elastic-creep tangential matrix for the formation of
a global stiffness matrix, which is used for the iteration combined with the BFGS update
and line searches. An equilibrium condition is achieved within the convergence tolerance at
every time-increment. The numerical solution scheme employed in the creep formulation is
unconditionally stable owing to the implicit method adopted for integration. However, a limit
should be placed on the time increment to achieve an accurate solution. The time increment
should be selected so that the strain and/or stress do not change excessively in a single step.

The total creep strain increment can be recovered by subtracting and {AeP) from
the total {Ae) in Eq. (6.4.19), i.e.,

The primary creep strain increment is saved in the database to take into account creep hard-
ening/softening effects, i.e.,

where

6.4.7 Verification and Validation

All aspects of creep capabilities were verified with respect t o the algorithm, accuracy, and
programming errors with a wide variety of problems in MSC/NASTRAN. All types of elements
(ID, 2D and 3D) have been tested and known solutions have been reproduced. Various creep
laws in the form of Eq. (6.4.22) and the direct input of rheological parameters reproduced
analytical solutions accurately with errors less than 0.1% in the creep strain a t the end of 70
steps. Creep laws in Eq. (6.4.21), however, produced a cumulative error of 15-19% in the creep
strain a t the end of 70 steps, because of the mismatch between the empirical formula and the
rheological model.

The creep analysis capability under variable temperatures is verified using empirical creep
laws recommended by ORNL. There are two different creep laws (types 111 and 121) for the
same material (type 304 stainless steel) established a t different temperatures (llOO°F And
1200°F) as shown in Fig. 6.4.3. The creep law type 111, defining the creep behavior at llOO°F,
is applied at the operating temperature of 1200°F with corrections for variable temperature.
The results are compared with the creep law type 121, which predicts proper creep behavior at
1200°F. The accuracy of the varying temperature case was found satisfactory only for a short
period of time or for small temperature variations. Nevertheless this feature is considered
useful and essential.

Coupling of the plastic deformation with creep is verified by reproducing the isochronous
stress-strain curve for stainless steel, type 304, as shown in Fig. 6.4.4. The data points are
obtained in 15 steps to the creep time of 100 hours.

The creep behavior is manifested in the relaxation process under constant strain. Figure
6.4.5 compares relaxation predictions by various methods. It is demonstrated that the current
method is superior t o the initial-strain method. Results from MSCINASTRAN analysis fall
between experimental data and the closed-form solution.

The creep response of various elements to the.stress reversal was thoroughly investigated.
The effects of the creep hardeninglsoftening were exhibited properly. However, the convergence
occasionally required a stiffness matrix update and a smaller time increment a t the onset of
stress reversal.

The solution t o the creep behavior of an infinitely long thick-walled cylinder subjected
t o internal pressure was presented by Greenbaum and Rubinstein [6.13]. They employed an
incompressible material with a simple empirical formula which accounts only for secondary
creep, to which an analytical solution exists. MSCINASTRAN reproduced the solution very
accurately using nine HEXA elements with axisymmetric and plane strain boundary conditions.

The creep behavior of a thick-walled pressure vessel with a flat-end closure was also analyzed
by Greenbaum and Rubinstein [6.13] under an internal pressure of 445 psi with a material
obeying an empirical creep law in the form of

A finite element model of the same pressure vessel was analyzed by MSCJNASTRAN using
72 solid elements with 355 active DOFs. Fig. 6.4.6 shows stress contours at t=3 hours. For a
model five times coarser, this solution represents an excellent agreement with that in Reference
[6.13].
 6..

Creep Law 121 at 1200°F

5. -
0
-

X
Y

Creep Law 1 11 Defined at 1 1OO°F

0 100 200 300 400 500

CREEP TIME (HOURS)
Figure 6.4.3. Creep Capability Verification for \;lriablr Temperature

24

Original Material
X MSCINASTRAN
0 ASME Code Prediction
4 -
0 I 1 I I I a

0 0.4 0.8 1.2 1.6 2.0 2.4

TOTAL STRAIN (Oh)
Figure 6.4.4. Isochronous Stress - Strain Cur
 0 MSCINASTRAN
-
--- Badani
Experimental
.....
0 Closed Form
initial Strain Method

0 20 40 60 80 100
TIME (HOURS)

Figure 6.4.5. Cornparision of Creep Relaxation Predictions

Level D ............400.
Level E ............ 500.
Level F ............ 600.
Level G ............700.
Level H ............ 800.
Level I ............ 900.
Level J .......... 1000.
Note: Stress Units in ksi

Figure 6.4.6. Effective Stress Distribution after 3 hours of Creep

6.4.8 Nozzle-to-Spherical Shell Attachment as Benchmark Problem

The High-Temperature Structural Design (HTSD) program of ORNL was instigated to develop
design methodologies for breeder reactor structures. Under the ORNL program, a nozzle-
to-spherical shell attachment model was presented as a benchmark problem with extensive
experimental 16.141 and analytical [6.15, 6.161 results. The model represents a liquid-metal fast
breeder reacter component, made of type 304 stainless steel, subjected t o successive cycles of
loading with internal pressure and end moment at llOO°F.

The model depicted in Fig. 6.4.7 represents this component using 1552 DOFs, which is an-
alyzed by MSCJNASTRAN with elasto-plastic-creep material and geometric nonlinear effects.
The creep characteristics are defined by a creep law in the form of Eq. (6.4.22) with ORNL
recommended values for parameters as follows:

A(a) = 6.985 x a2.444

R ( a ) = 7.032 x low4Exp (0.10720)
K ( a ) = 6.73 x lo-' [sinh ( 0 . 1 4 7 9 a ) ] ~ . ~

where the stress ( a ) and the time ( t )are measured in ksi and hour, respectively.

Typical results at selected points, where the plastic deformation is significant, are compared
with data in References [6.14,6.15, and 6.221 as shown in Figs. 6.4.8, 6.4.9(a) and 6.4.9(b). It
is noted that MSCINASTRAN used an isotropic, instead of a kinematic hardening rule as
employed by Levy [6.15]. Figs. 6.4.10(a) and 6.4.10(b) show effective-stress contours upon
unloading a t t=2546 and t=3400 hours, respectively. These results demonstrate proficient
capabilities of MSCJNASTRAN for a complex problem with combined nonlinearities.
 1
NOZZLE-TO-SPHERICAL SHELL

m
Overall Height: 24.06 in. 304 GRIDS
Sphere Radius: 12.03 in. 264 QUAD4s
Shell Thickness: 0.33 in.
Nozzle Radius: 1.984 in.
Nozzle Thickness:0.112 in.
Fillet Radius: 1.425 in.

MSC/GRASP ( UAX 4 25-JAN-85 16:04:26

Figure 6.4.7 Finite element model by hfSC/NASTRAN

.- .-.
1- 0 Experiment
MSC/NASTRAN
Pressure Loading.

TIME (HOURS)
Figure 6.4.8 Total circumferential strain at gauge location during
pressure loading
 5.01 f
)/((
- r4---' I
I
I
- MSCINASTRAN
--- Levy

s 4c, 1
?a
/ 1Q I
Unloaded
u
I-
0
1.0 -
-.-.20.261
0.0 ' I
,., Moment Load M Experimental data
exh~brteddr~ft
- 1.0 I I 1 I I I I I

1800 2000 2200 2400 2600 2800 3000 3200 3400 31

TlME (HOURS)

Figure 6.4.9(a) Total circumferential strain at gauge location (1)

during moment loading

1800 2000 2200 2400 2600 2800 3000 3200 3400 3600
TlME (HOURS)

Figure 6.4.9(b) Total circumferential strain a t gauge location (2)

during moment loading
Figure 6.4.10(a) Effective stress contour plot at t=2546 hours upon
loading

Numbers Represent
Stress in ksi

Figure 6.4.11(b) Effective stress contour plot a t t=3400 hours upon

loading
6.5 THERMO-ELASTICITY

Thermoelastic capability was implemented in SOL 66 (or SOL 106) of MSC/NASTRAN Ver-
sion 66. With this capability, the temperature dependent material properties for linear elastic
and nonlinear elastic materials are updated as the temperature distribution in the structure
changes.

6.5.1 Theoretical Basis

Thermo-elasticity in MSC/NASTRAN is an extension of nonlinear elasticity as described in

Section 6.3. The thermal strain is computed by

where
T = current temperature
TREF = reference temperature at which coefficient thermal expansion is measured
To = stress free temperature (initial temperature)
a(T) = coefficient of thermal expansion

The total stress and total strain relations are modified from Eq. (6.3.4) as
1 a(E,T)
{a} = --
E z [De(T)]{€ - ET}.
In the linear elastic case, Eq. (6.5.2) reduces t o

By neglecting the rate of change of the instantaneous tangent t o the uniaxial curve with respect
to temperature, the tangent material stiffness matrix may be obtained by

6.5.2 User Interface

The thermal-stress free temperature distribution, To, is used as a starting point for linear and
nonlinear thermal elastic analyses. The set ID of this temperature distribution is specified
above the Case Control section by the TEMP(IN1T) command. The initial temperature, To,
is used in Eq. (3.5.1) and other parts of the computation, i.e.,
To look up the material properties at temperature To for linear and upstream superele-
ments (these properties remain unchanged through the entire analysis),
 To be the default temperature distribution whenever TEMP(L0AD) is not specified for
a subcase,

r To be the reference (stress-free) temperature when TEMP(L0AD) is specified.

From Version 67, there is no default value for To and a fatal error message will be is-
sued if TEMP(1NIT) is not present. The set ID for the load temperature distribution, T , is
specified in the Case Control by TEMP(L0AD). For nonlinear elements, the TEMP(L0AD)
also specify the temperatures at which material properties are determined if Bulk Data en-
tries MATTi are present. The material properties will conform to TEMP(L0AD) without
requiring TEMP(MAT) or TEMP(B0TH). Each TEMP(1NIT) or TEMP(L0AD) points to
the Bulk Data TEMP, TEMPD, TEMPPI, TEMPRB on which the temperature distribution
is specified. The coefficient of thermal expansion reference temperature, TREFis input by the
TREF field on the MATi, which is different from Version 65. Notice that the combination of
TEMP(IN1T) and TEMP(L0AD) is new user interface introduced in Version 66. They are
recommended to be used for all types of thermo-elastic analyses available in SOL 66 and SOL
106 for Version 66 and beyond. Some old features such as TREF in MATi and TEMP(MAT)
which were valid prior t o Version 66 should not be used in versions following Version 65. For
example, TEMP(MAT) without TEMP(IN1T) would run through in Version 67. However, the
material properties are determined at the very beginning of the analysis based on the tempera-
ture distribution given in the first subcase. But these properties will stay constant throughout
the analysis. TEMP(MAT) may also be used in the nonlinear transient analsis (SOL 99 or
SOL 129), but it can only be applied to the upstream superelements. A fatal message will be
issued if TEMP(MAT) appears in a residual subcase.

Different options of thermo-elastic constitutive relations may be obtained by different com-

binations of input data:

r {a) = W [ D ~ ( T ) ] { ~ )

{a} = -[D,(T)]{C) (MAT1, MATT1, MATS1, TABLEST, TABLESl)

r {a} = W [ D e ] { c ) (MATl,MATSl,TABLEST,TABLESl)
where {a) and (€1are the stress and strain vectors, [D,(T)] the elasticity matrix from MATT1
data, E the effective elasticity modulus from TABLES1 data, and E the elasticity modulus
from MAT1 data.

Temperature dependent properties for elasticity matrix, i.e. [D,(T)], are input by a combi-
nation of the MATi, MATTi, and TABLEMi entries. For the nonlinear elastic material, which

ature dependent uniaxial stress-strain relations, i.e. =.

is specified by the MATSl and TABLESl, the TABLEST may be used to input the temper-
The TABLEST is referenced by
the MATSl and contains the TID of TABLES1 cards for given temperatures. A fatal error
message will be issued if both TABLES1 and TABLEST with the same TID are present. The
TABLEST entry is shown below followed by brief description of the fields:
 TID
T1 I1 T2 I2 T3 I3 ENDT

TID : Table identification number (integer > 0). This number must be identical to TID of
MATSl to be active.

Ti : Temperature values (real) which must be listed in ascending order.

Ii : Table identification number of TABLES1 entries (integer > 0). Ii must be unique with
respect to all TABLES1 and TABLEST table identification numbers.

ENDT : The end-of-table flag.

Possible thermal stress analysis types and the corresponding required input cards are shown
in tlie table below.

Types of Thermal Stress Analyses and Required Input Entries

Type of analysis Case Control Bulk Data
thermal loads only, no TEMP(IN1T) = m once TEMP,TEMPD..
temperature dependent TEMP(L0AD) = n each subcase MATi
material is present
+
thermal loads temperature same as above TEMP,TEMPD..
dependent linear materials MATi, MATTi
TABLEMi
+
thermal loads temperature same as above TEMP,TEMPD..
dependent nonlinear elastic MATi, MATS1
materials TABLEST
TABLES1
no thermal load, temperature same as above same as above
dependent material only but a=O.O on
MATi

Notice that temperature dependent linear composite materials (MAT2 with MATT2 or MATS
with MATTS) are available in MSCINASTRAN, but nonlinear elastic composite materials
(MATSlINONLINEAR with MAT2 or MATS) are not. Thermoplasticity is also not available
in MSCINASTRAN Version 67. Analysis capability for the creep (or viscoelastic) deformation
under the variable temperature became available from Version 66. In addition, the creep
characteristics and the elasticity properties both can be temperature dependent. The former
are specified by the CREEP entry (see Section 6.4 for more details), and the latter are input
with the same procedure as described above.

6.5.3 Validation Problem

A 1-inch cube, as shown in Fig. 6.5.1, is subjected to a uniaxial tension of 10,000 lbs and a
temperature change from 10o°F to 200°F, assuming that the former is the stress-free temper-
ature. The Young's modulus is 8.0 x lo6 psi at 100°F and 4.0 x lo6 psi at 200°F, and the
Poisson's ratio is 0.3 for all temperatures. For simplicity, the thermal expansion coefficient is
assumed to be 0. The structural response at 10o°F and 200°F, respectively, is determined.

Figure 6.5.1 Cube Subjected t o Uniaxial Tension

For verification purpose, two identical plate models were created and solved using different
approaches as shown in the attached MSCINASTRAN input data. In the first model, the
reference material properties were defined on a MAT1 entry, and their variations with different
temperatures were input through MATTl and TABLEMl entries. In the second model, the
reference properties were still input through a MATTl entry, however, a set of temperature
dependent stress-strain curves were specified using a combination of MATS1 and TABLES1
entries. The MATTl entry present in the second model is a dummy, which is required in V66
and V66A t o activate the thermal elastic capability. This requirement has been removed in
V67 and beyond. The results given by these two models are identical as shown below.

Temperature OX UY EX EY

100°F 10,000 psi 0 psi 1.25 x low3 -3.75 x

200°F 10,000 psi 0 psi 2.50 x -7.50 x
ID THPLATE, V66A $ SSH 12/12/90
TIME 5 $ CPU MIN.
SOL 66 $ NONLINEAR STATIC SOLUTION SEQUENCE
DIAG 8,50 $ PRINT MATRIX TRAILERS AND TRACE N-L ITERATION
CEND
TITLE = 2 DISCRETE PLATE MODELS UNDER UNIAXIAL TENSION
ECHO = SORT
SEALL = ALL
DISP = ALL
STRESS = ALL
ELFORCE = ALL
SPCFORCES = ALL
OLOAD = ALL
TEMP(INIT)=IOO
SUBCASE 100
NLPARM = 100
LOAD = I 0 1
SUBCASE 200
TEMP(L0AD) = 200
NLPARM = 200
LOAD = 101
BEGIN BULK
$
$ TEMPERATURE DISTRIBUTION
$
TEMPD 100 100. 200 200.
$
$ ITERATION STRATEGY
$
NLPARM 100 I YES
NLPARM 200 5 YES
$
$ FIRST MODEL
$
GRID 101 0 0. 0. 0.
GRID 102 0 I. 0. 0.
GRID 103 0 I. I. 0.
GRID 104 0 0. I. 0.
$
CQUAD4 101 100 101 102 103 104
$
PSHELL 100 120 1.0
$
MAT1 120 8.+6 3.0769+60.3
MATT1 120 151 152
$
TABLEMI 151
0. 8.0000+6100. 8.0000t6200. 4.0000+6ENDT
TABLEMI 152
 0. 3.0769+6100. 3.0769+6200.
$
FORCE 101 102 0 5.OE3 1.0
FORCE 101 103 0 5.OE3 1.0
$
$ SECOND MODEL
$
GRID 201 0 0. 0. 0.
GRID 202 0 I. 0. 0.
GRID 203 0 I. I. 0.
GRID 204 0 0. I. 0.
$
CQUAD4 201 200 201 202 203 204
$
.PSHELL 200 220 1.0
$
MAT1 220 8.+6 3.0769+60.3
MATT1 220
MATS1 220 221 NLELAST
TABLEST 221
+ST1 0. 25 1 100. 252 200. 253 ENDT
TABLES1 251
+SSI -1. -8.+6 0. 0. I. 8.+6 ENDT
TABLESI 252
+SS2 -1. -8.+6 0. 0. I. 8.+6 ENDT
TABLESI 253
tSS3 -1. -4.+6 0. 0. I. 4.t6 ENDT
$
FORCE 101 202 0 5.OE3 1.0
FORCE 101 203 0 5.OE3 1.0
ENDDATA
6.6 INITIAL ANISOTROPY

This material capability became available in SOL 66 (or 106) and SOL 99 (or 129) since Version
66A. It allows the following combinations of material property entries:
Elastic-plastic: MAT2 or MAT9 combined with MATSl(TYPE=PLASTIC)
Thermo-elastic: MAT2 combined with MATT2 or MAT9 with MATT9
Viscoelastic: MAT2 or MAT9 combined with CREEP
Notice that MAT2 or MAT9 represent elastic anisotropy. When they are combined with
plastic or creep material, they represent initial anisotropy because the plastic or creep (vis-
coelastic) deformation remains isotropic.

6.6.1 Theoretical Basis

The elasticity matrix [D,]appears in the incremental constitutive relations for elastic-plastic,
thermoelastic, and viscoelastic materials as shown in the preceding sections.

In those constitutive relations, the elastic properties are assumed to be isotropic so that
the elasticity matrix [D,]may be determined by two material constants, e.g. Young's modulus
E and Poisson's ratio v. This assumption was removed by simply replacing the isotropic
[D,]by an anisotropic elasticity matrix specified by MAT2 or MAT9 Bulk Data entries. The
orthotropic material (MAT8) is not applicable to this capability. This capability may be used
t o model nonlinear materials which exhibit significant initial anisotropy.

6.6.2 Validation Problems

A cube with dimensions of one unit was modeled with one QUAD4 element for testing the 2-D
case and one HEXA element for testing the 3-D case. First, isotropic properties were input
using MAT2 and MAT9 Bulk Data entries and the results were compared with the analytical
solutions. Then, genuine anisotropic properties were used and the results from the plate and
solid models were compared with each other.

The input data of three test decks, namely ANISOEP, ANISOC, and ANISOT, are listed
at the end of this section as examples. Each deck contains two discrete models, a plate and
a solid, which are subjected t o a simple tension force of 10,000 lb, respectively. In-plane
anisotropic elastic properties are assumed so that the plate and solid models are expected t o
give compatable results. ANISOEP is used t o test the elastic-plastic analyses, in which the
yield stress is assumed to be 5000 psi and the plastic hardening modulus t o be 5 x lo5 psi. In
ANISOEC, two subcases are specified. The first subcase is for a static response analysis which
is followed by the second subcase for a creep analysis of 100 hours. The material is assumed
t o remain linear elastic throughout the analysis. ANISOT is used t o test the thermo-elastic
analyses. A tension force of 10,000 lb is applied in the first subcase. In the second subcase, the
mechanical load remains unchanged but the temperature is increased, which reduces the value
of each elasticity modulus by one half. The results of these three test problems are tabulated
below:

ox ex E~ YSY
ANISOEP :
plate model 10,000 1.147 x -4.526 x -9.540 x
solid model 10,000 1.147 x -4.526 x -9.550 x lo-*
ANISOC :
plate model 10,000 2.560 x -7.054 x -9.548 x
solid model 10,000 2.560 x -7.050 x -9.551 x
ANISOT :
plate model 10,000 2.941 x 9.486 x -1.910 x
solid model 10,000 2.941 x 9.487 x -1.910 x
ID ANISOEP, V66 $ SSH 3/25/91
TIME 5
SOL 66
CEND
$
TITLE = MAT2 OR MAT9 COMBINED WITH MATSI(TYPE=PLASTIC).
SUBTITLE = 2 DISCRETE MODELS EACH CONTAINS I QUAD4 OR HEXA ELEMENT.
LABEL = UNIAXIAL TENSION CASE.
SEALL = ALL
DISP = ALL
STRESS = ALL
$
NLPARM = 100
LOAD = 101
$
BEGIN BULK
$1------2-------3-------4-------5-------6-------7-------8-------9------- lo-----$
$
$ ITERATION STRATEGY
$
NLPARM 100 10 AUTO
$
$ PLATE MODEL
$
GRID 201 0 0. 0. 0. 123456
GRID 202 0 I. 0. 0. 3456
GRID 203 0 1. I. 0. 3456
GRID 204 0 0. I. 0. 13456
CQUAD4 201 200 201 202 203 204
PSHELL 200 220 1.0
MAT2 220 7.9812E6-6.149E51.5136E64.9539E61.5136E63.0824E6
MATS1 220 PLASTIC 5 .E5 5.E3
FORCE 101 202 0 5.OE3 1.0 0.0 0.0
FORCE 101 203 0 5.OE3 1.0 0.0 0.0
$
$ SOLID MODEL
$
GRID 501 0 0.0 0.0 0.0 123456
GRID 502 0 1 .O 0.0 0.0 3456
GRID 503 0 1.0 1.0 0.0 3456
GRID 504 0 0.0 1.0 0.0 13456
GRID 511 0 0.0 0.0 1.0 12456
GRID 512 0 1.0 0.0 1.0 456
GRID 513 0 1.0 1.0 1.0 456
GRID 514 0 0.0 1.0 1.0 1456
CHEXA 501 500 501 502 503 504 511 512 +E51
+E51 513 514
PSOLID 500 520
MAT9 520 7.9812E6-6.149E50. 1.5136E60. 0. 4.9539E6+M51
+M51 0. 1.5136E60. 0. 1.E6 0. 0. 0. +M52
tM52 3.0824E60. 0. 3.E5 0. 3.E5
MATS1 520 PLASTIC 5.E5 5.E3
FORCE 101 502 0 5.OE3 0.5 0.0 0.0
FORCE 101 503 0 5.OE3 0.5 0.0 0.0
FORCE 101 512 0 5.OE3 0.5 0.0 0.0
FORCE 101 513 0 5.OE3 0.5 0.0 0.0
ENDDATA
ID ANISOC, V66 $ SSH 3/25/91
TIME 5
SOL 66
CEND
TITLE = MAT2 OR MAT9 COMBINED WITH CREEP
SUBTITLE = 2 DISCRETE MODELS, EACH CONTAINS ONE QUAD4 OR HEXA ELEMENT
LABEL = UNAXIAL TENSION CASE
SEALL = ALL
DISP = ALL
STRESS = ALL
$
SUBCASE 10 $ ELASTIC
LOAD = I
NLPARM = 10
SUBCASE 11 $ CREEP FOR 100 HOURS
LOAD = I
NLPARM = 20
$
BEGIN BULK
$1------2-------3-------4-------5-------6-------7-------8-------9------- lo---
$
$ ITERATION STRATEGY
$
NLPARM 10 I AUTO
NLPARM 20 5 20. AUTO
$
$ PLATE MODEL
$
GRID 201 0 0. 0. 0. 123456
GRID 202 0 1. 0. 0. 3456
GRID 203 0 1. I. 0. 3456
GRID 204 0 0. I. 0. 13456
CQUAD4 201 200 20 1 202 203 204
PSHELL 200 220 1.0
MAT2 220 7.9812E6-6.149E51.5136E64.9539E61.5136E63.0824E6
CREEP 220 CRLAW
+CP2I 222 3.476-4 2.08-4 2.085-11 2.094 1.02-11 7.43-4
FORCE I 202 0 0.5 1.+4 0.0 0.0
FORCE I 203 0 0.5 1.+4 0.0 0.0
$
$ SOLID MODEL
$
GRID 401 0 0. 0. 0. 123456
GRID 402 0 1. 0. 0. 3456
GRID 403 0 I. I. 0. 3456
GRID 404 0 0. 1. 0. 13456
GRID 411 0 0. 0. I. 12456
GRID 412 0 1. 0. I. 456
GRID 413 0 I. 1. I. 456
GRID 414 0 0. 1. 1. 1456
CHEXA 401 400 40 1 402 403 404 41 1 412 +E41
+E41 413 414
PSOLID 400 420
MAT9 420 7.9812E6-6.149E50. 1.5136E60. 0. 4.9539E6+M41
+M41 0. 1.5136E60. 0. 1.E6 0. 0. 0. +M42
+M42 3.0824E60. 0. 3.E5 0. 3.E5
CREEP 420 CRLAW +CP41
+CP41 222 3.476-4 2.08-4 2.085-11 2.094 1.02-11 7.43-4
FORCE I 402 0 0.25 1.+4 0.0 0.0
FORCE 1 403 0 0.25 1.+4 0.0 0.0
FORCE I 412 0 0.25 1.+4 0.0 0.0
FORCE 1 413 0 0.25 1.+4 0.0 0.0
ENDDATA
ID ANISOT, V66 $ SSH 3/25/91
TIME 5
SOL 66
CEND
$
TITLE = MAT2 COMBINED WITH MATT2 AND MAT9 WITH MATT9
SUBTITLE = 2 DISCRETE MODELS, EACH CONTAINS A QUAD4 OR HEXA ELEMENT
LABEL = UNIAXIAL TENSION
ECHO = SORT
SEALL = ALL
DISP = ALL
STRESS = ALL
TEMP(INIT)=IOO
$
SUBCASE 100
NLPARM = I00
LOAD = I01
SUBCASE 200
TEMP (LOAD)=200
NLPARM = 100
LOAD =I01
$
BEGIN BULK
$
TEMPD 100 100. 200 200.
NLPARM 100 I YES
$
$ PLATE MODEL
$
GRID 201
GRID 202
GRID 203
GRID 204
CQUAD4 201
PSHELL 200
MAT2 220
MATT2 220
TABLEMI 255
100.
TABLEMI 254
100.
TABLEMI 251
100.
TABLEMI 252
100.
TABLEMI 253
100.
FORCE 101
FORCE 101
$
$ SOLID MODEL
$
GRID 401 0 0.0 0.0 0.0
GRID 402 0 1.0 0.0 0.0
GRID 403 0 1.0 1.0 0.0
GRID 404 0 0.0 1.0 0.0
GRID 411 0 0.0 0.0 1.0
GRID 412 0 1.0 0.0 1.0
GRID 413 0 1.0 1.0 1.0
GRID 414 0 0.0 1.0 1.0
CHEXA 401 400 40I 402 403 404
+E41 413 414
PSOLID 400 420
MAT9 420 7.9812E6-6.149E50. 1.5136E60.
+M41 0. 1.5136E60. 0. i.E6 0.
+M42 3.0824E60. 0. 3.E5 0. 3.E5
MATT9 420 451 452 453
tMT41 453
tMT42 455
TABLEMi 455
100. 3.0824+6200. 1.5412+6ENDT
TABLEMi 454
100. 4.9539t6200. 2.4770t6ENDT
TABLEMI 451
100. 7.9812t6200. 3.9906t6ENDT
TABLEMI 452
100. -6.149t5200. -3.075t5ENDT
TABLEMi 453
100. 1.5136+6200. 0.7568+6ENDT
FORCE 101 402 0 l.OE4 0.25 0.0
FORCE 101 403 0 i.OE4 0.25 0.0
FORCE 101 412 0 1.OE4 0.25 0.0
FORCE 101 413 0 l.OE4 0.25 0.0
ENDDATA
Chapter 7

NONLINEAR STATIC ANALYSIS

7.1 USER INTERFACE

Nonlinear static analysis can be performed using SOLS 66 or 106. SOL 106 is a structured
DMAP (utilizing the new executive system constructs) implemented in Version 67. The func-
tionality of SOL 106 is identical t o the old DMAP, SOL 66. Modeling options are compatible
with linear analysis, i.e., A SOL 61 model can be converted t o SOL 66 analysis with a few
additional data entries relevant to the nonlinear analysis. The nonlinear properties and/or
effects are defined by nonlinear material data (e.g., MATS1, CREEP and TABLESl), gap
elements (GAP) for nonlinear interface, and PARAMeter LGDISP for geometric nonlinearity.

7.1.1 Case Control

Each subcase defines a set of loading and boundary conditions, which can be subdivided into
a number of increments. Loading conditions are specified using the same loads as in linear
static analysis, i.e., FORCEi, MOMENTi, PLOADi, GRAV, TEMP, etc. Constraints can be
specified using SPCs and MPCs, which can be changed from subcase to subcase. The enforced
motion can be applied using SPCD or non-zero value in SPC. The specified value for SPCD
or SPC is a measure of displacement with reference t o the undeformed postion. The enforced
motion is also subdivided in the subcase like the incremental loads. The unique data required
for SOL 66 or 106 is supplied on the NLPARM data entry, which controls the incremental
and iterative solution processes. The snap-through analysis or post-buckling analysis can be
performed using arc-length methods which are controlled by parameters in the NLPCI Bulk
Data entry.

The load step is labeled by the cumulative load factor. The load factor varies from 0 t o 1
in each subcase. Then the load step will end with 1, 2, 3, etc. for the first, the second and
the third subcase, respectively. There are advantages t o divide the entire loading history into
many subcases so that each subcase does not have excessive number of increments, e.g., not t o
exceed 20 steps. The data blocks containing solutions can be generated at each increment or
the end of each subcase depending on the intermediate output option specified in the INTOUT
field of NLPARM data entry. Data blocks are stored in the database for output process and
restarts. As such, the database size or the output quantities are better controlled with multiple
subcases. Furthermore, the nonproportional loading can only be achieved by using multiple
subcases.

Output requests for each subcase are processed independently. Requested output quantities
for all the subcases are appended after the computational process for actual output operation.
Available outputs are DISPLACEMENT, OLOAD, STRESS, ELFORCE, GRID point stresses
and SPCFORCE. However, element force and GRID point stresses are available only for linear
elements.

Restarts are controlled by parameters (LOOPID, SUBID and LOADINC) which can be
specified either in the Case Control section or Bulk Data section. Some optional parameters
(BUCKLE, LANGLE, MAXLP) are provided for additional control or capabilities. All the
superelement model generation options and matrix reduction options are allowed for the linear
portion of the structure.

7.1.2 Iteration Control: NLPARM Data

The most crucial data for successful nonlinear static solutions are contained in the Bulk Data
NLPARM entry, which defines strategies for the incremental and iterative solution processes. It
is difficult to choose the optimal combination of all the options for a specific problem. However,
based on a considerable number of numerical experiments, the default option was intended to
provide the best workable method for a general class of problems. Therefore, users with little
insight or experience in a specific application should start with the default option. However,
users with some experience and insight in a specific problem may change the default values
based on the following observations:
Computing cost for each line search is comparable t o that of an iteration.
The SEMI method usually provides better convergence than the AUTO method a t the
expense of higher computing cost.
Default tolerances for the convergence criteria may be somewhat conservative. However,
loose tolerances may cause difficulties in the subsequent steps.
The quasi-Newton method is effective in most problems. However, it seems t o have
adverse effects in some problems, e.g., creep analysis.
The line search method is effective t o cope with difficulties in convergence in some prob-
lems. More extensive line searches may be exercised by a large value of MAXLS and/or
a smaller value of LSTOL. On the other hand, line searches may have adverse effects in
some problems, e.g., plane stress plasticity.
 The arc-length method should be used if the problem involves snap-through or post-
buckling deformation. Then the Bulk Data entry NLPCI must be attached.
The NLPARM data format is shown below with default values wherever applicable:

NLPARM ID NINC DT KMETHOD KSTEP MAXITER CONV INTOUT

NLPARM 0. AUTO 5 25 PW NO +NP1

EPSU EPSP EPSW MAXDIV MAXQN MAXLS FSTRESS LSTOL

+NP1 1.OE-3 1.OE-3 1.OE-7 3 MAXITER 4 0.2 0.5 +NP2

MAXBIS M AXR RTOLB

+NP2 5 20. 20.

The ID field requires an integer t o be referenced by a Case Control command NLPARM.

The NINC field is an integer which specifies the number of increments to be processed in
the subcase. The total load specified in the subcase minus the load specified in the preceding
subcase is equally divided by this integer (NINC) t o obtain the incremental load for the current
subcase. Another subcase should be defined to change constraints or loading paths. However,
multiple subcases may be required in the absence of any changes in constraints or loads to use
a moderate value (e.g. not to exceed 20) for NINC. Use of a moderate value has advantages
in controlling database size, output size, and restarts.

The D T field requires a real number specifying the time increment for each load step in the
case of creep analysis. The time unit must be consistent with the unit used in the CREEP data
t o define the creep characteristics. The creep time increment should be determined properly
for efficiency and accuracy based on the creep rate represented by the creep law to be used.
The larger the creep rate, the smaller the time increment should be. No creep deformation is
considered with the default value of 0. It is noted that a creep analysis should be preceded by
a static analysis with at least one subcase, because the creep deformation requires stresses in
the material.

Stiffness matrix update strategies are determined by a combination of the data specified in
the two fields KMETHOD and KSTEP. Options for KMETHOD are AUTO, SEMI, or ITER.
The KSTEP field, which is an auxiliary to the KMETHOD field, should have an integer greater
than 1. With the AUTO option, the program automatically determines when to reevaluate the
stiffness matrix based on the rate of convergence. At each iteration, the computing time for
convergence without the stiffness matrix update is estimated and compared with the computing
time for the matrix update in order t o determine whether the update is more efficient. This
decision is deferred in the first two iterations after a new stiffness is obtained. If the solution
tends to diverge, however, the update decision will be made effective immediately. The stiffness
matrix will be updated upon convergence if the number of iterations required for convergence
is greater than KSTEP. The SEMI option is identical t o the AUTO option except for one
additional stiffness update after the first iteration which always occurs unless the solution
converges in a single iteration. With the ITER option, the stiffness matrix is updated at every
KSTEP iterations. Thus, the full Newton-Raphson iteration is exercised if KSTEP is 1. The
ITER method also allows a stiffness matrix update upon convergence (since Version 66) when
KSTEP < MAXITER. If KSTEP > MAXITER, the stiffness will never be updated. It is
noted that the original modified Newton's method is selected when KSTEP = MAXITER.

The MAXITER field is an integer representing the number of iterations allowed for each
load increment. If the number of iterations reaches MAXITER without convergence, the load
increment is bisected and the analysis is repeated. If the load increment cannot be bisected
(i.e., MAXBIS is reached or MAXBIS=O) and MAXDIV is positive, the best attainable solution
is computed and the analysis is continued to the next load increment. If MAXDIV is negative,
the analysis is terminated.

The convergence test is performed at every iteration with the criteria specified in the CONV
field. Any combination of U(for displacement), P(for load), and W(for work) may be specified.
All the specified criteria must be satisfied t o achieve convergence, except for an absolute
convergence condition, under which the solution is converged regardless of criteria. This subject
will be elaborated in the following section. The convergence tolerances are specified in the fields
EPSU, EPSP, and EPSW for U, P, and W criteria, respectively.

The INTOUT field requires a specification of YES, NO or ALL. If YES is specified, the
output requests (DISPLACEMENT, ELFORCE, STRESS, etc.), specified in the Case Control
Data, will be processed for every computed load step in the subcase. If NO is specified, the
output requests will be processed only for the last load step of the subcase. If ALL is specified,
the output requests are processed for every computed and user specified load increment. For
Newton's iteration methods (i.e., without NLPCI), the option ALL is equivalent t o the option
YES, since the computed load increment is equal t o the user specified load increment. For arc-
length methods (i.e., NLPCI command is specified), the computed load increment in general is
not the same as the user specified load increment, and is not known in advance. The option ALL
allows the user t o obtain solutions at the desired intermediate load increments. The INTOUT
data also affects the database if the PARAMeter SMALLDB (valid only in Version 65) has
a value of 1. With a small database option, the database will retain only those data blocks
corresponding t o the incremental load steps for which output is requested by an INTOUT field.
Then, the restarts will be restricted t o those starting from these steps. However, if the job is
terminated with an incomplete subcase for some reason (e.g. diverging, time expiration, etc.),
data blocks for all the interim steps of that subcase will be stored in the database to allow
flexible restarts.

The MAXDIV field requires an integer to specify a limit on the probable divergence con-
ditions allowed for each iteration to continue. There are two classifications in the divergence
condition: probable and absolute. The absolute divergence is treated as two occurrences of
the probable divergence. When the probable divergence occurs MAXDIV times or more and
if MAXDIV is positive, the current solution base (displacements) is retracted and the stiffness
matrix is updated in reference to the preceding iteration step. If the solution diverges again
in the same load increment while MAXDIV is positive, the best attainable solution is com-
puted and the analysis is continued to the next load increment. If MAXDIV is negative, the
analysis is terminated on second divergence. Refer to Section 3.3.2 for a detailed description
of divergence criteria.

The MAXQN field requires an integer to specify the maximum number of quasi-Newton
vectors t o be saved in the database. Quasi-Newton vectors are accumulated, if desired, until
MAXQN is reached. If the number of accumulated QN vectors is greater than or close to
MAXQN upon convergence, the stiffness matrix will be updated before the next increment is
processed. All the QN vectors already accumulated are purged upon a stiffness update and
the accumulation is re-initiated. The quasi-Newton update may be suppressed with a value of
0 for MAXQN. Details of the quasi-Newton method will be presented in the following section.
The line search is controlled by the fields MAXLS and LSTOL. The maximum number of
line searches allowed for each iteration is specified in the MAXLS field. The line search process
may be suppressed with a value of 0 for MAXLS. The LSTOL field requires a real number
between 0.01 and 0.9 t o specify a tolerance for the line search operation. The line search
operation will be conducted if the error defining the divergence rate is greater than LSTOL. If
the line search convergence is not achieved, the line search continues until the number of line
searches reaches MAXLS.

The FSTRESS field requires a real number (0.0 < FSTRESS < 1.0), representing a frac-
tion of the effective stress ( a ) used to limit the subincrement size in the material routines.
The number of subincrements in the material routines for elasto-plastic or creep deforma-
tion processes is determined such that a subincrement in the effective stress is approximately
FSTRESS*a. This data is also used t o establish an error tolerance in the yield function t o
alleviate accumulation and propagation of the error. If the error in the yield function exceeds
FSTRESS*B at the converging state, the program will terminate the job with a fatal error mes-
sage "ERROR EXCEEDED 20% OF CURRENT YIELD STRESS.." If the bisection option
is selected, bisection will be activated under this condition.

The MAXBIS field requires an integer t o specify the number of bisections allowed for a
load or arc-length increment. When the solution diverges, different actions are taken depending
on the sign of MAXBIS. If MAXBIS is positive, the stiffness matrix is updated on the first
divergence and the load increment is bisected on the second divergence. If MAXBIS is negative,
the load increment is bisected every time the solution diverges until the limit on bisection is
reached. If the solution does not converge after (MAXBISI bisections, the analysis is continued
or terminated depending on the sign of MAXDIV.

The MAXR field requires a real number to specify the bounds on the ratio of the adjusted
arc-length increment relative t o the initial value. In the adaptive load/arc-length increment,
the overall upper and lower bounds on the load/arc-length increment in the subcase are defined
as
1 Aln
< - < MAXR
MAXR - Al0 -
where Al, is the arc-length at step n and Alo is the original arc-length. The arc-length method
for load increments is selected by the NLPCI Bulk Data entry, which must have the same ID
as the NLPARM Bulk Data entry.

The RTOLB field requires a real number to specify the incremental rotation (in degrees)
allowed per iteration. The bisection is activated if the incremental rotation for any degree-of-
freedom (Ad,, Ad, or Ad,) excceeds the value specified for RTOLB. This bisection strategy
based on the incremental rotation is controlled by the MAXBIS field.

7.1.3 User Interface for Arc-Lengt h Met hods: NLPCI

Most of the parameters in the NLPARM Bulk Data entry are used t o control the iteration
strategy for arc-length methods. However, MAXLS is not applicable because the line search
procedure is not coupled with arc-length methods currently. The parameters which are appli-
cable only t o the arc-length methods are specified in the Bulk Data entry NLPCI, in connection
with the NLPARM data by the same ID. The NLPCI entry is shown below with default values,
followed by brief description of the fields:

NLPCI ID TYPE MINALR MAXALR SCALE DESITER MXINC

NLPCI CRIS 0.25 4. 0. 7 20

ID : must be identical to ID of NLPARM to be active.

TYPE : Constraint type (CRIS for Crisfield, RIKS for Riks, or MRIKS for modified Riks
method)

MINALR, MAXALR : For variable arc-length method, the arc-length is bounded by

MINALR 5 -5 MAXALR
Atold
where MINALR 5 1. and MAXALR > 1. The automatic adjustment is not activated if
MAXALR=MINALR=l.
 >
SCALE : Scaling factor for the magnitude of the load term in arc length. Real O., Default =
0. The constraint equation has a disparity in the dimension by mixing the displacements
with the load factor. The scaling factor (w) is introduced as user input so that the user
can make constraint equation unit-independent by a proper scaling of the load factor,
p . As the value of w is increased, the constraint equation is gradually dominated by the
load term. In the limiting case of infinite w, the arc-length method would be degenerated
t o the conventional Newton's method.

DESITER : Desired number of iterations for convergence to be used for adaptive arc-length
adjustment.

MXINC : Maximum number of controlled increments allowed within the specified load step.
There can be cases where the structure never snaps back and collapses t o failure. In
these cases, the analysis cannot be completed without bounding the maximum number
of increments. For such cases, this value should be confined to a reasonable limit, e.g.,
less than 100, t o prevent inadvertent depletion of the computing resources.

7.1.4 Iteration Related Output Data

At the end of every iteration, the relevant data from the iteration process are printed under
the following heading:
ITERATION : iteration count i

EUI : relative error in terms of displacements (See Section 3.6.3)

EPI : relative error in terms of loads (See Section 3.6.3)

EWI : relative error in terms of work (See Section 3.6.3)

LAMBDA : rate of convergence (defined in Section 3.3.1)

DLMAG : absolute norm of the load error vector

FACTOR : final value of the line search parameter (See Section 3.4)

E-FIRST : divergence rate, initial error before line search (See Section 3.4.3)

E-FINAL : error at the end of line search (See Section 3.4.3)

N-QNV : number of quasi-Newton vectors appended (See Section 3.5.3)

N-LS : number of line searches performed during the iteration

ENIC : expected number of iterations for convergence (See Section 3.3.1)

NDV : number of occurrences of probable divergence during the iteration (See Sec-
tion 3.3.2)

MDV : number of occurrences of bisection conditions in the material routine (exces-

sive stress increment) or in the rotation angle (excessive rotation) during the iteration
7.2 ITERATION MODULE AND SOLUTION SEQUENCE

In the heart of the nonlinear static solution sequences 66 or 106, there is a nonlinear module
which performs incremental and iterative processes. The solution sequence for nonlinear static
analysis (SOL 66) is depicted by flow diagrams in Figs. 7.2.1 and 7.2.2. The module, NLITER,
has evolved for a decade since Version 61. The DMAP interface, input/output data blocks and
parameters as of Version 67 are described below.

7.2.1 DMAP Interface of NLITER Module (Version 67)

NLITER CASES, CNVTST, PLMAT, YSMAT, KAAL, ELDATA,

KELMNL, LLLT, GMNL, MPT, DIT, MJJ, SLT1, CSTMS,
BGPDTS, SILS, USET / UGVMO, FGNL, ESTNLH,
CIDATA, QNV, FFGVH, MUGV, MESTNL, DUGVMO /
S,N,LOADFAC/ S,N,CONV=O/ S,N,RSTEP/ S,N,NEWP=+l/
S,N,NEWK=+l/ S,N,POUTF/ S,N,NSKIP/ V,Y,LGDISP/
S,N,MPTSIZE/ S,N,ITERID/ S,N,KMATUP/ S,N,LSTEP/
S,N,KTIME/ S,N,SOLCUR/ V,Y,TABS/ S,N,KFLAG=+I/
S,N, NBIS=O/ S,N, NBISO=O/ S,N, CURMU/
S,N,DELMU=l./ S,N,ARCLG=O./ S,N,ARCSIGN=l/
S,N, TWODIV=O/ V,N,LANGLES $
Note: New data in Version 67 are highlighted in italics.

7.2.2 Input Data Blocks

C A S E S Case Control data (Case Control subset for residual).
+
C N V T S T Convergence test vectors : AP,, Pg Qg, diag(KLLT).
P L M A T Load vectors for subcase (initial and final), !-size.
Y S M A T Constraint displacement vectors for subcase (initial and final), s-size.
K A A L Reduced element stiffness matrix (linear element only).
E L D A T A Element combination data, consisting of element thermal loads and element con-
nectivity (used in SDRNL).
K E L M N L Nonlinear element stiffness matrix.
LLLT Lower Triangular Factor for nonlinear elements (material and differential if positive
definite, else only material stiffness unless the arc-length methods).
G M N L Multipoint constraint equation matrix.
MPT Material Properties Table (NLPARM and NLPCI data included).
DIT Direct Input Table.
MJJ Mass matrix (g-size).
SLTl Static Load Table for follower forces.
CSTMS Coordinate System Transformation Table.
BGPDTS Basic Grid Point Definition Table.
SILS Scalar Index List.
USET Displacement Set Definition Table.
Note: The following data blocks may be purged - YSMAT (on first solution), GMNL, DIT,
SLT1, CSTMS, BGPDTS.

7.2.3 Output Data Blocks

UGVMO Input/output, displacement vector in g-set at converged step.

FGNL Nonlinear element force from last iteration.
ESTNLH Input/output, updated table of ESTNL at converged step.
CIDATA Input /output, miscellaneous data for arc-length methods.
SECALG Special root-selection flag for the false reversal case.
FULL Constraint status, currently used only with modified Riks.
1 signifies that the solution is on the constraint;
2 that it is off the constraint.
CURDEL The current value of the arc length.
ORGDEL The value of the arc length at the beginning of the subcase.
NEWMU The current value of p.
DMUAST If nonzero, the amount by which the projected value of p
exceeds the specified output value of p.
DENOML For the Riks method, the last denominator in the equation for the
computation of A p . For the other methods, DENOML is set t o zero.

{ A u : ) for the Riks method

{ u l - uO), (normalized), used for the second Crisfield root-selection algorithm
No data store for the modified Riks method

QNV Quasi-Newton vector file.

FFGVH Follower force vector for OLOAD output.
MUGV Displacement vector for K-update.
MESTNL Input/output, nonlinear Element Summary Table for K-update.
DUGVMO Input/output, incremental displacement vector in g-set between converged steps
(17,- Un-l),
to be stored at the output steps.

7.2.4 Parameters

LOADFAC Input/output-complex. Load factor for current module iteration, fractional value
between 0 to 1.

CONV Input/output-integer. On Input : Last iteration value (0 for initial iteration). On

Output : -1 if solution did not converge; $1 if solution converged.

RSTEP Input/output-integer. The count of arc-length increments.

NEWP Input /output-integer. +1 for a new subcase, -1 otherwise.
NEWK Input/output-integer. On Output : -1 if old stiffness matrix t o be retained; > 1 if
new stiffness matrix required.
POUTF Output-integer. On Output : -1 if stresses are not t o be printed; $1 if stresses are
t o be output. (material nonlinear elements)
NSKIP Input/output-integer. The number of records to skip into the Case Control record
(subcase ID). The value is set to -2 upon NOGO flag

LGDISP Input-integer. Problem nonlinear type flag. On Input : +1 for geometric nonlinear
problem. -1 otherwise.
MPTSIZE Input/output-integer-default 0. (No longer used in Version 67).
ITERID Input/output-integer. Iteration count.
KMATUP Input/output-integer. The current count of the stiffness matrix updates within
the increment.
LSTEP Input/output-integer. The value of the load step within a subcase. The value is
assumed t o be one at the start of a subcase and is incremented by one each time the
module converges. When LSTEP = NINC from the NLPARM card, NEWP is set fl.
KTIME Input/output-integer. The time-to-go value a t the start of the stiffness matrix for-
mulation loop within the DMAP. On output from the module, a negative value is stored,
denoting the increment of time between stiffness computation and module start.

SOLCUR Input/output-integer. The DMAP loop number at the current iteration step.
TABS Input-real. Absolute temperature conversion constant for creep analysis.
KFLAG Input/output-integer. Flag to invoke the K update at last converged step. Initialized
as $1 for the cold start and -1 for the restart in the DMAP. In the module it will be set
t o + l when the iteration is converged and K is updated. Set t o -1 when the iteration is
converged but K is not updated, or K is updated with CONV=-1.

NBIS Input/output-integer. The current count of the bisection. The value is set t o 0 at the
start of each user specified load step.

NBISO Input/output-integer. The previous count of the bisection. The value is set to 0 at
the start of each user specified load step and is set equal t o NBIS when the solution
converges.

CURMU Input/output-real. Current value of the load factor in the subcase. The value is
set to 0 at the beginning of the subcase.

DELMU Input/output-real. The ratio of current load increment t o the user specified load
increment. The value is set to 1 at the start of each user specified load step.

ARCLG Input/output-real. The arc-length at converged step, t o be stored in the database.

ARCSIGN Input/output-integer. The sign of ( P ) ~ { A P at
) the beginning of the subcase.
This is used in restarts in the post-buckling region.

TWODIV Input /output-integer . Number of diverging solutions in the same load step. If
this number reaches two, the solution process is terminated.

LANGLES Input-integer. Used to select the rotation vector approach t o large rotations by
a value of 2. The default (= 1) option is the gimbal angle approach.

7.2.5 Diagnostic Outputs

Diagnostic output is available in SOL 66 (or 106) if DIAG 50 or 51 is specified in the Executive
Control Section. The data that may be produced are listed below. When DIAG 50 is requested,
only the data marked with a plus (+) are printed. When DIAG 51 is selected, all the data are
printed at each iteration, allowing the user t o follow the nonlinear iteration calculations step
by step. Because of the volume of output produced, the use of DIAG 51 is recommended only
for small test problems. DIAG 51 is intended for debugging purposes, and users in general are
cautioned against its use.

The data marked with an asterisk (*) are used only for the arc-length method whereas the
data marked with a bullet ( 0 ) are used only for Newton's method.

For each entry into NLITER:

+ Subcase status data
+ NLPARM/NLPCI data
 Core statistics (ICORE, etc.)
Problem statistics (g-size, etc.)
File control block
Input file status
External load increment for subcase: A P
* Displacement increment: Aup
*+ Initial arc-length
Initial nonlinear force vector: Fg
Initial sum of nonlinear forces including follower forces: Fl
Initial displacement vector: ul
KFSNL*DELYS: [ L ~ ~ ] ~ { A u ~ }
Initial unbalanced load: Rl
* Cumulative displacement increment: {u,+~ - u,}
* Dot product determining sign: { A ~ ~ } ~ { u - ,u,)+ ~
* Projected NEWMU: pi+1

For each iteration:

Displacement increment: Aul
Initial energy
New displacement vector: ug
Nonlinear force vector: F,
Sum of nonlinear forces including follower forces: Fl
New displacement vector: ul
New unbalanced load: R1
Denominator of EUI
Denominator of EPI
Contraction factor: q
Remaining time
* Displacement increment: A U ~
* Arc-length ratio
* d p calculations

For each quasi-Newton vector set:

Condition number: X2
quasi-Newton vector: 6
quasi-Newton vector: y
1
Energy error: z = -
67.3

For each line search:

Previous line search factor: ak
Previous error: Ek
New line search factor: ak+1
 Save wid k p ~ e n df i l s
Initialize parameters
I

.
4.
Fetchdelta blocks
(W, *sTNL, m L )
9 S O U = SOLCW*l

-
+ .
ITERID * 0, WTW O
@5EQUIV PTELtWL, PPTEVII
i

Fur* d a t w e for 9 1 A U
St paruieters
Process M 8 and i f clone
4 (END)
Fetc?~W, N(;GV, NPGX, tiCJECT1
conpute load factor
Cenerhte 1086 vector
(DCttC.1, MLFG, aLLW
PPTEW, P E W L )
Chea ccrtstraint m q e

sinplsri ty process
1 for +set poinu
9 Reset LET parameters
USTPRT i f reqWteU
RWuce KGG to K M L
I

ns Set the clock for K upate

' Icc.pr
L
9 F O T ~EUWL,
rorn
KPICTNL. KGCNL
naft tor LGDIS
I

FB6UC;O KMmL t o KAANL

KLLT = M L + KAFINL YES
Form KFST and KSST 4
R B C U KDMi to KDLL for LWISP
K L U = KLLT + KDLL fOT W I S P
&cmrne KLLT or KLLR -b U L T
L

Figure 7.2.la Flow Diagram for Nonlinear Process in SOL 66

 .
-
m
P U = ~+~
DELF'M = DELPGMUGJ
Reduce trELPC;1 t o ELPLU
Emand MLPLV i n t o W V i l
A+ lRUiL #la PFI. t O DPLG
h l y for tiw Subcase
1

L
.
POUTOLD = PWTT*-I
ISELPL = D E L P L V + K W L +-
YES 1
v

.
+
MLPL = MERRPL

+
8 Fetm M U , U W , PTEUINL, PPTEVVI

NLCOHB - EUHTA, SLTi

NUTER - UGWtD, DFGNL, E m L , ERW'L, W

4
. Reset parmeters
Store ELDATA 0- for all
Update UGV = HUGV-MWS
only i f o-set exists

hpute = F -P +K U +K Y
S S f S f S s s

.
uplate W C T I
store CVECT1, WX, QW
Incrmt S O M

Figure 7.2.lb Flow Diagram for Nonlinear Process in SOL 66

 Fetcn ELDATk
8BCID = 0
nwm = o

E
ND = S)LPM-NU)@

K S
LEND = -1

oxpsi te e l m t
force, failure im
w sues recovery

4
A#lend O E S 4 U X -+ OESHU
AppClM 11CV +
6~- PGX -+P.R
Amend QCV -b QGS
AFpend C'uECTl --b CD16
AFpend N F N U -+ OEFNLU
qpem OESNU -+ OESNUX'

Figure 7.2.2 Flow Diagram for Appending Process in SOL 66

7.2 - 8
7.3 LOADS OVERVIEW

The static loads in nonlinear analysis consist of a subset of the total MSCINASTRAN static
load set. Most of the relevant loads data applicable t o the linear static analysis are also
applicable t o nonlinear static analysis except for the Bulk Data entry DEFORM. The enforced
displacements can be applied as loads using the Bulk Data entry SPCD, which defines an
enforced motion of a grid in a specific fixed direction. Each grid point with an enforced
displacement must also appear on an SPC or SPCl entry.

The loads are selected in each subcase by a Case Control command LOAD. The load for
a subcase is subdivided into the number of increments specified for the subcase. The solution
strategy in nonlinear is to apply the loads in an incremental fashion until the desired load level
is reached. The algorithms "remember" the loads from one subcase to the next. If the load
reaches the desired level in a subcase and if the load description is left out in the subsequent
subcase inadvertently, the nonlinear solution algorithm will begin in an incremental fashion to
remove the load. The loads described below make up the valid subset applicable for nonlinear
static analysis.

7.3.1 Concentrated Loads

The concentrated loads consist of two categories: those which are stationary in direction
through out the analysis and those which follow the grid motion. The first category consists
of the following:

FORCE, MOMENT, and SLOAD.

The second category consists of the following:

FORCEl, FORCE2, MOMENTI, and MOMENT2.

The three FORCE entries differ only in the manner in which the direction of the force
is specified. FORCE uses the components of a vector. FORCE1 uses two grid points, not
necessarily the same as the loaded grid points. FORCE2 defines the direction of the force as
a vector which is the vector product of two other vectors. The distinctions between the three
MOMENT entries are similar. The SLOAD entry is used t o apply loads t o scalar points.

7.3.2 Distributed Loads

The distributed loads consist of two categories: those which are stationary in direction through
out the analysis and those which follow the element motion. The first category consists of the
The second category consists of the following:

PLOAD, PLOAD2, and PLOAD4.

Loads may be applied to the interiors of finite elements. The PLOADl entry defines
concentrated and linearly distributed forces and moments to the BEAM element. However,
the stress calculation for nonlinear BEAM element ignores the distributed loads, but lump
them onto the nodal points. The PLOAD entry is exceptional in that it references three or
four arbitrarily located grids rather than an element. The PLOAD2 entry defines a uniform
static pressure load applied to the QUAD4 or TRIA3 elements. The PLOAD4 defines a pressure
load on surfaces of HEXA, PENTA, TRIA3, and QUAD4 elements which need not be normal
to the surface.

7.3.3 Mass Related Loads

The mass related loads are gravity and centrifugal loads, which require specification of mass
data. The gravity loads which can be specified in the Bulk Data entry

GRAV

are stationary in direction through out the analysis. It is used to define the direction and
magnitude of the gravity vector in some identified coordinate system. The components of the
gravity vector are multiplied by the mass matrix t o obtain the components of the gravity force
a t each grid point.

The centrifugal loads t o be specified in the Bulk Data entry

RFORCE

is used t o define the components of a spin vector which is used internally t o compute centrifugal
forces. Each component of the spin vector is multiplied by a scale factor.

7.3.4 Thermal Loads

Thermal loads are selected in each subcase by the Case Control command TEMP(L0AD).
Thermal loading which is a follower loading is accomplished through the following entries:

TEMP, TEMPD, TEMPPI, TEMPP3, and TEMPRB

The TEMP and TEMPD provide for grid point temperature specification which are then in-
terpolated to points within elements. For the nonlinear QUAD4 and TRIA3 elements the
T E M P P l provides for an average temperature and thermal gradient in the thickness direction
and the TEMPP3 provides for temperature at points through the thickness. The TEMPRB
provides for average temperature at ends A and B for the ROD and BEAM element. Aver-
age temperature specified directly for an element will take precedence over the temperature
interpolated from the elements connected grid points.

There was a limitation until Version 65 which required specification of only one increment
in the subcase where thermal load is first applied or removed. This limitation was removed in
Version 66. This limitation was caused by the initial temperature which was specified in the
material reference temperature field. In addition, the material temperature, TEMP(MAT),
was distinguished from the loading temperature, TEMP(LOAD), until Version 65. Segregat-
ing these temperatures is not meaningful in the nonlinear analysis. In order to maintain a
consistent temperature field, TEMP(MAT) was defaulted t o TEMP(L0AD) in Version 66.
In addition, a Case Control command TEMP(IN1T) was created for the initial temperature
distribution. For upward compatibility, Version 66 allowed TEMP(MAT) option in SOL 66,
which caused a number of errors. For these reasons, TEMP(MAT) option was abandoned in
Version 67 and TEMP(IN1T) was made a required data in case any thermal load is present in
the analysis.

The Case Control TEMP(IN1T) is required above subcase level. The use of TEMP(MAT)
or TEMP(B0TH) will cause a fatal error. Subcases that do not contain a TEMP(L0AD)
will default to TEMP(INIT), which implies no thermal load. Thermal loads are often used t o
provide initial strains.

7.3.5 Loads Combination

Various load specifications can be combined using

LOAD and (CLOAD, LSEQ).

The LOAD provides for the linear combination of static load sets. The static loads (LOAD,
FORCE, etc.) applied t o the upstream superelements cannot be referenced by a Case Control
command LOAD. The CLOAD entry is designed t o apply static loads t o upstream superele-
ments by combining loads defined in LSEQ entries. The (CLOAD, LSEQ) combination allows
the nonlinear algorithms t o apply in an incremental fashion on the upstream superelement
loads to the boundary of the residual structure.

The CLOAD defines a static load combination for superelement loads acting on residual
boundary and the LSEQ defines a static load combination for superelement upstream loads.
The LSEQ assigns load vectors t o the superelements and numerically labels them via the
DAREA field entry value. The LID and TID field entries point t o the appropriate load and
temperature distribution entries. The CLOAD defines a static load as a linear combination
of previously calculated superelement loads defined by the LSEQ. The IDVi fields correspond
directly t o the numeric labels defined via the DAREA fields of the LSEQ entries.

The CLOAD must be selected in the residual solution subcase via the case control CLOAD
entry. Any loads referenced by CLOAD should not be again referenced by LOAD until Ver-
sion 66, otherwise the load will be doubled: e.g., GRAV and TEMP. This deficiency has been
corrected in Version 67. The LSEQ is selected by a LOADSET Case Control command above
any residual subcase. Only one LOADSET may appear in Case Control. The hierarchy of the
loads data is shown schematically in Fig. 7.3.1. An example for the Case Control set-up is
shown below:

SEALL=ALL
LOADSET = 1000 $ Selects LSEQ 1000 for upstream loads
SUPER = ALL $ Identify SEs to process

DISP = ALL

$ NONLINEAR SOLUTIONS FOR RESIDUAL SUPERELEMENTS

SUBCASE 10
CLOAD = 1001 $ Refers to CLOAD Bulk Data
NLPARM = 12 $ Iteration control
..
SUBCASE 20
CLOAD = 1002
LOAD = 10 $ Residual SE load
NLPARM = 22
 Upstream SE

7 CLOAD

( Bulk CLOAD 1

defined
on LSEQ
DAREAl DAREA2

correlated
t o CLOAD

LSEQ +

LL Static Loads

Figure 7.3.1 Hierarchy of Load Data

7.4 NONPROPORTIONAL LOADING

The loading condition under which all the stress components increase in the same proportion
is called proportional loading, i.e.,

The stress gradient vector 1%) remains unchanged during proportional loading, and the stress
state may be expressed in terms of any one stress component. It can be shown that the
stress-strain relations reduce to stepwise-linear equations in the case of proportional loading.
Proportional loading is in general difficult t o achieve except for very simple and special cases
such as uniaxial loading. Notice that the linear strain path does not always result in the
proportional loading condition.

7.4.1 Validation of Nonproportional Loading Case

The general cases of nonproportional loading add more complexity to nonlinearity. Two classes
of problems in this category can be contemplated: a strain path dependent problem controlled
by a geometric (or essential) boundary condition and a stress path dependent problem con-
trolled by a force (or natural) boundary condition. Both cases were validated by a simple
model of a single HEXA element with eight Gauss quadrature points using the von Mises
yield criterion and the isotropic hardening rule. In the stress path dependent problem, the
loading was applied via the LOAD and PLOAD4 Bulk Data entries while in the strain path
dependent problem, the SPC and S P C l Bulk Data entries specified the loading conditions.
The material properties were specified using the MAT1 and MATS1 Bulk Data entries and the
solution strategy was provided by the NLPARM Bulk Data entries. The input data are shown
in Tables 7.4.1 and 7.4.2 for the stress and strain path dependent problems, respectively.

First, the solution was obtained for a prescribed stress path (shown in Fig. 7.4.l(a)), and
the strain path was plotted in Fig. 7.4.l(b). The solution was validated by obtaining a solution
to an inverted problem, in which the stress path was measured while the strain path was traced.
As illustrated in the figures, discrepancy between the two solutions is not discernible. Notice
that points A and B in Fig. 7.4.1 signify the elastic limit and the turning point into the
nonproportional loading, respectively.

By virtue of the subincremental scheme within the material routine, the nonproportional
loading did not pose any difficulty. In order t o examine the accuracy with respect t o the
number of increments in the nonproportional loading region, a parametric study was performed
by taking a various number of global increments (2 t o 10) between points B and D. Errors are
measured at points C and D for five different cases as tabulated below:
 Error in Stresses (%)
NO. of (5, 0 2

Inc. Point C Point D Point C Point D

2 0.486 0.162 0.185 0.208

I I I I I I

Note: Errors in B are less than 0.0001 %.

It is observed that smaller increments tend to reduce the error. However, an acceptable solution
was achieved with only two increments. Effectively, the entire loading path could have been
analyzed in three steps (one in proportional and two in nonproportional loading) for engineering
accuracy. The fact that the solution did not improve from eight to ten steps indicates that the
number of subincrements is reduced when the global increment size is sufficiently small due t o
the adaptive subincremental scheme.

7.4.2 Biaxial Loading Case for Neutral Loading

Neutral loading, which is a special case of nonproportional loading, is verified by a biaxially

loaded plate under the in-plane, plane-stress condition using the von Mises yield criterion and
kinematic hardening. The finite element model and the material properties are shown in Fig-
ures 7.4.2(a) and 7.4.2(b). The plate is modeled by four QUAD4 elements with the material
properties specified in the MAT1 and MATS1 Bulk Data entries. Appropriate boundary con-
ditions are provided by the S P C l and MPC Bulk Data entries. The loading is applied via the
FORCE Bulk Data entries and the solution strategy is specified in the NLPARM Bulk Data
entries. The input data are given in Table 7.4.3.

The loading path is shown in Fig. 7.4.2(c). The loading sequence starts with a uniaxial
tension in the x-direction t o 40.5 ksi, followed by a slight unloading to 40 ksi. The subsequent
loading path is defined by an elliptic locus a; - a,a, + at = 402. Although the locus seems
t o be a yield locus, it is slightly different because of the kinematic hardening rule employed in
the analysis. Therefore, the second loading phase is nearly, but not exactly, neutral loading.

