Madhujit Mukhopadhyay

Abdul Hamid Sheikh

Matrix and
Finite Element
Analyses
of Structures
Matrix and Finite Element Analyses of Structures
Madhujit Mukhopadhyay · Abdul Hamid Sheikh

Matrix and Finite Element

Analyses of Structures
Madhujit Mukhopadhyay Abdul Hamid Sheikh
Indian Institute of Technology Indian Institute of Technology
Kharagpur, India Kharagpur, India

ISBN 978-3-031-08723-3 ISBN 978-3-031-08724-0 (eBook)

https://doi.org/10.1007/978-3-031-08724-0

Jointly published with ANE Books Pvt. Ltd.

In addition to this printed edition, there is a local printed edition of this work available via Ane Books in
South Asia (India, Pakistan, Sri Lanka, Bangladesh, Nepal and Bhutan) and Africa (all countries in the
African subcontinent).
ISBN of the Co-Publisher’s edition: 978-81-80520-75-4

1st edition: © Authors 2015

© The Author(s) 2022
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or
information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publishers, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publishers nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publishers remain neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
To
Our Families
Ratna
Daisy
Kuttus
Aniq
Chuchu
Shaon
Arko
Preface

A revolutionary change has been brought about in the outlook of many fields of
engineering and science with the advent of the electronic digital computer. Amongst
them structural engineering forms an important discipline. With the innovations,
improvements and rapid progress of the development of sophisticated computing
systems, both in terms of hardware and software and with the fading away of the
boundaries of different disciplines and the enlargement of the interdisciplinary work,
never before has so much emphasis been laid in the computer aided analysis and
design as it is today. The main objective of the book is to acquaint the engineers
about the modern computer based techniques used in structural analysis.
This textbook is primarily written as a basic learning tool for senior undergraduate
and postgraduate students in the departments dealing with the mechanics of solid and
structural systems. The material of the book has been developed over a number of
years of teaching senior undergraduate and postgraduate students of different disci-
plines of IIT, Kharagpur. The materials of the book have been presented in sufficiently
simple manner. Basing on the knowledge of classical methods of structural analysis,
the reader is introduced to ‘Matrix and the Finite Element Methods’ and then exposed
to advanced topics in a more gradual manner. The book is self-contained in that most
of the background materials required for the analysis of structures have been dealt
with.
The presentation of the matter of the book is made as simple as possible so that
engineers who are interested may use this for his own purpose. A large number
of diagrams and illustrations make the text easy to understand. The book presents
‘matrix’, ‘finite element’, ‘computer’ and ‘structural analysis’ in a unified and inte-
grated manner. The role of computers is of paramount importance to the present
day structural analysis and design. Computer software and structural analysis have
progressed side by side throughout the text as the matter presented in the text are
primarily meant for large-sized structures. All the computer codes presented are in
two languages—‘FORTRAN’ and ‘C’. The computer programs in ‘C’ are becoming
quite popular now-a-days due to their versatility.
The senior author is grateful to his students—Dr. Manoranjan Barik of National
Institute of Technology, Rourkela and Dr. K. S. R. K. Murthy of Indian Institute of

vii
viii Preface

Technology, Guwahati for the assistance rendered by them. Most of the computer
programs in ‘C’ language have been developed by Mr. Dhirendra Kumar Singh, an
undergraduate student of the Department for which the authors are grateful to him.
The authors are very much thankful to Continuing Education Cell, IIT Kharagpur
for providing financial support for the preparation of the manuscript.
Sincere thanks are due to Mr. Parimal Kumar Roy for typing the manuscript and
preparing a large number drawings. The authors are thankful to Chinmoy Mukherjee
for preparing the necessary of drawings given to him. They also sincerely thank their
wives Ratna and Daisy for the patience and forbearance shown by them during the
preparation of the manuscript.

Kharagpur, India Madhujit Mukhopadhyay

Abdul Hamid Sheikh
About This Book

The main objective of the book is to acquaint the engineers about the computer-based
techniques used in structural analysis. This textbook is primarily written as a basic
learning tool for undergraduate and postgraduate students in the departments dealing
with the mechanics of solids and structural systems. The materials of the book have
been presented in sufficiently simple manner. Based on the knowledge of classical
methods of structural analysis, the reader is introduced to ‘Matrix and the Finite
Element Methods’ and then exposed to advanced topics in a more gradual manner.
The book is self-contained in that most of the background materials required for the
analysis of structures have been dealt with.
The presentation of the matter of the book is made as simple as possible so that
engineers who are interested may use this for its own purpose. A large number of
diagrams and illustrations may help the text easy to understand. The book presents
‘matrix’, ‘finite element’, ‘structural analysis’ and ‘computer implementation’ in
a unified and integrated manner. Computer programs and structural analysis have
progressed side by side throughout the text as the matter presented in the text are
primarily meant for large-sized structures. All the computer codes presented are in
two languages—‘FORTRAN’ and ‘C’.

ix
Contents

1 Basic Concepts of Structural Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Types of Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Objective of Structural Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Materials and Basic Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.5 General Methods of Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5.1 Equilibrium Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5.2 Compatibility Conditions . . . . . . . . . . . . . . . . . . . . . . . . 5
1.6 Force–Displacement Relationship . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.7 Statical Indeterminacy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.7.1 Plane Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.7.2 Space Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.8 Kinematic Indeterminacy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.9 Two Approaches of Structural Analysis . . . . . . . . . . . . . . . . . . . . . 11
2 Energy Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Principle of Virtual Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Principle of Complementary Virtual Work . . . . . . . . . . . . . . . . . . 15
2.4 Principle of Minimum Potential Energy . . . . . . . . . . . . . . . . . . . . 15
2.5 Principle of Minimum Complementary Energy . . . . . . . . . . . . . . 16
2.6 Castigliano’s Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.7 Determination of Displacements . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
References and Suggested Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3 Introduction to the Flexibility and Stiffness Matrix Methods . . . . . . 21
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 The Flexibility Matrix Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3 The Stiffness Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.4 Incorporation of Different Loading Conditions . . . . . . . . . . . . . . 50
3.5 Other Types of Loadings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.5.1 Treatment by the Flexibility Matrix Method . . . . . . . . 52

xi
xii Contents

3.5.2 Treatment by the Stiffness Method . . . . . . . . . . . . . . . . 54

3.6 Incorporation of Shear Deformation . . . . . . . . . . . . . . . . . . . . . . . . 58
3.7 Relation Between Flexibility and Stiffness Matrices . . . . . . . . . . 59
3.8 Equivalent Joint Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.9 Choice of the Method of Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 62
References and Suggested Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4 Direct Stiffness Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2 Local and Global Coordinate System . . . . . . . . . . . . . . . . . . . . . . . 75
4.3 Transformation of Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.3.1 Transformation of Member Coordinate Axes . . . . . . . 76
4.3.2 Transformation of Member Displacement
Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.3.3 Transformation of the Member Force Matrix . . . . . . . 79
4.3.4 Transformation of the Member Stiffness Matrix . . . . . 80
4.4 Transformation of the Stiffness Matrix of the Member
of a Truss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.5 Transformation of the Stiffness Matrix of the Member
of a Rigid Frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.6 Transformation of the Stiffness Matrix of the Member
of a Grillage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.7 Transformation of the Stiffness Matrix of the Member
of a Space Frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.8 Horizontally Circular Curved Beam Element . . . . . . . . . . . . . . . . 89
4.9 Overall Stiffness Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.10 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.10.1 Boundary Conditions Corresponding to Skewed
Supports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.11 Computation of Internal Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.12 Computer Program for the Truss Analysis by the Direct
Stiffness Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.13 Computer Program for the Frame Analysis by Direct
Stiffness Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.14 Computer Program for the Grillage Analysis by the Direct
Stiffness Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
References and Suggested Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5 Substructure Technique for the Analysis of Structural Systems . . . . 113
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.2 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.3 Direct Stiffness Method Restated . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.4 The Substructure Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
5.5 An Illustrative Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.6 Computer Program for the Truss Analysis
by the Substructure Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Contents xiii

References and Suggested Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

6 The Flexibility Matrix Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
6.2 Element Flexibility Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
6.3 Principle of Contragredience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.4 The Equilibrium Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
6.5 Construction of the Flexibility Matrix of the Structure . . . . . . . . 138
6.6 Matrix Determination of the Displacement Vector . . . . . . . . . . . . 139
6.7 Determination of Member Forces . . . . . . . . . . . . . . . . . . . . . . . . . . 140
6.8 Procedure of the Analysis of Statically Indeterminate
Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
6.9 Illustrated Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
6.10 Choice of the Released Structure . . . . . . . . . . . . . . . . . . . . . . . . . . 145
References and Suggested Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
7 Elements of Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
7.2 Some Notations and Relations in the Theory of Elasticity . . . . . 149
7.2.1 Surface and Body Forces . . . . . . . . . . . . . . . . . . . . . . . . 149
7.2.2 Components of Stresses . . . . . . . . . . . . . . . . . . . . . . . . . 150
7.2.3 Components of Strain . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
7.2.4 Stress–Strain Relationship . . . . . . . . . . . . . . . . . . . . . . . 151
7.3 Two-Dimensional Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
7.3.1 Plane Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
7.3.2 Plane Strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
7.3.3 Differential Equations of Equilibrium . . . . . . . . . . . . . . 154
7.4 Bending of Thin Plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
7.4.1 Basic Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
7.4.2 Deformation of the Plate . . . . . . . . . . . . . . . . . . . . . . . . . 156
7.4.3 Strain–Displacement Relationship . . . . . . . . . . . . . . . . 157
7.4.4 Stress–Strain Relationship . . . . . . . . . . . . . . . . . . . . . . . 157
7.4.5 Equilibrium Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 159
7.4.6 Differential Equation for Deflection . . . . . . . . . . . . . . . 160
7.4.7 Shearing Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
7.5 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
7.5.1 Simply Supported Edge . . . . . . . . . . . . . . . . . . . . . . . . . 161
7.5.2 Clamped Edge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
7.5.3 Free Edge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
7.5.4 Elastically Supported Edge . . . . . . . . . . . . . . . . . . . . . . . 163
7.5.5 Edge Having Elastic Rotational Restraint . . . . . . . . . . 164
7.6 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
References and Suggested Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
xiv Contents

8 Introduction to the Finite Element Method . . . . . . . . . . . . . . . . . . . . . . 167

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
8.2 The Finite Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
8.3 Brief History of the Development of the Finite Element
Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
8.4 Basic Steps in the Finite Element Method for the Solution
of Static Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
8.5 Advantages and Disadvantages of the Finite Element
Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
References and Suggested Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
9 Finite Element Analysis of Plane Elasticity Problems . . . . . . . . . . . . . 177
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.2 Three-Noded Triangular Element . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.2.1 Displacement Function . . . . . . . . . . . . . . . . . . . . . . . . . . 178
9.2.2 Displacement Function Expressed in Terms
of Nodal Displacements . . . . . . . . . . . . . . . . . . . . . . . . . 179
9.2.3 Strain–Nodal Parameter Relationship . . . . . . . . . . . . . . 180
9.2.4 Stress–Strain Relationship . . . . . . . . . . . . . . . . . . . . . . . 181
9.2.5 Derivation of the Element Stiffness Matrix . . . . . . . . . 182
9.2.6 Determination of Element Stresses . . . . . . . . . . . . . . . . 183
9.3 Criteria for the Choice of the Displacement Function . . . . . . . . . 184
9.4 Polynomial Displacement Functions . . . . . . . . . . . . . . . . . . . . . . . 185
9.5 Verification of the Convergence Criteria
of the Displacement Function of 3-Noded Triangular
Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
9.6 Number of Terms in a Polynomial . . . . . . . . . . . . . . . . . . . . . . . . . 186
9.7 Four-Noded Rectangular Element . . . . . . . . . . . . . . . . . . . . . . . . . 187
9.7.1 Displacement Function . . . . . . . . . . . . . . . . . . . . . . . . . . 188
9.7.2 Displacement Function in Terms of Nodal
Displacements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
9.7.3 Strain-Nodal Displacement Relationship . . . . . . . . . . . 190
9.7.4 Stress–Strain Relationship . . . . . . . . . . . . . . . . . . . . . . . 190
9.7.5 Derivation of the Element Stiffness Matrix . . . . . . . . . 191
9.7.6 Evaluation of Element Stresses . . . . . . . . . . . . . . . . . . . 191
9.8 A Note on the Rectangular Element . . . . . . . . . . . . . . . . . . . . . . . . 191
9.9 A Note on Element Stresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
9.10 Computer Program for the Plane Stress Analysis Using
Three–Noded Triangular Element . . . . . . . . . . . . . . . . . . . . . . . . . 192
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
10 Isoparametric and Other Element Representations
and Numerical Integrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
10.2 Shape Function or Interpolation Function . . . . . . . . . . . . . . . . . . . 197
10.3 Determination of Shape Functions . . . . . . . . . . . . . . . . . . . . . . . . . 198
Contents xv

10.3.1 Linear 2-D Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198

10.3.2 Quadratic 2-D Element . . . . . . . . . . . . . . . . . . . . . . . . . . 200
10.4 Plane Stress Isoparametric Linear Element . . . . . . . . . . . . . . . . . . 201
10.4.1 Displacement Function in Terms of Nodal
Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
10.4.2 Strain-Nodal Parameter Relationship . . . . . . . . . . . . . . 201
10.4.3 Evaluation of [B] Matrix . . . . . . . . . . . . . . . . . . . . . . . . . 202
10.4.4 Element Stiffness Matrix . . . . . . . . . . . . . . . . . . . . . . . . 204
10.4.5 Convergence of Isoparametric Elements . . . . . . . . . . . 204
10.4.6 Concept of Isoparametric Element . . . . . . . . . . . . . . . . 206
10.5 Numerical Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
10.5.1 Gaussian Quadrature Formula . . . . . . . . . . . . . . . . . . . . 207
10.5.2 Gaussian Integration of Two Variables . . . . . . . . . . . . . 207
10.6 Lagrangian Interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
10.7 Natural Coordinates and Higher Order Triangular
Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
10.7.1 One-Dimensional Element . . . . . . . . . . . . . . . . . . . . . . . 211
10.7.2 Higher Order Triangular Elements . . . . . . . . . . . . . . . . 212
10.8 The Quadratic Triangle for the Plane Stress Problem . . . . . . . . . 214
10.9 Numerical Integration of Area Coordinates . . . . . . . . . . . . . . . . . 216
10.10 Triangular Isoparametric Elements for the Analysis
of Plane Stress Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
10.11 Allman’s Triangular Plane Stress Element . . . . . . . . . . . . . . . . . . 218
10.12 Computer Program for the Solution of Plane Stress
Problem Using Isoparametric Element . . . . . . . . . . . . . . . . . . . . . 222
References and Suggested Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
11 Finite Element Analysis of Plate Bending Problems . . . . . . . . . . . . . . . 235
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
11.2 Beam Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
11.2.1 Displacement Function . . . . . . . . . . . . . . . . . . . . . . . . . . 236
11.2.2 Displacement Function in Terms of Nodal
Displacements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
11.2.3 Strain (Curvature)–Nodal Parameter
Relationship . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
11.2.4 Stress (Moment)–Strain (Curvature)
Relationship . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
11.2.5 Derivation of the Element Stiffness Matrix . . . . . . . . . 239
11.2.6 Determination of Equivalent Loading
on the Beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
11.3 Rectangular Plate Bending Element . . . . . . . . . . . . . . . . . . . . . . . . 241
11.3.1 Displacement Function . . . . . . . . . . . . . . . . . . . . . . . . . . 242
11.3.2 Displacement Function Expressed in Terms
of Nodal Displacements . . . . . . . . . . . . . . . . . . . . . . . . . 244
11.3.3 Strain–Nodal Parameter Relationship . . . . . . . . . . . . . . 245
xvi Contents

11.3.4
Stress (Moment)–Strain (Curvature)
Relationship . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
11.3.5 Derivation of the Element Stiffness Matrix . . . . . . . . . 247
11.4 Parallelogram Element of Plate Bending . . . . . . . . . . . . . . . . . . . . 249
11.4.1 Displacement Function . . . . . . . . . . . . . . . . . . . . . . . . . . 251
11.4.2 Displacement Function in Terms of Nodal
Displacements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
11.4.3 Curvature–Nodal Parameter Relationship . . . . . . . . . . 252
11.4.4 Moment–Curvature Relationship . . . . . . . . . . . . . . . . . . 253
11.4.5 Element Stiffness Matrix . . . . . . . . . . . . . . . . . . . . . . . . 254
11.5 Hermitian Polynomial Interpolation . . . . . . . . . . . . . . . . . . . . . . . . 255
11.6 A Conforming Plate Bending Element . . . . . . . . . . . . . . . . . . . . . . 256
11.7 Isoparametric Plate Bending Element . . . . . . . . . . . . . . . . . . . . . . 257
11.7.1 Displacement Function . . . . . . . . . . . . . . . . . . . . . . . . . . 257
11.7.2 Strain–Nodal Displacement Relationship . . . . . . . . . . . 259
11.7.3 Stress–Strain Relationship . . . . . . . . . . . . . . . . . . . . . . . 260
11.7.4 Element Stiffness Matrix . . . . . . . . . . . . . . . . . . . . . . . . 260
11.7.5 Reduced Integration Technique . . . . . . . . . . . . . . . . . . . 261
11.8 Smoothed Stresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
11.9 Triangular Plate Bending Elements . . . . . . . . . . . . . . . . . . . . . . . . 263
11.10 DKT Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
11.10.1 Constraint Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
11.10.2 Transformation Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . 266
11.10.3 Element Stiffness Matrix . . . . . . . . . . . . . . . . . . . . . . . . 269
11.11 The Patch Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
11.11.1 The Patch Test for the Plane Stress Element . . . . . . . . 273
11.11.2 The Patch Test for Plate Bending Elements . . . . . . . . . 273
11.12 Horizontally Curved Isoparametric Beam . . . . . . . . . . . . . . . . . . . 276
11.12.1 Displacement Function in Terms of Nodal
Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
11.12.2 Stress–Strain Relations . . . . . . . . . . . . . . . . . . . . . . . . . . 278
11.12.3 Strain–Displacement Relationship . . . . . . . . . . . . . . . . 279
11.12.4 Element Stiffness Matrix . . . . . . . . . . . . . . . . . . . . . . . . 280
11.13 Nonuniform Straight Beam Element . . . . . . . . . . . . . . . . . . . . . . . 280
11.14 Computer Program for Isoparametric Quadratic Bending
Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
References and Suggested Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
12 Finite Element Analysis of Shells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
12.2 Flat Shell Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
12.2.1 Transformation of the Stiffness Matrix
and Assembly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
12.3 Shell of Revolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
12.4 General Shell Finite Element of Triangular Shape . . . . . . . . . . . . 295
Contents xvii

12.4.1 Derivation of the Stiffness Matrix . . . . . . . . . . . . . . . . . 296

12.4.2 Consistent Load Vector . . . . . . . . . . . . . . . . . . . . . . . . . . 302
12.4.3 Condensation of Stiffness Matrix . . . . . . . . . . . . . . . . . 303
12.5 Isoparametric General Shell Element . . . . . . . . . . . . . . . . . . . . . . . 304
12.5.1 Geometry of the Shell Element . . . . . . . . . . . . . . . . . . . 305
12.5.2 Displacement Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
12.5.3 Strains Inside the Element . . . . . . . . . . . . . . . . . . . . . . . 308
12.5.4 Stress–Strain Relationship . . . . . . . . . . . . . . . . . . . . . . . 310
12.5.5 Stiffness Matrix of the Shell Element . . . . . . . . . . . . . . 311
12.6 Vertically Curved Beam Element . . . . . . . . . . . . . . . . . . . . . . . . . . 312
12.7 Computer Program for the Finite Element Analysis
of Shallow Shells of General Shape Using Triangular
Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
References and Suggested Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
13 Semi-analytical and Spline Finite Strip Method of Analysis
of Plate Bending . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
13.2 Beam Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
13.3 Model of the Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
13.4 The Displacement Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
13.5 Curvature-Nodal Parameter Relationship . . . . . . . . . . . . . . . . . . . 326
13.6 Moment—Curvature Relationship . . . . . . . . . . . . . . . . . . . . . . . . . 327
13.7 Strip Stiffness Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
13.8 Loading Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
13.9 Force Displacement Relationship . . . . . . . . . . . . . . . . . . . . . . . . . . 330
13.10 Spline Finite Strip Method of Analysis of Plate Bending . . . . . . 331
13.10.1 The Spline Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
13.10.2 Displacement Functions . . . . . . . . . . . . . . . . . . . . . . . . . 333
13.10.3 Strain–Displacement Relationship . . . . . . . . . . . . . . . . 334
13.10.4 Stiffness Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
13.10.5 The Loading Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
13.11 Computer Program for the Spline Finite Strip Method
of Analysis of Plates in Bending . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
References and Suggested Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
14 Dynamic and Instability Analyses by the Finite Element
Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
14.2 Dynamic Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
14.2.1 Torsional Vibration of Shafts . . . . . . . . . . . . . . . . . . . . . 343
14.2.2 An Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
14.2.3 Flexural Vibration of Beams . . . . . . . . . . . . . . . . . . . . . 347
14.2.4 In-Plane Vibration of Plates . . . . . . . . . . . . . . . . . . . . . . 348
14.2.5 Flexural Vibration of Plates . . . . . . . . . . . . . . . . . . . . . . 349
14.3 Elastic Instability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
xviii Contents

14.3.1 Column Instability Analysis . . . . . . . . . . . . . . . . . . . . . . 351

14.3.2 Plate Instability Analysis . . . . . . . . . . . . . . . . . . . . . . . . 356
References and Suggested Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360
15 The Finite Difference Method for the Analysis of Beams
and Plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
15.2 Finite Difference Representation of Derivatives . . . . . . . . . . . . . . 362
15.2.1 First Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
15.2.2 Second Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
15.2.3 Third Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
15.2.4 Fourth Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
15.3 Errors in the Finite Difference Expressions . . . . . . . . . . . . . . . . . . 364
15.4 Equivalent Concentrated Load . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
15.5 Boundary Conditions for Beam Bending . . . . . . . . . . . . . . . . . . . . 366
15.5.1 Simple Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
15.5.2 Fixed End . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
15.5.3 Free End . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
15.6 A Statically Determinate Static Problem . . . . . . . . . . . . . . . . . . . . 367
15.7 A Statically Indeterminate Static Problem . . . . . . . . . . . . . . . . . . . 370
15.8 Free Vibration of Beams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
15.9 Buckling of Columns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
15.10 Finite Difference Representation of the Plate Equation . . . . . . . . 375
15.10.1 A Plate Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
References and Suggested Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
16 Adaptive Finite Element Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
16.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
16.2 The Adaptive Finite Element Technique . . . . . . . . . . . . . . . . . . . . 381
16.3 Superconvergent Patch Recovery Technique . . . . . . . . . . . . . . . . . 382
16.4 Example of Verification of SPR . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
16.5 Error Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
16.6 ZZ Error Estimator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
16.7 ZZ—Refinement Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
16.8 Adaptive Mesh Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392
16.8.1 Mesh Generation Based on Mapping . . . . . . . . . . . . . . 394
16.8.2 Delaunay Triangulation Method . . . . . . . . . . . . . . . . . . 394
16.8.3 Domain Decomposition Method (Quadtree) . . . . . . . . 394
16.8.4 Advancing Front Technique . . . . . . . . . . . . . . . . . . . . . . 395
References and Suggested Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
17 Geometrical Nonlinear Finite Element Analysis . . . . . . . . . . . . . . . . . . 405
17.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
17.2 Nonlinear Equation Solving Procedures . . . . . . . . . . . . . . . . . . . . 405
17.2.1 Direct Iteration Method . . . . . . . . . . . . . . . . . . . . . . . . . . 406
17.2.2 Newton–Raphson Method . . . . . . . . . . . . . . . . . . . . . . . 407
Contents xix

17.2.3 Modified Newton–Raphson Method . . . . . . . . . . . . . . . 408

17.2.4 Incremental Techniques . . . . . . . . . . . . . . . . . . . . . . . . . 409
17.3 Formulation of the Geometric Nonlinear Problem . . . . . . . . . . . . 410
17.3.1 Equilibrium Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 411
17.3.2 Incremental Equilibrium Equation . . . . . . . . . . . . . . . . 412
17.4 Large Deflection Analysis of Plates in b-notation . . . . . . . . . . . . 413
17.5 Large Deflection Analysis of Plates in n-notation . . . . . . . . . . . . 418
17.6 Example of a Pin-Jointed bar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
17.7 Computer Program for Geometrically Nonlinear Analysis
of Plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
References and Suggested Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
18 Finite Element Method of Analysis of Stiffened Plates . . . . . . . . . . . . 431
18.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
18.2 Modeling the Plate and the Stiffener . . . . . . . . . . . . . . . . . . . . . . . 431
18.3 Rectangular Stiffened Plate Bending Element . . . . . . . . . . . . . . . 432
18.3.1 Stiffness Matrix of the Stiffener Element . . . . . . . . . . . 433
18.4 Isoparametric Stiffened Plate Bending Element . . . . . . . . . . . . . . 436
18.4.1 Stiffness Matrix of Arbitrarily-Oriented
Eccentric Stiffener . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436
References and Suggested Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
19 Selected Topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443
19.1 Rayleigh–Ritz Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443
19.2 An Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
19.3 Rayleigh–Ritz Finite Element Method . . . . . . . . . . . . . . . . . . . . . . 445
19.4 Weighted Residual Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 447
19.5 Galerkin Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
19.5.1 An Example of Galerkin Method . . . . . . . . . . . . . . . . . . 450
19.6 Galerkin Finite Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
19.7 Torsional Stiffness of Prismatic Beam Element . . . . . . . . . . . . . . 455
19.8 Torsion of Noncircular Sections . . . . . . . . . . . . . . . . . . . . . . . . . . . 458
19.9 Axi-symmetrical Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
19.10 Three-Dimensional Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463
19.10.1 Linear Element (8 Nodes) (Fig. 19.8a) . . . . . . . . . . . . . 463
19.10.2 Quadratic Element (20 Nodes) (Fig. 19.8b) . . . . . . . . . 464
19.10.3 Cubic Element (32 Nodes) (Fig. 19.8c) . . . . . . . . . . . . 464
19.10.4 A16 Noded Solid (Fig. 19.9a) . . . . . . . . . . . . . . . . . . . . 465
19.10.5 A24 Noded Solid (Fig. 19.9b) . . . . . . . . . . . . . . . . . . . . 466
References and Suggested Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 466

Appendix A: Fixed-End Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467

Appendix B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471
About the Authors

Prof. Madhujit Mukhopadhyay is an Ex-Dean (Faculty and Planning) and

Professor of Structural Engineering at Indian Institute of Technology, Kharagpur.
He obtained the B.E. degree from the University of Calcutta and the Ph.D. and the
D.Sc. degrees from IIT Kharagpur. His field of research interest lies in the application
of various numerical methods for the analysis of plates and shells, bare or stiffened,
isotropic or orthotropic. He has received awards for his research. A widely travelled
person, Prof. Mukhopadhyay published a large number of papers in international
journals in Aerospace, Civil, Mechanical and Ocean Engineering and is the author
of three textbooks.

Dr. Abdul Hamid Sheikh is an Associate Professor at the Indian Institute of Tech-
nology, Kharagpur. He obtained the B.E. and M.E. degrees from the University of
Calcutta and the Ph.D. degree from IIT Kharagpur. His research interest lies in finite
element and other numerical techniques and their applications to different types
of analyses for stiffened plates and shells, composite and smart laminates and allied
problems. He has published a large number of papers in leading international journals.

xxi
Chapter 1
Basic Concepts of Structural Analysis

1.1 Types of Structures

A simply supported beam may be treated as a one-dimensional structure, and neces-

sary quantities such as strength, rigidity or stability can be determined from the
knowledge of strength of materials. The same beam may also be analysed as a two-
dimensional or three-dimensional problem and its solution based on the theory of
elasticity is more rigorous and refined. Thus, the same structure can be idealized in
different ways and the result of the analysis depends to a great extent on the idealiza-
tion of the structure. For the convenience of classification, structures are organized
into the following three types:
1. Framed structures such as a truss, a frame, a beam or a grid are treated as a
combination of linear or uniaxial members. All the members are considered long
enough in comparison to their cross-sectional dimensions. The analysis of these
structures follows a particular pattern. Some structures of this category are shown
in Fig. 1.1.
2. Next category of structures includes plates, shells and planar elements which are
referred to as two-dimensional structures (Fig. 1.2). Just as the members of the
framed structures are represented by their lengths, plates and shells are repre-
sented by their surfaces. The analysis of these structures requires the knowledge
of the theory of elasticity.
3. Soil mass, rock foundations, complex machine parts, etc. are three-dimensional
bodies. For any detailed analysis, their actual behaviour in space is to be taken
into account.

1.2 Objective of Structural Analysis

The analysis of all the above structures involves the determination of

(a) displacements at salient points of the structure;
© The Author(s) 2022 1
M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_1
2 1 Basic Concepts of Structural Analysis

(a) Continuous beam (b) Grid (c) Frame

Fig. 1.1 Framed structures

(a) Web frame (b) Shell structure (c) Floor slab

of a tanker

Fig. 1.2 Two-dimensional structures

(b) internal forces (‘force’ is used here in general term. By force, we may mean a
translatory force, a bending moment, a twisting moment, etc.) in the members
of the structure on the basis of
(i) the given shape of the structure,
(ii) the sizes of the members,
(iii) the material properties of the members and
(iv) the applied loads.
In statically determinate structures, all the reactions can be determined from
the equations of equilibrium and then all the necessary quantities such as bending
moment, deflection, etc. can be obtained from the knowledge of elementary
mechanics of materials. Structures which are termed as statically indeterminate
contain more unknown reactions and/or internal forces than can be solved from
equations of statics. Solution of statically indeterminate structure therefore becomes
more complex and additional equations over and above those obtained from statics
are to be obtained on the basis of the deformation conditions of the structure.
1.4 Loads 3

The analysis of statically indeterminate structures finally reduces to the solution of

a series of linear algebraic equations. The number of such equations will essentially
depend on the method of analysis employed. The book is devoted primarily to the
statically indeterminate framed structures, plates and shells.
The classical method of analysis of structures has its own limitations in that only
simple structures can be dealt with. Modern structures are becoming complex in
shape, and uncertainties associated with the design of structures are getting reduced.
An awareness has grown to analyse the structure in truer perspective. In so doing,
many of the simplifying assumptions have to be done away with. With the advent
of electronic digital computers, mammoth calculations no longer pose any problem.
Its ultimate result is the development of matrix methods of structural analysis for
one-dimensional structure and the finite element method1 for plates, shells and more
complicated three-dimensional problems. These recent computer-oriented methods
are discussed in the following chapters.

1.3 Materials and Basic Assumptions

The analysis presented throughout unless otherwise specified is based on the

following assumptions:
1. The material is isotropic and homogeneous, i.e. the property of the material is
not changing from point to point.
2. The material with which the structure is made of is assumed to be linearly elastic,
i.e. it obeys Hooke’s law. For most cases, permissible stresses under working
loads are sufficiently low so that the linear behaviour of the material may be
justified.
3. Loads applied to the structure will not induce any effect of geometry changes.
4. Buckling aspects of the members in the general statical and dynamical analysis
are neglected.

1.4 Loads

A structure has to sustain the load coming on it. Before an analysis is made, load
coming on the structure must be ascertained. Some of the important types of loadings
are discussed here as follows:
1. Dead load—it is the self-weight of the structure.

1Finite element method can be applied to one-dimensional structure as well. Whereas matrix
methods of structural analysis have been developed based on intuitive reasoning, the finite element
method follows a more formal mathematical procedure.
4 1 Basic Concepts of Structural Analysis

2. Live load—it is the useful load carried by the structure. Furniture and human
weight in buildings, cranes in factory sheds and vehicles in bridges are examples
of live load. This load is determined from code specifications.
3. Impact load—it is due to sudden application of the load on a structure. The
jumping of a vehicle due to undulations on a surface causes impact which results
in stresses much more than the value it would attain had the load been applied
gradually.
4. Wind, earthquake and blast loads—these loads cause accelerations of the mass of
the structure and so must be dynamically considered. In fact that is the trend for
‘all highrise, tall and slender structures’. For conventional structures, these loads
are reduced to static loads, the values of which could be obtained by consulting
the codes of practice or from recorded data.
5. Self-straining load—it may occur due to the settlement of support, changes of
temperature, creep of the material of the structure and due to prestressing effects.
For structures floating on the sea, there will be loads on the structure due to buoy-
ancy and dynamic effects of the waves. If the floating structure is mobile, there will
be forces due to its own motion. For fixed ocean structures, there will be horizontal
forces due to water current. Loads to be encountered therefore depend on the type
of the structure.

1.5 General Methods of Analysis

Any method of structural analysis must satisfy the following two conditions:
1. Equilibrium and
2. Compatibility.

1.5.1 Equilibrium Conditions

Not only the entire structure, but also any part of the structure considered as a free
body should be in equilibrium under the action of the forces acting on it. The equa-
tions of equilibrium are obtained from statics. For a planar structure, there are three
equations of equilibrium. For three-dimensional structures, number of equations of
equilibrium is six. In the analysis to be followed, the equilibrium conditions in terms
of forces acting at the joints or nodes of the structure are formulated.
Similar consideration can be made for vibrating structures. The difference to be
made in this case is that in addition to all the static forces, the inertia forces and
the damping forces (if present) are to be taken into account using D’ Alembert’s
principle.
1.6 Force–Displacement Relationship 5

1.5.2 Compatibility Conditions

Deformation of the structure obtained should be compatible with the continuity of the
structure and the support conditions. Compatibility conditions refer to the continuity
of displacements throughout the structure, i.e. different parts of the structure must
fit together without discontinuity at all stages of loading. To illustrate this, a few
examples may be mentioned. There can be no rotation of the axis of the member
at a fixed support. At the unyielding support of a continuous beam, there can be no
deflection, and the rotation is same at both sides of the support. Well-known three
moment theorem is based on this compatibility condition and such similar equations
are frequently encountered in structural analysis.
Displacements at a joint must be identical to that of the individual members
meeting at that joint. For a plane problem, there can be three possible placements at
a joint-two translations and one rotation. For a three-dimensional problem, number
of possible displacements are six—three translations and three rotations.

1.6 Force–Displacement Relationship

When subjected to a force, a structural member undergoes deformation. A relation-

ship exists between the applied force and the displacement of the structure. This is
usually done in one of the following two ways.
A uniform cantilever beam of Fig. 1.3 is subjected to a load P at the free end and
the corresponding vertical deflection is X. The relationship between P and X may be
given as

X = FP (1.1)

where F is the deflection produced by a unit value of P at the point of deflection of

P and is called the flexibility of the beam.
Another way of relating P and X can be done as follows:

P = KX (1.2)

where K is the force corresponding to P required to produce unit displacement at

the point of application of P and is called the stiffness of the beam. For the uniform
beam of Fig. 1.3, F and K assume the following values:

L3
)
F= 3E I (1.3)
K = 3E I
L3

where L is the length of the beam and EI is the flexural rigidity. Equation (1.3) reveals
that the flexibility F and the stiffness K are inverse to one another.
6 1 Basic Concepts of Structural Analysis

Fig. 1.3 A cantilever beam P

(a)
l

(b)
K

l
(c)

The above concept has been extended to the solution of more general problems,
which involve the relation between the forces acting at a number of points in the
structure and the corresponding displacements.

1.7 Statical Indeterminacy

A structure remains in equilibrium under the action of external forces. A structure

whether considered as a whole or members or joints considered as freebody should
be in equilibrium under the action of external forces. The analysis of the structure
comprises the determination of reactions at supports and the internal forces (such as
axial force in the member of a truss). The types of structures for which the equations
of statics alone are not sufficient for the complete solution of the problem are referred
to as statically indeterminate structures. For such structures, the number of unknown
forces are more than the number of equations available from statics and the number
of forces in excess of static equations are termed as degrees of statical indeterminacy
or degrees of redundancy of the structure. Where the indeterminacy is due to the
reactions, it is stated as externally indeterminate and where it is due to internal forces
it is referred to as internally indeterminate. The redundancy of a structure may either
be external or internal or both.
1.7 Statical Indeterminacy 7

1.7.1 Plane Structure

For a truss-like structure having hinges at the joints (a joint is a point where two
or more members intersect), following relationship holds good if the structure is
statically determinate:

2j = m + 3 (1.4)

where j is the number of joints including supports and m is the number of members
of the truss. Equation (1.4) can be derived easily. For a statically determinate truss,
number of unknown reactions to be determined from the equilibrium conditions of
the overall structure is three. Other unknowns in the problem are the member forces
which are m in number. Total number of equations at each joint is 2. By equating the
number of unknowns to the number of equations of statics, Eq. (1.4) results.
For a statically indeterminate plane truss, the degree of indeterminacy is

i = (m + r ) − 2 j (1.5)

where r is the number of unknown reactions. For the truss of Fig. 1.4, m = 15, r = 4
and j = 8. Therefore, the degree of statical indeterminacy is 3. It may be noted that
the external indeterminacy is 1 and the internal indeterminacy is 2 for the truss.
Following similar reasoning it can be shown that a rigid plane frame is statically
determinate if

3 j = 3m + r − α (1.6)

where α is the number of internal releases. If at any particular point in the structure,
if the internal force remains zero for all possible types of loading, it is then said that
a release has been introduced at that point in the structure. Introduction of a hinge
at a point of a bending member means that bending moment at that point is zero
irrespective of the load applied.
For the beam of Fig. 1.5a, m = 2, j = 3, r = 4 and a = 1 (note that the hinge inside
a member is not a joint). For this beam 3 j = 9 and 3m + r − α = 9. Therefore,
Eq. (1.6) is satisfied and the beam is statically determinate. The frame of Fig. 1.5b

Fig. 1.4 Statically

indeterminate plane truss
8 1 Basic Concepts of Structural Analysis

is also statically determinate where 3 j = 9, m = 3, r = 4 and α = 1 and Eq. (1.6)

is satisfied.
The degree of indeterminacy of a rigid jointed plane structure is given by

i = 3m + r − 3 j − α (1.7)

For the frame of Fig. 1.6, j = 9, m = 10, r = 5 and α = 0. Therefore, the

degree of indeterminacy of the frame is 8. For the grid of Fig. 1.7, the reactions at
both supports are two couples and a vertical force. In this case, m = 2, j = 3, r = 6
and α = 0. Therefore, the degree of indeterminacy of grid is 3.

B C

(a) (b)

Fig. 1.5 Statically determinate structures

Fig. 1.6 A frame

1.7 Statical Indeterminacy 9

Fig. 1.7 A grid

1.7.2 Space Structures

The degree of statical indeterminacy of a space truss is given by

i = m +r − 3j (1.8)

The degree of indeterminacy of a space frame is given by

i = 6m + r − 6 j − α (1.9)

For the space truss of Fig. 1.8, m = 4, j = 5, r = 12 which on substitution into

Eq. (1.8) yields i = 1 which means that the given space truss is statically indeterminate
to first degree. For the space frame of Fig. 1.9, m = 8, j = 8 and r = 24 which on
substitution into Eq. (1.9) gives i = 24.

Fig. 1.8 A space truss

10 1 Basic Concepts of Structural Analysis

Fig. 1.9 A space frame

1.8 Kinematic Indeterminacy

The joints of a structure subjected to load undergo rotations and translations. At

each joint, possible joint displacements can be labelled such that they can be treated
as independent, i.e. each displacement can be arbitrarily changed without affecting
other displacements at that joint. The number of independent joint displacements of
a structure is called the degree of kinematic indeterminacy or the number of degrees
of freedom.
The propped cantilever AB of Fig. 1.10 has two joints A and B. Joint A is fixed;
so there cannot be any displacement there. The vertical movement of joint B is
restrained, but the horizontal translation and rotation of joint B are permitted. Hence,
the degrees of freedom of the beam is 2. In the analysis of such beams, the axial
deformation is neglected which means that the length of the beam remains same and
the displacement corresponding to axial movement need to be neglected. In that case,
the degree of kinematic indeterminacy becomes one.
Each of the three joints A, B and C of the rigid frame of Fig. 1.11 possesses three
degrees of freedom—two translations and one rotation. Joint E has one degree of
freedom—that of rotation. Total degrees of freedom then are 10. However, if the
axial deformation of the members is neglected, the horizontal translation at joints A,
B and C labelled as 2, 5 and 8 in the figure would be equal. The independent joint
displacements then would be the rotations at joints A, B, C and E and one horizontal
translation. Total degrees of freedom of the frame then becomes 5.

Fig. 1.10 A propped A B

cantilever
1.9 Two Approaches of Structural Analysis 11

Fig. 1.11 A rigid frame

3 6 9

1 5 B 4 8 7
2
A C

D E F
10

1.9 Two Approaches of Structural Analysis

All methods of structural analysis for statically indeterminate structures fall under
any of the following two categories: the force or the flexibility method and the
displacement or the stiffness method.
In the force method, the structure is first made statically determinate by releasing
certain redundants. The deformation of the released structure is thus inconsistent. For
example, if a roller support is removed, the force at the support is eliminated and at
the point of support, continuity is affected. To restore continuity in the displacement,
additional forces corresponding to the ‘released’ redundants are applied. This results
in a set of simultaneous equations, the solution of which gives the values of the
unknown forces. In this method, the forces are the unknowns.
In the displacement method, the structure is first made kinematically determi-
nate, i.e. inhibiting all possible joint displacements. In the restrained structure, joint
equilibrium is violated. To restore the equilibrium, displacements at the restrained
joints are allowed which results in a set of simultaneous equations of unknown
displacements. In this method, the displacements are the unknowns.
There are thus two approaches of analysis of structure depending on the choice
of modification of the original structure. The philosophy of both methods is again
the same and both methods finally result in a series of equations. In the subse-
quent chapters, these two methods are discussed in more detail for the case of linear
structures.
Exercise 1
Find out the degrees of statical and kinematic indeterminacies for the structures
shown in Fig. 1.1.
12 1 Basic Concepts of Structural Analysis

(a) (b)

(c) (d)

(All members in
horizontal plane)

Rigid joint

(e) (f)

(g) (h)

z
(I) (j)

Fig. 1.12 Various structures

Chapter 2
Energy Principles

2.1 Introduction

The variational or energy methods are of primary importance in the finite element
analysis [1–5]. These methods have been in use in the structural analysis for a consid-
erable length of time. But the development of the finite element method has brought
into being some of the sophisticated forms of these methods along with their scope
and limitations. In this chapter, important energy theorems which have been applied
in the subsequent chapters will be discussed [1, 5, 6].

2.2 Principle of Virtual Work

The principle of virtual work is a very useful approach to the solution of a variety of
problems of structural mechanics. This principle forms the basis of the subsequent
methods described in this chapter.
A body or a structure is considered to be in equilibrium under the action of body
forces {PB } whose components along three reference axes x, y and z are PBx , PBy and
PBz and surface forces {PS } whose components along x-, y- and z-axes are PSx , PSy
and PSz respectively. Entire structure is assumed to undergo fictitious displacement
(termed as virtual displacement) such that prescribed boundary conditions of the
structure are not violated. Components of the virtual displacement at each point are
δu, δv and δw. Total virtual work in the system consists of the work done by the
different components—the external force as well as the internal force.
The external virtual work δW E is the work done by the real loads acting on the
structure due to the virtual displacement. These loads include those distributed both
over the entire surface and the entire volume of the structure. Therefore, the virtual
work done by the external force is

© The Author(s) 2022 13

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_2
14 2 Energy Principles
⎧ ⎫ ⎧ ⎫
( ⎨ PSx ⎬ ( ⎨ PBx ⎬
δW E = { δu δv δw } PSy ds + { δu δv δw } PBy dv (2.1)
⎩ ⎭ ⎩ ⎭
s PSz v PBz

In Eq. (2.1), the integration is carried over the entire surface in the first term and
over the entire volume in the second term. Equation (2.1) can be written as
( (
δW E = δ{ f }T {PS } ds + δ{ f }T {PB } dv (2.2)
s v

⎧ ⎫
⎨u⎬
where { f } = v .
⎩ ⎭
w
For a general stress condition, there are six components of stresses
σx , σ y , σz , τx y , τ yz and τzx . The six components of strain in the virtual displace-
ment fields are δεx , δε y , δεz , δγx y , δγ yz and δγzx . The virtual internal work is
therefore
⎧ ⎫
⎪
⎪ σx ⎪⎪
⎪σ ⎪
⎪ ⎪
⎪
⎪ ⎪
⎪
( ⎪
⎨
y ⎪
⎬
σz
δU = { δεx δε y δεz δγx y δγ yz δγzx } dv
⎪
⎪ τ ⎪
v ⎪ xy ⎪
⎪ ⎪
⎪
⎪ ⎪
⎪ τ yz ⎪
⎪
⎩ ⎪
⎭
τzx
(
or δU = δ{ε}T {σ } dv (2.3)
v

The principle of virtual work states that the work done by the external forces due
to the virtual displacement for a structure in equilibrium is equal to the work done
by the internal forces for the virtual internal displacement. Mathematically,

δW E = δU
or − δU + δW E = 0 (2.4)

It may be noted that in the derivation of Eq. (2.4), the stress–strain relationship
has not been assumed. As such the principle of virtual work holds good for linearly
elastic as well as non-linear materials.
2.4 Principle of Minimum Potential Energy 15

2.3 Principle of Complementary Virtual Work

Unlike the previous section, the work done by a system of virtual forces during
actual deformation of the structure will now be investigated. The work done by the
virtual forces during actual displacements is referred to as complementary virtual
work. Imaginary system of body and surface forces is introduced into the actual
displacement field. The external complementary virtual work done is therefore given
by
( (
δW E∗ = { f }T δ{PS } ds + { f }T δ{PB } dv (2.5)
s v

A virtual state of stress δ{σ } results in the actual strain field {ε} which results in
the internal virtual work. It is expressed as
(
∗
δU = {ε}T δ{σ } dv (2.6)
v

The principle of complementary virtual work states that the complementary virtual
work done by the external virtual forces under the actual deformation of a structure
in equilibrium is equal to the complementary work done by the virtual stresses due
to internal strains. Mathematically,

− δU ∗ + δW E∗ = 0 (2.7)

2.4 Principle of Minimum Potential Energy

Instead of conceiving the necessary forces and displacements in terms of virtual

quantities, both the principle of virtual work and the principle of complementary
virtual work may be interpreted in terms of mathematical variations.
The displacement field for a structure in equilibrium be { f }. Let us consider
arbitrary displacements which are consistent with the constraints imposed on the
body. These arbitrary displacements differ from the actual displacements by δ{ f }. At
the constraints, the variation δ{ f } must vanish which introduces a concept equivalent
to that of virtual displacement introduced in the earlier section.
The external virtual work may be interpreted in terms of potential energy of the
system and may be considered as the work done during a variation δ{ f } in the
displacement field. The potential of the external forces is given by

δW E = −δVE (2.8)
16 2 Energy Principles

Substituting δW E from Eqs. (2.8) into (2.4) yields

δU + δVE = 0
or, δ(U + VE ) = 0
or, δ∅ = 0 (2.9)

where ∅ = U + VE is defined as the total potential energy of the system.

The principle of minimum potential energy states that among all displacement
fields that satisfy the prescribed constrained condition of the structure, those that
satisfy equilibrium conditions make the total potential energy a minimum.
Thus,

δ 2 ∅ = δ 2 (U + VE ) > 0 (2.10)

2.5 Principle of Minimum Complementary Energy

Similarly, the external complementary virtual work δW E∗ may be conceived as the

work done by a variation of external forces. A complementary potential function VE∗
may be introduced as

− δVE∗ = δW E∗ (2.11)

Substituting δW E∗ from Eqs. (2.11) into (2.7) yields

δU ∗ + δVE∗ = 0
or, δ (U ∗ + VE∗ ) = 0
or, δ∅∗ = 0 (2.12)

The principle of minimum complementary energy states that among all states
of stress that satisfy equilibrium conditions, those that satisfy the prescribed
displacement boundary conditions make the total complementary energy a minimum.
It may be noted that in applying the principle of minimum complementary energy,
it is the force which is to be varied, while in the principle of minimum potential energy,
the displacement is to be varied.

2.6 Castigliano’s Theorems

Castigliano’s theorems can be derived from the principle of minimum potential

energy and the minimum complementary energy.
2.6 Castigliano’s Theorems 17

The principle of minimum potential energy states that

δ∅ = δ (U + VE ) = 0 (2.13)

If a structure is in equilibrium under a set of forces P1 , P2 , P3 , . . . , Pn inducing

displacements X 1 , X 2 , X 3 , . . . , X n , then by the application of minimum potential
energy yields

∂U ∂U ∂U
δ X1 + δ X2 + · · · + − P1 δ X 1 − P2 δ X 2 · · · − Pn δ X n = 0
∂ X1 ∂ X2 ∂ Xn
(2.14)

Rearranging the terms gives

( ( ( ( ( (
∂U ∂U ∂U
− P1 ∂ X 1 + − P2 ∂ X 2 + · · · + − Pn ∂ X n = 0
∂ X1 ∂ X2 ∂ Xn
(2.15)

Since the variations are imaginary, the terms within the parentheses must vanish,
therefore,
∂U
⎫
= P1 ⎪
⎪
∂ X1 ⎪
∂U
∂ X2
= P2 ⎪
⎪
⎪
⎪
.. ⎬
. (2.16)
.. ⎪
⎪
⎪
⎪
. ⎪
⎪
⎪
∂U
∂ Xn
= P ⎭ n

Equation (2.16) is nothing but the mathematical form of what is known as

Castigliano’s first theorem. It states that if the strain energy U of an elastic structure
is expressed in terms of generalized displacements, then the first partial derivative of
U with respect to one of the generalized displacements gives the generalized force
acting at that point.
Similarly, the principle of minimum complementary energy states that

δ∅∗ = δ (U ∗ + VE∗ ) = 0 (2.17)

For a structure in equilibrium under a set of forces P1 , P2 , . . . , Pn inducing

displacements X 1 , X 2 , . . . , X n then by the application of the principle of minimum
complementary energy yields

∂U ∗ ∂U ∗ ∂U ∗
δ P1 + δ P2 + · · · + δ Pn − X 1 δ P1 − X 2 δ P2 · · · − X n δ Pn = 0
∂ P1 ∂ P2 ∂ Pn
( ∗ ( ( ∗ ( ( (
∂U ∂U ∂U ∗
or,
∂ P1
− X 1 δ P1 +
∂ P2
− X 2 δ P2 +
∂ Pn
− X n δ Pn = 0 (2.18)
18 2 Energy Principles

Variations of P1 , P2 , . . . , Pn are arbitrary and hence must vanish, therefore,

∂U ∗
⎫
= X1 ⎪
⎪
∂ P1 ⎪
∂U ∗
∂ P2
= X2 ⎪
⎪
⎪
⎪
.. ⎬
. (2.19)
.. ⎪
⎪
⎪
⎪
. ⎪
⎪
⎪
∂U ∗
∂ Pn
= X ⎭ n

Equation (2.19) is known as Engesser’s theorem and is valid for both linear and
non-linear elastic structures.
However, for linear elastic structure

U = U∗ (2.20)

So, Eq. (2.20) for elastic structure can be written as

∂U
⎫
= X1 ⎪
⎪
∂ P1 ⎪
∂U
∂ P2
= X2 ⎪
⎪
⎪
⎪
.. ⎬
. (2.21)
.. ⎪
⎪
⎪
⎪
. ⎪
⎪
⎪
∂U
∂ Pn
= X ⎭
n

Equation (2.21) is the mathematical statement of Castigliano’s second theorem.

It states that if the strain energy U of a linearly elastic structure is expressed in terms
of generalized forces, then the first partial derivative of U with respect to one of the
generalized forces gives the corresponding displacements at those points.

2.7 Determination of Displacements

If any arbitrary section of a one-dimensional structure of length L is subjected to the

simultaneous action of axial force N, bending moment M, shear force V and twisting
moment T, the total strain energy stored in the system is

(L (L (L (L
N2 dx M2 dx V2 T 2 dx
U = + + η dx +
2E A 2E I 2G A 2G J (2.22)
0 0 0 0
References and Suggested Readings 19

where A is the cross-sectional area of the member, I is the second moment of area
of the section, J is the polar moment of inertia of the section, E and G are Young’s
modulus and shear modulus of elasticity of the member and η is a factor dependent
on the cross-sectional properties of the member.
In order to determine a displacement X i of the structure, Castigliano’s second
theorem yields, i.e. from Eqs. (2.21) and (2.22)

(L (L (L (L
∂U N ∂N M ∂M V ∂V T ∂T
Xi =
∂ Pi
=
E A ∂ Pi
dx +
E I ∂ Pi
dx + η
G A ∂ Pi
dx +
G J ∂ Pi
dx (2.23)
0 0 0 0

The quantity ∂ N /∂ Pi is the value of the axial force at the arbitrary section due
to a force corresponding to a unit value of Pi . Similarly, δ M/δ Pi is the value of the
bending moment at that section due to a force corresponding to a unit value of Pi
and so on.
For determining the value of Pi , Eq. (2.23) indicates that two quantities N and
∂ N /∂ Pi , M and ∂ M/∂ Pi and so on are to be multiplied over the entire length of
the structure. This can be achieved by integrating the products. However, it becomes
simple in many cases to determine displacements by multiplying the area diagrams
given in Table 3.1 (next chapter).
No separate problem has been solved here either to demonstrate the application
of energy principles or to calculate the displacement. These will be indicated in
subsequent chapters.

References and Suggested Readings

1. J.H. Argyris, S. Kesley, Energy Theorems and Structural Analysis (Butterworths, London, 1960)
2. E.B. Becker, G.F. Carey, J.T. Oden, Finite Element Method, vol. I–V (Prentice Hall, 1981)
3. B.A. Finlayson, The Method of Weighted Residuals and Variational Principles (Academic Press,
1972)
4. R. Glowinsky, E.Y. Rodin, O.C. Zienkiewicz, Energy Methods in Finite Element Analysis (Wiley,
1979)
5. O.C. Zienkiewicz, R.L. Taylor, The Finite Element Method, vol. I, 4th edn (McGraw-Hill, 1989)
6. I.H. Shames, C.L. Dym, Energy and Finite Element Methods in Structural Mechanics (McGraw-
Hill, 1985)
Chapter 3
Introduction to the Flexibility
and Stiffness Matrix Methods

3.1 Introduction

It has been mentioned in Chap. 1 that there are two basic methods of structural
analysis—the flexibility method and the stiffness method. The purpose of this chapter
is to introduce these methods.

3.2 The Flexibility Matrix Method

The steps followed in the flexibility matrix method are described first. Then the
method is illustrated with a number of examples.
First, the degree of statical indeterminacy of the structure is determined. The
structure is then made statically determinate by removing the internal and/or external
forces whose total number is equal to the degree of statical indeterminacy. This
structure is referred to as the primary structure. Each force removed to convert the
structure into a statically determinate one is called a released force or a redundant
force. For example, if the end moment is released at a fixed support, then that support
is converted into a simple support. The analyst is free to make a suitable choice of
the releases. These released forces are the ones which are to be determined from the
analysis.
Secondly, displacements corresponding to the releases are to be determined for
the external load acting on the released structure. For example, if a roller support
is removed in a continuous beam, the beam will deflect downward at that support
point under the action of vertical loads acting downwards. The determination of
this vertical deflection and likewise displacements at all similar release points is the
second step to be followed.
Thirdly, the values of the displacements in the released structure corresponding
to a unit value of the released force acting in the same direction one at a time at that
point of release are to be determined.

© The Author(s) 2022 21

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_3
22 3 Introduction to the Flexibility and Stiffness Matrix Methods

Fig. 3.1 A continuous beam 2P P

A B C

I I

L/2 L/2 L/2 L/2

In the actual structure there are either no or some specified displacements at

the released points. So the displacements that are occurring at the points where the
forces are released are to be eliminated. For this purpose instead of a unit value of
the released forces, the redundant forces are given some unknown values which are
determined from the final equation after applying the principle of superposition of
displacements. This becomes the fourth step.
Lastly, once the released forces are known, the deformation and other necessary
quantities of the original indeterminate structure can be determined.
The sign convention to be adopted for the force or the flexibility method is first
discussed. For beams, hogging moment is negative and sagging moment is positive
and for frames, tension inside is positive and vice versa. For axial forces, tension is
positive and compression negative.
The method is first explained in details with an example. Then a few more exam-
ples follow. The continuous beam of Fig. 3.1 is to be analysed by the flexibility
method.
Step 1. The beam of Fig. 3.1 is statically indeterminate to the second degree.
So two reactions are to be removed. There are various possible ways of making
the choice. Three possible released structures are shown in Fig. 3.2 and the released
forces are shown by dotted lines and are numbered 1 and 2. It may be noted that other
possibilities also exist for making the structure of Fig. 3.1 statically determinate. In
Fig. 3.2(a), the fixed end A is converted to a hinge thus releasing the end moment
there and then at support B, a hinge is inserted thus removing the bending moment
at B.
Thus, the redundant forces considered in the analysis in this case are the end
moments at A and B. Another possible released structure is the cantilever beam of
Fig. 3.2(b). In this case, the redundant forces are two support reactions at B and C.
For the analysis presented here, the choice of the primary structure corresponding
to Fig. 3.2(a) is made and the decision in this case is arbitrary. The redundant forces
are R1 and R2 labelled as 1 and 2 in Fig. 3.2(a).
Step2. The displacements produced at the released points due to the external loads
for the structure of Fig. 3.2(a) are then determined.
The displacements in this case are X 1 and X 2 of Fig. 3.3(b). The bending moment
diagram of Fig. 3.3(b) due to the external load is shown in Fig. 3.4(a) and is referred
to as m0 diagram. In Fig. 3.4(b), the bending moment diagram for the same beam
has been shown for a unit load corresponding to R1 applied at release point 1. This
diagram is referred to as ml diagram. Similarly, m2 diagram of Fig. 3.4(c) refers to
3.2 The Flexibility Matrix Method 23

Fig. 3.2 Possible released 1 2

structure

A B C
L L

(a)

A B C

(b)

A B C

(c)

the bending moment diagram of the same beam due to a unit load corresponding
to R2 applied at point 2. Considering only the bending part of the strain energy and
applying Eq. (2.21) yields
(
∂U L
M ∂M
X1 = = dx (3.1)
∂R1 0 EI ∂R1

Equation (3.1) suggests that the value of X 1 can be obtained by multiplying the
bending moment diagram due to the applied load which corresponds to m0 diagram
of Fig. 3.4(a) to the bending moment diagram due to a unit load corresponding to R1
which corresponds to m1 diagram of Fig. 3.4(b) divided by EI. From Table 3.1, by
multiplying m0 diagram with m1 diagram, Eq. (3.1) yields

1 KiL
X1 = (1 + β)
6 EI
1 PL L
= (1 + 1/2)
6 2 EI

PL2
=
8EI
Similarly,
( ) ( )
1 1 PL L 1 1 PL L
X2 = 1 + + 1 +
6 2 2 EI 6 2 4 EI
24 3 Introduction to the Flexibility and Stiffness Matrix Methods

Fig. 3.3 The released R1 R2 R2

A B C
structure and the steps
followed in sequence
(a)
2P X2 P
X1

(b)

f11
1 f21

(c)

f12 1 1
f22

(d)

Fig. 3.4 Bending moment

diagrams of the released
structure PL/2
PL/4
x
(a) m0 diagram

1
x
(b) m1 diagram

1
x
(c) m2 diagram
3.2 The Flexibility Matrix Method 25

Table 3.1 Multiplication table for area diagram

3 PL2
=
16 EI
Step 3. The displacements at the release points corresponding to a unit value of
the released forces one at a time are next to be determined. They correspond to f 11
and f 2 l of Fig. 3.3(c) for a unit value of R1 and f 12 and f 22 of Fig. 3.3(d) for a unit
value of R2 .
Looking at Eq. (3.1) it may be noticed that in order to determine f 11 , m1 diagram
is to be multiplied by the same diagram and for f 12 , m1 diagram is to be multiplied by
the m2 diagram over the entire length of the beam. The process is repeated for other
displacements. The necessary multiplication of the area diagrams can be carried out
with the help of Table 3.1.

1 1×1×L L
f11 = =
3 EI 3EI
1 1×1×L L
f12 = f21 = =
6 EI 6EI
1 1×1×L 2L
f22 = 2 × =
3 EI 3EI
It may be noted that the quantities f 11 , f 12 , f 21 , and f 22 are independent of the
loading and as such are properties of the structure.
26 3 Introduction to the Flexibility and Stiffness Matrix Methods

Step 4. If R1 and R2 are the redundant forces, the displacement at A due to R1 and
R2 is given by

XR1 = f11 R1 + f12 R2

Similarly,

XR2 = f21 R1 + f22 Rl2

Now, as the total slope at A and relative slope at B are equal to zero, following
equations result by applying the principle of superposition of displacements
)
f11 R1 + f12 R2 + X1 = 0
(3.2)
f21 R1 + f22 R2 + X2 = 0

Equation 3.2 written in matrix form becomes

[ ]( ) ( ) ( )
f11 f12 R1 X1 0
+ = (3.3)
f21 f22 R2 X2 0

which in compact form is

[F] {R} + {X } = {0 } (3.4)

In order to determine {R}, Eq. (3.4) is rewritten as follows:

{R} = − [F] − 1 {X } (3.5)

The matrix [F] is known as the flexibility matrix and its elements f 11 , f 12 , etc. are
called flexibility coefficients or influence coefficients. f ij is defined as the displace-
ment at point i in the direction of unknown ‘force’ at i due to a unit value of the
‘force’ corresponding to unknown force at j and applied at j. It may be noted that the
flexibility matrix is symmetrical. This follows from Maxwell’s reciprocal theorem
and also from a careful study of Eq. (3.1). Substituting f 11 , f 12 , etc. and X 1 and X 2
for this problem of Fig. 3.1, Eq. (3.3) becomes
[ ]( ) ( )
L 21 R1 PL2 2
= −
EI 14 R2 16EI 3
[ ]
6EI 4 −1
[F] − 1 =
7L − 1 2

Therefore,
3.2 The Flexibility Matrix Method 27

Fig. 3.5 A rigid plane frame B 2I C

10 kN

12m
6m
I I 12m
A

( ) [ ]( )
R1 6EI PL2 4 −1 2
= − ×
R2 7L 16EI−1 2 3
( )
3PL 5
= −
56 4
15PL 3PL
R1 = − and R2 = −
56 14
Once R1 and R2 are determined, any other quantity of interest such as bending
moment, shear force at any point, any reaction, rotation or deflection at any section
may be obtained. A few more examples of the flexibility method are given below.

Example 3.1 Analyse the rigid frame of Fig. 3.5.

Two support reactions at A are considered here as redundant forces. The released
structure with redundant forces denoted as Rl and R2 respectively is shown in Fig. 3.6
(a). m0 diagram of Fig. 3.6(b) is the bending moment diagram of the released structure
due to the external load, ml diagram is that corresponding to redundant force R1 .
Similarly, m2 diagram represents the variation of the bending moment of the released
structure due to a unit value of the force corresponding to R2 , X 1 and X 2 are the
displacements produced at the released points corresponding to the released forces
R1 and R2 at A.
Following the same argument as discussed in the previous illustration, X 1 can be
obtained by multiplying the area diagrams m0 and m1. This can be done with the
help of Table 3.1. Thus,

1 60 × 6 × 6 1 6 × (60 + 2 × 120) × 6
X1 = − × + ×
6 EI 6 EI
1440
=
EI
1 120 × 12 × 12
X2 = ×
2 EI
8640
=
EI
28 3 Introduction to the Flexibility and Stiffness Matrix Methods

Fig. 3.6 Bending moment B C 10 kN

diagrams for the released
frame in kNm

X2
R1 A X1
D +
R2 120
(a) (b) m0 diagram

12
6 6

- - - 6 + 12

f21 f22
+
f11 + f12
6 12
(c) m1 diagram (d) m2 diagram

Next the flexibility coefficients are to be obtained in the same way as indicated in
the previous illustration. They are

1 6×6×6 6 × 6 × 12 432
f11 = × 3 × + =
3 EI 2 × EI EI
1 12 × 12 × 12 12 × 12 × 12 2016
f22 = × + =
3 2 × EI EI EI
216
f12 = f21 = −
EI
Similarly, assuming that there are no support displacements, the redundants R1
and R2 can be found by applying the principle of superposition as given in Eq. (3.3),
[ ]( ) ( )
1 432 − 216 R1 1 1440
= =
EI − 216 2016 R2 EI 8640

Solution of the above equation yields

( ) ( )
R1 − 4.9
=
R2 − 5.78

Example 3.2 For the frame with a tie-rod shown in Fig. 3.7, find out the force in the
tie-rod. The tie-rod is capable of taking up axial tensile force only. The following are
the particulars for the frame:
3.2 The Flexibility Matrix Method 29

Fig. 3.7 A frame with a 10 kN Rigid joint

tie-rod

3m

Tie-rod
8m

EA = 10 × 103 kN m2 for the tie-rod.

EI = 5 × 103 kN for the inclined members.
The force in the tie-rod is considered here as the redundant force. The tie-rod will
take up axial force only and for the inclined members the energy due to the bending
moment only is considered. X l can be determined exactly in the same way as has
been done in the previous example.

1 5 × 20 × 3 × 2
X1 = × = 4 × 10 − 2 m
3 5 × 103

For calculating the flexibility coefficient f 11 , the contribution due to axial force in
the tie-rod is to be taken into account. Therefore, if R1 is the axial force in the tie-rod,
then (from Fig. 3.8)
( (
M ∂M N ∂N
f11 = dx + dx
EI ∂R1 EA ∂R1
tie-rod
nclined
members
1 5×3×3×2 8×1×1
= × +
3 5 × 103 10 × 103

= 68 × 10 − 4 m

Applying the principle of superposition of displacements yields

68 × 10 − 4 R1 + 4 × 10 − 2 = 0

or
30 3 Introduction to the Flexibility and Stiffness Matrix Methods

10kN
3
20 20 +
+ + +

I I

X1 f11
5kN 5kN
(a) m0 diagram (b) m1 diagram

I I

(c) Axial compression in the

horizontal rod

Fig. 3.8 Bending moment (in KNm) and axial force diagrams (in KN) of the released frame

4
R1 = − × 10 2 = − 5.889
68
kN.
It may be noted that the force R1 in the tie-rod has been assumed as compressive
and the negative sign of the answer indicates that the force is, in fact, tensile.
Example 3.3 For the plane truss of Fig. 3.9, the forces in the members are to be deter-
mined. Select suitable redundants and determine the forces in them. Axial rigidity
of the horizontal and vertical members is EA and for diagonal members it is 2EA.
The plane truss of Fig. 3.9 has two degrees of statical indeterminacy. The diagonals
BF and BD are selected as redundants. These two bars are cut to make the structure
statically determinate. It must be noticed at this stage that the cut bars remain a part
of the released structure and its deformations are to be included in the calculation
of displacements. For convenience, the members are numbered and the results are
done in a tabular manner. Consider only the axial force components in Fig. 3.9.
The displacement obtained by the application of Castigliano’s second theorem then
would be
Σ ∂N L
Xi = N
∂Pi AE
3.2 The Flexibility Matrix Method 31

Q Q

Q F E D
2

3L

Q A
C
2 B
21 11
Q Q
16 4L 4L 16

Q Q

Q 2 3
2 R

2
R
1
7 9
8
1 4

10 11

6 5

Fig. 3.9 A redundant truss

Therefore, displacements corresponding to the release due to external load and

flexibility coefficients are given below (refer Fig. 3.10) (calculation of member forces
corresponding to the releases is given in Table 3.2):
(
) ( )( ) ( )
3 3L Q 4 4L 11
Xi = ( − Q) − + − − + − Q
5 AE 2 5 EA 12
( ) ( )
4 4L 25 5L QL
+ − Q× = −0.835
5 EA 48 2EA AE
( )( ) ( )
11 4 4L 55 5L QL
X2 = Q − + − Q × 1 × = − 5.79
12 5 AE 48 2AE AE
[( )( ) ( )( ) ]
3 3 3L 4 4 4L 1 × 1 × 5L
f11 = − − + − − + 2
5 5 AE 5 5 AE 2AE
12.28L
=
AE
( )( )
3 3 3L L
f12 = f21 = − − = 1.08
5 5 AE AE
32 3 Introduction to the Flexibility and Stiffness Matrix Methods

12.28L
f22 = .
AE
Applying the principle of superposition for displacements, it yields
[ ]( ) ( )
L 12.28 1.08 R1 QL 0.835
=
AE 1.08 12.28 R2 AE 5.790

Q Q
Q Q
2 3 Q
Q R2 2 X1
R1 7 X2
2 9
1 8
R1 R2 4

6 5
(a) (b)

1
1
f21 f12
f11 1 f22

(c) (d)

Fig. 3.10 Forces in the released truss

Table 3.2 Calculation of Forces in the Members of the Truss

Member Forces Forces Forces Final forces
Figure 3.10(b) Figure 3.10(c) Figure 3.10(d)
1 −Q −3/5 0 −1.016Q
2 −Q/2 −4/5 0 −0.5214Q
3 0 0 −4/5 −0.3753Q
4 0 0 −3/5 −0.2815Q
5 11
12 Q 0 −4/5 0.5414Q
6 11
12 Q −4/5 0 0.8953Q
7 − 25
48 Q 1 0 −0.4941Q
8 0 −3/5 −3/5 −0.2975Q
9 − 55
48 Q 0 1 −0.6767Q
10 0 1 0 −0.02674Q
11 0 0 1 0.4691Q
3.3 The Stiffness Method 33

The following are the values of R1 and R2 :

R1 = 0.023736Q
R2 = 0.4691Q

3.3 The Stiffness Method

The steps followed in the stiffness method are described first. Then the method
is explained with examples. Finally, a number of examples demonstrating the
applicability of the method to a variety of structural analysis problems is presented.
First step consists of the determination of the degrees of freedom of the structure.
This involves the identification of separate elements of the structure. The intersections
or interconnections between these elements are called joints or nodes. Displacements
corresponding to each degree of freedom referred to as nodal displacements are
numbered and positive directions are assumed.
The sign convention for the load is that if they are in the same direction as the
nodal displacements, they are assumed to be positive and vice versa (Fig. 3.11).
Secondly, all the nodal displacements are prevented and fixed-end forces for all
the members at desired location under external loading are determined. Restraining
forces are obtained as a sum of the fixed-end forces. Fixed-end forces for uniform
members for some typical loading cases are given in Appendix A.
Thirdly, the structure is assumed to deform corresponding to a unit value of one
of the nodal displacements while all other displacements are restrained. The forces
at all the required points are evaluated. The process is repeated for a unit value of
each of the nodal displacements.
Fourthly, the displacements required to eliminate the restraining forces in step
2 are determined. This requires the application of the principle of superposition
of forces. Final values of displacements are obtained from the solution of a set of
simultaneous equations.
Finally, the end forces of the different members are obtained on the basis of
superposing the fixed-end forces on the restrained structure to those caused by nodal
displacements.

θ Positive

θ Positive
(a) (b)

Fig. 3.11 Sign convention

34 3 Introduction to the Flexibility and Stiffness Matrix Methods

In this method, the displacements corresponding to the degrees of freedom are

determined, and then the forces in the members are evaluated.
The method is illustrated with an example. Let us consider the same continuous
beam of Fig. 3.1 and analyse it by the displacement method.
Step 1: Referring to Fig. 3.12, the continuous beam has two degrees of freedom—
the joint rotations at B and C. They are denoted as X 1 and X 2 .
Step 2: All joints where displacements can take place, i.e. joints B and C in this
case are clamped. In Fig. 3.12(b), the fixed-end moments of the members AB and
BC are determined. This follows a standard procedure. The unbalanced moments at
the joints are termed as R1 and R2 :
R1 = fixed-end moment at B from the member AB.
+ fixed-end moment at B from the member BC.
The signs of the moments are to be properly taken into account. Therefore,
referring to table in Appendix I for the particular loading case,

2PL PL PL
R1 = − + = −
8 8 8

L L

L 2P L P
2 2
B C
A
X1 X2

A B C
1
1
K11 K21
(a)

2P P
A B C

R1 R2

A B C

1
K12 K22
(b)

Fig. 3.12 Different steps in the displacement method

3.3 The Stiffness Method 35

4EI 2EI
1
L L
A B B C

1 4EI
L

Fig. 3.13 Forces due to unit rotation

Similarly, R2 can be determined. But joint C is the end of the member BC. The
unbalanced moment is simply the fixed-end moment at C for member BC under the
given loading.
Therefore,

PL
R2 = −
8
Step 3: Joints are now allowed to deform by a unit value of the nodal displacement
one at a time, while all other displacements are restrained. Referring to Fig. 3.12(c),
the joint B is allowed to rotate by unity in the counterclockwise direction (i.e. positive)
while joint C is kept fixed. The moment developed at B due to unit rotation at B is
termed K 11 and the moment induced at C due to this displacement is K 21 . In order
to determine K 11 , contribution from both the members AB and BC is to be taken into
account (refer Fig. 3.13),

4EI 4EI 8EI

K11 = + +
L L L
2EI
K21 =
L
Similarly,

2EI 4EI
K12 = and K22 = .
L L
The total moment at B consisting of the unbalanced moment acting at that joint
and the member end moments due to all possible rotations of the joints is equal to
zero. Similarly, end C being roller support, total moment at C will be equal to zero.
This follows from equilibrium conditions of the joints. By the application of the
principle of superposition of forces, following equations result:
)
K11 X1 + K12 X2 + R1 = 0
(3.6)
K21 X1 + K22 X2 + R2 = 0

Equation (3.6) written in matrix form becomes

36 3 Introduction to the Flexibility and Stiffness Matrix Methods
[ ]( ) ( ) ( )
K11 K12 X1 R1 0
+ = (3.7)
K21 K22 X2 R2 0

which in compact form becomes

[K] {X } + {R} = {0} (3.8)

The matrix [K] is called the stiffness matrix and its elements K 11 and K 12 , etc.
are called stiffness coefficients. K ij is defined as the force developed at point i corre-
sponding to the unknown displacement at i due to unit value of the displacement at
j corresponding to the unknown displacement at j. It may be noted that the stiffness
matrix is symmetric.
Substituting the value of K11 , K12 · · · · · · , R2 into Eq. (3.7),
[ ]( ) ( ) ( )
2EI 41 X1 PL 1 0
= − =
L 12 X2 8 1 0

Therefore,
( ) [ ] ( )( )
X1 L 2 −1 PL 1
= ×
X2 14EI − 1 4 8 1
( ) ( )
X1 PL2 1
=
X2 112EI 3

Usually, one is more interested in knowing the end forces rather than displace-
ments for most of the problems. So in the stiffness method, the forces in the members
are evaluated after the values of unknown displacements at joints are obtained. This
can be done in a systematic manner.
If a member of a continuous beam is isolated as shown in Fig. 3.14, the end
moments of the member are supposed to consist of the fixed-end moment and bending
moments developed at that end due to rotations at the two ends of the member.
In Fig. 3.14, the end moment at end 1 is given by

PM1 = PMF1 + PMX11 × X1 + PMX12 × X2 (3.9)

where PMF 1 is the fixed-end moment at 1, PMX 11 is the moment induced at 1 due
to a unit value of the rotation at 1 and PMX 12 is the moment induced at 1 due to a
unit value of rotation at 2.
Similarly,

PM2 = PMF2 + PMX21 X1 + PMX22 X2 (3.10)

Equations (3.9) and (3.10) when written in matrix form become

3.3 The Stiffness Method 37

X1
PM2
=

X2 PMF1
PM1 PMF2

1 PMX21

+ +
PMX11 PMX12 1
PMX22

Fig. 3.14 Contribution to the end moment of a member of a continuous beam

( ) ( ) [ ]( )
PM1 PMF1 PMX11 PMX12 X1
= + (3.11)
PM2 PMF2 PMX21 PMX22 X2

which in compact form becomes

{PM } = {PMF} + [PMX ] {X } (3.12)

Similarly, the end reactions can be shown to be equal to the superposition of reac-
tions from the fixed-end condition, to reactions developed due to nodal displacements.
In matrix notations, this can be written as

{PR} = {PRF} + [PRX ] {X } (3.13)

In order to find out member forces and reactions for any structure, Eqns. (3.12) and
(3.13) are to be used. The construction of {PMF}, {PRF}, [PMX], and [PRX] is to be
simultaneously developed with the formation of the stiffness matrix. However, final
solution of Eq. (3.13) is awaited till the unknown displacements {X} are evaluated.
Let us continue the problem of Fig. 3.1 and determine member end actions and
support reactions. The member forces to be determined are the end moment and the
shear force at each end of the member. The support reactions are four in number and
they are consequently numbered. The end actions and reactions are shown with their
positive sign convention in Fig. 3.15. For fixed-end condition, members AB and BC
are considered separately and shown in Fig. 3.16.
38 3 Introduction to the Flexibility and Stiffness Matrix Methods
⎡ PL
⎤
4
⎢ P ⎥
⎢ ⎥
⎢ − PL ⎥
⎢ 4 ⎥
⎢ ⎥
⎢ −P ⎥
{PMF} = ⎢ PL ⎥
⎢ 8 ⎥
⎢ P ⎥
⎢ 2 ⎥
⎢ PL ⎥
⎣− ⎦
8
− P2

and

PR2 PM1 PM3 PM4 PM5 PM7

B
A C

PM6 PM8
PM2

PR1
PR3 PR4

Fig. 3.15 The end moments and reactions of continuous beam of Fig. 3.1

2P

PL/4 – PL/4
A B

(a)
P P
P

PL/8 – PL/8
B C

(b) P/2
P/2

Fig. 3.16 Fixed-end forces of members

3.3 The Stiffness Method 39

2EI 4EI 4EI 2EI

L L L L

1
–6EI
6EI –6EI 6EI 2
2 2 2 L
L L L
(a)

0 2EI 4EI
0
L L
1
–6EI
0 0 6EI 2
2 L
L
(b)

Fig. 3.17 Member end forces and reactions due to unit value of the displacement

⎧ ⎫
⎪
⎪ P⎪
⎨ PL ⎪
⎬
{PRF} = 4 .
⎪
⎪
3P
⎪
⎩ P2 ⎪
⎭
2

Similarly, corresponding to a unit value of the displacement one at a time, the

member end forces and reactions are shown in Fig. 3.17. Based on the values indicated
in Fig. 3.17, {PMX}and {PRX} can be formed. They are.

⎡ 2EI ⎤
L
0
⎢ 6EI
0 ⎥
⎢ L2 ⎥
⎢ 4EI ⎥
⎢ 0 ⎥
⎢ L
6EI ⎥
⎢ 0 ⎥
[PMX ] = ⎢ L2 ⎥
⎢ 4EI 2EI
⎥
⎢ L L ⎥
⎢ 6EI 6EI ⎥
⎢ L2 L2 ⎥
⎣ 2EI 4EI ⎦
L L
6EI 6EI
L2 L2

and
⎡ 6EI ⎤
L2
0
⎢ 2EI
0 ⎥
[PRX ] = ⎢
⎣ 0
L ⎥
6EI ⎦.
L 2

− 6EI
L2
− 6EI
L2

Applying Eq. (3.12) yields

40 3 Introduction to the Flexibility and Stiffness Matrix Methods
⎧ ⎫ ⎡ ⎤
⎪
⎪ 2L ⎪
⎪ 2L 0
⎪
⎪ ⎪
⎪
⎪ 8 ⎪
⎪
⎪
⎢ 6 0 ⎥
⎢ ⎥
⎪
⎪ ⎪
⎪ ⎢ 4L 0 ⎥
⎪
⎪ −2L ⎪
⎪ ⎢ ⎥
⎪ ⎪ 2 ( )
P ⎨ −8 ⎬ EI ⎢ ⎥
⎢ 6 0 ⎥ PL 1
{PM } = + 2 ⎢ ⎥
8 ⎪
⎪ 1 ⎪
⎪ L ⎢ 4L 2L ⎥ 112EI 3
⎪
⎪ ⎪
⎪ ⎢ ⎥
⎪
⎪ 4 ⎪
⎪ ⎢ 6 6 ⎥
⎪
⎪ ⎪
⎪ ⎢ ⎥
⎪ −L ⎪
⎪ ⎪ ⎣ 2L 4L ⎦
⎪
⎩ ⎪
⎭
−4 6 6
⎧ ⎫
⎪
⎪ 30L ⎪ ⎪
⎪
⎪ ⎪
⎪
⎪ 118 ⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ −24L ⎪
⎪
⎪
⎨ ⎪
⎬
P −106
=
112 ⎪
⎪ 24L ⎪ ⎪
⎪
⎪ ⎪
⎪
⎪
⎪ 80 ⎪ ⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ 0 ⎪
⎪
⎩ ⎭
−32

This result agrees with those obtained from the flexibility method. Similarly, for
determining the reactions, Eq. (3.13) yields
⎧ ⎫ ⎡ ⎤
⎪
⎪4⎪ ⎪ 6 0
( )
⎨
P L⎬ EI ⎢ 2L 0 ⎥ PL2
⎢ ⎥ 1
{PR} = + 2 ⎣ 0 6 ⎦ 112EI 3
4 ⎪ 6
⎪ ⎭ ⎪
⎪ L
⎩
2 −6 −6
⎧ ⎫
⎪
⎪ 118 ⎪⎪
P ⎨ ⎬
30L
=
112 ⎪
⎪ 186 ⎪⎪
⎩ ⎭
32

It may be noted that whereas in the flexibility method, there are many possibilities
of choosing the released structure, in stiffness method the analyst is left with no
option, but to follow a rather set procedure. A few problems solved by the stiffness
method are given in the following examples.
Example 3.4 The plane frame of Fig. 3.18a is to be analysed by the displacement
method. All members are uniform and have flexural rigidity EI.
The axial deformation of the members will be neglected. The frame has three
degrees of freedom-one horizontal displacement denoted by X l which indicates
horizontal translation of the joint B or C which is same if the axial deformation
is neglected and two rotations denoted by X 2 and X 3 at joints B and C respectively.
The end moments of the members are also shown in Fig. 3.18b. R1 , R2 and R3 are
the reactions corresponding to the displacements X 1 , X 2 and X 3 , respecttively.
3.3 The Stiffness Method 41

P X3 PM P
2PL 4
PM3 2PL
C C B X1
B
L X2

L PM2
PM5 PM6
D P D P
2L

L PM1
A A
2L
(a) (b)

P 2PL 1 1
R3 K11
C B C B
R1
R2 K21
K31

D P D

A A
(c) (d)

K22 K33 1 K23

C B B C K13
K12
K23 1

1 1
D D

A A
(e) (f)

Fig. 3.18 Different steps to be followed in the rigid frame analysis by the stiffness method
42 3 Introduction to the Flexibility and Stiffness Matrix Methods

Next, all joints are clamped and the fixed-end moments at B and C of the members
are determined. Referring to Fig. 3.18c,

P
R1 = −
2
PL PL
R2 = − + − 2PL = −2PL
4 4
PL
R3 =
4
Therefore,
⎧ ⎫ ⎧ ⎫
⎨ R1 ⎬ ⎨ −2 ⎬
P
R = −8L
⎩ 2⎭ 4 ⎩ ⎭
R3 L

Next, the stiffness coefficients are to be determined. The forces developed due
to a unit horizontal displacement corresponding to X 1 are shown in Fig. 3.18d for
members AB and CD. K 11 consists of the contribution from the members AB and
CD. Thus
108 EI
K11 =
8 L3
Referring to Fig. 3.18d, K21 and K31 will assume the following values:

6EI
K21 = −
4L2
6EI
K31 = −
L2
In the same way, other stiffness coefficients may be determined. Referring to
Fig. 3.18(e),

6EI 8EI 2EI

K12 = − 2
, K22 = , K32 =
4L 2L 2L
Similarly,

6EI 2EI 12EI

K13 = − , K23 = , K33 =
L2 2L 2L
Therefore, the stiffness matrix is
3.3 The Stiffness Method 43
⎡ ⎤
108 −12L −48L
EI ⎣
[K] = −12L 32L2 8L2 ⎦
8L3
−48L 8L2 48L2

Finally, applying the principle of superposition of forces yields

⎡ ⎤⎧ ⎫ ⎧ ⎫ ⎧ ⎫
108 −12L −48L ⎨ X1 ⎬ ⎨ −2 ⎬ ⎨0⎬
EI ⎣ P
−12L 32L2 8L2 ⎦ X2 + −8L = 0
8L3 ⎩ ⎭ 4 ⎩ ⎭ ⎩ ⎭
−48L 8L2 48L2 X3 L 0
⎡ ⎤ ⎧ ⎫
92 12 90
P ⎨
3 2 ⎬
−1 L L
⎣ 12 1802
L
⎦
{X } = [K] {R} = − 18 8L
684EI L L L2 4 ⎩ ⎭
90
L
− 18
L2
207
L2
−L

Therefore,
⎧ ⎫ ⎧ ⎫
⎨ X1 ⎬ PL3 ⎨ 0.0694 ⎬
X = 0.5416/L
⎩ 2⎭ EI ⎩ ⎭
X3 −0.0625/L

The end forces in the members numbered PM 1 to PM 6 are to be determined.

Referring to Figs. 3.18 (b) and (c) yields.
⎡ ⎤
− 6EI2
4L 2L
2EI
0
⎢ − 6EI 4EI 0 ⎥
⎢ 4L2 2L ⎥
⎢ 2EI ⎥
⎢ 0 4EI 2L 2L ⎥
[PMX ] = ⎢ 4EI ⎥
⎢ 0 2EI ⎥
⎢ 6EI 2L 4EI 2L ⎥
⎣ − L2 0 L ⎦
− 6EI
L2
0 2EI
L

and
⎧ ⎫
⎪
⎪ − PL ⎪
⎪
⎪
⎪ PL ⎪
4
⎪
⎪
⎪ ⎪
⎪ PL ⎪
⎨ 4 ⎪
⎬
−4
{PMF} = .
⎪
⎪ 4 ⎪
PL
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ 0 ⎪ ⎪
⎪
⎩ ⎪
⎭
0

Therefore, applying Eq. (3.12) yields

44 3 Introduction to the Flexibility and Stiffness Matrix Methods
⎧ ⎫ ⎡ ⎤
⎪
⎪ −1 ⎪
⎪ −64L0
⎪
⎪ ⎪ ⎢ −68L0 ⎥
⎪
⎪
⎪ 1 ⎪⎪
⎪
⎪ ⎢ ⎥ ⎧ ⎫
⎨
PL −1 ⎬ EI ⎢⎢ 08L4L ⎥
⎥ PL3 ⎨ 0.0694 ⎬
{PM } = + ⎢ ⎥ 0.5416/L
⎪ 1 ⎪
4 ⎪ ⎪ 4L2 ⎢ 04L8L ⎥ EI ⎩ ⎭
⎪
⎪ ⎪
⎪ ⎢ ⎥ −0.0625/L
⎪
⎪ 0 ⎪
⎪ ⎣ −24016L ⎦
⎪
⎩ ⎪
⎭
0 −2408L

or
⎧ ⎫
⎪
⎪ 0.1875 ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ 1.2290 ⎪
⎪
⎪
⎪
⎨ ⎬
0.7707
{PM } = PL .
⎪
⎪ 0.6666 ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ −0.6664 ⎪
⎪
⎪
⎪
⎩ ⎭
−0.5415

Example 3.5 Let us now see the application of the displacement method in the case
of a truss of Fig. 3.19. The forces in each of the members are to be determined. EA
is same for all the members.

The only joint that can be displaced is joint E and for this plane truss, joint E can
have two translational displacements indicated by X 1 and X 2 in Fig. 3.20. R1 and R2
are the reactions acting at E corresponding to displacements X 1 and X 2 .
Joint E is kept fixed and reactions at E due to the external load in this case will
be the horizontal reaction acting to the left due to the force Q acting to the right.
Therefore,
( ) ( )
R1 0
=
R2 −Q

A B C D
θ1 θ2 θ2 θ1

E Q

Fig. 3.19 Truss problem

3.3 The Stiffness Method 45

PM1 PM2 PM3 PM4

A B C D A B C D

E X2 E K 21
1
X1 K 11
(b)
(a) A B C D

E K 22
1
K 12

(c)

Fig. 3.20 Different steps in the solution of truss problem by the stiffness method

Next, the stiffness coefficients are to be determined. Let us first look into the
member AE. The length of the member is given by

L
L1 =
sin θ1

Due to unit movement in the vertical direction of joint E, member AE elongates

[Fig. 3.20(b)] and the tensile force developed is T1 = EA L
sin2 θ1 .
and the vertical component of the tensile force is EAL
sin3 θ1 , and the horizontal
component is L sin θ1 cos θ1 acting to the left for this member., Therefore, in order
EA 2

to determine K 11 , the contribution of the vertical components of the forces from all
members meeting at that joint is to be taken into account.

EA
K11 = 2 (sin3 θ1 + sin3 θ2 )
L
Similarly, sum of the horizontal, components of the forces in the members will
give K 21 . In this case, it may be verified that this becomes zero. Thus, K 21 = 0.
For stiffness coefficients corresponding to unit horizontal displacement, let us
again focus our attention to the member AE. Tensile force developed in AE is
EA
L
cos θ1 sin θ1 and horizontal component of this force is EA L
cos2 θ1 sin θ1 . It may
46 3 Introduction to the Flexibility and Stiffness Matrix Methods

be noticed here that the sum of the vertical component of the forces in all members
meeting at E due to a unit horizontal movement is equal to zero. Therefore,

EA
K22 = 2 (cos2 θ1 sin θ1 + cos2 θ2 sin θ2 )
L

K21 = 0

Before proceeding with the final evaluation of the unknown displacement at joint
E, let us have a look into the member forces. We know

{PM } = {PMF} + [PMX ] {X }

In this case {PMF} is a null matrix and referring to Fig. 3:20(b) and 3.20(c), it
can be shown that
⎡ ⎤
sin2 θ1 cos θ1 sin θ1
EA ⎢
⎢ sin θ2 cos θ2
2
sin θ2 ⎥
⎥
[PMX ] = ⎣
L sin2 θ2 − cos θ2 sin θ2 ⎦
sin2 θ1 − cos θ1 sin θ1

Let us assume that θ1 = 45◦ and θ2 = 60◦ . Substituting these values in stiffness
coefficients and applying the principle of superposition,.
[ ]( ) ( ) ( )
EA 2.006 0 X1 0 0
+ =
L 0 1.14 X2 −Q 0

Solving the above equations,

X1 = 0

QL QL
X2 = = 0.8771
1.14AE AE
The member forces then are
⎡1 1
⎤
2 /2
⎢ 3 ⎥( )
EA ⎢ 3
⎥ 0 QL
{PM } = ⎢ /4 ⎥
4
L ⎢3 3⎥ 1
⎣ 4 − 4 ⎦ 1.14 AE
1
2
− 21
3.3 The Stiffness Method 47

C A

L/2
L B
L/2
Q

Fig. 3.21 A grid

⎧ ⎫
⎪
⎪ ⎪
0.438 ⎪
⎨ ⎬
0.380
= Q
⎪ −0.380 ⎪
⎪ ⎪
⎩ ⎭
−0.438

Example 3.6
The grid of Fig. 3.21 will be solved. The grid is a planar structure and the difference
between a grid and a rigid frame lies in the application of the load. The loading lies
in the plane of the frame in the case of rigid frame, whereas the grid is subjected to
loads perpendicular to its plane. Members are primarily subjected to flexural action
in the case of a rigid frame, whereas in a grid, members are subjected to bending as
well as twisting. Assume GJ/EI = 0.8 for both the members AB and BC. The joint
B is assumed to be rigid. The degrees of freedom for the grid of Fig. 3.21 is 3, of
which two are in rotation about two orthogonal axes in the horizontal plane and the
third is the vertical displacement. The internal forces of the members of the grid are
shown in Fig. 3.22(a).
The structure lies in the x–y plane and all applied forces act parallel to z-axis. Each
member experiences torsion as well as shear and bending. The internal forces acting
on a member in a plane grid are completely defined by two flexural end moments
and the torsional moment in the member. PM 5 and PM 6 are the twisting moments of
the members AB and BC respectively and their direction is positive when the arrows
indicating the moments act outwards. {R} matrix corresponding to the displacements
in this case is.
⎧ Q ⎫ ⎧ ⎫
⎨ 2 ⎬ Q ⎨ ⎬
4
{R} = 0 = 0
⎩ QL ⎭ 8 ⎩ ⎭
8
L

The stiffness matrix for the grid is next to be formed. The stiffness coeffi-
cients corresponding to a unit value of X 1 are determined as follows. Referring to
Fig. 3.22(b), two members AB and BC are shown separately for this unit displacement.
The values of the different quantities indicated in Fig. 3.22(b) are as follows:
48 3 Introduction to the Flexibility and Stiffness Matrix Methods

A PM1 PM4 C

PM
PM2 PM3 PM 6
5

B Y
L/2
L X3
X2 X1 X
L/2
Q
Z
(a)
K′11 K″11

B K′31 K″21 B
K′21 K″31
1
C C

(b)

K′22 K″32
A B B C
K′32 K″22

K′12 K″12

(c)

Fig. 3.22 Different steps in the analysis of the grid structure

' 12EI ' 6EI '

K11 = , K21 = − 2 , K31 = 0
L3 L
Similarly,

'' 12EI '' '' 6EI

K11 = , K21 = 0 , K31 =
L3 L2
The elements of the first column of the stiffness matrix will be obtained by taking
into account the contribution of both the members. Therefore,

24EI 6EI 6EI

K11 = 3
, K21 = − 2 , K31 =
L L L2
Next, let us find out the stiffness coefficients due to a unit displacement corre-
sponding to X 2 . Again, two members AB and BC are shown separately for this unit
displacement. The end B of the member AB is rotated flexurally. The end B of the
member BC is twisted. The different quantities of Fig. 3.22c are
3.3 The Stiffness Method 49

' 6EI '' 4EI '

K12 = − 2
, K22 = , K32 = 0
L L

'' '' GJ ''

K12 = 0, K22 = , K32 = 0
L
Therefore, combining the contribution of both the members meeting at the joint
yields

6EI 4EI GJ
K12 = − 2
, K22 = + , K32 = 0
L L L
In the same way, remaining stiffness coefficients may be determined. They are

6EI GJ 4EI
K13 = 2
, K23 = 0 , K33 = +
L L L
Once the stiffness coefficients are evaluated, the stiffness matrix becomes
⎡ ⎤
24 −6L 6L
EI ⎣ ⎦
[K] = 3 −6L (4 + β) L2 0
L
6L 0 (4 + β) L2

Substituting β = GJ
EI
= 0.8 in the above equation yields
⎡ ⎤
24 − 6L 6L
EI ⎣
[K] = 3 −6L 4.8L2 0 ⎦
L
6L 0 4.8L2

Applying the principle of superposition of forces yields

⎧ ⎫ ⎡ ⎤⎧ ⎫
⎨ −4 ⎬ 24 −6L 6L ⎨ X1 ⎬
Q EI
0 = 3 ⎣ −6L 4.8L2 0 ⎦ X2
8 ⎩ ⎭ L ⎩ ⎭
−L 6L 0 4.8L2 X3

Solution of the above equation gives

⎧ ⎫ ⎧ ⎫
⎨ X1 ⎬ QL ⎨ 6.36L ⎬
X = 79.20
⎩ 2⎭ 1658.88EI ⎩ ⎭
X3 36.00
⎧ ⎫
⎨ −31.68L ⎬
QL
= −39.60
829.44EI ⎩ ⎭
18.00
50 3 Introduction to the Flexibility and Stiffness Matrix Methods

The member forces can now be determined

⎧ ⎫ ⎧ ⎫
⎪
⎪ PMF1 ⎪⎪ ⎪
⎪ 0 ⎪⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ PMF ⎪
⎪ ⎪
⎪ 0 ⎪
⎪
⎪
⎨
2 ⎪
⎬ ⎪
⎨ QL ⎪⎬
PMF3
= 8
⎪
⎪ PMF4 ⎪⎪ ⎪
⎪ −8 ⎪
QL
⎪
⎪
⎪ PMF ⎪ ⎪ ⎪
⎪ 0 ⎪ ⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎩
5 ⎪
⎭ ⎪
⎩ ⎪
⎭
PMF6 0

Likewise, [PMX] may be determined

⎡ −6EI 2EI
⎤
L2 L
0
⎢ −6EI 4EI ⎥
⎢ 2 0 ⎥
⎢ L
6EI
L
4EI ⎥
⎢ 0 ⎥
[PMX ] = ⎢ L2 L ⎥
⎢ 6EI
0 2EI ⎥
⎢ L2 L ⎥
⎣ 0 0 GJ L
⎦
0 GJ L
0

Therefore,
⎧ ⎫ ⎧ ⎫ ⎧ ⎫
−6EI 2EI
⎪
⎪ PM1 ⎪ ⎪ ⎪
⎪ 0 ⎪⎪ ⎪
⎪ 0 ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
L 2 L ⎪
⎪
⎪
⎪ PM ⎪
⎪ ⎪
⎪ 0 ⎪
⎪ ⎪
⎪
−6EI 4EI
0 ⎪
⎪ ⎧ ⎫
⎪
⎨
2 ⎪
⎬ ⎪
⎨ QL ⎬⎪ ⎪
⎨ 6EI
L 2 L ⎪
4EI ⎬ 2 ⎨ −31.68L ⎬
PM3 0 QL
= 8
−QL + L2 L −39.60
⎪
⎪ PM4 ⎪ ⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
6EI
0 2EI ⎪
⎪ 829.44EI ⎩ ⎭
⎪
⎪ ⎪
⎪ ⎪
⎪
8 ⎪
⎪ ⎪
⎪
L 2 L ⎪
GJ ⎪ 18.00
⎪
⎪ PM5 ⎪ ⎪ ⎪
⎪ 0 ⎪⎪ ⎪
⎪ 0 0 L ⎪ ⎪
⎪
⎩ ⎪
⎭ ⎪
⎩ ⎪
⎭ ⎪
⎩ ⎪
⎭
PM6 0 0 GJ L
0
⎧ ⎫
⎪
⎪ 0.133 ⎪⎪
⎪
⎪ ⎪
⎪
⎪
⎪ 0.038 ⎪
⎪
⎪
⎨ ⎪
⎬
−0.017
= QL
⎪
⎪ −0.310 ⎪⎪
⎪
⎪ ⎪
⎪
⎪
⎪ 0.017 ⎪⎪
⎪
⎪
⎩ ⎭
−0.038

The equilibrium of joint B can be checked.