The analysis was performed in 36 increments (7 in radial loading and 29 in nonproportional

loading), which ran t o completion without any difficulty in convergence. The incremental
process progressed with 11 stiffness matrix updates, 145 iterations and 36 line searches. It is
noted that increments 12 and 13, where the incremental step size is larger than other steps,
required more iterations (7 and 10, respectively) and line searches (7 and 18, respectively).
The solution of the strain path (E, vs. E,) corresponding t o the loading path is compared with
the one presented by Chern [7.1], as shown in Fig. 7.4.2(d). The stress-strain response curves
in x and y components are plotted in Fig. 7.4.2(e). The response is highly nonlinear during
the second loading phase due t o the nonproportional loading path.
 a, (ksi)

7 Initial Yield Locus

a, (ksi)

-&- Stress Path Specified

+ Strain P a t h Specified (Inverted Case)

(a) Stress Path

(b) Strain Path

Figure 7.4.1 Validation of Nonproport ional Loading Case

 MPC RELATIONS(uy7 = uYs
2----------8------- 1
9
I
I
1
I
+ MPC RELATIONS
BI Ell I (ux3 = u x 6 = u x 9 )
II
I
IP
5
p4 I
II
II
I
1 &2

Width = Length = 1. inch

Thickness = 0.001 inch

Figure 7.4.2(a) Finite Element Model for Biaxial Loading

E = 30.0 x lo3 ksi

v = 0.3
H = 1.5 x lo3 ksi
Kinematic Hardening

7.4.2(b) Material Curve for Biaxial Loading

 bv (ksi)
w

a, (ksi)

Figure 7.4.2(c) Biaxial Loading Path

STEP NO.
1
2
3
4
5
6
7
8 to 36
 Chern
0 MSCINASTRAN

Figure 7.4.2(d) Corresponding Strain Path for Biaxial Loading

a, or a, (ksi)

Figure 7.4.2(e) Stress-Strain Response Curves

 Table 7.4.1 Input Data Listing for Stress Path Dependent Case
ID NLPATH2,V66 $ CJS 7-MAY-85, SHL 1981
SOL 66 $ NONLINEAR STATIC ANALYSIS
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
TIME 5 $ CPU MIN FOR VAX
CEND
TITLE = CODE TEST FOR NONLINEAR ELEMENTS WITH STRESS PATH 2
ECHO = UNSORT
SEALL = ALL
DISP = ALL
OLOAD = ALL
SPCF = ALL
STRESS(PL0T) = ALL
SPC = 100
SUBCASE 1195
LABEL = UNIAXIAL COMPRESSION
LOAD = 1195
NLPARM = 10
SUBCASE 1196
LABEL = PLANE STRAIN TENSION
LOAD = 1196
NLPARM = 20
OUTPUT(PLOT)
CSCALE = 1.3
PLOTTER NAST
SET I = ALL
VIEW 30., 20., 0.
PTITLE = ISOMETRIC VIEW
FIND SCALE, ORIGIN I, SET I
PLOT SET I, ORIGIN I, LABEL BOTH
PLOT STATIC 0, SET I, ORIGIN I
VIEW O., O., 0.
PTITLE = FRONT VIEW
FIND SCALE, ORIGIN I, SET I
PLOT SET 1, ORIGIN I
PLOT STATIC 0, SET I, ORIGIN I
AXES Z, Y, MX
PTITLE = TOP VIEW
FIND SCALE, ORIGIN I, SET 1
PLOT SET 1, ORIGIN1
PLOT STATIC 0, SET I, ORIGIN I
BEGIN BULK
$
NLPARM 10 2 AUTO UPW YES
NLPARM 20 4 SEMI 20 UPW YES
$$ GEOMETRY
GRID 1000 0. 0. 0.
GRID 1100 I. 0. 0.
GRID 1110 I. I. 0.
GRID 1010 0. I.
GRID 1001 0. 0.
GRID 1101 I. 0.
GRID 1111 I. I.
GRID 1011 0. I.
GRDSET
$$ CONNECTIVITY
CHEXA 100 I 1000 1100
+HXIOO 1111 1011
$
CELAS4 9999 1. 9999
SLOAD 100 9999 1.-5
$$ MATERIAL AND GEOMETRIC PROPERTIES
MAT1 10 30.+6 11.5+6 .3
MATS1 10 PLASTIC 3.+6
PSOLID I 10 0 .O
$$ BOUNDARY CONDITIONS
SPCI 100 123456 I000
SPCl 100 I 1010
SPCI 100 2 1001
SPCI 100 3 II00
$$ LOADING CONDITIONS
LOAD 1195 2.4 I. 123
LOAD 1196 2.4 I. 123 -1. 121
$ UNIAXIAL TENSION IN X
PLOAD4 121 100 36.+3 I100 llll
PLOAD4 121 100 36.+3 1000 1011
$ UNIAXIAL TENSION IN Z
PLOAD4 123 100 36.+3 I000 ill0
PLOAD4 123 100 36.+3 1001 llll
ENDDATA
 Table 7.4.2 Input Data Listing for Strain Path Dependent Case
ID NLPATH3, V66 $ QYC 8/24/90
SOL 66 $ NONLINEAR STATIC ANALYSIS
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
TIME 5 $ FOR VAX
CEND
TITLE = CODE TEST FOR NONLINEAR ELEMENTS WITH STRAIN PATH 3
ECHO = UNSORT
SEALL = ALL
DISP = ALL
OLOAD = ALL
SPCF = ALL
STRES = ALL
SUBCASE 10
LABEL = SUBCASE 10
SPC =I0
NLPARM = 10
SUBCASE 20
LABEL = SUBCASE 20
SPC = 20
NLPARM = 20
SUBCASE 30
LABEL = SUBCASE 30
SPC = 30
NLPARM = 20
SUBCASE 40
LABEL = SUBCASE 40
SPC = 40
NLPARM = 20
SUBCASE 50
LABEL = SUBCASE 50
SPC = 50
NLPARM = 20
SUBCASE 60
LABEL = SUBCASE 60
SPC = 60
NLPARM = 20
SUBCASE 70
LABEL = SUBCASE 70
SPC = 70
NLPARM = 20
SUBCASE 80
LABEL = SUBCASE 80
SPC = 80
NLPARM = 20
SUBCASE 90
LABEL = SUBCASE 90
SPC = 90
NLPARM = 20
SUBCASE 100
LABEL = SUBCASE 100
SPC = 100
NLPARM = 20
SUBCASE 110
LABEL = SUBCASE 110
SPC = 110
NLPARM = 20
SUBCASE 120
LABEL = SUBCASE 120
SPC = 120
NLPARM = 20
OUTPUT (PLOT)
CSCALE = 1.3
PLOTTER NAST
SET 1 = ALL
VIEW 30., 20., 0.
PTITLE = ISOMETRIC VIEW
FIND SCALE, ORIGIN 1, SET 1
PLOT SET 1, ORIGIN 1, LABEL BOTH
PLOT STATIC 0, SET 1, ORIGIN 1
VIEW O., O., 0.
PTITLE = FRONT VIEW
FIND SCALE, ORIGIN 1, SET 1
PLOT SET 1, ORIGIN 1
PLOT STATIC 0, SET 1, ORIGIN 1
AXES Z, Y, HX
PTITLE = TOP VIEW
FIND SCALE, ORIGIN 1, SET 1
PLOT SET 1,ORIGIN 1
PLOT STATIC 0, SET 1, ORIGIN 1
BEGIN BULK
$
NLPARM 10 1 SEMI UPW YES
NLPARM 20 1 SEMI 20 UPW YES
$$ GEOMETRY
GRID 1000 0. 0.
GRID 1100 1. 0.
GRID 1110 1. 1.
GRID 1010 0. 1.
GRID 1001 0. 0.
GRID 1101 1. 0.
GRID 1111 1. I.
GRID 1011 0. 1.
GRDSET
$$ CONNECTIVITY
CHEXA 100 1 1000 1100
+HXlOO 1111 1011
$
CELAS4 9999 1. 9999
SLOAD 100 9999 1.4
$$ MATERIAL AND GEOMETRIC PROPERTIES
MAT1 1 30.+6 11.5+6 .3 7.332-
MATS1 1 PLASTIC 3 .+6 1
PSOLID 1 1 0 0
$$ LOADING CONDITIONS
SPCI 10 I 1000 1010 1001
SPCl 10 2 1000 1100 1001
SPCl 10 3 1000 I100 1010
SPC 10 1100 1 2.-4 1110
SPC 10 I101 1 2.-4 1111
SPC 10 1001 3 -6.67-4 1101
SPC 10 1011 3 -6.67-4 1111
$
SPCl 20 1 1000 1010 1001
SPCl 20 2 1000 I100 1001
SPCl 20 3 1000 I100 1010
SPC 20 1100 1 1.07-3 1110
SPC 20 1101 1 1.07-3 llll
SPC 20 1001 3 -2.67-3 1101
SPC 20 1011 3 -2.67-3 1111
$
SPCl 30 1 1000 1010 1001
SPCl 30 2 1000 I100 1001
SPCl 30 3 1000 1100 1010
SPC 30 1100 1 3.12-3 1110
SPC 30 1101 1 3.12-3 1111
SPC 30 1001 3 -6.92-3 1101
SPC 30 1011 3 -6.92-3 1111
$
SPCl 40 1 1000 1010 1001
SPCl 40 2 1000 1100 1001
SPCl 40 3 1000 1100 1010
SPC 40 1100 1 5.17-3 1110
SPC 40 1101 1 5.17-3 1111
SPC 40 1001 3 -1.12-2 1101
SPC 40 1011 3 -1.12-2 1111
$
SPCl 50 1 1000 I010 1001
SPCl 50 2 1000 1100 1001
SPCl 60 3 1000 1100 1010
SPC 50 1100 1 7.21-3 1110
SPC 50 1101 1 7.21-3 1111
SPC 50 1001 3 -1.54-2 1101
SPC 50 1011 3 -1.54-2 1111
$
SPCl 60 1 1000 1010 1001
SPCl 60 2 1000 1100 1001
SPCl 60 1100 1010
SPC 60 9.26-3 1110
SPC 60 9.26-3 1111
SPC 60 -1.97-2 1101
SPC 60 -1.97-2 llll
$
SPCl 70
SPCl 70
SPCl 70
SPC 70
SPC 70
SPC 70
SPC 70
$
SPCl 80
SPCl 80
SPCl 80
SPC 80
SPC 80
SPC 80
SPC 80
$
SPCl 90 1010 1001
SPCl 90 1100 1001
SPCl 90 1100 1010
SPC 90 1.68-2 1110
SPC 90 1.68-2 llll
SPC 90 -2.92-2 1101
SPC 90 -2.92-2 1111
$
SPCl 100 1010 1001
SPCl 100 1100 1001
SPCl 100 1100 1010
SPC 100 2.021-2 1110
SPC 100 2.021-2 1111
SPC 100 -3.28-2 1101
SPC 100 -3.28-2 llll
$
SPCl 110
SPCl 110
SPCl 110
SPC 110
SPC 110
SPC 110
SPC 110
$
SPCl 120
SPCl 120
SPCl 120
SPC 120 1100 1 2.79-2 ill0 1 2.79-2
SPC 120 1101 1 2.79-2 ill1 1 2.79-2
SPC 120 1001 3 -4.03-2 1101 3 -4.03-2
SPC 120 1011 3 -4.03-2 ill1 3 -4.03-2
ENDDATA
 Table 7.4.3 Input Data Listing for Biaxial Loading Case
ID BIAXIAL, V66 $ FEB84
SOL 66 $ NONLINEAR STATIC ANALYSIS
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
TIME 5 $ CPU MIN FOR VAX
CEND
-
TITLE = QUAD4 PLASTIC ANALYSIS CHECK (M. WASYLCZAK, MCAUTO)
SUBTITLE = BIAXIAL LOADING CONDITION
LABEL = REF.: FOSTER WHEELER REPORT FWR-27, MARCH 1972, P. A-50
SEALL = ALL
STRESS = ALL
SPC = 10
MPC = 10
SUBCASE 1
LOAD = 1
NLPARM = 10
SUBCASE 2
LOAD = 2
NLPARM = 10
SUBCASE 3
LOAD = 3
NLPARM = 10
SUBCASE 4
LOAD = 4
NLPARM = 10
SUBCASE 5
LOAD = 5
NLPARM = 10
SUBCASE 6
LOAD = 6
NLPARM = 10
SUBCASE 7
LOAD = 7
NLPARM = 10
SUBCASE 8
LOAD = 8
NLPARM = 10
SUBCASE 9
LOAD = 9
NLPARM = 10
SUBCASE 10
LOAD = 10
NLPARM = 10
SUBCASE 11
LOAD = 11
NLPARM = 10
SUBCASE 12
LOAD = 12
NLPARM = 12
SUBCASE 13
LOAD = 13
NLPARM = 13
SUBCASE 14
LOAD = 14
NLPARM = 14
SUBCASE 15
LOAD = I 5
NLPARM = 10
SUBCASE 16
LOAD = 16
NLPARM = 10
SUBCASE 17
LOAD = 17
NLPARM = 10
SUBCASE 18
LOAD = 18
NLPARM = 10
SUBCASE 19
LOAD = 19
NLPARM = 10
SUBCASE 20
LOAD = 20
NLPARM = 10
SUBCASE 21
LOAD = 21
NLPARM = 10
SUBCASE 22
LOAD = 22
NLPARM = 10
SUBCASE 23
LOAD = 23
ILPARM = 10
SUBCASE 24
LOAD = 24
NLPARM = 10
SUBCASE 26
LOAD = 25
NLPARM = 10
SUBCASE 26
LOAD = 26
NLPARM = 10
SUBCASE 27
LOAD = 27
NLPARM = 10
SUBCASE 28
LOAD = 28
NLPARM = 10
SUBCASE 29
 LOAD = 29
NLPARM = 10
SUBCASE 30
LOAD =30
NLPARM = 10
SUBCASE 3 1
LOAD = 31
NLPARM = 10
SUBCASE 32
LOAD = 32
NLPARM = 10
SUBCASE 33
LOAD = 33
NLPARM = 10
SUBCASE 34
LOAD = 34
NLPARM = 10
SUBCASE 35
LOAD = 35
NLPARM = 10
SUBCASE 36
LOAD = 36
NLPARM = 10
B E G I N BULK
$$ PARAMETERS
PARAM NLAYERS I
$
NLPARM 10 I AUTO
NLPARM 12 1 AUTO
NLPARM 13 1 AUTO
+ILP13 3.53-3
NLPARM 14 I AUTO
$$ GEOMETRY
GRID I 0.0
GRID 2 0.5
GRID 3 1.0
GRID 4 0.0
GRID 5 0.5
GRID 6 1.0
GRID 7 0.0
GRID 8 0.5
GRID 9 1.0
$$ CONNECTIVITY
CQUAD4 I I I
CQUAD4 2 I 2
CQUAD4 3 1 4
CQUAD4 4 I 5
$$ P R O P E R T I E S
PSHELL I I .001
MAT1 I 30. +6 .3
MATS1 I PLASTIC 1.5+6
$$ CONSTRAINTS
MPC I 6 I I.
MPC I 9 I I.
MPC I 7 2 I.
MPC I 8 2 1.
SPCI 1 3456 I THRU
SPCl 1 2 I 2
SPC1 I I I 4
$$ LOADING
FORCE I 3 30.
FORCE 2 3 32.5
FORCE 3 3 35.
FORCE 4 3 37.5
FORCE 5 3 40.
FORCE 6 3 40.5
FORCE 7 3 40.
FORCE 8 3 40.5
FORCE 8 9 1.02
FORCE 9 3 41.5
FORCE 9 9 3.19
FORCE 10 3 43.
FORCE 10 9 6.89
FORCE 11 3 44.5
FORCE 11 9 11.53
FORCE 12 3 46.188
FORCE 12 9 23.05
FORCE 13 3 44.5
FORCE 13 9 32.966
FORCE 14 3 43.
FORCE 14 9 36.1
FORCE 15 3 41.5
FORCE 15 9 38.3
FORCE 16 3 40.5
FORCE 16 9 39.48
FORCE 17 3 40.
FORCE 17 9 40.
FORCE 18 3 37.
FORCE 18 9 42.44
FORCE 19 3 34.
FORCE 19 9 44.07
FORCE 20 3 30.
FORCE 20 9 45.41
FORCE 2 1 3 25.
FORCE 21 9 46.13
FORCE 22 3 20.
FORCE 22 9 46.05
FORCE 23 3 15.
FORCE 23
FORCE 24
FORCE 24
FORCE 25
FORCE 25
FORCE 26
FORCE 26
FORCE 27
FORCE 27
FORCE 28
FORCE 28
FORCE 29
FORCE 29
FORCE 30
FORCE 30
FORCE 31
FORCE 31
FORCE 32
FORCE 32
FORCE 33
FORCE 33
FORCE 34
FORCE 34
FORCE 36
FORCE 35
FORCE 36
FORCE 36
$
ENDDATA
7.5 CONSTRAINTS AND ENFORCED MOTION

In addition to the loading conditions, SOL 66 (or SOL 106) allows the user to change the
boundary conditions or enforce the displacements at the specified grid or scalar points.

7.5.1 Boundary Conditions

In the Case Control section, an SPC entry is used to select a single point constraint set (s-set)
which will be applied t o the structural model. The specified set identification must be identical
to the SID field of an SPC, SPCl or SPCADD Bulk Data entry. Notice that SPCADD entries
take precedence over SPC or SPCl entries. If both have the same SID, only the SPCADD
entry will be used.

A significant application of SPC is the imposition of boundary conditions. The PS field

in the GRID entry is also able to specify single-point constraints associated with a grid point.
But these constraints are so-called permanent constraints which can not be changed during the
analysis. An advantage of using SPC to specify boundary conditions is that these boundary
conditions can be changed from subcase to subcase by simply selecting a different SPC set
inside each subcase.

7.5.2 Multipoint Constraint

While SPC is t o constrain the motion by specifying a known value, MPC is to tie the motion of
a degree of freedom t o other degrees of freedom. An MPC entry in the Case Control section is
used t o select a multipoint constraint set t o be applied t o the structural model. The specified
MPC set identification must appear at least in one MPC or MPCADD Bulk Data entry.
Each MPC Bulk Data entry may be used to define a constraint equation involving a group of
degrees of freedom in which the first degree of freedom is assumed to be the dependent degree
of freedom and included in the m-set. All the degree of freedom in m-set will be condensed
out prior to the matrix operations. Their response will be directly recovered from those of
the independent degrees of freedom according t o the specified constraint equation. Similar to
SPC, MPC can also be changed from subcase t o subcase by specifying a different MPC set ID
inside each subcase in static analysis (SOL 66 or SOL 106).

7.5.3 Enforced Motion

The enforced displacements may be specified in the Bulk Data section using SPC or SPCD
entries. In the former case, the enforced displacement values may be input directly in the D
field of the SPC entry. Each SPC entry may define enforced displacements for up t o two grid
or scalar points. Several SPC entries which reference the same SID may be used if enforced
displacements for more than two grid or scalar points are desired. The only disadvantage of
this method is that the entire s-set must be redefined if the enforced displacement conditions
vary among subcases.

One way t o avoid this condition is t o define the enforced displacements using SPCD entries.
In this case, all the enforced degrees-of-freedom must be included in the s-set by SPC or S P C l
entries. The SPCD entry is requested as a LOAD, and therefore a LOAD set must also be
selected in the Case Control section. Since only the LOAD set needs to be changed instead
of the entire s-set, this method of applying the enforced displacements is more efficient than
the SPC entry method when more than one enforced displacement condition is applied. If
the enforced displacements are specified on both the SPCD and SPC entries, then the former
will override the latter. Other Bulk Data load entries such as FORCE, FORCE1, PLOAD,
PLOADi, etc. may also be combined with SPCD. However, it should be noted that the Bulk
Data LOAD combination entry will not combine an SPCD load.

The table below summarizes the data entries required for the enforced motions or the
boundary condition changes in MSC/NASTRAN.
Option Case Control Entries Bulk Data Entries
1 SUBCASE 1 SPC 100
SPC = 100 SPC 200
SUBCASE 2
SPC = 200
2 SUBCASE 1 SPC(or SPC1) 100
SPC = 100 SPCD 10
LOAD = 10 SPCD 20
SUBCASE 2
SPC = 100
LOAD = 20

7.5.4 Example

Fig. 7.5.1 shows a steel clip which will be pushed into a cavity whose height is 5.0 inches. It
is assumed that there is no friction force a t the contact surface. The clip is modeled with 24
elastic BEAM elements. Since a part of the clip bottom might lift off the ground during the
process, 7 GAP elements are provided at the clip-ground interface. The goal of the analysis is
t o predict the final configuration of the clip.

It is obvious from Fig. 7.5.1 that GRID 18 will be the first grid point to come in contact
with the ceiling of the cavity. Since the height of the clip at GRID 18 is 5.5 inches and that
of the cavity is 5.0 inches, an enforced displacement of -0.5 inches in the y direction is applied
at GRID 18 in the cold start run. The result shows that the height of the deformed clip a t
GRID 19 is 5.264 inches, which is still greater than the height of the cavity. Therefore, in
the first restart run the enforced displacement at GRID 18 is released and another enforced
displacement of -1.0 inches (i.e., 5.0 - 6.0 = -1.0) is applied in the y direction at GRID 19. Notice
that the specified enforced displacements should always refer to the undeformed configuration
and MSCJNASTRAN will automatically compute the increments. The analysis from this run
predicts that GRID 20 will attain a height of 5.156 inches, which is still greater than the cavity
height. A second restart is performed to release the constraint at GRID 19 and enforce the
displacement at GRID 20 by -1.5 inches in the y direction. At this time, the analysis results
indicate that no grid point in the model has a height over 5.0 inches, which means that the
entire clip is inside the cavity. The final deformed shape of the clip is shown in Fig. 7.5.1. The
predicted heights for some of the grid points are tabulated below at different stages with the
values greater than 5.0 in italics.

--
nr

--
-0-
22
'0-
/ .. ......

21
Undeformed

-e.. 20 [
'Y

I l l I

Figure 7.5.1. Elastic Clip Pushed into a Cavity

 Table 7.5.1. Input Data Listing for Elastic Clip
$RESTART $ added for the restart runs
ID CLIPB, V67 $ SSH 12/12/90
TIME 5 $
SOL 66 $ NONLINEAR STATIC ANALYSIS
DIAG 8,50 $
CEND
TITLE = INSERTING A CLIP INTO A CAVITY (BEAM MODEL)
SEALL = ALL
ECHO = BOTH
DISP = ALL
SPCF = ALL
$-------------------
$PARAM,SUBID,2 $ used in the 1st restart
$PARAM,LOOPID,2 $
$------------------
$PARAM,SUBID,3 $ used in the 2nd restart
$PARAM,LOOPID,7 $
$------------------
SUBCASE 10
SPC = 10
LOAD = 100
NLPARM = 1
$-------------------
$SUBCASE 20 $
$ SPC=20 $ added for the 1st restart
$ LOAD=200 $
$ NLPARM = 2 $
$-------------------
$SUBCASE 30 $
$ SPC=30 $ added for the 2nd restart
$ LOAD=300 $
$ NLPARM = 3 $
$-------------------
BEGIN BULK
PARAM LGDISP 1
$$ SOLUTION CONTROL
NLPARM I 2 0. SEMI 1
NLPARM 2 5 0. AUTO
NLPARM 3 4 0. AUTO
$
$$ GEOMETRY
$
GRID I
GRID 2
GRID 3
GRID 4
GRID 5
GRID 6
GRID 7
GRID 8
GRID 9
GRID 10
GRID 11
GRID 12
GRID 13
GRID 14
GRID 15
GRID 16
GRID 17
GRID 18
GRID 19
GRID 20
GRID 21
GRID 22
GRID 23
GRID 24
GRID 25
GRID 106
GRID 107
GRID 108
GRID 109
GRID 110
GRID Ill
GRID 112
$
$$ ELEMENT DEFINITION
$
CBEAM 101 100
CBEAM 102 100
CBEAM 103 100
CBEAM 104 100
CBEAM 105 100
CBEAM 106 100
CBEAM 107 100
CBEAM 108 100
CBEAM 109 100
CBEAM 110 100
CBEAM Ill 100
CBEAM 112 100
CBEAM 113 100
CBEAM 114 100
CBEAM 115
CBEAM 116
CBEAM 117
CBEAM 118
CBEAM 119
CBEAM 120
CBEAM 121
CBEAM 122
CBEAM 123
CBEAM 124
CORD2R I
+CORI 0.
CGAP 206
CGAP 207
CGAP 208
CGAP 209
CGAP 210
CGAP 211
CGAP 212
$
$$ PROPERTIES
$
PBEAM 100
MAT1 100
PGAP 200
PGAP 210
$
$$ LOADING
$
SPCI I0
SPCI 20
SPCI 30
SPCD 100
SPCD 200
SPCD 300
$
ENDDATA
7.6 RESTARTS
Restarts may be performed for various purposes. The most common reason is to extend a
normally or abnormally (due t o divergence or CPU time expiration, etc.) terminated analysis.
Restarts may also be used only for obtaining additional output. The present section will
concentrate on these two types of restarts. In addition, one may use SOL 66 t o preload a
nonlinear structure and then restart into either SOL 66 for a nonlinear buckling analysis, SOL
63 for a normal mode analysis, or SOL 99 for a nonlinear transient analysis.

7.6.1 Preliminaries

In Version 65 the data blocks corresponding to every converged load step in SOL 66 are saved
for restarts by default. If PARAM,SMALLDB is specified as 1, however, solution data blocks
corresponding to the intermediate load steps not requested for output (by INTOUT field in
NLPARM) are deleted from the database after a subcase is completed. In this case, the
restarts are restricted t o those starting from the solution with the corresponding data blocks
stored in the database, i.e., any converged load step in the current subcase or from the end
of any previous subcase. Note that the small database option (PARAM,SMALLDB,l) is only
applicable in Version 65.

In Version 66, the DMAP was modified to store only the data blocks for the output load
steps without requiring parameter SMALLDB. Therefore, the solution can only be restarted
from the end of a subcase if the default (INTOUT=NO) is used.

Further improvements were made in Version 67 so that the data blocks for the last converged
load step will be saved if the solution is terminated in the middle of a subcase due t o divergence
or consumption of the specified CPU time. Consequently, the solution can always be restarted
from the termination point regardless of the value of INTOUT in the NLPARM entry. In
addition, restarts from the unstable region for the arc-length methods were also made available
in Version 67.

7.6.2 Restart Procedure

Restarts in SOL 66 are controlled by three user parameters which may be specified either in
the Case Control section prior t o the subcases or in the Bulk Data section:

PARAM, LOOPID, t
PARAM, SUBID, m
PARAM, LOADINC, n (only applicable in V65 and V66)
The integer 1 identifies data blocks for a specific converged solution stored in the data base,
which may be used as initial condition for a subsequent restart. The value of LOOPID as well
as SUBID and LOADINC for each solution step is printed with the iteration information in
the output. PARAMeters SUBID and LOADINC refer t o the initial load step for the restart.
SUBID selects the subcase in which the restart analysis is initiated and LOADINC selects the
load step number in that subcase. The default values for LOADINC and SUBID are 1.

The simplest way t o perform a restart analysis is t o specify a new value for SUBID which is
greater than the SUBID value appearing in the last converged solution step of the previous run
(e.g., new SUBID = last SUBID + 1) and t o use the default for LOADINC (i.e., LOADINC
= 1). This means that the solution will restart into the very first load step of a new subcase.
If a subcase with the new SUBID value is added to the Case Control section of the restart
data deck, the solution will continue with the new LOAD and NLPARM data specified in that
subcase. If the new subcase is not added to the restart data deck, only the data recovery
(without any computation) will be processed in the rest art analysis.

For example, a load P is applied in one subcase with 5 increments, i.e., 0.2P for each incre-
ment. Assume that the solution diverges and the third load increment (assume its LOOPID =
3) is the last solution step saved for the restart. In the restart data deck, one may set the new
parameter values LOOPID=3, SUBID=2, LOADINC=l and add a second subcase in the Case
Control section. Assuming that the second subcase specifies the same LOAD and NLPARM
entries as the first subcase (i.e., load P with 5 increments), the solution will continue with a
load increment of ( P - 3 x 0.2P)/5 = 0.08P. If the second subcase is not added, only the data
recovery will be processed.

In Versions 65 and 66, a value greater than 1 may be specified for LOADINC if the new
SUBID value is the same as that which appeared in the previous run. Using the example
described in the preceding paragraph, one may also set LOOPID=3, SUBID=l, LOADINC=4
and leave out the second subcase. In this case, the solution will restart into the fourth load
increment to complete the remaining part of the first subcase. Since this feature has only
limited applications and is very difficult to use with the adaptive load/arc-length method, it
is removed from SOL 66 in Version 67.

7.6.3 Example

Fig. 7.6.1 shows a nearly-horizontal simple rod whose left end is hinged and right end is simply
supported. An elastic spring of 3 lbs/in. is connected in the vertical direction to the right
end of the rod, where a vertical force is applied. This is a snap-through problem with a single
degree-of-freedom. The force-displacement curve of the rod is shown in Fig. 7.6.2.

In the cold start run whose input data is given in Table 7.6.1, the loading is increased from
0 t o 2 lbs in two load increments and the analysis executed t o completion. A new subcase is
added in the restart run to increase the loading t o 4 lbs using two load increments. The input
data for this run is given in Table 7.6.2. This analysis diverges in the second load increment,
i.e.. LOOPID 3 (LOAD S T E P 1.5) is the last converged solution step which is saved for the
restart (except Version 66). T h e solution is restarted again using the arc-length method, whose
input d a t a is given in Table 7.6.3. In this run, Mi4SINC is intentionally set t o 2 in the SLPC'I
entry so that the analysis would terminate with a unstable solution. Finally, the solution is
restarted a third time t o reach the loading of 4 lbs. The input d a t a for the third restart is
given in Table 7.6.4.

Figure 7.6.1 Model of Rod with Spring

4oa- _ _ _ _ _ _ -- - - - -__-____
(- - --

3rd Restart

0.5 1 0 0 1 0 s 2.0

Figure 7.6.2 Force-Displacement Curve

 Table 7.6.1 Input Data Listing for Cold Start
ID SPROD,V67 $ SSH 6/17/91
TIME 5 $
SOL 66 $
DIAG 8,50 $
CEND
TITLE=SIMPLE ONE DOF GEOMETRIC NONLINEAR PROBLEM
SUBTITLE= COLD START RUN
LABEL=REF: STRICKLIN AND HAISLER; COMP. AND STRUCT.; 7:125-136 (1977)
ECHO=UNSORT
SEALL=ALL
OLOAD=ALL
DISP=ALL
SUBCASE 10 $ LOAD = 2 LB.
LOAD=I
NLPARM=I
BEGIN BULK
$ GEOMETRY
GRID I 0. 0. 0.
GRID 2 100. -1. 0.
$ CONNECTIVITY
CROD 10 10 I 2
CELASI 20 20 2 2 0 0
$ PROPERTIES
PROD 10 I .I
PELAS 20 3.
MAT1 1 10.E7
$ LOADING
FORCE I 2
$ PARAMETERS
PARAM LGDISP +I
NLPARM I 2
$
ENDDATA
 Table 7.6.2 Input Data Listing for First Restart
RESTART
ID SPRODRI,V67 $ SSH 6/17/91
TIME 5 $
SOL 66 $
DIAG 8,50 $
CEND
TITLE=SIMPLE ONE DOF GEOMETRIC NONLINEAR PROBLEM
SUBTITLE= FIRST RESTART (INCREASES THE LOAD UP TO 4 LBS)
LABEL=REF: STRICKLIN AND HAISLER; COMP. AND STRUCT.; 7:125-136 (1977)
$
ECHO=UNSORT
SEALL=ALL
OLOAD = ALL
DISP = ALL
PARAM, SUBID, 2
PARAM, LOOPID, 2
SUBCASE 10 $ LOAD = 2 LB.
LOAD=1
NLPARM=I
SUBCASE 20 $ LOAD = 4 LB.
LOAD=:!
NLPARM=2
BEGIN BULK
$ GEOMETRY
GRID I 0. 0. 0.
GRID 2 100, -1. 0.
$ CONNECTIVITY
CROD 10 10 I 2
CELASI 20 20 2 2 0
$ PROPERTIES
PROD 10 I .I
PELAS 20 3.
MAT1 I 10.E7
$ LOADING
FORCE I 2
FORCE 2 2
$ PARAMETERS
PARAM LGDISP +I
NLPARM I 2
NLPARM 2 2
+NLP21
+NLP22 0
$
ENDDATA
 Table 7.6.3 Input Data Listing for Second Restart
RESTART
ID SPRODR2,V67 $ SSH 6/17/91
TIME 5 $
SOL 66 $
DIAG 8,50 $
CEND
TITLE=SIMPLE ONE DOF GEOMETRIC NONLINEAR PROBLEM
SUBTITLE= 2ND RESTART (SWITCHES TO THE ARC-LENGTH METHOD)
LABEL=REF: STRICKLIN AND HAISLER; COMP. AND STRUCT.; 7:125-136 (1977)
ECHO=UNSORT
SEALL=ALL
OLOAD = ALL
DISP = ALL
PARAM, SUBID, 3
PARAM, LOOPID, 3
SUBCASE 10 $ LOAD = 2 LB.
LOAD=I
NLPARM=I
SUBCASE 20 $ LOAD = 4 LB.
LOAD=:!
NLPARM=2
SUBCASE 30 $ LOAD = 4 LB., ARC-LENGTH METHOD
LOAD=2
NLPARM=3
BEGIN BULK
$ GEOMETRY
GRID I 0. 0. 0.
GRID 2 100. -1. 0.
$ CONNECTIVITY
CROD 10 10 I 2
CELASI 20 20 2 2 0
$ PROPERTIES
PROD 10 I .I
PELAS 20 3.
MAT1 I 10.E7
$ LOADING
FORCE I 2
FORCE 2 2
$ PARAMETERS
PARAM LGDISP ti
NLPARM 1 2
NLPARM 2 2
+NLP2I
tNLP22 0
NLPARM 3 I
NLPCI 3
$
ENDDATA
 Table 7.6.4 Input Data Listing for Third Restart
RESTART
ID SPRODR3,V67 $ SSH 6/17/91
TIME 5 $
SOL 66 $
DIAG 8,50 $
CEND
TITLE=SIMPLE ONE DOF GEOMETRIC NONLINEAR PROBLEM
SUBTITLE= 3RD RESTART (RESTARTS IN THE UNSTABLE REGION TO COMPLETION)
LABEL=REF: STRICKLIN AND HAISLER; COMP. AND STRUCT.; 7:125-136 (1977)
$
ECHO=UNSORT
SEALL=ALL
OLOAD = ALL
DISP = ALL
PARAM, SUBID, 4
PARAM, LOOPID, 5
SUBCASE 10 $ LOAD = 2 LB. COLD START
LOAD=I
NLPARM=I
SUBCASE 20 $ LOAD = 4 LB. NORMAL RESTART
LOAD=2
NLPARM=2
SUBCASE 30 $ LOAD = 4 LB., RESTART WITH THE ARC-LENGTH METHOD
LOAD=:!
NLPARM=3
SUBCASE 40 $ LOAD = 4 LB., RESTART IN UNSTABLE REGION
LOAD=2
NLPARM=4
BEGIN BULK
$ GEOMETRY
GRID I 0. 0. 0.
GRID 2 100. -1. 0.
$ CONNECTIVITY
CROD 10 10 I 2
CELASI 20 20 2 2 0
$ PROPERTIES
PROD 10 I .l
PELAS 20 3.
MAT1 I 10.E7
$ LOADING
FORCE I 2
FORCE 2 2
$ PARAMETERS
PARAM LGDISP t1
NLPARM 1 2
NLPARM 2 2
+NLP2I
tNLP22 0
NLPARM 3 1
NLPCI 3
NLPARM 4 1
NLPCI 4
$
ENDDATA
7.7 VERIFICATION PROBLEM:
ANALYSIS of a PRESSURE VESSEL

This problem is used t o illustrate the nonlinear static analysis of pressure vessel using QUAD4
elements. T h e pressure vessel is illustrated in Fig. 7.7.1. T h e pressure vessel consists of a
spherical shell joined by a thin cylindrical shell a t the top. A 10' sector of the steel pressure
vessel is modeled using QUAD4 and PETU'TA elements as shown in Fig. 7.7.2. T h e intersection
of the spherical and cylindrical shells of the vessel is difficult t o model using plate elements.
This difficulty is emphasized here because changes in the plate thicknesses a t the junction
result in large differences in the final answers. T h e MSC/NASTRAN results agree reasonably
well with the experimental results of Dinno and Gill [7.3] as shown in Fig. 7.7.3. T h e input
d a t a for the model is listed in Table 7.7.1.

M ,
No s t r a i n hardening

Zierrkiewicz, The F i n i t e Element Method

McGraw-Hill, 1971, F i g u r e 1 8 . 1 0 , P. 4 7 4

Figure 7.7.1. Steel Pressure Vessel.

Figure 7.7.2. MSCINASTRAN Model (10% Sector of the Steel
Pressure Vessel)
 Vertical deflectlon of point 'A' ( x 10'~ In.)

Figure 7.7.3. Experimental Results vs. MSCINASTRAN

7.7 - 3
 Table 7.7.1 Input Data Listing for Pressure Vessel Model
ID VESSEL,V65 $ CJS 27-SEP-1985, 15-MAY-1984
SOL 66
DIAG 8,50
TIME 30
CEND
TITLE = PRESSURE VESSEL
SUBTITLE = MATERIAL NONLINEAR ANALYSIS
ECHO = UNSORT
SEALL = ALL
DISP = ALL
ELST = ALL
SUBCASE I
LABEL = PRESSURE LOAD OF 749.913 PSI
LOAD = 10
NLPARM = 10
SUBCASE 2
LABEL = PRESSURE LOAD OF 1080. PSI
LOAD = 20
NLPARM = 20
SUBCASE 3
LABEL = PRESSURE LOAD OF 1200. PSI
LOAD = 30
NLPARM = 30
SUBCASE 4
LABEL = PRESSURE LOAD OF 1400. PSI
LOAD = 40
NLPARM = 30
OUTPUT(PLOT)
CSCALE 1.3
SET I ALL
MAXI DEFO 1.0
VIEW 34.27, 23.17, 0.00
FIND SCALE ORIGIN 1 SET I
PLOT STATIC 0 SET I
AXES MY, MZ, X
VIEW O., S., 0.
FIND SCALE ORIGIN I SET I
PLOT STATIC 0 SET I
BEGIN BULK
$ PARAMETERS
PARAM AUTOSPC YES
$ PARAMETERS FOR NONLINEAR ITERATION
NLPARM 10 I AUTO W YES
NLPARM 20 3 AUTO W YES
NLPARM 30 5 AUTO W YES
$ DEFINE GRID LOCATIONS
CORD2C 10 0. 0. 0.
+C2C1 I. 0. 0.
CORD2S 20
+C2S2 I. 0.
GRID I 20
GRID 2 20
GRID 3 20
GRID 4 20
GRID 6 20
GRID 5 20
GRID 7 20
GRID 8 20
GRID 9 20
GRID 10 20
GRID 11 10
GRID 12 10
GRID 13 10
GRID 14 10
GRID 15 10
GRID 16 10
GRID 17 I0
GRID 18 10
GRID 19 10
GRID 20 10
GRID I01 20
GRID 102 20
GRID 103 20
GRID 104 20
GRID 105 20
GRID 106 20
GRID 107 20
GRID 108 20
GRID 109 20
GRID 110 20
GRID Ill 10
GRID 112 10
GRID 113 10
GRID 114 10
GRID 115 10
GRID 116 10
GRID 117 10
GRID 118 10
GRID 119 10
GRID 120 10
GRID 201
GRID 202
$ qUAD4 ELEMENTS
CqUAD4 I I
CqUAD4 2 I
CqUAD4 3 1
CqUAD4 4 I
CQUAD4 5 I 5 105 106 6
CQUAD4 6 I 6 106 107 7
CQUAD4 7 1 7 107 108 8
CQUAD4 8 1 8 108 109 9
CQUAD4 9 4 9 109 110 10
+CQO 0.25 0.25 0.28125 0.28125
CQUAD4 10 3 10 110 I11 I1
+CQl 0.28125 0.28125 0.3125 0.3125
CQUAD4 11 3 I1 ill 112 12
tCQ2 0.3125 0.3125 0.15625 0.15625
CQUAD4 12 5 12 112 113 13
tCQ3 0.15625 0.15625 0.125 0.125
CQUAD4 13 2 13 113 114 14
CQUAD4 I4 2 14 114 115 15
CQUAD4 15 2 15 115 116 16
CQUAD4 16 2 16 116 117 17
CQUAD4 17 2 17 117 118 18
CQUAD4 I8 2 18 118 119 19
$ QUAD4 CHARACTERISTICS
PSHELL I 1 0.25 I
PSHELL 2 I 0.125 I
PSHELL 3 I 0.3125 I
PSHELL 4 I 0.28125 I
PSHELL 5 I 0.15625 I
MAT1 1 2.912t7 0.3
MATS1 I PLASTIC I I 4.054t4
$ PENTA ELEMENT
CPENTA 101 4 20 1 19 119 202 20 120
$ PENTA CHARACTERISTICS
PSOLID 4 2 0
MAT1 2 2.912+7 0.3
$ DEFINE LOADING CONDITIONS
PLOAD2 10 749.913 I THRU 18
PLOAD2 20 1080. I THRU 18
PLOAD2 30 1200. I THRU 18
PLOAD2 40 1400. I THRU 18
PLOAD4 10 101 749.913
PLOAD4 20 101 1080.
PLOAD4 30 I01 1200.
PLOAD4 40 101 1400.
ENDDATA
MSCINASTRAN Nonlinear Analysis

HANDBOOK Volume I1

Version 67

Sang H. Lee
 Corporate Headquarters
The MacNeal-Schwendler Corporation
815 Colorado Boulevard
Los Angeles, CA 90041-1777
Tel: (213) 258-91 11 or (800) 336-4858
FAX: (213) 259-3838
Headquarters, European Operations
MacNeal-Schwendler GmbH
lnnsbrucker Ring 15
Postfach 801240
D - 81 612 Miinchen, GERMANY
Tel: (89) 431 9870
Telex: 523 784 MSG D
FAX: (89) 436 1716
Headquarters, Far East Operations
MSC Japan Ltd.
Entsuji-Gadelius Building
2-39, Akasaka 5-chome
Minato-ku, Tokyo 107, JAPAN
Tel: (03) 3505-0266
Telex: J23363 MSCWATA
FAX: (03) 3505-091 4

DISCLAIMER

The concepts, methods, and examples presented in this text are for illustrative and educational purposes
only and are not intended to be exhaustive or to apply to any particular engineering problem or design.
The MacNeal-Schwendler Corporation assumes no liability or responsibility to any person or company
I for direct or indirect damages resulting from the use of any information contained herein.
I
a1992 by The MacNeal-SchwendierCorporation
Printed in U.S.A.
All rights reserved.

MSC and MSCI are registered trademarks and service marks of The MacNeal-Schwendler Corporation. NASTRAN is a
registered trademark of the National Aeronautics and Space Administration. MSCINASTRAN is an enhanced, proprietary
version developed and maintained by The MacNeal-SchwendlerCorporation.

NA *V67 * Z * Z * Z * HB-NLN
Chapter

DIRECT TIME INTEGRATION

8.1 OVERVIEW

If the effects of inertia and damping forces are to be included in the nonlinear analysis, the
nonlinear transient response is analyzed by a step-by-step integration of an equation of motion,

By direct numerical integration, the equilibrium of Eq. (8.1.1) is satisfied at discrete time steps
with an interval of At. The equilibrium is ensured by iterations until the solution converges
to the preset error tolerance.

Since Version 62, several methods of improving the convergence of nonlinear problems
have been adopted in MSCINASTRAN [8.1]. These procedures, such as the quasi-Newton
update and line search process, provide users with additional options to adjust incremental
and iterative processes in search of more efficient and effective strategy for obtaining solutions.
The goal of the general-purpose program is to provide users with tools that are easy t o use.
As a step toward this goal, an adaptive time stepping algorithm has been implemented for the
nonlinear transient analysis.

A self-adaptive time stepping method has been implemented in MSCINASTRAN Version

66 for the transient analysis of nonlinear structures. The goal is to provide an automatic
method of implicit time integration, which renders accurate solutions efficiently t o any nonlin-
ear transient response problem. This goal was achieved by combining the automatic time step
adjustment with the bisection process in a coordinated manner.

Newmark's direct time integration method is implemented using the two-point recurrence
(or one-step) formula as a foundation of this self-adaptive time stepping algorithm [8.2]. The
optimal time step size, which is required for accuracy and efficiency, changes continuously in
the transient dynamic environment. The primary concept of automatic time step adjustment
is that the proper size of the time step can be predicted based on the dominant frequency
in the incremental deformation pattern at the previous time step. This concept presents a
deficiency of time lag involved in the prediction process. Furthermore, changes in nonlinearity
cannot be predicted from the deformation pattern at the previous time step. This deficiency
is overcome by the bisection process, which is activated when any difficulties arise in terms of
convergence during the iteration.

Frequent decomposition of the dynamic stiffness matrix is inevitable in adaptive time step-
ping using the implicit integration method. Efforts have been made to minimize the thrashing
in the algorithm (e.g., too frequent or repetitive adjustment of the time step size) and yet
to provide a solution reliably t o any type of problem. The iteration process for the equi-
librium employs expeditious methods such as BFGS updates and line searches as well as an
adaptive stiffness matrix update strategy. The present self-adaptive time stepping algorithm
for the nonlinear transient analysis is proving to be a robust and practical method, aimed at
efficiency, effectiveness, and user-friendliness.

The transient response analysis capability is provided in the self-contained solution se-
quences, SOLS 99 and 129. SOL 99, which is depicted by flow diagrams in Figs. 8.1.1 and
8.1.2, has evolved since Version 62. SOL 129 is a structured DMAP with the same function-
ality as SOL 99, which was implemented in Version 67 t o utilize the new executive system
constructs. Although these solution sequences are developed for nonlinear transient response
analysis, linear transient response analysis can be performed in these solution sequences in
order t o take advantage of the automatic time stepping and restart capabilities. The core part
of SOL 99 are the nonlinear modules (NLTRD and NLTRD2) which perform the incremental
and iterative processes for implicit direct time integration.
 . Save and Append F~les
ln~t~al~ze
parameters Generate FDV. UHVNL, ESTNL
Store and pnnt database

-
Fetch MESTNL
SOLCUR SOLPRE + 1

CASE - CASESXP. IMETHOD

Generate PDT, TEL

Process initial conditions 1
and store UHVNL

Reduce KJJNL -
Update stiffness KANL

KRDD = KDD + KDDNL

KDDNL * . . Reduce KDJJ 4 KDDD
KBDD KRDD + KDDD
=

. Create FDV tor IC

I

Create KSGG for SPCF

1 1
AM3=2MDD+ZiiiBDD .
1
AM2 = -zBDD
AM1 = PKBDD + AM3
Decompose AM1

Partition FSGV -
Run NLCOMB and NLTRD

if end of rubcar.
FSV
NLCOMB
NLTRDP
Print OLDDT

I I
Print STlME
Smra FSV, CASESX2. MPT
T E L UHVNL. FDV,
.
OESNLI PNL
Print LOOPID

SOLPRE = SOLPRE + 1

Figure 8.1.1 Flow Diagram for Nonlinear Process in SOL 99

8.1 - 3
 9 SDATREC

P OLTOP

Fetch UHVNL, FSV, TEL, GMD, GOD,

CASESX2, OESNLI , PNL,SMPT
TRLG ----t PPNL, PST
PSV = PST- FSV
Find the number of output time steps
TOLAPP TEL ---t OLB
Append
Append
PPNL --t PP

Append UHVNL
PSV

Append OESNL1
Append PNL
-
-
PSSD

- UHVF
OESNLX
PNU
MLOOP = MLOOP + 1
MLAST = LOOPID - MLOOP

Output Solution Set, OESNL2, NOLlN Force

XYplot Solution Set, OESNL2, NOLlN Force
I

Figure 8.1.2 Flow Diagram for Appending Process in SOL 99

8.1 - 4
8.2 THREE-POINT METHOD (NLTRD MODULE)

The three-point method is chosen to be compatible with MSCJNASTRAN linear transient in-
tegration method. The Newmark Beta method based on the two-step integration is combined
with the modifications t o Newton's method for nonlinear solutions [8.1]. The additional itera-
tion steps provide equilibrium solutions at each time step, thereby guaranteeing stability and
accuracy for arbitrary time step size. This method was implemented in the NLTRD module
in Version 62. This method can be used by selecting the AUTO or TSTEP method in the
TSTEPNL Bulk Data entry.

8.2.1 Basic Equations

Adding dynamic loads t o the basic equation given previously, we obtain a load equilibrium
error vector, {R,) at time step n by the equation:

where {P,) = Average load over the time period (tn-l < t , < tn+l)
{ U{ u = Corresponding acceleration and velocity vectors.
{Fn = Average elasto-plastic element total force vector.
( F = K'lL for linear problems.)
The above equation is solved at the reduced ( u d ) displacement vector size. The approxi-
mation errors due t o dynamic reduction methods are not included in the error vector {R,}.
Applying Newmark's averaging method over a finite time period, tn-l < t < tn+1, the static
forces are
+
{ F ) = { P F ( u n + l ) (1 - 2 P ) F ( u n ) f P F ( u n - I ) ) (8.2.2)
where ,O = Newmark Beta operator and F ( u n ) is the nonlinear force due t o a generalized
displacement vector {u,). An identical definition occurs for { P , ) from the applied loads a t
each time step.

From central finite differences, the acceleration and velocity vectors are:

) { ~ n +-
where { A ~ n + l = ) small angles and At = tn+1 - t , is the time step size.
l ~ n for
For large angle changes we require that {ii}, {u), and { A u ) are vectors in the glpbal
coordinate direction. The vector { u ) contains gimbal angles requiring a transformation [R(u,)]
such that
{ ~ n + l= +
) {un) [Rn]{Aun+l) (8.2.5)
Substituting Eqs. (8.2.2), (8.2.3) and (8.2.4) into Eqs. (8.2.1) defines the load error in terms
of the current vectors estimates:

and
{Aun+l} = {nun + dn)
At any time step the vectors {P,), {nu,), {F,), and {Fn-l) are known. In general the vectors
{d,}, {u,+~), and {Fn+l} = {F(u,+l)) must be found either by approximating {F,+l) or by
an iterative search.

Note that for a linear solution {F,) equals [l<]{u,}, where [ K ]is the reduced stiffness
matrix. In this case Eqs. (8.2.6) and (8.2.7) may be used t o solve for {u,+~} directly by
setting {R,) = (0). With ,D = 113 this will produce the linear SOL 27 MSCINASTRAN
results.

8.2.2 Nonlinear Iterations

For a nonlinaer solution, Eqs. (8.2.6) and (8.2.7) may be solved with Newton's method (or a
modified version). Using {d,} as the primary solution variable provides the following iteration
algorithm.

First assume a linear approximation:

{RL+'} - I,[
{R,} + dRn {dk+' - dk} = 0
dn
Let

Then, the estimated displacement change is

The algorithm is identical t o the static case except that instead of a tangent stiffness matrix,
the left hand side matrix, obtained from Eqs. (8.2.6) and (8.2.8) is

where [I;']= is the current tangent stiffness matrix.

At a new time step the load iterations may be started by assuming that {Au;+,} = {Au,}
and therefore:
{do,> = (0) (8.2.12)
However, t o be consistent with displacements, we must extrapolate the nonlinear forces:

Substituting Eqs. (8.2.12) and (8.2.13) into Eq. (8.2.6) for i = 0 results in the first estimate:

where 1

Here [A2] is a precalculated matrix. {Au,) and {F,) are previous results.

For subsequent iterations, ( i > 0) Eq. (8.2.6) becomes:

where
1 1
[A31 = I S M +z B 1 (8.2.17)

At each time step the code will iterate on Eq. (8.2.10) and Eq. (8.2.16) until {R;} passes
the convergence tests or the number of passes reaches an iteration limit. With a single step,
i = i, calculating only Ro, the results will be identical t o the existing NOLIN; results in
MS C/NASTRAN. For faster convergence the iterations may continue, the matrices may be
updated, and/or the time step size may be reduced.

8.2.3 Stability Analysis

In transient analysis, two types of instability could occur. The first is the familiar nonlinear
load iteration divergence which also occurs in static analysis. The second is the divergence
which grows with time in the transient integration. Both instabilities are caused by uncorrected
nonlinear equilibrium errors.

A convenient method, from von Neumann, for analyzing the stability limits is to assume
that the nonlinear forces are nearly linear and the error vector has a constant convergence. It
is assumed that the nonlinearity has a first order approximation:

IF(.)} - +A I ~ ~ ~ ] { u )
[IiT (8.2.18)
where A I c ~ difference
is the ~ ~ between the tangent stiffness matrix and its approximation 1CT.
The error vectors, {R), are assumed t o grow at the rate A, defined as:

Note that if IAI > 1 the system will be defined as unstable.

After lengthy calculations we may summarize the various criteria, assuming that the ma-
trices are reduced t o equivalent scalar modal quantities, the criterian for stable solutions are:
 For time steps with converged static iterations:

This is the same criteria as linear analysis.

For static load iterations:

This states that the mass and damping add to the effective linear stiffness [ K T ]and
improve the stability.

For time step integration with no intermediate static iterations, as in the standard New-
mark Beta method:
4
+
n K N L 5 (4p - 1)I<' -M
At2
(8.2.22)

This restriction is more severe than criteria (8.2.21) above, proving that the internal
iterations are more stable than the Newmark integration.
In summary, the method will have fewer divergence problems than either the static nonlin-
ear solution or the single step transient nonlinear methods. The better stability of the method
and the capability t o use larger time steps outweights the cost of a few internal iterations on
the static element forces.

8.2.4 DMAP Interface of NLTRD Module (Version 66)

NLTRD CASESX2,MESTNL,PDT,YS,KRDD,ELDATA,
KELMNL,AMlLL,GM,MPTS,DIT,AMlUU,DLTl,CSTMS,
BGPDTS,SILS,USETD,AM2,AMS,NLF'T,KSGG/
UHVNL,FSGV,ESTNL,FDV,OESNLl,PNL,TEL/
V,Y,BETA/S,N,CONV/S,N,RSTIME/S,N,NEWP/S,N,NEWK//
S,N,NSTEP/V,Y,LGDISP/S,N,MPTSIZE/S,N,ITERIDS//
S,N,KTIME/S,N,LASTUPD/S,N,NOGONL//V,Y,MAXLP
$ ttV66
Input Data Blocks:
CASESX2 Case control data.
MESTNL Nonlinear element summary table
PDT Applied loads, one column vector for each time step.
YS Enforced displacement.
KRDD Reduced stiffness in d-set
ELDATA Element data used to generate OESNL file.
KELMNL Nonlinear element stiffness matrix required for nonlinear force ( F ) computation.
AMlLL Lower triangular factor for nonlinear elements
GM Transformation matrix for multipoint constraint reduction.
MPTS Material properties table containing TSTEPNL data.
DIT Direct input tables.
A M l U U Upper triangular matrix for nonlinear elements.
DLTl Dynamic loads table for follower forces.
CSTMS Coordinate system transformation table.
BGPDTS Basic GRID point definition table.
SILS Scalar index list.
USETD Displacement set definition table.
AM2 Damping matrix in d-set divided by -At.
AM3 Matrix consisting of mass and damping terms: &Ad + B.
NLFT Nonlinear transient loads for NOLIN.
KSGG Partition of stiffness matrix for forces of constraints.
Output Data Blocks:
UHVNL InputJoutput, displacement and velocity vectors at converged time steps during the
direct integration, UHVNL consists of output displacement, velocity, and acceleration
vectors a t output time steps upon completion of the subcase.

FSGV Nonlinear element forces at constrained points, equivalent t o [K,,]{Y,) at output time
steps.

ESTNL Input/output, nonlinear Element Summary Table at the last converged step.
FDV Input/output, Two column vectors [F,, F ~at] the end of subcases. Element forces and
the current and the preceding steps (F,-l, F,) for matrix update.

OESNLl Output stress data for nonlinear elements.

PNL Nonlinear load vector appended at output time steps.
TEL Appended output time step list. Purged at the beginning of subcases.
Parameters:
BETA Input-real, integration parameter, default = 113.
CONV Input/output-integer, value from the last iteration for an input (0 for the initial
iteration), set t o -1 if not converged, f 1 if converged.

STIME Input/output-real, starting time or accumulated time used for restarts by user input.
NEWP Input/output-integer, set t o $1 for new subcase, -1 for no subcase switch.
NEWK Input/output-integer, set t o f l or +3 for stiffness update, -1 otherwise.
NSTEP InputJoutput-integer, current time step position for subcase, set t o 0 at the beginning
of the subcase.

LGDISP Input-integer, set t o +1 for geometric nonlinearity, -1 otherwise.

MPTSIZE Input/output-integer, the amount of open-core words required to hold the mate-
rial data, initialized t o 0 in MPL and retained during the problem solution.

ITERIDS Input /output-integer, iteration count within a time step.

KTIME Input/output-integer, time-to-go value at the beginning of K update in the DMAP
loop, on output from the module, a negative value is stored, denoting the increment of
time between the K computation and the start of the module.

LASTUPD Input/output-integer, the time step number of the last stiffness update. The
value 0 is used t o indicate if the stiffness update is performed due to the GAP element
during the iteration.

NOGONL Output-integer, set t o +1 to continue, -1 t o terminate.

MAXLP Input-integer, maximum limit allowed for element relaxation iteration and the ma-
terial subincrement processes.
8.3 TWO-POINT METHOD (NLTRD2 MODULE)

The multistep implicit integration method [8.1] is not suitable for the adaptive method. In
addition t o the extra memory requirement, the multistep integration scheme also introduces
difficulties in the starting algorithm, ending time step, and restarting. These difficulties become
more complicated and cumbersome when the adaptive time stepping is employed due t o the
unequal time intervals involved. Therefore, the two-point integration method [8.2], or one-step
integration, is introduced in Version 66 with module NLTRD2. This method can be selected
by specifying the ADAPT method in the TSTEPNL Bulk Data entry.

8.3.1 Newmark Integration

For the adaptive scheme, Newmark's method [8.3] is employed with the two-point recurrence
formula for one-step integration, i.e.,

and
= {un) + At { ~ n )+ 7 At {&+I - un> (8.3.2)
where {U), {u}, { u ) and At denote displacement, velocity, acceleration and the time step
increment, respectively. The subscript n designates the time step and the parameters (P and y )
are to be determined. An equilibrium equation t o be satisfied a t time step (ni-1) is:

where M and C denote mass and damping matrices, and {F) and {Pn+l)denote internal and
external forces, respectively. An incrementalliterative form of Eq. (8.3.3) can be recast for the
i-th iteration (denoted by superscripts) as:

satisfying
;+I - ui }
' + I i- + n + n+l
where

By solving Eqs. (8.3.1) and (8.3.2) for the velocity and the acceleration vectors, these
variables a t the i-th iteration of the time step ( n + l ) can be expressed in terms of displacements
as shown below:
where
{ u }= {u:+,} + {aui+l}
The governing equation for the Newton-Raphson iteration is obtained by substituting Eqs. (8.3.4)
and (8.3.5) into Eq. (8.3.3a) to obtain:
1
+ -C
M
Y
I
[pat. b' At + 1i-icl { h u i + l } = {R:+~}

where the residual load vector { R ; + ~is) expressed as

{ R : + ~ } = {Pn+l - ~:+1} -
M
pat. {u:+~ - Un - &) + ($ - 1) M {un}

- - Y C{
P At
+- } + ( - 1) c { u } - (1 - A { n } (8.3.7)

Notice that the tangential stiffness matrix, in Eq. (8.3.6) may be replaced by Kn
(modified Newton's iteration) or 2 which is a stiffness matrix evaluated at any preceding time
step.

8.3.2 Adaptation of Newmark's Method

An alternative expression for the load vector can be derived for y=0.5 by introducing

+
~ { u n } c { u ~ )= {Pn - Fn) (8.3.8)

By virtue of Eq. (8.3.8) the residual load error at each time step is effectively carried over
t o the next step and the error propagation is reduced, while the computation is significantly
simplified. Notice that the acceleration term in Eq. (8.3.7a) conveniently disappears when the
parameter L,3 has the recommended value of 0.25, i.e.,
If follower forces (i.e., applied forces change as a function of displacements) are involved,
however, the external load {Pn+l)should be updated at every iteration by

{ p ; + l ( ~ ) I = IPn+l(O)J + {fk(u7 tn+l)} + { ~ ' ( u tn+l)}

, (8.3.9)
where N' represents NOLINi forces at the i-th iteration, and f k ( ~ , t , + ~is) a correction for
follower forces as expressed below:
f k ( u , t n + l ) = g(tn+l>{pa(u)- P(O)J
where g(t) is a time dependent forcing function specified in the TLOADi entries and pi(^)
denotes static follower forces. It is noted that the residual vector in Eq. (8.3.7b) can be
computed without computing velocities and accelerations at every iteration.

8.3.3 Start-up Algorithm

The first iteration is performed with initial conditions

-
{u:+i} = {un} and {~:+1) = {K) (8.3.10)
which satisfy
{~:+l) {Fn) + 1cn{u,O+l - un)
It is noted that the initial equilibrium is satisfied by virtue of Eq. (8.3.8) without the need for
computing initial acceleration Uo, i.e.,

In this method of direct integration (based on the average acceleration scheme), the Eqs. (8.3.4))
(8.3.5) and (8.3.6) are reduced to:

4 4
{ u n + ~ )= {Un+l- un} -
At {On} - {Un}

Since the acceleration vector does not appear in Eq. (8.3.7b), {u) needs t o be computed
only for output purposes. However, Eq. (8.3.5a) requires an initial value, {uo). After some
numerical experiments, it has been determined that accelerations are most effectively evaluated
by the central difference method using velocities at three consecutive time steps at the end
of the integration process. For irregular time step intervals, the modified expression for the
acceleration is

{On+l} + (-At,+,
Atn - -)
Qtn
At,+, {On) - *
At n {&-I)] (8.3.5b)

for 0 < n < N , with N being the last time step. The initial and the final accelerations, {UO)
and {uN), are evaluated by an extrapolation using Eq. (8.3.5a).
8.3.4 DMAP Interface of NLTRD2 Module (Version 67)

NLTRD2 CASESX2, P D T , YS, ELDATA,

KELMNL, KDD, GM, MPT, DIT, KBDD, DLT1, CSTMS,
BGPDTS, SILS, USETD, BDD, MDD, NLFT/ UHVNL,
FSV, ESTNL, FDV, OESNL1, PNL, TEL, MUHVNL, MESTNL/
V,N,KRATIO=(l.,O.)/ S,N,CONV/ S,Y,STIME/
S,N,NEWP/ S,N,NEWDT=O/ S,N,OLDDT/ S,N,NSTEP/
V,Y,LGDISP/ S,N,MPTSIZE/ S,N,ITERIDS/ V,N,ITIME/
S,N,KTIME/ S,N,LASTUPD/ S,N,NOGONL/ S,N,NBIS/
V,Y,MAXLP=5/ V,Y,TSTATIC=-1/ V,N,LANGLES/ V,Y,NDAMP $
Input Data Blocks:
CASESX2 Case control data.
PDT Applied loads, one column vector for each time step.
YS Enforced displacement.
ELDATA Element data used to generate OSENL file.
KELMNL Nonlinear element stiffness matrix required for nonlinear force (F) computation.
KDD Stiffness matrix for linear portion in d-set.
GM Transformation matrix for multipoint constraint reduction.
MPT Material properties table containing TSTEPNL data.
DIT Direct input tables.
KBDD Tangential stiffness of residual structure in d-set.
DLTl Dynamic load table for follower forces.
CSTMS Coordinate system transformation table.
BGPDTS Basic GRID point definition table.
SILS Scalar index list.
USETD Displacement set definition table.
BDD Damping matrix in d-set.
MDD Mass matrix in d-set.
NLFT Nonlinear transient loads for NOLIN
Output Data Blocks:
UHVNL Input/output, displacement and velocity vectors a t converged time steps during the
direct integration, UHVNL consists of output displacement, velocity, and acceleration
vectors a t output time steps upon completion of the subcase.

FSV Nonlinear element forces at constrained points, equivalent to [liss]{Ys)

a t output time
steps.

ESTNL Input/output, nonlinear Element Summary Table at the last converged step.
FDV Input/output, Two column vectors [F,, P,] at the last converged step.
OESNLl Output stress data for nonlinear elements.
PNL Nonlinear load vector appended at output time steps.
TEL Appended output time step list. Purged at the beginning of subcases.
MUHVNL Input/output, displacement and velocity vectors (2 columns) for K update. It is
deleted at the beginning of a subcase (MUHVNL r UHVNL).

MESTNL Input/output, nonlinear Element Summary Table for K update.

Parameters:
KRATIO Input/output-complex, stiffness ratio t o be used for time step adjustment, initial-
ized t o (l.,O.).

CONV Input/output-integer, value from the last iteration for an input (0 for the initial
iteration), set t o -1 if not converged, +1 if converged.
STIME Input/output-real, starting time or accumulated time used for restarts by user input.
NEWP Input/output-integer, set t o +1 for new subcase, -1 for no subcase switch.
NEWDT Input/output-integer, used t o detect if the time step is adjusted in the previous
subcase. The value is set at the end of each subcase: 1 if ADJUST>O and 0 if AD-
JUST=O.

OLDDT Input/output-real, time step increment used in the previous iteration or time step;
t o be used after the matrix update or subcase switch.

NSTEP Input/output-integer, current time step position for subcase, set to 0 at the beginning
of the subcase.

LGDISP Input-integer, set t o +1 for geometric nonlinearity, -1 otherwise.

MPTSIZE Input/output-integer, the amount of open-core words required to hold the mate-
rial data, initialized t o 0 in MPL and retained during the problem solution.
ITERIDS Input/output-integer, iteration count within a time step.
ITIME Input-real, initial time at the beginning of a subcase.
KTIME Input/output-integer, time-to-go value at the beginning of K update in the DMAP
loop, on output from the module, a negative value is stored, denoting the increment of
time between the K computation and the start of the module.

LASTUPD Input/output-integer, the time step number of the last stiffness update. The
value 0 is used t o indicate if the stiffness update is performed due to the GAP element
during the iteration.

NOGONL Output-integer, set t o +1 to continue, -1 t o terminate.

NBIS Input/output-integer, current count of bisection.
MAXLP Input-integer, maximum limit allowed for element relaxation iteration and the ma-
terial subincrement processes.

TSTATIC Input-integer, default=-1, +1 allows static analysis by ignoring inertia and damp-
ing forces.

LANGLES Input-integer, used t o select the rotation vector approach t o large rotations by a
value of 2. The default (= 1) option is the gimbal angle approach.

NDAMP Input-real, default= 0.025. Introduces numerical damping with a value greater
than zero, which is used in

where 17 represents NDAMP.

8.4 ADAPTIVE TIME STEPPING

8.4.1 Introduction

It is desirable t o have a fully automated method of time integration which renders effective
and accurate solutions of nonlinear problems. This premise is wishful thinking, because such
a method is yet t o be found. Bergan et al. [8.4] have shone some light on this goal, but their
method has inherent flaws with regard to nonlinear problems. The dynamic response charac-
teristics of the structure may change due t o the nonlinearity (geometric or material) or the
mode of excitation. When the mode of response changes, the time step size should be adjusted,
requiring a decomposition. However, the stiffness matrix does not have to be reformulated un-
less it is so required due t o nonlinearity. The authors [8.4] did not distinguish between these
two cases, which are intrinsically different and therefore require distinct treatment as discussed
below. Nevertheless, their idea is salient and can be applied t o nonlinear problems with some
modification t o cope with the difficulties in adjusting the time step size which fluctuates with
the deformation pattern. The deficiency of this method stems from the time lag, which is
coped with by combining it with the bisection method in a coordinated manner. Divergence
activates the bisection process which is valid within the time interval in which the process is
activated. After the solution is achieved, the usual procedure will be resumed for the next time
step.

In the present implementation of the automatic time stepping for nonlinear applications,
the adjustment is designed t o be adaptive to the severity of the nonlinearity in the problem by
monitoring the changes in the stiffness. However, difficulties in the automatic time stepping
have been discovered, when plasticity or GAP elements are involved. In such cases an undesired
time step adjustment can be caused by a drastic change in the stiffness, which may be a
temporary situation. A filtering scheme has been devised t o suppress the effects of the spurious
mass or stiffness in the automatic time step adjustments. In addition, the bisection method
is chosen to be activated only at the time when divergence occurs. The bisection process is
coordinated with the stiffness matrix update so that changes in nonlinearities are properly
reflected while the bisection is in progress.

8.4.2 Automatic Time Stepping Algorithm

The proper time step size, required for accuracy and efficiency, changes continuously in the
transient dynamic environment. This size can be predicted from the dominant frequency, on,
in the incremental deformation pattern based on the Rayleigh's method [8.4]. The dominant
frequency is estimated at every converged time step using the displacement increment (as an
assumed vibration shape) and internal forces by
where {AU,) = { U , - U n - l ) , and { F , - Fn-l) represents a change in the internal forces which
are readily available in memory. A provision should be made to skip the time step adjustment
if {AU,) approaches zero, which is effectively detected by

II
''II max
< UTOL (defaulted t o 0.1)

where 11 u I(, is the maximum value of the norms computed throughout the preceding steps.

The next time step can be predicted by multiplying the current time step by a scaling
factor r , i.e.,
1 27 1
r = - -At,+, -
At n m w, At,
where it is assumed that the dominant period is marched through with m steps. However,
continuous adjustment of the step size at every time step is not a viable option because it
requires too frequent matrix decomposition. To prevent thrashing, a stepping function f ( r ) is
introduced such that
At,+, = f ( T ) At, (8.4.3)
where
f = 0.25 for r < 0.5Rb
f = 0.5 for 0.5Rb 5 r < Rb
f = 1.0 for Rbir<2
f = 2.0 for 2 5 r < 3/Rb
f = 4.0 for '7 1 3/Rb

with Rb being a user specified parameter defaulted t o 0.75.

Since At is allowed t o be quadrupled or quadrisected a t each time step, the time step will
be quickly adapted if the original At is not pertinent. While the adjustment algorithm is based
on the mode of response, however, At may have to be bounded t o represent the external loads
properly, e.g., the impulsive loading case. Upper and lower bounds of At will be defined as

At,,, = MAXR * DT (8.4.4)

and
At,;, = MIN (DT/2**MAXBIS, DT/MAXR)
where D T is the user specified step size, and MAXR and MAXBIS are additional parameters
to be specified in the TSTEPNL card. If the predicted time step size falls outside the bounds,
the time step will be set to the limit.

When the automatic time stepping option is selected, the user-specified DT will be used
only as an initial time step size and the automatic time stepping scheme will be activated. Since
the automatic time stepping scheme ignores user specified D T in time marching, a termination
criterion is required t o wrap up the subcase. The last step may be determined by terminating
the normal time stepping scheme when the currently processed step is close to the last step,
i.e., no further step will be taken if

where the right hand side represents the total duration for a subcase. The adjusted time
step will remain effective across the subcases. Notice that matrix decomposition is required
whenever the time step is adjusted.

It has been observed that the criterion in Eq. (8.4.1) is dictated by spurious mass or
stiffness terms when artificially low or high masses or stiffnesses are used. Large mass or large
spring approach t o the base motion is an example. It could be caused by massless point or
GAP elements as well. This undesired effect should be eliminated, which is accomplished
by checking the "apparent frequency" of each degree-of-freedom by comparing components of
{Fn - Fn-l}and {M AUn} , i.e.,

w9 = ABS [{ F Y 2;;
n 1]
where wi is the apparent angular frequency of the i-th DOF. The central idea is t o eliminate
the effects of the DOF that has extremely small or large apparent frequency compared with the
dominant frequency. The bounds are determined based on the reference frequency as follows:

where wTef is the dominant frequency previously computed. The dominant frequency should
then be computed by

where {AU;} is a modified vector with zeros inserted for the undesired DOF. Undesired time
step adjustment due t o the drastic change in stiffness of the plastic deformation is also properly
suppressed by this method. This is because the plastic deformation casts implications similar
t o the GAP in the automatic time stepping algorithm.

It has also been observed that m of 20, in Eq. (8.4.2)) is adequate if the deformation
remains linear but should be greater for nonlinear problems. Severity of the nonlinearity may
be represented by the stiffness ratio defined as:

where I?' represents the last updated stiffness matrix. The value of m can be a function of the
stiffness ratio. This is implemented as a default option in the following manner:
A* = if X2 2 1
A* = 1 if X2 < 1
a
 m = 20 for A* < 5
m = 40 for 5 5 A* < 1000
No Adjust for 1000 5 A*
This option will be superseded if the user specifies a value for m (MSTEP).

8.4.3 Bisection Algorithm

The automatic time stepping option is implemented to adapt the time step size primarily
to the dynamic environment based on the response pattern. This method, however, is not
likely to alleviate difficulties due t o the change in nonlinearity or divergence. Such difficulties
are resolved by using the bisection method. The material processing routine can detect a
change in nonlinearity. The bisection is activated by a material routine if an incremental
stress (Ai?) exceeds the specified tolerance (e.g., 20 % of the effective stress). The bisection
could also be initiated when a divergence is detected or anticipated. In addition, the bisection is
activated when the maximum number of iterations allowed for an increment is reached without
convergence. The divergence criterion is defined as

NDIV > MAXDIV (8.4.7)

where MAXDIV is a user-specified parameter (defaulted t o 2 ) and NDIV is an internal variable

which is initialized to zero at the beginning of a new time step and incremented as follows:

NDIV = NDIV +2 if Ef > 1 or E f < -10+12

and
NDIV=NDIV+l if -1012<~f<-1 orif I E ~ I > ~
with
Ef =
{auiIT{ R ~ ) and
E;
E; = -
{AUijT {Ri-1) G-l.
In addition, reaching MAXITER without convergence will be treated as a diverging case.

The stiffness will be updated at the last converged step when the first divergence is en-
countered. Then, an iteration is tried without bisection. Bisection normally begins upon the
second detection of divergence within a At, i.e.,
1
at, = - At.
2
Bisection continues until the solution converges, i.e.,
where k is a bisection count. Once the bisection is successful (rendering a converged solution),
the integration proceeds t o the next time step with the same increment ( A t k ) .Further bisection
may be required after achieving converged solution at the intermediate time steps as shown in
Fig. 8.4.1.