3.4 Incorporation of Different Loading Conditions

It has been seen in the previous section that the flexibility matrix and the stiffness
matrix of a structure are independent of the loading and are properties of the structure.
A structural engineer has to deal with different types of loading in analysing the
3.5 Other Types of Loadings 51

structure. If a structure is analysed by taking one loading condition at a time, then

the same flexibility and stiffness matrices are to be inverted depending on the method
chosen for the analysis, every time for a separate load. This is time-consuming and, in
fact, is not necessary, as all the loading conditions can be incorporated in one loading
matrix—different columns of the matrix representing different loading conditions.
So Eqns. (3.5) and (3.8) are modified as follows:

[R] −[F] −1 [X ]
= (3.14)
NR × NLC NR × NR NR × NLC

[X ] −[K] − 1 [R]
and = (3.15)
NP × NLC NP × NP NP × NLC

where.
NR is the number of redundants of the structure,
NP is the number of degrees of freedom,
NLC is the number of loading conditions for which results are required.
Each column of [P] and [X] refers to the corresponding values of the forces and
displacements of the structure for a particular condition.
Similarly, Eqns. (3.12) and (3.13) for determining reactions and member forces by
the displacement method can be modified to incorporate different loading conditions.
They are done as follows:

[PM ] [PMF] [PMX ] [X ]

= + (3.16)
NF × NLC NF × NLC NF × NP NP × NLC

[PR] [PRF] [PRX ] [X ]

= + (3.17)
NRE × NLC NRE × NLC NRE × NP NP × NLC

where
NF is the number of unknown member forces.
and NRE is the number of unknown reactions.

3.5 Other Types of Loadings

In addition to the conventional loading which act in the form of concentrated, uniform
and varying loads that act on a structure, an engineer has to deal with other effects
such as the variation in temperature, lack of fit, prestrain and settlement of supports,
etc.
For many reasons, supports of the structure yield. Though the value of the
settlement is small, the geometry is appreciably affected in the case of statically
indeterminate structure and considerable stresses and strains are introduced.
52 3 Introduction to the Flexibility and Stiffness Matrix Methods

Certain structures may be subjected to severe variation in temperature. These

temperature changes tend to alter the length or curvature of the member of the
structure; If these members are not allowed to undergo free movements, then they
will be strained.
Due to the error in fabrication, members sometimes become too long or too short
to be properly fitted in its place. The members in such cases are forced to fit during
erection. This results in members being stressed before any load is applied. A member
may be prestrained as is done in prestressed concrete beams which results in internal
forces in the members..
Matrix methods presented so far can be readily extended to incorporate the above
effects.

3.5.1 Treatment by the Flexibility Matrix Method

The effects of the variation in temperature, lack of fit, prestrain, settlement of support,
etc. can be catered in the flexibility matrix method by making Eq. (3.4) more general.
It becomes

[F] {R} + {X } + {XS } + {XLT } + · · · · · · = {0} (3.18)

where {X S }, {X T }… represent the displacements at the releases of the primary

structure due to prestrain, temperature changes, etc. Equation (3.18) suggests that
matrices {X S }, {X T }, etc. are to be added to the matrix {X}, displacements due to
external load acting on the released structure. The procedure has been explained with
the help of the following example.

Example 3.7 Analyse the continuous beam of Fig. 3.23 for (a) a unit vertical move-
ment, (b) a unit rotation of support B in the clockwise direction. The flexural rigidity
EI is constant throughout the beam. The beam has two degrees of statical indeter-
minacy. The internal support moments at B and C are considered as redundants.
The continuous beam is thus made statically determinate by introducing hinges at
supports B and C. The loading matrix generation will be dealt after the development
of the flexibility matrix. The steps followed to determine flexibility coefficients will
not be given here, as it has already been given in detail. The flexibility matrix for this
problem with the chosen redundants is given by.

[ ]
L 41
[F] =
6EI 14
[ ]
4 −1
and [F] −1 = 2EI
.
5L −1 4
3.5 Other Types of Loadings 53

A B C D

L L L

Fig. 3.23 Example 3.7

For the first loading case, displacements corresponding to the redundants are
denoted as X 11 and X 21 . Referring to Fig. 3.24(b), they assume the following values:

2 1
X11 = − and X21 =
L L
For this case, the solution is simple

[F] {R} = −{X }

{R} = −[F] −1 {X }

Therefore,
( ) [ ] ( ) ( )
R1 2EI 4 −1 1 −2 2EI 9
= − =
R2 5L −4 4 L 1 5L2 −6

Fig. 3.24 Loading of R1 R2

Example 3.7

A B C D
(a)

X21
1
A X11 C D
B
(b)

A B X11 C D
(c)
54 3 Introduction to the Flexibility and Stiffness Matrix Methods

1 2EIθ
5L
A B

Fig. 3.25 Freebody of AB

In the second loading case, the joint B is rotated by unit angle. Therefore, relative
rotation of support B is zero, whereas the relative rotation of support C be X 22 [
Fig. 3.24c].
Therefore,

X12 = 0 and X22 = θ (say)

The final equation becomes

( ) [ ]( )
R1 2EI 4 1 0
= −
R2 5L −1 4 θ

The above equation gives

2EI θ 8EI θ
R1 = and R2 = −
5L 5L
The values of R1 and R2 are given in terms of θ . So the value of θ is to be
determined. Considering the freebody diagram of the span AB (Fig. 3.25) of the
continuous beam.
where the joint B rotates by unity, the rotation due to couple at B should be equal
to unity

2EI θ L 15
· = −1 or θ = −
5L 3EI 2
On back substitution of θ , the final values of the redundant forces are

3EI 12EI
R1 = − and R2 =
L L

3.5.2 Treatment by the Stiffness Method

Like the flexibility method, separate or combined effects of temperature variation,

shrinkage, settlement of supports, etc. can be readily incorporated into the analysis
3.5 Other Types of Loadings 55

B I
C X1
X3 X2
2L

L
I I

A 2L

Fig. 3.26 A rigid frame

by the stiffness method. These effects are considered to act in the restrained structure.
Resulting restraining forces are evaluated and are added to the [R] matrix produced
by the loads. Thus Eq. (3.8) becomes

[K] [X ] + [R] + [RS ] + [RT ] + · · · · · · = [0] (3.19)

where [RS ], [RT ], etc. are the forces in the restrained structure due to settlement,
temperature variation, etc. The procedure is explained with the help of the following
examples.

Example 3.8 For the frame of Fig. 3.26, support D has vertical settlement of Δ
simultaneously with horizontal movement to the right of 2 Δ. EI is constant for all
members.

The example is solved by the displacement method. The stiffness matrix and its
inverse have been determined in Example 3.4.
⎡ ⎤
92 12 − 90
L2 ⎣ 12 180L L
⎦
[K] − 1 = L L2
− 18
L2
684EI
− L − L2 L2
90 18 20

For given joint displacements, the forces in the restrained structure are to be evaluated.
Referring to Fig. 3.27a, the restraining forces corresponding to X 1 , X 2 and X 3 for the
vertical settlement Δ of support D are.

6EI Δ 6EI Δ
R'1 = 0, R'2 = and R'3 =
4L2 4L2
56 3 Introduction to the Flexibility and Stiffness Matrix Methods

Δ
2Δ

(a) (b)

Fig. 3.27 Given point displacements of Example 3.8

Due to the horizontal movement 2 Δ of support D, the corresponding values are

[Fig. 3.27b].

24EI Δ 12EI Δ
R''1 = , R''2 = − , R''3 = 0
8L3 4L2
Therefore, the final values of restoring forces are obtained by adding the two cases
dealt separately as above.

24EI Δ
R1 = R'1 + R''1 =
8L3
6EI Δ
R2 = R'2 + R''2 = −
4L2
6EI Δ
R3 = R'3 + R''3 =
4L2
Final equations are
⎧ ⎫ ⎡ ⎤ ⎧ 2⎫ ⎧ ⎫
⎨ X1 ⎬ 92 12 − 90 ⎨ L⎬ ⎨ 82/L ⎬
L3 ⎣ 12 180 L L
⎦ 6EI Δ ΔL
X = − 18 −1 = − −174/L2
⎩ 2⎭ 684EI L L2 L2 4L2 ⎩ ⎭ 456 ⎩ ⎭
X3 − L − L2 L2
90 18 207
1 45/L2

After joint displacements are determined, the forces in the members can then be
evaluated.
Example 3.9 For the truss of Fig. 3.28, determine the forces in the members if the
member BE is Δ too short. EA is same for all members. The inverse of the stiffness
matrix for this truss worked out in Example 4.5 is.
3.5 Other Types of Loadings 57

A B C D
45° 60° 45°
60°

E X2
X1

Fig. 3.28 Example 3.9

[ ]
L 1.14 0
[K] − 1 =
2.28 EA 0 2.004

◦
√
In order to bring the member BE to joint E, a tensile force of ΔEA sin
L
60
= 32L Δ EA

is to be applied. This tensile force is applied at the joint E as shown in Fig. 3.29. The
reactions have to be in opposite directions.
Therefore,
( ) √ (√ )
R1 3 Δ EA 3
= − 2
1
R2 2L 2

Final equations then become

Fig. 3.29 External force B

concept in Example 3.9
60°

1 2 3 4

√3 ΔEA
2L
58 3 Introduction to the Flexibility and Stiffness Matrix Methods
( ) [ ]√ (√ )
X1 −L 1.14 0 3 Δ EA 3
= 1
X2 2.28 EA 0 2.004 4L 2
√ [ ] [ ]
3Δ 1.975 Δ 0.855
= − = −
4 × 2.28 2.004 2.28 0.868

Once the displacements at the joints are determined, the forces in the members of
the truss (based on Example 3.5) are
⎡1 1
⎤ ⎧ ⎫
⎪
⎪ √0 ⎪ ⎪
⎥ (−Δ) (
√2 )
2
EA ⎢
⎢ 3
⎥ 3
0.855 EA ⎨
3Δ
⎬
{PM } = ⎢ 4 ⎥ 4√ +
L ⎣ − 43 ⎦ 2.28
3 0.868 2L ⎪
⎪ 0 ⎪ ⎪
4 ⎩ ⎭
− 21
1
2
0
⎧ ⎫ ⎧ ⎫
⎪
⎪ 0.8615/2.28 ⎪⎪ ⎪
⎪ −0.3778 ⎪
⎪
EA Δ ⎨ 0.4203 ⎬ EA Δ ⎨ 0.4203 ⎬
= =
L ⎪ ⎪ −0.265/2.28 ⎪⎪ L ⎪⎪ −0.1160 ⎪
⎪
⎩ ⎭ ⎩ ⎭
5.5 × 10 − 3 −0.0002

3.6 Incorporation of Shear Deformation

The shear deformation can be included in the stiffness matrix. This is done on the
basis of flexibility coefficient.
Consider a cantilever beam AB of Fig. 3.30a. Two coordinates 1 and 2 have been
indicated. Applying Eq. (2.23), the flexibility coefficients after incorporating the
shear deformation are
( )
L3
f11 = 3EI + μAG
L
, f22 = EIL
(3.20)
f12 = f21 = L2 /2EI

where µ is a dimensionless coefficient and is dependent on the shape of the cross

section and is given by
( ( 2 )
Q dA
μ = I / A 2
(3.21)
b

where A = cross-sectional area,

b = width of the section,
Q = first moment of area of the part of the cross section above the.
Section considered, about the neutral axis.
3.7 Relation Between Flexibility and Stiffness Matrices 59

Fig. 3.30 Consideration for L 2

shear A B

1
(a)

2
4
EI, GA

1 3
(b)

The flexibility matrix then is

[ ]
L3 L2
+ L
μAG
[F] = 3EI
L2
2EI
L
(3.22)
2EI EI

The stiffness matrix corresponding to coordinates of Fig. 3.30a is

[ ]
−1 1 12EI
− 6EI
[K] e = [F] = L3 L2 (3.23)
1 + α − 6EI
L2
(4 + α) EI
L

where α = μAGL 12EI

2.

In a similar way, the stiffness matrix corresponding to coordinates of Fig. 3.30b

with the shear deformation considered can be determined. The final value for this
case is
⎡ 12 ⎤
L3
symmetric
EI ⎢ 62 (4 + α) ⎥
[K] e = ⎢ L L ⎥ (3.24)
1 + α ⎣ − 123 − 62 12 ⎦
L L L3
2 −α (4 + α)
6
L2 L
− L62 L

3.7 Relation Between Flexibility and Stiffness Matrices

In Fig. 3.31, a beam with two different moments of inertia at two portions are shown.
The unknown displacements at the intersection of the two parts denoted as X1 and
X2 corresponding to the deflection and the rotation and the forces corresponding to
X1 and X2 are R1 and R2 . So, total displacement can be written as
( )
X1 = f11 R1 + f12 R2
(3.25)
X2 = f21 R1 + f22 R2
60 3 Introduction to the Flexibility and Stiffness Matrix Methods

1
2

Fig. 3.31 A stepped beam

Here f 11 , f 12 , etc. are flexibility coefficients. So Eq. (3.20) can be written in compact
form as

{X } = [F] {R}

or {R} = [F] − 1 {X } (3.26)

[F] is the flexibility matrix.

Based on the same coordinate system, let us form the stiffness matrix. If forces
K 11 , K 21 are applied at 1 and 2 such that X 1 = 1 and X 2 = 0, then
( ) ( )
K11 −1 1
= [F] (3.27)
K21 0

Similarly, if forces K 12 , K 22 are applied at 1 and 2 such that X l = 0, but X 2 = 1,

then
( ) ( )
K12 0
= [F] − 1 (3.28)
K22 1

Combining Eqns. (3.22) and (3.23) yields

[ ] [ ]
K11 K12 10
= [F] − 1
K21 K22 01

or

[K] = [F] − 1 (3.29)

Hence, based on the same coordinate system, the stiffness matrix is the inverse of
the flexibility matrix and vice versa.
3.8 Equivalent Joint Loads 61

However, it must be noted that for the solution of the same problem by the flex-
ibility method and the stiffness method, different coordinate systems are usually
chosen—as in the first case the unknowns are redundant forces whereas in the other
case the unknowns are the degrees of freedom. So, for the same structure two coordi-
nate systems will not be same. Therefore, inverse of a stiffness matrix for the solution
of a structural problem is indeed a flexibility matrix, but it is not the same flexibility
matrix as one would obtain for the solution of the same problem.

3.8 Equivalent Joint Loads

Equation (3.8) for the displacement method is reproduced.

[K] [X ] = −[R] (3.30)

Reacting force matrix [R] is replaced by [P] such that

[P] = −[R] (3.31)

Equation (3.30) is thus modified with the help of Eq. (3.31) as

[K] [X ] = [P] (3.32)

The procedure of calculating [P] is same as that of [R] with the sign changed. [P]
matrix is the equivalent joint load matrix. The construction of [P] matrix is explained
with help of the following example.
The frame of Fig. 3.32 (a) is restrained at joints B and C. The reactions at
the restrained joints are determined. Different fixed-end forces are indicated in
Fig. 3.32(b). The sum of the reacting forces at joints B and C is shown in Fig. 3.32(c).
This is what is termed as equivalent joint loading. The part corresponding to the
degrees of freedom will be associated with the {P} matrix. If the frame has three
degrees of freedom shown as X 1 , X 2 and X 3 in Fig. 3.32(c), the corresponding {P}
matrix then will be
⎧ ⎫ ⎧ ⎫
⎨ P1 ⎬ ⎨ 4⎬
P = −2
⎩ 2⎭ ⎩ ⎭
P3 −4
62 3 Introduction to the Flexibility and Stiffness Matrix Methods

X2 3 kN/m

B C
X1
X3

8 kN
6m

3m
A D

4m

(a) 6 kN 6 kN

6 kN/m 4 kN/m 4 kN/m

B C 2 kN/m 4 kN/m
B 4 kN C 4 kN

6 kN 6 kN

A D

(b) (c)

Fig. 3.32 Calculation of equivalent joint loads

3.9 Choice of the Method of Analysis

Matrix methods of structural analysis are primarily meant for structures of complex
shapes which are large in size and for which hand computation is prohibitive. Matrix
methods are computer-oriented methods and their relative merits and demerits should
not be viewed with small sized problems. It has already been seen that there are two
basic approaches—the flexibility method and the stiffness method. Looking into
the fundamental philosophy of both these approaches, it is revealed that they are
practically same and in both a set of simultaneous equations is generated. The choice
of the method as such should depend primarily on the ease with which the problem
can be formulated. For a particular problem, one approach may appear easier than
the other.
With the flexibility method, the analyst is free to make his own choice of the
redundants and he has at his disposal a variety of choices. But in the stiffness method
there is no option on the selection of the restrained structure and as such the analysis
follows a rather set procedure. The analysis by the stiffness method is thus more
mechanical than the flexibility method. The economy of analysis has, however, to be
3.9 Choice of the Method of Analysis 63

viewed on the basis of the time taken by the computer for the solution of equations.
If the time taken by other operations is assumed to be same, the main time in the
computer is consumed in the solution of the simultaneous equations. Generally, the
elements of flexibility matrix are sparsely populated, whereas the problem formulated
by the stiffness method is banded in form. If advantage is taken of the banded matrix,
then less storage is required in the computer as well as less solution time. For many
problems, the stiffness method is thus better suited than the flexibility method. In
certain problems like truss-type structures such as transmission tower, lattice girders,
etc., where degrees of statical indeterminacy is much lesser than the degrees of kine-
matic indeterminacy, the flexibility method may appear more economical than the
stiffness method in spite of taking advantage of its banded nature. The displacement
method is, however, easy to formulate. In this book, more emphasis is laid on the
development and application of the stiffness method for the solution of structural
problems.
Exercise 3
Solve the problems 3.1 to 3.13 by the flexibility method. Select suitable redundants
and determine their values.
3.1 Write the flexibility matrix corresponding to the coordinate indicated.

1
2
EI Constant
L
(a)

1 2
L L L EI Constant

(b)

2
1

EI Constant for
L all Members

(c)
64 3 Introduction to the Flexibility and Stiffness Matrix Methods

Prob. 3.1
3.2 Determine the forces in the two springs at B and C. The beam AD has a
constant flexural rigidity EI and both the springs have stiffness 100/L3 .

L/2

L L L

Prob. 3.2
3.3 Determine the support reactions and draw the bending moment diagram for
the portal frame subjected to a point load of 120 kN, acting at the mid-point of
BC.

120 kN
B C

18 m EI Constant for
all Members

A D
30 m
3.9 Choice of the Method of Analysis 65

Prob. 3.3
3.4 Determine the end moments of the members and draw the bending moment
diagram for the gable frame. El is same for all members.

20m 20m

10 kN

18m
10m EI Constant

Prob. 3.4
3.5 Determine the end moments of the members and draw the bending moment
diagram for the closed rectangular portal frame. EI is same for all members.

200 kN

12m

9m
66 3 Introduction to the Flexibility and Stiffness Matrix Methods

Prob. 3.5
3.6 Determine the reactions at the supports. Also calculate the forces in the
members and deflection at the load point. EI is same for all members.

1m
Hinge
B C

1m 3m

3t

3m EI Constant

A D

Prob. 3.6
3.7 Determine the member forces of the trussed beam. Assume I as the moment
of inertia of the members and A as the area for all members.

P P
a a a

a
3.9 Choice of the Method of Analysis 67

Prob. 3.7
3.8 Determine the redundant forces.

2P P

L/2 L/2
A B C D

2I 2I I
L L L

Prob. 3.8
3.9 Find out the forces in all the members of the truss. The axial rigidity for the
horizontal and vertical members is 2EA and for the diagonal members EA.

P D

C 45° E
P

B F

A G
L
68 3 Introduction to the Flexibility and Stiffness Matrix Methods

Prob. 3.9
3.10 Evaluate the forces in all the members of the truss.

P
L
B C
P

A D

L L
L/2 Z L/2
A B C

2P P

Y
L

D X

Prob. 3.10 Prob. 3.11

3.11 A and C are simple supports. Two moments about x- and z-axes may be
considered as redundants. Joint B is rigid. EI/GJ = 0.8 for all the members.
Determine the redundants.
For problem 3.10, if the support settles by Δ, determine the member forces in the
truss.
Find the reactions at the supports of the continuous beam if joint C is displaced
by unity along with clockwise rotation of joint A of magnitude unity.
3.12 Solve the following problems by the stiffness method:
3.9 Choice of the Method of Analysis 69

B C
A

2L 3L

Prob. 3.13
3.14 Form the stiffness matrix based on the coordinate system indicated.

1 3
A D
2 B 4 C EI Constant
L L L

(a)

C 6
4
5 F

L
A

30°
B
1 B
3 E
2
2 60°

2I
E 1
L 60°

C
30°

A D D L/EA is Constant for

L all the Members

(b) (c)
70 3 Introduction to the Flexibility and Stiffness Matrix Methods

Prob. 3.14
3.15 Find all support rotations.

w/unit length 2w/unit length

B C
A

I 2I
L L

Prob. 3.15
3.16 Find all the member forces.

Prob. 3.16
3.17 For the grid shown, find the unknown joint displacements and the member
forces.

D
2
EI = 7300 kNm
2
GJ = 58000 kN

3m

A B C

1.2m

3m 3m
3.9 Choice of the Method of Analysis 71

Prob. 3.17
3.18 Determine the member forces for the truss.

B 50 kN

6m

C 20 kN

EA Constant
6m

A D
8m 8m

Prob. 3.18
3.19 Determine the member forces for the truss.

6m EA Constant

6kN

12m 4kN

10m
20m

Prob. 3.19
3.20 Find out the joint displacements and the member end forces for the rigid
frame, EI constant for all members.
72 3 Introduction to the Flexibility and Stiffness Matrix Methods

P
L
L

L/4
L
2L

EI Constant for all

Prob. 3.20
3.21 Determine the reactions and draw the B.M. diagram.

100 kNm

200 kNm
5m
kN

7.
60

400 kN
5m
7.

6m 6m 4m

Prob. 3.21
3.22 For the truss shown, determine the forces in the members, if AC is Δ too
long and the support D settles by Δ. The axial rigidity is 2AE for horizontal and
vertical members and EA for diagonal members.

B C

A D
References and Suggested Readings 73

Prob. 3.22
3.23 Determine the support reactions when the beam at B is rotated through an
angle θ in the counterclockwise direction.

A B C
EI Constant

L L

Prob. 3.23

References and Suggested Readings

1. J.M. Gere, W. Weaver Jr., Analysis of Framed Structures, 2nd edn. (D. Van Nostrand, USA,
1982)
2. A.S. Hall, R.W. Woodhead, Frame Analysis (John Wiley and Sons, USA, 1961)
3. K.H. Gerstle, Basic Structural Analysis (Prentice Hall, USA, 1974)
4. H. Kadestuncer, Elementary Matrix Analysis of Structures, McGraw-Hill Book Co., 1974.
5. W. MacGuire, R.H. Gallaghar, R.D. Zieman, Matrix Structural Analysis, 2nd edn. (John Wiley &
Sons, USA, 2000)
6. J.S. Przmieniecki, Theory of Matrix Structural Analysis, McGraw-Hill, 1968.
7. M.F. Rubinstein, Matrix Computer Analysis (Prentice Hall, USA, 1969)
8. C.K. Wang, Introductory Structural Analysis with Matrix Methods (Prentice Hall, USA, 1973)
Chapter 4
Direct Stiffness Method

4.1 Introduction

A more formalized approach to the stiffness method has been presented in this
chapter. It will be shown here that the overall stiffness matrix of the structure will be
generated automatically by the computer, based on the geometric and elastic proper-
ties of the individual element. The unknown displacements at the joints are obtained
from the overall solution of the problem. Once the displacements at the joints are
known, the stress resultants for any member can be obtained. In this approach, the
procedure becomes mechanical, the technique can be easily coded in the computer
and it results in a saving of storage space.

4.2 Local and Global Coordinate System

In this procedure, the stiffness matrix of an individual element is to be evaluated. In

order to do this, a clear concept of the local and global axis system and their relations
are necessary.
A local coordinate system is one that is defined for a particular element and not
necessarily for the entire body or structure. This may be explained with reference
to Fig. 4.1 where the bar of a truss has been indicated. This member develops axial
force due to the external load and as such the local axis system is considered along
the axis of the member. The stiffness matrix of this member in the local axis system
is given by
[ ]
EA 1 −1
[K ]e = (4.1)
L −1 1

where EA is the axial rigidity of the uniform member and L, its length. In a truss, the
members are oriented at different inclinations and as such, the stiffness matrix written

© The Author(s) 2022 75

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_4
76 4 Direct Stiffness Method

P4′ – X4′

P2′ – X2′ Y
2 P2′ – X2′ P3′ – X3′
2
1
P1′ – X1′
X 1 P1′ – X1′

(a) Local axis system (b) Global axis system

Fig. 4.1 Element of a truss

as above for different members, will have to be based on a different axis systems.
The difficulty which will result in such a case is that the member stiffness matrices
at the joints cannot be added. This is possible only when it is referred to a common
axis system. Global axis system is this common coordinate system which is valid
for the entire structure. The element stiffness matrix must be written in the global
coordinate system before assembly. The displacements that will be obtained from
the overall solution will be in a global coordinate system. It is, however, desirable to
determine the member forces in local coordinate system.
In order to understand how the conversion from local to global axis system is done
in the case of stiffness matrix and similar conversion associated with the loading
matrix and stress resultants, the knowledge of the transformation of axis system
is needed. In fact, it is of fundamental importance in structural analysis to obtain
transformations relating to the set of variables in one system to an equivalent set in
the second system.

4.3 Transformation of Variables

The discussion is initiated with the transformation of member coordinate axes.

4.3.1 Transformation of Member Coordinate Axes

Two sets of cartesian coordinate system are shown in Fig. 4.2. O, X , , Y , and Z ,
be the local axis system and O, X, Y and Z be the global axis system. Both sets are
right-handed. The direction cosines of X , with respect to X, Y and Z are l1 , m 1 and
n 1 . Similarly, direction cosines of Y , and Z , with respect to X, Y and Z are l2 , m 2 , n 2
and l3 , m 3 , n 3 respectively.
If a displacement vector {w , } has components wx, , w ,y and wz, along the local axis
system, the same vector denoted as {w} in the global axis system has components
4.3 Transformation of Variables 77

Fig. 4.2 Two sets of axis Y

Y′
system

X′

Z
Z′

wx , w y and wz respectively along that axis system, then it can be shown that
⎧ , ⎫
⎨ w x = l 1 w x + m 1 w y + n 1 wz ⎬
w , = l 2 w x + m 2 w y + n 2 wz (4.2)
⎩ ,y ⎭
wz = l 3 w x + m 3 w y + n 3 wz

Equation (4.2) written in matrix form becomes

⎧ ,⎫ ⎡ ⎤⎧ ⎫
⎨ wx ⎬ l1 m 1 n 1 ⎨ wx ⎬
w , = ⎣ l2 m 2 n 2 ⎦ w y
⎩ ,y ⎭ ⎩ ⎭
wz l3 m 3 n 3 wz

or

{w , } = [λ] {w} (4.3)

where [λ] is called rotation matrix. If both sets of reference axes are orthogonal, then

[λ] −1 = [λ] T (4.4)

which saves the time for computation, because the transpose of a matrix is easier to
obtain than its inverse (Fig. 4.3).
Usually, there are six displacement components at a point of an elastic structure—
three translations and three rotations. Just as the translational components discussed
above, the rotational components of the two axis systems can be related.
78 4 Direct Stiffness Method

Fig. 4.3 Generalized θy

displacements along local
θ′y
and global axis system
wy
w′y
θ′x
Y
Y′ w′x
X′

i X wx θx

Z
Z′

wz
w′z

θz

θ′z

4.3.2 Transformation of Member Displacement Matrix

A generalized displacement consisting of three components of translation and three

components of rotation, is given by
{ }T { }
X, = wx, w ,y wz, θx, θ y, θz,

or
( )
{ } { } . { }
X , T
= Δ i .. θ , i
,
(4.5)

The generalized displacement matrix is partitioned into translational and rotational

components and the subscript ‘i’ refers to the displacement parameters at point ‘i’
(generally they are, at the nodes i.e. joints). The generalized displacement component
in a local axis system can be related to the global axis system at node i by the following
relation:

⎧⎪{ ′}i ⎫⎪ ⎡[ ] [0] ⎤ ⎧⎪{ }i ⎫⎪

⎨ { ′} ⎬ = ⎢[0] [ ]⎥ ⎨ { } ⎬ (4.6)
⎩⎪ ⎪ ⎣⎢
i ⎭ ⎦⎥ ⎩⎪ i ⎭⎪

A straight member has two nodes at its two ends. If the axis system is considered
parallel at two nodes, then we have
4.3 Transformation of Variables 79
⎧ , ⎫ ⎡ ⎤⎧ ⎫
⎪
⎪ {Δ }1 ⎪
⎪ [λ] submatrices not ⎪
⎪ {Δ}1 ⎪⎪
⎨ , ⎬ ⎢ ⎥ ⎨ {θ }1 ⎬
{θ }1 ⎢ [λ] shown are zeros ⎥
= (4.7)
⎪
⎪ {Δ, } ⎪ ⎣ [λ] ⎦ ⎪ {Δ}2 ⎪
⎩ , 2⎪ ⎭ ⎪
⎩ ⎪
⎭
{θ }2 [λ] {θ }2

Subscript 1 refers to the displacement at node 1 and subscript 2 to that at node 2.

In compact form, Eq. (4.7) becomes
{ }
X , = [T ] {X } (4.8)

where [T ] is referred to as the member transformation matrix. [T ] is a square matrix

of order 12 × 12. It can be shown that

[T ] [T ]−1 = [T ] [T ]T = [I ]

i.e.

[T ]−1 = [T ]T (4.9)

It has already been seen that for the solution of a particular structure, the degrees of
freedom need not be six at each node and will be reduced depending on the problem
at hand. In that case, the size of [T ] is reduced.

4.3.3 Transformation of the Member Force Matrix

{ }
Let a member{ be}in equilibrium under the action of forces P , which results in
displacement X , in local coordinate system. Let {P} be the corresponding forces
in the global coordinate system.
For a virtual displacement, Eq. (4.8) becomes
{ }
δ X , = [T ] δ {X } (4.10)

Taking transpose of Eq. (4.10),

{ }T
δ X , = δ {X }T [T ]T (4.11)

Due to this virtual displacement, the virtual work done by the forces in local
coordinate system is
{ }T { , }
δ X, P

and the corresponding virtual work in the global coordinate system is

80 4 Direct Stiffness Method

δ {X }T {P}

Equating the virtual work in two coordinate systems,

{ }T { , }
δ X, P = δ {X }T {P} (4.12)

Combining Eqs. (4.11) and (4.12) yields

{ }
δ {X }T [T ]T P , = δ {X }T {P}

or
{ }
{P} = [T ]T P , (4.13)

Also,
{ }
P , = [T ] {P} (4.14)

As

[T ] [T ]T = [I ]

4.3.4 Transformation of the Member Stiffness Matrix

It is known that the element stiffness matrix in local coordinates is connected by the
following force–displacement relationship
[ ] [ ]
P, e
= [k]e X , e (4.15)
[ ]
X , e represents the displacements associated with that member. Premultiplying
both sides of Eq. (4.15) by [T ]T and substituting [X ]e from Eq. (4.10) into Eq. (4.15)
results
[ ]
[T ]T P , e = [T ]T [k]e [T ] [X ]e (4.16)

From Eqs. (4.14) and (4.16), we yields

[P]e = [T ]T [k]e [T ] [X ]e (4.17)

If the element stiffness matrix in global coordinates is defined as

4.4 Transformation of the Stiffness Matrix of the Member of a Truss 81

[K ]e = [T ]T [k]e [T ] (4.18)

then the force–displacement relation in the global coordinate system can be written
as

[P]e = [K ]e [X ]e (4.19)

Equations (4.13) and (4.18) give the relation of transforming the loading and the
stiffness matrix from, the local to the global axis system.
The derivation done so far will be illustrated with the elements of different
structures.

4.4 Transformation of the Stiffness Matrix of the Member

of a Truss

A typical member of a truss is shown in Fig. 4.4. As the member carries only the axial
force, the local axis system has X , -axis only. The local and global numbering of the
displacements has also been indicated in Fig. 4.4. The direction cosines of X , -axis
with X- and Y-axes are cos θ and sin θ respectively. The member is assumed to
have uniform axial rigidity AE and length L.
By elementary considerations, the stiffness matrix of the member in local
coordinates is

X′2
X4
X′ (Local axis)

2 X3

AE
L
Y X2
I Yi

θ
X 1 X1
(Global axis) Xi
X′1

Fig. 4.4 A typical member of a truss with a local and global axis system
82 4 Direct Stiffness Method
[ ]
EA 1 −1
[k]e = (4.20)
L −1 1

In order to obtain the transformation matrix, X 1, is to be related to X 1 , and X 2 and

X 2,are to be related to X 3 and X 4 . X 1 and X 2 are the displacements at node 1 parallel
to the global axis system. X 3 and X 4 are the corresponding displacements along the
global axis at node 2. Thus,

⎧ X1 ⎫
⎪⎧ X1′ ⎪⎫ ⎡l1 m1 0 0 ⎤ ⎪⎪ X 2 ⎪⎪
⎨X ′ ⎬ = ⎢0 0 l1 m1 ⎥⎥⎦ ⎨⎪ X 3 ⎬⎪
(4.21)
⎪⎩ 2 ⎪⎭ ⎢⎣
⎪⎩ X 4 ⎪⎭

where l1 = cos θ and m 1 = sin θ

Therefore,
[ ]
l1 m 1 0 0
[T ] =
0 0 l1 m 1

The element stiffness matrix in global coordinates as given in Eq. (4.18) can now
be calculated. For the truss, it is done as shown below:
⎡ ⎤
l1 0
[ ][ ]
⎢ m1 0 ⎥
[K ]e = ⎢ ⎥ E A 1 −1 l1 m 1 0 0
⎣0 l1 ⎦ L −1 1 0 0 l1 m 1
0 m1

or
⎡ ⎤
l12 Symmetrical
EA ⎢ 2
⎢ l1 m 1 m 1
⎥
⎥
[K ]e = ⎣ ⎦ (4.22)
L −l1 2
−l1 m 1 l1
2

−l1 m 1 −m 1 l1 m 1
2 2
m1

4.5 Transformation of the Stiffness Matrix of the Member

of a Rigid Frame

A typical member of a frame is shown in Fig. 4.5. The local axis system is shown
as X , , Y , and Z , and global axis system as X, Y and Z. The member is assumed to
have uniform flexural rigidity EI and uniform axial rigidity EA.
4.5 Transformation of the Stiffness Matrix of the Member of a Rigid Frame 83

Y′ X5

X′5 X′4

2 X4
X′4

X2 L
Y′ E A
Y EI, X6, X6′
X′2
Global 1 θ
Axis X1
X′1
X

X′

X2, X3′ , Z′
Z

Fig. 4.5 A typical member of a frame with local and global coordinate system

The axial deformation of the member will be considered. The other deformation
in the local axis system will be the displacement in the Y , -direction and rotation about
Z , -axis. The deformations in the global coordinates are the translations along X- and
Y-axes and rotation about Z , -axis. The direction cosines of X , with respect to X- and
Y-axes are cos θ and sin θ respectively. Six deformations—three at each node, are
numbered as shown in Fig. 4.5 for both the local and the global axis systems.
The stiffness matrix of the member of the frame in the local axis system is given
by
⎡ ⎤
EA
L
0 0 − ELA 0 0
⎢ 12E I 6E I
− 12E I 6E I ⎥
⎢ L3 L2
0 3 L2 ⎥
⎢ L
⎥
⎢ 4E I
0 − 6E I 2E I
⎥
[k]e = ⎢ L L2 L ⎥ (4.23)
⎢ EA
0 0 ⎥
⎢ L ⎥
⎣ Symmetrical 12E I
L3
− 6EL2
I ⎦
4E I
L

The relationship between local and global axis system is given by

84 4 Direct Stiffness Method

⎧ X1′ ⎫ ⎡ l1 m1 0 0 0 0 ⎤ ⎧ X1 ⎫
⎪ ′⎪ ⎢ 0 ⎥⎥ ⎪⎪ X 2 ⎪⎪
⎪ X 2 ⎪ ⎢ − m1 l1 0 0 0
⎪⎪ X 3′ ⎪⎪ ⎢ 0 0 1 0 0 0 ⎥ ⎪⎪ X 3 ⎪⎪
⎨ ⎬=⎢ ⎥⎨ ⎬ (4.24)
⎪ X 4′ ⎪ ⎢ 0 0 0 l1 m1 0⎥ ⎪ X 4 ⎪
⎪ X 5′ ⎪ ⎢ 0 0 0 − m1 l1 0⎥ ⎪ X 5 ⎪
⎪ ⎪ ⎢ ⎥⎪ ⎪
⎪⎩ X 6′ ⎪⎭ ⎢⎣ 0 0 0 0 0 1 ⎥⎦ ⎪⎩ X 6 ⎪⎭

where l1 = cos θ and m 1 = sin θ .

Therefore, for the member of the frame shown in Fig. 4.5, the transformation
matrix is given by
⎡ ⎤
l1 m1 00 0 0
⎢ −m 0⎥
⎢ 1 l1 00 0 ⎥
⎢ ⎥
⎢0 0 10 0 0⎥
[T ] = ⎢ ⎥ (4.25)
⎢0 0 0 l1 m1 0⎥
⎢ ⎥
⎣0 0 0 −m 1 l1 0⎦
0 0 00 0 1

The element stiffness match in the global axis system can now be calculated
according to Eq. (4.18). Instead of explicitly determining it, the necessary result may
be obtained by the matrix multiplication done by the computer.
If the axial force in the members of the rigid frame is neglected, there will be two
deformations in the local axis system at each node, i.e. X 1, and X 2, in node 1 and X 3,
and X 4, in node 2 (Fig. 4.6). However, global axis displacement labels remain the
same as the case with axial load.
⎡ 12E I ⎤
L3
Symmetrical
⎢ 6E2I 4E I ⎥
[k]e = ⎢ L L ⎥ (4.26)
⎣ − 3 − 6E2I 12E3 I
12E I ⎦
L L L
6E I
L 2
2E I
L
− 6E I
L2
4E I
L

and the transformation matrix for this case becomes

⎡ ⎤
−m 1 l1 0 0 0 0
⎢0 0 1 0 0 0⎥
[T ] = ⎢
⎣0
⎥ (4.27)
0 0 −m 1 l1 0⎦
0 0 0 0 0 1
4.6 Transformation of the Stiffness Matrix of the Member of a Grillage 85

X5
X′3

2 X4

X′4

X2
X′1
X6

1 θ
X1

X′2

X3

Fig. 4.6 A frame member neglecting axial deformation

4.6 Transformation of the Stiffness Matrix of the Member

of a Grillage

The plan of a typical member of a grid or grillage is shown in Fig. 4.7. The flexural and
torsional rigidities of the uniform members are EI and GJ respectively. In local axis
system, the deformations at each node are a vertical displacement and two rotational
displacements which are numbered in Fig. 4.7a. The deformation labels in global
coordinates are shown in Fig. 4.7b.
The stiffness matrix of the member of the grid in local axis system is given by
⎡ 12E I ⎤
L3
Symmetrical
⎢0 GJ ⎥
⎢ ⎥
⎢ 6E I L
4E I ⎥
⎢ L2 0 ⎥
[k]e = ⎢ 12E I L ⎥ (4.28)
⎢ − L3 0 − 6E I 12E I
⎥
⎢ L2 L3 ⎥
⎣0 − L 0
G J
0 GJ
L
⎦
6E I
L2
0 2E I
L
− 6EL2
I
0 4E I
L

The relationship of the deformation in the local and global axis system is given
by
86 4 Direct Stiffness Method

X′6
X′5

X4′ (+i ve axis downward)

GJ
EI,
L

X′3

X1′ (+i ve axis downward)

X′2

(a) Local axis system

X3

X2

X1′ (+i ve axis downward)

(b) Global axis

Fig. 4.7 A typical member of a grid

⎧ ⎫ ⎡ ⎤⎧ ⎫
⎪
⎪ X 1, ⎪
⎪ 1 0 0 0 0 0 ⎪
⎪ X1 ⎪⎪
⎪ X, ⎪
⎪ ⎪ ⎢0 ⎥ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎢ cos θ sin θ 0 0 0 ⎥⎪⎪ X ⎪
⎪
⎨ 2, ⎪
⎪ ⎬ ⎢ ⎪
⎥⎨ ⎪
2
⎬
X3 ⎢0 − sin θ cos θ 0 0 0 ⎥ X3
, =⎢ ⎥ (4.29)
⎪ X
⎪ 4⎪ ⎪ ⎢
⎪ 0 0 0 1 0 0 ⎥⎪ X ⎪
⎪
⎪ , ⎪
⎢ ⎥⎪⎪ 4⎪ ⎪
⎪
⎪ 5⎪
⎪ X ⎪
⎪ ⎣ 0 0 0 0 cos θ sin θ ⎦ ⎪
⎪
⎪
⎪ X5 ⎪⎪
⎪
⎩ ,⎭ ⎩ ⎪ ⎭
X6 0 0 0 0 − sin θ cos θ X6

In compact form, Eq. (4.29) is

{ }
X , = [T ] {X } (4.30)

Once the transformation matrix is obtained, the stiffness matrix of the member of
the global axis system can be obtained in the same way stated earlier.
4.7 Transformation of the Stiffness Matrix of the Member of a Space Frame 87

4.7 Transformation of the Stiffness Matrix of the Member

of a Space Frame

The possible displacements at the ends of a typical space frame member i are indicated
in Fig. 4.8a for local axis system and Fig. 4.8b for global axis system.
The member stiffness matrix of the space frame member in local axis system is
given by Eq. (4.31). In this equation, A x is the cross-sectional area of the member,
L is the length of the member, I x and Iz are the principal moments of inertia of the
cross section of the member with respect to ym and z m axes respectively.

11

Ym 8
10 Xm
7
5
j
i
2
9
1
4 i 12
3

Zm
(a) Local axis system

11

8
5
7 10
Y 2 j
i
9
X
i 1 4 12

Z 3

6
(b) Global axis system

Fig. 4.8 A typical member of a space frame

88 4 Direct Stiffness Method

⎡ EAx ⎤
⎢ L ⎥
⎢ ⎥
⎢ 0 12 EI z ⎥
⎢ L3 ⎥
⎢ ⎥
⎢ 0 12 EI y ⎥
0 Symmetrical
⎢ L3 ⎥
⎢ ⎥
⎢ 0 GI x ⎥
0 0
⎢ L ⎥
⎢ 6 EI y 4 EI y ⎥
⎢ 0 0 – 0 ⎥
⎢ L2 L ⎥
⎢ 6 EI z 4 EI z ⎥
⎢ 0 0 0 0 ⎥
⎢ L2 L2 ⎥
[k]e = ⎢ ⎥
EA EAx
⎢– x 0 0 0 0 0 ⎥
⎢ L L ⎥
⎢ 12 EI y 6 EI z 12 EI z ⎥
⎢ 0 – 0 0 0 – 0 ⎥
⎢ L3 L2 L3 ⎥
⎢ 12 EI y 6 EI y 12 EI y ⎥
⎢ 0 0 – 0 0 0 0 ⎥
⎢ L3 L2 L3 ⎥
⎢ GI GI x ⎥
⎢ 0 0 0 – x 0 0 0 0 0 ⎥
⎢ L L ⎥
⎢ 6 EI y 2 EI y 6 EI y 4 EI y ⎥
⎢ 0 0 – 0 0 0 0 0 ⎥
⎢ L2 L L2 L ⎥
⎢ 6 EI z 2 EI z 6 EI z 4 EI z ⎥
⎢ 0 0 0 0 0 – 0 0 0 ⎥
⎣ L2 L L2 L ⎦
(4.31)

If i and j, the two ends of the member has coordinates (xi , yi , z i ) and
(x j , y j , z j ), then the direction cosines of the member are given by

x j − xi y j − yi z j − zi
Cx = , Cy = and C z = (4.32)
L L L
The length L of the member is given by
/
L= (x j − xi )2 + (y j − yi )2 + (z j − z i )2 (4.33)

The rotation matrix [λ] (Eq. 4.3) is given by

⎡ ⎤
Cx C
/y Cz
⎢ − √C x C y C x2 + C y2 − √ y2 z 2 ⎥
C C
[λ] = ⎢
⎣ C x2 +C y2
⎥
C y +C z ⎦ (4.34)
−√ 2 2 0
Cz
√ 2 2
C x
C x +C z C x +C z

The transformation matrix of the space frame member can be formed accordingly.
The rotational matrix [λ] is valid for all positions of the member except when it
is vertical. [λ] matrix for this case is
4.8 Horizontally Circular Curved Beam Element 89
⎡ ⎤
0 Cy 0
[λ] = ⎣ − C y 0 0 ⎦ (4.35)
0 0 1

4.8 Horizontally Circular Curved Beam Element

Consider a structural member circularly curved in plan and is uniform. The member
will develop two moments P1 and P2 and a force P3 (the member is part of a grid
structure) as shown in Fig. 4.9. The end i is fixed, but the end j is free. Let us take an
infinitesimal arc length ds at an angle θ from ij subtending dθ at the centre.
Due to unit values of the forces at j the bending moments and torques developed
at ds is

M1 = sin θ, M2 = − cos θ, M3 = r sin θ

T1 = cos θ, T2 = sin θ, T3 = −r (1 − cos θ ) (4.36)

The flexibility coefficient (F j j )11 is obtained as [1]

( (
M12 T12
(F j j )11 = ds + ds (4.37)
E Iy G Ix
s s

or,

Fig. 4.9 Curved beam yj

element
P2
inθ xj
rs P1 φj
j

P3

zj
r(1
–c
os

y
) θ

x
θ
z
φ
dθ

c
90 4 Direct Stiffness Method

r r
(F j j )11 = (φ − sin φ cos φ) + (φ + sin φ cos φ) (4.38)
2E I y 2G Ix

Other flexibility coefficients can be similarly obtained. Each element of the flexi-
bility matrix consists of two parts—flexure and torsion, and each element is the sum
of the two.

[F j j ] = [F j j ] f lex + [F j j ]tor (4.39)

where
⎡ ⎤
φ − sin φ cos φ symmetrical
r ⎣ 2 ⎦
[F j j ] f lex = sin φ φ + sin φ cos φ
2E I y
r (φ − sin φ cos φ) −r sin φ
2
r (φ − sin φ cos φ)
2

(4.40a)
⎡ ⎤
φ + sin φ cos φ symmetrical
r ⎢ 2 ⎥
[F j j ]tor =
2G I x
⎣ sin φ φ + sin φ cos φ ⎦ (4.40b)
r (φ − 2 sin φ + sin φ cos φ) − r (2 − 2 cos φ − sin2 φ) r 2 (φ − sin φ cos φ)

Equations (4.40a) and (4.40b) are the matrices that indicate displacements in
the direction x j , y j and z j due to unit values corresponding to P1 , P2 and P3
respectively.
The stiffness matrix at the jth end can be obtained by inversion. Thus,

[k] j j = [F] −1
jj (4.41)

The stiffness matrix is to be expressed in structural directions by

[K j j ] = [R]T [k] j j [R] (4.42)

where
⎡ ⎤
cos φ j sin φ j 0
[R] j = ⎣ − sin φ j cos φ j 0 ⎦ (4.43)
0 0 1

Other parts of the member stiffness matrix are given by

[K i j ] = − [Ti j ] [K j j ] (4.44)

where
⎡ ⎤
1 0 yi j
[Ti j ] = ⎣ 0 1 − xi j ⎦ (4.45)
001
4.9 Overall Stiffness Matrix 91

Similarly,

[K ji ] = − [K ] j j [Ti j ]T (4.46)

and

[K i j ] = [Ti j ]T [K ] j j [Ti j ] (4.47)

4.9 Overall Stiffness Matrix

The stiffness matrix of an individual member of the structure forms the basic compo-
nent. Element stiffness matrices of all members are to be assembled to relate the form
and the displacement of the complete structure. The subsequent discussion is under-
taken to show how the contribution of the element stiffness matrix is taken care of
in the formation of the overall stiffness matrix, also known as the structure stiffness
matrix.
It may be noticed by looking into problems of Chap. 3 that the stiffness at a joint
is obtained by adding the stiffnesses of all the members meeting at the joint.
In the first step, the degrees of freedom of the structure are to be numbered, starting
with 1 and ending with NP, where NP is the total degrees of freedom. The restraints are
then numbered beyond NP. These numberings are referred to as degrees of freedom
corresponding to global numbering or simply the global degrees of freedom. If the
unknown displacements and the internal forces in the members are only to be deter-
mined, then all restraints can be given the number (NP + 1). This procedure will
save some storage space. However, in addition to the above two, if the reactions at
the supports are to be evaluated, then the restraints (i.e. the known displacements)
are to be numbered sequentially. The reaction aspect is put up till the next chapter. In
the meantime, the part of the stiffness matrix associated with the degrees of freedom
is discussed.
The force–displacement relationship for the complete structure is given by

[K ] [X ] = [P] (4.48)

Both [K] and [P] consist of contributions from individual element. Assembling
the contribution of element stiffness matrix of individual member to form an overall
stiffness matrix is now discussed. This is best demonstrated with an example.
Before further discussion is initiated, a revision of some of the terms is felt neces-
sary. The chapter has been started by introducing the concept of local axis system
and global axis system. We shall henceforth deal with two different numbering
in global axis system for the displacements as the member ends. The first set of
numbering is referred to as local numbering which will remain identical for every
member. This numbering refers to the typical displacement labels of the member. The
92 4 Direct Stiffness Method

members meet at a joint in the structure. The same member ends will have different
displacement labels when the total structure is concerned.
The continuous beam of Fig. 4.10 has four degrees of freedom-rotation at four
supports. The horizontal movement of the supports is ignored. In Fig. 4.10a, the
numbering from the left-hand side corresponding to global coordinates has been
indicated. In Fig. 4.10b, the local numbering of a typical member and, in Fig. 4.10c,
the positive direction of the displacements has been shown.
Calculation of the element stiffness matrix is next to be done. This is done for a
typical element on the basis of local axis system. For the problem at hand, the size
of the element stiffness matrix is 4 × 4. For the first element, they are, say,

Global 5162
Local 1234
⎡ ⎤
5 1 a11 a12 a13 a14
1 2 ⎢
⎢ a21 a22 a23 a24 ⎥
⎥
[K ]1 = (4.49)
6 3 ⎣ a31 a32 a33 a34 ⎦
2 4 a41 a42 a43 a44

Fig. 4.10 Local and global numbering of a continuous beam

4.9 Overall Stiffness Matrix 93

For members which are inclined and where local and global coordinates are not
coincident, the element stiffness matrix is to be evaluated on the basis of transforma-
tion as per Eq. (4.18). There is one-to-one correspondence between the joints of the
elements and that of the assemblage. The elements of the element stiffness matrix
of Eq. (4.49) should now be put in their proper place in the overall stiffness matrix.
Equation (4.32) indicates that local numbering 1 corresponds to global numbering
5 which means that a11 should occupy 5th row and 5th column (i.e. K 55 ) of the
overall stiffness matrix. Local member 2 corresponds to global number 1. Therefore
a12 , a21 , a22 will occupy the positions of K 51 , K 15 and K 11 of the overall stiffness
matrix. The proper location of all elements of Eq. (4.49) in the overall stiffness matrix
has been shown in Fig. 4.11.
The process is to be repeated for all the elements. It is to be borne in mind that
individual stiffness at a particular location in the overall stiffness matrix is to be
added. If the element stiffness matrices for element members 2 and 3 be then, the
overall stiffness matrix has been shown in Fig. 4.12. The steps to be followed in the
computer program have been shown in the flow chart of Fig. 4.13.

Global 6273
Local 1234
⎡ ⎤
6 1 b11 b12 b13 b14
2 2 ⎢
⎢ b21 b22 b23 b24 ⎥
⎥
[K ]2 =
7 3 ⎣ b31 b32 b33 b34 ⎦
3 4 b41 b42 b43 b44
Global 7384

Fig. 4.11 Assemblage into

global system of element of
Fig. 4.10
94 4 Direct Stiffness Method

Fig. 4.12 Overall stiffness matrix

Fig. 4.13 Flowchart of

assembly from the element
stiffness matrix to the overall
stiffness matrix
4.10 Boundary Conditions 95

Local 1234
⎡ ⎤
7 1 c11 c12 c13 c14
3 2⎢⎢ c21 c22 c23 c24 ⎥
⎥
[K ]3 = (4.50)
8 3 ⎣ c31 c32 c33 c34 ⎦
4 4 c41 c42 c43 c44

The overall loading matrix can be similarly formulated as an assemblage of the

contribution from the element loading matrix. For the formation of the overall loading
matrix, however, the load applied to the joint directly is to be taken care of.

4.10 Boundary Conditions

One practice of labelling is to put the number corresponding to the unknown displace-
ments first and then, the restrained displacements. This facilitates taking into account
the boundary conditions in an effective manner.
There are a number of ways of catering the boundary conditions. A few of the
approaches have been discussed here.
If the overall stiffness matrix is to be formed in half-band form, then the numbering
of displacements should be such that the bandwidth is minimum. For this case
displacement labels are put in a systematic manner irrespective of whether the joint
displacements are unknowns or restraints. The restrained displacements are taken
care of by assigning a high value (say 1025 ) in the diagonal element corresponding
to that displacement.
If the unknown displacements are labelled first, then the matrix operations can be
restricted up to unknown displacement labels and beyond that, the overall stiffness
matrix may be ignored.
Another approach to catering the boundary conditions will be to make the diagonal
element corresponding to that displacement label equal to unity, while all other
elements in the corresponding rows and columns as also the loading vector in that
row, are put equal to zero.
Another approach is not to assemble in the overall stiffness matrix, the contri-
butions from the nodes which have restrained displacements, i.e. row and column
corresponding to fixed displacement are eliminated.

4.10.1 Boundary Conditions Corresponding to Skewed

Supports

If the constraints applied at the nodes are in the direction of global axis system,
they can be taken care of as stated above. The same concept can be extended to the
96 4 Direct Stiffness Method

Fig. 4.14 Skewed support

problems where the boundary condition corresponding to a skew support is to be

incorporated.
A beam element is of Fig. 4.14 has a skewed roller support. Therefore, in the
overall formulation, the vertical displacement corresponding to the roller support is
zero. At the ith node, where the roller support is placed, displacement from the global
axis system is transformed to skew coordinates.

[X i ]G = [λ]i [X ]skew (4.51)

[λ]i is a simple point transformation involving the direction cosines which relate
to the global and skewed axis system. The transformation matrix for that clement
then becomes
⎡ ⎤
[I ]
⎢ [I ] ⎥
⎢ ⎥
⎢ [λ]i ⎥
⎢ ⎥
⎢ ⎥
⎢ . ⎥
[T ] = ⎢ ⎥ (4.52)
⎢ . ⎥
⎢ ⎥
⎢ . ⎥
⎢ ⎥
⎣ . ⎦
[I ]

where [I] is an identity matrix. There is one submatrix [λ] i in [T ] for each node with
skewed constraints.
The resulting stiffness matrix then becomes

[K ] = [T ]T [k]e [T ] (4.53)

This transformation is done at the individual stiffness matrix generation level, i.e.
before they are assembled to form the overall stiffness matrix.
After the unknown displacements are known for the overall solution, the skewed
displacements are to be converted back by global coordinates by using Eq. (4.51),
for obtaining the necessary stresses in the elements.
4.12 Computer Program for the Truss Analysis by the Direct Stiffness Method 97

The incorporation of the boundary conditions has been discussed in detail in the
computer programs and the examples presented later.

4.11 Computation of Internal Forces

The solution of Eq. (4.48) will yield the displacements at all joints. The end actions
of the members are next to be determined. This is possible on the basis of Eq. (3.12).

[P M]e = [P M F]e + [P M X ]e [X ]e (4.54)

or,

[P M] = [P M F]e + [k]e [X ]e (4.55)

The only change to be noted is that [PMX] is replaced by [k]e . [P M F]e matrix
for each element is given as an input to the computer.

4.12 Computer Program for the Truss Analysis

by the Direct Stiffness Method

Table 4.1 gives the listing of the computer program in ‘C’ developed for the analysis
of a plane truss by the direct stiffness method. It is also given in the CD (file name:
dir_tr.c) attached with this book. Different notations used and different steps followed
have been indicated with comment statement in the computer program. The computer
program for the same in FORTRAN (file name: DIR_TR.F) is also given in the CD.
Input data required for the computer program consists of the material and
geometric properties of the members and their global numbering. The x and y coor-
dinates of one end of the member with respect to the other end are specified as input.
The data of the loading matrix is also fed into the computer.

Example 4.1 The truss of Fig. 4.15 is to be analysed with the help of the computer
program in Table 4.1. The horizontal and vertical members have area 1 and the
inclined members 1.414. The loading has been shown in Fig. 4.15.

Global numbering has been indicated in Fig. 4.15a. For this problem, NT = 9, NP
= 8, NF = 10 and NLC = 1. It is assumed that E = 1. It may be noted that the total
degree of freedom for this problem is 8. All the restraints have been assigned the
number 9. The preparation of the input data corresponding to one typical member is
explained here.
The inclined member 4 is considered. A is assumed as the initial point and B as
the terminal point. The data for this member then is
98 4 Direct Stiffness Method

Table 4.1 Computer program for the analysis of plane truss by direct stiffness method
#include <stdio.h>
#include <math.h>
#include <dos.h>
#include <conio.h>
// FUNCTION INVER FOR MATRIX INVERSION
void inver(float a[][30],int n){
// VARIABLE DECLARATION
float max,temp,pivot;
int col,i,j,flag=1,index[20];
for(i=0;i<n;i++) index[i] = 0;
while(flag != 0){
max=-1.0;
for(i=0;i<n;i++){
if(index[i]==0){
temp = fabs(a[i][i]);
if(temp > max){
col=i;
max=temp;
}
}
}
if(max > 0){
index[col] = 1;
pivot = a[col][col];
a[col][col] = 1.0;
pivot = 1/pivot;
for(j=0;j<n;j++) a[col][j] *= pivot;
for(i=0;i<n;i++){
if((i-col) != 0){
temp = a[i][col];
a[i][col] = 0.0;
for(j=0;j<n;j++) a[i][j]
-= a[col][j] * temp;
}
}
}
else flag=0;
}
}
void planetruss()
{
/* NP TOTAL DEGREES OF FREEDOM
NT TOTAL DEGREES OF FREEDOM INCLUDING RESTRAINTS
NF NUMBER OF MEMBERS
NLC NUMBER OF LOADING CONDITIONS
E YOUNG’S MODULUS
STIFO THE OVERALL STIFFNESS MATRIX
*/
float stifo[30][30],x[30][4],fi[50][4],load[30][4];
float v[50],h[50],aa[50],xcos,xsin,xl,e,dirc,dirs,dircs;
int np1,np2,np3,np4,i,j,np,nt,nf,nlc,k,npg[40][4];
FILE *fpt1,*fpt2;

(continued)
4.12 Computer Program for the Truss Analysis by the Direct Stiffness Method 99

Table 4.1 (continued)

fpt1=fopen(“DIR_TR_IN”,”r”);
fpt2=fopen(“DIR_TR_OUT”,”w”);
fprintf(fpt2,”\tTruss Analysis\n”);
fscanf(fpt1,”%d%d%d%d%f”,&nt,&np,&nf,&nlc,&e);
fprintf(fpt2,”\tNT= %d\tNP = %d\tNF = %d\tNLC = %d\tE =
%.3f\n”,nt,np,nf,nlc,e);
// INITIALIZE STIFO
for(i=0;i<nt;i++)
{
for(j=0;j<nt;j++) stifo[i][j] = 0.0;
}
/* CALCULATE THE ELEMENT STIFFNESS MATRIX AND PUT THEM IN
THEIR PROPER LOCATION IN STIFO
NPG[I][1],NPG[I][2],NPG[I][3],NPG[I][4] ARE THE GLOBAL
NUMBERINGS CORRESPONDING TO THE LOCAL NUMBERING 1,2,3,4
H[I],V[I] ARE THE PROJECTIONS ALONG X AND Y AXIS OF THE
MEMBER
AA[I] IS THE CROSS-SECTIONAL AREA OF THE MEMBER */
fprintf(fpt2,
“ MEMBER NP1 NP2 NP3 NP4 H V A L COS SIN\n”);
for(i=0;i<nf;i++)
{
for(j=0;j<4;j++) fscanf(fpt1,”%d”,&npg[i][j]);
fscanf(fpt1,”%f%f%f”,&h[i],&v[i],&aa[i]);
xl = sqrt((h[i]*h[i])+(v[i]*v[i]));
xcos = h[i] / xl;
xsin = v[i] / xl;
np1 = npg[i][0]-1;
np2 = npg[i][1]-1;
np3 = npg[i][2]-1;
np4 = npg[i][3]-1;
fprintf(fpt2,”%3d%3d%3d%3d%3d”,i+1,np1,np2,np3,np4);
fprintf(fpt2,”%7.3f%7.3f%7.3f%7.3f%7.3f%7.3f\n”,h[i],
v[i],aa[i],xl,xcos,xsin);
dirc = e * aa[i] * xcos * xcos / xl;
dirs = e * aa[i] * xsin * xsin / xl;
dircs = e * aa[i] * xcos * xsin / xl;
stifo[np1][np1] = stifo[np1][np1] + dirc;
stifo[np1][np2] = stifo[np1][np2] + dircs;
stifo[np1][np3] = stifo[np1][np3] - dirc;
stifo[np1][np4] = stifo[np1][np4] - dircs;
stifo[np2][np2] = stifo[np2][np2] + dirs;
stifo[np2][np3] = stifo[np2][np3] - dircs;
stifo[np2][np4] = stifo[np2][np4] - dirs;
stifo[np3][np3] = stifo[np3][np3] + dirc;
stifo[np3][np4] = stifo[np3][np4] + dircs;
stifo[np4][np4] = stifo[np4][np4] + dirs;
stifo[np2][np1] = stifo[np1][np2];
stifo[np3][np1] = stifo[np1][np3];
stifo[np4][np1] = stifo[np1][np4];
stifo[np3][np2] = stifo[np2][np3];
stifo[np4][np2] = stifo[np2][np4];
stifo[np4][np3] = stifo[np3][np4];

(continued)
100 4 Direct Stiffness Method

Table 4.1 (continued)

}
for(j=0;j<nlc;j++)
{
for(i=0;i<np;i++) fscanf(fpt1,”%f”,&load[i][j]);
}
// OUTPUT THE STIFFNESS MATRIX
fprintf(fpt2,”\tThe Stiffness Matrix\n”);
for(i=0;i<np;i++)
{
for(j=0;j<np;j++) {
fprintf(fpt2,”%14.5e”,stifo[i][j]);
if((j+1)%5==0)
fprintf(fpt2,”\n”);
}
fprintf(fpt2,”\n”);
}
// INVERT MATRIX STIFO
inver(stifo,np);
// OUTPUT THE LOAD MATRIX
fprintf(fpt2,”\tThe Load Matrix\n”);
for(j=0;j<nlc;j++)
{
for(i=0;i<np;i++) {
fprintf(fpt2,”%14.5e”,load[i][j]);
if((i+1)%5==0)
fprintf(fpt2,”\n”);
}
fprintf(fpt2,”\n”);
}
// CALCULATE THE JOINT DISPLACEMNTS
for(i=0;i<np;i++)
{
for(j=0;j<nlc;j++)
{
x[i][j] = 0.0;
for(k=0;k<np;k++) x[i][j] += stifo[i][k] * load[k][j];
}
}
// OUTPUT MATRIX X
fprintf(fpt2,”\tThe X Matrix\n”);
for(j=0;j<nlc;j++)
{
for(i=0;i<np;i++){
fprintf(fpt2,”%14.5e”,x[i][j]);
if((i+1)%5==0)
fprintf(fpt2,”\n”);
}
fprintf(fpt2,”\n”);
}
// RESTRAINT DISPLACEMENTS ARE MADE ZERO IN NEXT 4 STATEMENTS
for(i=np;i<nt;i++)
{
for(j=0;j<nlc;j++) x[i][j] = 0.0;
}

(continued)
4.12 Computer Program for the Truss Analysis by the Direct Stiffness Method 101

Table 4.1 (continued)

// COMPUTE INTERNAL FORCES IN THE MEMBERS OF THE TRUSS
for(i=0;i<nf;i++)
{
xl = sqrt(h[i] * h[i] + v[i] * v[i]);
xcos = h[i] / xl;
xsin = v[i] / xl;
np1 = npg[i][0]-1;
np2 = npg[i][1]-1;
np3 = npg[i][2]-1;
np4 = npg[i][3]-1;
for(j=0;j<nlc;j++)
fi[i][j]=e*aa[i]/xl*(xcos*(x[np3][j]-x[np1][j])+xsin*
(x[np4][j]-x[np2][j]));

}
// OUTPUT MATRIX FI
fprintf(fpt2,”\tThe FI Matrix\n”);
for(j=0;j<nlc;j++)
{
for(i=0;i<nf;i++) {
fprintf(fpt2,”%14.5e”,fi[i][j]);
if((i+1)%5==0)
fprintf(fpt2,”\n”);
}
fprintf(fpt2,”\n”);
}

fclose(fpt1);
fclose(fpt2);

void main()
{planetruss();
}

1 2 7 8 1.0 − 1.0 1.414

The local numbering for this element has been shown as a dotted line in the figure.
B also could have been assumed as the initial point and A as the terminal point, Then
the data will be

7 8 1 2 − 1.0 1.0 1.414

The input data for the problem of Fig. 4.13 is given in Table 4.2. It may be noted
that two loading conditions have been considered.
102 4 Direct Stiffness Method

Fig. 4.15 Truss for solution by direct stiffness method

4.13 Computer Program for the Frame Analysis by Direct Stiffness Method 103

Table 4.2 Input data of truss

9 8 10 2 1.0
1 2 3 4 1.0 0.0 1.0
1 2 5 6 0.0 -1.0 1.0
5 6 9 9 0.0 -1.0 1.0
1 2 7 8 1.0 -1.0 1.414
5 6 3 4 1.0 1.0 1.414
5 6 7 8 1.0 0.0 1.0
5 6 9 9 1.0 -1.0 1.414
9 9 7 8 1.0 1.0 1.414
3 4 7 8 0.0 -1.0 1.0
7 8 9 9 0.0 -1.0 1.0
5.0 0.0 0.0 5.0 5.0 0.0 0.0 0.0
0.0 -10.0 0.0 -10.0 0.0 0.0 0.0 0.0

The computer output for the problem is given in Table 4.3. The forces in the
members can be checked with those obtained by Wang’s displacement method1 for
the same problem [2, 3]. They have been found to be of the same order.

4.13 Computer Program for the Frame Analysis by Direct

Stiffness Method

The computer program for the plane frame analysis by the direct stiffness method in
both the languages—C (file name: dir_fr.c) and FORTRAN (DIR_FR.F), are given
in the attached CD. Notations have been explained in the listing of the computer
program for both cases. Different steps followed in the computer code have been
explained with ‘comment’ statements.
In comparison to the computer program of the truss presented in the earlier section,
the present one has several new features. For the first time, we are encountering a
computer program where the overall stiffness matrix is stored in half-band form. The
displacement matrix has been obtained from a solution of simultaneous equations.
This is done by subroutine GAUSS.
The main program contains a number of subroutines—a subroutine for calculating
the element stiffness matrix in local coordinate system, a subroutine for calculating
the transformation matrix and another subroutine for computing the member forces.
The input data (Table 4.4) is simple to prepare. It consists of pertinent local and global
information. In addition to the material and geometric properties of the members,
the information regarding total degrees of freedom, total number of nodes, number
of restraints, number of loading conditions, etc., must be specified. This computer
program does not generate the loading matrix which, however, is fed directly as the
input data. The computer output has been indicated in Table 4.5.

1 It is a method where the stiffness matrix is formed in a different manner and this has not been
included in this book.
104 4 Direct Stiffness Method

Table 4.3 Output of the truss by direct stiffness method

Truss Analysis
NT= 9 NP = 8 NF = 10 NLC = 2 E = 1.000
MEMBER NP1 NP2 NP3 NP4 H V A L COS SIN
1 0 1 2 3 1.000 0.000 1.000 1.000 1.000 0.000
2 0 1 4 5 0.000 -1.000 1.000 1.000 0.000 -1.000
3 4 5 8 8 0.000 -1.000 1.000 1.000 0.000 -1.000
4 0 1 6 7 1.000 -1.000 1.414 1.414 0.707 -0.707
5 4 5 2 3 1.000 1.000 1.414 1.414 0.707 0.707
6 4 5 6 7 1.000 0.000 1.000 1.000 1.000 0.000
7 4 5 8 8 1.000 -1.000 1.414 1.414 0.707 -0.707
8 8 8 6 7 1.000 1.000 1.414 1.414 0.707 0.707
9 2 3 6 7 0.000 -1.000 1.000 1.000 0.000 -1.000
10 6 7 8 8 0.000 -1.000 1.000 1.000 0.000 -1.000
The Stiffness Matrix
1.49992e+000 -4.99925e-001 -1.00000e+000 0.00000e+000 0.00000e+000
0.00000e+000 -4.99925e-001 4.99925e-001
-4.99925e-001 1.49992e+000 0.00000e+000 0.00000e+000 0.00000e+000
-1.00000e+000 4.99925e-001 -4.99925e-001
-1.00000e+000 0.00000e+000 1.49992e+000 4.99925e-001 -4.99925e-001
-4.99925e-001 0.00000e+000 0.00000e+000
0.00000e+000 0.00000e+000 4.99925e-001 1.49992e+000 -4.99925e-001
-4.99925e-001 0.00000e+000 -1.00000e+000
0.00000e+000 0.00000e+000 -4.99925e-001 -4.99925e-001 1.99985e+000
0.00000e+000 -1.00000e+000 0.00000e+000
0.00000e+000 -1.00000e+000 -4.99925e-001 -4.99925e-001 0.00000e+000
2.99985e+000 0.00000e+000 0.00000e+000
-4.99925e-001 4.99925e-001 0.00000e+000 0.00000e+000 -1.00000e+000
0.00000e+000 1.99985e+000 0.00000e+000
4.99925e-001 -4.99925e-001 0.00000e+000 -1.00000e+000 0.00000e+000
0.00000e+000 0.00000e+000 2.99985e+000
The Load Matrix
5.00000e+000 0.00000e+000 0.00000e+000 5.00000e+000 5.00000e+000
0.00000e+000 0.00000e+000 0.00000e+000
0.00000e+000 -1.00000e+001 0.00000e+000 -1.00000e+001 0.00000e+000
0.00000e+000 0.00000e+000 0.00000e+000
The X Matrix
3.87295e+001 1.21818e+001 3.62750e+001 -2.81816e+000 1.89106e+001
9.63640e+000 1.60924e+001 -5.36360e+000
-1.09085e+000 -1.52731e+001 1.09083e+000 -1.52731e+001 -2.36347e+000
-7.45475e+000 2.36347e+000 -7.45474e+000
The FI Matrix
-2.45456e+000 2.54545e+000 9.63640e+000 -3.59981e+000 3.47126e+000
-2.81816e+000 -6.55685e+000 7.58528e+000 2.54545e+000 -5.36360e+000

2.18168e+000 -7.81832e+000 -7.45475e+000 -3.08536e+000 -3.08536e+000

4.72693e+000 -3.59953e+000 -3.59953e+000 -7.81832e+000 -7.45474e+000

Example 4.2 The frame shown in Fig. 4.16 is to be analysed. For this problem, NT
= 6, NLC = 1, NNODE = 4, NMEM = 3, NN = 2, NBH = 7. It may be noted
that all the restraints have been given a displacement label 7. For a small problem
of the type shown in Fig. 4.16, it is not possible to show the advantages of storing
the overall stiffness matrix in half-band form. For bigger size problems, storing the
overall stiffness matrix in half-band form will be obvious.
4.13 Computer Program for the Frame Analysis by Direct Stiffness Method 105

Table 4.4 Input data of the plane frame problem

4 3 3 2 6 6 1 1
0.0 0.0 5.0 8.0 13.0 8.0 18.0 0.0
3 1 1 2 2 4
7 7 7 1 2 3 1 2 3 4 5 6 4 5 6 7 7 7
7 200000000.
0.00036 0.00036 0.00036
0.6 0.6 0.6
0.0 -80.0 106.7 0.0 -80.0 -106.7
0 0 0 0 0 0
0.0 80.0 -106.7 0.0 80.0 106.7
0 0 0 0 0 0

Table 4.5 Computer output of the plane frame problem

NNODE = 4 NMEM = 3 NFREE = 3 NN = 2 NPM = 6 NT = 6 NLC = 1
The Coordinates of the Nodes
0.0000 0.0000 5.0000 8.0000
13.0000 8.0000 18.0000 0.0000
Node Numbering
3 1 1 2 2 4
The memberwise NPG Values
Member No. 1
7 7 7 1 2 3
Member No. 2
1 2 3 4 5 6
Member No. 3
4 5 6 7 7 7
NBH = 7 E = 2e+008
Moment of Inertia of Members
3.60000e-004 3.60000e-004 3.60000e-004
Cross-Sectional Areas
6.00000e-001 6.00000e-001 6.00000e-001
The OSTIF Matrix
9.276e+006 9.225e+006 -5.088e+003 -3.574e+006 -5.716e+006
-4.116e+003 0.000e+000 1.131e+007 4.152e+003 -5.716e+006
-9.147e+006 2.573e+003 -5.667e+006 0.000e+000 4.940e+004
4.116e+003 -2.573e+003 1.526e+004 8.826e+003 0.000e+000
0.000e+000 9.276e+006 2.208e+006 5.088e+003 -2.192e+006
0.000e+000 0.000e+000 0.000e+000 1.131e+007 -9.935e+002
1.351e+006 0.000e+000 0.000e+000 0.000e+000 0.000e+000
4.940e+004 6.883e+003 0.000e+000 0.000e+000 0.000e+000
0.000e+000 0.000e+000 1.575e+007 0.000e+000 0.000e+000
0.000e+000 0.000e+000 0.000e+000 0.000e+000
The LOAD Matrix
0.00000 -80.00000 106.70000 0.00000 -80.00000-106.70000
The DISPLACEMENT Matrix
0.53122 -0.86328 0.10734 -0.23611 -0.38370 0.07050
Load condition: 1
Member end forces of the member: 1
-152.044 -47.223-866.724 152.044 47.223 145.897
Member end forces of the member: 2
89.066 12.082-145.898 -89.066 147.918-494.837
Member end forces of the member: 3
110.253 21.265 494.837-110.253 -21.265-170.237
106 4 Direct Stiffness Method

Fig. 4.16 Frame for solution by direct stiffness method

The input data for the problem is given in Table 4.4. The computer output has
been given in Table 4.5.

4.14 Computer Program for the Grillage Analysis

by the Direct Stiffness Method

The computer program for the grillage analysis by the direct stiffness method in
both the languages—C and FORTRAN are given in the attached CD. Notations have
been explained in the listing of the computer program for both cases. Different steps
followed in the computer code have been explained with ‘comment’ statements.
In comparison to the computer program of the truss presented in the earlier section,
the present one has several new features. Much like the case of the frame analysis,
the overall stiffness matrix is stored in half-band form. The displacement matrix has
been obtained from a solution of simultaneous equations.
The main program contains a number of subroutines—a subroutine for calculating
the element stiffness matrix in local coordinate system, a subroutine for calculating
the transformation matrix and another subroutine for computing the member forces.
The input data (Table 4.6) is simple to prepare. It consists of pertinent local and global
information. In addition to the material and geometric properties of the members,
the information regarding, total degrees of freedom, total number of nodes, number
of restraints, number of loading conditions, etc., must be specified. This computer
program does not generate the loading matrix which, however, is fed directly as the
input data. The computer output has been indicated in Table 4.7.
The input data consists of among others, member geometries, their global
numbering and the geometric properties. A code LDIS is used as an input. This
4.14 Computer Program for the Grillage Analysis by the Direct Stiffness Method 107

Table 4.6 Input data for the grillage problem

4 3 1 2 6 4 0 0
1. .8 0.
1. 0. 0. 1. 1. 0 4 4 4 1 2 3 0. 0.
0. 1. 0. 1. 1. 0 1 2 3 4 4 4 0. 0.
0.5 -0.125 0.
0. 0. 0. 0. 0. 0.
-0.5 0. -0.125 -0.5 0.0 0.125

Table 4.7 Computer output of the grillage problem

Direct Stiffness Method of Analysis of Grillage
NT NP NLC NELM NPM NBH LGG LDIS
4 3 1 2 6 4 0 0
E = 1 G = 0.8 FF = 0
XX = 1.000 XY = 0.000 AA = 0.000 XI = 1.000 XJ = 1.000
SPDR = 0.000 Q = 0.000
NPG Values
4 4 4 1 2 3 LUND = 0
XX = 0.000 XY = 1.000 AA = 0.000 XI = 1.000 XJ = 1.000
SPDR = 0.000 Q = 0.000
NPG Values
1 2 3 4 4 4 LUND = 0
The Overall Stiffness Matrix
2.40000e+001 -6.00000e+000 -6.00000e+000 -3.60000e+001
4.80000e+000 0.00000e+000 7.20000e+000 0.00000e+000
4.80000e+000 7.20000e+000 0.00000e+000 0.00000e+000
1.00000e+025 0.00000e+000 0.00000e+000 0.00000e+000
The Overall Loading Matrix
Loading case: 1
5.00000e-001 -1.25000e-001 0.00000e+000
The Displacement Matrix
0.0381944 0.0217014 0.0477431
Stress in Member Number = 1
-1.7188e-001 -1.7361e-002 -1.3368e-001 1.7188e-001 1.7361e-002
-3.8194e-002
Stress in Member Number = 2
-1.7188e-001 3.8194e-002 1.7361e-002 -8.2813e-001 -3.8194e-002
3.1076e-001

takes care of the uniform load present in a member. For that member, the computer
program calculates the element loading matrix and puts it in the proper location of
the overall loading matrix. The computer program can cater shear deformation of the
member based on LGG value. If LGG = 0, the shear deformation is not included.

Example 4.3 The grid of Fig. 4.17 is to be analysed. It has been assumed that E =
1, I = 1, L = 1, P = 1, G = 0.8, J = 1.0 and A = 1. For this problem, NT = 4, NP
= 3, NLC = 1, NELM = 2, NPM = 6 and NBH = 4. All the restraints have been
given a displacement label 4. Like the frame problem of the previous section, this
problem is also of small size. As such the advantages of storing the overall stiffness
matrix in half-band form cannot be felt. Shear deformation has not been considered
in this problem.
108 4 Direct Stiffness Method

Fig. 4.17 Grillage for solution by the direct stiffness method

The displacement labels for a typical member corresponding to local coordinates

are shown in Fig. 4.17c. The data corresponding to item No. 3 of Table 4.6 for
member no. 1 is

1.0 0. 0. 1. 1. 4. 4. 4. 1 2 3 0 0

The input data for the problem is given in Table 4.6. The computer output for the
problem is given in Table 4.7. The values are nearly identical to those obtained from
Wang’s displacement method.
Exercise 4
4.14 Computer Program for the Grillage Analysis by the Direct Stiffness Method 109

1. Write a computer program for the solution of a plane frame by the direct stiffness
method. The computer program is to have the flexibility of considering both with
or without axial force.
2. Prepare the input data for the plane truss of Fig. below according to the computer
program of Table 4.1. Assume E =1.0, the areas of the members of top chord
and bottom chord are 2 units and that of all internal members 1.

Prob. 4.2

3. Prepare the input data for the grillage according to the program in Table 4.4.
Assume EI = 1.0 and GJ = 0.8 for all members. The shear deformation is not
to be included. The loads acting have been indicated in Prob. 4.3.

Prob. 4.3

4. With the help of computer programs find out the forces in the members of the
trusses shown. Cross-sectional areas of the outer members = 2 units and those
of internal members = 1.
110 4 Direct Stiffness Method

Prob. 4.4

5. Solve the plane frames by the direct stiffness method; neglect axial force for (a)
and (b), but incorporate axial force for (c). Assume EA = 0.1 El for (c).

P
0.3LL 0.2L I I 60kN
20 kN/m II I 80kN
2m2m

2EI 2.5I 2.5I

2.5I
2EI EI L
II I
3m

(a)
4m 4m 4m

(b)

40kN 40kN

200kN
60kN

o
75

L 0.75L

(c)

Prob. 4.5
References and Suggested Readings 111

6. Determine the member force for the grillages shown. Assume EI =1.25×109
kNm2 , GJ=1×109 kNmm2 .

Prob. 4.6

References and Suggested Readings

1. J.M. Gere, W. Weaver Jr., Analysis of Framed Structures, 2nd edn. (D. Van Nostrand, 1982)
2. C.K. Wang, Matrix Methods of Structural Analysis, International Textbook (Pasadena, USA,
1970)
3. C.K. Wang, Introductory Structural Analysis with Matrix Method (Prentice Hall, 1973)
4. W. McGuire, R.H. Gallagher, R.D. Ziemian, Matrix Structural Analysis, 2nd edn. (Wiley, USA,
2000)
Chapter 5
Substructure Technique for the Analysis
of Structural Systems

5.1 Introduction

The analysis of structural problems by the stiffness matrix method results in the
following system of simultaneous equations

[K ] [X ] = [P] (5.1)

A major part of the computer time (CPU time) is involved in the solution of
Eq. (5.1). The direct method of generating the structure stiffness matrix as a solid
matrix is suited only to problems of small size. For big size problems, advantage is
taken of the fact that the structure stiffness matrix [K] is symmetric and banded. [K]
is stored in half-band and the solution of simultaneous equations is obtained from
Gauss’s method for forward elimination and then backward substitution. This half-
band storage of [K] is a preferred approach for problems where all storage remains
within the core memory. The storage can be made more economical by storing the
half-band of [K] in skyline form.
For large-sized problems where half-band solution no longer becomes economic,
more sophisticated and specialized techniques are resorted to. Some of them are
the frontal solution technique [1, 2], sparse matrix techniques [3] and substructure
technique [4–6].
There are different approaches to the analysis of structural systems using substruc-
ture technique. In all of them, the total structure is divided into a number of substruc-
tures and the stiffness matrix of the substructure is expressed in terms of its boundary
displacements. The overall stiffness matrix of the structure is the assembly of these
reduced stiffness matrices. By dividing the total structure into substructures, the
number of equations that may be solved at any one time for the analysis of the total
system is considerably reduced.

© The Author(s) 2022 113

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_5
114 5 Substructure Technique for the Analysis of Structural Systems

5.2 Basic Concepts

The basic concept is explained with the plane truss in Fig. 5.1. To start with, the struc-
ture is to be conveniently partitioned into smaller structural systems. Each portion
of these partitioned structural systems is known as a substructure. Each substructure
consists of a group of elements.
Each substructure is analysed first by assuming the boundaries between the
substructures as restrained. This can be carried out by the direct stiffness method
discussed in the previous chapter considering each substructure as if it were a
complete system in itself. As the displacements at the substructure boundaries are
artificially constrained, joint displacements from this analysis will not represent the
true values of the displacements at the joints. Therefore, these displacements need
to be corrected. This correction is obtained from the analysis of the total structure
which is based on the displacements of the artificially restrained boundaries of the
substructure. Once the actual displacements of all the joints of the complete structure
have been evaluated, the member forces can then be determined.
The main advantage of analysing the structure using substructure technique can
be seen with respect to the plane truss in Fig. 5.1. By the direct stiffness method, the
number of unknown displacements for the truss is 24, as there are 12 free joints. When
analysed by the substructure technique according to the substructures in Fig. 5.1, a

Fig. 5.1 A plane truss with substructures

5.3 Direct Stiffness Method Restated 115

maximum number of unknown displacements to be catered at any one time is 8.

For substructures 1, 2 and 3, the unrestrained components of joint displacements
are 8, 4 and 4 respectively. For the analysis of the whole structure, eight displace-
ments are there corresponding to the artificial constraints. Thus, in comparison to
the direct stiffness method, the number of unknowns to be dealt with at any one time
is considerably lower.
While dividing a structure into a number of substructures, care should be taken
to partition the structure. Number of unknowns for the individual substructure and
those of the total structure considering the unknowns at the substructure joints should
preferably be of the same order for economical solution. Therefore, the partitioning
of the total structure into substructure must be judiciously done.

5.3 Direct Stiffness Method Restated

Equation (4.48) is rewritten with slight modification

[K ] c [X ] c = [P] c + [R] c (5.2)

where

[K ] c is the complete structure stiffness matrix (including all restraints),

[X ] c is the complete joint displacement matrix,
[P] c is the complete joint load matrix,
[R] c is the complete support reaction matrix.
While writing the elements of Eq. (5.2), it may be noted that the unknown displace-
ments are labelled first, then the boundary displacements [K ] c are partitioned into
groups associated with unrestrained and restrained joint displacements, as follows:

⎡[ K ] uu [ K ] ur ⎤ ⎡[ X ] u ⎤ ⎡[ P] u ⎤ ⎡[O] u ⎤
⎢[ K ] = + (5.3)
⎣ ru [ K ] rr ⎥⎦ ⎢⎣[ X ] r ⎥⎦ ⎣⎢[ P] r ⎦⎥ ⎢⎣[ R] r ⎥⎦

where

[K ]uu is the stiffness matrix associated with unknown joint displacements,

[K ]r u is the stiffness matrix associated with the restrained displacements of the
support developed by the unit values of unknown joint displacements,
[K ]ur is the stiffness matrix associated with the unknown joint displacements devel-
oped by unit values of the displacements corresponding to restrained support
displacements,
[K ]rr is the stiffness matrix associated with restrained support displacements
developed by unit values of restrained support displacements.
116 5 Substructure Technique for the Analysis of Structural Systems

Equation (5.3) can be resolved into two independent equations

[K ]uu [X ]u = [P]u (5.4)

[K ]r u [X ]u = [P]r + [R]r (5.5)

It may be noted that [X ]r = 0

Unknown joint displacements can be evaluated from Eq. (5.4)

[X ]u = [K ]−1
uu [P]u (5.6)

Once [X ]u has been obtained, support reactions can be determined by substituting

[X ]u into Eq. (5.5). Thus,

[R]r = [K ]r u [K ] −1
uu [P]u − [P]r (5.7)

5.4 The Substructure Technique

The complete structure is partitioned into a number of substructures. Each substruc-

ture needs to be analysed independently for
(a) the load applied within each substructure,
(b) possible substructure boundary displacements.
For the analysis of nth substructure of a given structural system, the equations
describing the static equilibrium of the joints of the substructure can be expressed in
the following form when its boundaries are fixed:

⎡[ K ] uu
n
[ K ] urn ⎤ ⎡[ X ] ufn ⎤ ⎡[ P] ufn ⎤ ⎡ [O] ⎤
⎢ n ⎥⎢ ⎥=⎢ ⎥+⎢ n ⎥ (5.8)
⎣⎢[ K ] ru [ K ] rrn ⎦⎥ ⎣⎢ [O] ⎦⎥ ⎣⎢[ P] rfn ⎦⎥ ⎣[ R] rf ⎦

Subscript f refers to the case where substructure boundaries are fixed. Equa-
tion (5.8) results in two equations

[K ]nuu [X ]nu f = [P]nu f (5.9)

[K ]rnu [X ]nu f = [P]rn f + [R]rn f (5.10)

From Eq. (5.9), one gets

[K ]rnu [X ]nu f = [P]rn f + [R]rn f (5.11)

5.4 The Substructure Technique 117

Substituting the value of [X ]nu f from Eq. (5.11) into Eq. (5.10) yields

[R]rn f = [K ]rnu ([K ]nuu )−1 [P]nu f − [P]rn f (5.12)

[R]rn f , the reactions developed at the fixed substructure boundaries, are to be

determined from Eq. (5.12). Therefore, in the first stage of analysis, the substructure
boundaries are kept fixed and [X ]nu f and [R]rn f for the nth substructure are determined
from Eqs. (5.11) and (5.12) respectively.
In the next stage of analysis, substructure boundaries are subjected to displace-
ments. These substructure boundary displacements are to be obtained from the anal-
ysis of the total structure, and this is discussed a little later. In the meantime, it is
assumed that substructure joints are given boundary displacements. The subscript d
refers to this case. Therefore, the response of substructure n to the application of true
boundary displacements with all other loadings removed is written. In this case,

[P]nud = [P]rnd = [0] (5.13)

The equilibrium equations for this case are

⎡[ K ] uu
n
[ K ] urn ⎤ ⎡[ X ] ud
n
⎤ ⎡ [0] ⎤
⎢ n ⎥ ⎢ n ⎥ ⎢ n ⎥
(5.14)
⎣⎢[ K ] ru
n
[ K ] rr ⎦⎥ ⎣⎢[ X ] rd ⎦⎥ ⎣[ R] rd ⎦

where [X ]rnd represents the actual displacements of the substructure boundaries.

Equation (5.14) can be expressed into two equations

[K ]nuu [X ]nud + [K ]nur [X ]rnd = 0 (5.15)

[K ]rnu [X ]nud + [K ]rr

n
[X ]rnd = [R]rnd (5.16)

Equation (5.15) yields

[X ]nud = −([K ]nud )−1 [K ]nur [X ]rnd (5.17)

Substituting [X ]nud from Eq. (5.17) into Eq. (5.16) yields

−[K ]rnu ([K ]nuu )−1 [K ]nur [X ]rnd + [K ]rr

n
[X ]rnd = [R]rnd

or

([K ]rr
n
− [K ]rnu ([K ])nuu )−1 [K ]nur ) [X ]rnd = [R]rnd (5.18)

Introducing,
118 5 Substructure Technique for the Analysis of Structural Systems

[K ]ns = [K ]rr
n
− [K ]rnu ([K ]nuu )−1 [K ]nur (5.19)

Equation (5.18) is written as

[K ]ns [X ]rnd = [R]rnd (5.20)

It may be recalled that [R]rnd refers to the reaction at the substructure bound-
aries to maintain equilibrium when it is subjected to boundary displacements [X ]rnd .
Therefore, [K ]ns can be appropriately described as the substructure stiffness matrix.
Thus, for each substructure, the substructure stiffness matrix as given by Eq. (5.19)
is to be obtained. These stiffness matrices are assembled to get the complete struc-
ture stiffness matrix [K ]c . The displacements associated with the complete structure
stiffness matrix are those on the substructure boundaries along with those at the real
boundaries. Therefore,

[K ]c [X ]c = [P]c + [R]c (5.21)

The evaluation [P]c in Eq. (5.21) needs some discussion. [R]rn f , the reactions at
the fixed boundaries of each substructure obtained from Eq. (5.12) with their sign
changed constitute one part of [P]c . To this is added the loads applied at the joints
linking the substructure. It is to be borne in mind that loads applied at the joints of
the substructure boundary will not be considered in the analysis of the substructure,
but will be taken into account when analysing the complete structure.
[X ]c of Eq. (5.21) has [X ]rnd of each substructure. Once [X ]c is evaluated on the
basis of Eq. (5.21), then selecting appropriate values of [X ]rnd for each substructure,
and using Eq. (5.17), [X ]nud can be obtained. True displacements of the unrestrained
joints of the substructure then are

[X ]nu = [X ]nu f + [X ]nud (5.22)

Once the true displacements of the joints are known, internal forces of the members
can then be computed.
Therefore, true displacements at the joint of the structure are determined in three
stages:
(a) After keeping the substructure boundaries fixed, the displacements at the unre-
strained joints of the substructure are determined. All loads acting in this
substructure are considered except those acting at the substructure boundaries.
The displacements in this case are obtained by Eq. (5.11) and the reactions at
the substructure boundaries by Eq. (5.12).
(b) Next, from the analysis of total structure based on the substructure stiffness
matrix defined by Eq. (5.19), the unknown displacements at the substructure
boundaries are obtained.
This is done by applying Eq. (5.21). Therefore, [X ]rnd values of each substructure
are obtained in this step.
5.5 An Illustrative Example 119

(c) Based on these [X ]rnd values, the displacements at all unrestrained joints of
substructure can then be obtained by Eq. (5.17).
(d) Finally, the displacements obtained from steps (a) and (c) are added to get the
true displacements.
In calculation at every step, it must be borne in mind that the unrestrained joints
are labelled first, then the restraints.