If no further bisection is required and the solutions converge without any difficulty, some
recovery process such as doubling time step may be activated. However, any adjustment t o
the time step would require additional matrix decomposition which offsets the advantages
of adjusting time step size. Therefore, it has been determined t o continue stepping through
the specified time interval to completion. The normal process should be resumed when the
user-specified time step at which bisection is activated is fully processed.

When the automatic time stepping is combined with the bisection method and if the bi-
section is activated, the automatic adjustment procedure will be deferred until the divergence
is trapped and the solution process is stabilized. The solution process is considered stabilized
when two successive solutions have converged without requiring further bisection. This is the
time when the automatic time step adjustment is resumed as illustrated in Fig. 8.4.1.

The maximum number of bisections is limited by a user-specified parameter MAXBIS

(defaulted to 5 ) . The bisection process is activated on an as-needed basis. Users also have
an option to suppress bisection by specifying MAXBIS=O. If the bisection is required more
than MAXBIS times, the solution process will continue without bisection by activating the
reiteration procedure, by which the same iteration process is repeated t o find the best attainable
solution. The best attainable solution is a solution obtained from the iteration that yields the
least average error. The average error is defined by averaging relative errors in displacements,
loads and energy. If the i-th iteration yields the least average error, the reiteration procedure
will end a t the i-th iteration and the normal time stepping procedure will be resumed.
 Notes:
1. The baseline represents a duration of a time step before bisection starts.
2. Numbers above the baseline denote a sequence of trial solutions.
3. X denotes converged solutions during the time step ~t .
4. denotes the step which triggers decomposition.
5. 0 denotes reference point for k update.
6. 7/~denotes critical period.
7. Automatic time stepping resumes at # if selected.
8. Numbers below the baseline denote a sequence of bisections.

Figure 8.4.1 Illustration of the Bisection Strategy

8.5 EXPEDIENT ITERATION STRATEGIES

The dynamic tangent matrix may change for the following reasons:
r Stiffness (K) changes due t o geometric, material and/or kinematic nonlinearity.

r Bisection or time-step adjustment causes changes in the effective stiffness.

While the second case only requires decomposition, the first case requires reformulation of the
new stiffness matrix as well. Examples of various cases for the update relative to the bisection
strategy are illustrated in Fig. 8.4.l(a-d). The adaptive method is based on the following
matrix update strategies:
Stiffness is updated at the previously converged position when the divergence is detected
for the first time at a given time step.
r Stiffness matrix update is allowed only once within the same time step.
r If the bisection is in progress, additional K-updates are allowed within a given At at
every KSTEP-th converged bisection, where KSTEP is a user-specified parameter in the
TSTEPNL entry. See Figure 8.4.l(d) for an example.
r Decomposition is performed within the module at every bisection or time step adjust-
ment.
r Whenever the decomposition is performed, the iteration starts from the previously con-
verged position.
r The stiffness matrix is updated at the beginning of each subcase.

The BFGS update [8.5] and the line search [8.6] processes are fully operational during the
iteration. Quasi-Newton (QN) vectors are continuously accumulated up t o MAXQN (user-
specified parameter) pairs until the new stiffness matrix is evaluated. Once the number of
QN vector pairs reaches MAXQN (defaulted t o l o ) , the QN vectors will be updated selectively
based on the condition number of the QN update matrix. As the condition number approaches
unity, the QN update matrix approaches identity matrix, which causes little effect on the matrix
t o be modified. Therefore, the QN vector pair, which renders the condition number closest
t o unity, is replaced or eliminated when a new candidate QN vector pair becomes available.
As a requirement for the reiteration procedure, the QN vectors t o be accumulated are stored
on a temporary basis until a converged solution is obtained. All the accumulated QN vectors
are purged if the stiffness matrix update or the decomposition (due t o the change in At) is
scheduled.

The iterative process for time step ( n + l ) can be summarized as follows:

1. Given at the beginning of the time step are At,, w T e j ,[ M I ,[C],{P,),{AF,), {F,), { U n ) ,
{u,), and the triangularized matrices ( L D L ~for) [A] = [& + c +
M 81.
2. Adjust the time step size, if applicable.
 3. Initialize {U,O+l), {F;+~)and {P,O+l).
4. Decompose, [A] = L D L ~if, # At,. Then, purge the QN vector file.

5. Compute {Ri,+l).
6. Solve for {nuif1)by forward and backward substitution, using the BE'GS update if
applicable.

7. Compute the line search error.

8. If divergence occurs, go t o step 12.
9. Proceed t o the next step if the line search error is less than the tolerance. Otherwise,
perform the line search process and go back to step 7.

10. Compute global error functions and check convergence.

11. If the solution has converged, go to step 16. Otherwise, save the QN vector on a tempo-
rary basis (if applicable) and go t o the next iteration in step 5.

12. If it is the first divergence or the divergence after the KSTEP-th converged bisection
step, update the stiffness matrix, and go back t o step 4. Otherwise, proceed to the next
step.

13. Bisect At and go back t o step 3. However, if the maximum number of bisections allowed
for the time step is reached, proceed to the next step.

14. Go to step 5 to continue iteration. However, if the maximum number of iterations allowed
for an increment is reached, proceed t o the next step.

15. Perform the reiteration procedure t o find the best attainable solution.

16. Accept the solution (Un+l and Fn+1) and append temporary QN vectors to the permanent
QN vector file.

17. Compute the velocity u n + l .

18. Advance t o the next time step with n = n + 1.

The adaptive time stepping method is implemented as an optional feature in SOL 99 with
a new module NLTRD2 for the nonlinear transient response analysis C8.71. Among numer-
ous enhancements from the DMAP standpoint, the efficiency of the adaptive time stepping
method stems from skipping about 50 statements by virtue of the internal decomposition
within NLTRD2 module. Another aspect t o consider in the DMAP is that the output time
steps cannot be predetermined in the variable time stepping. Some efforts were required to
bring together all the data blocks in compatible size and format.

The automatic time stepping and the bisection algorithms are flow-charted in Figs. 8.5.1
and 8.5.2, respectively. These algorithms are combined into the module NLTRD2 as shown in
Fig. 8.5.3. Control parameters for these algorithms t o be specified by the user are gathered in
the TSTEPNL Bulk Data entry.

Efforts have been made to minimize the thrashing in the algorithm and yet to provide a
reliable solution efficiently t o nonlinear problems. Thrashing of the decomposition is effectively
prevented by adopting a stepping function in the adjustment criterion. However, some combi-
nation of user parameters could cause thrashing such as excessive bisection, or repeated cycle
of time-step adjustment and bisection. Default values are tuned to near optimal condition
to the best of our experience. If necessary, the user should make a judicious selection of the
parameter values t o prevent any thrashing.
 calculate K(Au.), M(Aun}
(AFn )=(FnJ-{Fn.l1
,: =m
I IAFnXI [49.9]

(AU.) A [49.10]

DENOM+ { A < ) ~{KAU,}I

DENOMOS {MAU,) I

I
WORK={AU~
J{AF,)
WORK
[49.la]

At,, = MAXR-DT
NBISO=NBIS, NBISK=NBISK+l

Atn,! =MIN ( AL, , MAX ( d i n bf+t1)

0;.,=0:

Figure 8.5.1 Flow Diagram for Automatic Time Step Adjustment

 .NBIS=NBIS+l
NBISK=NBISK+l if NBISK>O

EXlT A: Reiterate for

best solution

EXlT B: Continue bisection

EXlT C: Update stiffness

EXlT D: Terminate job

Figure 8.5.2 Flow Diagram for the Bisection Algorithm

 NL21NT
.
-
'
Waale memmory
S.( KTIUE
NLPlNC

= -
A

hNSTEP

a .
1-(+A1

4EMIN-
-1
NDIV 0-I@
4

- NBlS4 NBlS I
WTER-NO
.;_R,_-
_$ _ _c)_ *_[ &_
+(p:
_ _i~,
_ _ _ _ _ _i_I.*-{ - - - - - -
M+$c]Ic
YES

NUIML CdcUa*N(u;)

. ,,ylm -4
. *Ilapolan P;" ( 0 )
c'"~-p;"(o)+~.,(t)[du",)
P(O)]+N(U:~;)

NUUOG - YES

~4---)I
.A.LDC'
- .
NBISK NBISK t l
.$ -3 'I + I R I ~
.d,' .f - 3"
- or-0
..m.lz. ROOT-no
(1.1
LSCT 0
.A,,?
hr..W ONV
.A'* "
am.

.&-lo" 1-E,

- MolV.1 It ~
t Y *
A p p d OESNLI. FSV PNL

-4)-
canp*I Fa
4 . RITER -YES

f-s.,m[dcur, -Pol
.~-p:"(~)*lp.~(uV M)

NUML v
. - _________----
Fy Fy1.

.en,~: F$[&Y.$c]@+&~
----
w7b1

- Fam DGMAG. DELW

a*-------

. p-ot,[Q~;l<E~amdi-l) ~(IE~I<(O'.ET~)I

- e - (d,}'{fW:
.
.
S.W U -
E' kr €FIRST LSCT 0
e.~s~ol~IE;l~ool-~~andlE'I>LST~

--. nd p d lor STATIC1 . NDIV-NDIVtlll

. (x x) 6nol E 9 n S H L 4
I E ~ I> I U 10"dh< 9
+
m dWL40
EXIT 1 K up6.n
.. sa
.-renuimp-
~~(aaum*llanutm
CONV paramelm
EXIT z mo(a w a r e
EXIT 3 ~bmmu(

. Q FIC\TTT for~the
n i 3 ~ ~ m NTJTRD2 Module
8.6 VERIFICATION PROBLEMS

Five example problems are presented in this section t o validate the effectiveness of the algorithm
and to verify the program's performance.

8.6.1 Elastic-Plastic Rod with an End Mass Subjected to a Step Loading

Consider a single-degree system (mass-rod model), in which the rod is made of an elastic-
perfectly plastic material. Fig. 8.6.l(a) shows the model subjected t o a step loading with four
different magnitudes: namely, 0.5Fy, 0.66Fy, 0.88Fy and F,, respectively, in which Fy is the
axial force corresponding t o the initial yield of the rod. As a result, the mass responds with
different patterns. All four cases are included in a single run by four disjoint models to test
the effectiveness of the automatic time step adjustment. The input data for this problem is
given in Table 8.6.1.

The theoretical solution can be considered in three different regimes: the elastic regime,
plastic, and rebounding (unloading) regime. In the elastic regime the solution can be obtained
simply by u = f ( 1 - coswt) which is valid throughout the time history if P 5 $F,. For
;F, < P 5 F,, there is a point in time where the stress in the rod reaches the yield point.
+
During the plastic deformation, the ~olutionmay be obtained from mu F, = P using proper
initial conditions. When the displacement reaches the maximum value, the system is governed
by an elastic equation for rebound. If P > F,, no rebound occurs and the mass will break
away from the rod.

Two analyses have been made: one with an initial time step of 0.001 second and the
other with 0.01 second. The displacement responses of the two analyses are almost identical
and agree with the closed-form solution [8.8] as shown by A, B, C and D in Fig. 8.6.l(b).
Also shown in Fig. 8.6.l(b) is the variation of the time step increment. In the first analysis,
the program detected that the time step is too small and automatically quadrupled it t o 0.004
second. In the second analysis when the initial time step was 0.01 second, the program reduced
the time step t o 0.005 second. No unnecessary time step adjustments were performed during
the elastic-plastic vibration. Bisection did not occur in this analysis.

8.6.2 Impulsively Loaded Clamped Plate

An elastic-plastic plate clamped at two ends is subjected t o an impulsive load over the center
portion as shown in Fig. 8.6.2(a). A quarter section of the plate is modeled by five QUAD4
elements with symmetric boundary conditions using MSCJNASTRAN. The impulsive load
was simulated by initial velocities applied a t the nodal points as shown in Fig. 8.6.2(a). The
material is elastic-perfectly plastic and the large displacement effect is also included in the
response analysis for the duration of one msec. The input data for this problem is given in
Table 8.6.2.

Four different initial time step sizes (namely, 0.1 msec, 0.05 msec, 0.02 msec and 0.0025
msec) were used in the analysis. Predicted time histories of the mid-span displacement of
the four cases are nearly identical as shown in Fig. 8.6.2(b). Also shown in Fig. 8.6.2(b)
are the experimental data obtained by Balmer and Witmer [8.9] and the solution from the
explicit integration obtained by Belytschko et al. [8.10]. This figure demonstrates that the
MSC/NASTRAN solution agrees more closely with the test results. Fig. 8.6.2(b) also shows
the variation of time increments due to the automatic time step adjustment. In all of the cases
the time step is adjusted to approximately 0.02 msec, from either the high or low value of a
specified initial increment size. Bisection was not activated in this analysis.

A question arose if a coarse mesh resulted in a more accurate solution by beneficially

filtering the high frequency response. In order to study this point, the analysis was repeated
with a 10-QUAD4 model for comparison with the 10-element model of Belytschko et al. [8.10].
The result is even more favorable as shown in Fig. 8.6.2(b). It is noted, however, that the time
step variation is quite different from the 5-element model. Except for the case of 0.0025 msec,
bisection processes were activated at the beginning to reduce At below 0.0125 msec. This
step size was maintained until approximately 0.4 msec, when At was doubled and remained
unchanged to completion in all four cases.

8.6.3 Simply Supported Beam with a Restrained Motion

A simply supported elastic beam is subjected t o a single cycle of a sinusoidal forcing function
at the quarter span as illustrated in Fig. 8.6.3(a). The beam is modeled using 20 linear BEAM
elements. A stopper (snubber) is present underneath the center of the beam with a clearance
of 0.02 inch. This gap is simulated by a displacement dependent force which is active only
when the beam is in contact with the snubber. The problem is linear except for the effect of
the stopper. The input data for this problem is given in Table 8.6.3.

To demonstrate the effectiveness of the automatic time step adjustment, the same model
was analyzed with two different initial time step sizes: 0.0002 sec and 0.002 sec. Each case was
analyzed twice: once with and once without using the automatic time step adjustment. The
displacement responses at the loading point for the four cases are compared in Fig. 8.6.3(b), in
which the time steps are indicated by the symbol x in the figure. It is demonstrated that the
time step size is adjusted to an equitable value (fluctuating between 0.0004 sec and 0.001 sec)
regardless of the initial value. When the automatic time step adjustment is not activated, the
solution process either wastes computing time by using time increments which are too small
or renders inaccurate solutions by using time step sizes which are too coarse. All the solutions
converged without activating the bisection process.

The model is modified for the performance comparison t o include 200 beam elements. The
duration of the analysis time span is also extended t o 0.1 sec. Five different time step sizes
(0.0001 sec, 0.0002 sec, 0.0005 sec, 0.001 sec, and 0.002 sec) have been used t o investigate the
efficiency and effectiveness of the adaptive method. All computer runs were made on the VAX
8700 system at its off-peak time. The results are tabulated below:

CPU Time Comparison

User-specified At(sec) 0.002 0.001 0.0005 0.0002 0.0001

Analysis with fixed 56 86 140 327 66 1
time step
Analysis with adaptive 135 145 144 150 158
time step algorithm

Note : Accuracy of the CPU time is estimated t o have a 5% deviation.

It is noted that the CPU time is increasing almost proportionally t o the decrease in the
time step size if the time step size is fixed. However, in the case of the adaptive time stepping
method the CPU time remains approximately constant regardless of the initial time step size.
The adaptive time stepping method provides the definite advantage which is the avoidance of
the trial-and-error process. Furthermore, it is likely to be more efficient in terms of computing
time than the fixed time stepping, because the proper time step size is not only difficult t o
predict in advance but it also changes throughout the analysis history.

8.6.4 Bouncing Weight on an Elastic Platform

A 4-lb weight is resting on a spring-supported platform which is initially compressed 4 inches in

the position as shown in Fig. 8.6.4(a). The platform is then released, and after moving 2 inches,
is halted by a stopper. The finite element model consists of one ROD element representing the
spring, a mass for the weight block, and two GAP elements simulating contacts between the
mass-platform and the platform-stopper interfaces as shown in Fig. 8.6.4(a). The gravity field
is applied and a small mass is assigned to the platform to suppress its jittering. The stopper is
represented by a fixed nodal point and the initial deformation in the spring by an equivalent
step force. The input data for this problem is given in Table 8.6.4.

A closed-form solution is obtained for the massless platform. While the weight is riding on
+
the platform, the motion is governed by mx bx = -mg with initial conditions xo = -4 and
k o = 0. The velocity of the weight block at the time of its departure from the platform can
be obtained by the conservation of energy. This velocity becomes the initial condition for the
equation h = vot - t g t 2 , which governs the motion after the weight departs from the platform.

The response predicted by MSCINASTRAN is compared with the theoretical solution as

shown in Fig. 8.6.4(b). The time steps are indicated by the symbol x in the figure. The
initial time step size was specified as 0.0025 sec, which was automatically adjusted to 0.01
sec a t the initial adjustment phase as shown in Fig. 8.6.4(b). In this analysis, bisection was
activated when the block was departing from the platform and changing the status of the GAP,
resulting in 0.005 sec. This analysis proves the effectiveness of the current adaptive method
for the problems involving the gaps and/or drastic changes of the structural stiffness.

8.6.5 Shallow Spherical Cap under a Concentrated Apex Load

A thin shell in the shape of a shallow dome is clamped at the periphery and subjected t o a step
loading of a concentrated force at the apex as shown in Fig. 8.6.5(a). The geometric nonlinear
effects are included, but the material properties are assumed to remain elastic througthout the
analysis.

Considering the symmetry, only one quarter of the cap is modeled using 27 QUAD4 ele-
ments with symmetric boundary conditions as illustrated in Fig. 8.6.5(a). The input data for
this problem is given in Table 8.6.5. This problem is chosen to demonstrate the performance of
the automatic time step adjustment and the bisection method combined. Hence, the analysis
was deliberately started with a relatively larger time step size, 20 psec. This time step size is
10 times larger than the size adopted in the previous study by Mondkar and Powell using an
axisymmetric model with ten 8-noded quadrilateral elements [8.11].

As expected, the solution diverged at the very first time step and the algorithm activated
the bisection process, resulting in At = 10 psec. After obtaining converged solutions at two
consecutive time steps, the program detected that the time step size was still too large and
adjusted it t o a smaller size, 2.5 psec. The analysis continued without any adjustment until
the displacement approached the peak, where At was adjusted to 2 psec (which was the lower
bound allowed for adjustment). The step size was again adjusted to 4 psec past the peak
response, which was later adjusted back t o 2 psec before completion. No further bisection
was required t o complete the analysis. The displacement response a t the apex is compared
with the results by Mondkar and Powell [8.11] in Fig. 8.6.5(b). Fig. 8.6.5(b) also shows the
variation of time step size due to the adaptive time step adjustment.

When the same model was analyzed again with a smaller At(i.e., 2 psec), the adjusted time
step size was reduced t o as low as 0.5 psec during the analysis. In this analysis, however, the
bisection did not occur. The resulting response agrees with that of the previous case shown in
Fig. 8.6.5(b) except for the minor details of spikes.
Figure 8.6.1 (a). Elastic-Plastic Rod Subjected to a Step Loading
 Displ.
4.
(in)

0.1 0.2
Time (sec)

0.01 --
DT = 0.01 sec
At
(sec) 0.005 - ...................................... -......................................... .."................
1, DT = 0.001 sec

Figure 8.6.1 (b). Elastic-Plastic Rod: Displacement Response

and Time Step Variation
 Finite Element Mesh

p= 2.61 x lo-' lb-sec2/i$

E = 10.4 x lo6 psi
a, = 41400 psi, yield stress
v = 5000 in/sec.
v = 3904 inlsec.

Figure 8.6.2(a). Impulsively Loaded Clamped Plate

Figure 8.6.2 (b). Clamped Plate: Central Displacement Response
and Time Step Variation
8.6 - 8
 20 Beam Elements

(~10010)
0.01 1 (sec)

Forcing Function Representing GAP

Figure 8.6.3(a). Simply Supported Beam with a Restrained Motion

 Time(sec .) Time(sec.)
with adjustment . without adjustment
User-specified At = 0.0002 Sec.

Time(sec.) Time(sec.)
with adjustment without adjustment
User-specified At = 0.002 Sec.

Figure 8.6.3(b). Simply Supported Beam: Loading Point

Displacement Time History
 I I

-i i stopper

1. Mass (4 lbs.)

Sprt ng k = 10 lb/in. (initially compressed 4 ins.)

P1 atform

Structural Configuration

w
closed gap
40 (lbs)

k = 10 lb/in

MSCJNASTRAN Model

Figure 8.6.4(a). Bouncing Weight on an Elastic Platform

8.6 - 11
 A A 4
16. +

14. "

Weight
12. "

g 10.

//
*,
W

\
Z
,(:

z2
+.a . 8.
a
W ,,
/da
t
0 Theoretical
b
8 8tt
+ MSCJNASTRAN
t3 6. I
i
"

t
8 8
t <
4. '* t
t

"
2.

0. a f

-2. + + -
0. 0.1 0.2 0.3 0.4 0.5 0.6

TIME (SEC)

Bisection
-
'om\

7.5
U
P

Figure 8.6.4(b). Bouncing Weight: Displacement Time History

 R = 4.75 in.
a = 0.90 in.
H = 0.08589 in
t =0.01576 in.
E = 10000 ksi, Young's modulus
v = 0.3, Poisson's ratio
p = 2.45 x lo-' lb-sec2/in4
P = 100. lbs., constant applied force

Figure 8.6.5(a). Shallow Spherical Cap under a

Concentrated Apex Load
 + MSCINASTRAN
0 Mondkar and Powell

Figure 8.6.5(b). Spherical Cap: Apex Displacement Response

 Table 8.6.1 Input Data Listing for Elastic-Plastic Rod

ID TBIGGS4,V66 $ SSH 10/22/87

DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
TIME 5 $ VAX TETRA
SOL 99 $
CEND
TITLEzELASTO-PLASTIC VIBRATION PROBLEM (LOAD = .5/.66/.88/1. FYI
SUBTITLE=NONLINEAR TRANSIENT ANALYSIS FOR 0.25 SEC.
LABEL=J. M. BIGGS, SECTION 2.7, P. 69-71
ECHO=BOTH
SET I = 11,21,31,41
DISP=I
ACCE=I
VELO=1
STRESS=I
SEALL=ALL
SUBCASE I
DLOAD=I00
TSTEPNL=IOO
OUTPUT(XYPLOT)
CSCALE = 1.5
XTITLE = TIME IN SECS
XGRID LINES = YES
YGRID LINES = YES
YTITLE = DISPLACEMENT GRID 1
XYPLOT DISP REsP/II (T2) ,21(~2) ,31(~2),41(~2)
YTITLE = VELOCITY GRID 1
XYPLOT VELO RESP/II(T2), 21( ~ 231 ) (~2)
~ ,4i(~2)
YTITLE = ACCELERATION GRID 1
XYPLOT ACCE RESP/II(T2) $21( ~ 231 ) (~2)
~ ,4i(~2)
YTITLE = STRESS IN RODS
XYPLOT STRESS RESP/11(2) ,21(2) ,31(2) ,41(2)
YTITLE = STRESS IN ROD
XYPLOT STRESS RESP /11(2) ,21(2) ,31(2) ~41(2)
BEGIN BULK
$ GEOMETRY AND CONNECTIVITY
GRID 11
GRID 12 240.
CROD 11 2 12 II
CMASS2 I1 10000. 11 2
GRID 21
GRID 22 240.
CROD 21 2 22 21
CMASS2 21 10000. 21 2
GRID 31
GRID 32 240.
CROD 31 2 32 31
CMASS2 31 10000. 31 2
GRID 41
GRID 42 240.
CROD 41 2 42 41
CMASS2 41 10000. 41 2
$ PROPERTIES
PROD 2 2 .6672
MAT1 2 30. +06
MATS1 2 PLASTIC 0 . 0
$ LOADING
DLOAD 100 I. I. 101
+DL1 I. 104
DAREA 201 11 2 22650.
DAREA 202 21 2 30000.
DAREA 203 31 2 40000.
DAREA 204 41 2 45300.
TLOADl 101 201 0 50
TLOADI 102 202 0 50
TLOADI 103 203 0 50
TLOADI 104 204 0 50
TABLED1 50
+ABC 0. 1. 10. 1. ENDT
$ PARAMETERS
PARAM WTMASS .0025907
TSTEPNL 100 50 .0050 I ADAPT
tTS2I
+TS22
ENDDATA
 Table 8.6.2 Input Data Listing for Impulsively Loaded Plate

ID GMNL08, V66 $ SHL 5/21/87 $IMPULSIVE LOAD

TIME 5 $ VAX8700(TETRA) CPU MIN.
SOL 99 $ NL-TRANSIENT SCR=YES
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATION
CEND
TITLE = NONLINEAR IMPULSIVELY LOADED CLAMPED PLATE (~6=1.+4) 99Iq4
SEALL = ALL
SPC = 200
IC = 300
TSTEP = 400
sTRE(PL0T) = ALL
$
SET I = I
DISP = I
VELO = I
ACCE = I
OUTPUT (XYOUT)
CSCALE 1.3
XGRID LINES = YES
YGRID LINES = YES
XTIT = TIME
XMAX = 1.0-3
$
YTITLE = D I S P L 1
XYPLOT DISP / l(T3)
YTITLE = V E L 0 C I
XYPLOT VELO / 1(T3)
YTITLE = A C C E L I
XYPLOT ACCE / l(T3)
YTITLE = D I S P L 3 I
XYPLOT DISP / 31(T3)
YTITLE = V E L 0 C 3 1
XYPLOT VELO / 31(T3)
YTITLE = A C C E L 3 I
XYPLOT ACCE / 31(T3)
YTITLE = S T R E S S I -
XYPLOT ELST / 1(3)
BEGIN BULK
TSTEPNL 400 200 5.0-6 I ADAPT
GRID I 0. 0.0
GRID 2 0. t.6
GRID I1 1. 0.0
GRID 12 I. t.6
GRID 21 2. 0.0
GRID 22 2. +. 6
GRID 31 3. 0.0
GRID 32
GRID 41
GRID 42
GRID 51
GRID 52
CQUAD4 I 100
CQUAD4 2 100
CQUAD4 3 100
CQUAD4 4 100
CQUAD4 5 100
PSHELL 100 100
+PS -.05 +.05
MAT1 100 10.4+6 .3
MATS1 100 PLASTIC 0.0
PARAM COUPMASSI
PARAM K6ROT 1.0+4
PARAM NLAYERS 5
PARAM LGDISP 1
PARAM W 4 1.2E4
SPCI 200 246
SPCI 200 123456
SPCI 200 156
TIC 300 I
TIC 300 2
TIC 300 11
TIC 300 12
ENDDATA
 Table 8.6.3 Input Data Listing for Beam with a Restrained Motion

ID TNOLIN, V66 $ SHL 5/19/87 $ MSC, D9918 $ KDB 12JUN84

TIME 5 $ VAX8700(TETRA) CPU MIN.
SOL 99 $ NONLINEAR TRANSIENT ANALYSIS
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
CEND
TITLE= DYNAMICS HANDBOOK DEMONSTRATION PROBLEM
SUBTITLE= DIRECT TRANSIENT RESPONSE, NONLINEAR FORCE
LABEL= NOLIN IN SOL 99
SEALL=ALL
ECHO=UNSORT
SPC=IOO2
SET 1=10005
SET 2=10010
SET 3=10005,10010
DISPL=3
$ ACCEL=2
VELO=3
OLOAD=I
NLLOAD=2
SUBCASE I $
DLOAD=30
TSTEPNL=20
NONLINEAR=I3 $ SELECT NONLINEAR FORCE
OUTPUT(XYPLOT)
CSCALE=I.8
PLOTTER NAST
CURVELINESYMBOL = I
XGRID=YES
YGRID=YES
XTITLE=TIME
YTITLE=DISP 10005 T3
XYPLOT DISP /10005(T3)
YTITLE=DISP 10010 T3
XYPLOT DISP /10010(T3)
YTITLE=APPLIED LOAD 10005 T3
XYPLOT OLOAD RESPONSE/l0005(T3) $ PLOT THE INPUT
YTITLE=NONLINEAR LOAD 10010 T3
XYPLOT NONLINEAR RESPONSE/IOO~O(T~)
BEGIN BULK
$.......2.......3.......4.......5.......6.......7......
$ MODELING INFORMATION FOR BEAM ONLY
CBAR 101 100 I0000 10001 0.0 0.0
CBAR 102 100 10001 10002 0.0 0.0
CBAR 103 100 10002 10003 0.0 0.0
CBAR 104 100 10003 10004 0.0 0.0
CBAR 105 100 10004 10005 0.0 0.0
CBAR 106 100 10005 10006 0.0 0.0
CBAR 107 100 10006 10007 0.0 0.0
CBAR 108 100 10007 10008 0.0 0.0
CBAR 109 100 10008 10009 0.0 0.0
CBAR 110 100 10009 I0010 0.0 0.0
CBAR Ill 100 10010 10011 0.0 0.0
CBAR 112 100 10011 10012 0.0 0.0
CBAR 113 100 10012 10013 0.0 0.0
CBAR 114 100 10013 10014 0.0 0.0
CBAR 115 100 10014 10015 0.0 0.0
CBAR 116 100 10015 10016 0.0 0.0
CBAR 117 100 10016 10017 0.0 0.0
CBAR 118 100 10017 10018 0.0 0.0
CBAR 119 100 10018 10019 0.0 0.0
CBAR 120 100 10019 10020 0.0 0.0
CONM2 12 10010 .I
GRID 10 50. -1.
GRID 10000 0.0 0.0 0.0
GRID 10001 5. 0.0 0.0
GRID 10002 10. 0.0 0.0
GRID 10003 15. 0.0 0.0
GRID 10004 20. 0.0 0.0
GRID 10005 25. 0.0 0.0
GRID 10006 30. 0.0 0.0
GRID 10007 35. 0.0 0.0
GRID 10008 40. 0.0 0.0
GRID 10009 45. 0.0 0.0
GRID 10010 50. 0.0 0.0
GRID lOO11 55. 0.0 0.0
GRID 10012 60. 0.0 0.0
GRID 10013 65. 0.0 0.0
GRID 10014 70. 0.0 0.0
GRID 10015 75. 0.0 0.0
GRID 10016 80. 0.0 0.0
GRID 10017 85. 0.0 0.0
GRID I0018 90. 0.0 0.0
GRID 10019 95. 0.0 0.0
GRID 10020 100. 0.0 0.0
MAT1 1000 3.+7 .3 .3
PARAM GRDPNT 10010
PARAM WTMASS .002588
PBAR 100 1000 0.31416 0.15708
++0000011. 0.
SPC 1002 10 123456
SPC 1002 10020 3 10000 3
$
$ MODELING INFORMATION FOR CENTER S P R I N G
CROD 10 10 10 I0010
MAT1 10 10. 0.
PROD 10 10 1.
MATSI,I0,,PLASTIC,0.0DIDI,3.0E8
$ LOADING AND SOLUTION INFORMATION
TLOAD2 30 33 0.0
DAREA 33 10005 3 47.2
TSTEPNL 20 199 .0002 I ADAPT
+TSI
+TS2
$
$ MODELING INFORMATION FOR NONLINEAR SPRING
$NOLINI SID GI CI S GJ CJ T
NOLINI 13 10010 3 1.0 10010 3 13
$TABLED1 ID +ABC
$+ABC XI Yl X2 Y2 X3 Y3
TABLED1 13 +TAB
+TAB -2.5E-2 4.95 -2.OE-2 0. 0. 0. ENDT
ENDDATA
 Table 8.6.4 Input Data Listing for Bouncing Weight Model

ID VGAPM,V66 $ SHL 12/11/82, SSH 1/13/88

TIME 5 $ FOR VAX
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
SOL 99
CEND
TITLE=GAP VERIFICATION
SUBTITLE=NONLINEAR TRANSIENT ANALYSIS
ECHO=UNSORT
SET 1=201,202
SET 2=200 $,203
SET 3=22 ,23
SET 4=20 $ CONROD
DISP=I
$ OLOAD=I
$ SPCF=2
ELFORCE=4
STRESS=3 $ FOR GAP FORCES
SEALL=ALL
LOADSET=20
SUBCASE I
DLOAD=210
TSTEPNL=21
SUBCASE 2
DLOAD=210
TSTEPNL=2I
OUTPUT (XYPLOT)
CSCALE=I.3
PLOTTER=NAST
XTITLE=TIME IN SEC.
YTITLE=DISPL. IN INCH
$ YMIN=O.
$ YMAX=I5.
XYPLOT DISP RESP/~OI (TI), 202(TI)
BEGIN BULK
TSTEPNL 21 200 0.0025 I ADAPT
+TSPI
+TSP2 0
$ PROPERTIES
PARAM W4 31.
$PARAM U3 1000.
$PARAM G 0.01
MAT1 I 100.
PGAP 3
+PG3 5.-3
PGAP 4 2.
+PG4 5.-3
$ LOADING
FORCE 201 20 1
GRAV 200 386.
LOAD 202 I. I.
LSEQ 20 30 1 202
$TLOAD2 210 301
TLOADI 210 30 1
TABLED1 220
+TAB1 0. 0. 0.001 1. ENDT
$ BOUNDARY CONDITIONS
GRDSET 23456
$ GEOMETRY AND CONNECTIVITY
GRID 200 -6. 123456
GRID 201 4.
GRID 202 4.1
GRID 203 6.
CONROD 20 200 201
CONM2 24 201
CONM2 21 202
CGAP 22 3 201
CGAP 23 4 20 1
ENDDATA
 Table 8.6.5 Input Data Listing for Shallow Spherical Cap

ID SCAP,V66 $KI-OOK KIM $ MAY 1987, GAD MSC, SAAB4

SOL 99 $ NONLINEAR TRANSIENT RESPONSE
DIAG 50 $ TRACE ITERATION PROCESS
TIME 60 $ CPU TIME IN MINUTES
CEND
TITLE = TEST PROBLEM 1.4
SUBTITLE= SHALLOW SPHERICAL CAP WITH CONCENTRATED APEX LOAD
LABEL = REF: MONDKAR AND POWELL, IJNME 11:499-520 (1977)
SEALL = ALL
SET 130 = 1 $ THRU 30
DISP = 130
SET 14 = 1 $ THRU 4
OLOAD = 14
SPC = 1234
SUBCASE I
DLOAD = 100
TSTEPNL = 250
OUTPUT(PLOT)
CSCALE 1.5
PLOTTER NAST
PAPER SIZE 29. X 21.
SET I = ALL
SET 2 = I THRU 6, 10, 11, 17, 18, 24, 25, 31, 32
AXES X, Y, Z
VIEW 34.27, 23.17, 0.
FIND SCALE ORIGIN 1 SET I
PLOT SET I LABEL BOTH SHRINK
FIND SCALE ORIGIN 1 SET I
PLOT STATIC 0 SET I
AXES Y, MX, Z
VIEW O., O., 0.
FIND SCALE ORIGIN I SET 2
PLOT STATIC 0 SET 2
OUTPUT(XYPLOT)
CSCALE I. S
PLOTTER NAST
XPAPER = 29.
YPAPER = 21.
CURVELINESYMBOL = 2
XGRID = YES
YGRID = YES
XTITLE = TIME
YTITLE = VERTICAL DISPLACEMENT AT GRID I
XYPLOT DISP /I(T3)
BEGIN BULK
TSTEPNL 250 80 5.-6 I ADAPT
 +TSI
+TS2 0
PARAM LGDISP +I
CORD2S 10
+I 1. 0.
GRID I 10
GRID 2 10
GRID 3 10
GRID 4 10
GRID 5 10
GRID 6 10
GRID 7 10
GRID 8 10
GRID 9 10
GRID 10 10
GRID I1 10
GRID 12 10
GRID 13 10
GRID 14 10
GRID 15 10
GRID 16 10
GRID 17 10
GRID 18 10
GRID 19 10
GRID 20 10
GRID 21 I0
GRID 22 10
GRID 23 10
GRID 24 10
GRID 25 10
GRID 26 10
GRID 27 10
GRID 28 10
GRID 29 10
GRID 30 10
GRID 31 10
GRID 32 10
GRID 33 10
GRID 34 10
GRID 35 10
GRID 36 10
' GRID 37 10
$
CqUAD4 I 1
CqUAD4 2 I
CqUAD4 3 I
CqUAD4 4 I
CQUAD4 5 I
CqUAD4 6 I
CQUAD4 7
CQUAD4 8
CQUAD4 9
CQUAD4 10
CQUAD4 I1
CqUAD4 12
CqUAD4 13
CQUAD4 14
CQUAD4 15
CQUAD4 16
CQUAD4 17
CQUAD4 18
CQUAD4 19
CQUAD4 20
CQUAD4 21
CQUAD4 22
CQUAD4 23
CqUAD4 24
CQUAD4 25
CQUAD4 26
CQUAD4 27
$
PSHELL I
MAT1 I
$
SPCI I
SPCI 2 5
SPCI 2 9
SPCl 3 THRU
SPCI 4 6
SPCI 4 THRU
SPCI 4 THRU
SPCI 4 THRU
SPCADD 1234 3
$
DAREA 10
TLOADI 100
TABLED1 2
+TBDI 0.0 ENDT
EIDDATA
Chapter 9

NONLINEAR TRANSIENT
RESPONSE ANALYSIS

9.1 USER INTERFACE

The input data required for SOL 99 or SOL 129 is a combination of direct time integration
control data, similar to SOL 69 (for direct linear transient with superelements), and nonlinear
modeling data similar t o SOL 66 (for nonlinear statics). The nonlinear properties and/or
effects are defined by nonlinear material data (MATS1 and TABLESI), gap elements (GAP)
for nonlinear interface, and PARAMeter LGDISP for geometric nonlinearity. The transient
effects are produced by time-dependent loading functions (TLOADi, DAREA, LSEQ, etc.),
damping (parameters, elements and material data), and mass properties. Massless degree of
freedom should be avoided for numerical stability and the small damping is also recommended.

The unique data required for SOL 99 or 129 is supplied on the TSTEPNL Bulk Data entry.
The TSTEPNL entry in itself is a combination of the TSTEP entry for direct time integration
and the NLPARM entry for nonlinear iteration control. Restarts are controlled by parameters
(LOOPID, STIME, SLOOPID and SDATA) which can be specified either in the Case Control
section or Bulk Data section. Some optional parameters (TSTATIC, NDAMP) are provided
for additional control or capabilities.

9.1.1 Case Control

Each subcase defines a time interval starting from the last time step of the previous subcase and
the interval is subdivided into small time steps. The output time is labeled by the cumulative
time, including all previous subcases. There are advantages t o divide the total duration of
analysis into many subcases so that each subcase does not have excessive number of time
steps, e.g., not t o exceed 200 steps. The data blocks containing solutions are generated at
the end of each subcase t o store in the database for output process and restarts. As such,
converged solutions are apt t o be saved at many intermediate steps in case of divergence and
more flexible control becomes possible with multiple subcases.

The input loading functions may be changed for each subcase or continued by repeating the
same DLOAD request. However, it is recommended to use the same TLOAD Bulk Data for
all the subcases in order t o maintain the continuity between subcases, because TLOADi data
defines the loading history as a function of cumulative time. Static loads (PLOADi, FORCEi,
MOMENTi) may be associated with time-dependent functions by the Bulk Data LSEQ which
can be selected by a Case Control command LOADSET. However, no thermal loads or enforced
displacements (SPCD) are not allowed in the nonlinear transient analysis. Nonlinear forces as
function of displacements or velocities (NOLINi) may be selected and printed by Case Control
commands NONLINEAR and NLLOAD, respectively. Each subcase may have a different time
step size, time interval, and iteration control selected by the TSTEPNL request. The Case
Control requests which may not be changed after the first subcase are: SPC, MPC, DMIG,
and T F .

Output requests for each subcase are processed independently. Requested output quantities
for all the subcases are appended after the computational process for actual output operation.
Available outputs are DISPLACEMENT, VELOCITY, ACCELERATION, OLOAD, STRESS,
FORCE, SDISPLACEMENT, SVELOCITY, SACCELERATION, NONLINEAR (NLLOAD)
and SPCFORCE. However, element force output and GRID point stresses are not available
for nonlinear elements.

Initial conditions (displacement or velocity) can be specified by the Bulk Data TIC se-
lectable by the Case Control command IC. If initial conditions were given, all of the nonlinear
element forces and stresses must be computed to satisfy equilibrium with the prescribed initial
conditions. On the other hand, initial conditions could be a part of the nonlinear analysis by
applying static analysis for the preload using PARAM, TSTATIC in the first subcase. Then
the transient analysis can be performed in the ensuing subcases. Associated with the adaptive
time stepping method, the PARAMeter NDAMP is used t o control the stability in the ADAPT
method. The parameter NDAMP represents the numerical damping (recommended value for
usual case is 0.01) which often required t o improve the stability and convergence in the contact
problems.

All the superelement model generation options and matrix reduction options are allowed
for the linear portion of the structure. The generalized dynamic reduction, component mode
synthesis, and Guyan reduction may be performed for upstream superelements. The residual
superelement may contain scalar degree of freedom representing linear modal formulations.
9.1.2 Implicit Integration Control: TSTEPNL Data

The input fields of the TSTEPNL Bulk Data entry specify the time step size, the number of
steps, and the output interval as well as the nonlinear iteration options. The TSTEPNL Bulk
Data is selectable by the Case Control command TSTEPNL. Although the same TSTEPNL
Bulk Data may be selected by more than one subcase, it is recommended t o select different
TSTEPNL entry for each subcase in preparation for changes in the restarts.

The choice of time step size is determined primarily by the frequency content of the input
load and the highest frequency mode-of-interest. A general guideline is that 7 or more steps
per cycle be provided for reasonable accuracy. Modes with shorter periods (higher frequency)
will be attenuated by the numerical process. Highly nonlinear problems may require smaller
step size. However, the adaptive time stepping capability is the vital part of SOL 99 or 129
and is recommended for any problem (linear or nonlinear). Nevertheless, the initial time step
size should be estimated by the user according t o the aforementioned principles.

A caution is necessary in using the automatic time step adjustment if the forcing function is
a short duration pulse. Since the automatic time step adjustment algorithm does not consider
the loading history, short pulses could be skipped if the time step is automatically adjusted t o
a larger value than the pulse duration. It is advised that a drastic change in the time step size
between subcases be avoided. The drastic change, e.g., ratio exceeding 1000, could cause a loss
of accuracy upon subcase switch. In such case, an intermediate subcase should be provided
for a transition period of short interval t o reduce the ratio.

The parameters for controlling the accuracy and stability of the incremental and iterative
process can be specified in the TSTEPNL Bulk Data entry. The controls are applicable t o the
automatic time step adjustment and bisection process in addition t o stiffness matrix updates,
BFGS updates and line searches similar t o those on the NLPARM Bulk Data entry. Since
default values are resulted from numerous test runs, the analysis should be started with the
default setting and changed if necessary. The TSTEPNL data format (Version 67) is shown
below with default values:
 Format:
1 2 3 4 5 6 7 8 9 10

TSTEPNL ID NDT DT NO METHOD KSTEP MAXITER CONV +TNLl

+TNLl EPSU EPSP EPSW MAXDIV MAXQN MAXLS FSTRESS +TNL2

+TNL2 MAXBIS ADJUST MSTEP RB MAXR UTOL RTOLB

E x a m ~ l ewith defaults:

TSTEPNL 250 1 ADAPT 2 10 PW +TNLl

+TNLl l.E-2 1.E-3 1.E-6 2 10 2 0.2 +TNL2

+TNL2 5 5 0 0.75 16.0 0.1 20.

The TSTEPNL Bulk Data entry is selected using ID by the Case Control command
TSTEPNL. Each subcase (residual superelement solutions only) requires a TSTEPNL entry.
Multiple subcases are assumed t o occur sequentially in time. Therefore, the initial conditions
of each subcase are defined by the end conditions of the previous subcase.

The NDT field specifies the number of time steps with DT as the size of each time step. The
total duration for the subcase can be assessed by multiplying NDT with DT (i.e., NDT*DT).
The time increment (At) remains constant during the analysis in AUTO and TSTEP options,
and is equal t o DT. However, the time increment (At) changes during the analysis in the
ADAPT option and the actual number of time steps will not be equal t o NDT. In ADAPT
option DT is used as an initial value for At.

The NO field specifies the time step interval for output, i.e., every ~0~~step solution is
saved for output. The data will be output at steps 0, NO, 2N0, ... etc., and the last converged
step for printing and plotting purposes. The Case Control command OTIME may also be used
t o control the output points.

The METHOD field selects an option for direct time integration and the stiffness matrix
update strategies among ADAPT, AUTO and TSTEP. If the AUTO option is selected, the
program automatically updates the stiffness matrix t o improve convergence while KSTEP value
is ignored. If the TSTEP option is selected, the program updates the stiffness matrix every
K S T E P ~increment
~ of time. If the ADAPT option is selected, the program automatically
adjusts the incremental time and uses the bisection algorithm in case of divergence. During
the bisection process in ADAPT option, stiffness is updated at every K S T E P ~successful
~
bisection. The ADAPT method allows linear transient analysis, but AUTO or TSTEP *ill
abort the run if the model does not have any data representing nonlinearity. The stiffness
matrix is always updated for a new subcase or restart, irrespective of the option selected.
 The number of iterations for a time step is limited t o MAXITER. If the solution does not
converge in MAXITER iterations, the process is treated as a divergent process, i.e., either
a bisection or stiffness matrix update takes place based on the value of MAXBIS. The sign
of MAXITER provides a control over the ultimate recourse (reiteration) in case of failure in
convergence or bisection. If the MAXITER is negative, the analysis is terminated when the
divergence condition is encountered twice during the same time step or the solution diverges
for five consecutive time steps. If MAXITER is positive, the program computes the best
attainable solution and continues the analysis on second divergence.

The convergence test is controlled by convergence test flags (U for displacement error test,
P for load equilibrium error test, W for work error test) and the error tolerances (EPSU, EPSP
and EPSW) which define the convergence criteria. All requested criteria (combination of U,
P, and/or W ) are satisfied upon convergence. It should be noted that at least two iterations
are necessary to check the displacement convergence criterion.

The MAXDIV field provides control over diverging solutions. Depending on the rate of
divergence, the number of diverging solutions (NDIV) is incremented by 1 or 2. The solution
is assumed to be divergent when NDIV reaches MAXDIV during the iteration. If the bisection
option is used with the ADAPT method, the time step is bisected upon divergence. Otherwise,
the solution for the time step is repeated with a new stiffness based on the converged state at
the beginning of the time step. If NDIV reaches MAXDIV twice within the same time step,
the analysis is terminated with a fatal message.

The BFGS quasi-Newton updates and the line search process work in the way as in static
analysis except for the default setting. The MAXQN field defines the maximum number of
quasi-Newton vectors t o be saved on the database and the MAXLS defines the number of
line searches allowed per iteration. Non-zero values of MAXQN and MAXLS activate the
quasi-Newton update and the line search process, respectively.

The FSTRESS field defines a fraction of the effective stress (5) which is used to limit the
subincrement size in the material routine. The number of subincrements in the material rou-
tines is determined such that the subincrement size is approximately FSTRESS * a (equivalent
stress). FSTRESS is also used to establish a tolerance for error correction in the elasto-plastic
material, i.e.,
Error in yield function < FSTRESS * a
If the limit is exceeded at the converging state, the program will exit with a fatal error message.
Otherwise, the stress state is adjusted to the current yield surface.

MAXBIS is the maximum number of bisections allowed for each time step (-9 5 MAXBIS
5 9). The bisection process is activated when divergence occurs and MAXBIS # 0. The
number of bisections for a time increment is limited t o (MAXBISI. If MAXBIS is positive and
the solution does not converge after MAXBIS bisections, the best solution is computed and
the analysis is continued t o the next time step. If MAXBIS is negative and the solution does
not converge in JMAXBISI bisection, the analysis is terminated.
 The parameter ADJUST allows the user t o control the automatic time stepping in the
ADAPT option. A value of zero for ADJUST turns off the automatic adjustment completely.
If ADJUST is positive, the time increment is continually adjusted for the first few steps until
a good value of At is obtained. After this initial adjustment, the time increment is adjusted
every ADJUST time steps only. A value of ADJUST an order greater than NDT will turn
off adjustements after the initial adjustement. Since the automatic time step adjustment is
based on the mode of response and not on the loading pattern, it may be necessary to limit the
adjustable step size when the period of the forcing function is much shorter than the period
of dominant response frequency of the structure. It is user's responsibility to ensure that
the loading history is properly traced with ADJUST option. The ADJUST option should be
suppressed for the duration of short pulse loading. If unsure, start with DT which is much
smaller than the pulse duration in order to properly represent the loading pattern.

MSTEP defines the desired number of time steps to obtain the dominant period response
accurately (10 _< Integer 5 200): RB defines bounds for maintaining the same time step for
the stepping function in the automatic time step adjustment method (0.1 <
Real 5 1.0).
Parameters MSTEP and RB are used to adjust the time increment during the analysis in the
ADAPT option. The adjustment is based on the number of time steps desired to capture the
dominant frequency response accurately. The time increment is adjusted as follows:

where
1
= MSTEP
with
f = 0.25 for r < 0.5 * RB
f = 0.5 for 0.5*RB 5 r < RB
f = 1.0 for RB r < 2<
f = 2.0 for 2. 5 r < 3./RB
f = 4.0 for >
r 3./RB
The recommmended value of MSTEP for nearly linear problems is 20. A larger value (e.g., 40)
is required for highly nonlinear problems. In the default option, the program automatically
computes the value of MSTEP based on the changes in the stiffness.

The MAXR field defines the maximum ratio for the adjusted incremental time relative t o
<
DT allowed for time step adjustment (1.0 5 Real 32.0). MAXR is used to define the upper
and lower bounds for adjusted time step size, i.e.,
DT DT
(2MAXBIS ' < At 5 MAXR * D T
s)
The UTOL field defines the tolerance on displacement increment below which there is no
time step adjustment (0.001 < Real 5 1.0). UTOL is used t o filter undesirable time step
adjustment, i.e., no time step adjustment is performed if

The RTOLB field defines the maximum value of incremental rotation (in degrees) allowed
per iteration t o activate bisection (Real > 2.0). The bisection is activated if the incremental
rotation for any degree-of-freedom (AB,, AO, or A0,) exceeds the value specified for RTOLB.
This bisection strategy based on the incremental rotation is controlled by MAXBIS field.

9.1.3 Iteration Related Output Data

During the incremental and iterative computation, the process information consisting of iter-
ation data is printed at the end of each iteration or time step. They are printed under the
following heading:
TIME: Cumulative time for the duration of the analysis
ITER: Iteration count for each time step
DISP: Relative error in terms of displacements defined as

for ADAPT method

for AUTO or TSTEP

where urnax= max(llulll, II~211,....., IIunll)

LOAD: Relative error in terms of load vectors defined as

for ADAPT method

for AUTO or TSTEP

where F and P are three-point average values for internal and external forces, respec-
tively.

WORK: Relative error in terms of work defined as

LAMBDA(1): Rate of Convergence in iteration where

LAMBDA(T): Ratio of the load error for two consecutive time steps computed only for
AUTO or TSTEP method Pn

LAMBDA-BAR: Average of LAMBDA(T) over the last 3 steps, computed only for AUTO
or TSTEP method

DLMAG: Absolute norm of the load vector, I( R 11. The absolute convergence is defined using
DLMAG by 1) R I(< 10-12.

FACTOR: Final value of the line search parameter

E-FIRST: Divergence rate, error before line search
E-FINAL: Error at the end of line search
NQNV: Number of quasi-Newton vectors appended
NLS: Number of line searches performed during the iteration
ITR DIV: Number of occurrences of divergence detected during the adaptive iteration by
the iteration module NLTRD2

MAT DIV: Number of occurrences of bisection conditions in the material routine (excessive
stress increment) or in the rotation angle (excessive rotation) during the iteration using
the ADAPT method

NO. BIS: Number of bisections executed for the current time interval
ADJUST: Ratio of time step adjustment relative to DT within a subcase
9.2 RESTARTS

Since SOL 66 (or 106) and SOL 99 (or 129) share the same database storage for nonlinear
tables and matrices, the restart system for transient analysis can use either a previous static
or transient nonlinear analysis as its initial conditions.

9.2.1 Restarting from SOL 66 into SOL 99

The options for a restart from SOL 66 into SOL 99 are static to static and static t o dynamic
analysis. For a restart from a previous static analysis, only the first subcase is affected. Simply
provide a database created in SOL 66 and specify the parameter

. PARAM, SLOOPID, N

where N is the printed value of LOOPID for the desired static solution. The initial transient
load should be identical t o static loads at the restart state. Constraint sets, direct input
matrices, mass, and damping may be changed.

It is noted that SOL 99 destroyed (until Version 66) the database for any subsequent static
analysis once a SOL 66 database is used for restart. It is recommended for the user t o restart
from a copy of the input SOL 66 database. This deficiency is corrected in Version 67.

9.2.2 Restarting within SOL 99

Restarting within SOL 99 allows static to static, static t o dynamic, and dynamic t o dynamic
analysis. Restarts from a previous nonlinear transient execution are available for a number
of cases. If the same model is t o be reexecuted, only the residual superelement needs to be
reassembled (SEMA, SELA = 0). If the final results from the previous transient run are to be
used as the initial conditions at t=O, add N dummy SUBCASE commands t o start the residual
Case Control execution and STIME = 0.

The normal restart for a transient run is t o be continued from the last step of a previous
subcase with different loads and/or TSTEPNL data. For the normal restart provide the
following parameters:
LOOPID = N : Start from the N-th subcase
STIME = t : Start from time t
Note that constraint sets should not be changed to avoid incompatible matrix sizes. The values
of LOOPID and STIME, which are printed with the iteration information for each subcase, can
be directly read from the printout of the previous run. If the adaptive time stepping algorithm
is not activated, the value o f t may also be calculated by the following equation:

t= NDTi r DT;;

where NDT; and DT; are the number of time steps and the time increment of the i-th subcase,
respectively.

If a SOL 99 run is terminated abnormally in the middle of a subcase, it may or may not
be restartable depending upon the cause of the abnormal exit. If the job is stopped due to a
diverging solution, it can be restarted either from the end of a previous subcase or from the last
saved solution step. The restart procedure for the former is identical t o that for the normal
restart as described in the preceding paragraph. The latter case also requires parameters
LOOPID and STIME, however, the input value for STIME differs depending on the value of
METHOD specified on the TSTEPNL entry. If METHOD = AUTO or TSTEP (the NLTRD
module), STIME is the time corresponding t o the last output step which may be calcualted
based on the output skip factor (i.e., the NO on the TSTEPNL entry). If METHOD = ADAPT
(NLTRD2 module), which is a new feature implemented in Version 66, the last converged
solution is treated as an output step and is always saved for the restart so that STIME can be
the time of the last converged step. The values of STIME and LOOPID can also be found in
the printout, if the ADAPT method is used. Once STIME and LOOPID are known, determine
the number of remaining time steps in the subcase and create a new TSTEPNL entry for the
remaining time. Insert a new subcase which references the new TSTEPNL entry prior t o the
remaining subcases in the Case Control section.

A solution may be terminated in the middle of a subcase due t o insufficient CPU time: (1)
the CPU time specified in the Executive Control section is insufficient so the run is forced t o
exit by the program, or (2) the CPU time exceeds the limit specified in the computer system
which leads to a sudden job abortion by the system. In the first case, MSCINASTRAN is
able t o detect the specified CPU time in the Executive Control section and automatically
activate the wrapping-up procedure before time expiration. When completed, the solution can
be restarted from the termination point as in the solution diverging case. In the second case
the solution can only be restarted from the end of a subcase.

Restarts may also be performed solely for data recovery by providing the following param-
eters:
SDATA = -1 : Recover data without running the solution module
LOOPID = N : from the 1st through the N-th subcases.
Note that solution sets DISP, VELO, ACCE, OLOAD, SPCF (printout and plotting) and
NLLOAD (plotting only) are recoverable while STRESS, SDISP, SVELO and SACCE sets are
not.
9.2.3 Example

A simply supported elastic beam is subjected t o a single cycle of a sinusoidal forcing function
at the quarter span as illustrated in Fig. 9.2.1. The beam is modeled using 20 linear BEAM
elements. A stopper (snubber) is present underneath the center of the beam with a clearance
of 0.02 inches. This gap is simulated by a displacement dependent force which is active only
when the beam is in contact with the snubber. The problem is linear except for the effect of
the stopper. The input data for this problem is given in Table 9.2.1.

One subcase with 100 time steps and an initial time step size of 0.1 msec was analyzed in
the cold start run. Since the adaptive time stepping algorithm was activated, the job ran t o
completion at 9.7 msec instead of the specified time span, 100 x 0.1 = 10 msec. A new subcase
with 600 time steps was added in the first restart run t o extend the time span for another 30
msec. The adaptive time stepping method was turned off and the execution time was specified
to be 0.8 min. As expected, the run was aborted a t 36 msec due t o insufficient time. In the
second restart run, the execution time was changed t o 5.0 min. and another subcase with 164
time steps and a time increment of 0.1 msec was added t o extend the analysis time span up t o
40 msec. The third restart was simply t o recover the DISP, VELO, ACCE, OLOAD, NLLOAD
and SPC output. The response at the loading point of the beam is shown in Fig. 9.2.2.

17-it--------50 in 50 in ---------+/ p = 0.3 lb/in3

20 Beam Element

0.02 in.
(~10010)

Forcing Function NOLINl Representing GAP

Figure9.2.1.SimplySupportedBeamWithaRestrainedMotion .
 VERT.
DISPL.
AT
POINT A

Figure 9.2.2. Simply Supported Beam: Loading Point Response

 Table 9.2.1. Input Data Listing for Simply Supported Beam

$RESTART $ ***added for the restart runs

ID TNOLIN, V66 $ SSH 7/10/91
TIME 5 $
SOL 99 $ NONLINEAR TRANSIENT ANALYSIS
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
CEND
TITLE = DYNAMICS HANDBOOK DEMONSTRATION PROBLEM
SUBTITLE = DIRECT TRANSIENT RESPONSE, NONLINEAR FORCE
LABEL = NOLIN IN SOL 99
SEALL = ALL
ECHO = UNSORT
SET I = 10005
SET 2 = 10010
SET 3 = 10000
SPC = 1002
$-----------------------
$PARAM,LOOPID,I *** used for the 1st restart
$PARAM,STIME,O.O093
$-----------------------
$PARAM,LOOPID,2 *** used for the 2nd restart
$PARAM,STIME,0.0236
$-----------------------
$ DISPL=2
$ VELO=2 *** used for the 3rd restart
$ ACCEL=2
$ OLOAD=I
$ NLLOAD=2
$ SPCF=3
$PARAM,LOOPID,3
$PARAM,SDATA,-I
$-----------------------
SUBCASE 1
DLOAD = 30
TSTEPNL = 10
NONLINEAR = 13 $ SELECT NONLINEAR FORCE
SUBCASE 2
DLOAD = 30
TSTEPNL = 20
NONLINEAR = 13 $ SELECT NONLINEAR FORCE
SUBCASE 3
DLOAD = 30
TSTEPNL = 30
NONLINEAR = 13 $ SELECT NONLINEAR FORCE
BEGIN BULK
$ MODELING INFORMATION FOR BEAM ONLY
CBAR 101 100 I0000 I0001 0. 0. 1.
CBAR 102 100 I0001 10002 0. 0. I.
CBAR 103 100 10002 10003 0.
CBAR 104 100 10003 10004 0.
CBAR 105 100 10004 10005 0.
CBAR 106 100 10005 10006 0.
CBAR 107 100 10006 10007 0.
CBAR 108 100 10007 10008 0.
CBAR 109 100 10008 I0009 0.
CBAR 110 100 10009 I0010 0.
CBAR Ill 100 10010 10011 0.
CBAR 112 100 I0011 10012 0.
CBAR 113 100 10012 10013 0.
CBAR 114 100 10013 10014 0.
CBAR 115 100 10014 10015 0.
CBAR 116 100 10015 10016 0.
CBAR 117 100 10016 10017 0.
CBAR 118 100 10017 10018 0.
CBAR 119 100 10018 I0019 0.
CBAR 120 100 10019 10020 0.
CONM2 12 I0010 .I
GRID 10 50. -1.
GRID 10000 0. 0. 0.
GRID 10001 5. 0. 0.
GRID 10002 10. 0. 0.
GRID 10003 15. 0. 0.
GRID 10004 20. 0. 0.
GRID 10005 25. 0. 0.
GRID 10006 30. 0. 0.
GRID 10007 35. 0. 0.
GRID 10008 40. 0. 0.
GRID 10009 45. 0. 0.
GRID 10010 50. 0. 0.
GRID 10011 55. 0. 0.
GRID 10012 60. 0. 0.
GRID 10013 65. 0. 0.
GRID 10014 70. 0. 0.
GRID 10015 75. 0. 0.
GRID 10016 80. 0. 0.
GRID 10017 85. 0. 0.
GRID 10018 90. 0. 0.
GRID 10019 95. 0. 0.
GRID 10020 100. 0. 0.
MAT1 1000 3.+7 .3 .3
PARAM GRDPNT 10010
PARAM WTMASS .002588
PBAR 100 1000 .31416 .I5708
+ P B I I. 0.
SPC 1002 10 123456
SPC 1002 10020 3 10000
$
$ MODELING INFORMATION FOR CENTER SPRING
CROD 10 10 10 10010
MAT1 10 10. 0.
PROD 10 10 I.
MATS1 10 PLASTIC 0. I I 3.+8
$ LOADING AND SOLUTION INFORMATION
TLOAD2 30 33 0. .011451 87.33 -90.
DAREA 33 10005 3 47.2
TSTEPNL 10 100 .0001 1 ADAPT
TSTEPNL 20 600 .00005 1 ADAPT
+TS21
+TS22 0
TSTEPNL 30 164 .00010 I ADAPT
+TS3I
+TS32 0
$
$ MODELING INFORMATION FOR NONLINEAR SPRING
NOLINI 13 I0010 3 I. 10010 3 13
TABLED1 13
+TAB -2.5-2 4.95 -2.0-2 0. 0. 0. ENDT
$
ENDDATA
9.3 DYNAMIC EFFECTS: MASS AND DAMPING

9.3.1 Mass Modeling

Mass properties of the finite element model may be specified in several different ways, in-
cluding mass elements (CONMI, CONM2, CMASSi), parameters (WTMASS, COUPMASS,
GRDPNT) and fields in material or property Bulk Data entries. They are described below:
The density (mass per unit volume) is specified in the RHO field of a MATi Bulk Data
entry and is used t o automatically compute the mass for all structural elements which
reference the MATi entry.

Nonstructural mass distributed over elements (mass per unit length for line elements or
mass per unit area for surface elements) is specified in the NSM field of property Bulk
Data entries. Applications of this feature are floor loads, electrical cables or thermal
insulating materials.

The mass elements have inertia properties directly specified by the user.

- The CONMl Bulk Data entry provides a general description of a concentrated mass
at a grid point of the structural model by defining a 6x6 symmetric matrix of mass
coefficients at the grid point. Applications of CONMl are rare.
- The CONM2 Bulk Data entry provides a more convenient definition of a concen-
trated mass at a grid point. The inputs t o CONM2 are the mass, the offset of the
center of mass from the grid point, and the moments and products of inertia about
the center of mass. As an option, the center of mass may be measured from the
origin of the basic coordinate system rather than as an offset from the grid point.
- The CMASSi Bulk Data entries provide mass coupling between any two degrees of
freedom. The form of the relationship is

where fl and f2 are the inertia forces acting at degrees of freedom 1 and 2, re-
spectively, and M is the mass coefficient (specified on the CMASSi entry or on the
PMASS entry if i = 2 or 4). In most applications, the second degree of freedom is
not specified. If this is the case, the inertia force fi = - M u l is generated, or in
other words, M is added t o the mass matrix in the diagonal position corresponding
t o ul. An important application of the CMASSi entry occurs in the recommended
method for specifying enforced motion a t grid points. Another application is for
modeling an inertia term in relative coordinates. If the ul and u2 terms are both
defined on the CMASSi entry, then inertia forces will result only when u l is not
equal t o u2.
 a The three parameters which relate to the specification of mass properties are

- PARAM, WTMASS, V1 specifies a factor by which t o multiply the mass input data
t o obtain dimensionally correct mass. For example, if the in-lb-sec system is used
and the mass input is specified in weight pounds, then V1 = & = 0.00259.
- PARAM, COUPMASS, 1 requests that the coupled mass option be used rather
than the lumped mass option (default) for all elements which support the coupled
mass option. In the lumped mass option, the distributed mass of an element is
simply divided up and the parts are located at the connected grid points. In the
coupled mass option, the distributed mass is replaced by a nondiagonal mass matrix
which couples the connected grid points. The coupled mass option is generally
more accurate, but uses more computer time. The lumped mass option is therefore
prefered for computational speed.
- PARAM, GRDPNT, V1 causes the Grid Point Weight Generator to be executed.
The value of V1 identifies a grid point at which the rigid body mass properties of
the structure will be computed. This capability can be used as a check on mass and
geometric input data.

9.3.2 Mass Matrix

For transient response analysis, the mass matrix is formulated as

where M is the mass matrix, Ml contains mass terms from the following sources
A 6x6 matrix of mass coefficients at a grid point defined on a CONMl entry

a A concentrated mass element defined on a CONM2 entry

a Structural mass for all elements, except plate elements without membrane stiffness, using
the mass density on the MATi entry

a Nonstructural mass for all elements specifying a value on the property entry

a Scalar masses defined on CMASSi entries

and M2 corresponds t o the direct input matrices which are generated by transfer functions
specified by T F Bulk Data entries or supplied directly via DMIG or DMIAX Bulk Data entries.

Unlike the stiffness matrix, the mass matrix is generated only once and is not updated
during the analysis. This situation should not cause much concern. The mass matrix for the
deformed structure is the same for the undeformed structure in material nonlinear analysis.
During large displacement analysis, precautions can be taken so that the mass matrix for the
deformed structure is very similar t o the one for the undeformed structure.
 The mass matrix is defined in the global coordinate system using the initial undeformed
geometry. During large displacement the structure deforms and rotates relative t o the global
coordinate system. This causes the element mass distribution and rotational inertias t o change.
The element mass distribution due to element deformation is usually minimal and will in most
cases cause little error. The changes in the rotational inertia due t o element rotation and
elongation are more severe and may cause large errors. The following guidelines will help
avoid most problems when using the large displacement option.
1. Do not use the coupled mass option (PARAM,COUPMASS,l). The coupled mass option
generates rotational inertia terms for many of the elements. These terms are defined in
the global coordinate system. As the elements rotate, the rotational inertia terms defined
in the global coordinate system change. This change, however, is not reflected in the
mass matrix since it is formulated only once and is not updated. By using the lumped
mass option (default), only translational terms are generated. Translational terms are
unaffected by element distortions.

2. If concentrated masses are used, then specify the same mass on all three translational
degrees of freedom. This is required t o keep the inertial forces consistent with the elastic
forces of the structure.
Consider the following example: a cantilevered beam with a mass at the free end defined

only in the y-direction. Initially the elastic bending and shear forces are in-line with the
inertial forces of the mass. As the beam is deformed by an end load the elastic forces
follow the deformation (stiffness matrix is updated). The mass is only defined in the
y-direction and will resist only the y component of the elastic force. The x component of
the elastic force will not be resisted. In the extreme case the elastic force is unresisted.

3. As with element rotational inertias, the same situation exists for concentrated rota-
tional inertias. Rotational inertias about the three directions of the global coordinate
are usually different. As the actual structure rotates at the concentrated mass point,
the concentrated mass will also rotate with respect t o the global coordinate system. As
it rotates, the inertias about the directions of the global system change. This change
in rotational inertia is not taken into account by the mass matrix. For this reason,
concentrated rotational inertia terms should not be used.

4. If rotational inertia terms are necessary then a rigid structure with concentrated trans-
lational masses should be used. This rigid structure should emulate the mass properties
of the concentrated mass. For example, a user wishes t o simulate a rigid mass with the
following properties:

Total mass = 10 kg
I,, = 10 kg . cm2
Ivy= 20 kg - cm2
Izz = 30 kg . cm2
 where the principal axes of the mass align with the global coordinate system. The
rotational inertia will be modeled by lumped masses (two mass points of equal magnitude
in each direction: m,, my and m,) at the ends of rigid bars (stiff CBEAMs). The bars
will extend an equal distance in the positive and negative directions from the grid along
the global coordinate system axes. The bar lengths and concentrated masses must satisfy
the following requirements:
2m, + 2my + 2m, = total mass (10 kg)
2myli + 2m,la = I,, (10 kg . cm2)
2m,l2 + 2m,1,2 = I,, (20 kg . cm2)
2m,12 + 2m,li = I,, (30 kg . cm2)

where l,, 1, and 1, are lengths of the rigid bars in the x, y and z-axes, respectively. There
are more unknowns than equations. In this example m, = 1 kg, m, = 1 kg, and m, =
3 kg were selected. Solving for I,, I,, and I , results in I, = f i cm, 1, = 4 cm, and
I, = 0 cm. This rigid structure will simulate the translational and rotational properties
of the concentrated mass as the structure rotates because the motion of the rigid bars is
taken into account during the stiffness matrix updating procedure.

9.3.3 Damping Modeling

Damping represents the energy dissipation observed in the structure and results from many
sources including
Viscous effects (dashpot, shock absorber)

Internal friction (characteristic of material type, i.e., hysterisis)

External friction (slippage in structural joints)

Structural nonlinearities (plasticity)

In MSC/NASTRAN, damping is divided into two types: viscous and structural. The two
types of damping perform different functions. Viscous damping is used t o simulate actual
structural damping elements whereas structural damping is used t o simulate the inherent
damping properties of materials. The two damping types can be combined t o model the
damping characteristics of a structure.

Viscous damping is input using discrete damping elements (CVISC and CDAMPi). Be-
cause stable damping situations are difficult t o model using discrete elements, the CVISC
and CDAMPi elements are not used t o add stability t o a structure during transient response
analysis. They are used mainly t o model actual damping components in a structure, such as
hydraulic dampers and viscous interface material. When using the discrete elements in this
situation, the major relative motion is assumed t o be between the coupled degrees of freedom
or along the initial axis connecting the grid points.
 Structural damping is a global damping proportional to the stiffness. The matrix is formed
from the input in the GE field of MATi or via PARAM G, PARAM W3, and PARAM W4.
Structural damping is included in a model in an attempt to simulate the intrinsic material
damping. The user should choose a realistic value (0.5% t o 4%), dependent upon the material
in the model. A light-weight metal structure may only have a 0.5% damping value, whereas a
composite material may have a 4% damping value. A change in the damping value may have
a significant effect on the calculation of the response. A minimum value of 0.03 x W4 x At is
recommended for stability, regardless of the material.

The viscous damping force is proportional to velocity, i.e.,

and the structural force is proportional t o displacement, i.e.,

Viscous and structural damping are related by

with

which results in an equivalence a t resonance with

Fig. 9.3.1 compares the viscous and structural damping forces as a function of frequency.
Damping due to internal material friction and damping due t o bolted or riveted joints behave
more like structural damping than viscous damping.
Damping
Force

I
Wn W

Figure 9.3.1 Comparison of Viscous and Structural Damping

Except for the overall structural damping parameter ( G ) ,all of the linear damping options
may be used in nonlinear transient analysis. The damping input options in the finite element
model are described below:
The CVISC Bulk Data entry defines a viscous damping element that may be connected
between any two grid points. It acts like a rod element with viscous extensional and
torsional properties rather than elastic properties.

The CDAMPi Bulk Data entries define a scalar damping element using a 2x2 viscous
damping matrix between any two degrees of freedom. The form of the relationship is

where fi and f2 are the damping forces acting at degrees of freedom 1 and 2, respectively,
and B is the damping coefficient (specified on the CDAMPi entry or on the PDAMP
entry if i = 2 or 4). Two degrees of freedom can be in any direction and may not act like
a rod element. If the second degree of freedom is not specified, then fi = B u l , i.e., B is
added t o the damping matrix in the diagonal position corresponding to ul.

Structural damping is specified in the GE field of a MATi Bulk Data entry. The effect
of GE is t o multiply the stiffness matrix for an element which references the MATi entry
+
by the scalar factor (1 ig) where g is the value of the GE field and i is the operator for
imaginary number.
 a PARAM, G specifies an overall structural damping factor (default = 0.0) for the linear
part of the stiffness matrix, in nonlinear transient analysis.
a PARAM, W3 selects the frequency (default = 0.0) to convert the overall structural
' damping on the linear elements to viscous damping.
a PARAM, W4 selects the frequency (default = 0.0) to convert the element structural
damping (including nonlinear elements) to viscous damping.
PARAM, NDAMP provides numerical damping in the ADAPT method of SOL 99 for
numerical stability (implemented in Version 67). The numerical damping increases as
the value of NDAMP increases, with zero being no numerical damping (default is 0.025).

9.3.4 Damping Matrix

For transient response analysis, the damping matrix is formulated as

where
B' : contains terms from viscous damping elements (CVISC, CDAMP)

B~ : direct input matrices generated by transfer functions (TF entries) or supplied by direct
matrix input (DMIG, DMIAX)
G : Overall structural damping coefficient, specified by PARAM G
W3 : frequency of interest, specified by PARAM W3
KI : global linear stiffness matrix

G , : structural damping constant, specified on MATi

W.4 : frequency of interest, specified by PARAM W4
K, : initial element stiffness matrix, generated for t=O
Like the mass matrix, the viscous damping matrix is generated only once and is not UP-
dated. For material nonlinear analysis, for instance, the viscous damping matrix for the de-
formed structure remains constant as for the undeformed structure based on the elastic mate-
rial. For the geometric nonlinear analysis, the viscous damping matrix will be inaccurate for
the deformed structure with large rotation because the direction has been changed and the
damping matrix is not rotated. In both cases caution must be exercised when using discrete
damping elements in transient analysis.
9.4 TRANSIENT LOADS AND INITIAL CONDITIONS

The methods employed t o define loads in nonlinear transient analysis are similar t o those used
in the linear solutions. A single degree of freedom or a set of GRID points may be loaded
with force pattern that varies with time. Functions may be tabular such as an earthquake or a
booster liftoff, or they may be simple analytic functions such as a sine wave. Simple static load
sets generated in both upstream superelements and the default residual may be used to create
the dynamic loads. They may be scaled and combined with other loads t o simulate complex
loading problems.

For special problems involving simple scalar nonlinearities, the traditional nonlinear loads
(NOLINi option) may be used in both types of transient solutions. These are useful for
simulating nonlinear damping mechanisms and rotational coupling not provided by the GAP
elements.

9.4.1 Transient Loads

Transient loads define the loadings as functions of time and the location. They can be a load
applied at a particular degree of freedom, pressure over the surface area, or the body force
simulating an acceleration. The time history is provided by TLOADi Bulk Data and the static
loads are converted t o dynamic loads using LSEQ option. The transient dynamic loads are
selected in the Case Control section, controlled by the following commands:
DLOAD: Required t o select TLOADi or DLOAD Bulk Data loads. Each stage of the transient
solution may be defined with a separate SUBCASE, each with its own DLOAD request.
The DLOAD data is restricted to act only in the time defined in its subcase.
LOADSET: Required for LSEQ processing. Must be placed above the solution subcases
and/or in the Superelement Subcases.
NOLINEAR: Used t o select a family of NOLINi functions. May not be changed between
solution subcases.
NLLOAD: Required for NOLINi output request.
Input Bulk Data associated with the transient loads in SOL 99 (or 129) are briefly described
below:
TLOADl, TLOAD2: Required data defining the coupling between Load factors (DAREA
Sets) and functions of time (TABLED, DELAY, or coefficients). Specifies the time func-
tions, Fi(t),for dynamic load i. TLOADl defines the load history in TABLED for-
mat while TLOAD2 defines the load history by an analytic expression. Note that each
TLOADi input must have a unique identification number.
DAREA: Optional Bulk Data defining simple load scale factors, A i j , for each degree of free-
dom, u j requested in set i. The DAREA factors may also be defined with static load
 data with the LSEQ Bulk Data. The name was chosen by a dynamicist who viewed the
problem in terms of time-dependent pressures and location-dependent "areas".

LSEQ: Generates transient load history for static loads and controls the assembly of static
load vectors t o be used in dynamic analysis. Each set is equivalent to a Subcase in Case
Control. It will cause a static load vector A; t o be generated and will label it with the
DAREA identification. This is useful for modeling problems with distributed transient
loading such as GRAV loads or pressurized areas.

DLOAD: Combines different TLOADi functions into a single set and scales each by a factor,
C k i . Performs the same function as the static LOAD Bulk Data input. Unlike statics,
this is the only method to combine loads in dynamic analysis.

TABLEDi, i=1,2,3,4 : Used with TLOADl input to specify general time varying tabular
functions. Since extrapolation is used beyond the range of data the user is cautioned to
be careful with the end points.

DELAY: Specifies the delay time for applying the forcing function defined in TLOADi t o
each GRID point. The Bulk Data DELAY is associated with the degree of freedom and
not the forcing function in each TLOADi. This option is useful for defining loads that
travel across a structure such as an oblique wave along a dam or a bump in the road
exciting the wheels of a traveling vehicle. This saves the analyst the effort of defining
multiple TLOADi inputs for simple time lags.

NOLINi, i=1,2,3,4 : Define specific nonlinear scalar loads as functions of velocity and/or
displacement. Basic analog nonlinear transfer functions are defined. However, these
functions do not generate tangent matrix terms and the Newton iteration method may
experience difficulties in convergence. Line search and BFGS solution methods are rec-
ommended for these models.

NOLINl : Nonlinear transient load as a tabular function

NOLIN2 : Nonlinear transient load as products of two variables
NOLIN3 : Nonlinear transient load as a positive variable raised t o a power
NOLIN4 : Nonlinear transient load as a negative variable raised t o a power
A diagram of input data linkages is shown in Fig. 9.4.1. The general form of the transient
load, Pkj, for each degree of freedom, uj , is:

where the factors are defined by various Bulk Data inputs as shown in Fig. 9.4.2.
Case Control Section

1 LOADSET 1
QDLOAD
I
I
i"""
Bulk Data Section

Figure 9.4.1 Dynamic Loads for Transient Analysis

 DLOAD
( x S >

TLOADl TLOADl TLOAD2 .

: TABLED1 TABLED2 ( ~ f ( t ) ):
: ( X f(t)) ( X f(t))
I :
-
I
-
4 _ _ - _ _ - _ _ _ _ _ _ _ - -I-- I
DAREA DAREA DAREA .
I : : I
I
I
............................

FORCE1 LOAD
. . . . .

PLOAD2
l . . . . .

I
1
1 MOMENT PLOAD4 I
I I
I I
SLT
L - - - - - - - - - _ - - - - - - - - - I

Note : Each load is scaled by [S x Si x f ( t ) ]at each time step in dynamic loads.

Figure 9.4.2 Example of Combining Loads

9.4.2 Enforced Motion

Excitation of the base of a structure with a time-dependent motion (displacement, velocity,

or acceleration) is provided by MSCJNASTRAN indirectly via the "large mass method". In
effect the user places a large mass element on the degrees of freedom representing the base and
then provides a large force proportional to the desired accelerations. If a very large mass (Mb)
is connected t o a degree of freedom on an unconstrained structure, the load of magnitude Pb
will produce an approximate base acceleration:

The accuracy of this approximation increases as Mb is made larger in comparison to the mass
of the structure. However, numerical difficulties arise if Mb is too large. It is recommended
that the value of Mb be approximately 1000 times the mass of the structure. The factor 1000
is a safe limit that will produce three digits of numerical accuracy.

When the large mass method is used, the function iib is normally specified as input loads,
and the value of Mb is input on DAREA or DLOAD Bulk Data entry, whichever is more
convenient. CMASSi or CONM2 entry should be used to input the large mass. CMASSi entries
allow different masses for different directions. The TLOADi inputs allow the specification of
displacements and velocities, which in turn are converted to acceleration by the program t o
compute the inertia force as depicted in Fig. 9.4.3.

A large stiffness method could be used in liu of the large mass approach. In this case, the
applied force would be computed by
Pb = Kbub
where Kb is the stiff stiffness and ub is the enforced displacement. The large stiffness method
has the advantage in case of enforced displacement, because it avoids the roundoff error of
numerical differentiation which occurs while computing accelerations from displacements in
the large mass approach.

For Nonlinear problems, the "large mass or large stiffness approach" requires special cau-
tion in choosing convergence error tolerances. The actual errors will be normalized by terms
proportional t o the large base forces. It may be necessary t o reduce the error tolerances on
the TSTEPNL inputs depending on the size of the large base mass.
 2nd derivative 1st derivative

( EQUIVALENT ACCELERATION 1

I LARGE MASS 1
4

0 EQUIVALENT LOAD

Figure 9.4.3 Options in TLOADi (Types 1, 2 and 3)

9.4.3 Initial Conditions

Initial velocities or displacements can be imposed by using the Bulk Data TIC which are se-
lected by a Case Control command IC. Alternatively, three methods are available for providing
a set of initial displacements and/or velocities for Nonlinear Transient Analysis:
1. TIC Bulk Data are used to specify point-by-point initial displacements and velocities.
As in linear analysis, the system also calculates internal equivalent loads to provide
equilibrium loads at the start. If uo and vo are the initial displacement and velocity
vectors, respectively, the equivalent loads are:

where [K] and [C] are the linear stiffness and damping matrices. Note that if the initial
state is nonlinear, the initial loads on the structure will not be in equilibrium. This will
 require iterations and may cause difficulty for the convergence. To be safe, use the TIC
data only for initial velocities and on linear sections of the structure.

2. Restarts from static analysis (SOL 66 or SOL 99 with TSTATIC) are highly recom-
mended for initially displaced and loaded structures. The initial static equilibrium will
be preserved if the initially supplied transient loads are identical t o the static loads. The
static solution (in SOL 66 or 106) may use enforced displacements (SPCD) and different
boundaries t o obtain the desired initial shape.

3. An Initial LLdummy"Subcase may be used to generate complicated initial conditions

without a restart. An extra subcase is added in front of the normal transient subcases
with loading functions designed to produce the desired initial condition by matching the
momentum. For instance, an initial velocity may be generated by a large impulsive load
applied for 5-10 small time steps. An initial deformation may be produced by a subcase
with large time steps (larger than the lowest mode's period) which eliminate the transient
behavior and produces a static solution if it converges. This method requires that the
other loads be shifted in time and will also cause a shift in the output response.
The user is also cautioned against supplying large initial loads in the structure that may
produce initial nonlinearities. The job may not be able t o complete the first time step. Solu-
tions should be obtained as follows:
a Restart from a Nonlinear static solution using the initial load.

a Start with a smaller load and increase it over a small time interval.

Reduce the size of the time steps by a few orders of magnitude for the first subcase.
9.5 Diagnostic Outputs

Diagnostic output is available in SOL 99 (or 129) if DIAG 50 or 51 is specified in the Executive
Control section. When DIAG 50 is requested, only the data marked with a plus ( f ) are printed.
When DIAG 51 is selected, all the data are printed at each iteration, allowing the user to follow
the nonlinear iteration calculations step by step. Because of the volume of output produced,
the use of DIAG 51 is recommended only for small test problems. DIAG 51 is intended for
debugging purposes, and users in general are cautioned against its use.

9.5.1 NLTRD Module for AUTO or TSTEP Method

For each entry into NLTRD:

+ Subcase status data
+ TSTEPNL data
Core statistics (ICORE, etc.)
Problem statistics (g-size, etc.)
File control block
Input file status

For each iteration:

Initial energy for line search
Nonlinear internal force: Fg
Sum of nonlinear forces including follower forces: Fd
NOLINi vector: Nd
Displacement vector: ud
Total internal force: Fd
Loading error vector: Rd
+ Iteration summary (Convergence factors, line search data, etc.)

For each quasi-Newton vector set:

Condition number: X2
Quasi-Newton vector: 6
Quasi-Newton vector: y
1
Energy error: z = 7
j' 7 3

For each line search:

Previous line search factor: crk
Previous error: Ek
New line search factor: a k + l
9.5.2 NLTRD2 Module for ADAPT Method

For each entry into NLTRD2:

+ Subcase status data
+ TSTEPNL data
Core statistics (ICORE, etc.)
Problem statistics (g-size, etc.)
File control block
Input file status

For each time step:

NOLINi vector: Nd
External load vector: Pd
Load vector including follower forces and NOLINs: Pd
Constant portion of residual vector: R&
Total internal force: Fd
Initial loading error vector: Rd

For each iteration:

Initial energy for line search
Nonlinear internal force: F'
Displacement vector: ud
Nonlinear internal force: Fd
Total internal force: Fd
NOLINi vector: Nd
Load vector including follower forces and NOLINs: Pd
Loading error vector: Rd
+ Iteration summary (Convergence factors, line search data, etc.)

For each quasi-Newton vector set:

Condition number: X2
Quasi-Newton vector: 6
Quasi-Newton vector: y
1
Energy error: z = -
6T.3

For each line search:

Previous line search factor: crk
Previous error: Ek
New line search factor: a k + l
For each converged time step:
Velocity vector: vd

For each time step adjustment:

Magnitude of the old velocity vector: u ,
Magnitude of the new velocity vector: u,+*
Generalized stiffness: DENOMl = Au: I< A u ,
Generalized mass: DENOM2 = Au: M Au,
Work = Au; AF,
Square of dominant frequency: wi
Stiffness ratio
Number of steps for the period of dominant frequency: MSTEP
Controlling ratio for time step adjustment: T
Chapter 10

SPECIAL APPLICATIONS

10.1 NONLINEAR BUCKLING ANALYSIS

10.1.1 Introduction

A simple approach to nonlinear buckling analysis was introduced in SOL 66 (or 106) by using a
quadratic extrapolation [10.1]. A similar approach to the limit point calculations was published
by Fujikake [10.2]. The distinction is that MSCINASTRAN uses two converged solutions
in the vicinity of the buckling point to form a differential stiffness [AIi], whereas Fujikake
uses a converged point and an estimated data point beyond the buckling point. To find an
instability.point within a small range of nonlinear domain, two methods of idealization can be
contemplated:
1. The tangent stiffness matrix is proportional to the external loads, which implies that the
+
critical load may be linearly interpolated, i.e., PC,= Pn XAP.
2. The tangent stiffness matrix is proportional to the displacement increments, which im-
plies that the critical displacements may be obtained by extrapolating from the current
+
state, i.e., Uc, = Un XAU.
It appears that Fujikake has chosen the first approach. However, he did not show how he
obtains the critical displacements. MSCINASTRAN has chosen the second approach. Since
the tangent matrix is assumed t o change linearly, the internal loads are quadratic function of
displacements. The distinction between the two approaches disappears in the linear bifurcation
problem.

Restart is required for a nonlinear buckling analysis after the non-positive definite stiff-
ness matrix is detected. Since the modified Newton's method is employed with quasi-Newton
updates in the general purpose program, tangential stiffness matrices in the two successive
increments are not immediately available in general. Details of the method are given below.
10.1.2 Formulation of Nonlinear Buckling Analysis

It can be shown that the eigenvalue problem for a nonlinear buckling analysis may be approx-
imated by
[Kn t X AK] ( 4 ) = (0) (10.1.1)
with
AK = Kn - Knvl
where Kn and Knvl are the stiffness matrices evaluated at the known solution points in the
vicinity of the instability.

The critical displacements upon instability may be estimated as

with
{Au) = {un) - {un-I>

Based on the virtual work principle,

where

The critical buckling load may be estimated as

{Per) = {Pn) + a {AP) (10.1.4)

with
{AP} = {Pn) - {Pn-1)

where
cu = { A u ) ~[Kn +
X AK] {AU}
{AUIT {AP)
This procedure was first implemented with DMAP ALTER in Version 64 of MSCINASTRAN
by a simple DMAP (Direct Matrix Abstraction Program) language. This capability has been
available with a control parameter BUCKLE in SOL 66 since Version 65.
10.1.3 Analysis Procedure

Nonlinear buckling analysis capability is provided in SOL 66 (or 106) and activated by a
user-specified parameter BUCKLE with a value of + l . The operational steps are
1. Run SOL 66 for static analysis and create a database for restarts. It is recommended t o
continue the nonlinear static analysis with small steps until a negative determinant of
the stiffness matrix is encountered.

2. Make a restart run as follows:

(a) Select PARAM entries SUBID, LOADINC and LOOPID for restart.
(b) Provide two small loading steps (below buckling). This may be provided by adding
a new subcase, skipping the rest of the increments of the subcase in which the
stiffness matrix becomes singular.
(c) Use "KSTEP=lV in the NLPARM entry for two load steps t o be solved so that the
stiffness matrix is updated for each solution.
(d) Include EIGB Bulk Data entry with a METHOD command in the Case Control
section for the eigenvalue analysis. SINV method is recommended for the eigenvalue
extraction.
(e) Provide mode shape PLOT data if desired.
(f) Include PARAM, BUCKLE, 1.

10.1.4 Verification Problem: Euler Column Buckling

This method should be applicable to linear problems because linear buckling is a special case of
nonlinear buckling. For verification, an Euler buckling problem of a pin-ended column is illus-
trated. The column with dimensions shown in Fig. 10.1.1 is modeled with 6 CBEAM elements
and subjected to a compressive axial load. The input data listing is shown in Table 10.1.1.

The buckling analysis was performed by two independent methods as follows:

1. Using SOL 65 (linear buckling analysis), restarted from SOL 61 (linear static analysis)
database.

2. Using SOL 66 with PARAM, BUCKLE, restarted from SOL 66 database.

Both procedures produced identical solutions with the critical buckling load (PC,)
of 120.38
lbs.
 r buckled shape

where n = l
E = 10.8 x lo6 psi
I = 4.069 x in4
L = 6in
P, = 120.48 1bs

Figure 10.1.1 Euler Buckling for Verification

 Table 10.1.1 Verification Problem Input Data
ID BBUCKL,V65 $ CJS, 7-MAY-85, SHL 5/2/84
SOL 66 $ NONLINEAR STATIC ANALYSIS
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
TIME 5 $ CPU MINUTES
CEND
TITLE = EULER BUCKLING OF BEAM-COLUMN
SUBTITLE = AXIAL FORCE, PIN-ENDED BOUNDARY
ECHO = UNSORT
SEALL = ALL
DISP = ALL
OLOAD = ALL
SPCF = ALL
STRESS = ALL
METHOD = 30
$----------------------*
$PARAM BUCKLE I *
$PARAM SUBID 2 * CHANGES REQUIRED
$PARAM LOADINC I * FOR BUCKLING
$PARAM LOOPID 1 *
$----------------------*
SUBCASE I
LOAD = 10
NLPARM = 10
SUBCASE 2
LOAD = 20
NLPARM = 20
OUTPUT(PLOT)
CSCALE = 1.3
PLOTTER NAST
SET I = ALL EXCEPT PLOTEL
MAXI DEFO .05
AXES Z, X, Y
VIEW 0.,0.,0.
PTITLE = SIDE VIEW
FIND SCALE ORIGIN I SET I
PLOT SET I ORIGIN I LABEL BOTH
PLOT STATIC 0 MAXI DEFO .05 SET I ORIGIN I
PLOT MODAL 0 SET I ORIGIN I
BEGIN BULK
$ PARAMETERS
PARAM LGDISP I
$ SOLUTION CONTROL
NLPARM 10 I AUTO UPW
NLPARM 20 4 AUTO I PW YES
$ EIGENVALUE ANALYSIS
EIGB 30 INV 0. 3. 20 2 2
+EIGB MAX
$ MATERIAL AND GEOMETRIC PROPERTIES
MAT1 1 10.8+6 .3
+MATI 7.8+4 7.8+4 4.5+4
MATS1 1 PLASTIC 1.225+6 1 2 7.8+4
PBEAM 3 I 3.125-2 4.069-5 1.628-4
+PB1 -.0625 -. 125 -.0625 .I25 .0625 .I25 .0625 -. 125
$ LOADING
FORCE 10 16 100. -1. 0. 0.
FORCE 20 16 140. -1. 0. 0.
$ GEOMETRY
GRID 1000 0. 1. 0. 123456
GRDSET 345
GRID 10 0. 0. 0. 12345
GRID 11 1. 0. 0.
GRID 12 2. 0. 0.
GRID 13 3. 0. 0.
GRID 14 4. 0. 0.
GRID 15 5. 0. 0.
GRID 16 6. 0. 0.
$ CONNECTIVITY
CBEAM I 3 10 II 1000
CBEAM 2 3 11 12 1000
CBEAM 3 3 12 13 I000
CBEAM 4 3 13 14 I000
CBEAM 5 3 14 15 1000
CBEAM 6 3 15 16 I000
$
ENDDATA
10.1.5 Example of Nonlinear Buckling

An elastic-plastic buckling analysis of a thin spherical shell is presented [10.1]. With a model
consisting of 16 shell elements (10 Q U A D 4 and one TRIAS), the nonlinear static analysis
determined that buckling would occur between 3500 psi and 3600 psi, where a negative deter-
minant of [I<]was first detected. The results are summarized in Fig. 10.1.2 and Table 10.1.2.

I/
h
s
m
m
a @ buckling point by MSCINASTRAN
+ Finite Element by MSCINASTRAN
1000 Finite Difference by ~ a o ~
@
= 2000 b
buckling pdnt by ~ujikakel
E x ~ e r i m n t aBuckling
l point2

Central Deflection. W (in)

Figure 10.1.2 Elastic-Plastic Buckling of a Clamped Spherical Cap

with Flat Spot (Load vs. Central Deflection)
 Table 10.1.2 MSCINASTRAN Output Summary

Subcase Load Factor Load (psi) No. of Iter. -u100(in) max ale (psi)
1 C 0.50 1000 5 1.739-3 3.532+4
o S 1.0 2000 7 3.719-3 7.555+4
-
1 t 0.20 2200 7 4.154-3 7.894+4****
d a 0.40 2400 8 4.640-3 8.061+4
2 r 0.60 2600 lo* 5.217-3 8.279+4
t 0.80 2800 lo* 5.927-3 8.564+4
R 1.0 3000 20 6.846-3 8.944+4
-
e 0.125 3100 7-K 7.418-3 9.179+4
s 0.25 3200 7 8.091-3 9.460+4
3 t 0.375 3300 9 8.918-3 T9.795+4
a 0.500 3400 17 10.032-3 1.024+5
r 0.625 3500 8-K 11.942-3 1.098+5
t 0.75 3600 2*** - -

* Max limit is reached.

** In this stage of the cold start run, the number of iterations reached the
prescribed maximum limit of 10 with e, = 3.34373-2. The iteration limit
was changed t o 20 and the convergence tolerances were relaxed before
the job restarted.
*** A negative term on factor diagonal in [K] detected. The execution was
terminated after seven iterations due t o two consecutive diverging
solutions with stiffness updating.
**** Plastic deformation initiated. Reference 1 reports that the yielding
initiated at 2600 psi.
-K [K] update during iteration.
 As shown in Table 10.1.3, a restart run was made with PARAM, BUCKLE starting from
the solution a t P=3300 psi. Solutions a t P=3400 psi and 3500 psi were repeated and a buckling
analysis was performed using the differential stiffness between those two loading steps. T h e
results of this eigenvaiue analysis are

X = 0.775 and cr = 0.466

from which the critical buckling load is calculated by

Pc, = P n + Q A P = 3500 + 0.466 x 100 = 3546.6 psi

and the maximum deflection a t the crown by

Uc, = Un + AAU = 0.01224 in.

The loading history to buckling point was obtained as shown in Fig. 10.1.2. Notice that Fujikake
[10.2] overestimated the maximum deflection at buckling. Fig. 10.1.3 shows the deformed shape
on buckling.

Undef ormed shape

Oefonned shape on buckling

Figure 10.1.3 Deformed Shape on Buckling

 This structure was analyzed again with a solid-element model. The buckling point is
somewhat underestimated with solid model, resulting in

PC, = 3520.8 psi and U,, = 0.0122 in.

a t the crown. Fig. 10.1.4 shows the deformed shape on buckling. It seems that the de-
formed shape a t buckling is as significant and useful as the buckling mode shape in the large-
displacement problem. Note that the mode shapes for bifurcation buckling may be quite
different.

,- UndetorW Shape

Figure 10.1.4 Deformed Shape Upon Buckling with a Solid Model

A few more analyses were attempted t o predict buckling point from the solutions at the
earlier stage of loading. When the solution points are farther removed from the actual buckling
point, the prediction becomes less reliable. Thus the user is cautioned t o be suspicious when
the cr: is greater than unity, in which the predicted buckling point is not close t o the instability
point detected by the static solution. In such cases, the solution may be easily validated by
repeating the process with new trial loads closer t o the estimated buckling load.
 Table 10.1.3 Example Problem Input Data
ID SBUCKL,V65 $ SHL 5/7/85 N6609
SOL 66 $ NONLINEAR STATIC ANALYSIS
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
TIME 15 $ CPU MINUTES
CEND
TITLE = ELASTIC-PLASTIC BUCKLING OF IMPERFECT SPHERICAL SHELL A1185G
SUBTITLE = HYDROSTATIC PRESSURE APPLIED, PERIPHERY CLAMPED
ECHO = UNSORT
SEALL = ALL
DISP = ALL
OLOAD = ALL
.
SPCF = ALL
STRESS = ALL
SPC = 10
METHOD = 30
$------------------------ *
$PARAM BUCKLE 1 *
$PARAM SUBID 4 * CHANGES REQUIRED
$PARAM LOADINC 1 * FOR BUCKLING
$PARAM LOOPID 10 *
$------------------------*
SUBCASE 1
LOAD = 1 0
NLPARM = 10
SUBCASE 2
LOAD = 20
BLPARM = 20
SUBCASE 3
LOAD = 30
NLPARM = 30
SUBCASE 4 $ ADDED FOR BUCKLING ANALYSIS
LOAD = 40
NLPARM = 40
OUTPUT(PLOT)
CSCALE = 1.3
PLOTTER NAST
SET 1 = ALL EXCEPT PLOTEL
SET 2 = 1000
SET 3 = ALL
VIEW 30.,20.,0.
MAXI DEFO .05
FIND SCALE ORIGIN 1 SET 3
PTITLE = ISOMETRIC VIEW
PLOT SET 1 ORIGIN 1 SET 2 SYMBOL 3
VIEW 90.,0.,0.
FIND SCALE ORIGIN 1 SET 1
PTITLE = SIDE VIEW
 PLOT SET I ORIGIN I SET 2 SYMBOL 3
PLOT STATIC 0 MAXI DEFO .05 SET I ORIGIN I SET 2 SYMBOL 3
PLOT MODAL 0 SET I ORIGIN I SET 2 SYMBOL 3
BEGIN BULK
$$ PARAMETERS
PARAM SMALLDB I
PARAM LGDISP 1
$$ SOLUTION CONTROL
NLPARM 10 2 AUTO PW
NLPARM 20 5 AUTO PW NO
NLPARM 30 8 AUTO YES
NLPARM 40 2 AUTO I YES
$$ EIGENVALUE ANALYSIS
EIGB 30 SINV 0. 2. 20 2 2 +EIGB
+EIGB MAX
$$ PROPERTIES
MAT1 I 10.8+6 .3
MATS1 I PLASTIC 1.225+6 I 2 7.8+4
PSHELL 2 I .0251 I
$$ BOUNDARY AND LOADING
SPCI 10 123456 131 132
PLOAD2 10 -2000. I0 THRU 25
PLOAD2 20 -3000. 10 THRU 25
PLOAD2 30 -3800. 10 THRU 25
PLOAD2 40 -3500. 10 THRU 25
$$ GEOMETRY AND CONNECTIVITY
CORD2S 100 0. 0. 0. 0. 0. I.
+C2SI I. 0. I.
CORD2S 200 0. 0. -. 32908 0. 0. 1.
+C2S2 I. 0. I.
GRDSET 100 345
GRID 1000 0. 0. 0. 123456
GRID 100 200 1.1506 0. 0. 0 12456
GRID 101 200 1.1506 .715 -5.
GRID 102 200 1.1506 .715 5.
GRID 103 200 1.1506 1.43 -5.
GRID 104 200 1.1506 1.43 5.
GRID 105 200 I. 1506 2.145 -5.
GRID 106 200 1.1506 2.145 5.
GRID 107 200 1.1506 2.86 -5.
GRID 108 200 1.1606 2.86 5.
GRID 109 200 1.1506 3.575 -5.
GRID 110 200 1.1506 3.575 5.
GRID ill 200 1.1506 4.29 -5.
GRID 112 200 1.1506 4.29 5.
GRID 113 200 1.1506 5.005 -5.
GRID 114 200 1.1506 5.005 5 .
GRID 115 200 1.1506 5.72 -5.
GRID 116 200 1.1506 5.72 5.
GRID 117 1.1506
GRID 118 1.1506
GRID 119 .8251
GRID 120 .8251
GRID 121 .8251
GRID 122 .8251
GRID 123 .8251
GRID 124 .8251
GRID 125 .8251
GRID 126 .8251
GRID 127 .8251
GRID 128 .8251
GRID 129 .8251
GRID 130 .8251
GRID 131 .8251
GRID 132 .8251
PLOTEL 1000 100
CTRIA3 10 100 101 102
CQUAD4 11 101 103 104
CQUAD4 12 103 105 106
CQUAD4 13 105 107 108
CQUAD4 14 107 109 110
CQUAD4 15 109 Ill 112
CQUAD4 16 Ill 113 114
CQUAD4 17 113 115 116
CQUAD4 18 115 117 118
CQUAD4 I9 117 I19 120
CQUAD4 20 119 121 122
CQUAD4 21 121 123 124
CQUAD4 22 123 125 126
CQUAD4 23 125 127 128
CQUAD4 24 127 129 130
CQUAD4 25 129 131 132
ENDDATA
10.2 NONLINEAR MODAL ANALYSIS

The modal analysis DMAP, SOL 63, has been modified for nonlinear modal analysis with the
nonlinear stiffness matrices obtained from the SOL 66 database. The rigid format DMAP
alter for SOL 63, RF63D89, is shown in Table 10.2.1. This rigid format alter can be used
in Version 66. The DMAP alter for Version 65 is listed in the Application Note, "Nonlinear
Normal Mode Analysis," dated January 1986. This is an alternate method to the SOL 64 -
SOL 63 sequence for nonlinear normal mode analysis, demonstrated in the Application Note
of May 1981. The main advantage of using SOL 66 instead of SOL 64 is that both geometric
and material nonlinearities can be included. In Version 67, this capability is built in SOL 106
with PARAM, NMLOOP.

This capability is demonstrated by four problems: (a) vibration of a linear material beam,
(b) vibration of a nonlinear material beam, (c) vibration of a plate, and (d) vibration of a
nonlinear stepped beam.

10.2.1 Analysis Procedure

In nonlinear static analysis using SOL 106, normal modes can be computed with the updated
stiffness if PARAM, NMLOOP, n is specified where n is the desired LOOPID. This option
also requires the presence of the METHOD command in the Case Control section and EIGR
or EIGRL entry in the Bulk Data section. If the restart procedure (from SOL 66 t o SOL 63)
is used with RF63D89, data blocks (ESTNL and UGV) are fetched for the nonlinear stiffness
from the SOL 66 run. LOOPID is used as a parameter t o choose the step at which the modal
analysis is desired. If LOOPID is not specified in the SOL 63 run or ESTNL is not found in
the data base, the linear normal modes will be computed.

The parameter LGDISP needs t o be set to 1in both SOLS 66 and 63 to obtain the differen-
tial stiffness matrix t o account for the prestressed condition. Since the superelement analysis
does not have the nonlinear capability, the nonlinear matrix generations for superelements will
be skipped. However, SEALL = n for the residual structure or SEALL = ALL should be used
to initiate the nonlinear stiffness matrix generation.

The stiffness matrix must be updated after the load step where the normal modes are
desired. In order t o force the update, the value of KSTEP in the NLPARM entry should be
set t o 1 with the AUTO, SEMI or ITER method.

The structure of the matrices used in SOL 66 and SOL 63 should be consistent. For
example, when dynamic reduction is t o be applied in SOL 63, the same scalar points should
be defined in SOL 66, even though these points are not actually used. See the examples and
the comments in Table 10.2.1 for other input items required for SOL 63.
10.2.2 Vibration of a Linear Material Beam

Fig. 10.2.1 shows a simply supported beam under axial force that is used to test this capability.
The total length of the beam is 1000 mm and the width and height of its cross section is 100
mm and 50 mm, respectively. The Young's modulus is 2.0684 x l o 5 MPa and the mass density
per unit length is 3.9167 x Nsec2/mm2 (tonlmm). An axial force of 1.0 x l o 7 N is applied
and the beam remains linear. The bending vibration frequencies will increase because of the
axial force. The beam is modeled using five BEAM elements.

E = 2.0684 x lo5 MPa

m = 3.9167 x Nsec2/mm2
P = 1.0 x l o 7 N

Figure 10.2.1 Simply Supported Beam.

To check the accuracy of the finite element beam model, first the theoretical natural fre-
quencies of a simply supported beam without axial force are compared with those obtained
from MSCINASTRAN. The theoretical values are obtained by

EI

The comparison is shown below:

Beam Vibration with P = 0.

Mode w (Theory) w (MSC/NASTRAN) Mode
1 7.3208 x lo2 7.3020 x lo2 1st bending
2 2.9281 x lo3 2.9043 x lo3 2nd bending
3 6.5883 x lo3 6.5079 x lo3 3rd bending
4 --- 8.0385 x lo3 1st longitudinal
5 1.1712 x lo4 1.1589 x l o 4 4th bending

The above table shows that MSCJNASTRAN values compare very well with the theoretical
values. Therefore, the model is quite accurate.
 The natural frequencies of the simply supported beam subjected t o constant axial force are
compared with the theoretical values obtained by

The comparison is shown below:

Linear Beam with P = 1. x lo7 N.
P = 1. x l o 7 N. , E = 2.0684 x lo5 MPa.

Mode w (Theory) w (MSCJNASTRAN) Mode

1 1.7481 x lo3 1.7339 x lo3 1st bending
2 4.3189 x. lo3 4.2828 x lo3 2nd bending
3 --- 8.0385 x lo3 1st longitudinal
4 8.1292 x lo3 8.0453 x lo3 3rd bending
5 1.3323 x l o 4 1.3207 x lo4 4th bending

The above tables show that the axial load affects the lower modes more than the higher modes.

10.2.3 Vibration of a Nonlinear Material Beam

This example calculates the natural frequencies of a simply supported beam (Fig. 10.2.1) sub-
jected t o an axial load of 1.0 x l o 7 N , and with the nonlinear elastic stress-strain relationship
shown in Fig. 10.2.2. The input data for this model are shown in Tables 10.2.2 and 10.2.3.

.001 .003 .005 .011

&
Figure 10.2.2 Nonlinear Stress-Strain Relationship for Simply Supported Beam
 Analysis results are compared with the theoretical solution as tabulated below. The theo-
retical values are calculated by using the tangent modulus at the final solution ( E = 0.031).

Nonlinear Beam with P = 1. x lo7 N.

P = 1. x lo7 N . , Eo = 2.0684 x lo5 MPa. , E = 5 . x lo4 MPa.
Mode w (Theory) w (MSC/NASTRAN) Mode
1 1.6277 x lo3 1.6768x lo3 1st bending
2 3.4859 x lo3 3.7978 x lo3 2nd bending
3 --- 6.0180 x lo3 1st longitudinal
4 5.7593 x lo3 6.5456 x lo3 3rd bending
5 8.5717 x lo3 9.9099 x lo3 4th bending

The frequencies of the lower modes have increased while those of the higher modes de-
creased. This is due t o the fact that the softening material effect is larger on the higher modes
than the longitudinal tension.

10.2.4 Plate Vibration

This example demonstrates geometric nonlinear modal analysis for a rectangular plate. The
plate has a thickness of 1.0 cm and the dimension is 2.0 m x 4.0 m. The Young's modulus is
2.0 x 10'' N/m2 and the mass density is 8.0 x lo3 kg/m3. The plate is clamped on all sides
and is subjected t o a pressure loading of 2.0 x lo5 P a normal t o its plane.

Due t o symmetry, only one quarter of the plate is modeled using 50 QUAD4 elements as
shown in Fig. 10.2.3. The clamped boundary condition is applied t o sides BC and CD by
constraining all six DOFs. The axisymmetric boundary condition is imposed by constraining
components 2, 4, and 6 on side AB and components 1, 5, and 6 on side AD. The input data
for this model are listed in Tables 10.2.4 and 10.2.5 for nonlinear static analysis and modal
analysis, respectively.
 Figure 10.2.3 Clamped Plate (Quarter Only, Due t o Symmetry)

At first, SOL 66 provides information about the nonlinear stiffness matrix. In the SOL 63
restart, dynamic reduction is used.

The results are tabulated along with frequencies of the linear system without the pressure
load. The lower modes are affected more by the pressure load than the higher modes, which
indicates the same tendency as in the beam problem.

Angular Frequencies of Plate.

Mode w (Linear) w (Nonlinear) Mode
1 9.2595 x lo1 4.7648 x lo2 1st bending
1 2 1 1.6710 x l o 2 / 5.5690 x lo2 / 2nd bending I
3 3.2677 x l o 2 7.1118 x lo2 3rd bending
4 4.6600 x l o 2 9.0174 x lo2 4th bending
5 1 5.3634 x lo2 1 9.4842 x l o 2 I 5th bending 1
10.2.5 Vibration of a Stepped Beam

This example demonstrates the geometric and material nonlinear modal analysis for a simply
supported stepped beam, in Fig. 10.2.4. The material is nonlinear elastic with the properties
shown in Fig. 10.2.2. The total length of the beam is 1000 mm and its cross sectional areas
vary from 120 mm to 40 mm. The Young's modulus is 2.0684 x lo5 MPa and the mass density
per unit length is 3.9167 x Nsec2/mm2 (tonlmm). An axial force of 1.5 x lo6 N is
applied, for which elements 1 and 2 remain linear. The stepped beam is modeled using five
BEAM elements. The input data for this model are listed in Tables 10.2.6 and 10.2.7.
 L = 1.0 x lo3 mm
E = 2.0684 x l o 5 MPa
m = 3.9167 x Nsec2/mm2
P = 1.5 x lo6 N

Figure 10.2.4. Stepped Beam.

The bending vibration frequencies are tabulated below. The first column represents the
linear system without the axial load. The second column shows the nonlinear system with the
axial load of 1.5 x lo6 N.

Angular Frequencies of Stepped Beam.

P = 1.5 x l o 6 N. .
---
E = 2.0684 x l o 5 MPa.
Mode w (Linear) w (Nonlinear) Mode
1 1.0112 x lo3 1.1142 x lo3 1st bending
1 2 1 1
5.1089 x lo3 4.9748 x lo3 / 2nd bending (
1 3 1 1
1.1614 x lo4 1.0806 x lo4 / 3rd bending /
1 4 1 I1
1.2437 x l o 4 1.2179 x lo4
I1 1st longitudinal 1
1 5 1 1.9544 x l o 4 1.8154 x lo4 4th bending /
The nonlinear frequencies, except the first, have decreased. This is because the softening
material effect is larger than the longitudinal tension. This result is consistent with the previous
analyses (Section 10.2.2 and 10.2.3).
 Table 10.2.1 DMAP Alters for SOL 63 (Version 66)
$ BEGINNING OF RF ALTER 63D89
$ + + + + + 22-DEC-1988 +++++++
$
$ RF63D89 -- ALTER FOR NORMAL MODES WITH NONLINEAR STIFFNESS
$
$ JCL
$ ATTACH DATABASE FROM SOL 66 COLD START
$
$ EXECUTIVE DECK INPUT
$ NO SPECIAL CARDS REQUIRED
$
$ CASE CONTROL DECK INPUT
$ SEALL = ALL
$ SUPER = N
$
$ BULK DATA DECK INPUT
$ PARAM,LGDISP,I- SELECT LARGE DISPLACEMENTEFFECTS (DEFAULT=-I)
$ PARAM,LOOPID,I- SELECTS NONLINEAR STIFFNESS (DEFAULT=O)
$
ALTER 483 $ JUST AFTER LABEL JMPK4GG
$ ACTUALLY JUST AFTER OLD DBSTORE K4JJ
$ NOTE KELM AND KDICT IF IN HERE NEVER STORED
TYPE DB,ESTNL,UGV $ PATH = SSNL, LOCATION = DBUP
COND LNSTIF, ACON $ SKIP THE ALTERS FOR SUPERELEMENTS
PARAM //lLT1/S,N,NOLOOP/V,Y,LOOPID=O/l $ READ LOOPID CARD
COND LNSTIF, NOLOOP $ NO LOOPID, GO TO LINEAR STIFFNESS
$DBFETCH /UGV,ESTNL,,,/SOLID/LOOPID//DBSET3 $ DATA FROM SOL 66 RUN
DBVIEW UGV66 =UGV (WHERE VERSION=RESTART AND SOLFETCH=LOOPID) $ +V66NX
DBVIEW ESTNL66=ESTNL (WHERE VERSION=RESTART AND SOLFETCH=LOOPID) $ +V66NX
EQUIV UGV66,UGV/ALWAYS/ESTNL66,ESTNL/ALWAYS $ +V66NX
PARAML ESTNL//JPRES'////S,N,NONLK$ CHECK PRESENCE OF ESTNL
COND LNSTIF, NONLK $ NO ESTNL, GO TO LINEAR STIFFNESS
TAI, MPT,ECTS,EPT,BGPDTS,SILS,ETT,CSTMS,DIT/
EST,DESTNL,GEI,GPECT,ESTL/V,N,LUSETS/S,N,NOESTL/
S,N,NP/2/S,N,NOGENL/SEID/S,Y,LGDISP=-I/
V,N,NLAYERS=5 $ GENERATE TABLE FOR LINEAR ELEMENT
COND JMPKGG ,NOKGGX $
EMG ESTL,CSTMS,MPT,DIT,GEOM2S,,,/KELM,KDICT,,,,/
S,N,NOKGXX/O/O/O/////////////K6ROT $ STIFFNESS FOR LINEAR ELEMENTS
EMA GPECT,KDICT,KELM,BGPDTS,SILS,CSTMS/KBJJZ $
EMG ESTL,CSTMS,MPT,DIT,,UGV,ETT,EDT/~LEL,KDLDI,,,,/
S,Y,NOD=I/O/O//NP $ DIFFERENTIAL STIFFNESS FOR LINEAR ELEMENTS
EMA GPECT,KDLDI,KDLEL,BGPDTS,SILS,CSTMS/KDLGG,/-I/$
ADD KB J JZ,KDLGG/KLTOT/$
EMG ESTNL,CSTMS,MPT,DIT,GEOM2S,,,/KELMNL,KDICTNL,,,,/
I/O/O//////////////V,Y,K6ROT $ STIFFNESS FOR NONLINEAR ELEMENTS
EMA GPECT,KDICTNL,KELMNL,BGPDTS,SILS,CSTMS/KBJJZNL,/$
ADD KLTOT ,KBJ JZNL/KBI/ $
EMG ESTNL,CSTMS,MPT,DIT,,UGV,ETT,EDT/KDELM,KDDICT,,,,/
I/O/O//NP/ $ DIFFERENTIAL STIFFNESS FOR NONLINEAR ELEMENTS
EMA GPECT,KDDICT,KDELM,BGPDTS,SILS,CSTMS/KBDJJ,/-I/$
DELETE /KJJZ,KELM,KDICT,,/ $ ++V66NX
ADD KBI,KBDJJ/KJJZ/$
JUMP JMPKGG $ SKIP LINEAR STIFFNESS GENERATION
LABEL LNSTIF $ LABEL FOR LINEAR STIFFNESS GENERATION
$ END OF RF ALTER 63D89
 Table 10.2.2 SOL 66 Input Data Listing for Uniform Nonlinear Beam
ID BEAMWT, V66 $ SHL 2/13/90 $ A0186H, CSR6347 $ JFC 13-DEC-1989
SOL 66
TIME 5
CEND
$ GENERATE NONLINEAR STIFFNESS TERMS FOR SOL 63 RUN
TITLE=EXTENSION OF S-S BEAM
SUBTITLE=FIVE BEAM ELEMENTS
SET 5 = 5
SET 6 = 6
DISP = 6
STRESS = 5
SEALL = ALL
SUBCASE I
LOAD=IOO
NLPARM=I0
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
SET I ALL
VIEW 0. , 0. , 0.
AXES MY , X , Z
FIND SCALE ORIGIN I SET I
PTITLE=SIDE VIEW
PLOT MODAL 0 SET I ORIGIN 1 SYMBOL 1 LABEL GRID
BEGIN BULK
$ PARAMETERS
PARAM LGDISP I
$
NLPARM 10 1 AUTO I PW YES
$ GEOMETRY
GRID I 0. 0. 0. 12346
GRID 2 200. 0. 0. 246
GRID 3 400. 0. 0. 246
GRID 4 600. 0. 0. 246
GRID 5 800. 0. 0. 246
GRID 6 1000. 0. 0. 2346
$ CONNECTIVITY
CBEAM I 1 1 2 0. I. 0.
CBEAM 2 I 2 3 0. 1. 0.
CBEAM 3 I 3 4 0. 1. 0.
CBEAM 4 I 4 5 0. I. 0.
CBEAM 5 1 5 6 0. I. 0.
$ BEAM PROPERTIES
PBEAM I I 5.+3 4.1667+61.0417+6
$ MATERIAL PROPERTIES
MAT1 I 2.0684+5 0.3 7.8334-9
MATS1 1 101 NLELAST
TABLES1 I01
+TBI 0. 0. .001 206.84 .003 500. .005 700. +TB2
+TB2 .Oil 1000. ENDT
$ LOADING
FORCE 100 6 1.+7 1.
$
ENDDATA
 Table 10.2.3 SOL 63 Input Data Listing for Uniform Nonlinear Beam
RESTART
ID BEAMWT63, V66 $ SHL 2/13/90 $ A01861, CSR6347R $ JFC 13-DEC-1989
DIAG 8
TIME 5
SOL 63 $ NORMAL MODES
COMPILE SOL63,SOUIN=MSCSOU,LIST,REF
$ RFALTER RF63D89 IN V66 AND V66A NEEDS THE FOLLOWING LINE INSERTED IN
$ THE ALTER.
$ FILE EST=OVRWRT/GEI=OVRWRT/CPECT=OVRWRT/KELM=OVRWRT/ICTRWRT $
ALTER 483 $
FILE EST=OVRWRT/GEI=OVRWRT/GPECT=OVRWRT/KELM=~~R~RT/KDICT=OVRWRT $
RFALTER RF63D89
CEND
TITLE=NONLINEAR VIBRATION OF S-S BEAM
SUBTITLE=FIVE BEAM ELEMENTS
DISP = ALL
SEALL = ALL
METHOD = 10
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
SET I ALL
VIEW 0. , 0. , 0.
AXES MY , X , Z
FIND SCALE ORIGIN I SET I
PTITLE=SIDE VIEW
PLOT MODAL 0 SET 1 ORIGIN 1 SYMBOL 1 LABEL GRID
BEGIN BULK
$ PARAMETERS
PARAM LGDISP 1
PARAM LOOPID I
PARAM COUPMASS1
$ GEOMETRY
GRID I 0. 0. 0. 12346
GRID 2 200. 0. 0. 246
GRID 3 400. 0. 0. 246
GRID 4 600. 0. 0. 246
GRID 5 800. 0. 0. 246
GRID 6 1000. 0. 0. 2346
$ CONNECTIVITY
CBEAM 1 1 1 2 0. I. 0.
CBEAM 2 1 2 3 0. I. 0.
CBEAM 3 I 3 4 0. I. 0.
CBEAM 4 I 4 5 0. I. 0.
CBEAM 5 I 5 6 0. I. 0.
$ BEAM PROPERTIES
PBEAM 1 1 5.+3 4.1667+61.0417+6
$ MATERIAL PROPERTIES
MAT1 I 2.0684+5 0.3 7.8334-9
$ EIGENVALUE EXTRACTION
EIGR 10 SINV 0. 4000. 8 8
$
ENDDATA
 Table 10.2.4. SOL 66 Input Data Listing for Plate
NASTRAN MESH $ PREFORT=2
ID PLATEWP, V66 $ SHL 2/13/90 $ MSC,AOI86J $ KOK 11/85
SOL 66
TIME 5
CEND
$ GENERATE NONLINEAR STIFFNESS MATRIX FOR SOL 63 RESTART
TITLE= GEOMETRIC NONLINEAR ANALYSIS OF A FLAT PLATE
SUBTITLE= PRESSURE LOAD WITH CLAMPED EDGES
LABEL= REF. APP. NOTE, MAY 1981
SEALL = ALL
DISP = ALL
STRESS = ALL
SPC = 10
SUBCASE I
LOAD=I
NLPARM=10
SUBCASE 2
LOAD=2
NLPARM=20
OUTPUT (PLOT)
CSCALE 1.3
PLOTTER NAST
SET I ALL
VIEW 30., 20., 0.
FIND SCALE ORIGIN I SET I
PTITLE=ISOMETRIC VIEW
PLOT MODAL 0 SET 1 ORIGIN 1 SYMBOL I
VIEW 0. , 0. , 0.
AXES Y , X , Z
FIND SCALE ORIGIN I SET I
PTITLE=SIDE VIEW
PLOT MODAL 0 SET I ORIGIN I SYMBOL I LABEL GRID
AXES X , Y , Z
FIND SCALE ORIGIN I SET I
PTITLE=FRONT VIEW
PLOT MODAL 0 SET I ORIGIN I SYMBOL I LABEL GRID
BEGIN BULK
$ PARAMETERS
PARAM LGDISP I
PARAM K6ROT 100.
$
NLPARM 10 2 AUTO PW YES
NLPARM 20 2 AUTO I PW YES
$ VERTEX GRID POINTS
EGRID 1
EGRID 2 2.
EGRID 3 2. 1.
EGRID 4 I.
$ GENERATE GRID POINT ARRAY
GRIDG 1 GRIDGI
+RIDGI 5 -4
$ GENERATE QUAD4 ELEMENTS
CGEN QUAD4 201 200
$ PROPERTIES
PSHELL 200 I .01
MAT1 I 2.+11
$ CONSTRAINTS
SPCG 2 I 156
SPCG 2 1 246
SPCG I I 123456
SPCADD 10 1 2
$ APPLIED LOADS
PLOAD4 I 201 1.+5 THRU 250
PLOAD4 2 201 2.+5 ,
THRU 250
$ THESE POINTS ARE FOR SOL 63
SPOINT 20001 THRU 20012
SPCl 3 0 20001 THRU 20012
$
ENDDATA
 Table 10.2.5 SOL 63 Input Data Listing for Plate
NASTRAN MESH $ PREFORT=2
RESTART
ID PLATEWP63, V66 $ SHL 2/13/90 $ A0581B $ JAJ 8 DEC 82
TIME 5
SOL 63
COMPILE SOL63,SOUIN=MSCSOU,LIST,REF
ALTER 483 $
FILE EST=OVRWRT/GEI=OVRWRT/GPECT=OVRWRT/KELM=OVRWRT/KDICTOVRWRT $
RFALTER RF63D89
CEND
TITLE=SYMMETRIC MODES
SUBTITLE=PRELOAD OF UNIFORM PRESSURE-CLAMPED EDGES
LABEL = REF. APP. NOTE OF MAY 1981
SEALL = ALL
SPC = 15
SET 1 = 0
DISP = ALL
METHOD = I
DYNRED = 1
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
SET I ALL
VIEW 30., 20., 0.
FIND SCALE ORIGIN I SET I
PTITLE=ISOMETRIC VIEW
PLOT MODAL 0 SET I ORIGIN I SYMBOL 1
VIEW 0. , 0. , 0.
AXES Y , X , Z
FIND SCALE ORIGIN I SET I
PTITLE=SIDE VIEW
PLOT MODAL 0 SET I ORIGIN I SYMBOL I LABEL GRID
AXES X , Y , Z
FIND SCALE ORIGIN 1 SET 1
PTITLE=FRONT VIEW
PLOT MODAL 0 SET I ORIGIN I SYMBOL I LABEL GRID
BEGIN BULK
$ PARAMETERS
PARAM LOOPID 4
PARAM LGDISP I
PARAM K6ROT 100.
$ VERTEX GRID POINTS
EGRID 1
EGRID 2 2.
EGRID 3 2. I.
EGRID 4 I.
$ GENERATE GRID POINT ARRAY
GRIDG I 10 -1 -2 -3 GRIDGI
tRIDG1 5 -4
$ GENERATE QUAD4 ELEMENTS
CGEN QUAD4 201 200 1
$ PROPERTIES
PSHELL 200 1 .01 1
MAT1 1 2.+11 .3 8000.
$ CONSTRAINTS
SPCG 2 1 156 AD
SPCG 2 i 246 AB
SPCG 1 1 123456 BC CD
$ SYMMETRIC BOUNDARY CONDITION FOR SOL 63
SPCADD 15 1 2
$ DATA FOR GENERALIZED DYNAMIC REDUCTION AND EIGENVALUE EXTRACTION - SOL 63
SPOINT 20001 THRU 20012
DYNRED 1 160. 8
EIGR 1 MGIV 0. 160. EIGRi
+IGRI MASS
ASETl 0 20001 THRU 20012
qSETi 0 20001 THRU 20012
$
ENDDATA
 Table 10.2.6. SOL 66 Input Data Listing for Stepped Beam
ID SBEAMWT, V66 $ SHL 1/13/90 $ A0186L KOK 12/85
TIME 5
SOL 66
CEND
$ GENERATE NONLINEAR STIFFNESS TERMS FOR SOL 63 RUN
TITLE=EXTENSION OF S-S STEPPED BEAM
SUBTITLE=FIVE BEAM ELEMENTS
DISP = A L L
STRESS = ALL
SEALL = ALL
SUBCASE I
LOAD=IOO
NLPARM=10
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
SET I ALL
VIEW 0. , 0. , 0.
AXES MY , X , Z
FIND SCALE ORIGIN 1 SET 1
PTITLE=SIDE VIEW
PLOT MODAL 0 SET I ORIGIN I SYMBOL I LABEL GRID
BEGIN BULK
$ PARAMETERS
PARAM LGDISP 1
$
NLPARM 10 I AUTO 1 PW YES
$ GEOMETRY
GRID 1 0. 0. 0.
GRID 2 200. 0. 0.
GRID 3 400. 0. 0.
GRID 4 600. 0. 0.
GRID 5 800. 0. 0.
GRID 6 1000. 0. 0.
$ CONNECTIVITY
CBEAM I II I 2 0. I.
CBEAM 2 12 2 3 0. I.
CBEAM 3 13 3 4 0. I.
CBEAM 4 14 4 5 0. 1.
CBEAM 5 15 5 6 0. 1.
$ BEAM PROPERTIES
PBEAM 11 I 1.44+4 1.728+7 1.728+7
PBEAM 12 I 1.00+4 8.333+6 8.333+6
PBEAM 13 1 6.40+3 3.413+6 3.413+6
PBEAM 14 1 3.60+3 1.080+6 1.080+6
PBEAM 15 I 1.60+3 2.133+6 2.133+6
$ MATERIAL PROPERTIES
MAT1 1 2.0684+5 0.3 7.8334-9
MATS1 I 101 NLELAST
TABLES1 101 +TB1
+TBI 0. 0. .001 206.84 .003 500. .005 700. +TB2
+TB2 .Oil 1000. ENDT
$ LOADING
FORCE 100 6 1.5+6 I.
$
ENDDATA
 Table 10.2.7. SOL 63 Input Data Listing for Stepped Beam
RESTART
ID SBEAMWT63, V66 $ SHL 2/13/90 $ A0186M $ KOK 12/85
DIAG 8
TIME 5
SOL 63
COMPILE SOL63,SOUIN=MSCSOUYLIST,REF
ALTER 483 $
FILE EST=OVRWRT/GEI=OVRWRT/GPECT=OVRWRT/KELM=OVRWRT/KDIC=OVR $
RFALTER RF63D89
CEND
TITLE=VIBRATION OF S-S STEPPED BEAM
SUBTITLE=FIVE BEAM ELEMENTS
SEALL = ALL
DISP = A L L
METHOD = 10
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
SET I ALL
VIEW 0. , 0. , 0.
AXES MY , X , Z
FIND SCALE ORIGIN I SET I
PTITLE=SIDE VIEW
PLOT MODAL 0 SET I ORIGIN I SYMBOL 1 LABEL GRID
BEGIN BULK
$ PARAMETERS
PARAM LOOPID 1
PARAM LGDISP I
PARAM COUPMASSI
$ GEOMETRY
GRID I 0. 0. 0.
GRID 2 200. 0. 0.
GRID 3 400. 0. 0.
GRID 4 600. 0. 0.
GRID 5 800. 0. 0.
GRID 6 1000. 0. 0.
$ CONNECTIVITY
CBEAM I 11 1 2 0.
CBEAM 2 12 2 3 0.
CBEAM 3 13 3 4 0.
CBEAM 4 14 4 5 0.
CBEAM 5 15 5 6 0.
$ BEAM PROPERTIES
PBEAM I1 I 1.44+4 1.728+7 1.728+7
PBEAM 12 I 1.00+4 8.333+6 8.333+6
PBEAM 13 I 6.40+3 3.413+6 3.413+6
PBEAM 14 I 3.60+3 1.080+6 1.080+6
PBEAM 15 I 1.60+3 2.133+6 2.133+6
$ MATERIAL PROPERTIES
MAT1 1 2.0684+5 0.3 7.8334-9
$ EIGENVALUE EXTRACTION
EIGR 10 SINV 0. 4000. 8 8
$
ENDDATA
10.3 STATIC ANALYSIS USING SOL 99

Static analysis can be performed using SOL 99 (or 129), the nonlinear transient solution
sequence. Since the effects of inertia and damping forces are ignored in the static analysis, the
automatic time stepping algorithm is not applicable. However, the bisection algorithm is active
with the adaptive stiffness matrix update strategy and other expedient iteration schemes such
as the BFGS updates and the line search process. The TSTATIC option was implemented
in Version 66 when the bisection was not available in SOL 66. The bisection algorithm is
implemented in SOL 66 (or 106) in Version 67.

While the static capability in SOL 99 provides convenience for transient response analysis
of a preloaded structure, it lacks some features provided by SOL 66. SOL 99 static capability
is confined to what SOL 99 can perform and abides by its limitations. Restrictions on the SOL
99 input data include the following:
No constraint changes (SPC, MPC) after the first subcase, including restarts
r No enforced displacements (SPCD) allowed
r No CREEP or thermal loads allowed
The SOL 99 static analysis is compared with SOL 66 for two example problems. In the first
example, material nonlinearity is combined with geometric nonlinear effects in the analysis of
a Z-shaped beam with a static load. The second example combines frictional contact with
geometric nonlinear effects in the rotation of a beam with a GAP element, and illustrates SOL
99 static and dynamic analysis using the same input data.

10.3.1 Input Data for SOL 99 Static Analysis

The important input data for static analysis using SOL 99 are described below:
r The parameter TSTATIC with a value of 1 must be included in the Case Control or
Bulk Data section in order t o activate the static analysis. TSTATIC with a value of
-1 (default) activates the dynamic analysis. The value of TSTATIC can be varied from
subcase t o subcase, thus allowing a static and dynamic analysis in a single run from the
same input data.

r The solution strategy is controlled by the TSTEPNL Bulk Data entry, in which the
ADAPT method (NLTRD2 module) must be specified for static analysis. A minimum of
4 time steps is required in the NDT field for each subcase. The MAXITER and MAXLS
fields should be assigned a larger value than the defaults (e.g., 25 and 4) in order t o allow
more iterations and line searches, which is the general strategy adopted in the NLPARM
for static analysis.

r The Bulk Data LSEQ, selected by LOADSET in the Case Control section, references
static load entries (i.e., FORCE, MOMENT, PLOAD, etc.) and provides static t o dy-
 namic load conversion. LSEQ also selects DAREA which is internally generated t o
define the points of dynamic load application and is referenced by TLOADi. The time-
dependent function for the dynamic load is specified by a TLOADi entry which is defined
by a tabular function via TABLED input or by an analytic function. The static loading
history should be defined following the same convention using TLOADi, i.e., time implies
the static load step in the subcase with TSTATIC=l. One advantage in SOL 99 static
analysis is that the loading path does not have t o be increased linearly within a subcase
as in SOL 66.

The parameter MAXLP determines the maximum number of internal subincrements in

the plasticity process. If a plastic material is used in the finite element model, MAXLP
with a value of 10 should be included in the input d a t a t o conform with the static default.
The default value for dynamic analysis is 5.

10.3.2 A Z-Shaped Beam Loaded with a Static Force

A Z-shaped cantilever beam, shown in Fig. 10.3.1, was proposed as a benchmark problem for
testing nonlinear finite element analysis codes [10.4]. The Z-shaped beam is loaded statically
with a vertical force of 2000 N a t one end and clamped a t the other end. The material is
elastic-plastic obeying the von Mises yield criterion and the isotropic hardening rule. The
problem is geometrically nonlinear, involving almost 90 degree rotation, but the follower force
effect is absent.

A = 68 m m 2 ; cross-sectional a r e a
I,,= 65.5067 m m 4 ; moment of inertia
E = 2 x lo5 n'/rnm2; Young's modulus
v= 0.3; Poisson's ratio
a, = 450 ~ / r n r n ' ; yield stress

Figure 10.3.1 Z-Shaped Cantilever B e a m

 The beam is modeled by ten BEAM elements of various lengths as seen in Fig. 10.3.1.
Smaller elements are used near the fixed end because of the large plastic deformation antici-
pated in this area. Material properties are given by the MAT1, MATS1, and TABLES1 Bulk
Data entries representing the elasto-plastic material as shown in Fig. 10.3.2. The clamped
boundary condition at the end of the beam is imposed by constraining all six components of
GRID 1. The load applied at the free end of the beam is specified by FORCE and LOAD Bulk
Data entries in SOL 66 and by FORCE, LSEQ, TLOAD1, and TABLED1 Bulk Data entries
in SOL 99. The NLPARM and TSTEPNL Bulk Data entries provide the solution strategy in
SOL 66 and SOL 99, respectively. The input data are given in Table 10.3.1 and 10.3.2.

Figure 10.3.2 Nonlinear Stress-Strain Relationship for Z-Shaped Beam

The analysis is challenging because the substantial loading results in very large rotation
and severe plastic deformation at the fixed end. The loading history, which is identical for
SOL 66 and SOL 99, is divided into 20 subcases. Each subcase specifies 10 increments, except
for subcase 2 and 3 which contain 25 increments to cope with the severe rotation and plastic
yielding in this domain.