5.5 An Illustrative Example

The truss in Fig. 5.2 is to be analysed by the substructure technique. Cross-sectional

areas of the vertical and horizontal bars = 200 mm2 and those of diagonal bars =
141.14 mm2 , E = 5.0 × 109 N/mm2 . All loads acting at different locations have
magnitude unity. The truss is split up into two substructures and they are shown in
Fig. 5.3 and are denoted as substructures 1 and 2 respectively.
Analysis of substructure 1
The displacements of the substructure are labelled. The unrestrained displacements
are denoted first by 1, 2, 3 and 4 and then the artificially restrained displacements
are labelled.
The structure stiffness matrix of this substructure is then generated. This is done
in the usual way by adding the member stiffness matrices with like labels. It assumes

Fig. 5.2 An illustrative

example
120 5 Substructure Technique for the Analysis of Structural Systems

Fig. 5.3 Substructures of

the example

⎡5 1 0 0 − 4 0 − 1 − 1⎤
⎢1 5 0 −4 0 0 − 1 − 1⎥⎥
⎢
⎢0 0 5 −1 −1 1 − 4 0 ⎥
⎢ ⎥
0 − 4 −1 5 1 −1 0 0⎥
[ K ]1c = ⎢ × 0.125 × 10 9
⎢ − 4 0 −1 −1 5 −1 0 0⎥
⎢ ⎥
⎢0 0 1 −1 −1 1 0 0⎥
⎢ −1 −1 − 4 0 0 0 5 1⎥
⎢ ⎥
⎣⎢ − 1 − 1 0 0 0 0 1 1 ⎦⎥

This matrix is partitioned as given in Eq. (5.8) as

⎡[ K ]1uu [ K ]1ur ⎤
[ K ]1c = ⎢ 1 ⎥
⎣⎢[ K ] ru [ K ]1rr ⎦⎥

The joint load matrix for this substructure is given by

5.5 An Illustrative Example 121
⎡ ⎤
1
⎢0⎥
⎢ ⎥
⎢1⎥
⎢ ⎥
⎢ ⎥
⎢1⎥
[P]c = ⎢ ⎥
1
⎢0⎥
⎢ ⎥
⎢0⎥
⎢ ⎥
⎣0⎦
0

It may be noted that loads at the substructures boundaries have not been considered
in the above [P]1c matrix.
The internal joint displacements are given by

[X ]1u f = ([K ]1uu )−1 [P]1u f

⎡ ⎤ ⎡ ⎤
10 −6 −1 −5 1
⎢ −6 30 5 25 ⎥ 2 × 10 −9 ⎢ ⎥
0
[X ]1u f = ⎢ ⎥
⎣ −1 5 10 6 ⎦ 11 ⎣ 1 ⎦
⎢ ⎥

−5 25 6 30 1
⎡ ⎤
4
2 × 10 −9 ⎢ 24
⎢ ⎥
⎥
= ⎣ 15 ⎦
11
31

The fixed boundary actions for substructure 1 are equal to

[R]r1 f = [K ]r1u [X ]1u f − [P]r1 f

⎡ ⎤ ⎡ ⎤ ⎡ ⎤
−4 0 − 1 1 4 0
10 9 ⎢⎢ 0 0 1 −1 ⎥ 2 × 10−9 ⎢ 24 ⎥ ⎢ 0 ⎥
⎥ ⎢ ⎥−⎢ ⎥
=
8 ⎣ −1 −1 −4 0 ⎦ 11 ⎣ 15 ⎦ ⎣ 0 ⎦
−1 −1 0 0 31 0
⎡ ⎤
0
2⎢⎢ ⎥ 16 ⎥
[R]r f =
1
⎣
88 88 ⎦
28

The substructure stiffness matrix for substructure 1 can now be evaluated

[K ]1s = [K ]rr
1
− [K ]r1u ([K ]1uu )−1 [K ]1ur
122 5 Substructure Technique for the Analysis of Structural Systems

⎡ ⎤ ⎡ ⎤
5 −1 0 0 −4 0 −1 1
⎢ −1 1 0 0 ⎥ 10 9 ⎢ 0 0 1 −1 ⎥ 109
=⎢ ⎥ ⎢ ⎥
⎣ 0 0 5 1 ⎦ 8 − ⎣ −1 −1 −4 0 ⎦ 8
0 011 −1 −1 0 0
⎡ ⎤ ⎡ ⎤
10 −6 −1 −5 −4 0 −1 −1
⎢ −6 30 5 25 ⎥ 2 × 10−9 ⎢ 0 0 −1 −1 ⎥ 109
⎢ ⎥ ×⎢ ⎥
⎣ −1 5 10 6 ⎦ 11 ⎣ −1 1 −4 0 ⎦ 8
−5 25 6 30 1 −1 0 0
( 5 6 7 8 − boundary displacement labels )
⎡ ⎤
0 00 0
⎢ 0 1 0 −1 ⎥ 109
=⎢ ⎥
⎣ 0 0 0 0 ⎦ 22
0 −1 0 1

Analysis of substructure 2
There are no interior displacements for substructure 2. The structure stiffness matrix
is same as that of [K ]uu part of substructure 1. Further actual boundary displacements
will have no effect on the substructure stiffness matrix. Therefore, the substructure
stiffness matrix for substructure 2 is
⎡ ⎤
5 1 0 0
⎢ 1 5 0 −4 ⎥ 109
[K ]2s = ⎢ ⎥
⎣ 0 0 5 −1 ⎦ 8
0 −4 −1 5

This type of substructure is a special case where there are no unrestrained

displacements.
Analysis of the complete structure
The displacements at the substructure boundary i.e. corresponding to the restrained
displacements in the restrained structure are labelled and shown in Fig. 5.4.
The substructure stiffness matrix for the two substructures has already been
obtained. The labels of these substructure stiffness matrices are considered in this
case.
However, in the general case, the labels are to be changed with respect to the total
structure. The complete structure stiffness matrix is obtained by adding the elements
of substructure stiffness matrix with the labels.
In this case,
5.5 An Illustrative Example 123

Fig. 5.4 Labelling of the

boundary displacements of
the complete structure

3 1

4
2

7 5

8
6

⎡ ⎤
55 11 0 0
⎢ 11 −48 ⎥ 9
[K ]c = ⎢
59 0 ⎥ 10
⎣0 0 55 −11 ⎦ 88
0 −48 −11 59

The joint load matrix for the total structure is obtained by adding the elements
with like labels of the reactions at the substructure boundaries with sign changed
when the substructure boundaries have been considered fixed with the loads at these
boundaries. Therefore,
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
0 1 44
1⎢ ⎥ ⎢ ⎥ ⎢
⎢ 16 ⎥ + ⎢ 0 ⎥ = 1 ⎢ 16 ⎥
⎥
[P]c = ⎣ ⎦ ⎣ ⎦
44 88 1 ⎣
44 132 ⎦
78 1 122

[X ]c is to be obtained from the solution of the complete structure.

[X ]c = [K ]−1
c [P]c
⎡ ⎤ ⎡ ⎤
1.43 −0.852 −0.144 −0.72 44
⎢ −0.852 3.6 ⎥ ⎢ ⎥
= 10−9 ⎢
4.26 0.72 ⎥ 1 ⎢ 16 ⎥
⎣ −0.144 0.72 1.43 0.852 44 132 ⎦
⎦ ⎣
−0.72 3.6 0.852 4.42 122
⎡ ⎤
−57.56
10−9 ⎢ ⎥
⎢ 564.91 ⎥
= ⎣
44 297.88 ⎦
677.62
124 5 Substructure Technique for the Analysis of Structural Systems

The real displacements at the unrestrained joints of the substructure 1 are next to
be determined. For substructure 1,
⎡ ⎤
−57.56
10−9 ⎢ ⎥
⎢ 564.91 ⎥
[X ]r1d = ⎣
44 297.88 ⎦
677.62

It may be noted that in this case [X ]r1d is same as [X ]1c and they are displacements
corresponding to labels 5, 6, 7 and 8 of substructure 1 which are same as labels, 1,
2, 3 and 4 of the complete structure.

[X ]1u = [X ]1u f − ([K ]1uu )−1 [K ]1ur [X ]r1d

⎡ ⎤ ⎡ ⎤
4 10 −6 −1 −5
−9 ⎢
2 × 10 ⎢ 24 ⎥ ⎢ ⎥
⎥ − ⎢ −6 30 5 25 ⎥ 2 × 10
−9
= ⎣ 15 ⎦ ⎣ −1 5 10 6 ⎦
11 11
31 −5 25 6 30
⎡ ⎤ ⎡ ⎤
−4 0 −1 −1 −57.56
⎢ 0 0 −1 −1 ⎥ 10 9
10 ⎢
−9 ⎥
⎢ ⎥ ⎢ 264.91 ⎥
⎣ −1 1 −4 0 ⎦ 8 × 44 ⎣ 297.88 ⎦
1 −1 0 0 677.62
⎡ ⎤
2.29
2 × 10−1 ⎢ ⎢ −16.48 ⎥
⎥
= ⎣
11 −23.52 ⎦
−13.45
⎡ ⎤
2.29
2 × 10−9 ⎢ −16.48 ⎥
[X ]r1d = ⎢ ⎥
11 ⎣ −23.52 ⎦
−13.45

For substructure 2, there are no unrestrained joints as the substructure boundaries

are fixed.
All the information is now available for the determination of internal forces in
the members. The procedure for determining internal forces is same as that treated
in earlier chapters.
5.6 Computer Program for the Truss Analysis by the Substructure Technique 125

5.6 Computer Program for the Truss Analysis

by the Substructure Technique

The computer program for the plane truss analysis by the substructure technique in
both the languages—C and FORTRAN, are given in the attached CD. Notations have
been explained in the listing of the computer program in both cases. Different steps
followed in the computer code have been explained with ‘comment’ statements.
The computer program presented here is more advanced than in earlier chapters,
and as such, it is to be carefully followed. There are many matrix operations to be
carried out. This is best illustrated with an example.

Example 5.1 The truss in Fig. 5.5 is to be analysed by the substructure technique.
In Fig. 5.5b, the truss is divided into two substructures and in Fig. 5.5c, into 3. The
preparation of the input data corresponding to Fig. 5.5c will be discussed.

Out of the three substructures, there are two substructure boundaries where there
are 4 unrestrained joints in all. Each joint has two degrees of freedom. Therefore, NS
= 8 and NPART = 3. Only one loading condition is considered, i.e. NLC = 1.
For substructure 1, the displacement labels are put as shown in Fig. 5.6a. The
unrestrained joints are numbered first, then the displacements at the substructure
boundary and, lastly, at the real boundary. For this substructure, then NI(IJ) = 7, NP
(IJ) = 3, NR(H) = 4 and NF(IJ) = 6. The member data corresponding to item no.
4 to be prepared is exactly same as those required by the direct stiffness method for
the truss.
Note that NSUB(IJ) for substructures 1 and 2 is 0, while that for substructure 3 is
4 which is the last global number of degrees of freedom of substructure 2 based on
the complete structure (see Fig. 5.6d). Also to be noted is the fact that, though there
may not be any load acting at the joint, still ND should not be made 0, but be given a
minimum value 1. The input data for the problem at hand is given in Table 5.1. The
output of the truss problem is presented in Table 5.2.
The values of the forces in the members of the same truss obtained by the
direct stiffness method as also by the substructure technique by dividing into two
substructures (Fig. 5.5b) are given in Table 5.3.
126 5 Substructure Technique for the Analysis of Structural Systems

Fig. 5.5 Details of a truss with substructures

5.6 Computer Program for the Truss Analysis by the Substructure Technique 127

6 2 5 2
5 1 4 1
2

3
6 1
8 4 7 4
7 3
5
6 3
8
Labelling of the complete structure
(a) Substructure 1

6 2 2 4
5 1 1
1 2 3

2 3
6 1
8 3 4 4
8 6
4 3 5 5
7 7
(b) Substructure 2 (c) Substructure 3

Fig. 5.6 Displacement labels of the individual substructure and the complete structure

Exercise 5

Assume E = 22.5 × 106 kN/mm2

5.1 Write a computer program in FORTRAN for the solution of a plane frame by
the substructure technique.
5.2 Extend the computer program given in Table 5.1 for the analysis of a space
truss.
5.3 Neglecting axial deformation and considering the moment of inertia of all
members to be same, determine the member forces by partitioning the structure
into three substructures.
5.4 Treating each element as a substructure, determine the complete structure
stiffness matrix and the member forces.
5.5 Prepare the input data for the truss to be solved by the computer program in
Table 5.1. Areas of all vertical and horizontal members = 5000 mm2 and those
of inclined members = 4000 mm2 .
5.6 For the frame, determine the member forces after dividing the frame into two
substructures EI is same for all.
128 5 Substructure Technique for the Analysis of Structural Systems

Table 5.1 Input data for the truss analysis by substructure technique
8 1 3
30000.0
7 3 4 6
3 8 1 2 0.0 120.0 3.0
4 5 1 2 120.0 0.0 3.0
6 7 1 2 120.0 120.0 3.0
4 5 3 8 120.0 -120.0 3.0
6 7 3 8 120.0 0.0 3.0
6 7 4 5 0.0 120.0 3.0
1
1 1 0.0
0
8 0 8 4
5 6 1 2 120.0 0.0 3.0
7 8 1 2 120.0 120.0 3.0
5 6 3 4 120.0 -120.0 3.0
7 8 3 4 120.0 0.0 3.0
0
6 2 4 6
5 6 3 4 0.0 120.0 3.0
1 2 3 4 120.0 0.0 3.0
7 8 3 4 120.0 120.0 3.0
1 2 5 6 120.0 -120.0 3.0
7 8 5 6 120.0 0.0 3.0
7 8 1 2 0.0 120.0 3.0
1
1 1 0.0
4
2
4 1 -50.0
8 1 -50.0
5.6 Computer Program for the Truss Analysis by the Substructure Technique 129

Table 5.2 Computer output for the truss analysis by substructure technique
Analysis of Truss by Substructure Technique
NS = 8 NLC = 1 NPART = 3 E = 3.00000e+004
Substructure No. = 1
MEMBER NP1 NP2 NP3 NP4 H V A
1 3 8 1 2 0.00000e+000 1.20000e+002 3.00000e+000
2 4 5 1 2 1.20000e+002 0.00000e+000 3.00000e+000
3 6 7 1 2 1.20000e+002 1.20000e+002 3.00000e+000
4 4 5 3 8 1.20000e+002 -1.20000e+002 3.00000e+000
5 6 7 3 8 1.20000e+002 0.00000e+000 3.00000e+000
6 6 7 4 5 0.00000e+000 1.20000e+002 3.00000e+000
Substructure No. = 2
MEMBER NP1 NP2 NP3 NP4 H V A
1 5 6 1 2 1.20000e+002 0.00000e+000 3.00000e+000
2 7 8 1 2 1.20000e+002 1.20000e+002 3.00000e+000
3 5 6 3 4 1.20000e+002 -1.20000e+002 3.00000e+000
4 7 8 3 4 1.20000e+002 0.00000e+000 3.00000e+000
Substructure No. = 3
MEMBER NP1 NP2 NP3 NP4 H V A
1 5 6 3 4 0.00000e+000 1.20000e+002 3.00000e+000
2 1 2 3 4 1.20000e+002 0.00000e+000 3.00000e+000
3 7 8 3 4 1.20000e+002 1.20000e+002 3.00000e+000
4 1 2 5 6 1.20000e+002 -1.20000e+002 3.00000e+000
5 7 8 5 6 1.20000e+002 0.00000e+000 3.00000e+000
6 7 8 1 2 0.00000e+000 1.20000e+002 3.00000e+000
ND = 2
The Pload Matrix
0.00000e+000 0.00000e+000 0.00000e+000 -5.00000e+001 0.00000e+000
0.00000e+000 0.00000e+000 -5.00000e+001
The XX Matrix
2.70091e-002 -1.97078e-001 9.36758e-002 -2.24087e-001 1.00731e-001
-1.97078e-001 3.40639e-002 -2.24087e-001
Member Force in Substructure No. = 1
NFSUB = 0
NPJ = 3 NRJ = 4 NFF = 6 NTJ = 7
Member Force in substructure No. 1
Member No. 1 -2.44521e+001
Member No. 2 -2.44521e+001
Member No. 3 3.45805e+001
Member No. 4 -3.61302e+001
Member No. 5 2.55479e+001
Member No. 6 2.02568e+001
Member Force in Substructure No. = 2
Member No. 1 -5.52912e+001
Member No. 2 7.48281e+000
Member No. 3 7.48282e+000
Member No. 4 4.47089e+001
Member Force in Substructure No. = 3
NFSUB = 4
NPJ = 2 NRJ = 4 NFF = 6 NTJ = 6
Member Force in substructure No. 3
Member No. 1 2.02568e+001
Member No. 2 -2.44521e+001
Member No. 3 -3.61302e+001
Member No. 4 3.45805e+001
Member No. 5 2.55479e+001
Member No. 6 -2.44521e+001
130 5 Substructure Technique for the Analysis of Structural Systems

Table 5.3 Comparison of the results of the truss in Fig. 5.5 obtained by various methods of analysis
Member no. Direct stiffness Two substructures (Fig. 5.5b) Three substructures (Fig. 5.5c)
1 −24.45208* −24.45208 −24.45208
2 −24.45208 −24.45208 −24.45208
3 34.58047 34.58043 34.58047
4 −36.13021 −36.13019 −36.13021
5 25.54792 25.54790 25.54792
6 20.25677 20.25678 20.25677
7 −55.29115 −55.29113 −55.29115
8 7.48281 7.48281 7.48281
9 7.48281 7.48281 7.48281
10 44.70885 44.70882 44.70885
11 20.25677 20.25677 20.25677
12 −24.45208 −24.45206 −24.45208
13 −36.13021 −36.13021 −36.13021
14 34.58047 34.58046 34.58049
15 25.54792 25.54788 25.54792
16 −24.45208 −24.45207 −24.45208

* Indicates compression

Prob. 5.3 Prob. 5.4

5.6 Computer Program for the Truss Analysis by the Substructure Technique 131

Prob. 5.5

5.7 Analyse the grid by dividing it into two substructures. The moment of inertia
and polar moment of inertia of each member are 1000 mm4 and 7000 mm4
respectively. Assume G = E/3.

Prob. 5.6 Prob. 5.7

5.8 The space truss is to be divided into two substructures. For the modified
computer program according to problem 5.2, prepare the input data for the
truss. The cross-sectional area of each member of the space truss in mm2 is
indicated adjacent to the member.
132 5 Substructure Technique for the Analysis of Structural Systems

Prob. 5.8

References and Suggested Readings

1. E. Hinton, D.R.J. Owen, Finite Element Programming (Academic Press, 1977)

2. B.M. Irons, A frontal solution program for the finite element analysis. Int. J. Num. Meth. Eng.
2, 5–32 (1970)
3. W.R. Spillers, Techniques for the analysis of large structures. J. Strl. Divn. Proc. ASCE 94,
2521–2534 (1968)
4. J.S. Przmieniecki, Theory of Matrix Structures (McGraw-Hill, 1968)
5. M.F. Rubinstein, Matrix Computer Analysis of Structures (Prentice Hall, 1966)
6. M.F. Rubinstein, Combined analysis of structure and recursion. J. Strl. Divn. Proc. ASCE 94,
231–235 (1967)
Chapter 6
The Flexibility Matrix Method

6.1 Introduction

The basic approach to the flexibility matrix method has been introduced in Chap. 3
where no attention has been paid to orient it towards the needs of the digital computer.
The purpose of this chapter is to introduce a more formalized approach for the analysis
performed by the flexibility matrix method. Unlike the stiffness method, it does not
follow a set procedure which is primarily due to the existence of various choices of
the primary structure. However, the complete formulation of the flexibility method
can be done in terms of matrices and matrix operations.

6.2 Element Flexibility Matrix

Due to the external loading, the element of the structure deforms in its own ‘standard’
way, e.g. the members of a truss can only have axial deformation, the deformation
of the member of frames or continuous beams where axial rigidity is not considered,
is generally denoted by the slopes at the member ends and so on. Defining the
deformation in this fashion results in the definition of flexibility matrices which
become typical of that structure and the approach becomes.
more mechanical as will be shortly seen.
The flexibility coefficient of the truss member of Fig. 6.1 (a) (which is say element
i) is

L
[F M]i = [ f 11 ] i = (6.1)
AE
The flexibility matrix of the beam element i of Fig. 6.1 (b) corresponding to the
labels shown are

© The Author(s) 2022 133

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_6
134 6 The Flexibility Matrix Method

Fig. 6.1 FM Matrix 1

L, AE

1
(a) Truss member

L, EA, EI, GJ

1 2
L, EI
2 3
(b) Beam 4

(c
c) A bending member with
axial force and twist

[ ] [ ]
f 11 f 12 L 21
[F M] i = = (6.2)
f 21 f 22 i
6E I 12

The element of Fig. 6.1(c) is subjected to axial force, bending moment and twisting
moment. The flexibility matrix of the member corresponding to the labels indicated
is
⎡ ⎤ ⎡ L L ⎤
f 11 f 12 f 13 f 14 3E I 6E I
0 0
⎢ f 21 f 22 f 23 f 24 ⎥ ⎢ L L
0 0 ⎥
[F M] i = ⎢
⎣ f 31
⎥ = ⎢ 6E I 3E I ⎥ (6.3)
f 32 f 33 f 34 ⎦ ⎣ 0 0 L
EA
0 ⎦
L
f 41 f 42 f 43 f 44 0 0 0 GJ

6.3 Principle of Contragredience

This principle is utilized in the construction of the flexibility matrix of the structure
in an automatic way.
A set of forces PM 1 , PM 2 and PM 3 and the corresponding displacements XM 1 ,
XM 2 and XM 3 exist at point O in a structure (Fig. 6.2). The effect of this force system
on another point O/ of the structure, the coordinates of O is (x, y) with O/ as origin,
is to be investigated.
Applying equations of statics, we get
6.3 Principle of Contragredience 135

Fig. 6.2 Forces and P2, X2 PM 2, XM2

displacements in the parallel
X
axis system

0
PM 1, XM1
PM 3, XM3 y
P1, X1
0′, P3, X3

⎫
P M 1 = P1 ⎬
P M 2 = P2
⎭
P M 3 = P3 + P2 x − P1 y

or,
⎧ ⎫ ⎡ ⎤⎧ ⎫
⎨ P M1 ⎬ 1 0 0 ⎨ P1 ⎬
P M2 = ⎣ 0 1 0 ⎦ P2 (6.4)
⎩ ⎭ ⎩ ⎭
P M3 −y x 1 P3

or,

{P M} = [T1 ] {P} (6.5)

Two systems of displacements are related thus

⎧ ⎫ ⎡ ⎤⎧ ⎫
⎨ X M1 ⎬ 10 y ⎨ X1 ⎬
X M2 = ⎣ 0 1 − x ⎦ X2 (6.6)
⎩ ⎭ ⎩ ⎭
X M3 00 1 X3

{X M} = [T2 ] {X } (6.7)

It can be easily shown that

[T1 ] [T2 ] T = [I ]

also,

[T2 ] [T1 ]T = [I ] (6.8)

Substituting [T 2 ] from the second relation of Eq. (6.8) into Eq. (6.7) yields

{X M} = ([T1 ] T ) −1 {X }
136 6 The Flexibility Matrix Method

or

{X } = [T1 ] T {X M} (6.9)

Thus, the principle of contragredience is illustrated by the following equations:

{P M} = [T1 ] {P}

and

{X } = [T1 ]T {X M}

The forces and displacements are said to transform contragrediently.

6.4 The Equilibrium Matrix

Consider the frame of Fig. 6.3. Suppose this frame needs to be analysed by the
flexibility matrix method. P1 and P2 are chosen as the redundants and PM 1, PM 2 ,
PM 3 and PM 4 are the end moments of the member. The relationship between P1 and
P2 and PM l , PM 2 , PM 3 and PM 4 is to be determined. In order to do that, unit values
corresponding to P1 and P2 are applied separately and the necessary equilibrium
diagrams are drawn as shown in Fig. 6.3 (d) and (e). Corresponding to Fig. 6.3(d),
the forces at member references are

P M1 = 1, P M2 = 0, P M3 = 0 and P M4 = 0

Similarly, corresponding to Fig. 6.3(e), the forces at member references are

P M1 = 0, P M2 = −V, P M3 = −V and P M4 = 0

Therefore, the following relationship can be written:

⎧ ⎫ ⎡ ⎤
⎪
⎪ P M1 ⎪
⎪ 1 0
( )
⎨ ⎬ ⎢0
P M2 −V ⎥⎥ P1
= ⎢
⎣0 (6.10)
⎪
⎪ P M3 ⎪
⎪ − V ⎦ P2
⎩ ⎭
P M4 0 0

{P M} = [E] {P} (6.11)

[E] matrix is known as the equilibrium matrix. Each column of the equilibrium
matrix is generated by applying the unit value of the redundant and finding out the
6.4 The Equilibrium Matrix 137

H
PM1 PM2
A B
PM3

V
PM4
C

(a) The frame (b) Member references

0 1
1
P1 H

P2 1
H
(c) System references (d) Reactions corresponding
to unit value of P1

1 w/unit length
1
H

1
1
H
(e) Reactions corresponding
(f) Loading
to unit value of P2

2 2
wH wH
Pl1 = Pl2 =
12 12

(g) Joint loading references

Fig. 6.3 A frame with member and system references

138 6 The Flexibility Matrix Method

forces at member references. Comparing eqns. (6.11) and (6.5) reveals that [E] =
[T 1 ]. Therefore, [E] is nothing but force transformation matrix.

6.5 Construction of the Flexibility Matrix of the Structure

Suppose for the frame of Fig. 6.3, H = V = L. The flexibility matrix of the system
consisting of the flexibility matrix of the member in their unconnected state is written
as
[ ]
[F M] = [F[0] M]1 [0]
[F M]2

or,
] [⎡[ ]⎤
21 00
L ⎢ ⎢ 12 00 ⎥⎥
[F M] = ⎢[ ] [ ]⎥ (6.12)
6E I ⎣ 0 0 21 ⎦
00 12

The displacements at the member ends are given by

{X M} = [F M] {P M} (6.13)

Similarly, the displacements corresponding to the redundants of the total structure

are

{X } = [F] {P} (6.14)

Substituting {PM} from Eq. (6.5) and changing [E] for [T 1 ] and replacing the
value of {XM} from Eq. (6.9), Eq. (6.13) becomes
( )−1
[E]T {X } = [F M] [E] {P} (6.15)

Premultiplying both sides of Eq. (6.15) by [E]T results

{X } = [E]T [F M] [E] {P} (6.16)

On comparing Eq. (6.16) with Eq. (6.14) immediately reveals that the structure
flexibility matrix is given by

[F] = [E]T [F M] [E] (6.17)

Therefore, for the frame of Fig. 6.3, the structure flexibility matrix is
6.6 Matrix Determination of the Displacement Vector 139

⎡ ⎤⎡ ⎤
2 1 0 0 1 0
[ ] [ ]
1 0 00 L ⎢ ⎢1 2 0 0⎥ ⎢0
⎥⎢ −L⎥ ⎥= L 2 −L
[F] =
0 −L L 0 6E I ⎣ 0 0 2 1⎦ ⎣0 − L ⎦ 6E I − L 4L 2
0 0 1 2 0 0

6.6 Matrix Determination of the Displacement Vector

The flexibility coefficients indicate displacements. Therefore, following the similar

procedure as above, the displacement vector {X} of Eq. (6.1) can be obtained.
The loading of the frame is indicated in Fig. 6.3 (f). X 1 , and X 2 are the displace-
ments corresponding to redundants (sec Fig. 6.3 (c)) of the primary frame due to
external loading. Let {PL} represent a vector of external loading usually repre-
senting the equivalent joint loading at salient points for the frame of Fig. 6.3. The
{PL} matrix is
( ) ( )
P L1 wL 2 1
=
P L2 12 1

Let [E 0 ] be the matrix which relates {PM} to {PL} such that

{P M} = [E 0 ] {P L} (6.18)

It has already been shown that (i.e. eqns. (6.13) and (6.9) are rewritten)

{X M} = [F M] {P M} (6.19)

Also,

{X } = [E]T {X M} (6.20)

Combining the three eqns. (6.18), (6.19) and (6.20) yields

( T )− 1
[E] {X } = [F M] [E 0 ]{P L} (6.21)

Premultiplying both sides of Eq. (6.21) by [E]T yields

{X } = [E]T [F M] [E 0 ] {P L} (6.22)

{X} represents the displacements caused by the redundants. For the frame of
Fig. 6.3, corresponding to PL 1 and PL 2 indicated, Eq. (6.18) becomes
140 6 The Flexibility Matrix Method
⎧ ⎫ ⎡ ⎤
⎪
⎪ P M1 ⎪
⎪ 1 0
( )
⎨ ⎬ ⎢
P M2 0 1⎥⎥ P L1
=⎢
⎣0
⎪
⎪ P M3 ⎪
⎪ 1 ⎦ P L2
⎩ ⎭
P L4 0 0
⎡ ⎤
1 0
⎢0 1⎥
i.e. [E 0 ] = ⎢
⎣0
⎥
1⎦
0 0
⎤⎡
⎡ ⎤
2100 10
[ ] ( )
1 0 0 0 L ⎢⎢ 1 2 0 0⎥⎥
⎢ 0 1 ⎥ wL 2 1
⎢ ⎥
{X } =
0 −L −L 0 6E I ⎣ 0 0 2 1 ⎦ ⎣ 0 1 ⎦ 12 1
Therefore, 0012 00
3 [ ]( ) 3 ( ) ( )
wL 2 1 1 wL 3 wL 3 1
= = =
72E I − L −2L 1 72E I −3L 24E I −1
It may be noted that only the non-zero components of equivalent joint loading
and not all possible components be considered in the construction of {PL} matrix.
Further, the sign of {PL} matrix need not coincide with those components of the
member reference system.

6.7 Determination of Member Forces

Once the redundant {P} are obtained from the solution of simultaneous equations
[Eq. (6.14)], the forces in the members are to be obtained. The member forces are
determined by the following equation:

{P M} = [E 0 ] {P L} + [E] {P} + {P M F} (6.23)

{PMF} constitutes the constraining forces corresponding to the member refer-

ences. For a bending member, the first part of Eq. (6.23) represents the forces in the
members due to the loading in the primary structure and the second part, that due to
redundants acting on the primary structure.
6.9 Illustrated Examples 141

6.8 Procedure of the Analysis of Statically Indeterminate

Structures

The procedure for analysis of statically indeterminate structure is as follows:

1. The structure is converted into a statically determinate structure and the
redundants are accordingly identified.
2. Select element references for which member flexibility matrix exists.
3. Select system references.
4. Select references corresponding to equivalent joint loading.
5. Generate [E], [FM], [E 0 ], [PMF] and [PL] matrices.
6. Form[F] = [E]T [FM] [E].
7. Form[X] = [E]T [FM] [E 0 ] [PL].
8. Solve for [P] from eqn. (6.14).
9. Determine [PM] from eqn. (6.23).

6.9 Illustrated Examples

Example 6.1 The continuous beam of Fig. 6.4 is to be analysed. The particulars of
the beam have been indicated in the figure.

Fig. 6.4 The beam of Example 6.1

142 6 The Flexibility Matrix Method

The beam is made statically determinate by removing the reactions of the roller
supports. All the necessary references have been shown in Fig. 6.4 (b), (c) and (d).
All the necessary matrices generated are given below:
⎡ ⎤ ⎡ ⎤ ⎧ ⎫
− L − 2L 2100 ⎪
⎪ wL ⎪ ⎪
⎢ 0 −L ⎥ ⎢ ⎥ ⎨ ⎬
⎢ ⎥ 1 ⎢1 2 0 0⎥ 0
[E] = ⎣ , [F M] = , {P L} =
0 −L ⎦ 6E I ⎣ 0 0 2 1 ⎦ ⎪
⎪
wL
⎪
⎩ wL2 2 ⎪
⎭
0 0 0012
⎧ ⎫ ⎡ ⎤ 12 ,
⎪−1⎪
⎪
⎨ ⎪
⎬
− L 1 − 2L 1
−1 ⎢ 0 1 − L 1⎥
{P M F} = wL , [E 0 ] = ⎢ ⎥
2

12 ⎪ − 1 ⎪ ⎣ 0 0 − L 1⎦
⎪
⎩ ⎪
⎭
−1 0 0 0 1

Therefore,
⎡ ⎤⎡ ⎤
2 1 0 0 − L − 2L
[ ]
−L 0 0 0 L ⎢ ⎢1 2 0 0⎥ ⎢
⎥⎢ 0 −L ⎥
⎥
[F] =
− 2L − L − L 0 6E I ⎣ 0 0 2 1 ⎦ ⎣ 0 −L ⎦
0 0 1 2 0 0
[ ]
L 2 5
=
6E I 5 16
⎡ ⎤⎡ ⎤⎧ ⎫
2 1 0 0 − L 1 − 2L 1 ⎪ ⎪ wL ⎪ ⎪
[ ] ⎨ ⎬
−L 0 0 0 L ⎢ ⎢1 2 0 0⎥ ⎢ 0 1 −L
⎥⎢ 1⎥⎥ 0
{X } =
− 2L − L − L 0 6E I ⎣ 0 0 2 1⎦ ⎣ 0 0 − L 1⎦ ⎪⎪
wL
⎪
⎩ wL2 2 ⎪
⎭
0 0 1 2 0 0 0 1 12
( )
wL 4 51
=
72E I 144

Applying Eq. (6.14), we get

[ ]( ) ( )
L3 2 5 P1 wL 4 51
=
6E I 5 16 P2 72E I 144

or,
( ) ( )
P1 wL 32
=−
P2 28 11

The member forces are given by

6.9 Illustrated Examples 143
⎧ ⎫ ⎡ ⎤ ⎧ ⎫ ⎡ ⎤ ⎧ ⎫
⎪
⎪ P M1 ⎪
⎪ −L 1 − 2L 1 ⎪ 12 ⎪
⎪ ⎪ − L − 2L ( ) ⎪
⎪ −1 ⎪⎪
⎪
⎨ ⎪
⎬ ⎢ ⎨ ⎪
⎥ wL ⎪ ⎬ ⎢ ⎥ ⎪
⎨ ⎪
⎬
P M2 ⎢ 0 1 −L 1⎥ 0 ⎢ 0 − L ⎥ −wL 32 wL 2 −1
=⎢ ⎥ +⎢ ⎥ +
⎪
⎪ P M3 ⎪
⎪ ⎣ 0 0 − L 1 ⎦ 12 ⎪⎪ 6 ⎪
⎪ ⎣ 0 − L ⎦ 28 11 12 ⎪
⎪ − 1 ⎪
⎪
⎪
⎩ ⎪
⎭ ⎩ ⎪
⎪ ⎭ ⎪
⎩ ⎪
⎭
P M4 0 0 0 1 L 0 0 −1
⎧ ⎫
⎪
⎪ 2L ⎪
⎪
2 ⎪
⎨ ⎪
⎬
wL 3L
=−
28 ⎪ ⎪ 3L ⎪
⎪
⎪
⎩ ⎪
⎭
0

Example 6.2 Analyse the truss of Fig. 6.5. All the particulars are given in the
diagram. Relative axial rigidities are given in brackets.

The member and system references have been indicated in Fig. 6.5(b) and (c).
Therefore, based on these references, the matrices which are input to the computer
are as follows (tensile force is considered as positive):
⎡ ⎤
0 − 3/4
⎢ 0 0⎥
⎢ ⎥
⎢ 0 0⎥
⎢ ⎥
⎢ 0 5/4 ⎥
⎢ ⎥
⎢ ⎥
⎢ − 0.8 0⎥
[E] = ⎢ ⎥
⎢ − 0.6 0⎥
⎢ ⎥
⎢ 1.0 0⎥
⎢ ⎥
⎢ 1.0 0⎥
⎢ ⎥
⎣ − 0.8 0⎦
− 0.6 0

Diagonals of [F M] = [ 3 8 4 4 2 3 3 5 3 4] .
Off-diagonals of [FM] are zero.

{P L} = {10}

[E 0 ]T = [ 3/4 0 1 − 5/4 0 1 0 0 0 0]

The flexibility matrix of the truss

[ ]
13.72 0
[F] = [E] T [F M] [E] =
0 7.9375
( )
18.000
{X } = [E] [F M] [E 0 ] {P L} =
T
79.375

In the solution of Eq. (6.14),

144 6 The Flexibility Matrix Method

Fig. 6.5 The truss of

Example 6.2

( )
1.312
{P} =
10.000

The axial forces in the member of the truss then are

6.10 Choice of the Released Structure 145

⎡ ⎤ ⎡ ⎤
0 − 3/4 3/4
⎢ 0 0⎥ ⎢ 0⎥
⎢ ⎥ ⎢ ⎥
⎢ 0 0⎥⎥ ⎢ 1⎥
⎢ ⎢ ⎥
⎢ 0 5/4 ⎥ ⎢ − 5/4 ⎥
⎢ ⎥( ⎢
) ⎢ ⎥
⎢ ⎥ ⎥
⎢ − 0.8 0 ⎥ 1.312 ⎢ 0⎥
{P M} = ⎢ ⎥ +⎢ ⎥ {10}
⎢ − 0.6 0 ⎥ 10.000 ⎢ 1⎥
⎢ ⎥ ⎢ ⎥
⎢ 1.0 0⎥ ⎢ 0⎥
⎢ ⎥ ⎢ ⎥
⎢ 1.0 0⎥ ⎢ 0⎥
⎢ ⎥ ⎢ ⎥
⎣ − 0.8 0⎦ ⎣ 0⎦
− 0.6 0 0
⎧ ⎫
⎪
⎪ 0⎪⎪
⎪
⎪ ⎪
⎪
⎪ 0⎪⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ 10 ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ 0 ⎪
⎪
⎪
⎨ ⎪
⎬
− 1.050
=
⎪
⎪ 4.000 ⎪⎪
⎪ 1.312 ⎪
⎪ ⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ 1.312 ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ − 1.050 ⎪
⎪
⎪
⎩ ⎪
⎭
− 0.787

6.10 Choice of the Released Structure

It is now realized that there is no unique choice regarding the primary structure. The
main disadvantage of the flexibility method is that the construction of the transforma-
tion matrices does not follow a set pattern. Both [E] and [E 0 ] matrices are input to the
computer and they are to be generated by hand calculation. For large-sized problems,
this generation becomes a tedious affair and takes away much of the charm of this
computer-oriented method. For this reason, the stiffness method is more popularly
applied for the solution of large-sized problems. However, in certain problems where
the degree of statical indeterminacy is much lesser than kinematical indeterminacy,
the flexibility method may be preferred. In the flexibility matrix method, the choice
of the primary structure is of paramount importance. While making the choice, the
146 6 The Flexibility Matrix Method

following points must be borne in mind:

(a) The complete primary structure should be statically determinate.
(b) The choice should be such that the generation of the necessary transformation
matrices [E] and [E 0 ] is not tedious.
(c) It may happen that depending on the choice of the released structure, the flexi-
bility matrix may become ill-conditioned. The primary structures which do not
result in a well-conditioned flexibility matrix of the structure should be avoided.

Exercise 6
The problems given are to be solved by the method presented in this chapter.
6.1. Analyse the continuous beam

Prob. 6.1

6.2 Analyse the grid. E = 2 × 105 N/mm2 , G = 0.6 × 105 N/mm2 , I = 4 × 108
mm4, J = 4 × 107 mm4 . All members are identical.
6.3 The cross-sectional properties of all members of the space frame are identical.
Determine the end moments of the members.

Prob. 6.2
6.10 Choice of the Released Structure 147

Prob. 6.3

6.4 Develop a computer program for the analysis of any framed structure by the
method presented in this chapter.
6.5 Analyse the frame.
6.6 Analyse the plane truss. The axial rigidity of the members of the truss is
identical.

Prob. 6.5

Prob. 6.6

6.7. Analyse the beam. Draw the bending moment and shear force diagrams.
6.8. For the grillage as shown, E = 2.5 G. For the loading indicated, determine
the variation of bending and twisting moment along the length of the member.
148 6 The Flexibility Matrix Method

Prob. 6.7

Prob. 6.8

References and Suggested Readings

1. A. S. Hall and R. W. Woodhead, Frame Analysis, John Wiley & Sons (1961)
2. H. C. Martin, Introduction to Matrix Methods of Structural Analysis, McGraw-Hill (1966)
3. J.H. Argyris, S. Kesley, Energy Theorems and Structural Analysis (Butterworths, London, 1960)
4. C. K. Wang, Introductory Structural Analysis with Matrix Methods, Prentice Hall (1973)
5. W. McGuire, R. H. Gallagher and R. D. Ziemian, Matrix Structural Analysis, Second Edition,
John Wiley & Sons (2000)
6. J. M. Gere and W. Weaver Jr, Analysis of Framed Structures, Second Edition, D. van Nostrand
(1982)
Chapter 7
Elements of Elasticity

7.1 Introduction

The theory of elasticity provides a more sophisticated tool for stress analysis than
the strength of materials approach. It tackles the problems in a more rigorous way.
Though the analysis caters more detailed aspects of the problem, the solution for
many of the problems is anything but simple. With the availability of high-speed
computers, it becomes somewhat a luxury to solve most of the structural problems
by the theory of elasticity approach.
For the basic formulation of plate and shell structures, the knowledge of the theory
of elasticity, however, is needed. Even any numerical analysis of plates and shells
requires those basic equations. Many books exist where the theory of elasticity, plates
and shells have been treated in detail to which an interested reader can refer [1–5].
This chapter, however, is limited only to the introduction of the basic relationships
which will be needed for the finite element analysis.

7.2 Some Notations and Relations in the Theory

of Elasticity

7.2.1 Surface and Body Forces

The force distributed over the surface of the body is called surface force, e.g.
hydrostatic pressure, frictional force, etc.
The force which is distributed over the volume of the body is called the body
force, e.g. self weight of a structure, magnetic force, inertia force, etc.

© The Author(s) 2022 149

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_7
150 7 Elements of Elasticity

7.2.2 Components of Stresses

The stress acting on an area can be split into a normal component and a tangential
component. If at point o of an element of sides dx, dy and dz which are parallel to
reference x-, y- and z-axis system is taken out of a stressed body, one normal stress
component and two shear stress components act all parallel to the reference axis
system. The components of the stresses along with the axis system are shown in
Fig. 7.1.
The sign convention adopted for normal stress is that it is positive when it produces
tension; therefore, if it produces compression, it is negative. Shear stresses shown in
Fig. 7.1 are accompanied by two subscripts—the first indicating the plane given by
its normal on which it acts and the second gives the direction of the component of
the stress. To explain it further, if we consider the two shearing stress components
acting on the plane which is perpendicular to x-axis, then the component which is in
the direction of y-axis will be τx y , and that in the z-direction will be τx z . The positive
directions of the components of shearing stress on any side of the element are taken
as the positive directions of the coordinate axes if a tensile stress on the same plane
would have the positive direction of the corresponding axis. Thus, if on any side, the
tensile stress is opposite to the positive direction of the reference axis, the positive
directions of the shearing stress components will then be reversed. All the stress
components shown in Fig. 7.1 are positive.
It is evident from Fig. 7.1 that there are three components of normal stress, σx ,
σ y and σz , and six components of shear stress, τx y , τ yz , τzx , τ yx , τzy and τx z . Further,
it is known that τx y = τ yx , τx z = τzx and τ yz = τzy . Therefore, to describe the
stress at a point, the information of six quantities σx , σ y , σz , τx y , τ yz , τzx are all that
is needed.

Fig. 7.1 Stressed element z σz

τzx
τxz
τzy

σx

τxy τyz τxy

O
A
σx τyx
dz
τxz σy dy

B dx C
7.2 Some Notations and Relations in the Theory of Elasticity 151

Fig. 7.2 Deformation of the dx

element
O A x
O′ v ∂y
v+ dx
A′ ∂x
dy u

∂u B′
B u + ∂y dy
y

7.2.3 Components of Strain

The deformation of the face OACB is shown in Fig. 7.2. The displacements of point O
of the deformed body are resolved into components u, v and w parallel to the reference
axes x, y and z respectively, for the infinitesimal element in Fig. 7.1. The side will
have two types of deformation. Point A will have displacement in the x-direction
which is
Therefore, the increase in length of the element whose length is dx, is ∂∂ux d x.
Hencc, thc strain in x-direction is ∂∂ux . Similarly, it can be shown that the strains in y-
and z-directions are ∂u∂y
and ∂w
∂z
respectively.
Point A also has a component of the displacement in the y-direction which is
v + ∂∂vx d x. Similarly, the displacement of point B in the x-direction is u + ∂u ∂y
dy.
∂v
Therefore, inclination of O’A’ from OA is ∂ x and that of O B from OB is ∂u
, ,
∂y
.
( )
∂v ∂u
Therefore, change in ∠AO B after deformation is ∂ x + ∂ y . By definition, this is
the shearing strain between the planes xz and yz. Similarly, other shearing strain
components can be determined.
Introducing the letter e indicates the normal strain and y for the shearing strain,
the following then is the strain–displacement relationship.
)
εx = ∂∂ux , ε y = ∂v
∂y
, εz = ∂w∂z
(7.1)
∂u
γx y = ∂ y + ∂ x , γ yz = ∂v
∂v
∂z
+ ∂w
∂y
, γzx = ∂w
∂x
+ ∂u
∂z

7.2.4 Stress–Strain Relationship

For a linearly elastic material, the following are the stress–strain relationships:
152 7 Elements of Elasticity
⎫
εx = E1 [σx − ν(σ y + σz )] ⎪
⎪
⎬
ε y = E1 [σ y − ν(σz + σx )]
εz = E1 [σz − ν(σx + σ y )] ⎪
⎪
τ τ τzx ⎭
γx y = Gx y , γ yz = Gyz , γzx = G

where v is the Poisson’s ratio of the material, E is Young’s modulus of elasticity and
G is the shear modulus of elasticity. Again E and G are related as follows:

E
G=
2(1 + ν)

7.3 Two-Dimensional Problems

7.3.1 Plane Stress

If a thin plate is loaded in its own plane in such a way that the forces are distributed
uniformly over the thickness, then the plate is said to be in a state of plane stress.
For the plate in Fig. 7.3, the stress components σx , σ y , and τx y represent the state of
stress and stress components σz = τ yz = τzx = 0.
Beams, buttress of dams, web frame of tankers and superstructure deck interaction
problems of ships are some of the examples of plane stress problems. The stress–strain
relations for the plane stress problem then become
⎧ ⎫ ⎡ ⎤⎧ ⎫
⎨ εx ⎬ 1
1 −ν 0 ⎨ σx ⎬
ε = ⎣ −v 1 0 ⎦ σy (7.2)
⎩ y ⎭ E ⎩ ⎭
γx y 0 0 2(1 + ν) τx y

Fig. 7.3 Thin plate under

plane stress
7.3 Two-Dimensional Problems 153

Equation (7.4) has resulted after making the necessary substitution of plane stress
conditions into Eq. (7.2) and replacing G in terms of E from Eq. (7.3). In compact
form, Eq. (7.4) can be written as

{ε} = [S]{σ } (7.3)

Therefore,

{σ } = [D]{ε} (7.4)

where
⎡ ⎤
1ν 0
E ⎣
[D] = [S]−1 = ν1 0 ⎦ (7.5)
1 − ν2
0 0 1−ν
2

7.3.2 Plane Strain

If the dimension of the body in the z-direction is very much larger in comparison
to other two dimensions, the section at any point of the body perpendicular to the
z-axis is constant and the necessary forces applied to the body do not vary along the
length, it is assumed that the deformation of the plane which is perpendicular to the
z-axis becomes independent of its distance from the z-axis, i.e. all cross sections are
assumed to be in the same condition. For this condition, longitudinal displacement
w is zero. Further, γ yz = γzx = εz = 0. For the analysis of this body, two sections at
a unit distance apart are then considered.
Culvert or tunnel, dam, retaining wall and cylindrical tubes subjected to internal
pressure are some of the examples of plane strain problems.
Since γ yz = γzx = 0, then from Eq. (7.2), τ yz = τzx = 0. σz , however, is not zero.
Since εz = 0, then from Eq. (7.2),

σz − ν(σx + σ y ) = 0

or

σz = ν(σx + σ y ) (7.6)

Thus, the plane strain problem like the plane stress problem reduces to the deter-
mination of σx , σ y and τx y as a function of x and y only. Substituting the values of
σz from Eq. (7.8) into the first two equations of Eq. (7.2) results in
154 7 Elements of Elasticity
)
εx = 1−ν 2
E
σx − υ(1+ν)
E
σy
(7.7)
εy = 1−ν 2
E
σ y − v (1+ν)
E
σx

Therefore, the stress–strain relation of a plane strain problem is

⎧ ⎫ ⎡ ⎤⎧ ⎫
ν
⎨ εx ⎬ 1 − ν 2 1 − 1−ν 0 ⎨ σx ⎬
⎢ ν ⎥
ε = ⎣ − 1−ν 1 0 ⎦ σy (7.8)
⎩ y ⎭ E ⎩ ⎭
τx y 0 2
0 1−ν τx y

or

{ε} = [S]{σ } (7.9)

Equation (7.11) can be rewritten as

{σ } = [D]{ε} (7.10)

where
⎡ ⎤
ν
1 0
E(1 − ν) ⎢ ν 1−ν
⎥
[D] = ⎣ 1 0 ⎦ (7.11)
(1 − 2ν)(1 + ν) 1−ν 1−2ν
0 0 2(1−ν)

Therefore, it is seen that there exists a great similarity between plane stress and
plane strain problems. It can further be verified that [D] matrix of plane stress can be
reduced to [D] matrix of plane strain by replacing E and υ in the former expression
by

E , ν
E, = and υ =
1 − ν2 1−ν

7.3.3 Differential Equations of Equilibrium

The rectangular element in Fig. 7.4 is in equilibrium in the stressed condition.