The solution produced by SOL 66 and SOL 99 was identical. Fig. 10.3.3 shows the deformed
shapes of the beam at progressively increasing loading stages. The load-deflection curve at the
free end is shown in Fig. 10.3.4. The numerical performance of each subcase, comparing SOL
66 and SOL 99, is tabulated below.
 Performance Comparison for 2-Shaped Beam

SOL 66 SOL 99
Subcase No. of No. of No. of No. of No. of No. of
ID Iter K Update Bis Iter K Update Bis

18 26 1 0 32 1 0
19 21 1 0 26 1 0
20 20 0 0 24 1 0
TOTAL 1495 179 23 2880 74 59
 Figure 10.3.3 Deformed Shapes of Z-Shaped Beam

Displacement (mrn)

Figure 10.3.4 Load-Deflection Curve for Z-Shaped Beam

 Table 10.3.1 SOL 66 Input Data Listing for 2-Shaped Beam
ID ZBEAM, V67 $ KOK 5-APR-85, SHL 9-MAY-85, CYP 19-OCT-87
TIME 30 $ CPU MINUTES FOR VAX
SOL 66 $ NONLINEAR STATIC ANALYSIS
DIAG 8, 50 $ PRINT MATRIX TRAILERS AND ITERATIONS
CEND
TITLE = Z-BEAM ANALYSIS WITH 10 BEAM ELEMENTS
SUBTITLE = COLOGNE CHALLENGE, AUG. 1984
SEALL = ALL
DISP = ALL
SUBCASE I
LOAD = 10
NLPARM = 10
SUBCASE 2
LOAD = 20
NLPARM = 25
SUBCASE 3
LOAD = 30
NLPARM = 25
SUBCASE 4
LOAD = 40
NLPARM = 10
SUBCASE 5
LOAD = 50
NLPARM = 10
SUBCASE 6
LOAD = 60
NLPARM = 10
SUBCASE 7
LOAD = 70
NLPARM = 10
SUBCASE 8
LOAD = 80
NLPARM = 10
SUBCASE 9
LOAD = 90
NLPARM = 10
SUBCASE 10
LOAD = 100
NLPARM = 10
SUBCASE I1
LOAD = 110
NLPARM = 10
SUBCASE 12
LOAD = 120
NLPARM = 10
SUBCASE 13
LOAD = 130
NLPARM = 10
SUBCASE 14
LOAD = 140
NLPARM = 10
SUBCASE 15
LOAD = 150
NLPARM = 10
SUBCASE 16
LOAD = 160
NLPARM = 10
SUBCASE 17
LOAD = 170
NLPARM = 10
SUBCASE 18
LOAD = 180
NLPARM = 10
SUBCASE 19
LOAD = 190
NLPARM = 10
SUBCASE 20
LOAD = 200
NLPARM = 10
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
MAXI DEFO 1.
SET 1 = ALL
VIEW O., O., 0.
AXES Z, X, Y
PTITLE = TOP VIEW
FIND SCALE ORIGIN 1 SET I
PLOT STATIC 0 MAXI DEFO 1. SET I ORIGIN 1
OUTPUT(XYPLOT)
CSCALE 1.3
PLOTTER NAST
XTITLE = FORCE (N)
YTITLE = TIP DISPLACEMENT (MM) AT NODES 5, 8, 11
XYPLOT = DISP RESP /5 (T2) ,8(~2),I1(T2)
BEGIN BULK
PARAM LGDISP 1
$$ SOLUTION CONTROL
NLPARM 10 10 AUTO 7
NLPARM 25 25 AUTO 7
$$ GEOMETRY
GRID I 0. 0. 0.
GRID 2 10. 0. 0.
GRID 3 20. 0. 0.
GRID 4 40. 0. 0.
GRID 5 60. 0. 0.
GRID 6 80. 10. 0.
GRID 7
GRID 8
GRID 9
GRID 10
GRID 11
GRID 99
$$ CONNECTIVITY
CBEAM 100 100
CBEAM 110 100
CBEAM 120 100
CBEAM 130 100
CBEAM 140 100
CBEAM 150 100
CBEAM 160 100
CBEAM 170 100
CBEAM 180 100
CBEAM 190 100
PLOTEL 200 I
$$ PROPERTIES
PBEAM 100 110 68. 65.5067 2266.667
MAT1 110 2.+5 .3
MATS1 110 111 PLASTIC 450.
TABLES1 Ill +TBSI
+TBS1 0. 0. 2.25-3 450. 8.25-3 570. 5.825-2 580. +TBS2
+TBS2 ENDT
$$ LOADING
FORCE 1000 I1
LOAD 10 I.
LOAD 20 I.
LOAD 30 I.
LOAD 40 I.
LOAD 50 I.
LOAD 60 I.
LOAD 70 I.
LOAD 80 I.
LOAD 90 1.
LOAD 100 I.
LOAD 110 I.
LOAD 120 1.
LOAD 130 I.
LOAD 140 I.
LOAD 150 I.
LOAD 160 I.
LOAD 170 I.
LOAD 180 I.
LOAD 190 I.
LOAD 200 I.
ENDDATA
 Table 10.3.2 SOL 99 Input Data Listing for 2-Shaped Beam
ID ZBEAMD,V67 $ SSH 3-DEC-87, SHL 9-MAY-85, CYP 19-OCT-87
TIME 30 $ CPU MINUTES FOR VAX
SOL 99 $ NONLINEAR TRANSIENT ANALYSIS
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
CEND
TITLE = 2-BEAM ANALYSIS WITH 10 BEAM ELEMENTS
SUBTITLE = COLOGNE CHALLENGE, AUG. 1984
SEALL = ALL
DISP = ALL
$--------------- $
LOADSET = 1000 $
$--------------- $
SUBCASE I
DLOAD = 100
TSTEPNL = 10
SUBCASE 2
DLOAD = 100
TSTEPNL = 25
SUBCASE 3
DLOAD = 100
TSTEPNL = 25
SUBCASE 4
DLOAD = 100
TSTEPNL = 10
SUBCASE 5
DLOAD = 100
TSTEPNL = 10
SUBCASE 6
DLOAD = 100
TSTEPNL = 10
SUBCASE 7
DLOAD = 100
TSTEPNL = 10
SUBCASE 8
DLOAD = 100
TSTEPNL = 10
SUBCASE 9
DLOAD = 100
TSTEPNL = 10
SUBCASE 10
DLOAD = 100
TSTEPNL = 10
SUBCASE I 1
DLOAD = 100
TSTEPNL = 10
SUBCASE 12
DLOAD = 100
TSTEPNL = 10
SUBCASE 13
DLOAD = 100
TSTEPNL = 10
SUBCASE 14
DLOAD = 100
TSTEPNL = 10
SUBCASE 15
DLOAD = 100
TSTEPNL = 10
SUBCASE 16
DLOAD = 100
TSTEPNL = 10
SUBCASE 17
DLOAD = 100
TSTEPNL = 10
SUBCASE 18
DLOAD = 100
TSTEPNL = 10
SUBCASE 19
DLOAD = 100
TSTEPNL = 10
SUBCASE 20
DLOAD = 100
TSTEPNL = 10
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
MAXI DEFO I.
SET I = ALL
VIEW O., O., 0.
AXES Z, X, Y
FIND SCALE ORIGIN 1 SET 1
PLOT TRANS 0 MAXI DEFO 1. SET I ORIGIN I
OUTPUT(XYPLOT)
CSCALE 1.3
PLOTTER NAST
XTITLE = FORCE (N)
YTITLE = TIP DISPLACEMENT (MM) AT NODES 5, 8, I1
XYPLOT DISP RESP /5(T2) ,8(T2), ll(T2)
BEGIN BULK
$------------------ $
PARAM TSTATIC 1
$------------------ $
PARAM LGDISP 1
PARAM MAXLP 10
$$ SOLUTION CONTROL
TSTEPNL 10 10 .I ADAPT 25
+TSI 4
TSTEPNL 25 25 .04 ADAPT 25
+TS2
$$ GEOMETRY
GRID I 0. 0. 0.
GRID 2 10. 0. 0.
GRID 3 20. 0. 0.
GRID 4 40. 0. 0.
GRID 5 60. 0. 0.
GRID 6 80. 10. 0.
GRID 7 100. 20. 0.
GRID 8 120. 30. 0.
GRID 9 140. 30. 0.
GRID 10 180. 30. 0.
GRID 11 200. 30. 0.
GRID 99 0. 200. 0.
$$ CONNECTIVITY
CBEAM 100 100 1 2 99
CBEAM 110 100 2 ' 3 99
CBEAM 120 100 3 4 99
CBEAM 130 100 4 5 99
CBEAM 140 100 5 6 99
CBEAM 150 100 6 7 99
CBEAM 160 100 7 8 99
CBEAM 170 100 8 9 99
CBEAM 180 100 9 10 99
CBEAM 190 100 10 11 99
PLOTEL 200 1 99
$$ PROPERTIES
PBEAM 100 110 68. 65.5067 2266.667
MAT1 110 2.+5 .3
MATS1 110 II I PLASTIC 450.
TABLES1 I l l +TBSl
+TBSI 0. 0. 2.25-3 450. 8.25-3 570. 5.825-2 580. +TBS2
+TBS2 ENDT
$----------------------------------------------------------------------------- $
$$ LOADING
LSEQ 1000 910 110
FORCE 110 1I 2000. 0. I. 0.
TLOADI 100 910 0 200
TABLED1 200 +TBDI
+TBD1 0. 0. 1. .05 20. 1. ENDT
$----------------------------------------------------------------------------- $
ENDDATA
10.3.3 Rotation of a Beam with Friction

A uniform beam has a length of 12 inches and a cross-sectional area of 0.15 in2. One end of
the beam is clamped and the other end is attached t o a GAP element as shown in Fig. 10.3.5.
A force of 50 lb acting in the -z direction is applied as a preload providing a friction force at
the end attached to the GAP. In the first phase, the beam is statically loaded by a 1000 in-lb
moment which is applied in the +z direction. In the second phase, the moment is removed
while the applied force remains. The unloading phase was analyzed statically (SOL 66) and
dynamically (SOL 99), and the beam responses were observed.

ix i" L = 12 in
A = 0.15 in2
/
4 I l l l I I l t I
, 1' E = 20 x l o 6 lb/in2
I = 2.813 x in4
P = 50 lb
M = 1000 in-lb

Figure 10.3.5 Uniform Beam with a G a p

The beam is modeled by twelve BEAM elements of equal length with linear elastic material.
Friction at one end of the beam is simulated by a GAP element, connected a t GRID 112 and
212, which is initially closed. The coefficient of static friction (p,) is 0.5, the coefficient of
kinetic friction ( p k ) is 0.3, the closed axial stiffness (li,) is l o 4 lb/in, and the transverse shear
stiffness ( l i t ) is 5.0 x lo3 lb/in (default value). The clamped boundary condition at the end
of the beam is imposed by constraining all six components of GRID 100. The applied load (at
GRID 112) is specified using FORCE, MOMENT and LOAD Bulk Data entries in SOL 66,
while in SOL 99 the Bulk Data entries LSEQ, DLOAD, TLOADl and TABLED1 are used in
addition t o FORCE and MOMENT entries. The solution strategy is provided by the NLPARM
Bulk Data entry in SOL 66 and by the TSTEPNL Bulk Data entry in SOL 99. The input data
for SOL 66 and SOL 99 are given in Table 10.3.3 and 10.3.4, respectively.

The analysis is performed in three subcases. The first subcase applies the preload (force of
50 lb) in 4 increments. The second subcase statically loads the beam with a 1000 in-lb moment
applied in 10 increments, resulting in a total rotation of 115.8' at the end. Notice that in SOL
99 the first and second subcase use PARAM,TSTATIC,l for the static analysis. The same
loading history is specified in the first and second subcase of the SOL 66 and SOL 99 analysis.
The deformed shapes of the beam during static loading are identical for SOL 66 and SOL 99
and are shown in Fig. 10.3.6. The deformed shapes are labeled from 1 t o 6, corresponding to
the tabulated applied moment and Bend, the rotation of the end of the beam about the z-axis.
The first deformed shape does not exhibit translation of the end point because the lateral
force induced by the applied moment is less than the friction force. When the induced lateral
force exceeds the frictional force, the end point starts to move as evidenced by the subsequent
deformed shapes in the figure.

In SOL 66 the third subcase statically unloads the beam by removing the applied moment
while retaining the 50 lb preload. The resulting deformed shapes are shown in Fig. 10.3.7. The
third subcase in SOL 99 uses PARAM,TSTATIC,-1 for dynamic unloading of the beam after
the moment is released, maintaining the preload on the GAP element. Fig. 10.3.8 shows the
rotation of the end of the beam in radians about the z-axis during dynamic unloading. The
angle (Bend)starts from 115.8' and diminishes as the beam vibrates due to friction.

The numerical performance of the second subcase in terms of the number of iterations,
stiffness updates, and bisections is tabulated below for SOL 66 and SOL 99. Notice that the
bisection algorithm was implemented in SOL 66 in Version 67 and was activated in the SOL
66 analysis.

Performance Comparison During Static Loading (Subcase 2)

No.of I
No.of /
No. of
iterations K updates bisections
SOL 66, V67 246 66 2
SOL 99, V67 1533 63 25

The table shows a variation in the efficiency of the two solutions. Different stiffness update
strategies are provided in SOL 66 and SOL 99, which have significant effects on the efficiency
of the iterative process. SOL 99 limits the number of stiffness updates within an increment
and updates in reference to the last converged step except for GAP induced update. SOL 66
has a more liberal stiffness update strategy, allowing numerous intermediate stiffness updates
as required for convergence based on the global divergence criteria in addition t o the GAP
induced updates. This signifies the effects of timely stiffness updates on the efficiency of the
iterative process.
 curve moment (in-lb) Oend
1 100 3.1'
2 200 17.3~
3 400 41.8'
4 600 66.4'
5 800 91.1'
6 1000 115.8'

Figure 10.3.6 Deformed Shapes of Beam During Static Loading (SOL 66 and SOL 99)

curve moment (in-lb) Oend

1 1000 115.8"
2 800 104.3'
3 600 80.1'
4 400 55.9'
5 200 31.6'
6 0 7.2'

Figure 10.3.7 Deformed Shapes of Beam During Static Unloading (SOL 66)
 2 . 0 4 2 . 0 6
T I ME I N SEC

F l XED/ F R E E BEAM. Wl T H A GAP AND A P P L l ED MOMENT AT F R E E END

MOMENT I S 1 0 0 0 I N-LB

Figure 10.3.8 Rotation (0,) at End of Beam During Dynamic Unloading

 Table 10.3.3 SOL 66 Input Data Listing for Beam with Friction
ID FRIBEAM66, V67 $ APR 15-AUG-90
TIME 30 $ CPU MINUTES
SOL 66 $ NONLINEAR STATIC ANALYSIS
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
CEND
TITLE = FIXED/FREEBEAM, WITH A GAP AND APPLIED MOMENT AT FREE END
SUBTITLE = MOMENT IS 1000 IN-LB
ECHO = BOTH
SEALL = ALL
SET 1 = 101 THRU 112
SET 2 = 100
SET 3 = 200
DISP = I
SPCF = 2
STRES = 3
SUBCASE I $ FORCE = 50 LB
NLPARM = 10
LOAD = I 0 1
SUBCASE 2 $ MOMENT = 1000 IN-LB
NLPARM = 20
LOAD = 2 0 1
SUBCASE 3 $ UNLOADING
NLPARM = 30
LOAD = I 0 1
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
MAXI DEFO .001
SET I = ALL EXCEPT GAP
SET 2 = ALL EXCEPT PLOTEL, EXCLUDE GRID POINTS 1000 THRU 1003
VIEW O., O., 0.
AXES MZ , X , Y
FIND SCALE ORIGIN I SET I
PTITLE = TOP VIEW
PLOT STATIC 0 MAXI DEFO .001 SET 1 ORIGIN I
BEGIN BULK
PARAM LGDISP I
$$ SOLUTION CONTROL
NLPARM 10 4 AUTO YES
NLPARM 20 10 AUTO YES
NLPARM 30 10 AUTO YES
$$ GEOMETRY
GRID 100 0. 0. 0.
GRID 101 I. 0. 0.
GRID 102 2. 0. 0.
GRID 103 3. 0. 0.
GRID 104 4. 0. 0.
GRID 105 5. 0. 0.
GRID 106
GRID 107
GRID 108
GRID 109
GRID 110
GRID Ill
GRID 112
GRID 212
$$ CONNECTIVITY
CBEAM I 10 I01 0.
CBEAM 2 10 102 0.
CBEAM 3 10 103 0.
CBEAM 4 10 104 0.
CBEAM 5 10 105 0.
CBEAM 6 10 106 0.
CBEAM 7 10 107 0.
CBEAM 8 10 108 0.
CBEAM 9 10 109 0.
CBEAM 10 10 110 0.
CBEAM 11 10 Ill 0.
CBEAM 12 10 112 0.
CGAP 200 20 112 I.
$$ PROPERTIES
PBEAM 10 II
MAT1 I1 2.E+7
PGAP 20 0.
+PGI .02
$$ LOADING
LOAD 101 I.
LOAD 201 I.
FORCE 100 112
MOMENT 200 112
$$ FRAME FOR PLOT
GRID 1000
GRID 1001
GRID 1002
GRID 1003
PLOTEL 200 1000
PLOTEL 201 I001
PLOTEL 202 1002
PLOTEL 203 1003
$
ENDDATA
 Table 10.3.4 SOL 99 Input Data Listing for Beam with Friction
ID FRIBEAM99, V67 $ APR 15-AUG-90
TIME 90 $ CPU MINUTES
SOL 99 $ NONLINEAR STATIC ANALYSIS
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
CEND
TITLE = FIXED/FREE BEAM, WITH A GAP AND APPLIED MOMENT AT FREE END
SUBTITLE = MOMENT IS 1000 IN-LB
ECHO = BOTH
SEALL = ALL
SET I = 101 THRU 112
SET 2 = 100
SET 3 = 200
DISP = I
SPCF = 2
STRES = 3
$--------------- $
LOADSET = 1000 $
$--------------- $
SUBCASE I $ FORCE = 50 LB
$------------------ $
PARAM TSTATIC I $
$------------------ $
TSTEPNL = 10
DLOAD = 101
SUBCASE 2 $ MOMENT = 1000 IN-LB
$------------------ $
PARAM TSTATIC I $
$------------------ $
TSTEPNL = 20
DLOAD = 201
SUBCASE 3 $ DYNAMIC UNLOADING
$------------------ $
PARAM TSTATIC -1 $
$------------------ $
TSTEPNL = 30
DLOAD = 101
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
MAXI DEFO .001
SET 1 = ALL EXCEPT GAP
SET 2 = ALL EXCEPT PLOTEL, EXCLUDE GRID POINTS 1000 THRU 1003
VIEW O., O., 0.
AXES MZ , X , Y
FIND SCALE ORIGIN I SET I
PTITLE = TOP VIEW
PLOT TRANS 0 TIME 0.0 2.0 MAXI DEFO .001 SET 1 ORIGIN I
OUTPUT(XYPLOT)
 CSCALE 1.3
PLOTTER NAST
XMIN = 2.0
XGRID LINES = YES
YGRID LINES = YES
XTITLE = TIME IN SEC
YTITLE = DISP AT GRID 112 ALONG Y
XYPLOT DISP REsP/II~(T~)
YTITLE = DISP AT GRID 112 ABOUT Z
XYPLOT DISP RESP/II2(R3)
BEGIN BULK
PARAM LGDISP I
$$ SOLUTION CONTROL
TSTEPNL 10 4 .25 ADAPT
+TSI 4
TSTEPNL 20 10 .I ADAPT
+TS2 4
TSTEPNL 30 50 .002 ADAPT
$$ GEOMETRY
GRID 100 0. 0. 0.
GRID 101 I. 0. 0.
GRID 102 2. 0. 0.
GRID 103 3. 0. 0.
GRID 104 4. 0. 0.
GRID 105 5. 0. 0.
GRID 106 6. 0. 0.
GRID 107 7. 0. 0.
GRID 108 8. 0. 0.
GRID 109 9. 0. 0.
GRID 110 10. 0. 0.
GRID Ill 11. 0. 0.
GRID 112 12. 0. 0.
GRID 212 12. 0. -.01
$$ CONNECTIVITY
CBEAM I 10 100 I01 0. 0.
CBEAM 2 10 101 102 0. 0.
CBEAM 3 10 102 103 0. 0.
CBEAM 4 10 103 104 0. 0.
CBEAM 5 10 104 105 0. 0.
CBEAM 6 10 105 106 0. 0.
CBEAM 7 10 106 107 0. 0.
CBEAM 8 10 107 108 0. 0.
CBEAM 9 10 108 109 0. 0.
CBEAM 10 10 109 II0 0. 0.
CBEAM 11 10 110 Ill 0. 0.
CBEAM 12 10 III 112 0. 0.
CGAP 200 20 212 112 I. 0.
$$ PROPERTIES
PBEAM 10 II .I5 2.813-4 2.813-4
MAT1 I1 2. E+7 . 3 7.25-4
PGAP 20 0. 0. I .Et4 5.Et3 .5 .3 +PGI
+PG1 .02
$---------------------------------------------------------------------------- $
$$ LOADING (SUBCASE 1 AND 3)
LSEQ 1000 910 100
FORCE 100 112 50. 0. 0. -1.
TLOADI 101 910 I 1I
TABLED1 I l l
+TBI 0. 0. I. I. 10. I. ENDT
$$ LOADING (SUBCASE 2)
LSEQ 1000 920 200
MOMENT 200 112 1000. 0. 0. I.
DLOAD 201 I. I. 101 I. 2001
TLOADI 2001 920 211
TABLED1 21 1 +TB2
+TB2 0. 0. I. 0. 2. I. ENDT
$---------------------------------------------------------------------------- $
$$ FRAME FOR PLOT
GRID 1000 -2. -2. 123456
GRID 1001 +13. -2. 123456
GRID 1002 +13. 10. 123456
GRID 1003 -2. 10. 123456
PLOTEL 200 1000 1001
PLOTEL 201 1001 1002
PLOTEL 202 1002 1003
PLOTEL 203 1003 1000
$
ENDDATA
Chapter 11

EXAMPLE PROBLEMS

11.1 SNAP-THROUGH ANALYSIS OF A SPHERICAL

DIAPHRAGM BY NEWTON'S ITERATION

11.1.1 Problem Description

A spherical diaphragm resting on a rigid backing plate is subjected to a uniform pressure as

shown in Fig. 11.1.1. It is desired to find the critical pressure loads for snap-through and
snap-back. The diaphragm has a thickness of 0.15 in. and the gap a t the center between the
diaphragm and the backing plate is 0.757 in. Material properties of the diaphragm are given
as follows:
Young's modulus: E = 2.0 x 105psi
Poisson's ratio: v = 0.28
The objective of this analysis is t o study the effects of NLPARM parameters on the con-
vergence characteristics when difficulties arise due to snap-through. The geometry and the
loadings are axisymmetric. For simplicity, it is further assumed that the deformation pattern
is confined to axisymmetric shapes.

Three cases are analyzed in this study according to the boundary conditions of the di-
aphragm edge, i.e.,
Clamped boundary
Simply supported boundary (and release of constraints)
Sliding and simply supported boundary
11.1.2 Finite Element Modeling and MSCINASTRAN Input Data

The main features of this problem are geometric nonlinearities due t o large displacements
and changes in constraints due t o the backing plate. A 10-degree sector of the diaphragm is
modeled using shell elements with axisymmetric boundary conditions. The contact process
between the diaphragm and the backing plate is simulated by gap elements as shown in Fig.
11.1.2. The flange portion is omitted in the model. Various boundary conditions are imposed
on the periphery for the three different cases to be investigated. The typical input data for the
clamped and the simply supported cases are listed in Tables 11.1.1 and 11.1.2, respectively.

Executive Control Section

SOL 66 (in Version 65) is used for the static analysis of the snap-through phenomena.

Case Control Section

Subcases are used t o change the loading path and/or boundary conditions. Within each
subcase, the loads are subdivided into equal increments and each load increment requires
iterations for convergence. As shown in Tables 11.1.1 and 11.1.2, Case Control data include
loading and unloading phases. In the case of the simply supported boundary in Table 11.1.2,
Subcase 7 is supplemented t o remove the constraints at the periphery after full unloading.

Bulk Data Section

A 10-degree sector of the diaphragm is modeled by one TRIA3 and six QUAD4 elements with
linear elastic material. The contact process between the diaphragm and the backing plate is
simulated by 13 GAP elements as shown in Fig. 11.1.2. A total of 28 GRID points are used
in the model.

The geometry and the output data are described in the cylindrical coordinate system. The
axisymmetric boundary condition is imposed by constraining components 2, 4, and 6 of the
grid points on the planes of symmetry. The apex point is allowed t o move only in the axis
of symmetry. Various boundary conditions are applied t o the GRID points on the periphery
(GRIDs 170 and 171) as follows:
Clamped case: all components (1-6) fixed
Simply supported case: components 1, 2, 3, 4, and 6 fixed

r Sliding and simply supported case: components 2, 3, 4, and 6 fixed

All grid points (GRIDs 200 - 261) simulating the backing plate are fixed. The number of total
DOFs of the problem is 168, and the number of active DOFs is 37 for the clamped case.
 Effects of geometric nonlinearity are included in the analysis by specifying PARAM, LGDISP,
+I. The external pressure applied t o the diaphragm surface in each subcase is specified using
PLOAD2. Smaller load increments are used around the loads where snap-through and snap-
back are anticipated to occur. The solution strategy is provided by the NLPARM Bulk Data,
which are varied from subcase t o subcase depending on the convergence characteristics.
 SPHERICAL DIAPHRAGM
AFTER SNAP-THROUGH 7 (UNDEFORMED)

I'

RIGID BACKING PLATE

Side View

Isometric View

Figure 11.1.1. Spherical Diaphragm Sujected to a Uniform Pressure

(a) Side View (b) Isometric View
 BOUNDARY CONDITIONS

Case 1
m I

Case 2

Figure 11.1.2. Spherical Diaphragm Model with Axisymmetric

Boundary Condition and Constraints by Gap Elements
 Table 11.1.1 Input Data for the Clamped Case

ID SNAPC,V66 $ SHL 5/2/85, GBL 11/22/86

TIME 30
SOL 66
DIAG 8,50
CEND
TITLE = SNAP-THROUGH ANALYSIS OF A SPHERICAL DOME
SUBTI = EDGE CLAMPED
ECHO = UNSORT
SEALL = ALL
SPC = 200
SET 1 = 100 110 120 130 140 150 160
SET 2 = 1 THRU 7
SET 3 = 170,200
DISPL = 1
SPCFO = 3
SUBCASE 1
SUBTI = 0.90 PSI
NLPAR = 1
LOAD = 90
SUBCASE 2
SUBTI = 1.0 PSI
NLPAR = 2
LOAD = 100
SUBCASE 3
SUBTI = 1.5 PSI
NLPAR = 3
LOAD = 150
SUBCASE 4
SUBTI = 0.6 PSI
NLPAR = 4
LOAD = 60
SUBCASE 5
SUBTI = 0.35 PSI
NLPAR = 5
LOAD = 35
SUBCASE 6
SUBTI = 0.0 PSI
NLPAR = 6
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
PAPER SIZE 29. X 21.
SET 1 ALL
SET 2 ALL EXCEPT GAP
MAXI DEFO ,001
FIND SCALE ORIGIN 1 SET 1
PTITLE=ISOMETRIC VIEW
PLOT SET 1, ORIGIN 1, LABEL
VIEW 0. , 0. , 0.
AXES MY , X , Z
FIND SCALE ORIGIN 1 SET 2
 PTITLE-ORTHOGRAPHIC VIEW
PLOT STATIC 0 M A X I DEFO .001 SET 2 SYMBOL 1
BEGIN BULK
$ PARAMETERS
PARAM K6ROT 1.0
PARAM LGDISP 1
NLPARM 1 6 AUTO YES +NL1
NLPARM 2 5 SEMI 2 YES +NL2
+NL2 5
NLPARM 3 2 AUTO YES +NL3
+NL3 5
NLPARM 4 3 AUTO YES +NL4
NLPARM 5 5 AUTO YES +NL5
+NL5 5
NLPARM 6 1 AUTO YES +NL6
$ BOUNDARY CONDITIONS
SPCl 200 12456 100
SPCl 200 246 110 120 130 140 150 160
SPCl 200 246 111 121 131 141 151 161
SPCl 200 123456 170 171
$ LOADING
PLOAD2 35 .35 1 THRU 7
PLOAD2 60 .60 1 THRU 7
PLOAD2 90 .90 1 THRU 7
PLOAD2 100 1.0 1 THRU 7
PLOAD2 150 1 . 5 1 THRU 7
$
$ GEOMETRY
$
CORD2C 1 0.0 0.0 0.0 0.0 0.0 1.0 +C 1
+C1 1.0 0.0 0.0
$ GRID POINTS
GRID 100 1 0.0 0.0 .503 1 6
GRID 110 1 1.082 0.0 .486 1 6
GRID 111 1 1.082 10. .486 1 6
GRID 120 1 2.163 0.0 .445 1 6
GRID 121 1 2.163 10. .445 1 6
GRID 130 1 3.243 0.0 .393 1 6
GRID 131 1 3.243 10. .393 1 6
GRID 140 1 4.322 0.0 .315 1 6
GRID 141 1 4.322 10. .315 1 6
GRID 150 1 5.401 0.0 .238 1 6
GRID 151 1 5.401 10. .238 1 6
GRID 160 1 6.479 0.0 .I53 1 6
GRID 161 1 6.479 10. .I53 1 6
GRID 170 1 7.557 0.0 .064 1 6
GRID 171 1 7.557 10. .064 1 6
$ ELEMENTS
CTRIA3 1 1 100 111 110
CqUAD4 2 1 110 111 121 120
CqUAD4 3 1 120 121 131 130
CqUAD4 4 1 130 131 141 140
CqUAD4 5 1 140 141 151 150
CqUAD4 6 1 150 151 161 160
CqUAD4 7 1 160 161 171 170
$ BACKUP O F G A P S
GRID 200 1 0.0 0.0 -.254 1
GRID 210 1 1.082 0.0 -.2533 1
GRID 211 1 1.082 10. -.2533 1
GRID 220 1 2.163 0.0 -.2354 1
GRID 221 1 2.163 10. -.2354 1
GRID 230 1 3.243 0.0 -.I862 1
GRID 231 1 3.243 10. -.I862 1
GRID 240 1 4.322 0.0 -.I062 1
GRID 241 1 4.322 10. -. 1062 1
GRID 250 1 5.401 0.0 -.1800-21
GRID 251 1 5.401 10. -. 1800-21
GRID 260 1 6.479 0.0 .0697 1
GRID 261 1 6.479 10. .0697 1
$ G A P S T O R E P R E S E N T BACKING P L A T E
CGAP 100 200 200 100 1.
CGAP 110 210 210 110 1.
CGAP 111 210 211 11 1 1.
CGAP 120 220 220 120 1.
CGAP 121 220 221 121 1.
CGAP 130 230 230 130 1.
CGAP 131 230 23 1 131 1.
CGAP 140 240 240 140 1.
CGAP 141 240 241 141 1.
CGAP 150 250 250 150 1.
CGAP 151 250 251 151 1.
CGAP 160 260 260 160 1.
CGAP 161 260 26 1 161 1.
$ PROPERTIES
MAT1 1 2.+5 .28 7.835-9
PSHELL 1 1 0.15 1 1.2 1
PGAP 200 .7570 1.+6 1.-6
PGAP 210 .7393 1.+6 1.-6
PGAP 211 .7393 1.+6 1.-6
PGAP 220 .6804 1.+6 1.-6
PGAP 221 .6804 1.+6 1.-6
PGAP 230 .5792 1.+6 1.-6
PGAP 231 .5792 1.+6 1.-6
PGAP 240 .4212 1.+6 1.-6
PGAP 241 .4212 1.+6 1.-6
PGAP 250 .2398 1.+6 1.-6
PGAP 251 .2398 1.+6 1.-6
PGAP 260 .0833 1.+6 1.-6
PGAP 261 .0833 1.+6 1.-6
$ P L O T DATA
PLOTEL 11 200 210
PLOTEL 12 210 220
PLOTEL 13 220 230
PLOTEL 14 230 240
PLOTEL 15 240 250
PLOTEL 16 250 260
PLOTEL 17 260 170
ENDDATA
 Table 11.1.2 Partial Listing of the Input Data for the Simply
Supported Case

I D SNAPS,V65 $ SHL 5/2/85, GEL 12/10/86

TIME 30
SOL 66
DIAG 8,50
CEND
TITLE = SNAP-THROUGH ANALYSIS OF A SPHERICAL DOME
SUBTI = EDGE SIMPLY SUPPORTED
SEALL = ALL
ECHO = UNSORT
SPC = 200
SET 1 = 100 110 120 130 140 150 160
SET 2 = 1 THRU 7
SET 3 = 170,200
DISPL = 1
SPCFO = 3
OLOAD = 3
SUBCASE 1
SUBTI = 0 . 5 PSI
NLPAR = 1
LOAD = 50
SUBCASE 2
SUBTI = 0 . 6 PSI, first snap-through
NLPAR = 2
LOAD = 60
SUBCASE 3
SUBTI = 0.7 PSI
NLPAR = 3
LOAD = 70
SUBCASE 4
SUBTI = 0.72 P S I , second snap-through
NLPAR = 4
LOAD = 72
SUBCASE 5
SUBTI = 0 . 8 PSI
NLPAR = 5
LOAD = 80
SUBCASE 6
SUBTI = 0 . 0 PSI, unloading
NLPAR = 6
SUBCASE 7
SUBTI = 0 . 0 PSI (EDGE CONSTRAINT REMOVED)
SET 5 = 100 110 120 130 140 150 160 170
DISPL = 5
SPC = 300
NLPAR = 7
OUTPUT (PLOT)
CSCALE 1 . 3
PLOTTER NAST
PAPER SIZE 29. X 21.
 SET 1 ALL
SET 2 ALL EXCEPT GAP
M A X I DEFO .001
FIND SCALE O R I G I N 1 SET 1
PTITLE=ISOMETRIC VIEW
PLOT SET 1 , ORIGIN 1, LABEL
VIEW 0 . , 0 . , 0 .
AXES MY , X , Z
FIND SCALE O R I G I N 1 SET 2
PTITLE=ORTHOGRAPHIC VIEW
PLOT STATIC 0 M A X I DEFO .001 SET 2 SYMBOL 1
BEGIN BULK
$ PARAMETERS
PARAM K6ROT 1.0
PARAM LGDISP 1
NLPARM 1 5 AUTO YES +NL1
NLPARM 2 5 AUTO 2 YES +NL2
+NL2 7 2
NLPARM 3 4 AUTO YES +NL3
NLPARM 4 4 SEMI 2 YES +NL4
+NL4 5 10
NLPARM 5 4 AUTO YES +NL5
NLPARM 6 8 AUTO YES +NL6
NLPARM 7 10 AUTO YES +NL7
$ BOUNDARY CONDITIONS
SPCl 200 12456 100
SPCl 200 246 110 120 130 140 150 160
SPCl 200 246 111 121 131 141 151 161
SPCl 200 12346 170 171
SPCl 300 12456 100
SPCl 300 246 110 120 130 140 150 160
SPCl 300 246 111 121 131 141 151 161
SPCl 300 2346 170 171
$ LOADING
PLOAD2 50 .50 1 THRU 7
PLOAD2 60 .60 1 THRU 7
PLOAD2 70 .70 1 THRU 7
PLOAD2 72 .72 1 THRU 7
PLOAD2 80 .80 1 THRU 7
$
$ GEOMETRY
$
*** The r e s t i s identical t o Table 1 1 . 1 . 1 ***
11.1.3 Analysis Results

Clamped Case

The load-deflection curve at the center point is shown in Fig. 11.1.3. The curve is constructed
by combining loading and unloading paths. The snap-through occurs between 0.96 psi and
0.98 psi, and the snap-back between 0.45 psi and 0.40 psi. Since the center point comes into
contact with the backing plate at p = 0.98 psi, the point does not translate beyond the gap
distance (0.757 in.) despite the increasing load. The load history between two points A and
B, where the snap-through and the snap-back begin t o occur, is conjectured as shown by the
dashed line in Fig. 11.1.3.

Fig. 11.1.4 shows the deformed shapes of the diaphragm for various loading and unloading
levels. It is shown that the curvatures of the diaphragm surface are reversed upon snap-through
or snap-back at the critical loads.

Simply Supported Case

In the simply supported case, snap-through occurs twice as shown in Fig. 11.1.5. The first
snap-through occurs between 0.54 psi and 0.56 psi, and the second occurs between 0.700 psi
and 0.705 psi. If it is unloaded before the second snap-through, the diaphragm snaps back at
a pressure between 0.50 psi and 0.48 psi. However, the snap-back does not occur when the
pressure is removed after the second snap-through as shown.

As shown in Fig. 11.1.6(a), two steps of snap-through assume distinct shapes. The first
snap-through can be considered as an interim snap-through by flipping into a shape of double
curvature. Closer examination of the analysis results reveals that the diaphragm comes into
contact with the backing plate near the periphery after the first snap-through. In fact, such
contacts stabilized the interim snap-through mode, which would not have existed otherwise.
The central portion of the diaphragm comes into contact with the backing plate in addition t o
the peripheral area upon the second snap-through.

The central area remains in contact with the backing plate during unloading until the
pressure is reduced t o 0.1 psi. When the pressure is removed entirely, the central area is lifted
off, leaving the peripheral area of the diaphragm (point A in Fig. 11.1.6(b)) in contact with
the backing plate. Based on these observations, the load history between points A and B can
be conjectured by the dashed line in Fig. 11.1.5.

It is interesting that the snap-back does not occur upon unloading in the elastic problem.
This is because the spring-back force is trapped by SPC forces a t the periphery. This SPC
force (Q) upon full unloading is computed as 19.9 lbs on each node a t the periphery, which
is equivalent t o a reaction force of 15.088 lbs/in. in the radial direction. It is attempted t o
release this reaction force by removing radial constraints on the periphery. The SPC force of
19.9 lbs is released in ten steps by Subcase 7. As shown in Fig. 11.1.6(b), the diaphragm snaps
back as the reaction force is released. When the reaction force is removed entirely at the end,
the diaphragm resumes the original shape as shown in the figure.

Fig. 11.1.7 shows the vertical displacement at the center and the radial displacement at
the periphery as functions of the SPC force. The snap-back occurs when the SPC force is
between 15.9 lbs and 13.9 lbs. Upon snap-back, the central displacement overshoots due to
the radial compression force remaining at the periphery. The periphery also slides back and
forth radially as the SPC force is diminished.

Sliding and Simply Supported Case

As the pressure is increased, the diaphragm undergoes a slight snap-through at a pressure

between 0.188 psi and 0.190 psi. Fig. 11.1.8 shows the load-deflection curve at the center
point. As shown, the snap-back occurs at a pressure between 0.184 psi and 0.182 psi.

11.1.4 Solution Progression

The convergence of the solution is the most important issue for the iterative and incremental
solution processes of nonlinear problems. The convergence characteristics are mainly influ-
enced by the increment size and the iteration strategy. These crucial data for controlling the
incremental and iterative processes are specified in the Bulk Data NLPARM.

The general strategy is t o start with default options and reasonable increments of the ap-
plied load. Nonlinear analysis requires restarts in general. If the solution has not converged at
a certain step, restarts are attempted from that step with different options in the NLPARM.
Major options that can be changed for better convergence are increment size and parameters
KMETHOD, MAXDIV, and MAXLS. It is recommended that the value of MAXDIV be in-
creased (5 to 7) for the subcase where snap-through or snap-back is anticipated. The increment
size was also reduced in the vicinity of the snap-through.

It was discovered in this study that the effect of the line search process is rather erratic
and unpredictable around the critical loads. The line search was helpful t o jump over the
unstable region in most cases, but on some occasions, the solution was more easily converged
by suppressing the line search. This behavior was scrutinized and the following observations
were contemplated:

r The convergence characteristics around the critical load (snap-through or snap-back) are
very sensitive t o the timing of the stiffness matrix update.
r The timing and the line search affect the base vector (displacements) for the matrix
update.
 A slight difference in the base vector for the update could lead the solution to either side
of divergence or convergence in the vicinity of the critical loads without due explanation.

Clamped Case

Table 11.1.3 shows the solution strategy and the performance of each subcase when the solu-
tions are obtained successfully. Subcases 1, 2, and 3 represent the loading phase, and Subcases
4, 5, and 6 represent the unloading phase. The snap-through occurs in Subcase 2, and the
snap-back occurs in Subcase 5. The solution has not converged with default options in the
vicinity of the critical loads. Restart runs were made with smaller load increments and differ-
ent iteration strategies until converged solutions were achieved. It is noted that more stiffness
updates, iterations, and/or line searches were required in these subcases t o obtain the solution
than in other subcases without an unstable region. In subcases without an unstable region,
solutions were easily achieved by default options.

Table 11.1.4 summarizes trial-and-errors attempted for Subcase 21increment 4 (p = 0.96 psi
- 0.98 psi), in which snap-through occurs. In the first trial, the program detects a negative term
on the factor diagonal of [K], and the problem stops due t o TWO DIVERGING SOLUTIONS
after 11 iterations with 3 stiffness updates. In the second and third trials, the problem does not
converge due t o the same causes. In the fourth trial, the problem stops before convergence due
to the maximum iteration limit of 20. Finally, when KMETHOD is set to SEMI, MAXDIV
t o 5, and KSTEP t o 2, the problem converges after 16 iterations with 6 stiffness updates. It
can be noticed that frequent stiffness updates at converged states (due t o KSTEP=2) expedite
convergence.
Table 11.1.3 Solution Strategy and Performance for Clamped Case

Subcase Load No. of Solution No. of No. of No. of

ID (psi) INC Strategy K Update Iter L.S.
1 0.90 6 Default 1 29 0
2* 1.OO 5 SEMI, KSTEP=2 12 29 41
MAXDIV= 5
3 1.5 2 MAXDIV= 5 1 11 16
4 0.60 3 Default 2 18 I
5* 0.35 5 MAXDIV= 5 4 35 16
6 0.00 1 Default 0 6 0
Note: Snap occurs at subcases denoted by *
 Table 11.1.4 Effects of NLPARM Parameters on Convergence
for Subcase 2/Increment 4 of Clamped Case

Trial Parameters No. of No. of No. of

Case KMETHOD KSTEP MAXDIV MAXLS K Update Iter L.S.
1 (AUTO) (7) (2) (5) 3 11 36
2 (AUTO) (7) 5 (5) 5 19 48
3 SEMI (7) 5 (5) 2 6 6
4 SEMI (7) 7 (5) 5 20 45
5 SEMI 2 5 (5) 6 16 41
) denotes default option
&
Note: 2. uccessful solution is achieved at the final trial

Simply Supported Case

Table 11.1.5 shows the solution strategy and the performance of each subcase upon successful
solutions. Subcases 1 through 5 represent the loading phase and subcase 6 represents the
unloading phase. After the pressure is removed entirely, the radial constraint of the diaphragm
is released, which is analyzed by Subcase 7. Another set, Subcase 3, 4, and 5 denoted by +,
analyzed the unloading path from 0.60 psi after the first snap-through occurs. The solution
requires strategies different from the default options for NLPARM in the vicinity of the critical
loads.

Tables 11.1.6 and 11.1.7 show the trial-and-errors t o obtain solutions for the first (Subcase
2/increment 3) and the second snap-through (Subcase 4/increment I), respectively. The first
3 trials in Table 11.1.6 were terminated due t o TWO DIVERGING SOLUTIONS.

Table 11.1.7 shows two trials terminated due t o the maximum iteration limit of 20, where
the program detects a negative term on the factor diagonal of [K] as well as the GIMBAL
ANGLE exceeding 90 degrees. It is noted that variations of MAXDIV and MAXLS have
significant effects on the performance while jumping over the unstable region for the snap-
through analysis.
 Table 11.1.5 Solution Strategy and Performance for Simply Supported Case

Subcase Load No. of Solution No. of No. of No. of

ID (psi) INC Strategy K Update Iter L.S.
1 0.50 5 Default 1 24 0
2* 0.60 5 KSTEP=2 7 25 8
MAXDIV=7
MAXLS=2
3 0.70 4 Default 0 17 0
4* 0.72 4 SEMI KSTEP=2 10 25 87
MA~DIV=~
MAXLS=10
5 0.80 4 Default 0 8 0
6 0.00 8 Default 2 30 3
7* 0.00 10 Default 7 75 18
3+ 0.50 4 Default 0 12 0
4+* 0.40 5 KSTEP=2 8 27 11
MAXDIV=7
5+ 0.00 5 Default 1 24 0
Notes: 1. Snap occurs at subcases denoted by *
2. +denotes restart for snap-back during unloading

Table 11.1.6 Effects of NLPARM Parameters on Convergence

for Subcase 21Increment 3 of Simply Supported Case

Trial Parameters No. of No. of No. of

Case KMETHOD KSTEP MAXDIV MAXLS K Update Iter L.S.
1 (AUTO) (7) (2) (5) 1 5 5
2 (AUTO) 2 5 (5) 2 8 24
3 (AUTO) 2 5 2 2 8 12
4 (AUTO) 2 7 2 3 11 8
) denotes default option
k
Notes: 2. uccessful solution is achieved at the final trial
 Table 11.1.7 Effects of NLPARM Parameters on Convergence
for Subcase 4/Increment 1 of Simply Supported Case

Trial - Parameters No. of No. of No. of

Case KMETHOD KSTEP MAXDIV MAXLS K Update Iter L.S.
1 (AUTO) (7) (2) (5) 3 12 12
2 (AUTO) 2 7 2 6 20 20
3 (AUTO) 2 7 0 6 20 0
4 SEMI 2 5 10 7 19 87

4
) denotes default option
Note: 2. uccessful solution is achieved at the final trial

Sliding and Simply Supported Case

As shown in Table 11.1.8, nine subcases are used for this analysis. Subcases 1 through 5
represent the loading phase, and Subcases 6 through 9 the unloading phase. The snap-through
occurs in Subcase 3, and the snap-back occurs in Subcase 8. Since the snap-through and the
snap-back in this case are moderate and shallow as shown in Fig. 11.1.8, the load is incremented
in very small steps around the critical load t o obtain a more descriptive load-deflection curve.
In fact, the coarser load increment rendered solutions by default options without giving details
of snap-through around the critical load.
Table 11.1.8 Solution Strategy and Performance for
Sliding and Simply Supported Case

Subcase Load No. of Solution No. of No. of No. of

ID (psi) INC Strategy K Update Iter L.S.
1 0.16 4 Default 1 24 0
2 0.18 1 Default 0 5 0
3* 0.19 5 MAXDIV=5 4 32 18
4 0.20 1 Default 0 4 0
5 0.60 4 Default 2 25 1
6 0.35 1 Default 0 6 0
7 0.19 1 Default 1 9 2
8* 0.18 5 Default 3 29 14
9 0.00 2 Default 2 8 16
Note: Snap occurs a t subcases denoted by *

11.1 - 16
11.1.5 Concluding Remarks

While default options in NLPARM are effective as a general strategy for ordinary problems,
some variations are required to overcome difficulties if divergence is encountered. It is discov-
ered that such variations are required to jump over the unstable region of the snap-through
and snap-back problems.
In general, more conservative methods are more reliable but also expensive. Helpful varia-
tions for snap-through problems are

SEMI rather than AUTO

Smaller value for KSTEP to allow more frequent stiffness updates at convergence
Larger value for MAXDIV to bypass the divergence criteria
Larger value for MAXLS to exercise more line searches

It is also observed that the solutions to the snap-back are easier t o achieve than to the snap-
through.
 ...!

..I x Loading
i @ Unloading A
...I @
i - - - Conjectured
."!
i I

1.00
i
A $
....
I
I g-' \
0
x \
\
..I \
Load \
i 5 \
(psi) ...I
. .:
\

.I
:.:
.
:
.
* .
:
.
. ..
\
\
\
\
\ @
@

*
: :
0.50...!. . 1 0
.
\ \

I \
0'
...I 0 - - M

B
i0
...,i
:
r
i i
...j ;
;*
i

..
i

."I.
!:

0.00 &."---"i" -7-.-.i.-7. ".-j" -7""-r" .-.. ..i"-. . -

TT.-.i"i"".?'-.-'i".".".?'" ....?'?'"."" . . . ~ . ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
0.00 0.25 0.5 0.75 1
Displacement (in)

Figure 11.1.3 Load vs. Vertical Displacement at the Center

(Clamped Case).
 'i UNDEFORMED
p = 0.75 p s i
I

FULLY UNLOADED

I / p = 0.45 p s i

Figure 11.1.4. Deformed Shapes of the Diaphragm (Clamped Case)

(a) Loading (b) Unloading
 1-00 . ......-.-.--.- "...."..""."""...--.-----.--.---.-....- .......-...... ....... ...................... .... ...........
..., j
i
....
i
...I x Loading
i
@ Unloading
...I

i - - - Conjectured 9
0.75 .-I
A
.": ?(X-,
j
...,j >i: \
\

x \
\ 0
j
\
\
\
i
Load I \ j
\ 0 i
(psi) \
\
\
...I

i :
\ 6 j

...:: ...
I
\
i
...: p
\

i
\
\
0 i
:. . \
0.25 .Ao \ j
: .
."!? x: \ 0 j
@
..
:.
:.
\ i
...% \ \0
@ f~
... 8.
\;
0 .oo ...'! .-------.------....------
' ! ! ! ! i ! i i i
....--...---. "@
i
.........
? ! ? ?
....
?
.................................
; ! I :
i
0.00 0.25 0.5 0.75 1.00
Displacement (in)

Figure 11.1.5 Load vs. Vertical Displacement at the Center

(Simply Supported Case).
 -----_-- - - ------_

BACKINGPLATE
-p = 0.72 psi

(a)

----------
' - - - - - - - - - - 0

-----
FINAL POSITION
I

/
I Q = 11.9 lbs
/FULLY UNLOADED ( p = 0 . 0 psi. Q = 19.9 1 b s )

Figure 11.1.6. Deformed Shapes of the Diaphragm (Simply Supported

Case) (a) Loading (b) Edge Constraint Removed after Unloading
 0.40

, . .. .
. I
.,.,... 3~""" I
I
0.20 - *....,..*x"'"-' I
+......-./K'.." I
....*' ...>c.""' I
I
.....',*.x...""'
,.-' &,,.~."-"
0.oo 49-------@------@------@-------@
------- --Q
I
!@"
I
Q..---- +,
-0.-7
I
I
I
Displacement
-0.20 - I
(in) II
I
x Vertical Displacement (Center) I
@ Radial Displacement (Boundary)
-0.40 - - - - Conjectured I
I
I
I
I
-0.60 - I

k......,,'........
,

"'X....,,,

-0.80 I I I

0 5 10 15 20
SPC Force (Q)

Figure 11.1.7 Vertical and Radial Displacements at the Center and

Boundary Points vs. SPC Forces at the Periphery.
Load
(psi)

Displacement (in)

Figure 11.1.8 Load vs. Vertical Displacement at the Center

(Sliding and Simply Supported Case).
11.2 PARAMETRIC STUDY ON BFGS UPDATES
AND LINE SEARCH METHOD IN SOL 66

11.2.1 Introduction

The quasi-Newton update and the line search methods are employed as options to accelerate
the convergence and improve the overall effectiveness of the modified Newton's iteration. When
the BFGS method was implemented and tested in MSC/NASTRAN, our experience showed
that the implementation scheme and adjustment of parameters had a significant impact on the
program's performance [11.1].The line search method is an expedient, but not an expeditious
method that could improve the computational efficiency. The line search method is modified
and implemented t o optimize its usefulness for a general class of problems in conjunction with
quasi-Newton updates [11.2].

Parametric studies are performed to validate the effectiveness and efficiency of the algo-
rithm. The study shows that the line search method is essential t o achieve convergence in some
highly nonlinear problems, but has little impact on the efficiency.

Various problems are examined to verify the algorithm's ability t o handle nonlinear prob-
lems involving softeninglhardening and buckling behavior with materialJgeornetric nonlin-
earities. The effectiveness of the solution process is studied for different values of the fields
MAXQN, MAXLS, and LSTOL in the NLPARM Bulk Data entry. Three example problems
are presented t o illustrate the convergence and efficiency characteristics of the quasi-Newton
update and the line search process. The analyses are performed using MSCINASTRAN Ver-
sion 65. The iteration strategy is based on the hybrid Newton's method, by which the stiffness
is updated only when it is necessary for efficiency and effectiveness [11.3]. Two cases are con-
sidered for each problem. In the first case, fields MAXLS and LSTOL are kept constant (equal
t o 5 and 0.5, respectively, unless otherwise stated), and field MAXQN is varied. In the second
case, field MAXQN is kept constant (equal t o 30), and fields MAXLS and LSTOL are varied.

11.2.2 The Cologne Challenge: a Z-Shaped Cantilever Beam

A Z-shaped cantilever beam as shown in Fig. 11.2.1 was proposed as a benchmark problem for
testing nonlinear capabilities of FEM codes [11.4]. The beam, when subjected t o the vertical
load at the end, undergoes very large displacement and plastic deformation with von Mises
yield criterion and isotropic hardening [11.5]. Two different models are used in the present
analysis: a plate element model (ZPLATE) and a beam element model (ZBEAM). The analysis
is difficult because of the highly nonlinear behavior resulting from the large rotation a t the
plastic hinge as shown in Fig. 11.2.2. The load-deflection curve a t the end point is shown in
Fig. 11.2.3. Notice that the nonlinearity of the system in the loading phase of 0-1000 N is
much more severe than that of 1000-2000 N. This makes the problem suitable for examining
the performance of nonlinear iteration algorithms in both highly and slightly nonlinear regions.
 plate thickness: 3.4 mm
plate width : 20 mm
tip load : 2000 N
Figure 11.2.1. Cantilevered 2-Shaped Beam (QUAD4 Model).

SHAPE

Figure 11.2.2. Deformed Shapes (QUAD4 model).

11.2 - 2
 I700
--- ZPLATE

Figure 11.2.3. Load-Deflection Curve.

Case 1: Tables 11.2.l(a) and 11.2.2(a) summarize the numerical performance of the BFGS
updates for ZPLATE and ZBEAM, respectively, by varying MAXQN from 0 to 30 with an
interval of 5. The number of QN updates reflects the cumulative QN updates counting all the
quasi-Newton vector pairs involved in every iteration. By increasing the value of MAXQN,
more QN updates are performed. When the QN update is suppressed (MAXQN=O), the beam
model fails to converge, whereas the plate model requires an excessive number of line searches,
stiffness updates and iterations for convergence. In terms of efficiency, both models seemed to
show most favorable results with a MAXQN of 20.
 Table 11.2.l(a). Numerical Performance as a function of MAXQN in ZPLATE

n LOAD I
COMPUTATION MAxQN '
I
H11 0 - 1000
(N) EFFICIENCY
I
No. of QN. update
. s
0 5 10
0 4191 4255 4995+
1 No. of line searches 1 503 327 260
15

256
20
4526
259
25
4476
252
30
4910
258
11 1130increments) I No.ofiterations i 1412 1326 1196 1195 1212 1158 1196'
severe nonlinearity No. of I< updates 225 148 147 140 138 141 140 '
No. of QN updates 0 2329 1462 2288 3882 4971 5537
1000 - 2000 No. of line searches 8 0 0 I 1 5 9 I
9
1
I (100 increments)
--
No. of iterations 1 578 1 565 / 433 1
mild nonlinearitv I No. of K u ~ d a t e s 1
total CPU time*, normalized
I
29 1
I
191 401
436
23
1 1.303 1 1.079 1 1.034 1 1.031 1
463
15
1.000 1
507
12
1.013
I
1
514
14
1.023
Notes:
* VAX 11/780 with VAX/VMS V4.5, not accurate as performance index
+
15 QN vectors dragging at 870N for 21 iterations

Table 11.2.2(a). Numerical Performance as a function of MAXQN in ZBEAM

Notes:
* VAX 111780 with VAX/VMS V4.5, not accurate as performance index
+ Diverging solution at load step 100-200 N.
- Reaching the iteration limit of 30 at load step 300-400 N.
Case 2: Tables 11.2.l(b) and 11.2.2(b) summarize the numerical performance of the line
search algorithm for ZPLATE and ZBEAM, respectively, by varying MAXLS from 0 to 6 with
LSTOL set at 0.5. It is clear from the tables that the effectiveness of the line search varies
with the extent of nonlinearity. Extensive line searches are performed in the highly nonlinear
region: the number of line searches over the number of iterations is around 22% in the plate
model and 29-33% in the beam model. In the slightly nonlinear region, the line search is hardly
required: the ratio of the number of line searches to iterations is around 1%in the plate model
and zero in the beam model. By increasing the value of MAXLS, more line search operations
are performed in the highly nonlinear region with more favorable results for MAXLS of 4 and
5. However, little difference is observed in the efficiency for RlAXLS between 3 and 6. In
the less nonlinear region on the other hand, MAXLS of 2 or 3 shotvs more efficient results as
illustrated in Table 11.2.l(b). When the line search is suppressed (MAXLS=O), both the plate
and beam models fail to converge at the onset of plastic deformation around the load of 200 N.
Such failures illustrate the necessity of combining the line search operation in the nonlinear
iteration algorithm.
Table 11.2.l(b). Numerical Performance as a function of MAXLS in ZPLATE

Notes:
* VAX 11/780 with VAX/VMS V4.5 (not accurate as performance index)
+ Diverging solution a t load step 200-300 N.

Table 11.2.2(b). Numerical Performance as a function of MAXLS in ZBEAM

Notes:
* VAX 111780 with VAXIVMS V4.5 (not accurate as performance index)
+ Diverging solution at load step 100-200 N.
- Reaching the iteration limit of 30 at load step 500-600 N.

11.2.3 Snap-Through and Snap-Back: a Spherical Shell with Backing Plate

A spherical diaphragm resting on the rigid backing plate is subjected t o a uniform pressure as
shown in Fig. 11.2.4. The analysis objective is t o find the critical pressure for snap-through and
snap-back under different boundary conditions: clamped, simply supported, and sliding/simply
supported cases. Additional constraints are applied after the snap-through by the backing
plate. The material is linear elastic, and the deformation pattern is assumed to be axisymmetric
because the diaphragm is connected t o a plunger which prevents the apex from rotating.

A 10-degree sector of the diaphragm is modeled using shell elements with axisymmetric
boundary conditions. The contact process between the diaphragm and the backing plate is
simulated by gap elements as shown in Fig. 11.2.5. The flange portion is omitted in the model.
The main features of these problems are geometric nonlinearities due t o large displacements,
follower forces, and changes in constraints due to the backing plate. The convergence charac-
teristics are influenced by the load increment and the iteration strategy. The increment size is
reduced in the vicinity of the snap-through and snap-back. The convergence behavior is rather
erratic and unpredictable around the critical loads.

SPHERICAL DIAPHRAGM
AFTER SNAP-THROUGH ---7 (UNDEFORMED)

0'757 in 'f I L RIGID BACKING P U T E

Side View

Isometric View

Figure 11.2.4. Spherical Diaphragm Subjected to a Uniform Pressure.

 BOUNDARY CONDITIONS

Case 1
n
8-3
8

Case 2

Case 3 g-3
GAP ELEMENT (1 3)

A SECTION OF A
SPHERICAL SHELL

RIGID BACKING PLATE 171

Figure 11.2.5. Spherical Diaphragm Model with Axisymmetric Boundary

Condition and Constraints by Gap Elements.
i) Clainped Case
The load-deflection curve a t the apex point for the clamped boundary conditions, con-
structed by combining loading and unloading paths, is shown in Fig. 11.2.6. The snap-through
occurs between 0.96-0.98 psi, and the snap-back between 0.45-0.40 psi. Since the apex point
comes into contact with the backing plate at 0.98 psi, the point does not translate beyond the
gap distance (0.757 in.) despite the increasing pressure. The load history between two points
A and B where the snap-through and the snap-back begin t o occur is conjectured as shown by
the dashed line.

-
-
-
--
w

+ Loading
- 0 Unloading C)
I
--- a Conjectured
-
I

A
\
-.
\
\
\ <)
\
\
Load \
(Psi) \
\
\
\

..
\
'\
\

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

Displacement (in)

Figure 11.2.6. Load vs. Vertical Displacement at the Center (Clamped Case).
 Case 1: The parametric study is performed by varying MAXQN from 0 to 30 with an
interval of 5. The results are summarized in Table 11.2.3(a) for loading and unloading phases
separately. Convergence near the snap-through point occurs only at a MAXQN of 10 and
20, which indicated the erratic nature of the problem. The unloading phase is analyzed by
restarting from the converged solution with a MAXQN of 20. The fact that the unloading is
easier than the loading for convergence is intriguing.

Table 11.2.3(a). Numerical Performance as a function of MAXQN

in t h e spherical diaphragm with clamped boundary

loading (snap-through)

Notes:
*VAX 8700 with VAX/VMS V4.5, CPU time normaIized
-t Diverging solution a t snap-through (0.96-0.98 psi)
Case 2: The parametric study is performed by varying MAXLS from 0 to 8. The results
are summarized in Table 11.2.3(b) for loading and unloading phases separately. As indicated
in the table, the convergence near the snap-through point requires MAXLS to be greater than
4. Computational efficiency appears t o be better with MAXLS of 5 or 7 than 6, which cannot
be rationalized. Again, the convergence is easier for unloading than the loading.

Table 11.2.3(b). Numerical Performance as a function of MAXLS

in t h e spherical diaphragm with clamped boundary
n LOAD I COMPUTATION I hf AXLS il

Notes:
* VAX 8700 with VAXIVMS V4.5
+ Diverging solution a t the pressure 0.96-0.98 psi
- ~eacling the iteration limit of 30 in the vicinity of the snap-through pressure
ii) Simply Supported Case
In the simply supported case, the snap-through occurrs twice as shown in Fig. 11.2.7. The
first snap-through occurrs at 0.54-0.56 psi, and the second occurrs at 0.70-0.705 psi. The first
snap-through brings the diaphragm into an interim stable mode by flipping it into a shape of
double curvature. The diaphragm comes into contact with the backing plate near the periphery
immediately after the first snap-through, stabilizing the interim position. The central area of
the diaphragm comes into contact with the backing plate upon the second snap-through. If
the pressure is unloaded before the second snap-through, the diaphragm snaps back at 0.50-
0.48 psi. Once the second snap-through occurs, however, the snap-back does not occur upon
unloading because the spring-back force is trapped by the constraint forces at the periphery.
Instead, the central area remains in contact with the backing plate during unloading until
the pressure is reduced to less than 0.1 psi (around point B). When the pressure is removed
entirely, the central area is lifted off, leaving the peripheral area in contact with the backing
plate. Based on these observations, the load history between points A and B can be conjectured
by the dashed line in Fig. 11.2.7.

Case 1: Table 11.2.4 summarizes the numerical performance as a function of MAXQN for
the first and second snap-through phases separately. The convergence behavior dictates that
the maximum number of line searches allowed for each iteration be set to 2 and 10 for the
first and the second snap-through, respectively. The first snap-through requires MAXQN to
be greater than 0 for convergence. The solution to the second snap-through phase can only be
obtained with a MAXQN of 15 and 20. Computational efficiency appeares t o be better with a
MAXQN of 20.
Table 11.2.4. Numerical Performance as a function of h/IAXQN in the
spherical diaphragm with simply supported boundary

Notes:
* VAX 8700 with VAX/VMS V4.5
+ Diverging solution at first and second snap-through
- Reaching the iteration limit of 30 at the second snap-through

Case 2: The first snap-through requires MAXLS to be greater than 1 for convergence and
the second snap-through converges only with MAXLS of G and 10 for the MAXQN range of 0
to 10. The parametric study is not fruitful for this case due t o the erratic numerical behavior.
 -
I

*
w

-
w ./
A
+-'., 3
I \

- f \
\
\\

\
c>
\
\
\
Load \ (>
(psi) \
\
\
\
\ ()
\
0 Unloading \

- - - Conjectured \
\
\
()

\
\
\
d
.
\ <>
\
\
\
\ )
\(
\
B
l l l l l l l l l l l l l l L" l l l l

Displacement (in)

Figure 11.2.7. Load vs. Vertical Displacement at the Center

(Simply Supported Case).
 iii) Simply Supported and Sliding Case
In the case where the periphery is allowed t o slide in the radial direction while it is simply
supported, the diaphragm undergoes a slight snap-through a t a pressure between 0.188-0.190
psi. The snap-back occurrs at a pressure between 0.184-0.182 psi as shown in Fig.ll.2.8. Since
the snap-through in this case is moderate and shallow, as shown in the figure, the pressure
is incremented in very small steps around the critical load t o obtain a more descriptive load-
deflection curve.

Load
(psi)

+ Loading
0 Unloading
--. Conjectured

Displacement (in)

'Figure 11.2.8. Load vs. Vertical Displacement at the Center

(Sliding and Simply Supported Case).
 Case 1: Table 11.2.5(a) summarizes the numerical performance as a function of MAXQN
for the loading and the unloading phases separately. The table shows that the solution diverges
around the snap-through or snap-back point for a MAXQN less than 10. This indicates that
BFGS update operations do help to overcome the difficulties near the singularity in the stiffness.
The computational efficiency improves slightly as MAXQN increases; the MAXQN values of
25 and 30 seem to be more efficient.

Table 11.2.5(a). Numerical Performance as a function of MAXQN in t h e

spherical diaphragm with sliding/simply supported boundary

n LOAD I COMPUTATION I MAXON II

(psi) EFFICIENCY 0 ( 5 1 10 1 15 -1 20 1 25 1 30
No. of QN- updates
- + I + 1 311 I 367 1 608 1 603 1 653 L
'

0 - 0.6 No. of line searches + +.----- 20 21 19 19 19

(15 increments) N o . -- 88 91 90 90 90
loading (snap-through) No. of K updates + + 10 7 6 5 5
No. of QN updates + + 193 231 183 245 171
0.6 - 0 No. of line searches + + 20 23 26 21 21 U
(9 increments) No. of iterations + + 61 61 59 59 55
unloading (snap-back) No. of K updates + + 7 6 7 5 5
-.
total CPU time*, normalized + + 1.108 1.115 1.074 1.000 1.027
Notes:
* VAX 8700 with VAX/VMS V4.5
+ Diverging solution at'snap-through or snap-back point
Case 2: Table 11.2.5(b) summarizes the numerical performance as a function of MAXLS
while LSTOL is fixed at 0.5. The table shows that the solution diverges around the snap-
through or snap-back for MAXLS less than 3. This indicates that line search operations do help
to overcome the difficulties near the singularity in the stiffness. The significance demonstrated
in Table 11.2.5(b) is the effectiveness of the line search, and the efficiency is not sensitive to the
value of MAXLS from 3 to 6. The parametric study on the line search tolerance (LSTOL) with
MAXLS of 5 is summarized in Table 11.2.5(c). Notice that when LSTOL is small (0.1-0.3),
the algorithm keeps thrashing on line search operations until the solution diverges at the snap-
through or snap-back point. It also indicates that the algorithm becomes inefficient if LSTOL
has too large a value, e.g., 0.7. The choice of 0.5 for LSTOL seems to be a good compromise
for convergence and efficiency of the line search method.
 Table 11.2.5(b). Numerical Performance as a function of MAXLS in the
spherical diaphragm with sliding/simply supported boundary

LOAD COMPUTATION RlAXLS

(psi) EFFICIENCY 0 1 2 3 4 5 6
0 - 0.6 No. of line searches 0 + + 16 18 19 21
(15 increments)
\-- - -~~
No. of iterations 87 + + 91 91 90 90 I
loading (snap-through) No. of I( updates
' 4 + + 5 5 5 5
0.6 - 0 No. of line searches + + + 22 22 22 22
(9 increments) No. of iterations + + + 5 9 62 62 62
unloading (snap-back) No. of K updates + + + 5 5 5 5
U
total CPU time* (normalized) I
+ I
+ +
I
1.010 1.003 1.008 1.000 U
Notes:
* VAX 111780 with VAXIVMS V4.5 (not accurate as performance index)
+ Diverging solution at snap-through or snap-back point
Table 11.2.5(c). Numerical Performance as a function of LSTOL in the
spherical diaphragm with sliding/simply supported boundary

Notes:
* VAX 11/780 with VAXIVMS V4.5 (not accurate as performance index)
+ Diverging solution at snap-through or snap-back point

11.2.4 Pre-Buckling Behavior: an Imperfect Spl~ericalCap

A spherical shell with an initial imperfection is analyzed [11.6]. The shell is subjected to an
external uniform pressure, while the periphery is clamped. The problem is assumed to remain
axisymmetric in geometry and loading throughout the deformation. The details of the geome-
try and the material are shown in Fig. 11.2.9. The initial imperfection is introduced by making
the radius of curvature near the apex greater than the shell radius. The material is elasto-
plastic with von Mises yield criterion and kinematic hardening, and the large displacement is
assumed in the analysis. The external pressure is gradually increased until the shell collapses
[11.7], as shown in Fig. 11.2.10. The finite element model shown in Fig. 11.2.11 represents a
10-degree sector in solid elements with axisymmetric boundary conditions.
Geometry: Material:
h = 0.0251 in. 7075-T6 Aluminum
R = 0.8251 in. E = 10.8 x 1O6 psi
R, = 1.1506 in. u = 0.3
a = 0.267 in. uy = 7.8 x 10' psi
y = 26)'
b = 0.14328 in.
a = 14.3065' Strain Hardening:
c = 0.32908 in. Slope E, = 1.1 x 1O6 psi
P = 37.7612' Kinematic

Boundary Condition:
Periphery Clamped

Figure 11.2.9. Imperfect Spherical Shell.

 -
Pressure, 3000 -
9 (psi)

@buckling point by MSCiNASTRAN

+ Finite Element by MSC/NASTRAN
Finite Difference by Kao
) Experimental Buckling Point

I
0 .002 .004 .006 .008 .010 .012 .014 ,016

Central Deflection, W (in.)

Figure 11.2.10. Elastic-Plastic Buckling of a Clamped Spherical

Cap with Flat Spot (Load vs. Central Deflection).

Figure 11.2.11. Deformed Shape upon Buckling with a Solid Model.

11.2 - 16
 For the parametric studies, numerous analyses are repeated for various values of MAXQN,
MAXLS and LSTOL with the pressure increased up to 3500 psi. In this problem, the loading
can be divided into two stages: the plastic deformation developing stage (0-3000 psi) and the
instability initiating stage (3000-3500 psi). Nonlinearity is fairly mild in both stages compared
to other examples. In the second stage, all the elements undergo plastic deformation.

Case 1: The results are summarized in Table 11.2.6(a), which illustrates the numerical
performance of the program as a function of MAXQN. Note that in Table 11.2.6(a) convergence
is achieved for all values of MAXQN except for the second loading stage and MAXQN=O, where
the geometric nonlinear effects become more prominent. The computational efficiency is most
favorable with a MAXQN of 20.
Table 11.2.6(a). Numerical Performance as a function of MAXQN
in t h e imperfect spherical cap

Notes:
* VAX 8700 with VAX/VhfS V4.5, not accurate as performance index
+ Diverging solution at load step 3400-3500 psi
Case 2: The results are summarized in Tables 11.2.6(b) and 11.2.6(c), which illustrate
the numerical performance of the program as functions of MAXLS and LSTOL. Note that
in Table 11.2.6(b) the convergence is not only achieved when the line search is suppressed
but also achieved most efficiently for MAXLS=O. When the line search operation is allowed
(MAXLS > 0, LSTOL=0.5), more line searches are performed in the second stage where
the geometric nonlinear effects become more prominent. The parametric study performed
on LSTOL with MAXLS=5 is summarized in Table 11.2.6(c). A quite natural consequence of
unnecessary line searches is exemplified by decreasing efficiency with smaller LSTOL. However,
Table 11.2.6(c) does demonstrate that the accurate line search does not expedite convergence
but rather deteriorates efficiency. This problem illustrates that the line search is unnecessary
for moderately nonlinear problems where the divergence is not an issue.
 Table 11.2.6(b). Numerical Performance as a function of MAXLS
in t h e imperfect spherical cap

Notes:
* VAX 8700 with VAX/VMS V4.5

Table 11.2.6(c). Numerical Performance as a function of LSTOL

in t h e imperfect spherical cap

I
instability developing No. of K updates 7 7 7 1 7 1 7 7 1 7
.. total CPU time* (normalized) 1.214 1.083 1.046 1 1.056 1 1.020 1.014 ( 1.000
Notes:
* VAX 11/780 with VAX/VMS V4.5

11.2.5 Concluding Remarks

The BFGS update method is an expedient and effective procedure t o prevent divergence in
complex nonlinear problems. It is also an expeditious method that improves computational
efficiency. If divergence is not an issue, the solution requires more line searches, iterations and
stiffness updates in the absence of BFGS updates, as demonstrated in the detailed parametric
studies. Preliminary tests showed that the current BFGS update scheme works well on the
negative-definite matrices, but did not work on the unsymmetric matrix.

Although the line search method appears theoretically attractive, the computational in-
tensity involved in the search process often prohibits indiscriminate use of this procedure in
nonlinear finite element analysis. The line search method is an effective procedure t o prevent
divergence in complex nonlinear problems. It is particularly useful t o cope with difficulties
when the stiffness approaches singularity. The line search process can help t o converge the
solution dramaticalIy in some cases, such as bypassing the unstable regime of the snap-through
problem without the aid of a displacement control method or when the structure stiffens dras-
tically as in the closing gap phenomenon. However, indiscriminate use of line searches causes
adverse effects on efficiency. In fact, the line search tends to deteriorate the efficiency in mod-
erately nonlinear problems where divergence is not encountered. Even when the line search is
valid, the numerical behavior associated with the line search is rather erratic. Moreover, the
line search diminishes the effectiveness and efficiency of the full Newton-Raphson iteration.

Optimal values of parameters and tolerances are extremely difficult to achieve. They are
not only problem dependent but also vary within a problem during the course of an incre-
mental process. Tolerances and parameter values in MSCINASTRAN were tuned based on
50 nonlinear test problems [11.8]. Default values are intended to provide the best workable
method for the wide spectrum of nonlinear problems without any insight or experience, but
they are by no means optimal.
11.3 CREEP ANALYSIS OF THICK-WALLED PRESSURE
VESSEL

11.3.1 Introduction

The purpose of this example is to illustrate the creep solution of a typically encountered
structure. The structure contains a complex stress patteren requiring an axisymmetric three-
dimensional solution. The material has the properties of an engineering metal subjected to
high pressure.

11.3.2 Problem Description

The creep behavior of a thick-walled pressure vessel with a flat-end closure [11.9] is analyzed
under an internal pressure of 445 psi. The geometry of the pressure vessel, as shown in Fig.
11.3.1, is defined as follows:
Overall length 0.89 inch
Outside radius 0.25 inch
Inside radius 0.159 inch
Wall thickness (uniform) 0.091 inch
Fillet radius 0.03 inch
The material has the following elastic properties:
Young's Modulus 20 x lo6 psi
Poisson's Ratio 0.3
and is assumed t o obey an empirical creep law in the form of

where a and t are measured in psi and hours, respectively.

11.3.3 Model Description

The modeling is simplified by axisymmetry and symmetry about the mid-plane. A finite
element model is constructed using 72 solid elements (4 PENTAs and 68 HEXAs) with 355
active degree-of-freedom. Appropriate boundary conditions are imposed for symmetry (using
SPC1, SPCADD and GRDSET entries) in the cylindrical coordinate system (CORDIC). A
local coordinate system (CORD2C) is used t o specify the geometry of a toroidal section.

The internal pressure is applied by PLOAD4 entries. Although the applied load does not
produce plastic deformation, all the elements are treated as being made of elasto-plastic-creep
material by attaching MATS1 and CREEP entries t o the MAT1 Bulk Data entry.
 Fig. 11.3.1 shows an isometric view of the model. Also shown is a model in the Reference
[11.9] for comparison. Table 11.3.1 shows a listing of the input data.

11.3.4 Solution Procedure

The creep analysis should be performed using SOL 66 of Version 63. The internal pressure
of 445 psi is applied in the first subcase for the initial static solution. Creep analyses are
performed in the subsequent subcases by applying a nonzero time increment while the applied
pressure is kept constant.

Six subcases (only two subcases are activated on the listing shown is Table 11.3.1) are set
up as follows:
-
Ending Creep
Subcase ID KMETHOD Increments At (Hour) Time (Hour)
100 AUTO 1 0. 0.
200 SEMIQN 5 0.02 0.1
300 AUTOQN 9 0.1 1.0
400 AUTOQN 10 0.1 2.0
500 AUTOQN 5 0.2 3.0
600 AUTOQN 5 0.4 5.0

The computer run was stopped after every subcase t o obtain the stress contour plot. Then
the analysis was continued by restarting from the last converged solution. Notice that the
restart run requires three PARAMeter entries (i.e., SUBID, LOADINC and LOOPID) to des-
ignate the solution from which the restart run starts.

11.3.5 Results and Discussion

The entire analysis took six runs (five restarts) to obtain solutions up to five hours of creep.
Each creep solution took three to four iterations t o converge (average 10 CPU minutes on the
VAX 111780). The results of MSCINASTRAN analysis are presented in Fig. 11.3.2 through
11.3.5 in comparison with those of the Reference.

Fig. 11.3.2 and 11.3.3 show stress contours at t = 0 and 3, respectively. The stress history
at two selected points (junction of the cylinder and end closure) are shown in Fig. 11.3.4. It
can be seen that the steady-state stress is reached after approximately three hours of creep.
Notice that MSCINASTRAN shows a higher stress level at the inside surface (where the stress
concentration occurs) throughout the analysis. Since the initial stress is similarly higher, this
error is actually a consequence of a coarse mesh. Also, notice that the point with higher stress
relaxes a little faster than might be expected.

It is not possible t o plot creep strain contours with the current version of MSCINASTRAN.
But the effective creep strains are printed with nonlinear stresses. Effective creep strains at five
sampling points at t = 3 hours are shown on the creep strain contour plot in the Reference.
The results are in good agreement with the Reference. More accurate solutions should be
obtained with a more refined mesh around the fillet area.

11.3.6 Input Data Listing for the Pressure Vessel

ID TWCAS,V65 $ SHL 6/21/83

SOL 66 $
DIAG 8, 50 $,20, 51
TIME 40 $ FOR VAX
CEND
TITLE = THICK WALLED CYLINDER SUBJECT TO INTERNAL PRESSURE
SUBTITLE = AXISYMMETRIC
ECHO = UNSORT
DISP = ALL
STRESS(PL0T) = ALL
GPSTRESS-ALL
STRFIELD=ALL
SEALL = ALL
SPC = 100
SUBCASE 100
LABEL = STATIC ANALYSIS
LOAD = 100
NLPARM = 10
SUBCASE 200
LABEL = CREEP 0.1 HOUR
LOAD = 100
NLPARM = 20
SUBCASE 300
LABEL = CREEP 1 HOUR
LOAD = 100
NLPARM = 30
SUBCASE 400
LABEL = CREEP 2 HOURS
LOAD = 100
NLPARM = 40
SUBCASE 500
LABEL = CREEP 3 HOURS
 LOAD = 100
NLPARM = 50
SUBCASE 600
LABEL = CREEP 5 HOURS
LOAD = 100
NLPARM = 60
OUTPUT(PLOT)
CSCALE 1 . 3
PLOTTER NAST
SET 1 ALL
AXES Z,X,Y
VIEW 3 0 . , 2 0 . , 0 .
FIND SCALE ORIGIN 1 SET 1
PTITLE=ISOMETRIC VIEW
PLOT SET 1 ORIGIN 1 LABEL POINT
VIEW 0. , 0. , 0 .
PTITLE=SIDE VIEW
FIND SCALE ORIGIN 1 SET 1
PLOT SET 1 ORIGIN 1 LABEL ELEMENT
PLOT STATIC SET 1 ORIGIN 1
OUTPUT(POST)
SET 10 = ALL
VOLUME 10 SET 10 SYSTEM CORD 1
BEGIN BULK
$ GENERAL
PARAM POST 0
NLPARM 10 1 AUTO
NLPARM 20 5 0.02
NLPARM 30 9 0.1
NLPARM 40 10 0.1
NLPARM 50 5 0.2
NLPARM 60 5 0.4
CORDlC 1 1000 1001 1002
GRID 1000 0 0. 0. 0. 0 123456
GRID 1001 0 0. 0.5 0. 0 123456
GRID 1002 0 0.5 0. 0. 0 123456
PLOTEL 1001 1000 1001
PLOTEL 1002 1000 1002
CORD2C 2 0.129 0.324 0. 0.129 0.324 -1. +COR2
+CUR2 0.129 0.354 0.
$ PROPERTIES
MAT1 1 20.0+6 0.3
MATS1 I PLASTIC 2.0+6 1 1
CREEP 1 CRLAW
+CRI 300 19.8-16 3.61
PSOLID 3 1 1
$ BOUNDARY CONDITIONS
SPCADD 100 200 300
SPC1 200 23456 1301
+SPC 1361 1362 1381
SPC1 300 12456 1100
$ LOADING CONDITIONS
LOAD 100 1. 445.
PLOAD4 200 104 1.
PLOAD4 200 108 1.
PLOAD4 200 112 1.
PLOAD4 200 116 1.
PLOAD4 200 120 1.
PLOAD4 200 204 1.
PLOAD4 200 208 1.
PLOAD4 200 212 1.
PLOAD4 200 216 1.
PLOAD4 200 220 1.
PLOAD4 200 224 1.
PLOAD4 200 304 1.
PLOAD4 200 308 1.
PLOAD4 200 312 1.
PLOAD4 200 316 1.
PLOAD4 200 320 1.
PLOAD4 200 324 1.
PLOAD4 200 328 1.
$ GRIDS R
GRDSET 1
GRID 1100 0.
GRID 1101 0.025

$***** OTHER GRID CARDS ARE OMITTED. *****

GRID 1201 2 0.121 15. 5.595-3

GRID 1396 0.159 -2. 0.324
$ ELEMENTS
CPENTA 101 3 1100 1101 1102 1120 1121 1122
$***** OTHER CONNECTIVITY CARDS ARE OMITTED. *****

CHEXA 328 3 1373 1374 1376 1375 1393 1394 +H328

+H328 1396 1395
ENDDATA
 Idealization of the pressure vessel with a flat end closure.

Figure 11.3.1. Finite Element Model of Pressure Vessel.

Effective stress contour? at t i m e = 0 for the pressure vessel w i t h n flat end closure.

Figure 11.3.2. Initial Stress Distribution.

11.3 - 8
 Effective stress contours at time = 3.0 hours for the pressure vessel with a flat end closure.

- - -----
EFFECTIVE STRESS AFTER 3 HOURS OF CREEP

L W 8 ............ 200.
L e d C ............300.
L w d 0 ..*......... 400.
I
i
L w d E ............ 500.
L w d F ............ 600.
Levd G ............ 700.
Levd H ............ 800.
L W l ............ 900.
L W J .......... 1000.
Not@:Stre88 Unit8 in k d

Figure 11.3.3. Effective Stress Distribution after 3 hours of Creep

11.3 - 9
 IJW

-
/ \
Im
1100
----- EFFKfNE STRESS A T INSIDE SURFACE

\/' \ / \ / \ /
lOQJ

>-. \

------ ---- -
-4
1'---
1

am..
f
m 700
3
.L
L

600'

WX)
-.-- ----
400

XI0

I
2QO
---- I
-
\
/ P U T E N D CLOSUIE
I
IW.
E f ~ m r v SE ~ U SAT CUI)IO( SUr*Q

0.001 0.01 0.1

I
1.0
x
10.0
M S C / N ~ M-
1W.O
T W - W

Figure 11.3.4. Stress Relaxation at the Junction of the Cylinder and

End Closure.

0.007

X MSC/NASTPAN
176 ~-, - Greenbaum, et. al.
\

Figure 11.3.5. Effective Creep Strain ( x I o - ~ ) after 3 hours of Creep

11.4 TRANSIENT RESPONSE OF AN IMPULSIVELY
LOADED CYLINDRICAL PANEL

11.4.1 Problem Description

A self-adaptive algorithm has been developed and implemented for the implicit time integra-
tion of the nonlinear transient response analysis (SOL 99 or 129) in MSCINASTRAN. In this
algorithm, the time increment for the next time step is estimated based on the deformation
pattern at the preceding step. The iteration process for equilibrium employs expeditious meth-
ods such as quasi-Newton updates and line searches, as well as an adaptive stiffness matrix
update strategy for efficiency. Convergence difficulties induced by inadequate prediction of the
time step size or the change in nonlinearities are tackled by the bisection method based on the
divergence criterion.

The divergence criterion can detect and overcome dynamic buckling problems. For instance,
dynamic buckling or snap-through may induce a singular or indefinite dynamic stiffness matrix.
The iteration is bound t o diverge with a nearly-singular dynamic stiffness matrix, which is
detected by the divergence criterion. Then the bisection is activated and the dynamic stiffness
is formed again. The dynamic stiffness is likely to be stable at this point and the difficulty
in the snap-through is coped with. Otherwise, the bisection will continue until the iteration
process is stabilized and a converged solution is found.

A 120' cylindrical panel is 12.56 inches long, with an inside radius of 2.9375 inches and
a thickness of 0.125 inches, as shown in Fig. 11.4.l(a). The ends of the panel are simply-
supported and the boundaries at the sides are fixed. The panel is impulsively loaded by
applying an initial velocity of vo = 5650 in./sec normal t o the region whose length and width
are 10.205 inches and 60' in angle, respectively. The objective of this analysis is t o illustrate
the applicability and effectiveness of the self-adaptive algorithm t o a nonlinear problem by
examining an impulsively loaded cylindrical panel.

11.4.2 Finite Element Model

This problem features material nonlinearity along with geometric nonlinearity due t o large
displacements. Due t o symmetry, only one half of the cylindrical panel is modeled using shell
elements. Three different meshes are used t o solve this problem:
6x16 mesh
- 6 circumferential elements, one element every 10 degrees
- 16 elements along the panel's length
r 8x16 mesh
- 8 circumferential elements, one element every 7.5 degrees
 - 16 elements along the panel's length

16x32 mesh

- 16 circumferential elements, one element every 3.75 degrees

- 32 elements along the panel's length

The undeformed shape of the 8x16 model is shown in Fig. 11.4.l(b), in which points A and
B (at z=6.28 in and z=9.42 in) indicate the nodal points where displacement responses are
measured. The input data for the 8x16 model is listed in Section 11.4.4. SOL 99 (in Version
67) is used for the nonlinear transient analysis of the impulsively loaded cylindrical panel. The
analysis is performed in five subcases of equal duration (200 psec). Subcase 1 contains the
loading conditions (IC for initial velocity), specified by TIC in the Bulk Data. The remaining
four subcases calculate the response to the impulsive loading.

The 6xl6,8x16, and 16x32 meshes used t o model the cylindrical panel contain 96, 128, and
512 QUAD4 elements, respectively. The geometry is described in the cylindrical coordinate
system, indicated by CORDlC. Boundary conditions, specified using S P C l entries, include
the constraint of all six degrees-of-freedom (DOF) for each GRID point at the fixed side of
the panel. Translational DOFs, namely components 1, 2, and 3, for the GRID points at the
simply-supported ends of the panel are also constrained. Symmetric boundary conditions are
imposed by constraining components 2 , 4 , and 6 of the GRID points along the side of the panel
on the symmetric plane.

Material nonlinearity is signified by the inclusion of MATS1. The material is assumed to

be elastic-perfectly plastic and is modeled using the von Mises yield criterion. The geometric
nonlinearity is also taken into account by specifying PARAM,LGDISP,l. The solution strategy
is controlled by the TSTEPNL data entry, which specifies the parameter ADAPT indicating
the adaptive strategy.

11.4.3 Analysis Results and Discussion

The analysis was performed for a duration of 1 msec with an initial time step size of 2 psec.
During the analysis of the 6x16 model, the automatic time step adjustment was activated three
times. The initial step size was quadrupled to 8 psec a t the end of 0.04 msec, reduced t o 4
psec at 0.22 msec, then reduced t o the original size a t 0.31 msec. During the analysis of the
8x16 model, the time step size was automatically adjusted nine times. The initial step size
was quadrupled a t the end of 10 psec, then reduced t o 4 psec in the next time step when the
bisection was activated. The adjustment was activated again, with the time step sizes ranging
from 8 psec t o 1 psec for the remainder of the analysis. The automatic time step adjustment
occurred eleven times in the analysis of the 16x32 model. The bisection was activated at 3
psec. Thereafter, A t was adjusted t o step sizes ranging from 4 psec t o 0.5 psec, having the
smallest value during peak displacements.
 The deformed shape of the 8x16 model at the peak displacement is shown in Fig. 11.4.l(c).
The displacement response of the midspan of the cylindrical panel, point A in Fig. 11.4.l(b),
is given in Fig. 11.4.2 for the three mesh sizes along with experimental results [11.10], and
the solution obtained by MSCIDYNA [11.11]. There is a significant improvement in the
MSCINASTRAN results as the mesh is refined. Notably, the 16x32 mesh solution is in very
good agreement with the experimental results. Refinement of the mesh causes the cylindrical
panel model t o become more flexible, thereby producing a more realistic displacement response
curve. The table below lists the maximum displacement of the midspan normalized to the
experimental results, and are compared with a solution using explicit integration [11.12].

Normalized Maximum Displacement at Midspan

mesh size MSCINASTRAN Kennedy et al. [11.12]

6x16 0.69 0.78
8x16 0.87 0.86
16x32 0.95 0.93

Kennedy et al. [11.12] stated that the increased number of quadrature points through the
thickness resulted in a stiffer behavior in the panel, compared with the solutions for the same
problem in Reference [11.13]. In order t o examine this point, MSCINASTRAN solutions were
obtained using three Gauss points through the thickness by including PARAM,NLAYERS,3 in
the Bulk Data. Fig. 11.4.3 compares the displacement response of the 8x16 model at points A
and B using three and five Gauss points. Contrary t o the observations in [11.12], the number of
Gauss points did not have a significant effect on the displacement response. This is attributed
t o the fact that the dominant reaction force stems from membrane action and the material is
perfectly-plastic. When the panel undergoes snap-through, the stresses in the crown line reach
the yield point and plastic yielding occurs in compression throughout the thickness. Varying
the number of Gauss points, therefore, should not greatly affect the results of this problem.
11.4.4 Input Data Listing for the 8x16 Model

ID CYROOF, V66 $ SHL, 5/27/87, DSNLTR

SOL 99 $ NONLINEAR TRANSIENT ANALYSIS
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
TIME 300 $ FOR VAX
CEND
TITLE=CYLINDRICAL PANEL UNDER IMPACT
ECHO=BOTH
SET 1=108,112
DISP=I
SEALL=ALL
SPC=IOO
SUBCASE 1
LABEL=LINEAR ELASTIC REGION
IC=I
TSTEPNL=10
SUBCASE 2
LABEL=RESPONSE TO IMPACT
TSTEPNL=lO
SUBCASE 3
TSTEPNL=IO
SUBCASE 4
TSTEPNL=10
SUBCASE 5
TSTEPNL=10
OUTPUT(PLOT)
CSCALE=1.3
PLOTTER NAST
SET I=ALL
AXES X,MZ,Y
VIEW 30.,20.,0.
MAXI DEFO 1.
PTITLE=ISOMETR,AIC VIEW
FIND
PLOT LABEL GRID-A
PLOT TRANSIENT DEFORMATION 1 TIME 4.E-5
OUTPUT(XYPLOT)
XTITLE= TIME IN SEC.
XGRID LINES=YES
YGRID LINES=YES
YTITLE=DISP. A T 1 0 8 % 1 1 2 ( T i )
XYPLOT DISP RESP/108(TI) ,112(T1)
BEGIN BULK
PARAM K6ROT 100.
PARAM LGDISP 1
TSTEPNL 10 100 2.-6 1 ADAPT
$ MATERIAL AND GEOMETRIC PROPERTIES
MAT1 1 10.5+6 0.33
MATS1 1 PLASTIC
PSHELL 10 1 0.125 1
$ BOUNDARY CONDITIONS
SPCl 100 123456 900 THRU
SPC1 100 246 100 THRU
SPC1 100 123 100 200
+SPC 700 800 900 116
+SPC1 616 716 816 916
$ LOADING CONDITIONS
TIC 1 101 1
- - *1 == $
=11 $
TIC 1 201 1 -5650.
- -- *1 == $
=11 $
TIC 1 301 1
- - *1 == $
=11 $
TIC 1 401 1
- - *1 == $
=11 $
TIC 1 501 1
- - *1 == $
=11 $
$ GEOMETRY AND CONNECTIVITY
CORDlC 1 1000 1001 1010
PLOTEL 1000 1000 1001
PLOTEL 1001 1000 1010
GRDSET,,,,,,l$
GRID 1000 0. 0. 0.
GRID 1001 0. 0. -1.
GRID 1010 0. 1. 0.
GRID 100 1 3.0 0. 0.
GRID 101 1 3.0 0. 0.785
GRID 102 1 3.0 0. 1.57
GRID 103 1 3.0 0. 2.355
GRID 104 1 3.0 0. 3.14
GRID
GRID
GRID
GRID
GRID
GRID
GRID
GRID
GRID
GRID
GRID
GRID
GRID
-
=15
GRID
-
=15
GRID
-
GRID
-
=15
GRID
-
=15
GRID
-
=15
GRID
-
=15
GRID
-
CQUAD4
CQUAD4
CQUAD4
CQUAD4
CQUAD4
CQUAD4
CQUAD4
CQUAD4
CQUAD4
CQUAD4 110
CQUAD4 111
CQUAD4 112
CQUAD4 113
CQUAD4 114
CQUAD4 115
CQUAD4 116
CQUAD4 201
- *1
= 14
CQUAD4 301
- *1
= 14
CQUAD4 401
- *1
=14
CQUAD4 501
- *1
= 14
CQUAD4 601
- *1
=14
CQUAD4 701
- *1
= 14
CQUAD4 801
-- *1
= 14
ENDDATA
 2.5 x /b - sec2/in4
10.5 x lo6 psi; Young's Modulus
0.33; Poisson's Ratio
4.4 X lo4 psi; Yield Stress
0 ; Plastic Modulus
5650 i n l s e c ; Initial Velocity

(a) Problem Description (with 6 X 16 mesh)

(b) 8 x 16 Model of Half the Panel

(c) 8 x 16 Deformed Shape a t Time = 0.349 msec

Figure 11.4.1 Impulsively Loaded Cylindrical Panel

Figure 11.4.2 Displacement Response at Midspan (z=6.28 in)
L

Time Cmsec'

Figure 11.4.3 Comparison of Gauss Integration Schemes

Using 8x16 MSCINASTRAN Model
11.5 CONTACT ANALYSIS FOR THE SHAFT-JOURNAL
INTERFACE IN SOL 66

11.5.1 Introduction

The GAP element is used to model the contact between surfaces. It is a nonlinear element
available in both the nonlinear static and nonlinear transient solution sequences, SOL 66 and
SOL 99, respectively. Difficulties encountered with the use of the GAP include the selection
of the GAP properties such as satisfactory open and closed stiffness values, selection of the
stiffness matrix update stategy, and efficient reanalysis (restarting). This example examines
these issues in SOL 66, using MSCJNASTRAN Version 65. All analyses were performed on a
VAX 8700. Although the same analysis can be performed more efficiently and effectively with
much less user-intervention by using the adaptive GAP elements in Version 67, guidelines and
principles suggested in this section is still valid for determining initial penalty values and the
preload for the adaptive GAP element.

The GAP is intended t o model point-to-point contact (with a bilinear stiffness and a fric-
tion) and is applied t o the surfaces which may come into contact. When the GAP is open
it has an axial stiffness, K b , which is typically very small and may be 0.0, provided that this
does not make the global stiffness matrix singular. While the GAP is open there is no lateral
stiffness and no frictional forces. However, there is a large axial stiffness, K,, when the GAP
is closed and significant compressive and shear forces may be present. When compressive force
exists in the GAP, it can carry any transverse shear load which is less than the coefficient of
friction times the normal load. When the shear load is greater than the friction force, slip
occurs.

For the GAP t o close, the structure must deform so that the GRIDS attached t o the GAP
deflect toward each other a distance Uo. Uo is the initial separation distance between the
surfaces as defined by the PGAP entry, not defined by the distance between the GRID points
attached t o the GAP element. Currently, the GAP element does not take large displacements
into account, i.e., the element coordinate system is not updated. This implies that the orien-
tation of the contact plane remains unchanged during deflection, which is a limitation t o be
considered before using GAP elements.

11.5.2 Problem Description

The model consists of a hollow steel shaft and a steel journal. The shaft is 20.0 inches long
with a 2.0 inch inside diameter and is 0.25 inches thick. It is completely fixed a t one end and
free a t the other end. The 9.0 inch long journal is 0.10 inches thick with 0.5 inch thick flanges
at each end. The journal is located 1.0 inch from the fixed end of the shaft and is free to slide
along the shaft (this rigid body motion is restrained with a spring in the finite element model).
There is a 0.001 inch clearance between the shaft and the journal. A 20000 lb point-load on the
 shaft acts radially inward and is located 5.0 inches from the free end. Analyses are performed
to determine the forces placed on the journal as the shaft deflects into the journal.

The three fundamental components of the finite element model are the shaft, the journal
and the gaps defining the contact surface. By taking advantage of the plane of symmetry
present in the structure, only half of it is modeled. A cylindrical coordinate system is used
to define 182 GRIDS for the 36 HEXA elements used to model the shaft and the 30 HEXA
elements for the journal. One element every 30 degrees is used in the circumferential direction.
This is a very coarse finite element mesh but adeqate for demonstration purposes.

Locations

Free EM

Figure 11.5.1 Finite Element Model Exposing GAP Locations

The boundary conditions include constraining all 6 degrees of freedom (DOF) of each GRID
point at the fixed end of the shaft. Symmetric boundary conditions are specified along the
edges on the symmetric plane for both the shaft and journal. Rotational DOFs attached to
the HEXA elements are also removed via SPCs.

Twenty-eight GAP elements connect the shaft to the journal, one at each GRID location
around the journal's flanges. The cylindrical coordinate system is used to define the x-axes of
the GAPs t o be the TI or radial direction. The space between the shaft and the journal is
0.001 inches. The initial gap opening, Uo, is set t o 0.001.

With the GAPs open, the stiffness between the shaft and journal is theoretically zero. The
constraints and boundary conditions are such that the journal is not restrained from the rigid
body motion in three degrees of freedom. These rigid body motions lead to singularities in the
stiffness matrix. The rigid body motions are rotation about the y-axis and translation along
the x and z-axes. To avoid singularities, a soft spring is attached to the edge of the journal
at GRID 30321 in the z-direction. This CELAS2 scalar spring element elastically restrains
translation in the z-direction. The spring constant should be as small as possible t o avoid the
singularity. A value of 10.0 lb/in. fails, however, an increase in magnitude to 100.0 lb/in.
is satisfactory. When the GAPs closed and the stiffness matrix is updated with the larger
stiffness value, K,, the other two rigid body motions are taken care of. However, they remain
singular until the contact occurs, thus a small nonzero Kb must be specified.

Due t o the symmetry condition, half of the 20000 lb force is applied t o GRID 10015 via
a FORCE entry. Because there are no material or geometric nonlinearities in the model, the
whole load is applied in 1 increment (INC = I on the NLPARM entry) in a single SUBCASE.
Once the load is applied, the GAPs either open or close and applying the load in multiple
increments has no affect on helping the solution to converge nor on the validity of the final
solution. To improve convergence, the preload, Fo on the PGAP entry, is set to 0.10 (Fo =
Uo X K b ) .

When superelements are used in SOL 66, the CLOAD and LSEQ entries must be used to
apply the loads t o upstream superelements.

11.5.3 Selection of GAP Stiffness

Kb must be large enough to prevent singularities in the initial stiffness matrix and small enough
t o present realistic answers. Caution is also required in choosing the closed GAP stiffness,
K,. A value that is too large may lead t o numerical ill-conditioning, slower convergence or
divergence. A K , value which is too small leads t o wrong answers. Experience shows that
good choices for Ii, and Kb are:

Closed stiffness K, > lo3 x max[K(journal), K(shaft)]

Open stiffness Kb < x min[K(journal), K(shaft)]

where K(journal) and K(shaft) are the local stiffnesses of the journal and shaft in the direction
of the GAP axis at the GAP location.

Using Roark and Young [11.14], the formulas for circular rings are used t o estimate the
local structural stiffness. The radial stiffness of the cylinder subject t o the radial load can be
estimated by

where E is the elastic modulus, I is the area moment of inertia and Rc is the approximate
radius t o the centroid.

Based on Eq. 11.5.2, the stiffness of the flange portion of the journal in Fig. 11.5.2 is
estimated to be 2.68 x 10' lb/in. and for the shaft 1.37 x l o 5 lb/in. In estimating the local
 -
stiffness of the shaft, the effective length of the shaft is assumed to extend 3 times the thickness
of the shaft beyond each edge of the flange.

F l rnge

R
OD ~ournrl

'Shaft

Figure 11.5.2 Detail of Contact Region for Determining Local Stiffness Values
Since two parallel rows of GAPs cover the contact region, half of the stiffness is attributed
t o each row of GAPs and the estimates for the local stiffness values are divided by 2. The
estimate for the local journal stiffness is 1.34 x los and for the shaft is .7 x lo5 lb/in. Hence.
from Eq. 11.5.1
Ii, = 1 . o x los

while I{, and should be halved a t the plane of symmetry.

11.5.4 Superelement and ASET

To minimize computer time and turn around time for each analysis, the use of superele-
ments (SEs) and ASET entries are implemented. Superelements are used for substructurillg
in MSCINASTRAN. ASET entries are used for Guyan reduction.

The time t o perform the analysis is decreased by reducing the size of the stiffness matrix
which is updated in SOL 66. A substantial reduction is made using static condensation via
ASET entries. By using ASET entries for nonlinear DOFs, only stiffness terms from nonlinear
elements will be retained for nonlinear iteration. DOFs t o be retained in the analysis set
are placed on ASET entries and all others are automatically ommitted. The complement of
the ASET entry is the OMIT entry. The use of the ASET or OMIT entry is based upon
convenience. The GAP'S axial DOFs are the only nonlinear variables in this problem. The
T 1 DOF from each GRID connected to each GAP are placed on ASET entries. The nonlinear
stiffness matrix is thereby reduced from 452 DOF t o 56 DOF, reducing substantially the time
required for each iteration and stiffness matrix update.

The benefit from using superelements is realized in reanalysis, specifically, when upstream
superelements remain unchanged. By placing the nonlinear elements in the residual structure
(SE 0) and the linear elements in an upstream superelement (SE l o ) , only SE 0 requires
reanalysis, not upstream SE 10. (The use of either the SESET entry and/or field 9 of the
GRID entry t o place GRID points in superelements is a matter of convenience.) When restarts
are performed with changes t o the GAP properties, only the residual structure is updated. For
instance, t o restart with a new Ka value requires only to change the value of I<, on the PGAP
entry and to perform an SEALL operation on SE 0, the residual structure.

Table 11.5.1 compares run times of analyses done with and without static condensation via
ASETs and with and without superelements. In order t o emphasize the benefit that comes
from reducing the size of the nonlinear stiffness matrix and using superelements, two models
are compared, a 66-element model and a 900-element model. In the smaller model the ratio
of nonlinear DOFs (A-set) t o linear DOFs (0-set) is about 1:7. In the big model the ratio of
A-set to 0-set DOFs is about 1:153.

Table 11.5.1. Normalized values of elapsed CPU time for analyses with and without ASETs
and/or SUPERELEMENTS. (Results for each model normalized t o 1.)

All DOFs Nonlinear All DOFs Nonlinear

in ASET by GAP DOFs in ASET by GAP DOFs
default (T1)inASET default (T1)in ASET
No SEs 1.00 .88 1.OO .36
GAP GRIDs
in SE 0 .87 .89 .55 .53
GAP GRIDs
in SE 0 .59 .61 .08 .06
(restart)
No. Elements 66 900
11.5.5 Analysis Procedure

Several analyses are performed t o demonstrate how results vary depending on the value of li,
used. Comparisons are done for two models: a 66-element model with one element every 30
degrees ( a very coarse mesh), a 132-element model with one element every 15 degrees. The
results are compared in Fig. 11.5.3. It is considered that the best possible solution is achieved
when the G A P forces become stationary with respect t o I<,.

2700 - 66-El m n t Model

-
a \ /

u
-
I \

2600

2500 -
2400 '

2300 -
132-El emant Model
-- V
-
a

I \

-ow-
= aest p o ~ ~ i b sl oe l u t i o n
K rec-rma for efficiency

X K, for Best possible s o l u t i o n

9 X, r e s u l t s tn "HI 4158 *ax r a t i o 17-

Ka
Figure 11.5.3. K, vs. axial G A P forces for 66-element and 132-element models
T h e NLPARM entry provides for three types of nonlinear solution methods (ITER, SEMI
and AUTO), which control the stiffness matrix update strategy. T h e ITER method which
updates the stiffness matrix after each iteration ( K S T E P = l ) is referred t o as full Newton's
method. Variations of all three methods are used in several analyses and are tabulated in
Table 11-5.2, which summarizes results from restarts of the residual structure. Only axial
G A P DOFs are in the nonlinear stiffness matrix via ASET entries. All cases use li, = 10"
and K b = 100. T h e severe nonlinear effects (rapidly changing stiffnesses) make it difficult to
obtain a converged solution. Inspection of the convergence factors and line search d a t a from
the nonlinear iteration output is required to make a judicious selection of NLPARM parameters
for better convergence.

Table 11.5.2 Variation of Iteration Strategy in Version 65

(Numbers in parenthesis or blank indicate default value.)
Note that the performance is significantly improved in Version 67.

12 ITER stopped due to 2 diverging sols

13 ITER 2 stopped due to MAXITER limit
14 ITER 2 0 stopped due to MAXITER limit
15 ITER 1 converges in 63. sec
16 ITER 1 0 0 converges in 43. sec

11.5.6 Summary of Results

The load deforms the shaft into an elliptical shape near the free end which then contacts the
journal over a small area at a few locations shown in Fig. 11.5.4. At these locations the GAPS
close and the stiffness changes from Kb to K,.
I. Section A-A Section 0-0 section C-c
Section 3-5

@ = CLOSED

Figure 11.5.4 Sections from Fig. 11.5.1 Indicating Which GAPS Close
Fig. 11.5.3 shows that the 66-element model converges to the best possible solution with
I<, = 10". However, acceptable results are achieved with K , = lo8 and the answers are less
than than 5% off when K , is as small as lo7. I(, = lo8 is considered the smallest value which
provides a good solution, hence recommended for efficiency. The best possible solution for the
132-element model is achieved over a larger range of h', values. A value of I<, spanning six
orders of magnitude provides results that differ by less than 1%. The best possible solution
is achieved with K , = 10'' while Ii, = l o 7 gives a solution that is less than 1% different
from the best possible solution. For the 132-element model, K , = l o 7 is recommended for
efficiency. The range of acceptable K, values increases as the mesh is refined. This trend
continues until the "best" mesh is used, which in this case is about 1 element every 10 degrees
in the circumferential direction.

All three of the iteration methods determined identical solutions as shown in Table 11.5.2.
However, the default values on the NLPARM entry do not yield a converged solution. The
full Newton's method found the solution most efficiently (which is not the case in Version 67).
The time to reach a converged solution was further minimized by restricting the number of line
searches done. Optimal iteration strategy varies with each individual problem. Nevertheless.
the full Newton's method is considered the most conservative and reliable method.

The use of ASETs to reduce the problem size for the nonlinear iterations is strongly rec-
ommended if the model has high ratios of linear to nonlinear elements. The large 900-element
model in Table 11.5.1 uses 64% less CPU time when the nonlinear stiffness matrix is reduced
via ASET entries to include only the essential nonlinear terms.
 Superelements are especially effective when restarts are performed. Without superelements,
the entire analysis must be repeated. However, when superelements are combined with ASETs
for the large model, the restart uses only 6% of the CPU as in the original analysis. These
numbers will vary depending on factors such as: number of stiffness matrix updates, iterations,
ratio of linear to nonlinear elements.

11.5.7 Friction in the GAP

The effects of friction in the GAPs are considered in a separate analysis. The analysis is of the
same model but with two loading conditions. The initial load closes the GAPs (SUBCASE I ) ,
the second one loads the closed GAPs in shear (SUBCASE 2 and 3). When the shear force
is greater than the friction force, slip occurs. Before the full slip condition, the shear force in
each GAP equals K t (transverse stiffness from PGAP entry) times the shear displacement. K t
is an elastic stiffness (ideally rigid) used t o simulate the stationary condition under the critical
load a t which slip occurs. The data is nearly identical t o the data in Table 11.5.3 but with
modifications to the PGAP and ASET entries, additional load cases and NLPARM entries.

Friction coefficients of 0.25 are used for the GAP y and z-directions. With friction included,
the GAP'S shear DOFs (y and z) become nonlinear DOFs and, as such, need t o be included in
the analysis set. DOFs 2 and 3 for each GRID attached t o a GAP are placed on ASET entries
(except where SPCed).

The CELAS spring used t o restrain the journal is not a structural member, however its
stiffness (100 lb/in) provides an adequate stability t o the model. The transverse stiffness,
K t , connected in series with the CELAS should be the predominant stiffness to resist the
transverse loading. As such, the value of K t should be orders of magnitude larger than 100. A
parametric study using K t equal t o lo3, lo4, and lo5 found the minimum load at which full
slip occurs. With Kt equal t o l o 4 and lo5, the same load produces slip. At K t equal t o lo3,
a slightly higher load is required for slip. At very large K t , e.g. l o 5 or higher, convergence is
difficult t o achieve and caution is needed in selecting the loads and iteration parameters. For
best efficiency, K t which is two orders of magnitude larger than the spring stiffness is proven
sufficient (i.e., K t = lo4).

The load in SUBCASE 1 is the concentrated force. The second load, SUBCASEs 2 and 3,
is a line load along the edge of the journal acting in the +z-direction. A 90 lb force is applied to
each of the seven GRIDS along the edge of the journal. All but one of the GAPs (no. 43) slip.
In SUBCASE 3 the load is increased t o 97 lbs in 1 lb increments. The journal slips completely
a t 95 lbs with a total shear force equal t o about 1311 Ibs. As the remaining load increments
are applied, the journal does not experience large displacements as would be expected in the
full slip condition. This is due t o the spring attached along the z-axis.
 Table 11.5.3 MSCINASTRAN Input Data Listing for Case 16.
ID GAP, APPNOTE $ BY BOB LOUWERS
SOL 66
TIME 10
$ CONCENTRIC CYLINDERS CONNECTED BY GAPS.
CEND
TITLE = KA = l.E+9; KB = 100.; INITIAL GAP OPENING = 0.001; SE MODEL.
SUBTITLE = METHOD = ITER WITH 1 INC; K UPDATE AFTER EACH ITERATION
LABEL = ASETS AND SUPERELEMENTS
$SET 111 = 0 $ FOR RESTARTS
$SEALL = 111 $ FOR RESTARTS
SEALL = ALL
SUPER = ALL
ECHO = BOTH
SPC = 100
LOADSET = 950
$STRESS = ALL $ ADDITIONAL OUTPUT REQUESTS
$DISPLA = ALL $ FOR UPSTREAM DATA RECOVERY
$ELFORC = ALL
SET 10 = 10 THRU 50 $ FORCES IN GAP ELEMENTS
ELFORCE = 10
SUBCASE 1
CLOAD = 901
NLPARM = 2001
$
$ Remove ' I $ " from the following Case Control entries to include
$ shear loads when friction effects are included in the Bulk Data
$ as discussed in Section 7.0.
$
$SUBCASE 2
$ CLOAD = 902 $ SHEAR LOADING FOR GAP ELEMENTS
$ NLPARM = 2002
$SUBCASE 3
$ CLOAD = 903 $ SHEAR LOADING FOR GAP ELEMENTS
$ NLPARM = 2003
OUTPUT(PLOT)
*
*
*
BEGIN BULK
$
$ USED TO DEFINE SHAFT, JOURNAL AND THE RADIAL DIRECTION OF GAP AXES.
$
CORD2C 1 0 0. 0. 0. 0.
+CORD 1. 0. 0.
$
$ GRID POINTS WHICH DEFINE SHAFT AND JOURNAL.
$
GRID 10000 1 2. 0.0 0.0 1
*
*
GRID 30621 1 2.75 180. 10. 1 10
$
$ CHEXA ELEMENTS WHICH DEFINE SHAFT AND JOURNAL.
$
CHEXA 100 12 10000 10001 10011 10010 10100 10101 +MOO000
+M00000 10111 10110

CHEXA 405 12 20511 30510 30520 20521 20611 30610 +MOO065

+MOO065 30620 20621
$
MAT1 13 3.+7 .3
PSOLID 12 13
$
$ CGAPS DEFINE CONNECTIVITY - USE COORD SYSTEM 1 TO DEFINE GAP X AXES.
$
CGAP 10 15 10011 20000 1
*

CGAP 46 15 10614
$
$ UO FO
PGAP 15 0.001 .10
$
$ SYMMETRIC BOUNDARY CONDITIONS
$
SPCl 100 246 10000
+MOO068 10006 10010 10011

*
SPCl 100 246 30600 30601 30610 30611 30620 30621
$
$ CONSTRAIN ROTATIONAL DOFS ON ALL SOLID HEXA ELEMENTS.
$

SPCl 100 456 10000 THRU 40000

$
$ BOUNDARY CONDITIONS FOR FIXED END OF SHAFT.
$
SPCl 100 123456 10000 10010 10100 10110 10200 10210 +MOO066
+MOO06610300 10310 10400 10410 10500 10510 10600 10610
$
$ PLACE AXIAL DOFS CONNECTED TO GAPS IN THE ANALYSIS SET.
$
ASETl 1 10011 10012 10013 10014
*
*
ASET1 1 21100 21101 31100 31101
$
$ PLACE GRIDS ATTACHED TO GAPS IN THE RESIDUAL STRUCTURE.
$
SESET 0 10011 10111 10211 10311 10411 10511 10611
*
*
SESET 0 30001 30101 30201 30301 30401 30501 30601
$
$ SOFT SPRING TO PREVENT RIGID BODY MOTION OF JOURNAL IN T3 DIRECTION.
$
CELAS2 76 100.0 30321 3
$
$ CLOAD, LSEQ REQUIRED FOR LOADING WHEN SUPERELEMENTS ARE USED.
$
CLOAD 901 1. 1. 975
LSEQ 950 975 1000
$
FORCE 1000 10015 10000.0 -1. 0. 0.
$
$ NLPARM PROVIDES PARAMETERS FOR NONLINEAR ANALYSIS ITERATION STRATEGY.
$
$ INC KMETHOD KSTEP
NLPARM 2001 1 ITER 1 +NLP
$ MAXQN MAXLS
+NLP 0 0
$
$ The following Bulk Data e n t r i e s d e f i n i n g t h e s h e a r loads and NLPARM e n t r i e s
$ f o r SUBCASE 2 and 3 a r e used when SUBCASE 2 and 3 a r e included.
$
$ SUBCASE 2 LOADS AND NLPARM CARD
$
CLOAD 902 1. 1. 976
LSEq 950 976 1002
FORCE 1002 10015 10000.0 -1. 0. 0.
FORCE 1002 20010 90. 0. 0. 1. O
*
*
*
FORCE 1002 20610 90. 0. 0.
NLPARM 2002 4 ITER 1

+NLP2 0 0
$
$ SUBCASE 3 LOADS AND NLPARM CARD
$
CLOAD 903 1. 1. 977
LSEq 950 977 1003
FORCE 1003 10015 10000.0 -1. 0. 0.
FORCE 1003 20010 97. 0. 0. 1.0
*
*
*
FORCE 1003 20610 97. 0. 0. 1. O
NLPARM 2003 7 ITER 1 YES +NLP3
+NLP3 0 0
$
$ Remove t h e "$" t o include t h e following r e q u i r e d s h e a r DOFs when
$ friction effects are used i n SUBCASE 2 and 3 .
$
$ASETI 3 10011 10012 10013 10014
$ASETI 23 10111 10112 10113 10114
*
*
*
$ASET1 23 20500 20501 30500 30501
$ASETl 3 20600 20601 30600 30601
$
$ To include t r a n s v e r s e s t i f f n e s s and f r i c t i o n e f f e c t s f o r SUBCASE 2
$ and 3 , r e p l a c e t h e above PGAP card with t h e following PGAP card.
$
$ Uo Fo Ka Kb Kt mu-y mu-z
$PGAP 15 0.001 .10 1.E9 1.E2 1.E4 0.25 0.25
ENDDATA
11.6 IMPACT ANALYSIS IN SOL 99: STRESS WAVE
PROPAGATION IN AN ELASTIC ROD

11.6.1 Problem Description

A rod with a fixed end is struck by a moving mass at the other end. Let vo be the initial
velocity of the mass which is considered rigid. The velocity of particles at the end of the rod
at the instant of impact (t = 0) is vo. Determine the time history of stress and displacement
at various cross sections of the rod during the impact. Find the duration of the impact and
the maximum stress in the rod for various values of mass ratio, M / m .

Rod mass = m

L = 100.0 m, rod length

A = 1.0 m2, cross-sectional area of the rod
E = lo3 N/m2, Young's Modulus
v = 0.3, Poisson's ratio
p = 0.1 Kg/m3, mass density
m = pAL = 10.0 Kg, rod mass
11.6.2 Theoretical Solution

Let M be the mass of the moving body and m be the total mass of the rod. Denoting the
stress at the free end of the rod by a and the velocity of the body by v, we have

where c = is the velocity of sound in rod material. As such, the stress in the wave is
determined by the ratio of the two velocities. The initial compressive stress in the rod is

00 = 210- (11.6.2)
Considering the force balance at the free end of the rod during the impact, the equilibrium
eauation can be written as

where A is the cross-sectional area of the rod.

Let the time interval ( T )for the stress wave, traveling a complete cycle in the rod, be
2L
T=- (11.6.4)
C

Solving the equilibrium equation for stress at the end struck in various time intervals, we
obtain [11.15]

o = So(t) for O<t<T

a = Sl(t)+So(t-T) for T s t <2T
o=S2(t)+Sl(t-T) for 2T5t<3T

etc.

where

S o ( t ) = ooe - 2 ( m l M ) ( t l T )

Since the stress wave is reflected when it arrives at the fixed end of the rod, the stress value
is twice the intensity of the incident wave. The general expression for the stress a t the fixed
end during any time interval (12 t n)T 5 t < ( $ + n)T is therefore

11.6.3 MSCINASTRAN Solution

The rod is modeled with twenty CONRODs and a GAP element. One end of the rod is
constrained completely and the other is left free axially. One CONM2 is used to represent
the rigid mass, and the mass value is varied for various ratios of m t o M. A GAP element
connects the CONM2 and the free end of the rod with a zero initial opening. An initial velocity
of vo = -O.lm/sec is specified for both the CONM2 and the free end nodal point using TIC
card. A time step is chosen so that twenty vibration modes of the rod model can be adequately
represented. A small damping (0.4% at resonance) is introduced in the rod elements t o smooth
out the undesired oscillations. The consistent mass matrix option (PARAM, COUPMASS) is
recommended for more stable responses.

11.6.4 Comparison of Results

The instant when the stress a t the free end (or force in the GAP element) vanishes signifies
the end of the impact. The durations of impact for four distinct values of M/m are determined
from MSCINASTRAN analyses and compared with the calculations of Saint-Venant in the
following table.

Table 11.6.1 Duration of Impact (sec)

I Theory 131168 F 7 0 8 r 5 . 9 0 0 1 7.419

% Difference 1 0.1 % / 0.0 % 13.56 %* 0.01 %/
* The discrepancy is explained below.

The MSCINASTRAN results given in the above table are obtained using the prescribed
model of 20 CONRODs. They match the theoretical values closely except for M/m = 4. It
was discovered, however, that this appreciable discrepancy is caused by a limitation of the
theoretical solution. The theoretical solution is valid only if the mass body stays in contact
with the rod. With a more refined model, MSCINASTRAN reveals that there is a momen-
tary opening near the theoretical solution (5.86 seconds in MSCINASTRAN) while the real
departure of the mass body from the rod occurs at 6.7 seconds. Thus, we may conclude that
MS.C/NASTRAN predicts the true end of impact.
 Time histories of displacement at various grid points and stresses at various elements are
obtained from MSC/NASTRAN analyses for several different values of M/m. Illustrations in
Fig. 11.6.1 and Fig. 11.6.2 are typical examples of the transient responses associated with the
free end of the rod. A plot of displacement vs. time in Fig. 11.6.1 for the mass point and the
free end point clearly shows a good departure of the mass body, consistent with the theory.
In Fig. 11.6.2, a stress vs. time plot for the element at the free end is given in comparison t o
the theoretical one for the end point. It should be noted that due to the stress gradients in
both time and space, the element stress output can not reproduce the theoretical stress curve
exactly at a grid point.

The maximum stress during impact always occurs at the fixed end, but in different time
intervals, for different values of Mlm. This shifting in time interval causes the discontinuity of
the slope of amax/aovs. M/m curve shown in Fig. 11.6.3. In general, element stresses are lower
than the theoretical values due t o discretization error as expected. For each continuous and
smooth curve segment, the MSCjNASTRAN result improves in agreement with the theoretical
solution as it approaches the end of the segment. It is shown that a refined model improves
the results slightly.

11.6.5 Input Data Listing

NASTRAN BUFFPOOL=37
NASTRAN SYSTEM(66)=112
INIT MASTER(S1
INIT SCRATCH(MEM=150)
ID HAMMER,V66 $ SSH 6/22/87, SHL 5/28/87 $ TKT 5-10-1985, SHL 6/4/85
TIME 30 $ FOR VAX
SOL 99 $ NONLINEAR TRANSIENT ANALYSIS
DIAG 8,50 $ PRINT MATRIX TRAILERS AND ITERATIONS
CEND
TITLE= TRANSIENT RESPONSE OF SHOCK WAVE IN BAR -- IMPACT
SUBTITLE= BAR WITH FIXED END STRUCK BY A MOVING MASS AT THE OTHER END
SET 1=21,99
SET 2=101,120,899
SET 3=99,21
SEALL=ALL
DISPLACEMENT=I
VELOCITY=3
STRESS-2
SUBCASE 1 $ UP TO 3 SEC,
IC=I
TSTEPNL=10
SUBCASE 2 $ UP TO 5 SEC.
TSTEPNL=20
$SUBCASE 3
$ TSTEPNL=30
OUTPUT(XYPLOT)
CSCALE=1.3
XAXIS=YES
YAXIS=YES
XTITLE=TIME
YTITLE=DISPLACEMENT
TCURVE= DISP.(T1) AT MASS PT.(GP99),FREE END(GP21)
XYPLOT DISP /99(T1) ,21(T1)
YTITLE=STRESS
TCURVE= STRESS AT FREE END (ELEMENT 120)
XYPLOT STRESS /120(2)
TCURVE= STRESS AT FIXED END (ELEMENT 101)
XYPLOT STRESS /I01(2)
TCURVE= FORCE IN THE GAP (ELEMENT 899)
YMAX=2.
XYPLOT STRESS /899(2)
BEGIN BULK
$.......2.......3.......4.......5.......6.......7.......
GRID 1 0.0 0.0 0.0
GRID 2 5. 0.0 0.0
GRID 3 10. 0.0 0.0
GRID 4 15. 0.0 0.0
GRID 5 20. 0.0 0.0
GRID 6 25. 0.0 0.0
GRID 7 30. 0.0 0.0
GRID 8 35. 0.0 0.0
GRID 9 40. 0.0 0.0
GRID 10 45. 0.0 0 .O
GRID 11 50. 0.0 0.0
GRID 12 55. 0.0 0.0
GRID 13 60. 0.0 0.0
GRID 14 65. 0.0 0.0
GRID 15 70. 0.0 0.0
GRID 16 75. 0.0 0.0
GRID 17 80. 0.0 0.0
GRID 18 85. 0.0 0 .O
GRID 19 90. 0.0 0.0
GRID 20 95. 0.0 0.0
GRID 21 100. 0.0 0.0
GRID 99 100. 0.0 0.0
$ ROD ELEMENTS
CONROD 101 1 2
CONROD 102 2 3
CONROD 103 3 4
CONROD 104 4 5
CONROD 105 5 6
CONROD 106 6 7
CONROD 107 7 8
CONROD 108 8 9
CONROD 109 9 10
CONROD 110 10 11
CONROD ill 11 12
CONROD 112 12 13
CONROD 113 13 14
CONROD 114 14 15
CONROD 115 15 16
CONROD 116 16 17
CONROD 117 17 18
CONROD 118 18 19
CONROD 119 19 20
CONROD 120 20 21
$ GAP ELEMENT
CGAP 899 90 21
PGAP 90
$ MASS ELEMENT
CONM2 999 99
$ INITIAL CONDITION
TIC 1 99 1
TIC 1 21 1
$ MATERIAL PROPERTY OF ROD
MAT1 100 1 .E+3
PARAM W4 1.570796
PARAM COUPMASS1
$ TIME STEP DEFINITION
TSTEPNL 10 120 0.025 AUTO
TSTEPNL 20 80 0.025 AUTO
TSTEPNL 30 75 0.010 AUTO
ENDDATA
Figure 11.6.1. Displacement vs. Time for M l m = 1.
Figure 11.6.2 Comparison of Free End Stress
(Time History for M / m = 1)
5.

4 ..
- 'Exact' Solution
A 20 CONROO Yodel
A Refined Yodel

Mass R a t i o M/m

Figure 11.6.3. Comparison of Impact Stresses.

11.6 - 9
11.7 EFFECTS OF KGROT ON GEOMETRIC
STIFFENING OF THE THIN SHELL MODEL

11.7.1 Theoretical Basis for Parameter KGROT

Stiffness is not defined for the rotation normal to the surface of shell elements (QUAD4 and
TRIA3). Unlike linear analysis, this degree of freedom can not be constrained in the geomet-
ric nonlinear analysis. It is noted that QUADR and TRIAR elements have stiffness in this
direction, but they are not equipped with nonlinear features. In order t o remove the singu-
larity in the out-of-plane rotation (the sixth DOF), an artificial stiffness Ke, proportional to
the user-specified parameter KGROT, is provided for the normal rotation in the QUAD4 and
TRIA3 shell elements [11.16]. The pseudo stiffness Ke is added by the parameter K6ROT to
the relative rotation of the element t o grid rotation as follows:

I<* for (a, - 8,) = * G * t * I JI * K6ROT

with
0, = rotation of a GRID from global displacement

0, = 5
I
(avz- $); rotation measured in the element
where G is the shear modulus, t is the thickness, and J is the Jacobian, representing the area
of the element. The requirements for Ke are
The unit should be in moments.

Pass the constant strain patch test.

No effect on the rigid body rotation.

Insensitive to the mesh size.

The value should be small enough t o maintain accuracy.

11.7.2 Difficulties in the Thin Shell Model

The thin shell structure is flexible in the lateral direction a t the beginning of the analysis. The
differential stiffness develops as the deformation proceeds, and the shell structure becomes
progressively stiffer due t o the nonlinear geometric effect. Geometric stiffening of thin shells
often poses difficulties in convergence during the initial loading stage. The difficulties arise from
ill-conditioned stiffness due t o a very large ratio between bending and membrane stiffnesses.
Consider the stiffness of a (L x L x t ) quadrilateral element:

Et3
kb 0: - and k, cc E t
L2
Thus the membrane t o bending stiffness ratio ( k , / k b ) is dictated by ( L / t ) 2 . As such, the
lateral pressure on the thin shell causes excessively large rotations due to bending of the
initially soft stiffness. The large rotation could easily involve a larger displacement than 1.57
radian (equivalent t o 90') for rotation, which causes singularity in the gimbal angle calculation.

The gimbal angle for large rotations is applicable to the QUAD4 and TRIA3 elements. At
each iteration, the gimbal angle is accumulated as

68, +
= (A8, sin 8, A8, cos 8,)/ cos 8,
68, = A@,cos 8, - A8, sin 8,
68, = AO, + +
[(AO, sin 8, At), cos 8,)/ cos O,] * sin 8,
where A8 is the incremental rotation in the global coordinate system and SO is the net incre-
mental gimbal angle. A mathematical singularity occurs when the incremental gimbal angle
component 8, = zf90°. If l8,l 2 90°, the user warning message "Gimbal Angle Exceeds 90
Degrees" is issued. This is usually caused by numerical ill-conditioning during the nonlinear
iterations rather than by physical rotation exceeding 90'. The problem is most likely aggra-
>
vated beyond the occurance of (8,l 90' and the solution diverges. This limitation has been
eliminated in Version 67 by introducing a set of auxiliary angles t o avoid division by cos 90'.

The divergent process can be contained by using a proper value of KGROT. This is because
excessive 0, rotation of the thin shell structure propagates to the sixth DOF, even though there
is no normal rotation in the physical problem, and causes excessive normal rotation. KGROT,
which provides a stiffness for the normal rotation, can favorably affect the iterative process.
Recommendations t o alleviate the convergence difficulties when the gimbal angle exceeds 90
degrees are:
r If the physical problem specifies the rotation of 8, beyond 90°, the coordinate system
should be changed so that 8, 5 90' (this limitation is removed in Version 67).

r If the problem diverges due t o numerical condition, a smaller load increment with a
proper value of KGROT should be used.

11.7.3 Parametric Studies

Four problems are examined t o study the effect of the parameter K6ROT on the iterative
process of thin shell structures subjected t o geometric stiffening. The problems are
r QUAD4 model of rectangular plate with line load

r TRIA3 model of rectangular plate with line load

r QUAD4 model of square plate with uniform pressure

QUAD4 model of square plate with concentrated load

SOL 66 in Version 66 is used in the analysis of the problems. The main feature of these problems
is geometric nonlinearity due t o geometric stiffening. The effects of geometric nonlinearity are
included in the analysis by specifying PARATVI, LGDISP,l.

The convergence and efficiency characteristics of the solution process are observed in or-
der t o determine the optimal range of values for KGROT. To obtain an even distribution of
converging and diverging solutions for each problem, the value of K6ROT is varied from 1 to
lo9, incremented by a factor of 10 in each run. The performance (i.e., the number of stiff-
ness updates, iterations, and line searches) of the successful executions is presented for the
corresponding KGROT value.

11.7.4 QUAD4 Model of Rectangular Plate with Line Load

A rectangular plate has dimensions of 200 mm x 400 mm and a thickness of 0.5 mm. The
plate is simply supported at the 200 mm sides and is subjected t o a line load of 400 N/mm
acting in the -2 direction as shown in Fig. 11.7.1.

The plate is modeled by 8 QUAD4 elements with linear elastic material. The simply
supported boundary condition is imposed by constraining components 1, 2, and 3 of the GRID
points at the 200 mm sides of the plate. The line load a t GRID 2, 5, and 8 is specified using
FORCE and LOAD Bulk D a t a entries. The input d a t a are given in Table 11.7.1 for K6ROT
= lo4.

Figure 11.7.1 QUAD4 Model of Rectangular P l a t e w i t h Line Load.

 Table 11.7.1 Input Data Listing for QUAD4 Model with Line Load
ID TKGROTI, V66 $ SHL 3/16/90, TLL 4/21/89
TIME 5 $ CPU MINUTES
SOL 66 $ NONLINEAR STATIC ANALYSIS
DIAG 50
CEND
TITLE = S.S. RECTANGULAR PLATE WITH 8 QUAD4 ELEMENTS AND LINE LOAD
SEALL = ALL
DISP = ALL
OLOAD = ALL
SUBCASE 1
NLPARM = 10
LOAD = 100
SUBCASE 2
NLPARM = 20
LOAD = 200
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
SET 1 = ALL
MAXI DEFO = 1.
VIEW 30., 20., 0.
FIND SCALE ORIGIN 1 SET 1
PTITLE = ISOMETRIC VIEW
PLOT SET 1 ORIGIN 1 SYMBOL 1 LABEL GRID
PLOT STATIC 0 MAXI DEFO 1. SET 1 ORIGIN 1 SYMBOL 1 LABEL GRID
BEGIN BULK
$$ PARAMETERS
PARAM KGROT l.E+4
PARAM LGDISP I
$
NLPARM 10 4 AUTO YES
NLPARM 20 5 AUTO YES
$$ GEOMETRY
GRID 1
GRID 2
GRID 3
GRID 4
GRID 5
GRID 6
GRID 7
GRID 8
GRID 9
GRID 10
GRID 11
GRID 12
GRID 13
GRID 14
GRID 15
$$ CONNECTIVITY
CQUAD4 I I I
CqUAD4 2 I 2
CQUAD4 3 I 4
CqUAD4 4 I 5
CQUAD4 5 1 10
CQUAD4 6 I 11
CQUAD4 7 I 14
CqUAD4 8 I 15
$$ PROPERTIES
MAT1 10 2.07E+5
PSHELL I 10 .5
$$ LOADING
FORCE 100 2
FORCE 100 5
FORCE 100 8
LOAD 200 I. 10.
$
ENDDATA
 Table 11.7.2 summarizes the results of the KGROT parametric study. The solution con-
verges only for KGROT = lo4. For K6ROT >lo5, the solution diverges at the first load
increment. The analysis stops for KGROT = l o 2 because the maximum number of iterations
is reached in the first increment of the second subcase. For the remaining KGROT values of
1, 10, and lo3, the analysis stops due to a diverging solution. For this problem, the optimum
value of KGROT is lo4. The load-deflection curve of the plate at the center (GRID 8) is given
in Fig. 11.7.2.

Table 11.7.2 Results of Parametric Study for KGROT:

QUAD4 Model with Line Load
( K Updates/Iterations/Line Searches )
KGROT
INC LOAD 1 10 lo2 lo3 104** u8

Notes : 1. Discrepancies in U8 with different values of K6ROT is less than 0.0003% at *.

2. Solution diverges a t the first increment beyond K 6 ~ 0 T = 1 0 ' .
3. ** indicates the most effective case.
 6,
a-
*,
63
m-
-
m -
63-
E a-
Em-
\
L
--
-a-
m-
COW-

c,
na -
-
CS-
-0
0
--
n -
J a -
m-
0
"
C --
-*
(D-
7
-_
6)-
m

, ~ . . ~ " " ~ " " l " ' ' l l ' ' ' l l ' " 1 " " 1 '
>L
El 5 1D 15 dW L - 3U jL

C)e f 1 ec t i or1 CI t 8 I IIIIII

Figure 11.7.2 Load-Deflection at Center of QUAD4 Model with Line ~ o a d .

11.7.5 TRIA3 Model of Rectangular Plate with Line Load

This problem was chosen t o be identical to the previous one, but with TRIA3 instead of
QUAD4 elements in the model. T h e loading had t o be reduced, however, because the solution
did not converge with the original loading of 400 N/mm. Fig. 11.7.3 shows the rectangular
plate with dimensions of 200 mm x 400 mm and a thickness of 0.5 mm. The plate is simply
supported a t the 200 mm sides and is subjected t o a line load of 140 N/mm acting in the - z
direction.

The plate is modeled by 16 TRIA3 elements with linear elastic material. The -simply
supported boundary condition a t the 200 mm sides of the plate is imposed by constraining
components 1, 2, and 3 of the GRID points. The line load a t GRID 2, 5, and 8 is specified
using FORCE and LOAD Bulk Data entries. Table 11.7.3 shows the input d a t a for KGROT
= lo2.

Figure 11.7.3 TRIA3 Model of Rectangular Plate with Line Load.

 Table 11.7.3 Input Data Listing for TRIA3 Model with Line Load
ID TK6ROT2, V66 $ CSK 3/16/90
TIME 5 $ CPU MINUTES
SOL 66 $ NONLINEAR STATIC ANALYSIS
DIAG 50
CEND
TITLE = S.S. RECTANGULAR PLATE WITH 16 TRIA3 ELEMENTS AND LINE LOAD
SEALL = ALL
DISP = ALL
OLOAD = ALL
SUBCASE I
NLPARM = 10
LOAD = 100
SUBCASE 2
NLPARM = 20
LOAD = 200
OUTPUT (PLOT)
CSCALE 1.3
PLOTTER NAST
SET I = ALL
MAXI DEFO = 1.
VIEW 30., 20., 0.
FIND SCALE ORIGIN I SET I
PTITLE = ISOMETRIC VIEW
PLOT SET I ORIGIN I SYMBOL I LABEL GRID
PLOT STATIC 0 MAXI DEFO I. SET I ORIGIN I SYMBOL I LABEL GRID
BEGIN BULK
$$ PARAMETERS
PARAM K6ROT 1 .E+2
PARAM LGDISP I
$
NLPARM 10 4 AUTO I -50 YES
NLPARM 20 5 AUTO I -50 YES
$$ GEOMETRY
GRID I
GRID 2
GRID 3
GRID 4
GRID 5
GRID 6
GRID 7
GRID 8
GRID 9
GRID 10
GRID I1
GRID 12
GRID 13
GRID 14
GRID 15
$$ CONNECTIVITY
CTRIA3 I I I
CTRIA3 2 I I
CTRIA3 3 I 10
CTRIA3 4 I 10
CTRIA3 5 I 2
CTRIA3 6 I 2
CTRIA3 7 I II
CTRIA3 8 I II
CTRIA3 9 I 13
CTRIA3 10 I 13
CTRIA3 I1 I 8
CTRIA3 12 I 8
CTRIA3 13 I 12
CTRIA3 14 I 12
CTRIA3 15 I 7
CTRIA3 16 I 7
$$ PROPERTIES
MAT1 10 2.07E+5
PSHELL I 10 .5
$$ LOADING
FORCE 100 2
FORCE 100 5
FORCE 100 8
LOAD 200 I. 10.
$
ENDDATA
 The results of the KGROT parametric study are summarized in Table 11.7.4. The solution
converges for K6ROT values of 10, l o 2 , lo3, and lo5. The analysis stops due t o a diverging
solution a t the first load increment for K6ROT >
lo6 and the remaining K6ROT values of 1
and lo4. In this problem, the optimum value of KGROT is l o 2 because it is the most efficient
in terms of the number of stiffness updates, iterations, and line searches. The load-deflection
curve of the plate a t the center (GRID 8) is given in Fig. 11.7.4.

Table 11.7.4 Results of Parametric Study for KGROT:

T R I A 3 Model with Line Load
( K Updates/Iterations/Line Searches )
K6ROT
INC LOAD 1 10 lo2** 1o3 lo4 lo5 u
8

Notes : 1. Discrepancies in U8 with different values of KGROT is less than 0.011% at *.

2. Solution diverges a t the first increment beyond K6ROT=106.
3. ** indicates the most effective case.
 Clef lect iut-I i l l 8 k 111111 I

Figure 11.7.4 Load-Deflection at Center of TRIA3 Model with Line Load.

11.7.6 QUAD4 Model of Square Plate with Uniform Pressure

Fig. 11.7.5 shows a square plate which has a length of 400 mm and a thickness of 0.5 mm. The
plate is simply supported a t three of its sides and is subjected t o a uniform pressure loading
of 1 N/mm2 applied normal t o its plane.

The plate is modeled by 16 QUAD4 elements with linear elastic material. The simply
supported boundary condition is imposed by constraining components 1, 2, and 3 of the G R I D
points a t three sides of the plate. The uniform pressure loading is specified using PLOAD2
and LOAD Bulk Data entries. The input d a t a are given in Table 11.7.5 for K6ROT = lo3.

Figure 11.7.5 QUAD4 Model of Square P l a t e with Uniform Pressure.

11.7 - 13
 Table 11.7.5 Input Data Listing for QUAD4 Model with Uniform Pressure
ID TK6ROT3, V66 $ RA 3/7/90
TIME 5 $ CPU MINUTES
SOL 66 $ NONLINEAR STATIC ANALYSIS
DIAG 50
CEND
TITLE = S.S. SQUARE PLATE WITH 16 QUAD4 ELEMENTS AND UNIFORM PRESSURE
SEALL = ALL
DISP = ALL
OLOAD = ALL
SUBCASE I
NLPARM = 10
LOAD = 100
SUBCASE 2
NLPARM = 20
LOAD = 200
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
SET I= ALL
MAXI DEFO = 1.
VIEW 30., 20., 0.
FIND SCALE ORIGIN I SET I
PTITLE = ISOMETRIC VIEW
PLOT SET I ORIGIN ISYMBOL ILABEL GRID
PLOT STATIC 0 MAXI DEFO I.SET I ORIGIN 1 SYMBOL I LABEL GRID
BEGIN BULK
$$ PARAMETERS
PARAM K6ROT I .E+3
PARAM LGDISP I
$
NLPARM 10 4 AUTO YES
NLPARM 20 5 AUTO YES
$$ GEOMETRY
GRID 1
GRID 2
GRID 3
GRID 4
GRID 5
GRID 6
GRID 7
GRID 8
GRID 9
GRID 10
GRID 11
GRID 12
GRID 13
GRID 14
GRID 15
GRID 16 0.
GRID 17 100.
GRID 18 200.
GRID 19 300.
GRID 20 400.
GRID 21 0.
GRID 22 100.
GRID 23 200.
GRID 24 300.
GRID 25 400.
$$ CONNECTIVITY
CQUAD4 I I I
CQUAD4 2 I 2
CQUAD4 3 1 3
CQUAD4 4 I 4
CQUAD4 5 1 6
CQUAD4 6 I 7
CQUAD4 7 I 8
CQUAD4 8 I 9
CQUAD4 9 I II
CQUAD4 10 I 12
CQUAD4 I1 1 13
CQUAD4 12 I 14
CQUAD4 13 1 16
CQUAD4 14 I 17
CQUAD4 15 I 18
CqUAD4 16 I 19
$$ PROPERTIES
MAT1 10 2.07E+5
PSHELL I 10 .5
$$ LOADING
PLOAD2 100 -.I I THRU
LOAD 200 I. 10. 100
$
ENDDATA
 Table 11.7.6 summarizes the results of the K6ROT parametric study. The solution diverges
at the third load increment of the second subcase for K6ROT = lo5 and at the first load
increment of the first subcase for K6ROT 2 lo6. The analysis stops at the first load increment
for K6ROT = 1 when the maximum number of iterations is reached. For the remaining K6ROT
values of 10 through lo4, the solution converges. The optimum value of K6ROT is lo3 in this
problem. The load-deflection curve of the plate at GRID 23 is given in Fig. 11.7.6.

Table 11.7.6 Results of Parametric Study for K6ROT:

QUAD4 Model with Uniform Pressure
( K Updates/Iterations/Line Searches )
K6ROT
INC LOAD 1 10 lo2 lo3** lo4 lo5 u23

Notes : 1. Discrepancies in UZ3with different values of K6ROT is less than 0.0087% at *.

2. Solution diverges at the first increment beyond K 6 ~ 0 T = 1 0 ~ .
3. ** indicates the most effective case.
 Dei I E : C t I C I ~ I (IIITII 1

Figure 11.7.6 Load-Deflection at Grid 23 of QUAD4 Model with Uniform Pressure.

11.7.7 QUAD4 Model of Square Plate with Concentrated Load

This problem is similar to the previous one, but has a different loading condition. A square
plate has a length of 400 mm and a thickness of 0.4 mm. T h e plate is simply supported on
three sides and is subjected t o a concentrated load of 13,200 N acting in the -2 direction as
shown in Fig. 11.7.7.

The plate is modeled by 16 QUAD4 elements with linear elastic material. The simply
supported boundary condition is imposed by constraining components 1, 2, and 3 of the GRID
points at three sides of the plate. The concentrated load a t GRID 19 is specified using FORCE
and LOAD Bulk D a t a entries. Table 11.7.7 shows the input d a t a for K6ROT = lo3.