Stresses acting at the different faces of the element are shown in Fig. 7.4. Body
forces per unit volume acting in X and Y directions are Fx and Fy respectively. The
equation of equilibrium of the forces in the x-direction is
( ) ( )
∂σx ∂τx y
σx + d x dy − σx d x + τx y + dy d x − τx y d x + Fx d x d y = 0
∂x ∂y
7.3 Two-Dimensional Problems 155

Fig. 7.4 Element in equilibrium

or,

∂σx ∂τx y
+ + Fx = 0 (7.12)
∂x dy

Similarly, the equilibrium of the forces in the y-direction will result in the other
equation

∂τx y ∂σ y
+ + Fy = 0 (7.13)
dx dy

It may be noted that the third equilibrium equation, i.e. the moment of the forces
about any point in space is zero, results in τx y = τ yx . This has already been considered
and accordingly, no distinction between τx y and τ yx has been made as is evident in
Fig. 7.4.
It may, further, be seen that a two-dimensional theory of elasticity problem
involves three unknowns—σx , σ y and τx y . Only two equilibrium equations are avail-
able, therefore, the problem is statically indeterminate. Further discussions regarding
the solution of plane problems will not be made, as they are not needed as far as the
finite element analysis is concerned. However, before the topic is closed, the equi-
librium equations of three-dimensional problems are given which are nothing, but
the extension of eqns. (7.14) and (7.15).
∂τ ⎫
∂σx
∂x
+ ∂ yx y + ∂τ∂ zx z + Fx = 0 ⎪
⎬
∂τ yx ∂σ ∂τ
∂x
+ ∂ yy + ∂ zyz + Fy = 0 (7.14)
∂τzx ∂τ ⎪
⎭
∂x
+ ∂ yzy + ∂σ∂z
z
+ Fz = 0
156 7 Elements of Elasticity

7.4 Bending of Thin Plates

The theory of bending of plates which assumes the material of the plate to be linearly
elastic and the deformation of the plate to be small is known as the small deflection
theory of plates or simply as the classical theory of thin plates.
The plane parallel to the two faces of the plate and bisecting the thickness of the
plate in the undeformed state is called the middle plane or middle surface of the
plate. A body having the middle surface in the form of a plane and whose thickness
is sufficiently small compared to its other two dimensions is called a thin plate.

7.4.1 Basic Assumptions

The classical theory of thin plates is based on the following assumptions:

(1) A normal to the middle plane of the plate before bending transforms into a
normal to the middle plane after bending.
(2) The normal stress component perpendicular to the plane of the plate is small
compared with other stress components and is neglected in the stress–strain
relationship.
(3) The middle plane of the plate remains unstrained after bending.

7.4.2 Deformation of the Plate

The coordinate system shown for the plate is given as x- and y-axes in the middle
plane of the plate and z-axis perpendicular to the middle plane.
Figure 7.5 indicates the section of a plate parallel to xz-plane. After bending, a
point P on the middle plane is deflected to P , with a deflection w. Another point Q,
at a distance z from the undeformed middle plane, is displaced to Q , which is on
the normal to the middle plane after bending. The displacement of the point Q in the
x-direction is given by.

∂w
u = −z (7.15)
∂x
Similarly,

∂w
v = −z (7.16)
∂y
7.4 Bending of Thin Plates 157

Fig. 7.5 Deformation of the

plate in xz-plane P
x
Z
Q
z

P′

Q′ ∂w
∂x

7.4.3 Strain–Displacement Relationship

Referring to Eq. (7.1) and combining it with Eqns. (7.17) and (7.18) yields
⎫
εx = ∂∂ux = −z ∂∂ xw2
2
⎪
⎬
ε y = ∂v = −z ∂∂ yw2
2

∂y (7.17)
⎪
⎭
γx y = ∂u + ∂∂vx = −2z ∂∂x ∂wy
2

∂y

7.4.4 Stress–Strain Relationship

Assumption (2) states that σz = 0. Further, τx z = τ yz = 0. Therefore, Eq. (7.2)

becomes
⎫
εx = E1 (σx − νσ y ) ⎬
ε y = E1 (σ y − νσx ) (7.18)
τ ⎭
γx y = Gx y

Solution of Eq. (7.20) yields

⎫
σx = 1−νE ⎪
2 (ε x + ν ε y ) ⎬

σ y = 1−ν 2 (ε y + ν εx )
E
(7.19)
( 1−ν ) ⎪
τx y = 1−ν
E
2 2
γx y ⎭

Substituting εx , ε y and γx y from Eq. (7.19) into Eq. (7.21) yields

158 7 Elements of Elasticity
( )⎫
∂ 2w ∂ 2w ⎪
σx = − 1−ν
E z
2 + ν ∂ y2 ⎪
( ∂ x2 )⎬
2

∂ w ∂ 2w
σ y = − 1−ν 2 ∂ y 2 + ν ∂ x 2 ⎪
E z (7.20)
−E z ∂ 2w
⎪
⎭
τx y = 1−ν 2 (1 − ν) ∂ x ∂ y

Once the stress components have been expressed in terms of the curvatures, the
bending moment and twisting moment per unit length of the plate can be obtained.
The bending moments have two components parallel to xz-plane and yz-plane
respectively. Thus,
( h/2
Mx = σx zdz (7.21)
−h/2

Substituting σx from Eq. (7.22) into Eq. (7.23) and noting that the deflection w is
not a function of z, Eq. (7.23) becomes
( 2 ) ( h/2 ( 2 )
E ∂ w ∂ 2w ∂ w ∂ 2w
Mx = − + ν z 2
dz = −D + ν (7.22)
1 − ν2 ∂ x 2 ∂ y2 −h/2 ∂x2 ∂ y2

where D denotes [Eh 3 /12(1 − ν 2 )] and is called the flexural rigidity of the plate.
Similarly, it can be shown that
( )
∂ 2w ∂ 2w
M y = −D + ν (7.23)
∂ y2 ∂x2

The twisting moment is given by

( h/2
Mx y = −M yx = − τx y · zdz
−h/2

or

∂ 2w
Mx y = D (1 − ν) (7.24)
dx dy

The negative sign of the integral has been put because of the fact that for positive
τx y and positive z, d Mx y is negative.
The moment–curvature relationship in matrix notations then is
⎧ ⎫ ⎡ ⎤⎧ ∂ 2 w ⎫
⎨ Mx ⎬ 1ν 0 ⎨ ⎪ − ∂x2 ⎪
⎬
M y = D⎣ ν 1 0 ⎦ − ∂ y2
∂2w
(7.25)
⎩ ⎭ ⎪
⎩ 2 ∂ 2w ⎪
⎭
Mx y 0 0 1−ν
2 ∂x ∂y

In compact form, it is written as

7.4 Bending of Thin Plates 159

{M} = [D]{k} (7.26)

It may be mentioned here that Eq. (7.27) is valid for isotropic plate only. For
orthotropic plates, where the plate has two different properties in two orthogonal
directions, the moment–curvature relationship becomes
⎧ ⎫ ⎡ ⎤⎧ ∂ 2 w ⎫
⎨ Mx ⎬ D x D1 0 ⎪ ⎨ − ∂2x 2 ⎪
⎬
M ⎣
= D1 D y 0 ⎦ − ∂∂ yw2 (7.27)
⎩ y ⎭ ⎪
⎩ ⎪
⎭
2 ∂∂x ∂wy
2
Mx y 0 0 Dx y

7.4.5 Equilibrium Equations

Though it has been assumed that plane sections normal to the middle plane before
bending remain plane after bending which signifies that τx z and τ yz are negligible,
yet the shear forces acting at the different edges of the element will be considered.
All the necessary forces in the element of a plate are shown in Fig. 7.6. A lateral load
p per unit area acts on the plate.
Considering the equilibrium of all the forces in the z-direction yields
( ) ( )
∂ Qx ∂ Qy
Qx + d x dy − Q x dy + Q y + dy d x − Q y d x + p d x dy = 0
∂x ∂y
or

Qy
Qx
M yx
M xy Mx
y x
z P My

dy
dx ∂ Mxy
∂My Mxy + dx
My + dy ∂x
∂y
∂ Qx
Qx + dx
∂ M yx ∂x
M yx + dy ∂Mx
∂y Mx + dx
∂ Qy ∂x
Qy + dy
∂y

Fig. 7.6 Forces acting on the plate

160 7 Elements of Elasticity

∂ Qx ∂ Qy
+ +p=0 (7.28)
∂x dy

Taking moments of all the forces acting on the element with respect to the x-axis
and neglecting higher order terms,
( ) ( )
∂ Mx y ∂ My
Mx y + d x dy − Mx y dy + M y d x − M y + dy d x + Q y d x d y = 0
∂x ∂y
or
∂ Mx y ∂ My
− + Qy = 0 (7.29)
∂x ∂y

Similarly, taking moment of all the forces on the y-axis will yield

∂ Mx ∂ M yx
+ − Qx = 0 (7.30)
∂x ∂y

Equations (7.30), (7.31) and (7.32) can be combined into a single equation given
by

∂ 2 Mx ∂ 2 Mx y ∂ 2 My
− 2 + + p = 0--o (7.31)
∂x2 ∂x ∂y ∂ y2

7.4.6 Differential Equation for Deflection

Substituting Mx , M y and Mx y in terms of curvatures as given by Eq. (7.27) into

Eq. (7.33) results
[ ( )] [ ] [ ( )]
∂2 ∂ 2w ∂ 2w 2∂ 2 ∂ 2w ∂2 ∂ 2w ∂ 2w
−D + ν − D(1 − ν) + −D + ν +p=0
∂x2 ∂x2 ∂ y2 ∂x ∂y ∂x ∂y ∂ y2 ∂ y2 ∂x2

or

∂ 4w ∂ 4w ∂ 4w p
+ 2 + = (7.32)
∂x2 ∂ x 2∂ y2 ∂ y4 D

Equation (7.34) is known as the plate equation.

7.5 Boundary Conditions 161

7.4.7 Shearing Forces

Equation (7.32) is rewritten as

∂ Mx ∂ Mx y
Qx = − (7.33)
∂x ∂y

Substituting M x and M y from Eq. (7.27) into Eq. (7.35) yields

( )
∂ 3w ∂ 3w
Q x = −D + (7.34)
∂x 3 ∂ x ∂ y3

Similarly, it can be shown that

( )
∂ 3w ∂ 3w
Q y = −D + (7.35)
∂x ∂y
2 ∂ y3

7.5 Boundary Conditions

The boundary conditions discussed below are with respect to the rectangular plate.

7.5.1 Simply Supported Edge

If the edge x = a is simply supported, the deflection and bending moment along the
edge must be zero. That is,
( )
∂ 2w ∂ 2w
(w)x=a = 0 and (Mx )x=a = −D + ν =0
∂x2 ∂ y2

But w = 0 along the edge x = a implies that

∂w ∂ 2w
= = 0 along this edge
∂y ∂ y2

The boundary conditions for a simply supported edge are therefore equivalent to
( )
∂ 2w
(w)x=a = 0 and =0 (7.36)
∂x2 x=a
162 7 Elements of Elasticity

7.5.2 Clamped Edge

If an edge at x = a is clamped, the deflection, as well as the slope of the middle

plane, is zero, that is,
( )
∂w
(w)x=a = 0 and =0 (7.37)
∂x x=a

7.5.3 Free Edge

If the edge x = a is free, then the bending moment, the twisting moment and the
vertical shearing force at that edge are zero, i.e.

(Mx )x=a = 0, (Mx y )x=a = 0 and (Q x )x=a = 0 (7.38)

It was shown by Kirchoff that three boundary conditions are too many and that two
conditions are sufficient for a complete determination of deflection. Kirchoff replaced
two conditions prescribing M xy and Qx by a single equation. This is explained below.
On the edge x = a, for an element of length dy, the twisting moment is M xy dy,
which may be replaced by two vertical forces of magnitude M xy , at a distance dy
apart. For the
[ immediate next ] element of length dy, the twisting moment assumes a
∂M
magnitude Mx y + ∂ yx y dy dy, which may be replaced by vertical forces of magni-
∂ Mx y
tude Mx y + ∂y
dy as shown in Fig. 7.7 by dotted lines. At point H, there acts,
∂M
therefore, a net vertical force of magnitude ∂ yx y For positive M xy , this statically
equivalent vertical force is acting upward which is negative according to the sign
convention. Distribution of M xy is, therefore, equivalent to a distribution of shearing
forces of intensity.
( )
∂ Mx y
Q ,x = − (7.39)
∂ y x=a

Hence, the shearing forces along a free edge x = a becomes

( )
∂ Mx y
(Vx )x=a = Qx − =0 (7.40)
∂y x=a

Replacing the values of Q x and Mx y from Eq. (7.36) and Eq. (7.27) respectively
in Eq. (7.42) yields
7.5 Boundary Conditions 163

z ∂ M xy
M xy + dy M xy
∂y

dy
dy
M xy

y ∂ M xy
M xy + dy
∂y

Fig. 7.7 Edge effect of torsional moment

[ ]
∂ 3w ∂ 3w
+ (2 − ν) =0 (7.41)
∂x3 ∂ x ∂ y2 x=a

The other boundary condition to be satisfied at a free edge is

(Mx )x=a = 0

[ ]
∂ 2w ∂ 2w
+ ν =0 (7.42)
∂x2 ∂ y2 x=a

Equations (7.43) and (7.44) may be taken as boundary conditions at a free edge.

7.5.4 Elastically Supported Edge

If an edge x = a is elastically supported and the translational stiffness or spring

constant is K s , shown in Fig. 7.8(a), the boundary conditions for this edge will be
[ ]
D ∂ 3w ∂ 3w
(w)x=a =± + (2 − ν) (7.43)
Ks ∂ x 3 ∂ x ∂ y 2 x=a

and
[ ]
∂ 2w ∂ 2w
(Mx )x=a = + ν =0 (7.44)
∂x2 ∂ y2 x=a
164 7 Elements of Elasticity

Fig. 7.8 Elastically translational and rotational supports

If this elastic support is provided by a beam supporting the edge of the plate, then
Eq. (7.43) becomes
[ ] [ ]
∂ 4w ∂ 3w ∂ 3w
E Ib 4 = ∓D + (2 − ν) (7.45)
∂y x=a ∂x3 ∂ x ∂ y2 x=a

where I b represents the moment of inertia of the beam.

7.5.5 Edge Having Elastic Rotational Restraint

If an edge has an elastic rotational restraint k r , then the boundary conditions will be

(w) x=a = 0

and
( ) [ ]
∂w D ∂ 2w ∂ 2w
=∓ + ν (7.46)
∂x x=a kr ∂ x 2 ∂ x 2 x=a

7.6 Concluding Remarks

No attempt will be made to solve the plate differential Eq. (7.34). The closed-bound
solutions for various types of boundary conditions for some plates of certain shapes
References and Suggested Readings 165

under typical loading conditions have been obtained and results have been presented
in a tabulated or graphical form [4, 6].
Here, only the derivation of the basic relationship of the plane stress and plate
bending problems of the plate has been carried out which will be required in the
subsequent chapters when the finite element analysis is presented. For the same
reason, no exercise is given in this chapter.

References and Suggested Readings

1. A. Green and W. Zerna, Theoretical elasticity, second edition, Oxford University Press, 1968.
2. Fifth Conference on Electronic Computation, Proc. ASCE, J. Strl. Divn., 97, Jan., 1971.
3. I. S. Sokolnikoff, Mathematical Theory of Elasticity, McGraw-Hill, 1966.
4. R. Szilard, Theory and analysis of plates: classical and numerical methods, Prentice Hall, 1974.
5. S. Timoshenko and J. N. Goodier, Theory of elasticity, second edition, McGraw Hill, 1951.
6. S. Timoshenko and S. Woinowsky-Krieger, Theory of plates and shells, second edition, McGraw-
Hill, 1959.
Chapter 8
Introduction to the Finite Element
Method

8.1 Introduction

In all the previous chapters, we have seen how the matrix methods are applied to one-
dimensional or skeletal structures. For the analysis, we consider one member of the
structure at a time. The choice of the element is obvious—the member is the discrete
element considered. However, when we focus our attention on a continuum, such
as two-dimensional structures like plates and shells, we are immediately faced with
the difficulty regarding the choice of discretization, as no obvious choice exists. In
such cases, like conventional one-dimensional structures which are considered as an
assemblage of structural elements interconnected at their intersections, the continuum
will be assumed to consist of imaginary divisions. Thus, the real continuum is divided
into a finite number of discrete elements as shown in Fig. 8.1.

8.2 The Finite Element Method

In a continuum that is divided into a mesh, two adjacent elements placed side by
side will have a common edge (Fig. 8.1). Its true representation will result in a
more complicated analysis. Therefore, in order to make the analysis simpler, it is
assumed that the elements are connected at the nodal points and it is only there
that the continuity requirement is to be satisfied. Once the discretization is made,
the analysis follows a rather set procedure. The analysis presented here regarding
the finite element method is throughout based on the displacement approach. The
stiffness matrix of the individual elements needs to be formulated. The forces actually
distributed in the real structure are transformed to act at nodal points. In the finite
element analysis, the continuum is divided into a ‘finite’ number of elements, having
‘finite’ dimensions and reducing the continuum having ‘infinite’ degrees of freedom
to ‘finite’ degrees of freedom. The term ‘finite element’ was coined by Clough [1].

© The Author(s) 2022 167

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_8
168 8 Introduction to the Finite Element Method

Fig. 8.1 A continuum Boundary of region

divided into finite elements of interest
Typical node

y
x

Typical element

It has been found that the accuracy of the solution in general increases with the
increase in the number of elements considered. However, involving more number of
elements will take more computer time for obtaining the solution. A question that
arises in the mind of the analyst is what should be the optimum number of divisions
that would give a satisfactory solution. The answer given by the experts to this query
is that (a) the required subdivision of the continuum is to be based on past experience
and (b) if an analysis is attempted for the first time, then convergence is to be tested
for varied mesh gradings. For regions where a higher concentration of stresses is
expected, such as around the openings, a finer mesh division is to be made (Fig. 8.2).
For the last few years, adaptive finite element procedures have been developed to
generate a mesh which would indicate accurate value.
In the finite element method, the problem is formulated in two stages:
(a) the element formulation and
(b) the system formulation.
The first stage involves the derivation of the element stiffness matrix that yields
a relationship between nodal forces and nodal displacements. The next stage is the

Fig. 8.2 Mesh division of a

plate with hole
8.3 Brief History of the Development of the Finite Element Method 169

formulation of the stiffness and loads of the entire structure. The resulting solution
of it yields the deformation and the internal forces at the points desired.

8.3 Brief History of the Development of the Finite Element

Method

Although the name ‘finite element’ is of recent origin, the concept has been used
for centuries. The basic philosophy is to replace the actual problem with a simpler
model which will closely approximate the solution of the problem at hand.
One of the earlier problems dealt with this concept was by ancient mathematicians,
who found the circumference of the circle by approximating it as a polygon. Each
side is a finite element according to present-day notations. Depending on whether the
approximating polygon is inscribing or circumscribing the circle, a lower or upper
bound solution to the true circumference can be obtained. Further, as the number
of sides of the polygon is increases, the approximate value of the circumference
converges towards the true value. The finite element concept led to an accurate
evaluation of the unknown quantities in those early days as can be found in ancient
records.
The development of the finite element method has been carried out in two stages.
In the initial stages, the theory was formulated on the basis of applying the elementary
theory of structural analysis with intuitive reasoning. Argyris and Kesley [2] have
pioneered the concept of applying energy principles to the formulation of structural
analysis problems. The starting point of the finite element method is a paper by
Turner, Clough, Martin and Topp [3]. It presented the application of simple finite
elements (pin-joined bar and triangular plate with in-plane load) for the analysis of
aircraft structures.
The second stage of the development started with a more rigorous study of the
convergence aspect of the finite element method. One such study by Melosh [4] led
to the formulation of the finite element method based on the principle of minimum
potential energy. Soon after that, de Veubeke [5] introduced stress or equilibrium
elements based on the principle of minimum complementary energy. More or less
at the same time, the concept of hybrid element was established by using the dual
principles of minimum potential energy and minimum complementary energy was
put forward by Pian [6].
The concept that the formulation of the finite element method can be set up
by the variational principles, led to a revolution in its application. The structural
analysis problems could now be placed on a more general footing. Load acting on
the structure, be it a body force or a surface force can now be elegantly expressed
in terms of generalized nodal forces. Further, the conversion of the physical nature
of the problem into a mathematical model freed the method from its only field of
application, structural mechanics and paved the way for its application to other fields.
170 8 Introduction to the Finite Element Method

In fact, any physical problem for which a variational formulation is possible can now
be solved by the finite element analysis.
According to the type of models of the element which exist, they are of three types-
displacement model based on the principle of minimum potential energy, the stress
or equilibrium model based on the principle of minimum complementary energy,
and the hybrid model which takes into account the dual nature of the variational
principles.
During the last two decades, the finite element method has made rapid strides. The
theories are now placed on a firm mathematical footing. The generality and versatility
of the method are now thoroughly exposed. A number of international conferences
have been held in different parts of the world [7–21, 40–44]. As far as the structural
analysis is concerned, from the initial attention focused on the elastic analysis of
plane stress and plate bending problems, the method has been successfully extended
to the cases of analysis of three-dimensional structures [5, 45, 46], curved structures
[22, 47–49], stability and vibration problems [23–28, 50–53], nonlinear analysis
taking into account the large deflection and material nonlinearity either individually
or combined [29–31, 54–56].

8.4 Basic Steps in the Finite Element Method

for the Solution of Static Problems

The following are the steps adopted for analysing a structural engineering static
problem by the finite element method [32, 33].
Step 1. Discretization of the domain
The continuum is divided into a number of finite elements by imaginary
lines or surfaces. The interconnected elements may have different sizes and
shapes. The success of this idealization lies in how closely this discretized
continuum represents the actual continuum. The choice of the simple
element or higher order element, straight or curved, its shape, refinement
is to be decided before the mathematical formulation starts.
Step 2. Identification of variables
It is assumed that the elements are connected at their intersecting points
referred to as nodal points. At each node, unknown displacements are to
be prescribed. They are dependent on the problem at hand. For example,
in a plane stress problem, the minimum unknowns are two linear transla-
tions at each node. In a simpler model of the plate bending problem, the
unknowns at a node are the deflection and two rotations. In ordinary cases,
to every unknown displacement at a node, there corresponds a nodal force
which can be physically visualized. But it has already been discussed that
the finite element method transcends the physically based reasoning. The
problem may be identified in such a way that in addition to the displacement
8.4 Basic Steps in the Finite Element Method for the Solution of Static Problems 171

which occurs at the nodes depending on the physical nature of the problem,
certain other quantities such as ‘strain’, ‘curvature’, etc., may need to be
specified as nodal unknowns for the element, which, however, may not have
a corresponding physical quantity in the generalized forces. The value of
these quantities can, however, be obtained from variational principles.
Step 3 Choice of approximating functions
Once the variables and local coordinate system have been chosen, the next
step is the choice of the displacement function. In fact, it is the displace-
ment function that is the starting point of the mathematical analysis. This
function represents the variation of the displacement within the element.
The function can be approximated in a number of ways. The displacement
function may be approximated in the form of a linear function or higher
order functions. A convenient way to express it is by using polynomial
expressions. A typical question is posed in the finite element analysis, that
is, the solution may converge to the exact values either by increasing the
degrees of the polynomial or by decreasing the element size. This aspect
will be discussed later.
The shape of the element or the geometry may also be approximated.
The coordinates of corner nodes define the element shape accurately if the
element is actually made of straight lines or planes. This simplification of
geometry may sometimes lead to wrong results, e.g. flat elements in a shell.
The analyst will also have to decide on the weightage to be given to the
geometry and displacements for a particular problem.
Step 4 Formation of the element stiffness matrix
After the continuum is discretized with desired element shapes, the element
stiffness matrix is formulated. This can be done in a number of ways. Basi-
cally, it is a minimization procedure whatever may be the approach adopted.
For certain elements, the form involves a great deal of sophistication. With
the exception of a few simple elements, the element stiffness matrix for
majority of elements is not available in explicit form. As such, they require
numerical integration for their evaluation.
The geometry of the element is defined with reference to the global
frame. In many problems such as those of rectangular plates, the global
and local axis systems are coincident and, for them, no further calculation
is needed at the element level beyond computation of element stiffness
matrix in local coordinates. Coordinate transformation must be done for all
elements where it is needed.
Step 5 Formation of the overall stiffness matrix
After the element stiffness matrices in global coordinates are formed, they
are assembled to form the overall stiffness matrix. The assembly is done
through the nodes which are common to adjacent elements. At the nodes,
the continuity of the displacement function and possibly their derivatives
are established. The overall stiffness matrix is symmetric and banded.
172 8 Introduction to the Finite Element Method

Step 6 Incorporation of boundary conditions

The boundary restraint conditions are to be imposed in the stiffness matrix.
There are various techniques available to satisfy the boundary conditions.
In some of these approaches, the size of the stiffness matrix may be reduced
or condensed in its final form. To ease the computer programming aspect
and elegantly incorporate the boundary conditions, the size of the overall
stiffness matrix is kept the same.
Step 7 Formation of the element loading matrix
The loading forms an essential parameter in any structural engineering
problem. The loading inside an element is transferred at the nodal points
and a consistent element loading matrix is formed. Sometimes, based on
the typicality of the problem, the loading matrix may be simplified.
Step 8 Formation of the overall loading matrix
Like the overall stiffness matrix, the element loading matrices are assem-
bled to form the overall loading matrix. This matrix has one column
per loading case and it is either a column vector or a rectangular matrix
depending on the number of loading conditions.
Step 9 Solution of simultaneous equations
All the equations required for the solution of the problem are now devel-
oped. In the displacement method, the unknowns are the nodal displace-
ments. Rarely is the stiffness matrix stored as a solid matrix. Advantages
are taken from the symmetric nature of the problem and banded properties.
The Gauss elimination and Cholsky’s factorization are the most commonly
used procedures for the solution of simultaneous equations. These methods
are well suited to a small or moderate number of equations. For large-sized
problems, frontal technique is one of the methods of obtaining solutions.
For systems of large order, Gauss–Seidel or Jacobi iterations are more
suited.
Step 10 Calculation of stresses or stress resultants
In the previous step, nodal displacements are calculated and these values
are utilized for the calculation of stresses or stress resultants. This may be
done for all elements of the continuum or it may be limited only to some
predetermined elements.
Results may be obtained by graphical means. It may be desirable to
plot the contour of the deformed shape of the continuum. The contour of
the principal stresses may be one of the sought-after items for a certain
category of problems.
8.5 Advantages and Disadvantages of the Finite Element Method 173

8.5 Advantages and Disadvantages of the Finite Element

Method

Physical problems result in differential equations. To obtain the exact solution of

them in closed-bound form poses a formidable challenge. Numerical methods are
employed to obtain an approximate solution. There are many approximate methods
such as Rayleigh–Ritz method, Galerkin method, the least square method and the
finite difference method that are often used to obtain a necessary solution. The finite
element method stands superior to all of them. The main advantage of the finite
element method is that in this method, approximations are confined to relatively
small subdomains, whereas, in other methods, the admissible function satisfies the
boundary condition of the entire domain which becomes extremely difficult when the
domain has an irregular shape. In the finite element method, the admissible functions
are valid over the simple domain and have nothing to do with the boundary, however,
simple or complex it may be.
Thus, the main advantage of the finite element method can be put in one sentence –
the physical problems which were so far intractable and complex for any closed-
bound solution can now be analysed by the finite element method. The advantages
in relation to the complexity of the problem are stated below:
1. The method can be efficiently applied to cater irregular geometry. It can take care
of any type of boundary.
2. Material anisotropy and nonhomogeneity can be catered without much difficulty.
3. Any type of loading can be handled.
Based on the above discussion, one may form the idea that the finite element
method is the most efficient for the analysis of any type of structural engineering or
physical problem. There are many types of problems where some other method of
analysis which may prove to be more efficient than the finite element method. One
of the major disadvantages of the method is the time involved in the solution of the
problem. In many problems related to vibration and stability, the time consumed in
carrying out the finite element analysis may turn out to be very high.
It may therefore be a luxury to undertake vibration and stability analysis of simpler
structures where the application of even simpler computer methods such as the finite
strip [34] or other semianalytical methods [35–37] will lead to more economical
solution. But those methods will work within their limitations and will not be as
versatile as the finite element method.
In many cases, one is so much carried away by the versatility of the finite element
method that one loses sight of its limitations. It must be remembered that in whatever
sophisticated manner the problem might have been formulated and solved, it has been
done so within the frame of its assumptions. Proper engineering judgment is to be
exercised in interpreting the results.
In the finite element method, if the size of the problem is relatively large, it may
lead to round-off errors. A computer works with a limited number of digits. Solving
the problem on the basis of restricted numbers of digits may not yield the desired
174 8 Introduction to the Finite Element Method

degree of accuracy or it may give erroneous results in certain cases. The magnitude of
these round-off errors varies with the problem, the problem description and computer
configuration. For many problems, however, the increase in number of digits for the
purpose of calculation improves the accuracy. An error analysis can be made to this
effect [38, 39].
When a continuum is discretized, an infinite degree of freedom system is converted
into a model having a finite number of degrees of freedom. In continuum functions
which are continuous are now replaced by ones which are piece-wise continuous
within individual elements. Thus, the actual continuum is represented by a set of
approximations.
Injudicious choice of the shape of the element will lead to a considerable error in
the solution. Triangular elements in the shape of equilateral or rectangular elements
in the shape of a square will always perform better than those having unequal length
of the sides. For very long shapes, the attainment of convergence will be extremely
slow.
Simplification of the actual boundary is another source of error. The domain may
be reduced to the shape of a polygon. If the mesh is refined, then the error involved
in the discretized boundary may be reduced.

References and Suggested Readings

1. R. W. Clough, The finite element method in plane stress analysis, Proc. Second Conf. Electronic
Computation, ASCE, Pittsburg, 1960.
2. J.H. Argyris, S. Kesley, Energy Theorems and Structural Analysis (Butterworths, London,
1960)
3. M. J. Turner, R. W. Clough, H. C. Martin and L. J. Topp, Stiffness and deflection analysis of
complex structures, J. Aero. Sc., 23, Sept., 1956, pp. 805–824.
4. R. J. Melosh, Basis of derivation of matrices for direct stiffness method, J. AIAA, 1(7), 1963.
5. B. F. de Veubeke, Upper and lower bounds in matrix structural analysis, in “Matrix Methods
of Structural Analysis”, AGARD, 72, 1964, pp. 165−201.
6. Y. K. Cheung, The Finite Strip Method in Structural Analysis, Pergamon Press, 1976.
7. Conference on “The Use of Digital Computers in Structural Engineering”, The University of
New Castle, England, July, 1966.
8. S. J. Fenves, N. Perrone, A. R. Robinson and W. C. Schnobrich (Editors), Numerical and
Computer Methods in Structural Mechanics, Academic Press, 1972.
9. Fifth Conference on Electronic Computation, J. Strl. Divn., Proc. ASCE, 97, Jan., 1971.
10. Finite Element Review, Advanced Engineering Corporation, Linkuping, Sweden, 1979.
11. R. H. Gallaghar, Y. Yamada and J. T. Oden (Editors), Recent Advances in Matrix Methods of
Structural Analysis and Design, Tokyo, Aug. 25–30, 1979, University of Alabama Press, 1980.
12. R. Glowinski, E. Y. Rodin and O. C. Zienkiewicz, Energy Methods in Finite Element Analysis,
John Wiley and Sons, 1979.
13. ISD/ISSC Symposium on Finite Element Techniques, Stuttgart, West Germany, June, 1969.
14. ISSC Symposium on “High Speed Computing in Elasticity”, Liege, Belgium, 1970.
15. J. C. McCutcheon, M. S. Mirza and A. A. Mufti (Editors), Proceedings on Specialized Confer-
ence on the Finite Element Method in Civil Engineering, (held at McGill University, Canada
1972), Engineering Institute of Canada, Ottawa, London, 1972.
16. J. T. Oden, O. C. Zienkiewicz, R. H. Gallaghar and C. Taylor (Editors), Finite Elements in
Fluids, v. I and II, John Wiley and Sons, 1979.
References and Suggested Readings 175

17. V. Palmano and A. Kabila (Editors), International Conference on Finite Element Methods,
University of New South Wales, Australia, 1975.
18. Proceedings of the First Conference on Matrix Methods of Structural Mechanics, Wright-
Patterson Air Force Base, Dayton, Ohio, AFFDL-TR-66–80, October 26–28, 1965.
19. Proceedings of the Second International Symposium in Computing Methods in Applied Science
and Engineering, Iria, Versailles, France, 1975.
20. Symposium on “Applied Finite Element in Civil Engineering”, Vanderbilt University,
Tennessee, USA, Nov. 1969.
21. O. C. Zienkiewicz, R. Lewis and E. Stagg (Editors), Numerical Methods in Offshore
Engineering, John Wiley & Sons, 1977.
22. G.A. Mohr, Numerically integrated triangular element for doubly curved thin shells. Comput.
Struct. 11, 565–573 (1980)
23. K. J. Bathe, Finite Element Procedures, Prentice-Hall, 1996.
24. A. Chajes, Principles of Structural Stability Theory, Prentice-Hall, 1974.
25. A.N. Chen, A finite element study and bifurcation and limit point buckling of elasto-plastic
arches. Comput. Struct. 60(2), 189–196 (1996)
26. R. H. Gallaghar, Finite element analysis fundamentals, Prentice-Hall, 1975.
27. J. Padavan, Nonlinear vibrations of general structures. J. of Sound and Vibration 72, 427–442
(1980)
28. G. B. Warburton, The dynamical behaviour of structures, Pergamon Press, 1976.
29. R. Szilard, Theory and analysis of plates, Prentice-Hall, 1974.
30. M.M. Shokrieh, L.B. Lessard, Effects of material non-linearity on the three-dimensional stress
state of pin-loaded composite laminate. J. Comp. Materials 30(7), 839–861 (1996)
31. D. Venugopal Rao, A. H. Sheikh and M. Mukhopadhyay, A finite element large deflection
analysis of stiffened plates, Computers and Structures, 47(6), 1993, pp. 987–993.
32. M. Mukhopadhyay, Free vibrations of rectangular plates having different degrees of elastic
restraint of rotation. J. Sound and Vibration 67(4), 459–468 (1979)
33. M. Mukhopadhyay, Vibrations, Dynamics and structural systems, Oxford and IBH, 2nd Edition,
2000.
34. D. D. Milasinovic, The finite strip in computational mechanics, faculties of civil engineering,
subotica, Belgrade, 2003.
35. M. Mukhopadhyay, Response of rectangular plates having different degrees in elastic restraint
in harmonic excitation. Ocean. Eng., Pergamon Press. 8(5), 497–505 (1981)
36. M. Mukhopadhyay, Buckling analysis of rectangular ship plating. Int. Shipbuild. Prog. 26(297),
89–97 (1979)
37. M. Mukhopadhyay, A semi-analytic solution for the free vibration of annular sector plates. J.
Sound Vib. 63(1), 87–95 (1979)
38. R. J. Melosh, Structural engineering analysis by finite elements, Prentice-Hall, 1990.
39. R. J. Melosh, Estimating manipulation errors for finite element analysis, finite elements in
analysis and design, 3(3), Oct. 1987.
40. 11th Annual workshop on “the finite element method in biomedical engineering, biomechanics
and related fields”, University of Ulm, Germany, 11–14 July, 2004.
41. Finite element circus, Fall 2002, Penn State University, Oct. 25–26, 2002.
42. International Conference on Finite Element Methods: superconvergence, post-processing and
a post-priori estimates, University of Jyvaskyla, Finland, July 1–6, 1996.
43. International Conference on the Finite Element Modelling, The Luxembourgh “Henry Tudor”
Research Laboratory, Luxembourgh, 13–14 November, 2003.
44. International Conference on “ p and hp Finite Element Methods: Mathematics and Engineering
Practice”, Washington University, May 31−June 2, 2000.
45. G. Dhondt, The finite element method for three-dimensional thermo-mechanical applications,
Wiley Europe, 2004.
46. Y.C. Yip, R.C. Averill, A three-dimensional laminated plate finite element with high order
zig-zag sublimate. J. Comput. Eng. Sci. 2(1), 137–180 (2001)
176 8 Introduction to the Finite Element Method

47. N. Rattanawangcharoen, H. Bai, A.H. Shah, A 3D cylindrical finite element model for thick
curved beam stress element. IJNME 59, 511–531 (2004)
48. T.Y. Yang, Finite Element Structural Analysis (Prentice-Hall, Englewoods, Cliffs, 1986)
49. Jing Li, A Geometrically Exact curved beam theory and its finite element formulation /imple-
mentation, master of science thesis in aerospace engineering, virginia polytechnique institute,
USA, 2000.
50. M. Petyt, Introduction to Finite Element Vibration Analysis, 2nd edn. (Cambridge University
Press, Cambridge, U.K., 1998)
51. M. Mukhopadhyay, Vibrations, dynamics and structural systems, Oxford & IBH, 2nd edition,
2000.
52. S.Y. Wang, S.T. Quek, N.K. Ang, Dynamic stability analysis of finite element modelling of
piezo-electric composite. Int. J. Solids Struct. 41(3), 745–764 (1987)
53. K. Chandrasekhara, K. Bhatia, Active buckling control of smart composite plates – a finite
element analysis. Smart Mater. Struct. 2, 31–39 (1993)
54. M. A. Crisfield, Non-linear finite element analysis of solids and structures, v. I and II, John
Wiley and Sons, 1991 and 1997.
55. P.K. Kythe, D. Wee, An Introduction to linear and non-linear finite element analysis
(Birkhauser, Boston, 2002)
56. J. T. Oden, Finite elements of non-linear continua, McGraw-Hill, 1972.
Chapter 9
Finite Element Analysis of Plane
Elasticity Problems

9.1 Introduction

The present chapter deals with the element representation of plane stress prob-
lems. The derivation of element stiffness matrices is demonstrated for two types
of elements—three-noded triangular element and four-noded rectangular element.
Some of the higher order elements are then discussed. Isoparametric element will,
however, be presented in the next chapter. A computer program for the solution of
PLANE stress problems using triangular element has been presented. An example
worked out with the help of the computer program has also been given.
The membrane part of the stress condition of some structures such as stiffened
plate, folded plate, shells etc. can be represented by the plane stress elements. The
difference between plane stress and plane strain problems has been discussed in
Chap. 7. The finite element analysis of these two problems is identical except for the
[D] matrix, which relates the stress and the corresponding strain.

9.2 Three-Noded Triangular Element

The simplest element available in the analysis of plane elasticity problems is the
3-noded triangular element. This element is most versatile in that it can fit into any
irregular boundary—the curved boundaries are to be approximated by a series of
straight lines. The element is also referred to as constant strain triangle, because
it will be seen in the derivation of element stiffness matrix that the strain remains
constant at any point within the element.
A triangular element of arbitrary shape with nodes at the corners denoted as 1, 2
and 3 is shown in Fig. 9.1. The co-ordinates of 1,2 and 3 are (x 1 , y1 ), (x 2 , y2 ) and
(x 3 , y3 ). The thickness of the element is t. The displacements parallel to the axis
systems at points 1, 2 and 3 are (u1 , v1 ), (u2 , v2 ) and (u3 , v3 ). Forces corresponding to
the displacements are (Px 1 , Py1 ), (Px 2 , Py2 ) and (Px 3 , Py3 ). Therefore the complete

© The Author(s) 2022 177

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_9
178 9 Finite Element Analysis of Plane Elasticity Problems

v3, Py3

(x3, y3)
3 u3, Px3

v2, Py2

v1, Py1
u2, Px2
2 (x2, y2)
1 u1, Px1
(x1, y1)

Fig. 9.1 Three-noded triangular plane stress element

displacements and force vectors for these elements are

{ X }eT = { u 1 v1 u 2 v2 u 3 v3 } (9.1)

{ P }eT = { P x1 P y1 P x2 P y2 P x3 P y3 } (9.2)

9.2.1 Displacement Function

Each node is free to move in x and y directions so that there are two degrees of
freedom per node. Therefore, for 3 nodes there are in all six degrees of freedom. The
displacement functions for u and v are represented by polynomials. Total number of
constant terms to be associated with the polynomial will be six, because the number
of constant terms equals the total degrees of freedom for the element. Therefore, for
this case, the displacement function is represented by two linear polynomials
( )
u = α1 + α2 x + α3 y
(9.3)
v = α4 + α5 x + α6 y

or
9.2 Three-Noded Triangular Element 179
⎧ ⎫
⎪
⎪ α1 ⎪⎪
⎪
⎪ ⎪
⎪
⎪
⎪ α2 ⎪⎪
⎪
⎪
( ) [ ] ⎨ ⎬
u 1x y000 α3
= (9.4)
v 0001x y ⎪ ⎪ α4 ⎪⎪
⎪
⎪ ⎪
⎪
⎪
⎪ α ⎪
⎪
⎪
⎩
5 ⎪
⎭
α6

or

{ f } = [C] {α} (9.5)

9.2.2 Displacement Function Expressed in Terms of Nodal

Displacements

Using Eq. (9.4) nodal displacements can be expressed in terms of respective nodal
coordinates. Thus,
⎧ ⎫ ⎡ ⎤ ⎧ ⎫
⎪
⎪ u1 ⎪⎪ 1 x1 y1 0 0 0 ⎪
⎪ α1 ⎪⎪
⎪
⎪ ⎪
⎪ ⎢0 0 0 1x y ⎥ ⎪ ⎪ ⎪
⎪
⎪
⎪ v ⎪
⎪ ⎢ ⎥ ⎪
⎪ α ⎪
⎪
⎪
⎨
1 ⎪
⎬ ⎢
1 1 ⎪
⎥ ⎨ ⎪
2
⎬
u2 ⎢ 1 x2 y2 0 0 0 ⎥ α3
= ⎢ ⎥ (9.6)
⎪
⎪ v2 ⎪⎪ ⎢ 0 0 0 1 x2 y2 ⎥ ⎪ ⎪ α4 ⎪
⎪
⎪
⎪u ⎪ ⎪
⎪ ⎢ ⎥ ⎪
⎪α ⎪ ⎪
⎪
⎪
⎪ 3 ⎪ ⎣ 1 x3 y3 0 0 0 ⎦ ⎪ ⎪ 5⎪
⎪
⎩ ⎪
⎭ ⎪
⎩ ⎪ ⎭
v3 0 0 0 1 x3 y3 α6

or, { X }e = [ A ] { α } (9.7)

The constant terms associated with the polynomial can now be determined from
Eq. (9.7). Thus

{ α } = [ A ]− 1 { X }e (9.8)

Substituting the value of {α} from Eq. (9.8) into Eq. (9.5), yields

{ f } = [C] [A]− 1 {X }e (9.9)

or { f } = [ [I ] N1 [I ] N2 [I ] N3 ] {X }e (9.10)

where [I] is a 2 × 2 identity matrix

a1 + b1 x + c1 y
N1 = ,
2Δ
180 9 Finite Element Analysis of Plane Elasticity Problems

a2 + b2 x + c2 y
N2 =
2Δ
a3 + b3 x + c3 y
and N3 = 2Δ

where

a1 = x2 y3 − y2 x3 , b1 = y2 − y3 , and c1 = x3 − x2
a2 = x3 y1 − x1 y3 , b2 = y3 − y1 , and c2 = x1 − x3
a3 = x1 y2 − x2 y1 , b3 = y1 − y2 , and c3 = x2 − x1
⎡ ⎤
1 x1 y1
2Δ = det ⎣ 1 x2 y2 ⎦
1 x3 y3
= 2 × area of the triangular elements 123

It may be mentioned that the absolute value of the area of the triangle is to be
considered. Equation (9.9) can be written as

{ f } = [N ] {X }e (9.11)

It will be shown in later chapter that [N] denotes the shape function matrix and
[N ] = [C] [A] −1 . . In the present case, the size being small, [A] has been inverted
algebraically. For elements where [A] is larger, the inversion has to be done by the
computer.
Whether from Eq. (9.4) or Eq. (9.11), it is seen that displacements along the
edge vary linearly and its magnitude is dependent on the coordinates at the two
nodes connecting that edge. Therefore, the compatibility of displacements of adjacent
triangular elements with a common boundary is ensured.

9.2.3 Strain–Nodal Parameter Relationship

Strain components for a plane stress problem are given by

⎧ ⎫ ⎧ ⎫
∂u
⎨ εx ⎬ ⎪
⎨ ∂x
⎪
⎬
∂v
{ε} = εy = ∂y (9.12)
⎩ ⎭ ⎪
⎩ ∂u ∂v ⎪
⎭
γx y ∂y
+ ∂x

Equation (9.12) can be written as

9.2 Three-Noded Triangular Element 181
⎡ ⎤
∂
∂x
0 ( )
⎢ ∂ ⎥ u
{ε} = ⎣ 0 ∂y ⎦ (9.13)
∂ ∂ v
∂y ∂x

Combining Eqs. (9.11) and (9.13) yields

⎡ ∂ N1 ∂ N2 ∂ N3
⎤
∂x
0 ∂x
0 ∂x
0
⎢ ∂ N1
0 ∂∂Ny2 0 ∂∂Ny3 ⎥
{ε} = ⎣ 0 ∂y ⎦ {X }e (9.14)
∂ N1 ∂ N1 ∂ N2 ∂ N2 ∂ N3 ∂ N3
∂y ∂x ∂y ∂x ∂y ∂x

Performing all the necessary differentiation, Eq. (9.14) reduces to

⎡ ⎤
b1 0 b2 0 b3 0
1 ⎣
{ε} = 0 c1 0 c2 0 c3 ⎦ (9.15)
2Δ
c1 b1 c2 b2 c3 b3

In compact form, Eq. (9.15) becomes

{ε} = [B] {X }e (9.16)

A careful look into [B] matrix will immediately reveal that all the elements of
[B] are constants, which indicates that the strain is constant at any point within the
element. Strains are the first derivatives of displacements. So it is obvious that when
displacement functions are linear polynomial, the strain will be constant.