Figure 11.7.7 QUAD4 Model of Square Plate with Concentrated Load

11.7 - 18
 Table 11.7.7 Input Data Listing for QUAD4 Model with Concentrated Load
ID TK6ROT4, V66 $ RA 3/7/90
TIME 5 $ CPU MINUTES
SOL 66 $ NONLINEAR STATIC ANALYSIS
DIAG 50
CEND
TITLE = S.S. SQUARE PLATE WITH 16 QUAD4 ELEMENTS AND CONCENTRATED LOAD
SEALL = ALL
DISP = ALL
OLOAD = ALL
SUBCASE 1
NLPARM = 10
LOAD = 100
SUBCASE 2
NLPARM = 20
LOAD = 200
OUTPUT(PLOT)
CSCALE 1.3
PLOTTER NAST
SET I = ALL
MAXI DEFO = I.
VIEW 30., 20., 0.
FIND SCALE ORIGIN 1 SET I
PTITLE = ISOMETRIC VIEW
PLOT SET I ORIGIN I SYMBOL I LABEL GRID
PLOT STATIC 0 MAXI DEE0 I. SET I ORIGIN I SYMBOL I LABEL GRID
BEGIN BULK
$$ PARAMETERS
PARAM K6ROT l.E+3
PARAM LGDISP I
$
NLPARM 10 4 AUTO YES
NLPARM 20 5 AUTO YES
$$ GEOMETRY
GRID I
GRID 2
GRID 3
GRID 4
GRID 5
GRID 6
GRID 7
GRID 8
GRID 9
GRID 10
GRID 11
GRID 12
GRID 13
GRID 14
GRID 15
GRID 16 0.
GRID 17 100.
GRID 18 200.
GRID 19 300.
GRID 20 400.
GRID 21 0.
GRID 22 100.
GRID 23 200.
GRID 24 300.
GRID 25 400.
$$ CONNECTIVITY
CQUAD4 I 1 I
CQUAD4 2 1 2
CQUAD4 3 I 3
CQUAD4 4 I 4
CQUAD4 5 I 6
CQUAD4 6 I 7
CQUAD4 7 1 8
CQUAD4 8 I 9
CQUAD4 9 I II
CQUAD4 10 I 12
CQUAD4 I1 I 13
CQUAD4 12 1 14
CQUAD4 13 1 16
CQUAD4 14 1 17
CQUAD4 15 I 18
CQUAD4 16 I 19
$$ PROPERTIES
MAT1 10 2.07E+5
PSHELL I 10 .4
$$ LOADING
FORCE 100 19
LOAD 200 I. 6.
$
ENDDATA
 The results of the K6ROT parametric study are summarized in Table 11.7.8. The results
are similar t o those of the previous problem of a square plate with a uniform pressure. The
solution converges for K6ROT values of 10, lo2, lo3, and lo4. The analysis stops at the first
load increment for K6ROT = 1 and K6ROT = l o 5 because the solution does not converge
in MAXITER iterations. For K6ROT 2 lo6, the analysis stops at the first load increment
due to a diverging solution. The optimum value of K6ROT is also lo3 in this problem. The
load-deflection curve of the plate a t GRID 19 is shown in Fig. 11.7.8.

Table 11.7.8 Results of Parametric Study for KGROT:

QUAD4 Model with Concentrated Load
( K Updates/Iterations/Line Searches )
K6ROT
-

INC LOAD 1 10 lo2 lo3** lo4 lo5 u19

1-1 550 MAXITER 3/38/18 4/38/18 3/39/16 4/38/20 MAXITER 6.4118

1-2 1100 1/15/1 1/15/1 1/15/1 1/16/1 8.095

Notes : 1. Discrepancies in U19 with different values of K6ROT is less than 0.12% at *.
2. Solution diverges a t the first increment beyond K 6 R 0 ~ = 1 0 ~ .
3. ** indicates the most effective case.
Figure 11.7.8 Load-Deflection at Grid 19 of QUAD4 Model with Concentrated Load.
11.7.8 0bservations

At the beginning of the analysis, the plate is flexible. As the deformation increases, however,
the plate becomes progressively stiffer due to the nonlinear geometric effect. Bending of the
initially soft plate causes excessive rotation which propagates t o the sixth DOF. This excessive
rotation in the sixth DOF is alleviated by a larger value of KGROT. Therefore, small values
of KGROT (< lo2) make the solution convergence very difficult. In fact, the solution did not
converge for KGROT = 1 in any of these test problems. Values of KGROT that are too large ( 2
lo6) also cause the solution to diverge in all of the problems. Since an excessively large value
of K6ROT will lock the varying strain within an element by enforcing the rotation measured in
the element (R,) to conform t o the GRID point rotation (O,), the solution will be inaccurate
even if it converges with a large KGROT value. The parametric study thus indicates that the
iterative process in geometric stiffening is highly sensitive to KGROT, whose proper value can
contain the divergent process. The optimal range of values is found to be lo2 2 KGROT 2 lo4.
11.8 ADAPTATION OF CREEP ANALYSIS CAPABILITY
TO GENERAL VISCOELASTIC MATERIALS

11.8.1 Introduction

MSCINASTRAN's creep capability was developed based on the Maxwell-Kelvin rheological

model, which can be applied t o any viscoelastic materials [11.17,11.18]. To use this capability,
users need t o define the creep characteristics either by empirical creep laws in the form of

with

or by defining the rheological creep parameters (C,, Cp and ICp) as functions of stresses, so
that the creep strain is represented by

If the user should model a material that is characterized by experimental data or by a creep
law different from what is provided in MSCINASTRAN, some kind of curve fitting is required
t o find equivalent coefficients or parameter values corresponding t o the predefined empirical
law (types 111 through 222) or the Maxwell-Kelvin rheological model.

The least square fit is described t o determine rheological parameters (Cs,Cp and Ii, as
functions of stresses) corresponding t o the Findley's law representing the creep behavior in
plastics as an example. An interactive program is developed t o perform this numerical anal-
ysis procedure. This procedure should be followed in any similar application of the general
viscoelastic or creep analysis.

11.8.2 Formulation for Least Square Fit of Rheological Parameters

Let us determine the parameter values (C,, Cp and K,) at a constant stress ( a ) for the Maxwell-
Kelvin rheological model represented by

which should simulate the behavior of plastics expressed by the Findley's law,
u u u
cf = EO sinh -
00
+ &ttnsinh -
at
--
E
 n = 0.19
EO = 0.0067
Et = 0.0011
a. = at = 8500 psi
E = 1.2 x lo6 psi

Using the data points (tj, a): on the reference curve in Eq. (11.8.2), the least square fit can
be employed t o minimize the error, i.e.,

min. error = x3
6;

where
sj = &$ - &3f
Eq. (11.8.4) should be linearlized with respect to C,,Cp and I<pto apply an iterative and
incremental process, i.e.,
c, = c; AC,+

The first order approximation of Eq. (11.8.4) is obtained by a Taylor series with a substitution
of Eq. (11.8.5), i.e.,

where
8eT
- - - --at
ac, c,z

In order t o satisfy the minimum error in Eq. (11.8.3), the following condition has to be
satisfied, i.e.,
When Eq. (11.8.6) is substituted into Eq. (11.8.8), there are three linear equations with
AC,, AC, and A& as unknown variables, i.e.,
a&; i A&] =
8" j
C[-(E~
ac, ac, ACp + EK j
ac, - h;)]

where the derivatives are defined in Eq. (11.8.7), and E~ and ~f in Eqs. (11.8.1) and (11.8.2),
respectively. By solving the simultaneous equations in Eq. (11.8.9) for AC,, AC, and Air',,
these unknown parameters can be updated by Eq. (11.8.5) for the next iteration.

The rheological parameters may be evaluated by the least square fit of the strain-rate
instead of the strain itself. For this approach, Eqs. (11.8.1), (11.8.2), (11.8.4), (11.8.6) and
(11.8.7) should be replaced by the following equations:
d
= - + -0e - ( K ~ l C ~ ) t (11.8.1b)
cs c,
0
if = wttn-' sinh - (11.8.2b)
dt
6 = iT- if (11.8.4b)

6 ?? iT- i*+ PAC,

aiT 8iT
+ PAC, atT
+ -A& (11.8.6b)
ac, acp ari,
aiT
- - --
0
(11.8.7b)
acs CS~
aiT -
- d A-
-[ A t - lle-(Kp/Cp)t
acP c,"CP
air
- - --Ot e - ( K ~ / c ~ ) t
a~r, cP2

11.8.3 Computational Process

The strain function in Eq. (11.8.1), which requires curve fitting, has two variables, namely,
stress and time. The experimental data are usually measured at constant stress levels, which
become the basis for the empirical formula or the curve fitting. Therefore, the least square
fit can be performed as formulated in the preceding section with data points in the operating
time domain while the stress level is fixed. Newton's iteration starts to minimize the error in
Eq. (11.8.3). The simultaneous equations in Eq. (11.8.9) are solved at every iteration and the
iteration continues until the convergence is achieved. Upon convergence, a set of rheological
parameters (C,, C, and K,) has been determined for a constant stress. The least square fit is
performed incrementally by sweeping along the entire operating stress range.

The iteration starts with initial values for Cs, C, and K,. Initial values should not be too
far from the solution for a good convergence. The initial values can be appropriately estimated
by considering a physical interpretation of the creep strain curve in relation to the rheological
model in Eq. (11.8.1). The slope at a long time span ( t f ) approximates the secondary creep
rate, i.e.,

from which

The total primary creep strain a t a long time span can be approximated by

from which -
The slope at t=O can be related t o Cp by

from which

For a = 1000 psi and tf = 500000 hours

1000
Ii;, N = 8.1427 x lo5 psi
1.526 x - 2.982 x
However, Cp cannot be evaluated by Eq. (11.8.12) because if approaches infinity at t = 0.
Instead, Cp may be estimated by using lip,C, and ~f a t some finite time span, e.g., for
t = 10000 hours and a = 1000 psi
11.8.4 Analysis Procedure and Results

The computation process, as presented in the preceding sections, has been coded. Table
11.8.5 shows the source listing of this program. The analysis starts by running the program
"RHEOLOGY", which is an interactive program. The program prompts six questions t o be
answered as shown in Table 11.8.1. These questions pertain to:
1. Least square fit on the strain curve or strain-rate curve (default = strain curve).

2. Title t o appear in the output (no default).

3. Error tolerance for the convergence criterion (default =

4. Maximum time for creep (default = 5 x lo5) and the number of data points to be used
for curve fitting (default = 200).

5. Maximum operating stress (default = 1500) and stress increments (default = 100)
6. Output file names for time vs. strain data (default = RHEOLOGY.OUT) and stress vs.
rheological parameters (default = RHEOLOGY.PAM).
Upon completion of the dialogue, the iterative and incremental computation process pro-
ceeds with the iteration information printed on the terminal screen as shown in Table 11.8.1.
Upon termination of the job, there will be two output files as shown in Tables 11.8.2 and 11.8.3,
respectively. Table 11.8.3 shows rheological parameter values (C,, Cp and Kp) as functions
of stress, which can be converted into MSC/NASTRAN data specified on the CREEP and
TABLES1 Bulk Data entries.

An interactive graphics program (DISPLY) is available t o plot curves for the data generated
by the program RHEOLOGY. Table 11.8.6 shows the source listing of the DISPLY program
(to be linked with Tektronix PLOT10 package).

The DISPLY program starts by prompting questions as shown in Table 11.8.4. The re-
sults of processing the file "RHEOLOGY.OUTV in Table 11.8.2 are shown in Fig. 11.8.1 and
11.8.2. Fig. 11.8.3, 11.8.4, and 11.8.5 show plots of rheological parameters saved in the file
"RHEOLOGY.PAMV in Table 11.8.3.

The least square fit on the strain-rate curve was attempted t o determine the rheological
parameters. This approach was not fruitful because the Hessian matrix became singular within
the operating time domain.

11.8.5 Recommendations

The rheological parameters corresponding t o any empirical creep law may be obtained by using
the program "RHEOLOGY". The listing in Table 11.8.5 shows FUNCTION EPSF defining
the Findley7s creep law in Eq. (11.8.2) and its time derivative. This FUNCTION has to be
redefined for a different type of empirical formula. It is also possible t o use experimental data
for the curve fitting in place of the empirical law with slight modifications t o the current version
of "RHEOLOGY".

The least square fit can also be applied to the creep laws defined by types 111 through 222
to be specified in the CREEP Bulk Data entry. For those creep laws, coefficients "a" through
"g" should be determined, rather than the rheological parameters, by the curve fitting. In this
case, the time and the stress must be treated as a unified variable. This means that all the
data points obtained by sweeping through the time domain, nested under the stress increments
across the operating stress range should be included in the iterative process for the least square
fit.
 Table 11.8.1 Run Progression of the Program "RHEOLOGY"
$RUN RHEOLOGY

* PERFORMS LEAST SQUARE FIT FOR RHEOLOGICAL PARAMETERS

** WRITTEN BY SANG H. LEE.

2. TYPE A TITLE(MAX. 40 CHARACTERS).

LEAST SQUARE FIT ON FINDLEY'S CREEP LAW FOR RHEOLOGICAL MODEL.
3. TYPE IN AN ERROR TOLERANCE FOR ITERATION. (DEFAULT= 1.E-6)
I.E-8
4. DEFINE AN X-AXIS IN TIME.
TYPE MAXIMUM TIME DOMA1N:TMAX (DEFAULT=5.E5)

HOW MANY POINTS WOULD YOU USE FOR CURVE FIT? (DEFAULT=200)

5. DEFINE THE RANGE OF SECOND VARIABLE, STRESS.

TYPE IN THE MAXIMUM OPERATING STRESS: SMAX (DEFAULT=I500)

6. OUTPUT FILE NAMES:

TYPE FILE NAME FOR TIME VS. STRAIN; DEFAULT=RHEOLOGY.OUT

TYPE FILE NAME FOR STRESS VS. CS/CP/KP; DEFAULT=RHEOLOGY.PAM

I LEAST SQUARE FIT AT STRESS 1.000000E+02

ITERATION CS ERROR
0 1.68046073+12 2.4587280E-08
I 1.4885375E+ii 5.4180528E-06
2 2.7227526E+11 1.078400iE-06
3 4.5875800E+iI 1.8038159E-07
4 6.7277285E+lI 2.2354318E-08
5 8.1780644E+ii 5.3207905E-09
6 8.5469646E+ii 4.7783750E-09
7 8.55766203+11 4.7770992E-09
8 8.5539796E+ii 4.7770010E-09

2 LEAST SQUARE FIT AT STRESS 2.000000E+02

ITERATION CS CP KP ERROR
0 1.6803444E+12 1.1390349E+iO 8.1731938E+05 9.83670033-08
I 1.4883239E+ii 1.20487013+10 9.6204513E+05 2.1678970E-05
 Table 11.8.2 StrainIStrain-Rate vs. Time (RHEOLOGY.OUT File Content)

I AT STRESS 1.0000E+02 CS= 8.5494E+lI CP= l.OllOE+lO KP= 1.0044E+06

TIME STRAIN/EMPIRICALRHEOLOGICAL STR-RATE/EMPIRICAL RHEOLOGICAL

2.500000E+03 5.271788E-05 2.218916E-05 4.349165E-09 7.832982E-09
5.000000E+03 6.077321E-05 3.956276E-05 2.480685E-09 6.136065E-09
7.500000E+03 6.600114E-05 5.317984E-05 1.786231E-09 4.812336E-09
1.000000E+04 6.996244E-05 6.386654E-05 1.414938E-09 3.779724E-09
1.250000E+04 7.318766E-05 7.226731E-05 1.180974E-09 2.974205E-09
1.500000E+04 7.592629E-05 7.888488E-05 1.018834E-09 2.345837E-09

2 AT STRESS 2.0000E+02 CS= 8.5488E+lI CP= 1.0108E+IO KP= 1.0042E+06

TIME STRAIN/EMPIRICAL RHEOLOGICAL STR-RATE/EMPIRICAL RHEOLOGICAL

2.500000E+03 1.054546E-04 4.4385583-05 8.698933E-09 1.566844E-08
5.000000E+03 1.2156643-04 7.913807E-05 4.961714E-09 1.227393E-08
7.500000E+03 1.3202303-04 1.0637613-04 3.572710E-09 9.625984E-09
1.000000E+04 1.399461E-04 1.2775233-04 2.830072E-09 7.560396E-09
1.250000E+04 1.463970E-04 1.4455593-04 2.3621113-09 5.9490923-09

3 AT STRESS 3.0000E+02 CS= 8.5476E+ii CP= 1.0103E+10 KP= 1.0040E+06

TIME STRAIN/EMPIRICAL RHEOLOGICAL STR-RATE/EMPIRICAL RHEOLOGICAL

2.500000E+03 1.5822903-04 6.6607143-05 1.3049913-08 2.351199E-08
5.OOOOOOE+03 1.8239943-04 1.187549E-04 7.443429E-09 1.841699E-08
7.500000E+03 1.980861E-04 1.596242E-04 5.359683E-09 1.444280E-08
 Table 11.8.3 C,/C,/I(, vs. Stress (RHEOLOGY.PAM File Content)

LEAST SqUARE FIT ON FINDLEY'S CREEP LAW FOR RHEOLOGICAL MODEL

 Table 11.8.4 Run Progression of the Program "DISPLY"

$RUN DISPLY
THIS IS A GENERAL PURPOSE PROGRAM FOR XYPLOT
WRITTEN BY SANG H. LEE.

TYPE IN BAUD RATE

I. TYPE A TITLE(MAX. 40 CHARACTERS).

DEFAULT IS TO READ THE FIRST LINE FROM INPUT FILE

2. NUMBER OF DATA POINTS?

DEFAULT WILL INCLUDE ALL THE DATA POINTS IN THE FILE

3. X-AXIS DEFINITION.
LINEAR OR LOGARITHMIC? TYPE LN OR LG

WHICH DOMAIN DO YOU WANT TO DISPLAY? TYPE XMIN,XMAX

TYPE A LABEL(MAX. 10 CHARACTERS).

TIME IN HOUR
4 . Y-AXIS DEFINITION.
LINEAR OR LOGARITHMIC? TYPE LN OR LG

RANGE FOR Y-AXIS WILL BE SET AUTOMATICALLY.

TYPE A LABEL(MAX. 10 CHARACTERS).
CREEP STRAIN
5 . GRID LINES? Y OR N.

6. OVERLAY TWO CURVES? Y OR N.

Y
TYPE I, 2, 3 OR 4 TO SPECIFY INPUT DATA FORMAT:
I IF INPUTS ARE X, Yi, Y2
2 IF INPUTS ARE Xi, X2, Y
3 IF INPUTS ARE XI, Yi, X2, Y2
4 IF INPUTS ARE X, DUMMY, DUMMY, Yl, Y2
I
IST CURVE WILL BE A SOLID LINE; 2ND WILL BE DASHED
TYPE A LEGEND FOR SOLID LINE.
FINDLEY
TYPE A LEGEND FOR DASHED LINE.
RHEOLOGY
7. TYPE THE FILE NAME.
NUMBER OF DATA POINTS= 200

HIT RETURN TO CONTINUE

8 Q)
Q
0
w
c(

QO
L:
.-
u r n
 Table 11S . 5 . Source Listing for the Program "RHEOLOGY"
C
C
C PROGRAM RHEoLOGY(INPUT,OUTPUT,TAPE~=OUTPUT)
L
C BY SANG H. LEE, 11/6/87
C
C THIS PROGRAM PROVIDES RHEOLOGICAL PARAMETERS KP, CP AND CS,
c CORRESPONDING TO THE GIVEN EMPIRICAL CREEP LAW (FINDLEY'S)
C USING THE LEAST SQUARES FIT.
C
COMMON / /TITLE(20)
CHARACTER NAME1*20,NAME2*2O,IBK*l
DATA IBK/' ' /
C INPUT DATA INTERACTIVELY
C
CALL INTERACT(NF ,ERROR,TMAX ,ANDP ,SMAX ,DSIGMA,NAMEl,NAME21
C
c ...................................................................
C
C DEFAULT VALUES ARE DEFINED HERE.
C
IF(ERROR.EQ.O.)ERROR=l.E-6
TMIN=O.
IF(TMAX.EQ.O.)TMAX=500000.
IF(ANDP.EQ.0.) ANDP=200.
DTIME=(TMAX-TMIN)/ANDP
TINIT=DTIME*4.
IF(SMAX.EQ.O.)SMAX=l500.
IF(DSIGMA.EQ.O.) DSIGMA=IOO.
C CREATE AN EXTERIAL FILE.
IF(NA~EI.EQ.IBK~NAME~=~RHEOLOGY.OUT~
IF(NAME~.EQ.IBK)NAME~=~RHEOLOGY.PAM~
OPEN(UNIT=7,FILE=NAMEl,STATUS=JNEW')
OPEN(UNIT=8,FILE=NAME2,STATUS='NEWJ)
C
C INPUT TITLE
WRITE(8,lOOi) (TITLE(1) ,I=1,20)
WRITE(8,1002)
1001 FORMAT(X ,2OA4)
1002 FORMAT(/SX,JSTRESS',10X,JCS'14X'CP'14X'KP'~
C NUMBER OF LINES
NL=O
SIGMA=O.
100 SIGMA=SIGMA+DSIGMA
IF(SIGMA.GT.SMAX)GO TO 200
NL=NL+I
CALL PROCESS(NL,NF,SIGMA,DTIME,THAX,TINIT,ERROR~
WRITE(7,1003)
 1003 FORMAT( 1H )
GO TO 100
200 CONTINUE
CLOSE(UNIT=7)
CLOSE(UNIT=8)
STOP
END
C
c ....................................................................
C
FUNCTION EPSF(N,STR,TIM)
C
C FINDLEY'S CREEP LAW AND ITS TIME DERIVATIVE
C
IF(N.EQ.l)THEN
EPSF=O.O067*SINH(STR/8500.)
1 +O.OOll*TIM**O.l9*SINH(STR/8500.)-STR/1.2E6
ELSE
EPSF=0.19*0.0011*TIM**(O.19-1.)*SINH(STR/8500.)
ENDIF
C
RETURN
ENTRY EPSFDT(STR, TIM)
EPSFDT=0.19*0.00ll*TIM**(O.l9-l.)*SINH(STR/8500.)
RETURN
END
C
c ...................................................................
C
FUNCTION EPSR(N,TIM)
COMMON/STHD/STRESS,CS,CP,KP
REAL KP
C
C RHEOLOGICAL MODEL
C
IF(N.EQ.l)THEN
EPSR=STRESS*TIM/CS+STRESS*(l.-EXP(-KP*TIM/CP))/KP
ELSE
EPSR=STRESS/CS+STRESS*EXP(-KP*TIM/CP)/CP
ENDIF
RETURN
C DERIVATIVES
ENTRY FUlF(N,TIM)
IF(N.EQ.l)THEN
FUlF=-STRESS*TIM/CS**2
ELSE
FUlF=-STRESS/CS**2
ENDIF
RETURN
 ENTRY FU2F(N,TIM)
IF(N.EQ.~)THEN
FU~F=-STRESS*TIM*EXP(-KP*TIM/CP)/CP**2
ELSE
FU~F=STRESS*(KP*TIM/CP-I.)*ExP(-KP*TIM/CP)/CP**~
ENDIF
RETURN
ENTRY FU~F(N,TIM)
1F(N.EQ.l)THEN
FU~F=STRESS*EXP(-KP*TIM/CP)*(~./KP+TIM/CP)/KP-STRESS/KP**~
ELSE
FU~F=-STRESS*TIM*EXP(-KP*TIM/CP)/CP**2
ENDIF
RETURN
END
C
C
SUBROUTINE INTERACT(NF, ERROR,TMAX ,ANDP ,SMAX ,DSIG ,NAME1 ,NAME21
CHARACTER NAME1*20,NAME2*2O
COMMON / /TITLE(~O)
DATA IY,IN,IBK/'Y','N',' ' /
WRITE(6,999)
999 FORMAT(/X1* PERFORMS LEAST SQUARE FIT FOR RHEOLOGICAL PARAMETERS'/
1 lOX,)** WRITTEN BY SANG H. LEE. I / )
C
C LEAST SQUARE FIT ON THE STRAIN OR STRAIN RATE?
C
NF=O
100 WRITE(6,998)
998 FORMAT(X,'I. LEAST SQUARE FIT ON THE STRAIN RATE?',
1 2x, Y OR B (DEFAULT) '
READ(S,IO13)IANS
IF(IANs.Eq.IBK)IANS=IN
IF(IANS.EQ.IY)NF=2
IF(IANS.EQ.IN)NF=~
IF(NF.EQ.0)GO TO 100
C INPUT TITLE
300 TYPE 1000
ACCEPT 1001,(TITLE(1) ,I=1,201
1000 FORMAT(X,'2. TYPE A TITLE(MAX. 40 CHARACTERS).')
1001 FORMAT(20A4)
C ERROR TOLERANCE
TYPE 1002
1002 FORMAT(X,'B. TYPE IN AFJ ERROR TOLERANCE FOR ITERATION.'
1 ,X,'(DEFAULT= 1.E-6)')
ACCEPT 1003,ERROR
1003 FORMAT(F20 .O)
C X-AXIS IS TIME.
TYPE 1004
 1004 FORMAT(X,'4. DEFINE AN X-AXIS IN TIME.')
20 TYPE 1005
1005 FORMAT(5XIJTYPEMAXIMUM TIME DOMA1N:TMAX (DEFAULT=5.E5l1)
ACCEPT 1003,TMAX
1006 FORMAT(2Fl5.0)
C DATA POINTS IN TIME
TYPE 1007
1007 FORMAT(5X1'HOW MANY POINTS WOULD YOU USE FOR CURVE FIT?'
1 ,2X' (DEFAULT=200)' )
ACCEPT 1003,ANDP
C SECOND VARIABLE IS STRESS
TYPE 1008
1008 FORMAT(X,'S. DEFINE THE RANGE OF SECOND VARIABLE, STRESS.')
TYPE 1009
1009 FORMAT(5X,*TYPE IN THE MAXIMUM OPERATING STRESS: SMAX'
1 ,2XJ(DEFAULT=1500)')
ACCEPT 1003, SMAX
TYPE 1010
1010 FORMAT(SX,'TYPE IN THE STRESS INCREMENTS SWEEPING STRESS RANGE'
1 ,2X'(DEFAULT=lOO)')
ACCEPT 1003,DSIG
C OUTPUT FILE NAMES FOR STRAIN VS. TIME AND CS/CP/KP VS. STRESS
TYPE 1014
1014 FORMAT(XIJ6. OUTPUT FILE NAMES:')
TYPE 1011
READ(5,1013)NAMEl
loll FORMAT(6XI1TYPEFILE NAME FOR TIME VS. STRAIN;'
1 X,'DEFAULT=RHEOLOGY.OUTJ)
TYPE 1012
READ(5,1013)NAME2
1012 FORMAT(5X,'TYPE FILE NAME FOR STRESS VS. CS/CP/KP;'
1 X'DEFAULT=RHEOLOGY.PAMJ)
1013 FORMAT(A)
RETURN
END
C
C
SUBROUTINE PROCESS(NL,NF,SIGMA,DELTIME,TMAX,TINIT,ERLMT)
COMMON/STHD/STRESS,CS,CP,KP
DIMENSION X(9Ol) ,Y(SOI) ,YRATE(901)
REAL KP
C
STRESS=SIGMA
I=O
X(I)=O.
C
10 I=I+l
X(I)=X(I-l)+DELTIME
IF(X(I) .GT.TNAX)GO TO 200
 NP=I
Y(I)=EPSF(~,STRESS,X(I)
YRATE(I)=EPsF(~,STRESS,X(I))
GO TO 10
C READ 1001,X(I) ,Y(I), J
C IF(J.EQ.O)GO TO 10
C NP=I
C 1001 FORMAT(2F10.0,50XD 110)
200 CONTINUE
WRITE(6,1000)NL, STRESS
CALL GUESS(TMAX ,TINIT)
IF(NF.EQ.1)THEN
CALL CURVFIT(NF,ERLHT,X,Y,NP)
ELSE
CALL CURVFIT(NF,ERLMT,X,YRATE,NP)
ENDIF
C
WRITE(7,999)NL,STRESSlCSDCPDKP
WRITE(8,1001)STRESSDCSDCPlKP
WRITE(7,1002)
999 FORMAT(I4.2X'AT STRESS',IPEII.4,
1 X,3HCS=,Eii.4,X,3HCP=,Eil.4,X,3RKP=,E11.4/)
1000 FORMAT(lHO,I5,3X,'LEAST SQUARE FIT AT STRESS' ,iPEi5.6/)
1001 FORMAT(5(1PE16.6))
1002 FORMAT(8XD'TIME STRAIN/EMPIRICAL RHEOLOGICAL',
1 2XD1STR-RATE/EMPIRICALRHEOLOGICAL')
C;

DO 99 I=i,IP
STRAIN=EPSR(I,x(I))
STRATE=EPSR(2,X(I))
WRITE(T,~OO~) X(1) ,Y(I) ,STRAIN,YRATE(I) ,STRATE
99 CONTINUE
RETURN
END
C
C
SUBROUTINE GUESS(TIMEF,TIMEI
C COMPUTE INITIAL VALUES OF CS, CP AND KP FOR ITERATION.
C
COMMON/STHD/STRESS,CS,CP,KP
REAL KP
C
cS=STRESS/EPSFDT(STRESS ,TIHEF)
TEMP=EPSF(~,STRESS,TIMEF)
TEMP=TEMP-STRESS*TIMEF/CS
KP=STRESS/TEMP
TEMP=EPSF(I,STRESS,TIMEI)-STRESS*TIMEI/CS
TEMP=TEMP*KP/STRESS
TEMP=-ALOG(I.-TEMP)
 CP=KP*TIMEI/TEMP
C
RETURN
END
C
C
SUBROUTINE CURVFIT(NF,ERLMT,X,YY,NN)
DIMENSION X( 1) ,Y(9OI) ,C(20) ,YY( 1)
COMMON/STHD/STRESS,CS,CP,KP
REAL KP
C
ERROR= 1.
M=3
I=O
WRITE(6,999)
50 ERR=O.
DO 20 N=I,NN
Y(N)=YY(N)-EPSR(NF,X(N))
ERR=ERR+Y(N)**2
20 CONTINUE
C
WRITE(~,IOO~)I,CS,CP,KP,ERR
C CONVERGENCE TEST
IF(ERR.LT.ERLMT**~)CO TO 100
IF(ABS(ERROR/ERR-~.).LT.ERLMT)GO TO 100
IF(ERR.GT.~OOO.*ERROR)GO TO 200
IF(I.GT.30)GO TO 200
C
I=I+l
ERROR=ERR
CALL LSTSQ(NF,ERLMT,X,Y,NN,C~M)
CS=CS+C( 1)
CP=CP+C(2)
KP=KP+C(3)
GO TO 5 0
200 WRITE(6,1003)
100 CONTINUE
C
999 F O R M A T ( ~ X , J I T E R A T I O N J ~ X , ~ H C S , ~ O X , ~ H C P , ~ O X , ~ H K P , ~ X , ~ H E R R O R )
1001 FORMAT(2FlO. 0)
1002 F O R M A T ( X , I ~ , ~ ( I P E ~ ~ . ~ ) )
1003 FORMAT (/ ,5X,' ITERATION DOES NOT CONVERGE. ' )
RETURN
END
C
C
SUBROUTINE LSTSQ (NF,ERLMT,X,Y ,N,c ,MI
c COMMON/STHD/STRESS,CS~CP~KP
DIMENSION X(1) ,Y(I) ,C(20) ,F(20,901) ,A(20,20) ,B(20)
C REAL KP
C DERIVATIVES
C FulF(T)=-STRESS*T/CS**2
C FU~F(T)=-STRESS*T*EXP(-KP*T/CP)/CP**2
C FU~F(T)=STRESS*EXP(-KP*T/CP)*(~./KP+T/CP)/KP-STRESS/KP**~
C
DO I J=l,N
F(I,J)=FuIF(NF,X(J))
~ ( 2J)=FU2F(NFtX(J))
,
1 F(M, J)=Fu~F(NF,X(J))
C
DO 3 I=l,M
DO 3 K=l,I
A(K,I)=o.
DO 2 J=l,N
2 A(K,I)=A(K,I)+F(I, J)*F(K, J)
3 A(I.K)=A(K,I)
C
DO 4 K=l,H
B(K)=o.
DO 4 J=1,1
4 B(K)=B(K)+Y(J)*F(K,J)
c CALL GAUSID(A,M,B,C,ERLMT)
CALL ELIM(A,M,B,C)
RETURN
END
C
C
SUBROUTINE ELIM(AA ,N,BB ,X 1
DIMENSION AA(20,20) ,BB(20) ,A(20,21) ,x(20) ,ID(20)
NN=N+l
DO 100 I=l,l
A(I,NN)=BB(I)
ID(I)=I
DO 100 J=l,N
loo A(I, J)=AA(I, J)
K=1
1 CALL EXCHI(A,N,NN,K,ID)
2 IF(A(K,K))3,999,3
3 KK=K+I
DO 4 J=KK,IN
A(K, J)=A(K, J)/A(K,K)
DO 4 I=l,N
'IF(K-I)41,4,41
41 A(I, J)=A(I,J)-A(I,K)*A(K, J)
4 CONTINUE
K=KK
IF(K-I)l,2,5
5 DO 10 I=l,N
 DO 10 J=i,N
IF(ID(J)-1)10,6,10
6 X(I)=A(J,NN)
10 CONTINUE
RETURN
999 WRITE(6,lOOO)
1000 FORMAT(5X,'NO UNIQUE SOLUTION.')
RETURN
END
C
C
SUBROUTINE EXCHI( A ,N,NN ,K,ID)
DIMENSION ~ ( 2 0 ~ 2 1,ID(20)
)
NROW=K
NCOL=K
B=ABS(A(K,K))
DO 2 I=l,N
DO 2 J=l,N
1F(ABS(A(I, J))-B)2,2,21
21 NROW=I
NCOL=J
B=ABS(A(I, J))
2 CONTINUE
IF(NR0U-K)3,3,31
31 DO 32 J=K,NN
C=A(NROW, J)
A(NROW, J)=A(K,J)
32 A(K,J)=C
3 CONTINUE
IF(NC0L-K)4,4,41
41 DO 42 I=l,N
C=A(I ,NCOL)
A(I ,NCOL)=A(I,K)
42 A(I,K)=C
I=ID(NCOL)
ID(NCOL)=ID(K)
ID(K)=I
4 CONTINUE
RETURN
END
C
C
SUBROUTINE GAUSID(A,N,B,X,ERR)
DIMENSION A(20,20) ,B(20) ,C(20,20) ,X(20)
K=O
NN=N+l
DO 11 I=l,N
IF(A(ID1))12,6,12
12 X(I)=I.
 C(I,NN)=B(I)/A(I,I)
DO 11 J=i,N
1 1 c(I,J)=A(I,J)/A(I,I)
1 CONTINUE
E=O.
C
DO 3 I=l,N
P=C(I,NN)
DO 2 J=l,N
P=P-c(I,J)*x(J)
2 CONTINUE
X(I)=X(I)+P
E=E+ABS (PI
3 CONTINUE
IF(E-ERR)4,4,5
4 RETURN
5 K=K+l
IF(IO0-K)6,I,l
6 WRITE(6,lOOO)
RETURN
1000 FORMAT(5X,26H GAUSID DOES NOT CONVERGE.)
END
 Table 11.S.6. Source Listing for the Program "DISPLY"

PROGRAM DISPLY SHL 11/16/87

GENERALIZED PLOT PROGRAM

C O M M O N / D T / N D P , X X ~ (,wi(9oi)
~O~) ,xx2(901) ,~~2(901)
COMMON/TTL/TITLE(~O) ,XLABEL(~),YLABEL(~)SOLID(4),DASHED(4)
CHARACTER NAME*20
DATA 1Y81Ns1Ps1D,LNsLGs1BK/'Y','N','P','D1,'LN','LG',' '/
WFtITE(6,999)
999 FORMAT(X, 'THIS IS A GENERAL PURPOSE PROGRAM FOR XYPLOT' /
1 IOX,'WRITTBN BY SANG H. LEE.')
w~ITE(6,998)
998 FORMAT(/,xS'TYPE IN BAUD RATE')
READ(SD1009)BRATE
IBRATE=BRATE/IO
IF(IBRATE.Eq.O)IBRATE=960
CALL INITT(IBRATE)
CALL TERM(2,1024)
C
400 CALL ERASE
CALL CHRSIZ(3)
CALL ANMODE
C INPUT TITLE
TYPE 1000
ACCEPT 1001, (TITLE(1) ,I=1,20)
1000 FORMAT(XD1l. TYPE A TITLE(MAX. 40 CHARACTERS). ' /
1 SX, 'DEFAULT IS TO READ THE FIRST LINE FROM INPUT FILE')
1001 FORMAT(2OA4)
c DATA POINTS(NDP)
TYPE 1002
ACCEPT lOO9,ANDP
NDP=ANDP
IF(NDP .EQ .O)NDP=SOI
1002 FORMAT(XS12. NUMBER OF DATA POINTS?'/
1 ,10XS1DEFAULTWILL INCLUDE ALL THE DATA POINTS IN THE FILE.')
C X-AXIS
TYPE 1004
IXA=2
10 TYPE 1005
ACCEPT lOO6,IANS
IF(IANS.EQ.IBK)IXA=O
IF(IANS .EQ.LN) IXA=O
IF(IANS.EQ.LG)IXA=I
IF(IXA.NE.2)GO TO 20
TYPE 1007
GO TO 10
 20 TYPE 1008
ACCEPT 1009,XMIN,XMAX
21 TYPE 1010
ACCEPT 1001,(XLABEL(I),I=I,4)
IF(XLABEL(I) . ~ q . ')GO TO 21
1003 FORMAT(II)
1004 FORMAT(X,'3. X-AXIS DEFINITION.')
1005 FORMAT(5XD1LINEAR OR LOGARITHMIC? TYPE LN OR LG '1
1006 FORMAT(AZ)
1007 FORMAT(X,'ANSWER NOT ACCEPTABLE, TRY AGAIN. ' )
1008 FoRMAT(~X,'UHICH DOMAIN DO YOU WANT TO DISPLAY? TYPE XMIN~XMAX')
1009 FORMAT(~FIO.O)
1010 FORMAT(~X, 'TYPE A LABEL(MAX. 10 CHARACTERS) . '1
C Y-AXIS
TYPE 1011
IYA=2
30 TYPE 1005
ACCEPT 1006,IANS
IF(IANS.EQ.IBK)IYA=O
IF(IANS.E~.LN)IYA=O
IF(IABS.E~.LG)IYA=I
IF(IYA.NE.~)GO TO 40
TYPE 1007
GO TO 30
40 TYPE 1012
41 TYPE 1010
- ACCEPT 1001,(YLABEL(I) ,I=l,3)
IF(YLABEL(I) .EQ. ')GO TO 41
1011 FORMAT(X,'4. Y-AXIS DEFINITION. '1
1012 FORHAT(X ,'RANGE FOR Y-AXIS WILL BE SET AUTOMATICALLY. '
C GRID OPTION
60 TYPE 1013
ACCEPT lOO6,IANS
NG=2
IF(1ANS. EQ .IBK)NG=O
IF(IANS.EQ.IY)NG=I .
IF(IANS.EQ.IN)NG=O
IF(NG.NE.~)GO TO 60
TYPE 1007
GO TO 50
1013 FORMAT(X,'5. GRID LINES? Y OR N. '1
C OVERLAY
60 TYPE 1014
IDASH=O
ACCEPT 1006,IANS
IF(IANS.EQ.IBK)IANS=IN
IF(IANS.EQ.IY)GO TO 100
IF(IANS.EQ.II)GO TO 200
TYPE 1007
 GO TO 60
1014 FORMAT(X,'6. OVERLAY TWO CURVES? Y OR N. ')
C INPUT DATA FORMAT
100 IDASH=O
TYPE 1016
ACCEPT 1003,IDASB
IF(IDASH.Eq.0 .OR. IDASH.GT.4)GO TO 100
C LEGEND
TYPE 1015
TYPE 1017
ACCEPT 1001,(SOLID(1). I=1,4)
TYPE 1018
ACCEPT iooi,(DASHED(I) ,1=1,4)
1015 FORMAT(3X. 'IST CURVE VILL BE A SOLID LINE; 2ND VILL BE DASHED')
1016 FORMAT(3X. 'TYPE 1, 2. 3 OR 4 TO SPECIFY INPUT DATA FORMAT:' /
1 iOXIJ1 IF INPUTS ARE XI Y1, Y2'/
2 1OXIJ2 IF INPUTS ARE XI, X2, Y ' /
3 iOX,'3 IF INPUTS ARE Xi, Yi, X2, Y2'/
4 lOX,'4 IF INPUTS ARE XI DUMMY. DUMMY, Y1, ~2')
1017 FoRMAT(~X,'TYPE A LEGEND FOR SOLID LINE.' ) 4:
1018 FORMAT(5X1 'TYPE A LEGEND FOR DASHED LINE.' )
C DATA INPUT
200 INPUT=I
TYPE 1021
READ(5,1022)NAME
iO21 FORMAT(XIJ7. TYPE THE FILE NAME.',/)
1022 FORMAT(A)
c CALL ASSIGN(I,NAME,-1)
OPEN (UNIT=INPUT,FILE=NAME,READONLY,STATUS='OLD ,ERR=~OO)
C CALL BELL
IF(TITLE(I). ~ q . J )READ(INPUT. 1024) (TITLE(1). I=1120)
GO TO 500
300 READ(INPUT, 1024) (TITLE(I) . I = W )
500 CALL PREP(IXA,IYA,XMIN,XMAX,NG,INPUT~IDASH)
C
450 TYPE 1020,NAME
ACCEPT 1006,IANS
IF(IANS.EQ.IY)GO TO 300
IF(IANS. Eq .1N)GO TO 700
TYPE 1007
GO TO 450
1020 FORMAT(X 'MORE PLOTS TO PROCESS IN FILE ' ,AI2, ' Y OR I ? ' //)
700 CLOSE(UNIT=INPUT)
C
C FOR MORE FILES TO PROCESS.
350 TYPE 1023
ACCEPT IOO6,IANS
IF(1ANS. Eq.1Y)GO TO 400
IF(IANS.E~.IN)STOP
 TYPE 1007
GO TO 350
1023 FORMAT(XJMORE PLOTS TO PROCESS? Y OR N'//)
1024 FORMAT(~OA~,//)
END
C
C
SUBROUTINE PREP(IXA,IYA,XMIN,XMAX,NC,INPUT,IDASH~
COMMON/TTL/TITLE(~~) ,XLABEL(4) JLABEL(4) ,SOLID(4),DASHED(~)
COMMON/DT/NDP,XX~(~O~) ,~i(9oi) ,xx2(901) ,~~2(901)
DIMENSION GINC(20).TINC(20)
DATA T 1 N C / 1 . , 1 . , 2 . , 2 . , 4 ~ ~ 5 ~ ~ 5 ~ ~ 5 ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
1 10.,10.,10:,10.,10.,10./
DATA G1NC/2.,2.,5.,5.,5.,10.,10.,10.~10.~10.,10.,10.,10.~10.~
1 20.,20.,20.,20.,20.,20./
DATA IY,IN/'YJ,'NJ/
C INPUT DATA
I=O
100 I=I+i
XX2(1)=0.
YY2(1)=0.
I F ( I D A S H . E ~ . O ) R E A D ( I N P U T ~ ~ O O ~ ~ E N D = ~ O ),YYi(I)
XX~~I~
IF(IDASH.Eq. l)READ(INPUT, 1OO3~END=iO)XXi~I) ,YYi(I) ,YY~(I)
IF(IDASH.Eq. l)XX2(I)=XX1(1)
IF(IDASH.E~.~)READ(INPUT,1003,END=10)XX1~1~ JXX2(1) ,YYi(I)
IF(IDASH.E~.~)YY~(I)=YYI(I)
IF(IDASH . ~ q~)READ(INPUT,
. 1003,END=lO)XXi(I) ,YYi(I) ,xx~(I),YY~(I)
IF(IDASH.Eq.4)THEN
READ(INPUT, 1003,END=l0)XXi(I) ,DUMMY~DUHMY~YY~(I) ,YY2(1)
XX2(I)=XXI(I)
ENDIF
IF(XXI(I).E~.O.)THEN
IF(XXZ(1) .EQ.O.)THEN
IF(YYl(1) .Eq.O.)THEN
IF(YY~(I).E~.O.)THEN
GO TO 10
ENDIF
ENDIF
ENDIF
ENDIF
GO TO 100
10 CONTINUE
NDP=I-1
TYPE 1008,NDP
C BOUNDARY
CALL BOUND(XXI,NDP,IMINX,IMAXX,NSCALX~
IF(XM1N.Eq.XMAX)THEN
XMIN=IMINX*lO.**NSCALX
XMAX=IHAXX*IO.**NSCALX
 ENDIF
CALL BOUND(YYI ,NDP,IMINI, IMAXI ,NSCALI)
IF(IDASB.E~.O)GO TO 50
CALL BOUND(YY2, NDP ,IMIN2,IMAX2,N S C A L ~ )
IF(NSCAL~.GT.NSCALI)GO TO 30
W(NSCAL2. LT .NSCAL1)GO TO 40
IF(IMAX~.GT.IMAX~)IMAX~=IMAX~
IF(IMIN~.LT.IMIN~)IMIN~=IHIHZ
GO TO 50
30 NSCALl=NSCAL2
IMAXl=IMAX2
40 IMINl=IMIN2
50 YMIN=IMIN1*10.**NSCAL1
YMAX=IMAXI*lO.**NSCALl
C TICK MARK
200 ITX=(IMAXX-IMINX)/IO+~
ITY=(IMAXI-IMIN~)/~o+~ '

TINX=TINC(ITX)*lO. **NSCALX
TINY=TINC(ITY)*IO. **NSCALl
D TYPE lO1l,ITX,ITY,IMAXl,IHINI,NSCALl
C GRID INCREMENTS
GINX=GINC(ITX)*lO.**NSCALX
GINY=GINC(ITY)*IO.**NSCALl
D TYPE lOlO,TINX,TINY,GINX,GINY
D TYPE lOll,IXA,IYA,NG,IDASH
TYPE lOOS,XMIN,XMAX
TYPE lOO2,YMIN,YMAX
C CALL BELL
TYPE 1004
ACCEPT 1006,DUMMY
300 CALL GENPLT
i (IXA,IYA ,XMIN,XMAX ,YMIN,YMAX,NG,IDASH,TINX,TINY,GINX,GIHY)
350 TYPE 1005
ACCEPT lOO6,IANS
IF(IANS.E~.IY)GO TO 400
IF(IANS.EQ.IN)RETURN
GO TO 350
400 TYPE 1007
ACCEPT 1003,XXMIN,XXMAX
IF(XXMIN.E~.XXMAX)GO TO 500
XMIN=XXMII
XMAX=XXMAX
500 TYPE 1012
ACCEPT lOO3,YYMIN,YYMAX
IF(YYMIN.E~.YYMAX)GOTO 300
YMIN=YYMII
YMAX=YYMAX
GO TO 300
1001 FORMAT(X,lOA4)
1002 FORMAT(// ,X,'YMIN= ' ,E10.3,5X,' YMAX= , ~ 1 03)
.
1003 FORMAT(BN,5F16.0)
1004 FORMAT(//, X, HIT RETURN TO CONTINUE. ' )
1005 FORMAT(XDICHANGE DOMAIN? Y OR N1//)
1006 FORMAT(AZ)
1007 FORMAT(X,'TYPE XMIN,XMAX' )
1008 FORMAT(//, X , 'NUMBER OF DATA POINTS=' ,IS)
1009 FORMAT(//,X, 'XMIN= ' ,E10.3,5XD' XMAX= ' ,~10.3)
1010 FORMAT(/,X,~EI~.~)
1011 FORMAT(//,X,~~I~)
1012 FORMAT(X, 'TYPE YMIN ,YMAX1)
END
C
C
SUBROUTINE BOUND(YY,NDP,IMIN,IMAX,NSCALE)
DIMENSION YY (NDP)
NSCALE=O
C YMIN AND YMAX
YMIN=YY(11
YMAX=YY( 1
DO 100 I=2,NDP
IF(YY(I) .LE.YMAX)GO TO 90
YMAX=YY(I)
GO TO 100
go IF(YY(I). GE.YMIN)GO TO 100
YMIN=YY(11
100 CONTINUE
C TYPE 999,YMIN,YMAX
C TYPE 999,YY
C 999 FORMAT(/5E16.3)
C WHICH IS DOMINANT? YMAX OR YMIN
NOSWAP=-I
IF(ABS(YMAX). GE.ABS(YWIN) )NOSWAP=I
IF(NOSWAP.EQ.1)GO TO 200
C TYPE 1001,NOSWAP
ciooi FORMAT(/IIO)
YMAXX=YMAX
YMAX=YMIP
YMIN=YMAXX
C IMAX AND PSCALE
200 IF (ABS(YMAX).GE.IOO.)GO TO 210
IF(ABS(YMAX) .LT. 10. )GO TO 220
GO TO 300
210 NSCALE=NSCALE+I
YMAX=YMAX/IO.
GO TO 200
220 NSCALEzNSCALE-1
YMAX=YMAX*lO.
GO TO 200
 300 IMAX=YMAX
IF(IMAX.NE.YMAX)IMAX=IMAX+NOSWAP
C IMIN
YMIN=YMIN/IO.**NSCALE
IMIN=YMIN
IF(IMIN.GT.YMIN)IMIN=IMIN-NOSWAP
IF(NOSWAP. Eq .l)RETURN
IMINX=IMIN
IMIN=IMAX
IMAX=IMINX
RETURN
END
C
C
SUBROUTINE GENPLT(IXA,IYA,XMIN,XMAX,
I ~MIN,YMAX,NG,IDASH,TINX,TINY,GINX,GINY)
COMMON/DT/NDP ,XXi(901) sYY1(901) ,xx2(901) ,YY2(901)
C~MM~N/TTL/TITLE(~O)XLABEL(4) ,YLABEL(4) ,SOLID(4),DASHED(4)
INTEGER IBUF(20)
CHARACTER ISTR*80
CALL ERASE
C FRAME
CALL ~~1~~0(120,840,150~600)
XL=(XMAX-XMINI*~.001
YL=(YMAX-YMIN)*I. 001
CALL VWINDO(XMIN ,XL,YMIN,YL)
CALL MOVEA(XM1N ,YMIN)
CALL DRAWA(XMAX ,YMIN)
CALL DRAWA(XMAX ,YHAX)
CALL DRAWA(XMIN,YMAX)
CALL DRAWA(XM1N ,YMIN)
C TICK MARKS
IF(XMII.GT.O.)GO TO 80
XTK=O .
70 CALL MOVEA(XTK ,WIN)
CALL DRWREL(Os3)
CALL MOVEA(XTK ,YMAX) '

CALL DRUREL(0, -3)

XTK=XTK-TIIX
IF(XTK.GT.XMIN)GO TO 70
80 XTK=TIBX
so IF(XTK.GT.XMIN)GO TO 100
XTK=XTK+TINX
GO TO 90
ioo IF(XTK.GE.XMAX)G~ TO 200
CALL MOVEA (XTK ,YMIN)
CALL DRWREL(0,B)
CALL MOVEA(XTK,YMAX)
CALL DRWREL(0 ,-3)
 XTK=XTK+TINX
GO TO 100
200 IF(YMIN.GT.O.)GO TO 209
YTK=O.
201 CALL MOVEA(XM1N ,YTK)
CALL DRWREL(3 ,O)
CALL MoVEA (XMAX ,YTK)
CALL DRWREL(-3 ,0)
YTK=YTK-TINY
IF(YTK.GT.YMIN)GO TO 201
209 YTK=TINY
210 IF(YTK.GT.YMIN)GO TO 250
YTK=YTK+TINY
GO TO 210
250 IF(YTK .GE .YMAX) GO TO 300
CALL MOVEA (XMIN,YTK)
CALL D R U R E L ( ~ , ~ )
CALL MOVEA(XMAX,YTK)
CALL DRWREL(-3, 0)
YTK=YTK+TINY
GO TO 250
C GRID LINES
300 IF(NG.EQ.O)GO TO 400
IF(XMIN.GT.O.)GO TO 309
XGD=O.
301 CALL MOVEA(XGD ,YMIN)
CALL DRAUA (XGD ,YMAX)
XGD=XGD-GINX
IF(XGD.GT.XMIN)GO TO 301
309 XGD=GINX
310 IF(XGD. GT. XM1I)GO TO 320
XGD=XGD+GINX
GO TO 310
320 IF(XGD.GE.XMAX)GO TO 350
CALL MOVEA (XGD,YMIN)
CALL DRAWA(XGD ,YMAX)
XGD=XGD+GINX
GO TO 320
350 IF(YMIN.GT.O.)GO TO 355
YGD=O.'

351 CALL MOVEA(XM1N ,YGD)

CALL DRAMA (XMAX,YGD)
YGD=YGD-GIHY
IF(YGD.GT.YMIN)GO TO 351
355 YGD=GINY
360 IF(YGD .GT.YMIN)GO TO 370
YGD=YGD+GINY
GO TO 360
370 1F(YGD. GE.YMAX)GO TO 400
 CALL MOVEA(XMIN,YGD)
CALL DRAWA(XMAX, YGD)
YGD=YGD+GINY
GO TO 370
C CURVES
400 CALL MOVEA(XX~(~) ,YYl(l))
DO 410 I=2,NDP
410 CALL DRAWA(XX~(I) ,YYi(I))
IF(IDASH.E~.O)GOTO 500
CALL MOVEA(XX~(I) ,YY2(1))
DO 420 I=2,NDP
420 CALL DASHA(XX~(I),YYZ(I) ,341
C SCALE
500 CALL MOVEA(XMIN ,YMIN)
CALL MOVREL(-20, -20)
WRITE(ISTR,~OO~)XMIN
CALL CHTOIN (ISTR,IBUF,12)
CALL AOUTST(12,IBUF)
TYPE 1001,XMIN
CALL MOVEA(XMAX,YMIN)
CALL MOVREL(-50, -20)
WRITE(ISTR,~~~~)XMAX
CALL CHTOIN (ISTR,IBUF ,12)
CALL AOUTST( 12,IBUF)
TYPE 1001,XMAX
CALL MOVEA (XMIN ,YMII)
CALL MOVREL(-110,2)
WRITE(ISTR, lOO2)YMIN
CALL CHTOIN (ISTR ,IBWs 12)
CALL AOUTST( 12,IBUF)
TYPE lOO2,YMII
CALL MOVEA(XM1H ,YMAX)
CALL M0VREL(-110, 2)
WRITE(ISTR,IOO~)YMAX
CALL CHTOIN (ISTR,I B W ,12)
CALL AOUTST( 12,IBUF)
C TYPE 1002,YMAX
C LABELING
CALL MOVABS (500,100)
CALL AOUTST( 16,XLABEL)
C TYPE 1003,(XLABEL(1) ,I=I,4)
CALL MOVABS (20,460)
CALL AOUTST( 12,YLABEL)
c TYPE 1003,(YLABEL(I1, I=ls3)
C TITLE
CALL MOVABS( 10,30)
CALL CHRSIZ(2)
CALL AOUTST(80 ,TITLE)
CALL CHRSIZ(3)
C TYPE 1003,(TITLE(1) ,I=l,lO)
C LEGEND
IF(IDASH.Eq.O)GO TO 600
CALL M0VABS(650,700)
CALL DRWREL(IOO,O)
CALL MOVREL(30,O)
CALL AOUTST(l6, SOLID)
TYPE 1003,(SOLID(I),I=1,4)
CALL MOVABS(650,675)
CALL DSHREL(IOO,O,B~)
CALL MOVREL(30,O)
CALL AoUTST(16,DASHED)
600 CONTINUE
C CALL HDCOPY
CALL HOME
C CALL BELL
CALL ANMODE
RETURN
1001 FORMAT(1PE8.lD4X)
1002 FORMAT(IPEl0.2,4X)
1003 FORMAT(X, 1OA4)
END
C
C
SUBROUTINE CHTOIN (ICH,IBUF,NC)
INTEGER IBUF(1)
CHARACTER*80 ICH
C
DO 100 I=l,NC
11 = ICHAR(ICH(I:I)
CALL CFS (IBUF,Il,I,O)
100 CONTINUE
RETURN
END
 BIBLIOGRAPHY
[1.1]R. H. MacNeal, Ed., HANDBOOK FOR LINEAR ANALYSIS, MSCINASTRAN Ver-
sion 64, The MacNeal-Schwendler Corporation, August 1985.

[1.2] D.N. Herting, R. L. Harder, K. K. Karlsten and R. N. Swami, "Architecture of Nonlinear

Analysis," Conference on Finite Element Methods and Technology, Pasadena CA, March
19-20, 1981.
[1.3] D. N. Herting and R. N. Swami, "Efficient Methods for Static Nonlinear Analysis of
Large Order Problems," ASME Symposium on Nonlinear Finite Element Analysis of
Shells, Washington D.C., November 15-20, 1981.
[1.4] D.N. Herting, "Advances in Nonlinear Analysis in MSC/NASTRAN," MSCINASTRAN
User's Conference, Pasadena CA, March 18-19, 1982.
[1.5] D. N. Herting, "The Reality of Cost-Effective Nonlinear Finite Element Analysis,"
AUTOFAC 5 Conference Proceedings, CASA Association of SME, Detroit MI, November
14-17, 1983.
[2.1] R. H.Gallagher, Finite Element Analysis Fundamentals, Prentice-Hall, 1975.

[2.2] 0. C. Zienkiewicz, The Finite Element Method in Engineering Science, Third Edition,
McGraw-Hill, 1977.
[2.3] B. Irons and S. Ahmad, Techniques of Finite Elements, Wiley & Sons, 1980.

[2.4] K. J. Bathe, Finite Element Procedures in Engineering Analysis, Prentice-Hall, 1982.

[2.5] I. S. Sokolnikoff, Mathematical Theory of Elasticity, Second Edition, McGraw-Hill, 1956.

[2.6] M. F. Rubinstein, Structural Systems - Statics, Dynamics and Stability, Prentice-Hall,

1970.
[3.1] R. H.Pennington, Introductory Computer Methods and Numerical Analysis, Second Edi-
tion, Macmillan, 1970.

[3.2] E. K. Blum, Numerical Analysis and Computation: Theory and Practice, Addison-
Wesley, 1972.
[3.3] S. H. Lee, "Basic Considerations for Expedient Solution Strategies in Nonlinear Finite
Element Analysis," Proceedings, ASME Region IX Computer Applications Symposium,
Los Angeles, pp. 4/1-18, October 1988.
[3.4] D.G. Luenberger, Linear and Nonlinear Programming, Second Edition, Addison-Wesley,
1984.

Bibliography - 1
[3.5] C. S. Davis, "Line Search Methods for Extended Penalty Function Environments," Int.
J. Numerical Methods in Engineering, Vo1.15, 1980.
[3.6] M. A. Crisfield, "A Faster Modified Newton-Raphson Iteration," Computer Meth. Ap-
plied Mechanics and Engineering, Vo1.20, 1979.
[3.7] M. A. Crisfield, "Accelerating and Damping the Modified Newton-Raphson Method,"
Computers & Structures, Vo1.18, 1984.

[3.8] H. Matthies and G. Strang, "The Solution of Nonlinear Finite Element Equations," Int.
J . Numerical Methods in Engineering, Vo1.14, 1979.
[3.9] L. Zhang and D. R. J. Owen, "A Modified Secant Newton Method for Nonlinear Prob-
lems," Computers & Structures, Vo1.15, 1982.
[3.10] S. H.Lee, "Rudimentary Considerations for Effective Line Search Method in Nonlinear
Finite Element Analysis," Computers & Structures, Vo1.32, No.6, pp. 1287-1301, 1989.
[3.11] S. H. Lee, "Rudimentary Considerations for Effective Quasi-Newton Updates in Non-
linear Finite Element Analysis," Computers 63 Structures, Vo1.33, No.2, pp. 463-476,
1989.
[3.12] K. J. Bathe and A. P. Cimento, "Some Practical Procedures for the Solution of Non-
linear Finite Element Equations," Computer Meth. Applied Mechanics and Engineering,
V01.22, 1980.

[3.13] P. G.Bergan and R. W. Clough, "Convergence Criteria for Iterative Processes," AIAA
Journal, Vol.10, No.8, 1972.
[3.14] S. H.Lee, "Rudimentary Considerations for Effective Convergence Criteria in Nonlinear
Finite Analysis," Proceedings, 1988 ASME Int. Computers in Engineering Conference,
San Francisco, Vo1.3, pp. 593-598, August 1988.
13.151 E. Ramm, "The Riks/Wempner Approach - An Extension of the Displacement Control
Method in Nonlinear Analysis," Recent Advances in Nonlinear Computational Mechanics,
Pineridge Press, 1982.
[3.16] M.A. Crisfield, "A Fast Incremental/Iterative Solution Procedure that Handles 'Snap-
Through'," Computers & Structures, Vo1.13, 1981.
[3.17] G.A. Wempner, "Discrete Approximations Related to Nonlinear Theories of Solids,"
Int. J. Solids and Structures, Vo1.7, 1971.

[3.18] E.Riks, "The Application of Newton's Method to the Problem of Elastic Stability," J.
of Applied Mechanics, Vo1.39, 1972.

[3.19] R. Kao, "A Comparative Study on the Elastic-Plastic Collapse Strength of Initially
Imperfect Deep Spherical Shells," Int. J. Num. Mech. Eng., Vo1.17, pp. 427-444, 1981.

Bibliography - 2
[3.20]S. H. Lee and D. N. Herting, "Comments on 'A Simple Approach to Bifurcation and
Limit Point Calculations'," Int. J. Numerical Methods Eng., Vo1.21, pp. 1935-1937,1985.
[4.1]H.G.Schaeffer, "MSC/NASTRAN Primer, Static and Normal Modes Analysis," Wallace
Press, 1979.
[4.2]M. Okabe, Y. Yamada and I. Nishiguchi, "Basis Transformation of Trial Function Space
in Lagrange Interpolation," Computer Methods in Applied Mechanics and Engineering,
Vol. 20, pp. 85-99, 1980.
[4.3]T.J. R. Hughes, R. L. Taylor, J. L. Sackman, A. Curnier and W. Kanoknukulchai, "A
Finite Element Method for a Class of Contact Problems," Comp. Meth. in Appl. Mech.
Eng., Vol. 8, pp. 249-279, 1976.
[4.4]K. J. Bathe and A. Chaudhary, "A Solution Method for Planar and Axisymmetric Con-
tact Problems," Int. J. Num. Meth. Eng., 21, pp. 65-88, 1985.
[4.5]A. B. Chaudhary and K. J. Bathe, "A Solution Method for Static and Dynamic Analysis
of Three-dimensional Contact Problems with Friction," Computers & Structures, Vol.
24, No. 6, pp. 855-873, 1986.
[4.6]J . C. Simo, P. Wriggers and R. L. Taylor, "A Perturbed Lagrangian Formulation for the
Finite Element Solution of Contact Problems," Comp. Meth. Appl. Mech. Eng., 50,
pp. 163-180, 1985.
[4.7]J . W. Ju and R. L. Taylor, "A Perturbed Lagrangian Formulation for the Finite Element
Solution of Nonlinear Frictional Contact Problems," J. Theo. and Appl. Mech., 7, 1988.

[4.8]J. A. Landers and R. L. Taylor, "An Augmented Lagrangian Formulation for the Fi-
nite Element Solution of Contact Problems," UCB/SESM-85/09, Dept. of Civil Eng.,
University of California, Berkeley, 1985.
[4.9]L. T . Campos, J. T. Oden, and N. Kikuchi, "A Numerical Analysis of a Class of Contact
Problems with Friction in Elastostatics," Comp. Meth. Appl. Mech. Eng. , 34, pp. 821-
845, 1982.
[4.10]J. H. Cheng and N. Kikuchi, "An analysis of Metal Forming Processes using Large
Deformation Elastic-Plastic Formulations," Comp. Meth. Appl. Mech. Eng., Vol. 49,
pp. 71-108, 1985.
[4.11]S. H. Lee, "Some Salient Considerations for Computational Procedure in Elasto-plastic
Material Model," Proc. of MSCINASTRAN Users Conference, Seoul, Korea, October
22-23, 1990.
[4.12]S. H.Lee and S. S. Hsieh, "Expedient Implicit Integration with Adaptive Time Stepping
Algorithm for Nonlinear Transient Analysis," Comp. Meth. Appl. Mech. Eng., Vol. 81,
No. 22, pp. 151-172, 1990.

Bibliography - 3

........ ...... _..... ....... ..... ____

[4.13] R. F. Steidel, Jr., An Introduction to Mechanical Vibrations, John Wiley & Sons, New
York, 1971.
[4.14] S. Timoshenko, D. H. Young and W. Weaver, Jr., Vibration Problems in Engineering,
Fourth Edition, pp. 198-199, John Wiley & Sons, New York, 1974.
[4.15] S. Timoshenko and J. N. Goodier, Theory of Elasticity, Third Edition, pp. 409-421,
McGraw-Hill, New York, 1970.
[4.16] S. H.Lee, S. S. Hsieh and T. L. Bock, "Adaptive Arc-Length Methods in MSC/NASTRAN,"
proceedings of 1990 MSC World Users Conference, 5/1-22, Los Angeles, March 28, 1990.
[5.1] R. Szilard, Theory and Analysis of Plates, Prentice-Hall, 1974.
[5.2] S. H. Lee and A. Raiten, "Characterization of Geometric Nonlinearity with Some Salient
Examples," Proceedings of the MSCINASTRAN User's Conference, 15, Tokyo, Japan,
October 17-18, 1991.
[5.3] 0. C. Zienkiewicz, The Finite Element Method in Engineering Science, Third Edition,
McGraw-Hill, 1977.
[5.4] H. D. Hibbitt, "Some Follower Forces and Load Stiffness," Int. J. Num. Met. in
Engineering, Vo1.14, pp. 937-941, 1979.
[5.5] D. N. Herting and G. W. Haggenmacher, "Pressure Follower Matrix for Geometric Non-
linear Finite Element Analysis," Proceedings, MSCINASTRAN User's Conference, Los
Angeles, 1987.
[5.6] T . R.Kane, P. W. Likins, and A. Levinson, Spacecraft Dynamics, McGraw Hill, 1983.
[5.7] S. P. Timoshenko and J . M. Gere, Theory of Elastic Stability, McGraw-Hill, 1961.
[5.8] H. D.Hibbit, P. V. Marcal and J. R. Rice, "A Finite Element Formulation for Problems
of Large Strain and Large Displacement," Int. J. Solids Structures, Vo1.6, 1970.
[5.9] R. H. McMeeking and J. R. Rice, "Finite Element Formulations for Problems of Large
Elastic-Plastic Deformation," Int. J. Solids Structures, Vol.11, 1975.
[5.10] J. C. Nagtegaal, "Some Computational Aspects of Elastic-Plastic Large Strain Analy-
sis," Int. J. Numerical Methods in Engineering, Vo1.17, 1981.
[5.11] J. H. Argyris and M. Kleiber, "Incremental Formulation in Nonlinear Mechanics and
Large Strain Elasto-Plasticity: Natural Approach (Part 1 and 2)," Computer Methods in
Applied Mechanics and Engineering, Vol.11, 1977 and Vo1.14, 1978.
[5.12] N. T . Tseng and G. C. Lee, "Inelastic Finite Strain Analysis of Structures Subjected
to Nonproportional Loading," Int. J. Numerical Methods in Engineering, Vo1.21, 1985.
[5.13] L. E. Malvern, Introduction to the Mechanics of a Continuous Medium, Prentice-Hall,
1969.

Bibliography - 4
[5.14]Y. C. Fung, Foundation of Solid Mechanics, Prentice-Hall, 1965.

[5.15]G. E. Mase, Continuum Mechanics, McGraw-Hill, 1976.

[5.16]W.Prager, Introduction to Mechanics of Continua, Ginn and Company, 1961

[6.1] R. Hill, The Mathematical Theory of Plasticity, Oxford a t the Clarendon Press, First
Edition 1950; Fifth Edition, 1967.

[6.2]E. G. Thomsen, C. T . Yang and S. Kobayashi, Mechanics of Plastic Deformation in

Metal Processing, The MacMillan Company, New York.

[6.3]S. H. Lee, "Some Salient Considerations for Computational Procedure in Elastoplas-

tic Material Model," Proceedings, MSCINASTRAN User's Conference, Seoul, Korea,
October 22-23, 1990.

[6.4]S. H. Lee, ('Self-Adaptive Algorithm for Efficient Computation of Elasto-plastic Material

Process," Proceedings, 1991 ASME Int . Computers in Engineering Conference, Santa
Clara, Calif., August 18-22, 1991.

[6.5]0. C. Zienkiewicz, The Finite Element Method, Third Edition, McGraw-Hill, 1977

[6.6]G. C. Nayak and 0. C. Zienkiewicz, "Convenient Form of Stress Invariants for Plasticity."
Journal of the Structural Division, ASCE, Vo1.98, N. ST4, April 1972.

[6.7]D. R. J. Owen and E. Hinton, Finite Elements in Plasticity, Pineridge Press, 1980.
[6.8]Y.N. Yamada and T. Sakurai, "Plastic Stress Strain Matrix and Its Application for the
Solution of Elastic-Plastic Problems by the Finite Element Method," Int. J. Mech. Sci.,
Vol.10, pp. 343-354, 1968.

[6.9]H.Armen, "Assumptions, Models and Computational Methods for Plasticity," Trends

in Computerized Structural Analysis and Synthesis, ed. A. K. Noor and H. G. McComb,
Jr., pp. 161-174,1978.

[6.10]R. S. Rivlin, "Some Topics in Finite Elasticity," Structural Mechanics: Proceedings of

the First Symposium on Naval Structural Mechanics, Pergamon Press, New York, pp.
169-198,1960.

[6.11]Y. Yamada, H.Takabatake and T . Sato, "Effect of Time-dependent Material Properties

on Dynamic Response," Int. J. Numerical Methods in Engineering, Vol. 8, pp. 403-414,
1974.

[6.12]M. P. Badani, ''A New Method t o Solve Nonlinear Creep Problems in Structural Me-
chanics Using Incompatible Isoparametric Finite Elements," Ph.D. Dissertation, U. C.
Berkeley, 1980.

[6.13]G. A. Greenbaum and M. F. Rubinstein, "Creep Analysis of Axisymmetric Bodies

Using Finite Elements," Nuclear Engineering and Design, Vo1.7, 1968.

Bibliography - 5
[6.14]J . A. Clinard et al., "Elevated-Temperature Deformation and Failure Testing and Anal-
ysis of Nozzle-to-Spherical Shells: Specimens NS-2 and NS-I," Oak Ridge National Lab-
oratory, ORNL-5939, January, 1983.
[6.15]A. Levy, "High-Temperature Inelastic Analysis," Computers and Structures, Vo1.13,
1981.
[6.16]A. Levy and A. B. Pifko, "On Computational Strategies for Problems Involving Plas-
ticity and Creep," International Journal for Numerical Methods in Engineering, Vo1.17,
1981.
[6.17]S. H. Lee, "Application of the Rheological Model to the Creep Analysis Coupled with
Plastic Deformation," Transactions, The 8th SMIRT Conference, Brussels, Vol. L, pp.
85-92, August, 1985.
[6.18]S. H. Lee, "Generalized Viscoelastic Model for Creep Analysis Coupled with Plastic
Deformation," Int. J . Numerical Methods in Engineering, John Wiley & Sons, Vo1.26,
pp. 153-165, 1988.
[6.19]H. Kraus, Creep Analysis, John Wiley and Sons, 1980.
[6.20] N. H. Polakowski, and E. J. Ripling, Strength and Structure of Engineering Materials,
Prentice-Hall, 1966.
[6.21] R. C. Juvinall, Stress, Strain and Strength, McGraw-Hill, 1967.

[6.22]A. Levy, "High-Temperature Inelastic Analysis of a Pressure Vessel Nozzle Attach-

ment," Grumman Aerospace Corporation, ORNL/Sub-448513, June 1980.
[6.23] S. H.Lee, "Creep Analysis Capability in MSCINASTRAN," Proceedings, MSC User's
Conference, Pasadena CA, March 1984.
[7.1]J. M. Chern, "Intermediate Heat Exchanger for Fast Flux Test Facility: Evaluation of
Inelastic Computer Programs," FWR-27, Foster Wheeler Corporation, March 1972.
[7.2]R. Kao, "A Comparative Study on the Elastic-Plastic Collapse Strength of Initially
Imperfect Deep Spherical Shells", International Journal of Numerical Methods in Engi-
neering, Vo1.17, No.3, 1981.
[7.3]Dinno and Gill, "An Experimental Investigation into the Plastic behavior of Flush Noz-
zles in Spherical Pressure Vessels," Int. J. Mesh. Sci., Vo1.7, p. 817, 1965.
[8.1] D. N. Herting, "A General Purpose Nonlinear Transient Integration System", Innovative
Methods for Nonlinear Problems, edited by W. K. Liu, T . Belytschko and K. C. Park,
Pineridge Press, 1984.
[8.2]S. H.Lee and S. S. Hsieh, "Expedient Implicit Integration with Adaptive Time Stepping
Algorithm for Nonlinear Transient Analysis", Int. J. Computer Methods in Applied
Mechanics and Engineering, Vo1.81, No. 22, pp. 151-172, 1990.

Bibliography - 6
[8.3] N. M. Newmark, "A Method of Computation for Structural Dynamics," J. of Engineering
Mechanics Division of ASCE, July, 1959.
[8.4] P. G. Bergan and E. Mollestad, "An Automatic Time Stepping Algorithm for Dynamic
Problems," Computer Methods in Applied Mechanics and Engineering, Vo1.49, pp. 299-
318, 1985.
[8.5] S. H. Lee, "Rudimentary Considerations for Effective Quasi-Newton Updates in
Nonlinear Finite Element Analysis," Computers & Structures, Vo1.33, No.2, pp. 463-476,
Pergamon Press, 1989.

[8.6] S. H . Lee, "Rudimentary Considerations for Effective Line Search in Nonlinear Finite
Element Analysis," Computers & Structures, Vo1.32, No.6, pp. 1287-1301, 1989.
[8.7] S. H. Lee, T. L. Bock and S. S. Hsieh, "Adaptive Time Stepping Algorithm for Nonlinear
Transient Analysis,'' Proceedings, 1988 MSC World Users Conference, Los Angeles, Vo1.2
(54), March 1988.
[8.8] J. M. Biggs, Structural Dynamics, pp. 69-76, McGraw-Hill, 1964.

[8.9] H. A. Balmer and E. A. Witmer, "Theoretical-Experimental Correlation of Large Dy-

namic and Permanent Deformation of Impulsively Loaded Simple Structures," Air Force
Flight Dynamic Laboratory, Report FDP-TDR-64-108, Wright-Patterson AFB, Ohio,
July, 1964.

[8.10] T . B. Belytschko, J. I. Lin and C. S. Tsay, "Explicit Algorithms for the Nonlinear
Dynamics of Shells," Computer Methods in Applied Mechanics and Engineering, Vo1.42,
pp. 225-251,1984.
[8.11] D. P. Mondkar and G. H. Powell, "Finite Element Analysis of Nonlinear Static and
Dynamic Response," Int. J. Numerical Methods in Engineering, Vol.11, pp. 499-520,
1977.
[10.1] S. H. Lee and D. N. Herting, "Comment on 'A Simple Approach to Bifurcation and
Limit Point Calculations,' " Int. J. Numerical Methods Eng., Vo1.21, pp. 1935-1937,
1985.
[10.2] M. Fujikake, "A Simple Approach to Bifurcation and Limit Point Calculations," Int.
J. Numerical Methods Eng., Vo1.21, pp. 185-191, 1985.
[10.3] R. Kao, "A Comparative Study on the Elastic-Plastic Collapse Strength of Initially
Imperfect Deep Spherical Shells," Int. J. Numerical Methods Eng., Vo1.17, pp. 427-444,
1981.
[10.4] W. P. Schulze Schwering, "The Cologne Challenge," Finite Element News, No.4, 1984.

[11.1] S. H. Lee, "Rudimentary Considerations for Effective Quasi-Newton Updates in Non-

linear Finite Element Analysis," Computer 0 Structures, Vo1.33, No.2, pp. 463-476.

Bibliography - 7
[11.2] S. H. Lee, "Rudimentary Considerations for Effective Line Search Method in Nonlinear
Finite Element Analysis," Computers 6 Structures, Vo1.32, No.6, pp. 1287-1301, 1989.

[11.3] S. H. Lee, "Basic Considerations for Expedient Solution Strategies in Nonlinear Finite
Element Analysis," Proceedings, the ASME Region IX Computer Applications Sympo-
sium, Los Angeles, pp. 4/1-18, 1988.

[11.4] W. P. Schulze Schwering, "The Cologne Challenge," Finite Element News, No.4, 1984.

[11.5] S. H. Lee, "The Cologne Challenge-The Replies (Part IV)," Finite Element News, No.5,
pp. 24-25, 1986.

[11.6] R. Kao, "A Comparative Study on the Elastic-Plastic Collapse Strength of Initially
Imperfect Deep Spherical Shells," Int. J. Numerical Methods Eng., Vo1.17, pp. 427-444,
1981.

[11.7] S. H. Lee and D. N. Herting, "Comments on 'A Simple Approach t o Bifurcation and
Limit Point Calculations'," Int. J. Numerical Methods Eng., Vo1.21, pp. 1935-1937,
1985.

[11.8] S. H. Lee, "Rudimentary Considerations for Effective Convergence Criteria in Nonlinear

Finite Element Analysis," Proceedings, 1988 ASME Int. Computers in Engineering
Conference, San Francisco, Vo1.3, pp. 593-598,1988.

[11.9] G. A. Greenbaum and M. F. Rubinstein, "Creep Analysis of Axisymmetric Bodies

Using Finite Elements," Nuclear Engineering and Design, Vo1.7, 1968.

[11.10] L. Morino, J. W. Leech, and E. A. Witmer, "An Improved Numerical Calculation

Technique for Large Elastic-Plastic Transient Deformations of Thin Shells: Part 2 -
Evaluation and Applications," ASME Journal of Applied Mechanics, Vo1.38, pp. 429-
435, 1971.

[11.11] C. Tsay and R. Keene, "MSC/DYNA Verification and Demonstration Problem Man-
ual," 1989.

[11.12] J. M. Kennedy, T. Belytschko, and J. I. Lin, "Recent Developments in Explicit Finite

Element Techniques and Their Application t o Reactor Structures," Nuclear Engineering
and Design, Vo1.97, pp. 1-24, 1986.

[11.13] T . Belytschko, J . I. Lin, and C. S. Tsay, "Explicit Algorithms for the Nonlinear
Dynamics of Shells," Computer Methods in Applied Mechanics and Engineering, Vo1.42,
pp. 225-251,1984.

[I 1.141 Roark and Young, Formulas for Stress and Strain, McGraw-Hill, Fifth Edition, p. 220.

[11.15] S. Timoshenko and J. N. Goodier, Theory of Elasticity, pp. 497-504,2nd ed., McGraw-
Hill, New York, 1951.

Bibliography - 8
[11.16] R. H. MacNeal and R. L. Harder, "A Refined Four-Noded Membrane Element with
Rotational Degrees of Freedom," Computers & Structures, Vol. 28, No. 1, pp. 75-84,
1988.

[11.17] S. H. Lee, "Application of Rheological Model t o Creep Analysis Coupled with Plastic
Deformation," Transactions, The 8th SMiRT Conference, Brussels, Vol. L, pp. 85-92,
August, 1985.

[11.18] S. H. Lee, "Generalized Viscoelastic Model for Creep Analysis Coupled with Plastic
Deformation," Int. J. Numerical Methods in Engineering, Vol. 26, pp. 153-165, 1988.

Bibliography - 9
 APPENDIX A
Nonlinear Bulk Data Description
 CGAP BULK DATA ENTRY (V67)

Bulk Data Entry: CGAP - Gap element connection

Description: Defines a gap or frictional element of the structural model for nonlinear analysis.

Format:
1 2 3 4 5 6 7 8 9 10

CGAP EID PID GA GB X1,GO X2 X3 CID

CGAP 17 2 110 112 13

-
Field Contents

EID Element identification number (Integer > 0).

PID Property identification number of a PGAP entry (Integer 2 0, or blank; if blank,

value of EID will be used). See Remark 1.

GA,GB Connected grid points at ends A and B (Integers > 0; GA # GB). See Remark 2.

Xl,X2,X3 Components of the orientation vector v', given in the global coordinate system of
end A (Real). See Remark 2.

GO Grid point identification number t o optionally supply the orientation vector 5. Vec-
tor Cis from GA to GO (Integer).

CID Coordinate system identification number (Integer, default 0 refers t o the basic co-
ordinate system). See Remark 2.

Remarks:

1. The gap element is intended for the nonlinear solution sequences 66, 99, 106, and 129.
However, it will produce a linear stiffness matrix for the other solutions, but remains
 linear with the initial stiffness. The stiffness used depends on the value for the initial
gap opening (Uofield in the PGAP entry).
2. The gap element coordinate system is defined by one of two methods given below:
(a) If the grid points GA,GB are not coincident, then line AB is the element x-axis and
the orientation vector v'lies in the x-y plane (like the beam element).
(b) If the grid points GA,GB are coincident, coordinate system CID is used for the
element, i.e., the element x-axis is in the T 1 direction, the y-axis in the T2 direction,
etc.
The element coordinate system does not rotate as a result of deflections. Initial gap
openings are defined on the PGAP entry and not by the GA-GB separation distance.
Forces, which are requested with STRESS or ELSTRESS case control commands, are
output in the element coordinate system. F, is positive for compression.
 CREEP BULK DATA ENTRY

Bulk Data Entry: CREEP - Creep Characteristics.

Description: Defines creep characteristics based on the experimental data or known empirical
creep law. This entry will be activated if MATl with the same MID is being used and the
NLPARM entry is prepared for creep analysis.

Format:
1 2 3 4 5 6 7 8 9 10
I
CREEP MID RT EXP FORM TIDKP TIDCP TIDCS THRESH

TYPE a b c d e f g

CREEP 8 0. l.E-9 CRLAW

121 6.985E-6 2.444 7.03254 0.1072 6.733-9 0.1479 3.

-
Field Contents

MID Identification number of a MATl entry (Integer > 0).

RT Reference temperature (To)at which creep characteristics are defined (Real). De-
fault = 0.0. See Remark 1.
-AH
EXP Temperature dependent term, e m , in the creep rate expression (0.0 < Real 5
1.0). Default = 1.OE-9. See Remark 1.

FORM Specifies the form of the input data defining creep characteristics: CRLAW (em-
pirical creep law) or TABLE (tabular input data of creep model parameters). See
Remark 2.

TIDKP Identification number of a TABLES1 entry for specification of the creep model
parameter li'p(a) (Integer > 0). See Remark 3.
 TIDCP Identification number of a TABLES1 entry for specification of the creep model
parameter Cp(u) (Integer > 0). See Remark 3.

TIDCS Identification number of a TABLES1 entry for specification of the creep model
parameter C,(u) (Integer > 0). See Remark 3.

THRESH Threshold limit for creep process corresponding to the threshold strain (0.0 < Real
< 1.OE-3). Default = 1.OE-5. Threshold stress under which creep does not occur
is computed as THRESH*E, where E is the Young's Modulus.

TYPE Specifies the identification number of the empirical creep law type (111 5 Integer
< 300). See Remark 4.

a-?Z Coefficients of the empirical creep law specified in the TYPE field (Real). See
Remark 4.

Remarks:

1. Creep law coefficients (a-g) are usually determined by least squares fit of experimental
data obtained under a constant temperature. The reference temperature (To) at which
creep behavior is characterized must be specified in the RT field if the temperature of
the structure is different from this reference temperature. The unit of the temperature
input ( O F or OC) must be indicated in the PARAM entry TABS as follows:

PARAM, TABS, 273.16 (if Celsius is used)

PARAM, TABS, 459.69 (if Fahrenheit is used)

When the correction for the temperature effect is required, the temperature distribution
must be defined in the Bulk Data entries (TEMP, TEMPPI, and/or TEMPRB) which
are selected by the Case Control command, TEMP(L0AD) = SID, within the subcase.

From the thermodynamic consideration, the creep rate is expressed as

where A H is the free energy of activation

R is the gas constant (= 1.98 cal/mole OK)
T is the absolute temperature
A is strain/sec per activation
If the creep characteristics are defined at the temperature To (RT field), the creep rate
 at the temperature T is corrected by a factor

where (e*) is to be specified in the EXP field.

2. If FORM = CRLAW, the creep law type and the coefficients (a-g) must be specified
in the CREEP continuation. If FORM = TABLE, the creep model parameters (ICp,
C,, and C,) are t o be specified in the TABLES1 entries whose identification numbers
appear in the TIDKP, TIDCP, and TIDCS fields, respectively. In this case, the CREEP
continuation field should not be used.
3. The creep model parameters (K,, C, and C,) represent parameters of the uniaxial rhe-
ological model shown in Figure 1. Tabular entries (xi, yi) in the TABLES1 entry will
consist of (a;, I<,;), (o;, Cp;), or (a;, C,;) for the input of ICp, Cp, and C,, respectively.
In the case of a linear viscoelastic material, the parameter (ICp, Cp, and C,) values are
constant and two data points with an identical parameter value should be specified for
each parameter.

The creep model parameters must have positive values as shown in Figures 2-4. If the
table look-up results in a negative value, the value will be reset t o zero and a warning
message (TABLE LOOK-UP RESULTS IN NEGATIVE VALUE OF CREEP MODEL
PARAMETER IN ELEMENT ID = ****) will be detected.
4. Two different empirical creep laws are available at present. The first class of creep law is

Oak Ridge National Laboratory recommends the parameters A(a), R(o), and I i ( o ) in
the form given in the following table.

Parameter Type 1 Type 2

A@) aab aExp(bo)

R(a) cExp(da) cod

K(a) e[sinh(f a)]g eExp( f a )

Each of the three digits (e.g., 111, 121, etc.) in the TYPE field selects type 1 or 2 for
the parameters A(a), R(a), and K ( a ) , respectively.

The second class of creep law (TYPE = 300) is expressed as

&'(a, t ) = a ob t d

where the values b and d are limited to the ranges

1.0 < b < 8.0

 and
0.2 < d < 2.0
The coefficient g should be blank for creep law type xx2 material and coefficients c, e, f,
and g should be blank for creep law type 300. The coefficients (a-g) are dependent on
the structural units. Caution must be exercised t o make these units consistent with the
rest of the input data.
5. Creep analysis requires an initial static solution a t t=O, which can be obtained by a
subcase with an NLPARM on which DT=O.

Primary
4 Elastic Creep Secondary Creep

++
+++.
+
K, (0)

c, (0) - o(t>

Figure 1. Uniaxial Rheological Model.

a(ksi)

Figure 2. Creep Model Parameter K , as a Function of Stress a.

A -6
 Figure 3. Creep Model Parameter C, as a Function of Stress u.

c. -
Ki hours
ln3

Figure 4. Creep Model Parameter C, as a Function of Stress a.

 MATSl BULK DATA ENTRY (V67)

Bulk Data Entry: MATSl - Stress-dependent nonlinear material properties

Description: Specifies material properties which are stress-dependent for use in material non-
linearity applications. This entry will be activated if MATl with the same MID is being used
in the nonlinear solution sequences (66 and 99).

Format:

MATS1 MID TID TYPE H YF HR LIMIT1 LIMIT2

1

MATS1 17 28 PLASTIC 0.0 1 1 2.+4

-
Field Contents

MID Identification number of a MATl entry (Integer > 0).

TID Identification number of a TABLES1 or TABLEST entry (Integer 2 0). If H is

given, then this field should be blank. See Remark 1.

TYPE Specifies the type of material nonlinearity: NLELAST (Nonlinear elastic) or PLAS-
TIC (Elastoplastic). See Remarks 2 and 3.

H Work hardening slope (slope of stress vs. plastic strain) in units of stress (Real).
For elastic-perfectly plastic cases, H=O.O. For more than a single slope in the plastic
range, the stress-strain data must be supplied on a TABLES1 entry referenced by
TID and this field must be blank. See Remark 3.

YF Yield function (Integer) is selected by one of the following values:

1 for von Mises yield criterion (Default)
2 for Tresca yield criterion
3 for Mohr-Coulomb yield criterion
4 for Drucker-Prager yield criterion
 HR Hardening rule (Integer) is selected by one of the following values:
1 for Isotropic hardening (Default)
2 for Kinematic hardening
3 for combined Isotropic and Kinematic hardening

LIMITl Parameter representing an initial yield point (Real). See Remark 4.

LIMIT2 Parameter representing the internal friction angle for the Mohr-Coulomb and
Drucker-Prager yield criteria. (0 5 Real < 45O). See Remark 4.

Remarks:

1. If TID is given, TABLES1 entries (xi, y;) of stress-strain data ( E ~Yk)

, must conform to
the following rules (see Figure 1):
(a) If TYPE is PLASTIC, the curve must be defined in the first quadrant. The first
point must be at the origin (xl = 0, yl = 0) and the second point (x2, y2) must
be at the initial yield point (Y1 or 2c) specified on the MATS1 entry. The slope of
the line joining the origin to the yield stress must be equal t o the value of E on the
MAT1 entry.
(b) If TYPE is NLELAST, the full stress-strain curve (-oo < x < oa) may be defined
in the first and the third quadrant to accommodate different uniaxial compression
data. If the curve is defined only in the first quadrant, then the curve must start at
the origin (xl = 0, yl = 0) and the compression properties will be assumed identical
t o tension properties.
(c) If TYPE is PLASTIC and TID is given, it may not reference a TABLEST entry.

2. If TYPE is NLELAST, the stress-strain data given in the TABLES1 entry will be used
to determine the stress for a given value of strain. The values H, YF, HR, LIMITl, and
LIMIT2 will not be used in this case.

For nonlinear elastic temperature-dependent material properties in SOL 66, TID must
reference a TABLEST entry. Also a MATT1 entry should be supplied t o specify the
temperature dependence of material properties.

3. If TYPE is PLASTIC, either the table identification, TID, or the work hardening slope,
H, may be specified but not both. If the table ID is omitted, the work hardening slope,
H, should be specified in field 5 unless the material is perfectly-plastic. The plasticity
modulus (H) is related to the tangential modulus (ET) by
 where E is the elastic modulus and ET = $$ is the slope of the uniaxial stress-strain
curve in the plastic region. See Figure 2.

4. LIMITl and LIMIT2 are parameters used in the yield function and must be defined as
follows:

Yield Funtion LIMIT1 LIMIT2

von Mises (1) Yield stress in tension, Y1 Not used
or Tresca (2)
Mohr-Coulomb (3) 2*Cohesion, 2c (in stress units) Angle of internal
or Drucker-Prager (4) friction q5 (in degrees)

Y (or a)

I I
I I E: = Effective Plastic Strain
I I

Figure 1. Stress-Strain Curve Definition with TABLES1 Entry.

Figure 2. Stress-Strain Curve Definition for a Single H.
 NLPARM BULK DATA ENTRY (V67)

Bulk Data Entry: NLPARM - Parameters for Nonlinear Static Analysis

Description: Defines a set of parameters for nonlinear analysis iteration strategy

Format:
1 2 3 4 5 6 7 8 9 10

NLPARM ID NINC DT KMETHOD KSTEP MAXITER CONV INTOUT

EPSU EPSP EPSW MAXDIV MAXQN MAXLS FSTRESS LSTOL

MAXBIS MAXR RTOLB

NLPARM 15 5 0. AUTO 5 25 PW NO

1.E-3 1.E3 1.E7 3 25 4 0.2 0.5

5 20. 20.

Field Contents

ID Identification number (Integer > 0). See Remark 1.

NINC Number of increments (0 < Integer < 1000). See Remark 2.

DT Incremental time interval for creep analysis (Real 2 0). Default = 0.; no creep. See
Remark 3.

KMETHOD AUTO, SEMI or ITER for controlling stiffness updates. Default = AUTO. See
Remark 4.

KSTEP Number of iterations before the stiffness update in ITER option. Default = 5. See
Remark 5 for AUTO and SEMI methods.
MAXITER Limit on number of iterations for each load increment (Integer > 0). Default = 25.
See Remark 6.

CONV Flags (U, P, W or any combination) for convergence criteria. Default = PW. See
Remark 7.

INTOUT Intermediate output Flag (YES, NO or ALL). Default = NO. See Remark 8.

EPSU Error tolerance for displacement (U) criterion (Real > 0.0). Default = 1.OE-3.

EPSP Error tolerance for load (P) criterion (Real > 0.0). Default = 1.OE-3.

EPSW Error tolerance for work (W) criterion (Real > 0.0). Default = 1.OE-7.

MAXDIV Limit on the number of diverging conditions per iteration before the solution is
assumed to diverge. (Integer # 0). Default = 3. See Remark 9.

MAXQN Maximum number of Quasi-Newton correction vectors t o be saved on the data base
(Integer 2 0). Default = MAXITER. See Remark 10.

MAXLS Maximum number of line searches allowed for each iteration (Integer 2 0). Default
= 4. See Remark 11.

FSTRESS Fraction of effective stress (F)used t o limit the subincrement size in the material
routines (0. < Real < 1.0). Default = 0.2. See Remark 12.

LSTOL Line search tolerance (0.01 5 LSTOL < 0.9). Default = 0.5. See Remark 11

MAXBIS Maximum number of bisections allowed for each load increment (-10 < MAXBIS
< 10). Default = 5. See Remark 13.
-

MAXR Maximum ratio for the adjusted arc-length increment relative t o the initial value
<
(1. 5 MAXR 40.). Default = 20. See Remark 14.

RTOLB Maximum value of incremental rotation (in degrees) allowed per iteration to activate
bisection (Real > 2.0). Default=20. See Remark 15.
Remarks:

1. The NLPARM Bulk Data entry is selected by the Case Control command NLPARM =
ID. Each solution subcase requires an NLPARM command.

2. In case of static analysis (DT = 0.) using Newton methods, NINC is the number of
equal subdivisions of the load change defined for the SUBCASE. Applied loads, gravity
loads, temperature sets, enforced displacements, etc. define the new loading conditions.
The differences from the previous case are divided by NINC to define the incremental
values. In case of static analysis (DT = 0.) using arc-length methods, NINC is used t o
determine the initial arc-length for the subcase, and the number of load subdivisions will
not be equal to NINC. In case of creep analysis (DT > O.), NINC is the number of time
step increments.

3. The unit of D T must be consistent with the unit used on the CREEP entry that defines
the creep characteristics. Total creep time for the subcase is DT multiplied by the value
in field NINC, i.e., DT*NINC.

4. The stiffness update strategy is selected in the KMETHOD field.

(a) If the AUTO option is selected, the program automatically selects the most efficient
strategy based on convergence rates. At each step the number of iterations required
t o converge is estimated. Stiffness is updated, if (i) estimated number of iterations
t o converge exceeds MAXITER, (ii) estimated time required for convergence with
current stiffness exceeds the estimated time required for convergence with updated
stiffness, and (iii) solution diverges (See Remarks 9 and 13 for diverging solutions).
(b) If the SEMI option is selected, the program for each load increment (i) performs a
single iteration based upon the new load, (ii) updates the stiffness matrix, and (iii)
resumes the normal AUTO option.
(c) If the ITER option is selected, the program updates the stiffness matrix a t every
KSTEP iterations and on convergence if KSTEP 5 MAXITER. However, if KSTEP
> MAXITER, stiffness matrix is never updated. It should be noted that Newton-
Raphson iteration strategy is obtained by selecting the ITER option and KSTEP
= 1, and Modified Newton-Raphson iteration strategy is obtained by selecting the
ITER option and KSTEP = MAXITER.

5. For AUTO and SEMI options, the stiffness matrix is updated on convergence if KSTEP
is less than the number of iterations that were required for convergence with the current
stiffness.

6. The number of iterations for a load increment is limited t o MAXITER. If the solution does
not converge in MAXITER iterations, the load increment is bisected and the analysis
is repeated. If the load increment can not be bisected (i.e., MAXBIS is reached or
 MAXBIS = 0) and MAXDIV is positive, the best attainable solution is computed and
the analysis is continued to the next load increment. If MAXDIV is negative, the analysis
is terminated.

7. The test flags (U = Displacement error, P = Load equilibrium error, and W = Work
error) and the tolerances (EPSU, EPSP and EPSW) define the convergence criteria. All
the requested criteria (combination of U, P and/or W) are satisfied upon convergence.
Please refer t o Handbook for Nonlinear analysis for more details on convergence criteria.

8. If INTOUT = YES: output requests for displacements, element forces and stresses etc.,
are processed for every computed load increment.
NO: output requests for displacements, element forces and stresses, etc.
are processed only for the last load step of the subcase.
ALL: output requests for displacements, element forces and stresses, etc.
are processed for every computed and user specified load increment.

For Newton family of iteration methods (i.e., no NLPCI command) the option ALL
is equivalent to option YES, since the computed load increment are always equal to the
user specified load increment. For arc-length methods (i.e., NLPCI command is specified)
the computed load increment in general are not going t o be equal to the user specified
load increment, and are not known in advance. The option ALL allows the user to obtain
solutions at the desired intermediate load increments.

9. The ratio of energy errors before and after the iteration is defined as divergence rate
(E", i.e.,

Depending on the divergence rate, the number of diverging iteration (NDIV) is incre-

..
mented as follows:
NDIV = NDIV
NDIV = NDIV
+ 2, if E' 1 1 or Ei < -1012
+ 1, if -1012 < E; < -1.
The solution is assumed t o diverge when NDIV 2 JMAXDIVI. If the solution diverges and
the load increment can not be further bisected (i.e., MAXBIS is reached or MAXBIS is
zero), the stiffness is updated based on previous iteration and the analysis is continued.
If the solution diverges again in the same load increment while MAXDIV is positive,
the best attainable solution is computed and the analysis is continued to the next load
increment. If MAXDIV is negative, the analysis is terminated on second divergence.

10. The BFGS update is performed if MAXQN > 0. As many as MAXQN quasi-Newton
vectors can be accumulated. The BFGS update with these QN vectors provide a secant
modulus in the search direction. If MAXQN is reached, no additional QN vectors will
 be accumulated. Accumulated QN vectors are purged when the stiffness is updated and
the accumulation is resumed.

11. The line search is performed as required, if MAXLS > 0. In the line search the displace-
ment increment is scaled t o minimize the energy error. The line search is not performed
if the absolute value of the relative energy error is less than the value specified in LSTOL.

12. The number of subincrements in the material routines (elastoplastic and creep) are deter-
mined such that the subincrement size is approximately FSTRESS * T (equivalent stress).
FSTRESS is also used t o establish a tolerance for error correction in the elastoplastic
material, i.e.,
error in yield function < FSTRESS * 7
If the limit is exceeded at the converging state, the program will exit with a fatal error
message. Otherwise the stress state is adjusted to the current yield surface.

13. Number of bisections for a load incrementlarc-length are limited to IMAXBISI. Differ-
ent actions are taken when the solution diverges depending on the sign of MAXBIS. If
MAXBIS is positive the stiffness is updated on first divergence and the load is bisected
on second divergence. If MAXBIS is negative the load is bisected every time the solution
diverges until the limit on bisection is reached. If the solution does not converge after
IMAXBISI bisections, the analysis is continued or terminated depending on the sign of
MAXDIV (See Remark 9).

14. MAXR is used in the adaptive load incremntlarc-length method to define the overall
upper and lower bounds on the load incrementlarc-length in the subcase, i.e.,
1 Aln
< - 5 MAXR
MAXR - Alo
where Al, is the arc-length at step n and Alo is the original arc-length. Arc-length
method for load increments is selected by an NLPCI Bulk Data Entry. This entry must
have the same ID as the NLPARM Bulk Data Entry.
15. The bisection is activated if the incremental rotation for any degree-of-freedom (AB,,
Ae, or Ae,) exceeds the value specified for RTOLB. This bisection strategy based on
the incremental rotation is controlled by MAXBIS field.
 NLPCI BULK DATA ENTRY (V67)

Bulk Data Entry: NLPCI - Parameters for Arc-length Methods in Nonlinear Static Analysis.

Description: Defines a set of parameters for the arc-length incremental solution strategies.
This entry is activated for a particular subcase if an NLPARM with the same ID is being used
in the nonlinear solution sequence (SOL 66).

Format:

NLPCI ID TYPE MINALR MAXALR SCALE DESITER MXINC

NLPCI 15 CRIS 0.25 4.0 0. 12 20

-
Field Contents

ID Identification number (Integer > 0) of an associated NLPARM entry. See Remark

1.

TYPE Constraint type (CRIS, RIKS, or MRIKS). Default = CRIS. See Remark 2.

MINALR Minimum allowable arc-length adjustment ratio between increments for the adaptive
arc-length method (0. < Real 5 1.). Default = 0.25. See Remark 3.

MAXALR Maximum allowable arc-length adjustment ratio between increments for the adap-
tive arc-length method (Real 2 1.). Default = 4.0. See Remark 3.

SCALE Scale factor (w) for controlling the loading contribution in the arc-length constraint
(Real 2 0.). Default = 0. See Remark 2.

DESITER Desired number of iterations for convergence to be used for adaptive arc-length
adjustment (Integer > 0). Default = 12. See Remark 4.
 MXINC Maximum number of controlled increment steps allowed within a subcase (Integer
> 0). Default = 20. See Remark 5.

Remarks:

1. The NLPCI Bulk Data entry is selected by the Case Control command NLPARM = ID.
There must also be an NLPARM Bulk Data entry with the same ID. However, in case
of Creep analysis (DT # 0. in NLPARM Bulk Data entry) the arc-length methods can
not be activated, and the NLPCI Bulk Data entry is ignored if specified.

2. The available constraint types are as follows

CRIS: {u; - uO,)~{U; - u:) + w2(pi - p 0)2 - - a12n
RIKS: (21; - U ~ - ~ ) ~ {-Uu:)~ + w 2 ~ p i A P =
1 0
MRIKS: {uk - U ; - ' ) ~ { U ~ - ~ - u:) + w2api(pi-I - PO) = 0
where w is the user-specified scaling factor (SCALE), p is the load factor and A1 is
the arc-length. The constraint equation has a disparity in the dimension by mixing the
displacements with the load factor. The scaling factor (w) is introduced as user input
so that the user can make constraint equation unit-independent by a proper scaling of
the load factor, p. As the value of w is increased, the constraint equation is gradually
dominated by the load term. In the limiting case of infinite w, the arc-length method
would be degenerated t o the conventional Newton's method.

3. The MINALR and MAXALR fields are used to limit the adjustment of the arc-length
from one load increment to the next by

MINALR -5 MAXALR
Alold
The arc-length adjustment is based on the convergence rate (i.e., number of iterations
required for convergence) and the change in stiffness. For constant arc-length during
analysis, use MINALR = MAXALR = 1.

4. The arc-length A1 for the variable arc-length strategy is adjusted based on the number
of iterations that were required for convergence in the previous load increment (I,,,)
and the number of iterations desired for convergence in the current load increment ( I d ) ,
as follows: -

5. The MXINC field is used to limit the number of controlled increment steps in case the
solution never reaches the specified load. This field is useful in limiting the the number
of increments computed for a collapse analysis.
 PBCOMP BULK DATA ENTRY (V67)

Bulk Data Entry: PBCOMP - Beam Property (Alternate Form of PBEAM)

Description: Alternate format to PBEAM entry to define properties of a uniform cross-sectional

beam referenced by CBEAM entry, This entry is also designed to specify lumped areas of the
beam cross section for nonlinear analysis and/or composite analysis.

Format:

-
Field Contents

PID Property identification number (Integer > 0). See Remark 1.

MID Material identification number (Integer > 0). See Remark 2 and 3.

A Area of beam cross section (Real > 0.)

I1 Area moment of inertia in plane 1 about the neutral axis (Real > 0.). See Remark
4.
 I2 Area moment of inertia in plane 2 about the neutral axis (Real > 0.). See Remark
4.

I12 Area product of inertia (Real, default = O., 1112- > 0.). See Remark 4.

J Torsional stiffness parameter (Real > O., default = 0.). See Remark 4.

NSM Nonstructural mass per unit length (Real > O., default = 0.)

K1, K2 Shear stiffness factor K in (KAG) for plane 1 and plane 2 (Real > 0, default = 1.0).
See Remark 5.

M I , M2 (y,z) coordinates of center of gravity of nonstructural mass (Real, default = 0). See
Figure l a , Section 1.3.

N1, N2 (y,z) coordinates of neutral axis (Real, default = 0). See Figure l a , Section 1.3.

SECTION Symmetry option to input lumped areas for the beam cross-section (integer 0 to 5,
default = 0). See Figure lg, Section 1.3, and Remark 6 .

Yi, Zi (y,z) coordinates of the lumped areas in the element coordinate system. (Real, no
default). See Remark 1.

Ci Fraction of the total area for the ith lumped area (Real > O., default = 0.).

MIDi Material identification number for the ith integration point (Integer > 0). See Re-
mark 3.

SOUTi Stress output required (YES) or not (NO) for the various lumped areas (Character,
default = NO). See Remark 7.

Remarks:

1. The PID number must be unique with respect to other PBCOMP entries as well as
PBEAM entries. The PBCOMP entry creates equivalent PBEAM entries in a preface
operation. The third continuation type may be repeated 18 more times. A maximum
of 21 continuations is allowed, i.e., a maximum of 20 lumped areas may be input if
SECTION=5. If SECTION=l through 4, the total number of areas input plus the total
number generated by symmetry must not exceed 20. If the user does not specify these,
 the program defaults t o the elliptically distributed 8 nonlinear rods as usual. See Figure
lg, Section 1.3.

2. For structural problems, PBCOMP entries may reference only MAT1 material entries.
For material nonlinear analysis, the material should be perfectly plastic since the plastic
hinge formulation is not valid for strain hardening. For heat transfer problems, PBCOMP
entries may reference only MAT4 and MAT5 material entries.

3. The values Eo and Go are computed based on the value of MID on the parent entry.
MIDis will follow the same symmetry rules as Ci depending on the value of SECTION.
If MID; field on a continuation is blank, the value will be that of MID on the parent
entry. MID; values may be input on continuations without the corresponding y;, z;, and
C; values, t o allow different stress-strain laws.

4. If the lumped cross-sectional areas are specified, fields 11, 12, and 112will be ignored.
These and other modified values will be calculated based on the input data (y;, zi, C;,
MID;) as follows:

where n is the number of lumped cross-sectional areas specified.

5. Blank fields for K1 and K2 are defaulted to 1.0. If a value of 0.0 is used for K1 and K2,
the transverse shear stiffness becomes rigid and the transverse shear flexibilities are set
to 0.0.

6. As can be seen from Figure lg, Section 1.3, if the user chooses to leave the SECTION
field blank; the program defaults to the elliptically distributed 8 nonlinear rods, similar
to the PBEAM entry. For this particular case it is illegal to supply C; and MID; values.
For a doubly symmetric section (SECTION = I), if the lumped areas are specified on
either axis, the symmetry option will double the areas. For example for the section shown

points 2 and 4 are coincident and so are points 6 and 8. In such cases, it is recommended
that users input the value of area as half of the actual value at point 2 to get the desired
effect.

7. For the case where the user specifies 11,I2 and II2 on the parent entry, he may specify the
stress-output location on continuation entries. The (y,z) coordinates specified on these
entries will serve as stress output locations with the corresponding (2;'s set to 0. Stress
output is provided at the first four lumped area locations only. If one of the symmetry
options is used and fewer than four lumped areas are input explicitly, the sequence of
output locations in the imaged quadrants is shown in Figure lg. For one specific example,
in the model shown in Remark 6, output can be obtained at points 1 and 2 and in the
image points 3 and 4.

The SOUTi field is a provision for features to be implemented later. Stress for the
lumped areas is output regardless of the value of this field. Its main function is to allow
the program to determine if the first continuation is intended for general beam data (the
data starting with K1) or data at a beam station (the data starting with Yl). If the
sixth field of this entry contains a string (such as "YESn or "NOn), the entry is regarded
as data for a beam station. If it is blank or real, it is intended t o be for general beam
data.
Stated another way, the PBCOMP entry may be used without any continuations, or
without the continuation for general beam data ( K l , K2, . . ., SECTION) but with the
beam station data. This practice is consistent with similar provisions on the PBEAM
entry.
 PGAP BULK DATA ENTRY (V67)

Bulk Data Entry: P G A P - Gap Element Properties

Description: Defines the properties of the Gap element

Format:
1 2 3 4 5 6 7 8 9 10

PGAP PID uo Fo Ka Kb Kt PI P2

TMAX MAR TRMIN

PGAP 2 0. 0. 1.E6 l.E-2 1.E5 0. 0. +PG

+PG 0. loo. 0.001

-
Field Contents

PID Property identification number (Integer > 0).

Uo Initial gap opening (Real). Default = 0.0. See Remark 1.

Fo Preload (Real). Default = 0.0. See Remark 1.

K, Axial stiffness for the closed gap, i.e. U, - Ub > Uo, (Real > 0.0). See Remark 2.

Kb Axial stiffness for the open gap, i.e. U, - Ub < Uo, (Real 2 0.0). Default = 10-'Ka.

Kt Transverse stiffness when the gap is in contact (Real >

0.0). Default = pll'i,.
Recommended range of K t is O.OIKa 5 I'it 5 K,, See Remark 3.
 PI Coefficient of static friction (p,) for the adaptive gap element or coefficient of friction
>
in y transverse direction (p,) for the nonadaptive gap element (Real 0.0). Default
= 0.0.

PP Coefficient of kinetic friction ( p k )for the adaptive gap element or coefficient of

friction in z transverse direction (p,) for the nonadaptive gap element (Real 2 0.0,
for adaptive gap element p2 5 p1). Default = p l .

TMAX Maximum allowable penetration used in the adjustment of penalty values (Real).
Default = 0.0. Positive value activates the penalty value adjustment. See Remark
4.

MAR Maximum allowable adjustment ratio for adaptive penalty values ,'A and Kt (1 <
MAR < lo6). Default = lo2. See Remark 5.

TRMIN Fraction of TMAX, defining the lower bound for the allowable penetration (0.0 5
Real < 1.0). Default = 0.001. See Remark 6.

Remarks:

1. Figures 1 through 3 show the GAP element and the force-displacement curves used in
the stiffness and force computations for the element.

2. For most contact problems, K , (penalty value) should be chosen to be three orders of
magnitude higher than the stiffness of the neighboring grid points. A much larger K ,
value may slow down convergence or cause divergence, while a much smaller K , value
may result in an inaccuracy. This value is adjusted as necessary if TMAX > 0.

3. When the GAP is open, there is no transverse stiffness. When the GAP is closed and if
there is friction, the GAP has the elastic stiffness ( K t )in the transverse direction until
the friction force is exceeded and the slip starts to occur.

4. There are two kinds of GAP elements: adaptive GAP and nonadaptive GAP. If TMAX
>0.0, the adaptive GAP element is chosen. When TMAX = 0.0, penalty values will not
be adjusted but other adaptive features will be active (the gap induced stiffness update,
gap induced bisection, and subincremental process). The value of TMAX = -1.0 selects
the nonadaptive (old) GAP element. The recommended allowable penetration TMAX is
about 10% of the element thickness for plates or equivalent thickness for other elements
which are connected to the GAP.

5. The maximum adjustment ratio MAR is used only for the adaptive GAP element. Upper
 and lower bounds of the adjusted penalty are defined by
Kinit
< K < Kinit MAR
MAR- -
where Kinit is the user-specified value of K , or Kt.
6. The parameter TRMIN is used only for the penalty value adjustment in the adaptive
GAP element. T h e lower bound for the allowable penetration is computed by TRMIN *
TMAX. T h e penalty values are decreased if the penetration is below the lower bound.

Figure 1. The GAP Element.

 F, (compression)
t

tension I
uo
+ : (compression)

ua - Ub
Figure 2. Force-Deflection Curve for Nonlinear Analysis.

Nonlinear Shear

Figure 3. Shear Force for GAP Element.

 TABLEST BULK DATA ENTRY (V67)

Bulk Data Entry: TABLEST - Material Property Temperature-Dependence Table

Description: Specifies the material property tables for nonlinear elastic temperature-dependent
materials

Format:
1 2 3 4 5 6 7 8 9 10

TABLEST TID blank blank blank blank blank blank

T1 I1 T2 I2 T3 -etc.-

TABLEST 101

150.0 10 175.0 20 -etc.- ENDT

-
Field Contents

TID Table identification number (Integer > 0)

Ti Temperature values (Real)

Ii Table identification numbers of TABLES1 entries (Integer > 0)

Remarks:

1. Ii must be unique with respect to all TABLES1 and TABLEST table identification num-
bers.

2. Temperature values must be listed in ascending order.

3. The end of the table is indicated by the existence of ENDT in either of the two fields fol-
lowing the last entry. An error is detected if any continuations follow the entry containing
the end-of-table flag ENDT.

4. This table is referenced only by MATS1 entries which define nonlinear elastic (TYPE'=
NLELAST) materials.
 TABLES1 BULK DATA ENTRY (V67)

Bulk Data Entry: TABLESl - Material Property Table, Form 1

Description: Defines a tabular function for stress-dependent material properties such as the
stress-strain curve and creep parameters

Format:
1 2 3 4 5 6 7 8 9 10

TABLES1 ID blank blank blank blank blank blank blank

xl Yl x2 Y2 x3 Y3 x4 Y4

TABLES1 32

0.0 0.0 .01 10000. .02 15000. ENDT

Field Contents

ID Table identification number (Integer > 0)

xi, yi Tabular entries (Real)

Remarks:

1. The xi must be in either ascending or descending order, but not both.

2. Jumps between two points (xi = are not allowed.

3. At least two entries must be present.

4. Any x-y entry may be ignored by placing SKIP in either of the two fields used for that
entry.
5. The end of the table is indicated by the existence of ENDT in either of the two fields fol-
lowing the last entry. An error is detected if any continuations follow the entry containing
the end-of-table flag ENDT.

6. TABLES1 is used to input a curve in the form of

where X is input to the table and Y is returned. The table look-up yT(x), x = X, is
performed using linear interpolation within the table and linear extrapolation outside
the table using the last two end points at the appropriate table end.
 TSTEPNL BULK DATA ENTRY (V67)

Bulk Data Entry: TSTEPNL - Parameters for Nonlinear Dynamic Analysis.

Description: Defines a set of parameters for nonlinear dynamic solution strategy.

Format:
1 2 3 4 5 6 7 8 9 10

TSTEPNL ID NDT DT NO METHOD KSTEP MAXITER CONV

EPSU EPSP EPSW MAXDIV MAXQN MAXLS FSTRESS

MAXBIS ADJUST MSTEP RB MAXR UTOL RTOLB

TSTEPNL 250 1 ADAPT 2 10 PW

l.E-2 l.E-3 l.E-6 2 10 2 0.2

5 5 0 0.75 16.0 0.1 20.

-
Field Contents

ID Identification number (Integer > 0). See Remark 1.

NDT Number of time steps of value DT (Integer 2 4). See Remark 2.

DT Time increment (Real 2 0.0). See Remark 3.

NO Time step interval for output: every ~0~~step solution will be saved for output.
(Integer > 0). Default = 1. See Remark 4.

METHOD AUTO, TSTEP, or ADAPT for dynamic stiffness update and the direct time inte-
gration strategy. Default = ADAPT. See Remark 5.
KSTEP Time step interval in TSTEP method for stiffness update or stiffness update interval
in bisection process for ADAPT method. (Integer > 0). Default = 2. See Remark
6.

MAXITER Limit on number of iterations for each time step (Integer # 0). Default = 10. See
Remark 7.

CONV Flags (U, P, W or any combination) for convergence criteria. Default = PW. See
Remark 8.

EPSU Error tolerance for displacement (U) criterion (Real > 0.0). Default = 1.OE-2.

EPSP Error tolerance for load (P) criterion (Real > 0.0). Default = 1.OE-3.

EPSW Error tolerance for work (W) criterion (Real > 0.0). Default = 1.OE-6.

MAXDIV Limit on the number of diverging conditions for a time step before the solution is
assumed t o diverge (Integer > 0). Default = 2. See Remark 9.

MAXQN Maximum number of quasi-Newton correction vectors t o be saved on the data base
(Integer 2 0). Default = 10. See Remark 10.

MAXLS Maximum number of line searches allowed per iteration (Integer > 0). Default = 2.
See Remark 10.

FSTRESS Fraction of effective stress (7)used to limit the subincrement size in the material
routines (0. < Real < 1.0). Default = 0.2. See Remark 11.

MAXBIS Maximum number of bisections allowed for each time step (-9 5 MAXBIS 5 9).
Default = 5. See Remark 12.

ADJUST Time step skip factor for automatic time step adjustment (Integer 2 0). Default =
5. See Remark 13.

MSTEP Desired number of time steps to obtain the dominant period response accurately
(10 5 Integer 5 200). Default = variable between 20 and 40. See Remark 14.

RB Define bounds for maintaining the same time step for the stepping function in the
automatic time step adjustment method (0.1 5 Real 5 1). Default = 0.75. See
Remark 15.

MAXR Maximum ratio for the adjusted incremental time relative t o D T allowed for time
step adjustment (1. Real 5 32.). Default = 16. See Remark 16.
 UTOL Tolerance on displacement increment below which there is no time step adjustment
(0.001 < Real 5 1.0). Default = 0.1. See Remark 17.

RTOLB Maximum value of incremental rotation (in degrees) allowed per iteration to activate
bisection (Real > 2.0). Default=20. See Remark 18.

Remarks:

1. The TSTEPNL Bulk Data entry is selected by the Case Control command TSTEPNL
= ID. Each subcase (residual superelement solutions only) requires a TSTEPNL entry.
Multiple subcases are assumed to occur sequentially in time. So the initial conditions of
each subcase are defined by the end conditions of the previous subcase.

2. NDT is the number of time steps in AUTO and TSTEP options (KMETHOD field). In
ADAPT option NDT is used to define the total duration for analysis by multiplying NDT
with DT (i.e., NDT*DT). Since, the time increment (At) changes during the analysis in
ADAPT option, the actual number of time steps, in general, will not be equal to NDT.

3. The time increment (At) remains constant during the analysis in AUTO and TSTEP
options, and is equal to DT. In ADAPT option DT is used as a initial value for At.

4. For printing and plotting the solution, data will be output at steps = 0, NO, 2N0, ...
etc., and the last converged step. The Case Control command OTIME may also be used
t o control the output points.

5. The stiffness update strategy as well as the direct time integration method is selected in
the METHOD field.
(a) If the AUTO option is selected, the program automatically updates the stiffness
matrix to improve convergence. The KSTEP value is ignored.
(b) If the TSTEP option is selected, the program updates the stiffness matrix every
K S T E P ~increment
~ of time.
(c) If the ADAPT option is selected, the program automatically adjusts the incremental
time and uses bisection. The ADAPT method allows linear transient analysis, but
AUTO or TSTEP will abort the run if the model does not have any data representing
nonlinearity.
The stiffness matrix is always updated for a new subcase or restart, irrespective of the
option selected.
 6. During the bisection process in ADAPT option, stiffness is updated every KSTEP con-
verged bisected solution to reduce computing cost.

7. The number of iterations for a time step is limited to MAXITER. If MAXITER is

negative, the analysis is terminated when the divergence condition is encountered twice
during the same time step or the solution diverges for five consecutive time steps. If
MAXITER is positive, the program computes the best solution and continues the analysis
on second divergence during a time step. If the solution does not converge in MAXITER
iterations, the process is treated as a divergent process (See Remark 9)

8. The convergence test flags ( U = Displacement error test, P = Load equilibrium error
test, W = Work error test) and the error tolerances (EPSU, EPSP and EPSW) define
the convergence criteria. All requested criteria (combination of U, P, and/or W) are
satisfied upon convergence. It should be noted that at least two iterations are necessary
t o check the displacement convergence criterion.
9. MAXDIV provides control over diverging solutions. Depending on the rate of divergence,
the number of diverging solutions (NDIV) is incremented by 1 or 2. The solution is
assumed t o diverge when NDIV reaches MAXDIV during the iteration. If the bisection
option is used (allowed MAXBIS times) with the ADAPT method, the time step is
bisected upon divergence. Otherwise, the solution for the time step is repeated with a
new stiffness based on the converged state at the beginning of the time step. If NDIV
reaches MAXDN again within the same time step, the analysis is terminated.

10. Non-zero values of MAXQN and MAXLS activate the quasi-Newton update and the line
search process, respectively.

11. The number of subincrements in the material routines are determined such that the
subincrement size is approximately FSTRESS * 5 (equivalent stress). FSTRESS is also
used t o establish a tolerance for error correction in the elastoplastic material, i.e.,

error in yield function < FSTRESS * B

If the limit is exceeded at the converging state, the program will exit with a fatal error
message. Otherwise, the stress state is adjusted t o the current yield surface.

12. The bisection process is activated when divergence occurs and MAXBIS # 0. The number
of bisections for a time increment is limited to (MAXBISI. If MAXBIS is positive and the
solution does not converge after MAXBIS bisections, the best solution is computed and
the analysis is continued to the next time step. If MAXBIS is negative and the solution
does not converge in (MAXBISI bisection, the analysis is terminated.

13. The parameter ADJUST allows the user t o control the automatic time stepping in the
ADAPT option. A value of zero for ADJUST turns off the automatic adjustment com-
pletely. If ADJUST is positive, the time increment is continually adjusted for first few
 steps till a good value of At is obtained. After this initial adjustment, the time increment
is adjusted every ADJUST time steps only. A value of ADJUST a n order greater than
NDT will turn off adjustements after the initial adjustement. Since the automatic time
step adjustment is based on the mode of response and not on the loading pattern, it may
be necessary t o limit the adjustable step size when the period of the forcing function
is much shorter than the period of dominant response frequency of the structure. It is
user's responsibility t o ensure that the loading history is properly traced with ADJUST
option. The ADJUST option should be suppressed for the duration of short pulse load-
ing. If unsure, start with D T which is much smaller than the pulse duration in order t o
properly represent the loading pattern.

14. The recommmended value of MSTEP for nearly linear problems is 20. A larger value
(say 40) is required for highly nonlinear problems. In the default option, the program
automatically computes the value of MSTEP based on the changes in the stiffness.

15. Parameters MSTEP and RB are used t o adjust the time increment during analysis in
the ADAPT option. The adjustment is based on the number of time steps desired t o
capture the dominant frequency response accurately. T h e time increment is adjusted as
follows:
At,+, = f ( r ) At,
where

with
f = 0.25 for r < 0.5 * RB
f = 0.5 for 0 . 5 s R B 5 r < RB
f = 1.0 for RB 5 r < 2
f <
= 2.0 for 2. r < 3./RB
f >
= 4.0 for r 3./RB

16. MAXR is used t o define the upper and lower bounds for adjusted time step size, i.e.,
DT DT
(2MAXBIS ' m)< At 5 MAXR * D T

17. UTOL is used t o filter undesirable time step adjustment, i.e., no time step adjustment
is performed if

18. T h e bisection is activated if the incremental rotation for any degree-of-freedom (AB,,
AB, or AO,) exceeds the value specified for RTOLB. This bisection strategy based on
the incremental rotation is controlled by MAXBIS field.
Index
absolute convergence, 3.6-5, 9.1-8 buckling, 10.1-1
absolute divergence, 3.3-3 bulk data, 1.4-6
adaptive
arc-length method, 3.7-4
bisection, 3.7-6, 8.4-4 case control, 1.4-4, 7.1-1, 9.1-1
GAP, 4.3-6 Cauchy's deformation tensor, 5.6-2
Newmark's method, 8.3-2 Cauchy stress tensor, 5.6-3
penalty value adjustment, 4.3-11,12 CDAMPi, 9.3-6
solution algorithm, 3.1-1 combined hardening, 6.2-10
time stepping, 8.4-1 constitutive relations, 6.1-1, 5.6-10
Almansi finite strain, 5.6-2 constraint, 7.5-1
ALRATIO, 3.7-5, 7.1-6 contact analysis, 4.3-1, 11.5-1
angular velocity tensor, 5.6-5 continuum mechanics, 5.6- 1
anisotropic friction, 4.3-13 contraction factor, 3.6-4
anisotropy, 6.6- 1 convective coordinates, 5.2- 1
apparent Poisson's ratio, 6.2-27 convective rate of change, 5.6-5
arc-length method, 3.7-1, 7.1-6 convergence, 3.6- 1
associated flow rule, 6.2-12,13,14 absolute convergence, 3.6-5, 9.1-8
automatic time stepping, 8.1-1, 8.4-1 criteria, 3.6-1
AUTOSPCR, 1.4-8 information output, 7.1-7, 9.1-7
auxiliary angle, 5.4-2 rate, 3.3-2
tolerance, 3.6-5
coordinate system, 2.3-1
basic coordinate system, 2.3-1, 5.2-2 convective, 5.2-1
Bauschinger Effect, 1.5-2, 6.2-2 element, 5.2-2
BEAM, 4.1-3,6 isoparametric, 4.2- 1
BEND, 4.1-4 material, 4.1-4
BETA, 1.4-8 coordinate transformations, 2.3-1, 5.2-2
BFGS update, 3.5-1,2, 11.2-1 corotational stress rate tensor, 5.6-5
criteria, 3.5-3 Coulomb friction law, 4.3-7
for arc-length method, 3.5- 11 COUPMASS, 9.3-3
biaxial loading, 7.4-2 CREEP, 4.1-2, 6.1-5
bilateral stress-strain relation, 6.3-4 creep
bisection algorithm, 8.4-4, 8.5-5 analysis, 6.4-1,4, 11.3-1
boundary conditions, 2.1-2, 7.5- 1 strain, 11.8-1
BUCKLE, 1.4-8, 10.1-3 strain rate, 6.4-2,4, 11.8-3

Index - 1
Crisfield's method, 3.7-3, 7.1-6 E-FIIL'AL, 7.1-7, 9.1-8
criteria EIGB, 2.5-1
BFGS update, 3.5-3 elasto-plastic-creep relations, 6.4-6
convergence, 3.6-1 elasto-plastic material, 6.2-19
divergence, 3.3-2, 3.6-6, 8.4-4 element
line search, 3.4-3 coordinate system, 2.3-1, 4.2-8, 5.2-2
time expiration, 3.3-3 matrix, 4.2-5,6, 5.1-2
yield, 6.2-3 nonlinear analysis, 4.1-2
critical buckling load, 10.1-2 stiffness, 2.2-3, 4.2-6
CSHEAR, 4.1-4 enforced motion, 7.5-1, 9.4-5
CVISC, 9.3-6 ENIC, 7.1-7
cyclic loading, 1.5-2, 6.2-29 EPI, 7.1-7
equilibrium, 2.2-3, 3.2-1, 8.2-1
error functions, 3.6-3
damping error vector, 3.2-1, 8.2-1
matrix, 9.3-7 EUI, 7.1-7
modeling, 9.3-4 Eulerian strain, 5.6-2,3,11
numerical, 1.4-9, 8.3-6, 9.1-2 Eulerian formulation, 5.6-1
DBDRNL, 1.4-8 EWI, 7.1-7
deformation gradient, 5.6-1,3 Executive System, 1.3-1
DESITER, 7.1-6 EXP, 6.1-6
DFP method, 3.5-2
diagnostic outputs (DIAG 50,51),
SOL 66 and 106, 7.2-4 FACTOR, 7.1-7, 9.1-8
SOL 99 and 129, 9.5-1 FBS, 3.5-5
differential stiffness, 3.2-1, 5.1-5 FIAT (File Allocation Table), 1.3-1
direct time integration, 8.1-1 finite element method, 2.2-1, 4.1-1
discrete damping elements, 9.3-4 finite strain, 5.6-2
discretization, 2.1-2 follower force, 5.1-5
displaced element coordinate system, 2.3-1, follower matrix, 3.7- 1, 5.3-2
5.2-3 frame-indifference, 5.6-5
displacement sets, 2.4- 1 friction element, 4.3-3
divergence criteria, 3.3-2, 3.6-6, 8.4-4 frictional yield function, 4.3-7
DLMAG, 7.1-7, 9.1-8 FSTRESS, 7.1-5, 9.1-5
DMAP, 1.3-1 functional module, 1.3-1
doubling scheme, 3.4-4,13
Druker-Prager yield criterion, 6.2-5
DT, 7.1-3, 9.1-4 G, 9.3-7
dynamic loads, 9.4-1 GAP element, 4.3-1, 11.5-1
Gaussian elimination method, 3.2- 1
generalized effective plastic strain, 6.2-14
effective plastic strain increment, 6.2-14 generalized logarithmic strain, 5.6-10
effective strain for nonlinear elastic, 6.3-1 geometric nonlinearity, 1.5-1, 2.1-2, 5.1-1
E-FIRST, 7.1-7, 9.1-8 geometric stiffness, 5.1-5

Index - 2
gimbal angles, 5.4-1
gimbal lock, 5.4-2 Lagrange multiplier method, 4.3-1
GINO, 1.3-1 Lagrangian description, 5.6-3
global coordinate system, 2.3-1 Lagrangian formulation, 5.6-1
Green's deformation tensor, 5.6-2 Lagrangian strain, 5.6-2,3
Green's finite strain, 5.6-2 Lagrangian strain rate, 5.6-10
LAMBDA, 7.1-7, 9.1-8
LANGLE, 1.4-9, 5.4-1,3
Herzian problem, 4.3-25 large rotation, 5.1-2, 5.4-1, 5.5-1
Hessian matrix, 3.5-1 left stretch tensor, 5.6-3
hierarchy of nonlinear looping, 1.3-2 LGDISP, 1.4-9, 4.1-2, 5.1-2
hyperelastic material, 5.6-8,11 LIMIT1, LIMIT2, 6.1-2
hysteresis, 6.2-29 line elements, 4.1-2
line search, 3.4-1, 11.2-1
IFP (Input File Processor), 1.3-1 algorithm, 3.4-10
Illinois algorithm, 3.4-4 criteria, 3.4-3
impact analysis, 11.6-1 iteration, 3.4-11
implicit integration, 8.1-1, 9.1-2 linear interpolation, 3.4-4,13
increment, 2.5- 1,2 load,
initial anisotropy, 6.6-1 dynamic, 9.4-1
input data structure, 1.4-1 static, 7.3-1
integration superelement, 7.3-3
direct time, 8.1-1 LOADINC, 1.4-9
Gauss, 4.2-6 loading and unloading, 6.2-1
implicit, 8.1-1, 9.1-2 local coordinate system, 2.3-1
Newmark, 8.3-1 logarithmic strain, 5.6-4,10
three-point method, 8.2-1 LOOPID, 1.4-9
two-point method, 8.3-1 LSEQ, 10.3-1
INTOUT, 3.7-6, 7.1-4 LSTOL, 3.4-4, 7.1-5
isoparametric elements, 4.2- 1
isotropic hardening, 6.2-9 mass
iteration, 2.5-2, 8.2-2,3, 8.5-1 matrix, 9.3-2
modeling, 9.3- 1
Jaumann stress, 5.6-5,6,10 properties, 9.3-1
MAT2, 6.6-1
MAT9, 6.6-1
KGROT, 1.4-8, 11.7-1 material coordinate system, 2.3-1, 4.114
Kelvin-Maxwell model, 6.4-3, 11.8-1 material derivative, 5.6-4,10
kinematic hardening, 6.2-9 material nonlinearity, 1.5-1, 2.1-2, 6.1-1
Kirchhoff & Love, plate theory, 5.1-1 material properties entries, 6.6-1
Kirchhoff stress, 5.6-3,10 material stiffness, 2.2-2
KMETHOD, 3.3-1,2, 11.1-12 MATS1, 6.1-2
KSTEP, 3.3-1, 7.1-3, 9.1-4 MAXALR, 3.7-5, 7.1-6

Index - 3
MAXBIS, 7.1-5, 9.1-5 nonlinear characteristics, 1.5-1
MAXDIV, 3.3-2, 7.1-5, 9.1-5, 11.1-12 nonlinear elasticity, 6.3-1
MAXITER, 7.1-4, 9.1-4 nonlinear element, 1.2-1, 4.1-2
MAXLP, 1.4-9 properties, 1.2-3
MAXLS, 3.4-4, 7.1-5, 9.1-5 nonlinear modal analysis, 10.2-1
MAXQN, 3.5-5, 7.1-5, 9.1-5 nonlinear static analysis, 7.1-1, 11.1-1
MAXR, 7.1-5, nonlinear transient response analysis, 9.1-1
Maxwell-Kelvin model, 6.4-3, 11.8-1 nonproportional loading, 7.4- 1
method of false position, 3.4-2,5 normality principle, 6.2- 13
MINALR, 3.7-5, 7.1-6 N-QNV, 7.1-7, 9.1-8
modal coordinate system, 2.3-1 numerical damping, 1.4-9, 8.3-6, 9.1-2
modeling, 1.5-1
Mohr-Coulomb yield criterion, 6.2-5
MPC, 2.4-1,3 one-step integration, 8.3-1
MPL (Module Properties List), 1.3-1 open/close boundary, 4.3-9
MXINC, 7.1-7 OSCAR, 1.3-1
output
diagnostics, 7.2-4, 9.5-1
natural strain, 5.6-4 interpretation for large rotation, 5.4-6
NDAMP, 1.4-9, 8.3-6, 9.1-2 iteration process, 7.1-7, 9.1-7
NDIV, 3.3-3, 8.4-4 line search, 3.4-8
neutral loading, 7.4-2
NEWK, 3.3-1
Newmark integration, 8.3-1 parameters, 1.4-8
Newton-Raphson method, 2.5-2,4, 3.2-1 parametric studies
NINC, 3.7-4, 7.1-3 automatic time stepping, 8.6-1
NLAYERS, 1.4-9 BFGS updates, 11.2-1
NLITER, 7.2-1 KGROT, 11.7-2
input data block, 7.2-1,2 line search, 11.2-1
output data block, 7.2-2,3 penalty value in GAP, 4.3-17
parameters, 7.2-3,4 static solution method, 11.1-14
NLPARM, 2.5-1, 7.1-2 penalty GAP element, 4.3-2,3
NLPCI, 7.1-6 penalty values
NLTRD module, 8.2-1,4, 9.5-1 adaptive adjustment, 4.3-11
NLTRD2 module, 8.3-1,4, 8.5-6, 9.5-2,3 as input parameters, 4.3-6
NMATD, 6.3-2 selection of Gap stiffness, 11.5-3
NMEP, 6.2-24 Piola-Kirchhoff stress, 5.6-3
NMEVD, 6.3-2 PGAP, 4.3-5
NMLOOP, 1.4-10, 10.2-1 plane strain, 6.2-17,18
nodal forces, 2.2-2,4.2-7 plane strain element, 4.1-5
NOGO flag, 3.3-3 plane stress, 6.2-18
NOLIN, 9.4-1 plasticity, 6.2-1
non-adaptive GAP element, 4.3-13 plate & shell elements, 4.1-4
nonlinear capabilities, 1.2- 1 plotting output codes, 1.4-17

Index - 4
polar decomposition, 5.6-2 user interface, 7.1-1
post-buckling analysis, 3.7-1,12 SOL 99 and 129, 1.4-1, 1.5-1, 8.1-3
Prandtl-Reuss stress-strain relation, 6.2-12 capabilities, 1.2-2
Preface Operation, 1.3-1 examples, 11.4-1, 11.6-1
primary creep, 6.4- 1 static analysis, 10.3-1
probable divergence, 3.3-3 user interface, 9.1-1
program architecture, 1.3-1 solid elements, 4.1-4
proportional loading, 6.2-1 solution method, 2.5-1, 3.1-1
spatial derivative, 5.6-4
SPC, 2.4-2,4
QUAD4, 7.8-1, 11.7-2 spin tensor, 5.6-4
quadratic convergence, 3.2-2 SSG3, 1.4-11
quadratic interpolation, 3.4-13 stability analysis, 8.2-3,4
quasi-Newton ( Q N ) update, 3.5-1, 11.2-1 static loads, 7.3-1
stress stiffness, 5.1-4
rate of deformation tensor, 5.6-4 stretching tensor, 5.6-4
reduction of system equations, 2.4-1 stick-slip behavior, 4.3-8
restarts stick-slip boundary, 4.3-9
in SOL 66 and 106, 7.6-1 stiffness
in SOL 99 and 129, 9.2-1 differential, 3.2-1, 5.1-5
RF63D89, 10.2-7 element, 2.2-3, 4.2-6
rheological model, 6.1-8, 6.4-1, 11.8-1 geometric, 5.1-5
right stretch tensor, 5.6-3 material, 2.2-2, 6.5-1
Rik's method, 3.7-2, 7.1-6 stress, 5.1-4
Ritz method, 2.2-1 updates, 3.2-1, 3.3-1
ROD element, 4.1-2,3 STIME, 1.4-11, 9.2-1
rotation vector, 5.4-3 strain hardening, 6.2-1,9
RT, 6.1-6 strain-rate increment,
RTOLB, 5.4-6, 7.1-6, 9.1-7 stress data recovery, 4.2-9
structual damping, 9.3-4,5
SUBID, 1.4-11
SCALE, 7.1-7 subincremental algorithm, 4.3-9
SDATA, 1.4-10, 9.2-2 superelement, 7.3-3
secondary creep, 6.4-1 surface elements, 4.1-4
shape function, 4.2-1,2,3 SYSTEM(141), 3.4-13
shell element, 4.1-4 SYSTEM(145), 3.5-12
slip yield surface, 4.3-8
SLOOPID, 1.4-10, 9.2-1
SMALLDB, 1.4-10 TABLEST, 6.5-2
snap-through, 3.7-1,9, 11.1-1 TABS, 1.4-11
SOL 64, 1.1-1, 5.5-2 tangent material stiffness, 6.5-1
SOL 66 and 106, 1.4-1, 1.5-1, 7.2-6 tangential stiffness, 3.2-2
examples, 11.1-1, 11.3-1, 11.5-1 TEMP, 6.5-1,2,3
tertiary creep, 6.4-1

Index - 5

- - - - -
TESTNEG, 1.4-12, 3.3-1 GAP, 4.3-14
thermal strain, 6.5-1 initial anisotropy, 6.6-1
thermal stress analyses, 6.5-3 large rotation, 5.5-1
required input, 6.5-3, 7.3-4 nonlinear buckling analysis, 10.1-7
thermo-elasticity, 6.5-1 nonlinear modal analysis, 10.2-1
three-point method, 8.2-1 nonproportional loading, 7.4-1
THRESH, 6.1-6 plasticity, 6.2-27
time expiration criteria, 3.3-3 restart in SOL 66, 7.6-2
total Lagrangian, 5.6-6,7 restart in SOL 99, 9.2-3
transverse shear stiffness, 4.3-6 static analysis in SOL 66, 7.7-1
Tresca yield criterion, 6.2-5 static analysis in SOL 99, 10.3-2
TRIA3, 11.7-2,8 surface contact, 4.3-25, 4.3-29
TSTATIC, 1.4-12, 10.3-1 thermo-elasticity, 6.5-3
TSTEPNL, 8.2-1, 8.3-1, 9.1-1,2, 10.3-1 transient response analysis, 8.6-1
two-point method, 8.3- 1 viscoelastic, 6.4-1, 11.8-1
viscous damping, 9.3-4,5
volume integration, 4.2-6,7
uniaxial stress-strain, 6.3-3 von Mises yield criterion, 6.2-3
uniaxial stress, 6.2-18,19 vorticity tensor, 5.6-4
UPARTN, 2.4-2
updated Lagrangian, 2.2-2, 5.6-1,6,7
user interface, 1.4-1 W3, 1.4-12
arc-length, 7.1-6 W4, 1.4-12
buckling analysis, 10.1-2 weighted normalization, 3.6-3,4
creep analysis, 6.1-5
GAP element, 4.3-4
geometric nonlinearity, 5.1- 1 yield criteria, 6.2-1,3
loads, 7.3-1, 9.3-1 yield function, 6.2-6
material nonlinearity, 6.1-1 yield function derivatives, 6.2-15
nonlinear modal analysis, 10.2-1 yield stress, 6.2-1
static analysis, 7.1-1 yield surface, 6.2-1
thermo-elasticity, 6.5-1
transient analysis, 9.1-1

velocity gradient, 5.6-4

verification and validation problems
adaptive time stepping, 8.6-1
arc-length method, 3.7-9
buckling analysis, 10.1-3
follower forces, 5.3-7
creep, 6.4-9, 6.4-13
enforced displacement, 7.5-2
friction, 4.3-19

Index - 6