9.2.4 Stress–Strain Relationship

It has been seen in Eq. (9.12) that the stress–strain relationship for the plane stress
problem is given by

{σ } = [D] {ε} (9.17)

Replacing {ε} from Eq. (9.16) into Eq. (9.17) yields

{σ } = [D] [B] {X }e (9.18)

182 9 Finite Element Analysis of Plane Elasticity Problems

9.2.5 Derivation of the Element Stiffness Matrix

The element stiffness matrix is derived here by applying Castigliano’s theorem. Total
strain energy of the plate element is given by
(
1 { }
U = σx εx σ y ε y τx y γx y dv (9.19)
2 v
⎧ ⎫
( ⎨ σ ⎬
1 { } x
or, U = εx ε y γx y σ y dv
2 v ⎩ ⎭
τx y
(
1
or, U = { ε }T { σ } dv (9.20)
2 v

Substituting {ε} and {σ } from Eqs. (9.16) and (9.18) into Eq. (9.20) yields
(
1
U = ( [B] {X }e )T [D] [B] {X }e dv
2 v
or, (9.21)
(
1
U = {X }eT [B]T [D] [B] {X }e dv
2 v

The integrand is constant. As such Eq. (9.21) becomes

1
U = {X }eT [B]T [D] [B] {X }e Δ . t (9.22)
2
where t is the thickness of the plate.
Now applying Castigliano’s theorem (Eq. 9.16),

∂U
= [B]T [D] [B] Δ · t · {X }e
∂{X }e

or, {P}e = [K ]e {X }e (9.23)

where [K ]e = [B]T [D] [B] Δ · t (9.24)

[K]e has been explicitly evaluated and the final values for a plate element are given
as Eq. (9.25).
9.2 Three-Noded Triangular Element 183
⎡ ⎤
D1 b12
⎢ + D12 c12 ⎥
⎢ ⎥
⎢ ⎥
⎢ D1 D2 b1 c1 D1 c12 ⎥
⎢ ⎥
⎢ + D12 b1 c1 + D12 b12 ⎥
⎢ ⎥
⎢ D1 b2 b1 D1 D2 b2 c1 D1 b22 ⎥
⎢ ⎥
t ⎢
⎢ + D12 c1 c2 + D12 b1 c2 + D12 c22
⎥
⎥
[k]e = ⎢
4Δ ⎢
⎥
⎥
(9.25)
D1 D2 b1 c2 D1 c1 c2 D1 D2 b2 c2 D1 c22
⎢ ⎥
⎢ + D12 b2 c1 + D12 b1 b2 + D12 b2 c2 + D12 b22 ⎥
⎢ ⎥
⎢ D1 b32 ⎥
⎢ D1 b1 b3 D1 D2 b3 c1 D1 b2 b3 D1 D2 b3 c2 ⎥
⎢ ⎥
⎢ + D12 c1 c3 + D12 b1 c3 + D12 c2 c3 + D12 b2 c3 + D12 c32 ⎥
⎢ ⎥
⎣ D1 D2 b1 c3 D1 c1 c3 D1 D2 b2 c3 D1 c2 c3 D1 D2 b3 c3 D1 c32 ⎦
+ D12 b3 c1 + D12 b1 b3 + D12 b3 c2 + D12 b2 b3 + D12 b3 c3 + D12 b32

For plane stress

( )
D1 = E/ 1 − ν 2 , D2 = ν, D12 = D1 (1 − D2 )/2.
For plane strain
−ν)
D1 = (1 +E(1ν) (1 − 2ν)
, D2 = 1 −ν ν , D12 = D1 (1 2− D2 ) .

9.2.6 Determination of Element Stresses

After all the unknown displacements are obtained, the stresses in the elements are to
be obtained as the final step. Equation (9.18) is rewritten as

{σ } = [D] [B] {X }e

For this element, Eq. (9.18) becomes

⎡ ⎤
D b D1 D2 c1 D1 b2 D1 D2 c2 D1 b3 D1 D2 c3
1 ⎣ 1 1
{σ } = D1 D2 b1 D1 c1 D1 D3 b2 D1 c2 D1 D2 b3 D1 c3 ⎦ {X }e
2Δ
D12 c1 D12 b1 D12 c2 D12 b2 D12 c3 D12 b3
(9.26)

The loading on the plate or more precisely the evaluation of the loads acting at
the nodes can be done for the plane stress problem by simply lumping as shown in
the example presented later in the chapter. The general treatment of the load acting
on the element will be done in the next chapter. The remaining steps required for
the complete solution of the problem are done by the computer. But those steps are
more or less similar to the approach presented by the direct stiffness method as will
be evident in the computer program presented later.
184 9 Finite Element Analysis of Plane Elasticity Problems

9.3 Criteria for the Choice of the Displacement Function

As already mentioned, the choice of the displacement function is the starting point
of the mathematical analysis, and its selection has a lot of bearing on the accuracy of
the results. On the basis of this assumed displacement function, the problem reduces
from infinite degrees of freedom to some finite number. If the choice is improper
then the result will not converge to the exact solution irrespective of the fineness of
the subdivision. In order that the simulated structure deforms as closely as possible
to the actual structure, the displacement function chosen should satisfy the following
three minimum requirements.
1. The chosen displacement function should ensure continuity within the element
and also compatibility between the adjacent elements.
The polynomial expressions representing the displacement function obviously
ensure the continuity of displacements within the element. There should not be
any kinking, gaps or overlaps at the edge common to adjacent elements (Fig. 9.2).
This condition is referred to as the ‘inter-elemental continuity’ condition. This
condition is satisfied only when displacements along the edge are dependent only
on the nodal displacements on that edge.
2. The chosen displacement function should include the rigid body modes of the
displacement of the element. This suggests that rigid body displacements at the
nodes must not induce any straining of the element.
In order to avoid complicated expressions, the terms pertaining to the rigid body
modes have not been sometimes considered in the displacement function. Such
formulation in curvilinear coordinates revealed that due to the violation of this
condition, the convergence of the result has only slowed down, but the results
have indeed converged to the correct value (9.1).
3. The chosen displacement function must represent the constant strain state of the
element.

Fig. 9.2 Discontinuity at the element boundary

9.5 Verification of the Convergence Criteria of the Displacement … 185

1 (constant – 1 term)
x y (constant – 2 term)
x xy y2
2
(constant – 3 term)
x3 x2y xy2 y3 (constant – 4 term)
x x3y x2y2 xy3 y4
4
(constant – 5 term)
x5 x4y x3y2 x2y3 xy4 y5 (constant – 1 term)
Fig. 9.3 Pascal triangle

This requirement suggests that as the subdivisions of a body are increased so that
elements get smaller so as to reach an infinitesimal size, the strains in each element
assume constant value. More serious consequences result from the violation
of this condition—the result converges to some incorrect value, which may be
significantly different from the correct one.

9.4 Polynomial Displacement Functions

One of the prime considerations to be given for the choice of the displacement
function is that it should not be biased by the local coordinate system. This condition
is referred to as geometric invariance, geometric isotropy or spatial isotropy. This
suggests that for two-dimensional problems, the x and y terms should be balanced
in the expression or there should not be more weightage of either x or y terms. The
geometric isotropy condition can be met by choosing terms from Pascal’s triangle
given in Fig. 9.3.
This immediately indicates the number of terms present in a complete polynomial
of a particular order.
The terms of the displacement function should always be chosen from lower order
polynomials and gradually moved towards the next higher order and so on. These
lower order terms help in satisfying the convergence criteria.

9.5 Verification of the Convergence Criteria

of the Displacement Function of 3-Noded Triangular
Element

The displacement function of the 3-noded triangular element given by Eq. (9.3) is
reproduced below.
( )
u = α1 + α2 x + α3 y
(9.3)
v = α4 + α5 x + α6 y
186 9 Finite Element Analysis of Plane Elasticity Problems

Let the three conditions required to satisfy the convergence criteria be studied
with respect to the u and v polynomial expression of Eq. (9.3).
1. Condition 1: Both u and v being polynomial series satisfy the continuity require-
ment within the element. It has been shown in Art 9.2 that the inter-elemental
continuity along any edge also exists.
2. Condition 2: The rigid body u and v displacements terms are α 1 and α 4 . The
rigid body rotation terms are α 3 and α 5 . All these constant terms are not zero.
Therefore, the displacement function contains terms depicting the rigid body
modes.
3. Condition 3: For the plane stress problem, three strain terms are εx , εy and γ xy .
The constant strain terms in the displacement function are α 2 , α 3 , α 5 and α 6 ,
which are non-zero. So condition 3 is also satisfied.

9.6 Number of Terms in a Polynomial

It has been mentioned that the number of constants to be associated with a polynomial
equals the total degrees of freedom of the element. Let the background of it be seen
with respect to the constant strain triangular element.
The direct strain in x-direction is constant and is given by

∂u
εx = = α2 (9.27)
∂x
Integrating Eq. (9.27) with respect to x yields

u = α2 x + f (y) (9.28)

Similarly, v = α6 y + g(x) (9.29)

∂u ∂v
Again γx y = + = c (9.30)
∂y ∂x

Combining Eqs. (9.28), (9.29) and (9.30), yields

dg(x) d f (y)
+ = c (9.31)
dx dy

Sum of the derivations of two independent functions in x and y can be equal to a

constant if both are separately equal to constants.

dg(x)
Let = α5 (9.32)
dx
9.7 Four-Noded Rectangular Element 187

Integrating Eq. (9.32) with respect to x, yields

g(x) = α5 x + α4 (9.33)

d f (y)
Similarly, = α3 (9.34)
dy

Integrating Eq. (9.34) with respect to y yields,

f (y) = α3 y + α1 (9.35)

Substituting the values of g(x) and f (y) from Eqs. (9.33) and (9.35) into Eqs.
(9.28) and (9.29), result in
)
u = α1 + α2 x + α3 y
(9.36)
v = α4 + α5 x + α6 y

9.7 Four-Noded Rectangular Element

For structures having regular shaped boundaries, rectangular elements can be effec-
tively employed for the analysis. The plate is divided into a suitable number of rect-
angular elements. It may be associated with other elements depending on the nature
of the structure. A skew plate of Fig. 9.4 is divided into triangular and rectangular
elements. In the following, the derivation of element stiffness matrix of the rectan-
gular element will be presented. The basic steps will be the same as that adopted for
the constant strain triangular element.
A typical rectangular element having sides a and b is shown in Fig. 9.5. The
thickness of the element is t. The nodes are numbered such that the coordinates of
1,2,3 and 4 are (0,0), (0,b), (a,0) respectively. The displacements at nodes 1, 2 etc.
parallel to the axis system are (u1 , v1 ), (u2 , v2 ) etc. and the corresponding forces are

Fig. 9.4 Mesh division of a

skew plate
188 9 Finite Element Analysis of Plane Elasticity Problems

Fig. 9.5 Rectangular y

element (with corner nodes)
2 4

x
1 a 3

(Px 1 , Py1 ), (Px 2 , Py2 ) etc. The displacement vector and the force vector are then
given by
{ } ⎫
{X }eT = u 1 v1 u 2 v2 u 3 v3 u 4 v4 ⎪
⎬
and (9.37)
{ }⎪
⎭
{P}eT = P x1 P y1 P x2 P y2 P x3 P y3 P x4 P y4

9.7.1 Displacement Function

Like triangular plane stress element, the state of displacement at any point (x, y) can be
represented by two components u and v. To be consistent with 8 nodal displacements
at four corners of the rectangle, the displacement equation to be chosen should contain
8 arbitrary constants. The functions expressed as a polynomial then are
)
u = α1 + α2 x + α3 y + α4 x y
(9.38)
v = α5 + α6 x + α7 y + α8 x y
⎧ ⎫
⎪
⎪ α1 ⎪
⎪
⎪
⎪ ⎪
⎪
⎪ α2 ⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ α ⎪
⎪
( ) [ ]⎪
⎨
3 ⎪
⎬
u 1 x y xy 0 0 0 0 α4
{f} = = (9.39)
v 0 0 0 0 1 x y xy ⎪⎪ α5 ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ α ⎪
⎪
⎪
⎪
6 ⎪
⎪
⎪
⎪ α ⎪
⎪
⎪
⎩
7 ⎪
⎭
α8
{
and f } = [C]{a} (9.40)
9.7 Four-Noded Rectangular Element 189

It may be verified that the displacement function satisfies all the requirements for
convergence criteria. The functions being linear satisfy inter-elemental continuity.

9.7.2 Displacement Function in Terms of Nodal

Displacements

Nodal displacements can be expressed in terms of its nodal coordinates with the help
of Eq. (9.39)
⎧ ⎫ ⎡ ⎤ ⎧ ⎫
⎪
⎪ u1 ⎪⎪ 100 0 0 0 0 0 ⎪
⎪ α1 ⎪
⎪
⎪
⎪ ⎪
⎪ ⎢000 ⎥ ⎪
⎪ ⎪
⎪
⎪
⎪ v ⎪
⎪ ⎢ 0 1 0 0 0 ⎥ ⎪
⎪ α2 ⎪
⎪
⎪ ⎪ ⎥ ⎪ ⎪
1
⎪ u2 ⎪ ⎢10b ⎪ ⎪
⎪
⎪ ⎪
⎪ ⎢ 0 0 0 0 0 ⎥ ⎪
⎪ α3 ⎪
⎪
⎪
⎨ ⎪
⎬ ⎢ ⎥ ⎪
⎨ ⎪
⎬
v2 ⎢000 0 1 0 b 0 ⎥ α4
= ⎢ ⎥ (9.41)
⎪ u3 ⎪⎪ ⎢1a0 0 0 0 0 0 ⎥ ⎪ α5 ⎪
⎪
⎪ ⎪ ⎢ ⎥ ⎪
⎪ ⎪
⎪
⎪
⎪ v3 ⎪⎪ ⎢000 ⎥ ⎪
⎪ α6 ⎪
⎪
⎪
⎪ ⎪
⎪ ⎢ 0 1 a 0 0 ⎥ ⎪
⎪ ⎪
⎪
⎪
⎪
⎪ u
⎪
⎪
⎪ ⎣1ab ab 0 0 0 0 ⎦ ⎪
⎪
⎪ α7
⎪
⎪
⎪
⎪
⎩
4 ⎪
⎭ ⎪
⎩ ⎪
⎭
v4 000 0 1 a b ab α8

or

{X }e = [A] {α} (9.42)

From Eq. (9.42), one can write

{α} = [A]−1 {X }e (9.43)

The algebraic equations given by Eq. (9.41) are very simple and therefore [A−1 ]
can be easily obtained. It is given below
⎡ ⎤
1 0 0 0 0 0 0 0
⎢ −1/a 0 0 0 1/a 0 0 0 ⎥
⎢ ⎥
⎢ −1/b 0 ⎥
⎢ 1/b 0 0 0 0 0 ⎥
⎢ ⎥
⎢ 1/ab 0 −1/ab 0 −1/ab 0 1/ab 0 ⎥
[A]−1 = ⎢ ⎥ (9.44)
⎢ 0 1 0 0 0 0 0 0 ⎥
⎢ ⎥
⎢ 0 −1/a 0 0 0 1/a 0 0 ⎥
⎢ ⎥
⎣ 0 −1/b 0 1/b 0 0 0 0 ⎦
0 −1/ab 0 −1/ab 0 −1/ab 0 1/ab
190 9 Finite Element Analysis of Plane Elasticity Problems

Replacing (α) from Eq. (9.43) into Eq. (9.40) yields

{ f } = [C] [A]−1 {X }e (9.45)

or { f } = [N ] {X }e (9.46)

9.7.3 Strain-Nodal Displacement Relationship

Strain components for a plane stress problem are given by

⎧ ⎫ ⎡ ⎤
∂ ( )
⎨ εx ⎬ 0
⎢ ∂x ∂ ⎥ u
{ε} = ε = ⎣ 0 ∂y ⎦ (9.47)
⎩ y ⎭ ∂ ∂ v
εx y ∂y ∂x

Performing the necessary differentiation on Eq. (9.38), Eq. (9.47) can be written
as
⎧ ⎫ ⎡ ⎤
⎨ εx ⎬ 010y0000
ε = ⎣ 0 0 0 0 0 0 1 x ⎦ {α} (9.48)
⎩ y ⎭
γx y 001x010y

or { ε } = [ G ] {α } (9.49)

Combining Eq. (9.48) within Eq. (9.43), yields

{ ε } = [ G ] [ A ]−1 { X }e

or { ε } = [ B ] { X }e (9.50)

The strain variation within the elements is linear as given by Eq. (9.49).

9.7.4 Stress–Strain Relationship

Stresses and strains are related as follows

{σ } = [ D] {ε} (9.51)

Replacing {ε} from Eq. (9.50) into Eq. (9.51), yields

9.8 A Note on the Rectangular Element 191

{ σ } = [D] [B] {X }e (9.52)

9.7.5 Derivation of the Element Stiffness Matrix

It has already been seen that the stiffness matrix of the element is given by
(
{k}e = [B]T [D] [B] dv (9.53)
v

So for an element of constant thickness, Eq. (9.53) becomes

(
{k}e = [B]T [D] [B] d x d y (9.54)
A

In order to evaluate the integral, [B]T [D] [B] are first to be calculated and then the
elements of the resulting matrix are to be evaluated.[K]e has been explicitly evaluated
and is given by Eq. (9.55) for an orthotropic plate element.

9.7.6 Evaluation of Element Stresses

After the unknown displacements are evaluated, the stresses in the element can be
determined by Eq. (9.56).

9.8 A Note on the Rectangular Element

Strains varying linearly within the element suggest that stresses also vary linearly
within the element. It may be remembered that stresses are constant within a triangular
element. If the same structure is to be solved by using first triangular elements and
then by rectangular elements, it has been found that in order to attain a particular
level of accuracy, the plate is to be divided into a much lesser number of meshes in
the case of rectangular elements.
192 9 Finite Element Analysis of Plane Elasticity Problems

9.9 A Note on Element Stresses

It has been shown that element stresses do not satisfy the equilibrium conditions
for the individual element. This suggests that the applied loads do not balance the
internal stresses at a nodal point. As a consequence, it has been found that the stresses
at a common node obtained from different elements do not attain a unique value.
What should then be the correct value of the stress at the node? In one approach,
the stresses are calculated at all the nodes of the element and the average value is
assigned at the centroid of the element. In another approach, the stresses at a particular
node are taken as the average values obtained from all the adjacent elements common
to that node. For a more detailed discussion about element stresses the reader may
look into the chapter on Adaptive Finite Elements analysis.

[K ]e
⎡ ⎤
4Dx p −1
p = a
⎢ + 4Dx y p b ⎥
⎢ ⎥
⎢ 3D1 4D y p ⎥
⎢ ⎥
⎢ + 3Dx y + 4Dx y p −1 ⎥
⎢ ⎥
⎢ 2Dx p −1 − 3D1 4Dx p −1 ⎥
⎢ ⎥
⎢ ⎥
⎢ −4Dx y p + 3Dx y +4Dx y p ⎥
⎢ ⎥
⎢ − 3D1 −4D y p −3D1 4D y p ⎥
⎢ ⎥
t ⎢
⎢ + 3Dx y +2Dx y p −1 − 3Dx y +4Dx y p −1 ⎥
⎥
=
12 ⎢ − 4Dx p −1 3D1 −2Dx p −1 3D1 4D1 p −1 ⎥ (9.55)
⎢ ⎥
⎢ +2Dx y p + 3Dx y −2Dx y p + 3Dx y ⎥
⎢ 4Dx y p ⎥
⎢ ⎥
⎢ 3D1 2D y p −1 3D1 −2D y p −3D1 4Dx p ⎥
⎢ ⎥
⎢ − 3Dx y 4Dx y p −1 + 3Dx y −2Dx y p −1 −3Dx y 4Dx y p −1 ⎥
⎢ ⎥
⎢
⎢ −2Dx p −1 −3D1 − 4Dx p −1 −3D1 2Dx p −1 3D1 4Dx p −1 ⎥
⎥
⎢ −2Dx y p + 3Dx y +2Dx y p −3Dx y − 4Dx y p + 3Dx y +4Dx y p ⎥
⎢ ⎥
⎣ −3D1 −2D y p −3D1 2D y p 3D1 − 4D y p 3D1 − 4D y p ⎦
−3Dx y −2Dx y p −1 +3Dx y − 4Dx y p −1 − 3Dx y +2Dx y p −1 + 3Dx y +4Dx y p −1
{σ }e
⎡ ⎤
−D1 (b − y) −D1 D2 (a − x) −D1 y D1 D2 (a − x) D1 (b − y) −D1 D2 x D1 y D1 D2 x
1 ⎢ ⎥
= ⎣ −D1 D2 (b − y) −D1 (a − x) −D1 D2 y D1 (a − x) D1 D2 (b − y) −D1 x D1 D2 y D1 x ⎦ {X }e
ab
−Dη (a − x) −Dη (b − y) +Dη (a − x) −Dη y −Dη x Dη (b − y) Dη x Dη y
(9.56)

9.10 Computer Program for the Plane Stress Analysis

Using Three–Noded Triangular Element

The computer program in both C-language and FORTRAN for the solution of plane
stress problem using constant strain triangular elements has been given in the attached
CD. The notations have been explained at the beginning of the computer programs
and different steps followed there have been explained.
To use this computer program, every element data is to be prepared, i.e., the
coordinates of each node of the element and global numbering of the nodes are input
to the computer program.
9.10 Computer Program for the Plane Stress Analysis … 193

The computer program for CST elements consists of a main program and four
subroutines. The entire input data is to be prepared on the basis of the main program.
It will then call the subroutines to perform the necessary functions. In the computer
program, the element stiffness matrix is generated in subroutine TELMS. The
assembly of element stiffness matrices to form an overall stiffness matrix is obtained
in subroutine ASSEM. The overall stiffness matrix is stored in half band form. The
simultaneous equations are solved by using subroutine GAUSS. The element stresses
are generated in subroutine TSTRM.
The preparation of the data is explained with the help of an example problem
shown in Fig. 9.6. Due to symmetry, only the bottom half of the plate is analysed.
The boundary conditions indicated in the computer program are corresponding to
this problem and they have not been presented in a general manner.
The element numbers have been indicated within small circles shown in Fig. 9.6b.
In all, there are 9 nodes and each node has two degrees of freedom so that NT =
18. The half bandwidth NBH = 10, NLC = 1. The preparation of data for a typical
element say 5 is explained here. The node numbers corresponding to element 5 are
2,3 and 5. Any one node can be taken as the origin. Suppose node 2 is taken as the
origin. The positive direction of the axis system has been shown in Fig. 9.6b The data
for element coordinates and node numbers corresponding to the READ statement of
the computer program will be.

0. 0.5 0. 0. 0. 0.25 1.0

235

The input data corresponding to ‘A’ and ‘B’ are irrevalent and so are ‘NX’ and
‘NY’ for a more general problem. The input data of the problem is given in Table
9.1 and the computer output has been presented in Table 9.2.
Exercise 9
9.1. Deduce the stiffness matrix of a rectangle element in terms of normalized
coordinates ξ = x/a and η = y/b.
9.2. For the parallelogram plane stress element, derive the stiffness matrix in terms
of ξ and η (Fig. 9.4)
9.3. For the parallelogram plane stress element, derive the stiffness matrix in terms
of x and y.
9.4. The stiffness matrix of the parallelogram element is once formed in terms
of ξ and η and then in terms of x and y. Do you think that the displacement
function in both cases will satisfy the convergence criteria?
9.5. Write the computer program for the analysis of rectangular plates using
rectangular plane stress elements.
9.6. What changes are to be incurred in the computer program in order to convert
it for analysis of plates using rectangular elements? Make the necessary
modifications.
194 9 Finite Element Analysis of Plane Elasticity Problems

Fig. 9.6 The plate example

9.10 Computer Program for the Plane Stress Analysis … 195

Table 9.1 Input data for the plane stress problem using cst element
1.0 0.3 1.0 0.5
3 2 18 10 1 6 8 2 2 2
0.0 0.5 0.5 0.0 0.0 0.25 1.0
1 2 5
0.0 0.5 0.0 0.0 0.25 0.25 1.0
1 5 4
0.0 0.5 0.0 0.0 0.0 0.25 1.0
4 5 7
0.0 0.0 -0.5 0.0 0.25 0.25 1.0
5 8 7
0.0 0.5 0.0 0.0 0.0 0.25 1.0
2 3 5
0.0 0.0 -0.5 0.0 0.25 0.25 1.0
3 6 5
0.0 0.5 0.5 0.0 0.0 0.25 1.0
5 6 9
0.0 0.5 0.0 0.0 0.25 0.25 1.0
5 9 8
5 1 -0.125
11 1 -0.25

9.7. Modify the computer program so that a plate can be used as either a triangular
element or as a rectangular element or both.
9.8. Derive the stiffness matrix for the triangular element and verify the results
from Eq. (9.25)
9.9. If a triangular element is subjected to distributed in plane forces X1 and Y1
per unit area in the direction of x and y axis, then by applying the principle of
virtual work or by any other method, determine the equivalent nodal forces.
9.10. Construct stress matrix {σ } for a triangular plane stress element for an
isotropic plate.
9.11. Compute the plane strain stiffness matrix for the triangular element. The
element has unit thickness. ν = 0.3.
196 9 Finite Element Analysis of Plane Elasticity Problems

Table 9.2 Computer output for the plane stress problem using CST element
PLANE STRESS PROBLEMS FOR TRIANGULAR ELEMENTS
D = 1.000000 PR = 0.300000 A = 1.000000 B = 0.500000
NN = 3 NFREE = 2 NT = 18NBH = 10 NLC = 1 NENP = 6
NELM = 8 NLDT = 2 NX = 2 NY = 2
Element-wise X coordinates, Y coordinates and thickness
0.0000 0.5000 0.5000 0.0000 0.0000 0.2500 1.0000
0.0000 0.5000 0.0000 0.0000 0.2500 0.2500 1.0000
0.0000 0.5000 0.0000 0.0000 0.0000 0.2500 1.0000
0.0000 0.0000 -0.5000 0.0000 0.2500 0.2500 1.0000
0.0000 0.5000 0.0000 0.0000 0.0000 0.2500 1.0000
0.0000 0.0000 -0.5000 0.0000 0.2500 0.2500 1.0000
0.0000 0.5000 0.5000 0.0000 0.0000 0.2500 1.0000
0.0000 0.5000 0.0000 0.0000 0.2500 0.2500 1.0000
Element-wise node numbers
1 1 2 5
2 1 5 4
3 4 5 7
4 5 8 7
5 2 3 5
6 3 6 5
7 5 6 9
8 5 9 8
The Loading Matrix
0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000 -1.25000e-001
0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000
-2.50000e-001 0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000
0.00000e+000 0.00000e+000 0.00000e+000
The Displacement Matrix
-1.91045e-001 2.39539e-016 -2.85745e-001 -2.75714e-017 -6.01211e-001
-1.05971e-015 -1.41929e-001 1.77456e-002 -2.06899e-001 2.95080e-002
-4.51145e-001 -4.59991e-002 -6.17142e-016 6.99983e-016 -1.59153e-015
6.48527e-016 -1.54133e-015 -5.00765e-016
The Stress Matrix
1 -1.69220e-001 6.72660e-002 1.21300e-001
2 -1.19390e-001 3.51654e-002 8.46100e-002
3 -1.66192e-001 -1.20840e-001 2.27401e-001
4 -3.89116e-002 -1.29705e-001 3.18307e-001
5 -6.54422e-001 -7.82946e-002 1.21300e-001
6 -5.97461e-001 -3.63235e-001 1.72789e-001
7 -4.76145e-001 4.11529e-002 6.35986e-001
8 -3.89116e-002 -1.29705e-001 3.18307e-001

Bibliography

1. K.H. Murray, Comments on the convergence of the finite element solutions. J. AIAA 8, 815–816
(1970)
2. O.C. Zienkiewicz, R.L. Taylor, The Finite Element Method, vols. I and II (McGraw-Hill), p. 989
3. S.S. Rao, The Finite Element Method in Engineering, 2nd edn (Pergamon Press, 1989)
4. R.D. Cook, D.S. Malkus, M.E. Plesha, Concepts and Applications of Finite Element Analysis,
3rd edn (Wiley & Sons, 1989)
Chapter 10
Isoparametric and Other Element
Representations and Numerical
Integrations

10.1 Introduction

When a structure subjected to plane stress is analysed by employing three-noded

triangular or rectangular elements, two disadvantages may occur. Firstly, the structure
may have irregular curved boundaries and as such more number of straight edge
elements will be needed to approximate the boundary. Secondly, a large number
of these simple elements are needed to solve the problem to the desired degree of
accuracy so as to require a considerable amount of computer time. Isoparametric
element formulation that is introduced here is free from these drawbacks.
Isoparametric elements are capable of representing the boundary curvature. These
elements are mapped into rectangular elements having a specified number of nodes
and then the necessary finite element formulation is made. Thus, isoparametric
elements enable the effective handling of structures with arbitrary shapes.

10.2 Shape Function or Interpolation Function

Let us have a relook into Eqs. (9.10) and (9.11) and rewrite them as
⎧ ⎫
⎪
⎪ u1 ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
( ) [ ⎪ ⎪
]⎨
v1 ⎪
⎬
u N1 0 N2 0 N3 0 u2
= (10.1)
v 0 N1 0 N2 0 N3 ⎪⎪ v2 ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ u3 ⎪
⎪
⎪
⎩ ⎪ ⎭
v3

or

{ f } = [N ] {X } (10.2)
© The Author(s) 2022 197
M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_10
198 10 Isoparametric and Other Element Representations and Numerical Integrations

A careful look into Eq. (10.1) will reveal that N 1 = 1 at node 1 and zero at
other nodes. Likewise N 2 = 1 at node 2 and zero at other nodes and so on. N 1 , N 2
etc. are called shape functions or interpolation functions. It is termed as interpolation
function because it is on the basis of this function that the field variable (displacement
function in the present case) can be determined by interpolation at any point inside
the element or at any point on the boundary. It is called the shape function because it
is dependent on the shape of the element. There is one shape function corresponding
to each degree of freedom of the element. The shape functions may be polynomials
of any order, but the convergence criteria of Sect. 9.3 must be satisfied.

10.3 Determination of Shape Functions

Comparing Eq. (9.9) with Eq. (9.11), it is revealed that

[N ] = [C] [A]−1 (10.3)

The shape function matrix [N] can be derived from the displacement function as
the starting point. But certain difficulties may be encountered in the process. It may
be observed that for some elements, the inversion of [A] is considerably tedious and
further in some cases [A] may become singular, e.g., for a triangular plate bending
element where two sides are at right angles. The shape functions used in isoparametric
elements remove these disadvantages, as they can be constructed directly, based on
the basic property of shape function i.e.
)
Ni = 1( at (ξi ,) ηi )
(10.4)
Ni = 0 at ξi , η j (i /= j )

It is to be borne in mind that the edges of the element as it actually occurs may be
irregular, but they are mapped into some regular shapes which facilitate the analysis.
In Fig. 10.1 an irregular shaped quadrilateral is mapped into a square. Similarly, in
Fig. 10.2 the elements having highly curved edges are mapped into a square. The
number of nodes in the isoparametric element is the same as the number of nodes
of the physical element and there is a node to node correspondence between the two
elements. This aspect will be discussed later in this chapter. In the meantime shape
functions for the square elements are discussed. The sides of the square elements are
based on normalized coordinates such that their faces have values ±1.

10.3.1 Linear 2-D Element

The shape function for the linear 2-D element of Fig. 10.1b is given by
10.3 Determination of Shape Functions 199

y (–1,1) η (1, 1)
2 1 2

1 ξ

4
3 4 3
(–1,–1) (1,–1)
x

Fig. 10.1 Transformation from physical to isoparametric coordinates for the linear element

x
η
3 1(–1,1) 2(0,1) 3(1,1)
2

1 4
8 4
ξ
8 (–1,0) (1,0)
5
6

7 7(–1,–1) 6(0,–1) 5(1,–1)

x

(a) Physical coordinates (b) Isoparametric coordinates

Fig. 10.2 Transformation from physical to isoparametric coordinates for quadratic element

⎫
N1 = 1
(1 − ξ ) (1 + η) ⎪
⎪
4
⎬
N2 = 1
4 (1 + ξ ) (1 + η)
(10.5)
N3 = 1
(1 + ξ ) (1 − η) ⎪
⎪
4 ⎭
N4 = 1
4 (1 − ξ ) (1 − η)

It is seen that at node 1, ξ = – 1 and η = 1 and when these values are substituted in
N 1 of Eq. (10.5), N 1 becomes equal to 1. Similarly, if the coordinates of other nodes
are substituted into N 1 of Eq. (10.5), N 1 becomes zero. Therefore, the basic require-
ment of shape function as given by Eq. (10.4) is satisfied. Other shape functions may
be similarly verified.
For this four-noded element, the variation of the shape function is linear. Therefore,
this is referred to as a linear isoparametric element. As the shape function varies
linearly on all sides, the continuity is satisfied.
The shape functions given by Eq. (10.5) can be written more compactly as given
below:
200 10 Isoparametric and Other Element Representations and Numerical Integrations

1
Nr = (1 + ξ0 ) (+η0 ) (10.6)
4
where ξ0 = ξ ξr and η0 = η ηr

10.3.2 Quadratic 2-D Element

Referring to Fig. 10.3b, the shape function for the corner nodes is given by

1
Nr = (1 + ξ0 ) (1 + η0 ) (ξ0 + η0 − 1) (10.7)
4
The shape function for mid-side nodes 2 and 6 (i.e. where ξr = 0) are

1( )
Nr = 1 − ξ 2 (1 + η0 ) (10.8)
2
and those for nodes 4 and 8 (i.e. ηr = 0) are

1 ( )
Nr = (1 + ξ0 ) 1 − η2 (10.9)
2
Shape functions for cubic and other higher order elements can be similarly written
(10.1).

y η

2 1
1

b –1 ξ

1
x
4 3 –1
a

(a) Physical coordinates (b) Isoparametric coordinates

Fig. 10.3 Rectangular element and its isoparametric mapping

10.4 Plane Stress Isoparametric Linear Element 201

10.4 Plane Stress Isoparametric Linear Element

A rectangular element with node numbers and actual coordinates has been indicated
in Fig. 10.3. The element stiffness matrix of this element considering it as a linear
isoparametric element will be derived. The procedure to be presented is valid for
any odd shaped quadrilateral element with nodes at four corners. The same analysis
can be extended without any difficulty to quadratic or other higher order plane stress
elements.
As indicated in the previous chapter, the displacement is specified in the two
orthogonal directions in the plane of the plate. The displacement field is given by
( )
u
{f} = (10.10)
v

where u and v are the displacements parallel to x and y-directions. The nodal
parameters are
( )
.. .. ..
{X }eT = u 1 v1 .u 2 v2 .u 3 v3 .u 4 v4 (10.11)

10.4.1 Displacement Function in Terms of Nodal Parameters

The shape function defines the geometry as well as the displacement field for the
isoparametric element. Thus, at any point
)
x = N1 x 1 + N2 x 2 + N3 x 3 + N4 x 4
(10.12)
y = N1 y1 + N2 y2 + N3 y3 + N4 y4

and
)
u = N1 u 1 + N2 u 2 + N3 u 3 + N4 u 4
(10.13)
v = N1 v1 + N2 v2 + N3 v3 + N4 v4

For linear element, N 1 to N 4 are given by Eq. (10.5) in terms of ξ and η.

10.4.2 Strain-Nodal Parameter Relationship

The strain matrix associated with the plane stress problem is given by
202 10 Isoparametric and Other Element Representations and Numerical Integrations

⎧ ⎫ ⎧ ∂u ⎫ ⎡ ∂N ⎤
⎨ εx ⎬ ⎪ ⎨ ∂x ⎪
⎬ 0 ( )
r

∂v
Σ
4 ∂x
⎢ 0 ∂ Nr ⎥ u r
{ε } = ε y = ∂y = ⎣ ∂y ⎦ (10.14)
⎩ ⎭ ⎪⎩ ∂u + ∂v ⎪
⎭ ∂ Nr ∂ Nr vr
γx y ∂y ∂x
r =1
∂y ∂x

Thus, according to the previous notation

⎡ ∂N ⎤
Σ
4 Σ
4 ∂x
r
0
⎢ 0 ∂ Nr ⎥
[B] = [B]r = ⎣ ∂y ⎦ (10.15)
r =1 r =1 ∂ N r ∂ Nr
∂y ∂x

The shape function derivatives may be calculated using the chain rule
)
∂ Nr
∂x
= ∂ Nr ∂ξ
.
∂ξ ∂ x
+ ∂ Nr
∂η
. ∂∂ηx
(10.16)
∂ Nr
∂y
= ∂ Nr ∂ξ
.
∂ξ ∂ y
+ ∂ Nr
∂η
. ∂η
∂y

Equation (10.16) can be written as

( ) [ ]( )
∂ Nr ∂ξ ∂η ∂ Nr
∂x ∂x ∂x ∂ξ
∂ Nr = ∂ξ ∂η ∂ Nr
∂y ∂y ∂y ∂η

or
( ) ( )
∂ Nr ∂ Nr
∂x −1 ∂ξ
∂ Nr = [J ] ∂ Nr (10.17)
∂y ∂η

where
( )
∂x ∂y
∂ξ ∂ξ
[J ] = ∂x ∂y (10.18)
∂η ∂η

[J], the matrix of first derivatives, is referred to as the Jacobian matrix

10.4.3 Evaluation of [B] Matrix

The nodes indicated as 1, 2, 3 and 4 in Fig. 10.3 have coordinates (0, b), (a, b) (a, 0)
and (0, 0) respectively.
Substituting the values of N 1 to N 4 from Eq. (10.5) into Eq. (10.17) and then,
differentiating x and y with respect to both ξ and η respectively,

∂x ∂ N1 ∂ N2 ∂ N3 ∂ N4
= x1 + x2 + x3 + x4
∂ξ ∂ξ ∂ξ ∂ξ ∂ξ
10.4 Plane Stress Isoparametric Linear Element 203

a a a
=0+ (1 + η) + (1 − η) + 0 =
4 4 2
Similarly,

∂y b b
= − (1 + η) + (1 + η) + 0 + 0 = 0
∂ξ 4 4
∂x a a
= 0 + (1 + ξ ) − (1 + ξ ) + 0 = 0
∂η 4 4

and
∂y b b b
= (1 − ξ ) + (1 + ξ ) + 0 + 0 =
∂η 4 4 2

∂x ∂y
Therefore, substituting the values of ∂ξ
, ∂ξ etc. into Eq. (10.18), yields.
[ ]
a/2 0
[J ] = (10.19)
0 b/2

Therefore,
[ ] [ ]
∂ξ ∂η
−1 ∂x ∂x 2/a 0
[J ] = ∂ξ ∂η = (10.20)
∂y ∂y 0 2/b

Thus,

∂ξ 2 ∂η 2 ∂η ∂ξ
= , = and = =0
∂x a ∂y b ∂x ∂y

Now, Eq. (10.16) applied to node 1 yields

∂ N1 ∂ N1 ∂ξ ∂ N1 ∂η (1 + η)
= . + . = − (10.20a)
∂x ∂ξ ∂ x ∂η ∂ x 2a

and
∂ N1 ∂ N1 ∂ξ ∂ N1 ∂η (1 − ξ )
= . + . =− (10.21b)
∂y ∂ξ ∂ y ∂η ∂ y 2b

Similarly, applying Eq. (10.16) to other nodes, yields

∂ N2 1 + η ∂ N2 1+ξ
= , =
∂x 2a ∂y 2b
204 10 Isoparametric and Other Element Representations and Numerical Integrations

∂ N3 1 − η ∂ N3 (1 + ξ )
= , =−
∂x 2a ∂y 2b
∂ N4 1 − η ∂ N4 (1 − ξ )
=− , =− (10.22)
∂x 2a ∂y 2b

Substituting the different values of the derivatives obtained in terms of ξ and η as

given by Eqs. (10.21a, 10.21b) and (10.22) into Eq. (10.15) yield
⎡ ⎤
− (1+η)
2a
0 (1+η)
2a
0 (1−η)
2a
0 − (1−η)
2a
0
[B] = ⎣ 0 1−ξ
2b
0 (1+ξ )
2b
0 − (1+ξ
2b
)
0 − (1−ξ 2b
)⎦
(10.23)
1−ξ
2b
− (1+η)
2a
1+ξ
2b
1+η
2a
− (1+ξ
2b
) (1+η)
2a
− (1−ξ
2b
)
− (1−η)
2a

10.4.4 Element Stiffness Matrix

Following steps similar to those presented in the previous chapter, the element
stiffness matrix can be shown as
( (
[k]e = t [B]T [D] [B] d x d y (10.24)

As already seen [B] is expressed in terms of ξ and η. The differential area dx dy

is to be replaced by

d x d y = |J | dξ dη (10.25)

| J | is the determinant of Jacobian matrix [J ]. From Eqs. (10.24) and (10.25),

[k]e can be written as
( (
[k]e = t [B]T [D] [B] |J | dξ dη (10.26)

Absolute value of | J | is to be considered.

It is realized that explicit evaluation of [k]e is tedious even for this simplest case
of the linear element. The practice that exists for determining [k]e is through the use
of numerical integration.

10.4.5 Convergence of Isoparametric Elements

In order that convergence criteria may be satisfied, the conditions of rigid body
displacements, constant strain criteria and the inter-elemental compatibility condition
10.4 Plane Stress Isoparametric Linear Element 205

must be fulfilled. In the case of two nodes along the edge, the shape function is linear,
for 3-noded edge it is quadratic and for 4-noded edge it is cubic. So the displacements
are uniquely defined which ensures the continuity along the edge or rule out any
discontinuity between adjacent elements common to the edge. The validity of other
two criteria is shown below with respect to the plane stress element.
A linear plane stress field that satisfies both the conditions required for rigid body
displacements and constant strain is given by
)
u = α1 + α2 x + α3 y
(10.27)
v = α4 + α5 x + α6 y

At node r, Eq. (10.27) assumes

)
u r = α1 + α2 xr + α3 y4
(10.28)
vr = α4 + α5 xr + α6 yr

In isoparametric formulation

Σ
n Σ
n
u= Nr u r and v = Nr vr (10.29)
r =1 r =1

where n is the number of nodes.

Substituting from Eq. (10.28) into Eq. (10.29) yields
⎫
Σ
n Σ
n Σ
n
u = α1 Nr + α2 Nr xr + α3 Nr yr ⎪
⎪
⎬
r =1 r =1 r =1
Σ n Σ n Σ n (10.30)
v = α4 Nr + α5 Nr xr + α6 ⎪
Nr yr ⎪
⎭
r =1 r =1 r =1

Σ
n Σ
n
Further noting that x = Nr xr and y = Nr yr Eq. (10.30) becomes
r =1 r =1

⎫
Σ
n
u = α1 Nr + α2 x + α3 y ⎪
⎪
⎬
r =1
Σ n (10.31)
v = α4 Nr + α5 x + α6 y ⎪
⎪
⎭
r =1

Displacements of Eq. (10.31) and Eq. (10.27) are same provided

Σ
n
Nr = 1 (10.32)
r =1

Although it may not look apparent, but Eq. (10.32) is valid for all types of
isoparametric elements.
206 10 Isoparametric and Other Element Representations and Numerical Integrations

10.4.6 Concept of Isoparametric Element

The element mentioned above has been termed as isoparametric element because the
shape function parameter defining the displacements is the same as that defining the
geometry.
However, it is not necessary that the geometry and displacements may be defined
by the same order model. The element in which geometry is represented by a lower
order model than displacements is called subparametric (Fig. 10.4a). The element
in which the converse is true is called superparametric element (Fig. 10.4b). For
elements with straight edges, it is sufficient if the geometry is defined by 4-corner
nodes whereas a higher order displacement field may be needed for the analysis. The
choice of the type of element isoparametric, subparametric or superparamatric – will
depend on the type of structure and the desired analysis. However, it is more important
for the analysis of solids (10.1). For isoparametric, subparametric and some forms
of superparametric elements, convergence criteria are satisfied.
It may be noted that the shape function which defines geometry is only reflected
in the Jacobian matrix.

10.5 Numerical Integration

Numerical integration is the method of evaluating the value of a definite integral

based on a certain numerical scheme. It is referred to as mechanical quadrature
when applied to the integration of a single variable, and mechanical cubature when
applied to the integration of two variables. However, in finite element literature, this
distinction between quadrature and cubature is not made and all numerical integration
schemes are referred to as quadrature.

(a) Subparametric (b) Superparametric

o – points defining geometry

– points defining displacements

Fig. 10.4 Superparametric and subparametric elements

10.5 Numerical Integration 207

Quadrature formulae may be divided into two main classes:

1. Newton-Cotes formula in which the absicca is usually divided into equal
intervals.
2. Gaussian quadrature in which no restriction on equal spacing of absicca is made,
although there may be other specific restrictions on the absicca and weight.
In Newton–Cotes formula, (n + 1) equally spaced sampling points are used to
integrate exactly a polynomial of maximum order n. On the other hand, in Gauss
quadrature, for evaluation of a polynomial of maximum order (2n – 1), n unequally
spaced sampling points are needed. For performing the necessary integration to
evaluate the element stiffness matrix in the finite element analysis, Gauss quadrature
is employed, as it is usually most effective.

10.5.1 Gaussian Quadrature Formula

Gauss sought answer to the following question:

(b
If a f (x) d x is to be computed from a given number of values of f (x), what selected values
of abscissa in general give the most accurate result?

He observed that points on the abscissa need not be equally spaced, but they
should be symmetrical about the midpoint of the interval of integration. The detailed
derivation has been given in the books of numerical analysis and as such it will not
be discussed here. Only the procedure of evaluation will be indicated. For a function
of one variable, the Gaussian quadrature rule is

(1 Σ
n
I = φ (ξ ) dξ = Hi φ (ξi ) (10.33)
−1 i=1

where n is the number of sampling points, ξi is the position corresponding to ith

point, φ (ξi ) and Hi are the values of the function and the weighting coefficients at
that point (Fig. 10.5). The sampling points and weighting coefficients up to n = 10
is given in Table 10.1.

10.5.2 Gaussian Integration of Two Variables

When a function having two variables are to be numerically integrated by Gaussian

quadrature formula, it is done thus
208 10 Isoparametric and Other Element Representations and Numerical Integrations

φ(ξi)
ξ
ξi

Fig. 10.5 Function φ (ξ)

(1 (1
I = φ (ξ, η) dξ dη
−1 −1 (10.34)
Σm Σ n ( )
= Hi H j φ ξi , η j
i=1 j=1

In Eq. (10.34) the number of integrating points in both directions have been
assumed to be different.
The above scheme can be extended to 3-dimensional integration.

Example 10.1 Evaluate the integral

(1 (1 2 2
−1 −1 x y d x d y using Gauss quadrature formula.
Let
( 1( 1
I = x 2 y2 d x d y
−1 −1

Both x and y in the above integral have quadratic terms. If n = 1, then polynomial
of maximum order (2 × 1 – 1) or 1 can be evaluated exactly. So n = 1 is insufficient
in the present case. If n = 2, then the polynomial of maximum order (2 × 2 – 1) or 3
can be evaluated exactly. So for performing the above integration, a 2-point Gaussian
quadrature for both x and y is employed.

Σ
m Σ
n
I = Hi H j φ (xi , yi )
i=1 j=1

Here m = n = 2
Therefore,
10.5 Numerical Integration 209

Table 10.1 Abscissae and weight coefficients of the gaussian quadrature formula
(1 Σ
n ( )
−1 f (x) d x = Hi f a j
j=1

±a H
n=1
0.00000 2.00000
n=2
0.57735 02691 89626 1.00000 00000 00000
n=3
0.77459 66692 41483 0.55655 55555 55556
0.00000 00000 00000 0.88888 88888 88889
n=4
0.86113 63115 94053 0.34785 48451 37454
0.33998 10435 84856 0.65214 51214 62546
n=5
0.90617 98459 38664 0.23692 68850 56189
0.53846 93101 05683 0.47862 86704 99366
0.00000 00000 00000 0.56888 88888 88889
n=6
0.93246 95142 03152 0.17132 44923 79170
0.66120 93864 66265 0.36076 15730 48139
0.23861 91860 83197 0.46791 39345 72691
n=7
0.94910 79123 42759 0.12948 49661 68870
0.74143 11855 99394 0.27970 53914 89277
0.40584 51513 77397 0.38183 00505 05119
0.00000 00000 00000 0.41795 91836 73469
n=8
0.96028 98564 97536 0.10122 85362 90376
0.79666 64774 13627 0.22238 10344 53374
0.52553 24099 16329 0.31370 66458 77887
0.18343 46424 95650 0.36268 37833 78362
n=9
0.96816 02395 07826 0.08127 43883 61574
0.83603 11073 26636 0.18064 81606 94857
0.61337 14327 00590 0.26061 06964 02935
0.32425 34234 03809 0.31234 70770 40003
0.00000 00000 00000 0.33023 93550 01260
n = 10
0.97390 65285 17172 0.06667 13443 08688
0.86506 33666 88985 0.14935 13491 50581
0.67940 95682 99024 0.21908 63625 15982
0.43339 53941 29247 0.26926 67193 09996
0.14887 43389 81631 0.29552 48247 14753
210 10 Isoparametric and Other Element Representations and Numerical Integrations

I = H1 H1 φ (x1 , y1 ) + H1 H2 φ (x1 , y2 ) + H2 H1 φ (x2 , y1 )

+H2 H2 φ (x2 , y2 )
( )2 ( )2 ( )2 ( )2
= 1 × 1 × − √13 × − √13 + 1 × 1 × − √13 × √13
( )2 ( )2 ( )2 ( )2
+1 × 1 × √13 × − √13 + 1 × 1 × √13 × √13
= 4
9

10.6 Lagrangian Interpolation

It is now realized that it is always preferable to express directly the assumed displace-
ment function in terms of its nodal degrees of freedom. Lagrangian interpolation
provides an easy means of generating shape functions.
Consider a line in Fig. 10.6c. It is divided into equally spaced points 1, 2, 3, …
k, …, n. The coordinates of these points are ξ1 , ξ2 , ξ3 , …, ξk , …, ξn . The Lagrange
polynomials satisfy the shape function properties given by Eq. (10.4). For different
nodes, they are

(ξ − ξ2 ) (ξ − ξ3 ) · · · · · · (ξ − ξn )
N1 =
(ξ1 − ξ2 ) (ξ1 − ξ3 ) · · · · · · (ξ1 − ξn )
(ξ − ξ1 ) (ξ − ξ2 ) · · · · · · (ξ − ξk−1 ) (ξ − ξk+1 ) · · · · · · (ξ − ξn )
Nk = (10.35)
(ξk − ξ1 ) (ξk − ξ2 ) · · · · · · (ξk − ξk−1 ) (ξk − ξk+1 ) · · · · · · (ξk − ξn )

In two dimensions, the shape function is given by

Ni j = Ni N j (10.36)

(i,
i j)

(a) Quadratic (b) Quartic

1 2 3 K–1 K K+1

(c) A line

Fig. 10.6 Intervals for Lagrangian interpolation

10.7 Natural Coordinates and Higher Order Triangular Elements 211

where N i and N j are one-dimensional values along their respective axes as given by
Eq. (10.35).
These shape functions are indeed very handy. But it suffers from two drawbacks.
The presence of a number of internal nodes makes the subsequent analysis compli-
cated. Secondly, polynomial terms contain some higher order terms, which remain
devoid of lower order terms that affect the convergence of the problem.

10.7 Natural Coordinates and Higher Order Triangular

Elements

A natural coordinate system is nothing but a local coordinate system which is

dimensionless and whose maximum value can be unity.

10.7.1 One-Dimensional Element

For the line element of Fig. 10.7, the natural coordinates are L 1 and L 2 such that L 1
= l 1 /l and L 2 = l 2 /l. For this line, the following relationship holds good
)
L1 + L2 = 1
(10.37)
x1 L 1 + x2 L 2 = x

It may be noticed that at point 1, L 1 = 1 and L 2 = 0 and at point 2, L 1 = 0 and L 2

= 1. From Eq. (10.37), the following equations are obtained
( ) [ ]( )
L1 1 x2 −1 1
= (10.38)
L2 l −x1 1 x

Differentiation of such polynomials with respect to x is given by the formula

Σ ∂ Li ∂
2 ( )
d 1 ∂ ∂
= . = − (10.39)
dx i=1
∂ x ∂ Li l ∂ L2 ∂ L1

Integration of such polynomials with respect to x is given by

Fig. 10.7 Line element x x1(1,0) (L1, L2) x2(0,1)

L2 L1

L
212 10 Isoparametric and Other Element Representations and Numerical Integrations
(
a!b!
L a1 L b2 d x = l (10.40)
l (a + b + 1)!

10.7.2 Higher Order Triangular Elements

The natural coordinate system for triangular elements is known as area coordinates.
If the total area of a triangle be A and a point P inside the triangle is such that the
total triangle is divided into three areas A1 , A2 and A3 as shown in Fig. 10.8, then the
area coordinates are given by

L 1 = A1 /A, L 2 = A2 / A, L 3 = A3 /A (10.41)

In Cartesian coordinate system, the following relationship exists (Fig. 10.8),

⎧ ⎫ ⎡ ⎤⎧ ⎫
⎨x ⎬ x1 x2 x3 ⎨ L 1 ⎬
y = ⎣ y1 y2 y3 ⎦ L 2 (10.42)
⎩ ⎭ ⎩ ⎭
1 1 1 1 L3

The last of the Eq. (10.42) indicates that two of the coordinates L1 , L2 and L3 are
independent, Inverting Eq. (10.42), yields
⎧ ⎫ ⎡ ⎤⎧ ⎫
⎨ L1 ⎬ 1 ⎣ 1 1 1⎦ ⎨ ⎬
b c a x
L2 = b2 c2 a2 y (10.43)
⎩ ⎭ 2A ⎩ ⎭
L3 b3 c3 a3 1

where b1 = y2 – y3 , c1 = x 3 – x 2 and a1 = x 2 y3 – x 3 y2 , a2 , b2 , c2 etc. can be defined

by chain rule.

Fig. 10.8 Area coordinates

10.7 Natural Coordinates and Higher Order Triangular Elements 213

Fig. 10.9 Quadratic triangle 1

6 4

3 5 2

The values of L 1 , L 2 and L 3 are identical with those of N 1 , N 2 and N 3 of Eq. (10.10)
for 3-noded triangle. Therefore, for 3-noded triangle, the area coordinates are the
shape functions.
Same concept can be extended to quadratic and other higher order triangles. The
shape functions for a quadratic triangle of Fig. 10.9 are given as follows:
For corner nodes

N1 = L 1 (2L 1 − 1) and so on (10.44)

For midside nodes

N4 = 4L 1 L 2 (10.45)

The differentiation rule to be applied is

∂ Σ ∂ Li 18 Σ bi ∂ 8
= . = .
∂x i=1
∂ x ∂ Li i=1
2 A ∂ Li

∂ Σ ∂ Li 18 Σ ci ∂ 8
= . = . (10.46)
∂y i=1
∂ y ∂ Li i=1
2 A ∂ Li

The integration rule to be applied is

(
r !s!t!
L r1 L s2 L t3 d x d y = A (10.47)
A (r + s + t + 2)!
214 10 Isoparametric and Other Element Representations and Numerical Integrations

10.8 The Quadratic Triangle for the Plane Stress Problem

The quadratic triangle of Fig. 10.9 is considered. It has straight edges and mid-side
nodes. The displacement vector is given by

{X }eT = {u 1 u 2 u 3 u 4 u 5 u 6 v1 v2 v3 v4 v5 v6 } (10.48)

The displacement functions are given by

Σ
6 Σ
6
u= Ni u i , v = Ni vi (10.49)
i=1 i=1

where the shape functions for the quadratic triangle (based on Eqs. (10.44) and
(10.45)) are
)
N1 = L 1 (2L 1 − 1), N2 = L 2 (2L 2 − 1), N3 = L 3 (2L 3 − 1)
(10.50)
N4 = 4L 1 L 2 , N5 = 4L 2 L 3 , N6 = 4L 3 L 1

Next the strain-nodal displacement relationships are to be obtained.

∂u Σ ∂ Ni
6
εx = = ui (10.51)
∂x i=1
∂x

Again,

∂ Ni Σ ∂ Ni ∂ L j
3
= (10.52)
∂x j=1
∂L j ∂x

So,
( )
∂ N1 ∂ N1 ∂ L 1 ∂ N1 ∂ L 2 ∂ N1 ∂ L 3
= + + (10.53)
∂x ∂ L1 ∂ x ∂ L2 ∂ x ∂ L3 ∂ x

Using Eqs. (10.43) and (10.50), Eq. (10.53) becomes

∂ N1 b1
= (4L 1 − 1) (10.54)
∂x 2A
Similarly,

∂ N2 ∂ N3
, and so on can be obtained
∂x ∂x
10.8 The Quadratic Triangle for the Plane Stress Problem 215

Equation (10.51) can be written as

1
εx = [(4L 1 − 1) b1 (4L 2 − 1) b2 (4L 3 − 1) b3
2A
⎧ ⎫
⎪
⎪ u1 ⎪⎪
⎪
⎪ ⎪
⎪
⎪
⎪ u ⎪
⎪
⎨ ⎪
⎪ 2
⎬
u3
4 (L 2 b1 + L 1 b2 ) 4 (L 3 b2 + L 2 b3 ) 4 (L 1 b3 + L 3 b1 )] (10.55)
⎪
⎪ u ⎪
⎪ 4⎪
⎪ ⎪
⎪
⎪
⎪ u5 ⎪⎪
⎪
⎩ ⎪ ⎭
u6

Similarly, εy and γxy can be expressed in terms of nodal displacements. Combining

all three, the strain matrix can now be written as
⎧ ⎫
⎨ εx ⎬
{ε} = ε y = [B] {X }e (10.56)
⎩ ⎭
γx y

Equation (10.55) reveals that the strain is linearly varying within the element.
The formulation can be simplified on the following basis. The strain matrix {ε}
can interpolate from the matrix {εc } indicating the strains at the corner nodes 1, 2
and 3.
P }T
{ε} = [ Q ] {εc } = [ Q ] εx1 εx2 εx3 ε y1 ε y2 ε y3 γx y1 γx y2 γx y3 (10.57)

where εx1 is εx for node 1 and so on.

⎡ ⎤
L1 L2 L3 0 0 0 0 0 0
[Q] = ⎣ 0 0 0 L 1 L 2 L 3 0 0 0 ⎦ (10.58)
0 0 0 0 0 0 L1 L2 L3

The strain–displacement relationship is now to be written for {εc }.

{εc } = [H ] {X }e (10.59)

where [H] is a matrix of order 9 × 12.

H 11 = 3b1 /2A, H 12 = 3b2 /2A and so on. Therefore, [H] is independent of x and y.
Combining Eqs. (10.57) and (10.59), we get

{ε} = [Q] [H ] {X }e (10.60)

Comparing Eq. (10.56) with Eq. (10.60), yields

216 10 Isoparametric and Other Element Representations and Numerical Integrations

[B] = [Q] [H ] (10.61)

The element stiffness matrix is given by

( (
[k]e = [B]T [D] [B] d V = [H ]T [Q]T [D] [Q] d V [H ] (10.62)
V V

or,
(
[k]e = t [H ]T [Q]T [D] [Q] d A [H ] (10.63)
A

For efficient evaluation of [k]e in Eq. (10.63), the integration rule given by Eq. (10.47)
may be used. Numerical integration of area coordinates is given below.

10.9 Numerical Integration of Area Coordinates

If f (L 1 , L 2 , L 3 ) is a function in terms of area coordinates L 1 , L 2 and L 3 , then its

integral is of the following form.
( 1 ( 1−L 1
I = f (L 1 , L 2 , L 3 ) d L 2 d L 1 (10.64)
0 1

In order to numerically evaluate I, it is written as

Σ
n
I = Wi f i (10.65)
i=1

where n is the number of sampling points fi is the value of f (L 1 , L 2 , L 3 ) at a

specified point in the triangle, W i is the weight appropriate to that point. A series of
necessary sampling points and weights are given in Table 10.2 [10.8].

10.10 Triangular Isoparametric Elements for the Analysis

of Plane Stress Problems

We take a quadratic triangle of arbitrary shape, shown in Fig. 10.10.

Like other isoparametric elements, we assume the following
Σ Σ Σ Σ
u= Ni u i , v = Ni vi , x = Ni x i , y = Ni yi (10.66)
10.10 Triangular Isoparametric Elements for the Analysis of Plane Stress Problems 217

Table 10.2 Numerical integration formulae for triangles

Order Figure Point Triangular coordinates Weights
Linear a 1/3, 1/3, 1/3 1
a

Quadratic a 1/2, 1/2, 0 1/3

a b
b 0, 1/2, 1/2 1/3
c 1/2, 0, 1/2 1/3
c

Cubic a 1/3, 1/3, 1/3 – 27/48

b b 0.6, 0.2, 0.2 25/48
c a
d
c 0.2 0.6, 0.2
d 0.2, 0.2, 0.6
Quintic a 1/3, 1/3, 1/3 0.22500, 00,000
b f d
a
b α 1 , β1 , β1 0.13239, 41,527
g
c
e c β1 , α 1 , β1 0.12593, 91,805
d β1 , β1 , α 1
e α 2 , β2 , β2
f β2 , α 2 , β2
g β2 , β2 , α 2

Fig. 10.10 Quadratic 3

triangle of arbitrary shapes
y,v 5
6

1
2
4

x,u

For quadratic triangle, the shape functions Ni are the same as those given in
Eq. (12.50). It has been shown in Art. 10.4 that one of the requirements for the compu-
tation of the stiffness matrix is the evaluation of the Jacobian matrix. It consists of
terms such as
∂x ∂y ∂y
∂ξ
, ∂ξ , ∂∂ηx and ∂η
(10.67)
∂x
∂ξ
= ∂∂ξN1 x1 + ∂∂ξN2 x2 + ∂ N3
∂ξ
x3 + .....................

Therefore, ∂∂ξNi is to be evaluated. In order to do that it may be noted that area coor-
dinates are not independent. There are two independent quantities, as the constraint
relation L 1 + L 2 + L 3 = 1 has to be satisfied.
The followings are assumed
218 10 Isoparametric and Other Element Representations and Numerical Integrations
⎫
L1 = ξ ⎬
L2 = η (10.68)
⎭
L3 = 1 − ξ − η

Now,

∂ Ni ∂ Ni ∂ L 1 ∂ Ni ∂ L 2 ∂ Ni ∂ L 3 ∂ Ni ∂ Ni
= . + . + . = − (10.69)
∂ξ ∂ L 1 ∂ξ ∂ L 2 ∂ξ ∂ L 3 ∂ξ ∂ L1 ∂ L3

Similarly,

∂ Ni ∂ Ni ∂ Ni
= − (10.70)
∂η ∂ L2 ∂ L3

Rest of the procedure is similar to that given in Art. 10.4. The stiffness matrix of
the triangular element is
(
[k]e = [B]T [D] [B] td A (10.71)
A

where t is the element thickness.

The matrix [B] consists of the terms expressed in area coordinates. Equa-
tion (10.71) is to be evaluated accordingly.

10.11 Allman’s Triangular Plane Stress Element

Allman’s triangular finite element for the plane elasticity analysis has three degrees
of freedom per node. With three corner nodes, total d.o.f per element is nine. Three
d.o.f per node include 2 translations and a drilling d.o.f.
In the derivation of the stiffness matrix, it is assumed that the displaced shape of
the element side is parabolic due to unequal drilling degrees of freedom at connecting
nodes as shown in Fig. 10.11. The edge displacement of an element side consists
of two parts: (a) a straight shape associated with the nodal d.o.f and (b) a parabolic
shape associated with drilling d.o.f as shown in Fig. 10.11.
The displacement at any point on the side of the element is first to be found out.
Let the edge normal displacement due to the unequal drilling d.o.f at the connecting
nodes be δ (Fig. 10.1b), then

δ = a0 + a1 s + a2 s 2 (10.72)

Boundary conditions are

10.11 Allman’s Triangular Plane Stress Element 219

v3

w3 u3 L/2

δ
L/2
v2

w1
2
u1 u2
w2
v1 (b) Edge normal displacement
due to drilling d.o.f
(a) Allman’s plane triangle
v3

3 u3 3 u3 v3

v6 v5
3
5 u5
6 u6
v2
v2
v1 v4
2
4 u2
u4 2 u2 2
1 u1
(c) Linear strain triangle (d) Displaced shape due to translateral
degrees of freedom

Fig. 10.11 Allman’s triangle

⎫
δ = 0 when s=0 ⎬
δ |= 0 when | s=L (10.73)
∂δ | ∂δ | ⎭
∂s s=L
− ∂s s=0 = w2 + w1

Substituting the boundary conditions of Eq. (10.73) into Eq. (10.72) and then
solving the resulting equation yields the values of the constants

a0 = 1
w1 − w2 w1 − w2
a1 = − , a2 = − (10.74)
2 2L
Therefore, Eq. (10.72) takes the following form when the constants are substituted
from Eq. (10.74),

(w2 − w1 ) s (L − s)
δ= (10.75)
2L
Since the angle between normal to the side and x-axis is α, the components in the
direction u and v are δ cos α and δ sin α respectively.
220 10 Isoparametric and Other Element Representations and Numerical Integrations

The displacement at any point on the side of the element connecting nodes i and
j are
( ) ( ) ( ) ( )
u L − s ui s uj s (L − s) ( ) cos α
= + + w j − wi (10.76)
v L vi L vj 2L sin α

The displacements at mid-sides 4, 5 and 6 are as follows (Fig. 10.11 (c))

( ) ( ) ( ) ( )
u4 1 u1 1 u2 w2 − w1 y2 − y1
= + + (10.77)
v4 2 v1 2 v2 8 x1 − x2
( ) ( ) ( ( ) )
u5 1
1 u3 w3 − w2 y3 − y2
u2
+ = + (10.78)
v5 2 v3
2 8 v2 x2 − x3
( ) ( ) ( ) ( )
u6 1 u1 1 u3 w1 − w3 y1 − y3
= + + (10.79)
v6 2 v1 2 v3 8 x3 − x1

A relationship is now to be found out between 12 translational d.o.f of the original

element having mid-side nodes to 9 degrees of freedom at the corner nodes.
They are

u 1 = u 1 , v1 = v1 , u 2 = u 2 , v2 = v2 , u3 = u3, v3 = v3 ,
u1 u2 y2 − y1 v1 v2 x1 − x2
u4 = + + (w2 − w1 ), v4 = + + (w2 − w1 )
2 2 8 2 2 8
u2 u3 y3 − y2 v2 v3 x2 − x3
u5 = + + (w3 − w2 ), v5 = + + (w3 − w2 )
2 2 8 2 2 8
u1 u3 y1 − y3 v2 v3 x3 − x1
u6 = + + (w1 − w3 ), v6 = + + (w1 − w3 )
2 2 8 2 2 8
(10.80)

The above equations in matrix form can be written as

{u 1 v1 u 2 · · · · · · v6 }T = [T ] {u 1 v1 w1 · · · · · · v3 w3 } (10.81)

[T ] is of size 12 × 9.
For the linear strain triangle (Fig. 10.11c),
10.11 Allman’s Triangular Plane Stress Element 221
⎧ ⎫
⎪
⎪ u1 ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪ v ⎪⎪
⎪ 1⎪
⎪ ⎪
⎪ u2 ⎪
⎪
( ) [ ]⎪
⎨ ⎪
⎪ ⎪
⎬
u N1 0 N2 0 N3 0 N4 0 N5 0 N6 0 .
=
v 0 N1 0 N2 0 N3 0 N4 0 N5 0 N6 ⎪⎪ . ⎪⎪
⎪
⎪ ⎪
⎪
⎪ . ⎪
⎪ ⎪
⎪
⎪ ⎪
⎪
⎪ . ⎪
⎪ ⎪
⎩ ⎪
⎪ ⎭
v6
⎧ ⎫
⎪
⎪ u1 ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪ v ⎪ ⎪
⎪ 1⎪
⎪ ⎪
⎪
⎪
⎪ w 1⎪⎪
⎪
⎨ ⎪ ⎬
.
= [N ] [T ] (10.82)
⎪
⎪ . ⎪⎪
⎪
⎪ ⎪
⎪
⎪
⎪ . ⎪⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ v 3 ⎪
⎩ ⎪ ⎭
w3

Following appropriate steps, it can be shown that the stiffness matrix of the
Allman’s triangle is
( (
[k]e = [T ]T [B]T [D] [B] [T ] d x d y (10.83)

where
⎡ ∂N ∂ N2
⎤
∂x
1
0 ∂x
· · · · · · · · · ∂∂Nx6 0
⎢ ∂ N1
0 · · · · · · · · · 0 ∂∂Ny6 ⎥
[B] = ⎣ 0 ∂y ⎦ (10.84)
∂ N1 ∂ N1 ∂ N2 ∂ N6 ∂ N6
∂y ∂x ∂y
· · · · · · · · · ∂y ∂x

3 × 12

and
⎡ ⎤
1ν 0
Et ⎣
[D] = ν1 0 ⎦
1 − ν2
0 0 1−ν
2
222 10 Isoparametric and Other Element Representations and Numerical Integrations

10.12 Computer Program for the Solution of Plane Stress

Problem Using Isoparametric Element

The computer program in FORTRAN for the solution of plane stress problems using
isoparametric linear element has been given in Table 10.3 and is also given in
the attached CD. The notations used have been explained at the beginning of the
computer program and different steps have been indicated with ‘comment’ state-
ments. The computer program as given solves only skew or rectangular orthotropic
plates. However, with slight modification, the computer program can be used for
the solution of plates of any general shape and material properties. The computer
program in C for the same analysis has also been given in the CD.
Numerical integration has been performed by Gauss method. Subroutine GACUV
supplied the necessary GAUSS-point values and their weighting function for two-
point and three-point integration. GACUV, as given, is limited to 3 Gauss-point
values.
The preparation of the input data of the computer program is pretty simple. The
input data for the plate of Fig. 10.12 is shown in Table 10.4. The plate is analysed for
bottom half and boundary conditions have been incorporated in the main computer
program. The element and node numbers have been shown in Fig. 10.12 in which
the plate is divided into 2 × 2 mesh. The computer output has been given in Table
10.5.
Exercise 10
10.1 Write the shape functions of the quadratic triangle and verify the shape
function properties.
10.2 Write the interpolation function for a line segment having 3 nodes in terms
of natural coordinates.
10.3 For a cubic triangle, write the shape functions in terms of area coordinates
and verify the properties of the shape function.
10.4 Verify the convergence criteria for the 3-noded triangle where shape functions
have been expressed in terms of area coordinates.
10.5 For the quadratic triangle, derive the stiffness matrix of the element.
10.6 Determine the shape functions for isoparametric cubic element.
10.7 For the derivation of the stiffness matrix of the rectangular element presented
in the chapter by using isoparametric element, evaluate explicitly the stiffness
matrix of the element.
10.8 For the computer program for the solution of plane stress problems by the
use of isoparametric linear element presented in this chapter, what changes
in the computer program will be needed if a plane strain problem is to be
solved by the use of isoparametric quadratic element?
10.9 For isoparametric quadratic element, develop expressions for the nodal forces
due to the presence of body forces.
10.10 Construct a displacement field parallel to x-axis (i.e. u) for the element.
10.12 Computer Program for the Solution of Plane Stress Problem … 223

Table 10.3 Finite element analysis of plane stress problems of plates using isoparametric linear
element
C ***************************************************************
**
C ***************************************************************
**
C FINITE ELEMENT ANALYSIS OF PLATES USING ISOPARAMETRIC LINEAR
C PLANE STRESS ELEMENT
C ISOTROPIC PLATE CASE
C ***************************************************************
***
DIMENSION STIFO(120,45),X(120,2),RGD(3,3),XG(4),YG(4),N
OD(100,4),
* ZX(4),ZY(4),SFD(4),SFDE(4),BB(3,8),ASAT(8,8),STIFF(8,8),
* PLOAD(120,2),DISP(8,2),STRESS(12,2),SFDZ(4)
COMMON STIFO,PLOAD
OPEN(7,FILE=’isoin’)
OPEN(8,FILE=’isout’)
C ***************************************************************
*****
C ***************************************************************
*****
C A LENGTH ALONG X-AXIS
C B LENGTH ALONG Y-AXIS
C THICK THICKNESS OF THE PLATE
C D FLEXURAL RIGIDITY OF THE PLATE
C PR POISSON’S RATIO
C EXL SIZE OF THE ELEMENT IN THE X-DIRECTION
C EYL SIZE OF THE ELEMENT IN THE Y-DIRECTION
C PHAI ANGLE BETWEEN THE SIDES IN DEGREES (=90 FOR RECTANGULAR
)
C NX,NY MESH DIVISIONS IN THE X- AND Y-DIRECTIONS
C NQ NO. OF GAUSS POINTS FOR THE EVALUATION OF STIFF
C NN NO. OF NODES IN AN ELEMENT
C NFREE DEGREES OF FREEDOM OF A NODE
C NT TOTAL DEGREES OF FREEDOM OF THE ENTIRE PLATE
C NBH HALFBAND WIDTH OF THE STIFFNESS MATRIX
C NQL NO. OF GAUSS POINTS IN THE EVALUATION OF THE LOADING MA-
TRIX
C NLC NO. OF LOADING CONDITIONS
C NELM TOTAL NO. OF ELEMENTS
C NENP TOTAL DEGREES OF FREEDOM OF EACH ELEMENT
C DXZ PARTIAL DERIVATIVE OF X WITH RESPECT TO ZI
C DXE PARTIAL DERIVATIVE OF X WITH RESPECT TO ETA
C DYZ PARTIAL DERIVATIVE OF Y WITH RESPECT TO ZI
C DYE PARTIAL DERIVATIVE OF Y WITH RESPECT TO ETA
C ZAC VALUE OF THE JACOBIAN
C DNX DERIVATIVE OF THE SHAPE FUNCTION W R TO X
C DNY DERIVATIVE OF THE SHAPE FUNCTION W R TO Y
C SFD SHAPE FUNCTION
C NOD GLOBAL NO. OF NODES OF AN ELEMENT
C ASAT PART OF THE ELEMENT STIFFNESS MATRIX
C RGD RIGIDITY MATRIX
C X DISPLACEMENT MATRIX
C PLOAD OVERALL LOADING MATRIX

(continued)
224 10 Isoparametric and Other Element Representations and Numerical Integrations

Table 10.3 (continued)

C STRESS STRESS VECTOR
C STIFF ELEMENT STIFFNESS MATRIX
C STIFO OVERAL STIFFNESS MATRIX
C XG X-COORDINATE OF THE ELEMENT
C YG Y-COORDINATE OF THE ELEMENT
C ZX NORMALISED X-COORDINATE
C ZY NORMALISED Y-COORDINATE
C SFDE PARTIAL DERIVATIVE OF SFD W R TO ETA
C SFDZ PARTIAL DERIVATIVE OF SFD W R TO ZI
C ***************************************************************
****
C ***************************************************************
****
C ***************************************************************
*****

100 FORMAT(2X,6G13.6)
101 FORMAT(16I5)
C*********************************************************************
*******
C INPUT OF MATERIAL AND GEOMETRIC PROPERTIES OF THE PLATE
C ***************************************************************
*******
C ***************************************************************
*******
103 READ(7,*) PHAI,A,B
READ(7,*) NN,NFREE,NQ,NLC,NX,NY
READ(7,*) D,PR,THICK
C ***************************************************************
***
C OUTPUT OF THE MATERIAL PROPERTIES OF THE PLATE
C ***************************************************************
****
C ***************************************************************
****
WRITE(8,*)’ ANALYSIS OF PLANE STRESS BY ISOPARAMETRIC
PLANE
* STRESS ELEMENT’
EXL=A/NX
EYL=B/NY
NELM=NX*NY
NX1=NX+1
NY1=NY+1
NT=(NX1*NY1)*NFREE
NBH=(NX1+2)*NFREE
NENP=NN*NFREE
WRITE(8,*)’ EXL EYL PHAI PR D
THICK’
WRITE(8,100)EXL,EYL,PHAI,PR,D,THICK
WRITE(8,*)’ NN NFREE NT NELM NENP NQ NBH NLC NX NY’
WRITE(8,101)NN,NFREE,NT,NELM,NENP,NQ,NBH,NLC,NX,NY
PI=3.1415927
PHAI=PI*PHAI/180.0

(continued)
10.12 Computer Program for the Solution of Plane Stress Problem … 225

Table 10.3 (continued)

C ***************************************************************
***
C ***************************************************************
***
C INPUT AND OUTPUT OF PLATE PROPERTIES ARE OVER
C ***************************************************************
***
C NODE NUMBERS FOR DIFFERENT ELEMENTS ARE GENERATED
C ***************************************************************
****
C ***************************************************************
****
NX1=NX+1
DO 10 II=1,NELM
NXI= (II-1)/NX
DO 11 I=1,2
11 NOD(II,I)=NXI*NX1+(II-NX*NXI)+I-1
DO 14 I=3,4
14 NOD(II,I)= NX1*(NXI+1)+(II-NX*NXI)+4-I
10 CONTINUE
C ***************************************************************
***
C ***************************************************************
***
C RIGIDITY MATRIX GENERATION STARTS
C ***************************************************************
***
C ***************************************************************
****
DO 23 I=1,3
DO 23 J=1,3
23 RGD(I,J)=0.0
PRR=PR*PR
RGD(1,1)= D/(1.0-PR*PR)
RGD(1,2)= PR*RGD(1,1)
RGD(2,1)= RGD(1,2)
RGD(2,2)= RGD(1,1)
RGD(3,3)= (1.0-PR)*RGD(1,1)/2.0
WRITE(8,*)’ RGD MATRIX’
WRITE(8,100)((RGD(I,J),I=1,3),J=1,3)
C ***************************************************************
****
C ***************************************************************
*****
C RIGIDITY MATRIX GENERATION OVER
C ***************************************************************
*****
DO 12 I=1,NENP
DO 12 J=1,NENP
12 STIFF(I,J)=0.0
C ***************************************************************
*****

(continued)
226 10 Isoparametric and Other Element Representations and Numerical Integrations

Table 10.3 (continued)

C ***************************************************************
*****
C ELEMENT COORDINATE INFORMATION IS PRESCRIBED
C ***************************************************************
*****
C ***************************************************************
****
XG(1)= EYL*COS(PHAI)
YG(1) = EYL*SIN(PHAI)
XG(2)= EXL+XG(1)
YG(2)= YG(1)
XG(3)= EXL
YG(3)= 0.0
XG(4)= 0.0
YG(4)= 0.0
ZX(1)= -1.0
ZY(1)= 1.0
ZX(2)= 1.0
ZY(2)= 1.0
ZX(3)= 1.0
ZY(3)= -1.0
ZX(4)= -1.0
ZY(4)= -1.0
C ***************************************************************
******
C ***************************************************************
*******
C ELEMENT STIFFNESS MATRIX GENERATION STARTS
C ***************************************************************
*****
C ***************************************************************
*****
DO 16 I=1,NQ
DO 16 J=1,NQ
790 CALL GACUV(I,J,ZI,ETA,HI,HJ,NQ)
791 DO 19 L=1,4
ZIZ= ZI*ZX(L)
ETAZ= ETA*ZY(L)
SFD(L)= 0.25*(1.0+ETAZ)*(1.0+ZIZ)
SFDZ(L)= 0.25*(1.0+ETAZ)*ZX(L)
19 SFDE(L)= 0.25*(1.0+ZIZ)*ZY(L)
DO 20 II=1,3
DO 20 JJ=1,NENP
20 BB(II,JJ)= 0.0
DXZ=0.0
DYZ=0.0
DXE=0.0
DYE=0.0
DO 21 KK=1,4
DXZ= DXZ+SFDZ(KK)*XG(KK)
DYZ= DYZ+SFDZ(KK)*YG(KK)
DXE= DXE+SFDE(KK)*XG(KK)
DYE= DYE+SFDE(KK)*YG(KK)

(continued)
10.12 Computer Program for the Solution of Plane Stress Problem … 227

Table 10.3 (continued)

21 CONTINUE
ZAC= DXZ*DYE - DYZ*DXE
DXZI= DYE/ZAC
DYZI= -DYZ/ZAC
DXEI= -DXE/ZAC
DYEI= DXZ/ZAC
DO 22 II=1,NN
K=2*(II-1)
DNX= SFDZ(II)*DXZI + SFDE(II)*DYZI
DNY= SFDZ(II)*DXEI + SFDE(II)*DYEI
BB(1,K+1)= DNX
BB(2,K+2)= DNY
BB(3,K+1)= DNY
BB(3,K+2)= DNX
22 CONTINUE
C ***************************************************************
*****
C STIFF MATRIX IS SYMMETRIC, ONLY UPPER HALF OF THE TRIANGLE IS
STORED
C ***************************************************************
******
DO 24 M=1,NENP
DO 24 N=1,NENP
IF(N-M) 24,611,611
611 ASAT(M,N)= 0.0
DO 24 L=1,3
DO 24 KI=1,3
ASAT(M,N)= ASAT(M,N)+ BB(L,M)*RGD(L,KI)*BB(KI,N)*ZAC*THICK
24 CONTINUE
DO 25 M=1,NENP
DO 25 N=1,NENP
IF (N-M) 25,615, 615
615 STIFF(M,N)= STIFF(M,N)+ASAT(M,N)*HI*HJ
25 CONTINUE
16 CONTINUE
DO 620 I=1,NENP
DO 620 J=1,NENP
620 STIFF(J,I)=STIFF(I,J)
WRITE(8,*)’ THE STIFF MATRIX’
WRITE(8,100)((STIFF(I,J),J=1,NENP),I=1,NENP)
C ***************************************************************
********
C ***************************************************************
********
C ELEMENT STIFFNESS MATRIX GENERATION OVER
C ***************************************************************
*******
C ***************************************************************
********
C ASSEMBLY OF OVERALL STIFFNESS MATRIX IN HALFBAND
C ***************************************************************
*******

(continued)
228 10 Isoparametric and Other Element Representations and Numerical Integrations

Table 10.3 (continued)

DO 31 I=1,NT
DO 31 J=1,NBH
31 STIFO(I,J)= 0.0
DO 26 II=1,NELM
CALL ASSEM(NOD,NN,NFREE,II,STIFF)
26 CONTINUE
C ***************************************************************
*****
C ***************************************************************
*****
C PLOAD MATRIX IS INITIALIZED IN DO LOOP 32
C ***************************************************************
******
C ***************************************************************
*******
DO 32 I=1,NT
DO 32 J=1,NLC
32 PLOAD(I,J)= 0.0
C ***************************************************************
***
C PLOAD MATRIX DATA ARE GIVEN AS THE INPUT
C ***************************************************************
***
312 READ(7,*)INP,JNP,VALUE
IF(INP) 314,314,310
310 PLOAD(INP,JNP)= PLOAD(INP,JNP)+VALUE
GO TO 312
C ***************************************************************
***
C ***************************************************************
***
C FITTING OF BOUNDARY CONDITIONS START
C ***************************************************************
***
314 NNO=NT/NFREE
NX1= NX+1
NBX= NNO-NX
DO 37 MN= NBX,NNO
NEWN= MN*NFREE-1
NEW2= NEWN+1
STIFO(NEWN,1)= 0.1E20
STIFO(NEW2,1)=0.1E20
37 CONTINUE
DO 877 MN=1,NX1
NEWN=MN*NFREE
877 STIFO(NEWN,1)=0.1E20
C **************************************************************
C ***************************************************************
C BOUNDARY CONDITION FITTING IS OVER
C ***************************************************************
*
C ***************************************************************
*

(continued)
10.12 Computer Program for the Solution of Plane Stress Problem … 229

Table 10.3 (continued)

C GAUSS IS A SUBROUTINE TO SOLVE SIMULTANEOUS EQUATIONS WHERE
C THE COEFFICIENT MATRIX IS STORED IN HALFBAND FORM
C ***************************************************************
***
WRITE(8,*)’ THE STIFO MATRIX’
WRITE(8,100)((STIFO(I,J),J=1,NBH),I=1,NT)
320 CALL GAUSS(X,NBH,NT,NLC)
WRITE(8,*)’ THE DISPLACEMENT MATRIX’
WRITE(8,100)((X(I,J),I=1,NT),J=1,NLC)
C ***************************************************************
**
C STRESS GENERATION STARTS
C ***************************************************************
**
WRITE(8,*)’ THE STRESS MATRIX’
DO 41 II=1,NELM
DO 518 LL=1,4
DO 519 L=1,4
ZIZ= ZX(LL)*ZX(L)
ETAZ= ZY(LL)*ZY(L)
SFD(L)= 0.25*(1.0+ETAZ)*(1.0+ZIZ)
SFDZ(L)= 0.25*(1.0+ETAZ)*ZX(L)
519 SFDE(L)= 0.25*(1.0+ZIZ)*ZY(L)
DO 520 K=1,3
DO 520 KK=1,NENP
520 BB(K,KK)=0.0
DXZ=0.0
DYZ=0.0
DXE=0.0
DYE=0.0
DO 521 KK=1,4
DXZ=DXZ+SFDZ(KK)*XG(KK)
DYZ=DYZ+SFDZ(KK)*YG(KK)
DXE= DXE+SFDE(KK)*XG(KK)
521 DYE= DYE+SFDE(KK)*YG(KK)
DXZI= DYE/ZAC
DYZI= -DYZ/ZAC
DXEI=- DXE/ZAC
DYEI= DXZ/ZAC
DO 522 KK=1,NN
K=2*(KK-1)
DNX= SFDZ(KK)*DXZI + SFDE(KK)*DYZI
DNY= SFDZ(KK)* DXEI +SFDE(KK)* DYEI
BB(1,K+1)= DNX
BB(2,K+2) = DNY
BB(3,K+1) = DNY
522 BB(3,K+2)= DNX
DO 570 L=1,NLC
DO 570 KK=1,NN
DO 570 I=1,2
K=(NOD(II,KK)-1)*NFREE+I
J=(KK-1)*NFREE+I
570 DISP(J,L)= X(K,L)

(continued)
230 10 Isoparametric and Other Element Representations and Numerical Integrations

Table 10.3 (continued)

DO 525 J=1,NLC
DO 524 M=1,3
STRESS(M,J)= 0.0
DO 524 L=1,3
DO 524 K=1,NENP
524 STRESS(M,J)= STRESS(M,J)+ RGD(M,L)*BB(L,K)*DISP(K,J)
WRITE(8,*)’ ELE NO= ‘,II,’ NODE NO= ‘,LL,’ LOAD COND= ‘
* ,J
WRITE(8,100)(STRESS(M,J),M=1,3)
525 CONTINUE
518 CONTINUE
41 CONTINUE
GO TO 103
END
C *************************************************************
C *************************************************************
SUBROUTINE GACUV(I,J,ZI,ETA,HI,HJ,NQ)
IF(NQ-2) 2,1,2
1 IF(I-1) 4,3,4
3 ZI= -0.577350269189626
HI= 1.000000000000000
GO TO 7
4 ZI= 0.57735026918926
HI= 1.000000000000000
7 IF(J-1) 5,6,5
6 ETA= -0.57735026918926
HJ= 1.000000000000000
GO TO 18
5 ETA= 0.577350269189626
HJ= 1.00000000000000000
GO TO 18
2 IF (I-1) 8,9,8
9 ZI= -0.774596669241483
HI=0.5555555555555556
GO TO 12
8 IF (I-2) 10,11,10
11 ZI= 0.000000000000000
HI= 0.8888888888888889
GO TO 12
10 ZI= 0.774596669241483
HI= 0.555555555555556
12 IF(J-1) 14,15,14
15 ETA= -0.774596669241483
HJ= 0.55555555555555556
GO TO 18
14 IF(J-2) 16,17,16
17 ETA=0.00000000000000000
HJ=0.888888888888888889
GO TO 18
16 ETA= 0.774596669241483
HJ= 0.5555555555555556
18 RETURN
END
C ***********************************************************
10.12 Computer Program for the Solution of Plane Stress Problem … 231

1/unit length

1 2 3
1 0.125
t = 1.0
1 2
v = 0.3 6
y 4 0.25
3 4
x 7 9
1 0.5 8 0.5

Fig. 10.12 A rectangular plate

Table 10.4 Input data for the plane stress problem

90.0 1.0 0.5
4 2 2 1 2 2
1.0 0.25 1.0
5 1 -0.125
11 1 -0.25
0 0 0.0

1 2 3

8 4

7 6 5

Prob. 10.10
232 10 Isoparametric and Other Element Representations and Numerical Integrations

Table 10.5 Computer output for the plane stress problem using isoparametric element
Analysis of Plane Stress by Isoparametric Plane Stress Element
Exl = 0.500 Eyl = 0.250 Phi(F) = 90.000 Pr = 0.250 D =
1.000 Thick = 1.000
NN = 4 NFREE = 2 NT = 18 NELM = 4 NENP = 8
NQ = 2 NBH = 10 NLC = 1 NX = 2 NY = 2
The RGD Matrix
1.06667e+000 2.66667e-001 0.00000e+000
2.66667e-001 1.06667e+000 0.00000e+000
0.00000e+000 0.00000e+000 4.00000e-001
The Element Stiffness Matrix
4.44445e-001 -1.66667e-001 -4.44444e-002 3.33333e-002
-2.22222e-001 1.66667e-001 -1.77778e-001 -3.33333e-002

-1.66667e-001 7.77778e-001 -3.33333e-002 2.88889e-001

1.66667e-001 -3.88889e-001 3.33333e-002 -6.77778e-001

-4.44444e-002 -3.33333e-002 4.44444e-001 1.66667e-001

-1.77778e-001 3.33333e-002 -2.22222e-001 -1.66667e-001

3.33333e-002 2.88889e-001 1.66667e-001 7.77778e-001

-3.33333e-002 -6.77778e-001 -1.66667e-001 -3.88889e-001

-2.22222e-001 1.66667e-001 -1.77778e-001 -3.33333e-002

4.44444e-001 -1.66667e-001 -4.44444e-002 3.33333e-002

1.66667e-001 -3.88889e-001 3.33333e-002 -6.77778e-001

-1.66667e-001 7.77778e-001 -3.33333e-002 2.88889e-001

-1.77778e-001 3.33333e-002 -2.22222e-001 -1.66667e-001

-4.44444e-002 -3.33333e-002 4.44445e-001 1.66667e-001

-3.33333e-002 -6.77778e-001 -1.66667e-001 -3.88889e-001

3.33333e-002 2.88889e-001 1.66667e-001 7.77778e-001

The Overall Stiffness Matrix

4.44445e-001 -1.66667e-001 -4.44444e-002 3.33333e-002 0.00000e+000
0.00000e+000 -1.77778e-001 -3.33333e-002 -2.22222e-001 1.66667e-001

1.00000e+019 -3.33333e-002 2.88889e-001 0.00000e+000 0.00000e+000

3.33333e-002 -6.77778e-001 1.66667e-001 -3.88889e-001 0.00000e+000

8.88889e-001 0.00000e+000 -4.44444e-002 3.33333e-002 -2.22222e-001

-1.66667e-001 -3.55556e-001 0.00000e+000 -2.22222e-001 1.66667e-001

1.00000e+019 -3.33333e-002 2.88889e-001 -1.66667e-001 -3.88889e-001

0.00000e+000 -1.35556e+000 1.66667e-001 -3.88889e-001 0.00000e+000

4.44444e-001 1.66667e-001 0.00000e+000 0.00000e+000 -2.22222e-001

-1.66667e-001 -1.77778e-001 3.33333e-002 0.00000e+000 0.00000e+000

1.00000e+019 0.00000e+000 0.00000e+000 -1.66667e-001 -3.88889e-001

-3.33333e-002 -6.77778e-001 0.00000e+000 0.00000e+000 0.00000e+000

8.88889e-001 0.00000e+000 -8.88889e-002 0.00000e+000 0.00000e+000

0.00000e+000 -1.77778e-001 -3.33333e-002 -2.22222e-001 1.66667e-001

(continued)
10.12 Computer Program for the Solution of Plane Stress Problem … 233

Table 10.5 (continued)

1.55556e+000 -3.72529e-009 5.77778e-001 0.00000e+000 0.00000e+000
3.33333e-002 -6.77778e-001 1.66667e-001 -3.88889e-001 0.00000e+000

1.77778e+000 1.49012e-008 -8.88889e-002 0.00000e+000 -2.22222e-001

-1.66667e-001 -3.55556e-001 0.00000e+000 -2.22222e-001 1.66667e-001

3.11111e+000 -3.72529e-009 5.77778e-001 -1.66667e-001 -3.88889e-001

0.00000e+000 -1.35556e+000 1.66667e-001 -3.88889e-001 0.00000e+000

8.88889e-001 1.49012e-008 0.00000e+000 0.00000e+000 -2.22222e-001

-1.66667e-001 -1.77778e-001 3.33333e-002 0.00000e+000 0.00000e+000

1.55556e+000 0.00000e+000 0.00000e+000 -1.66667e-001 -3.88889e-001

-3.33333e-002 -6.77778e-001 0.00000e+000 0.00000e+000 0.00000e+000

1.00000e+019 1.66667e-001 -4.44444e-002 -3.33333e-002 0.00000e+000

0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000

1.00000e+019 3.33333e-002 2.88889e-001 0.00000e+000 0.00000e+000

0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000

1.00000e+019 1.49012e-008 -4.44444e-002 -3.33333e-002 0.00000e+000

0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000

1.00000e+019 3.33333e-002 2.88889e-001 0.00000e+000 0.00000e+000

0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000

1.00000e+019 -1.66667e-001 0.00000e+000 0.00000e+000 0.00000e+000

0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000

1.00000e+019 0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000

0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000 0.00000e+000

The Displacement Matrix

-1.56967e-001 -2.77353e-021 -2.78398e-001 1.82393e-021 -6.12379e-001
8.95011e-021 -1.19232e-001 -2.18991e-002 -1.82388e-001 -3.04039e-002
-4.91564e-001 5.42437e-002 -6.60647e-021 -6.10388e-021 -1.87891e-020
-9.06911e-021 -1.21044e-020 7.17248e-021
The STRESS Matrix
Element No. = 1 Node No. = 1 Load Condition = 1
-2.35693e-001 2.86729e-002 -6.03769e-002
Element No. = 1 Node No. = 2 Load Condition = 1
-2.26622e-001 6.49602e-002 -1.53617e-001
Element No. = 1 Node No. = 3 Load Condition = 1
-1.02302e-001 9.60402e-002 -1.60421e-001
Element No. = 1 Node No. = 4 Load Condition = 1
-1.11374e-001 5.97529e-002 -6.71808e-002
Element No. = 2 Node No. = 1 Load Condition = 1
-6.80062e-001 -4.83998e-002 -1.53617e-001
Element No. = 2 Node No. = 2 Load Condition = 1
-7.70352e-001 -4.09563e-001 -1.93304e-001
Element No. = 2 Node No. = 3 Load Condition = 1
-7.17436e-001 -3.96334e-001 -1.25586e-001

(continued)
234 10 Isoparametric and Other Element Representations and Numerical Integrations

Table 10.5 (continued)

Element No. = 2 Node No. = 4 Load Condition = 1
-6.27145e-001 -3.51707e-002 -8.58987e-002
Element No. = 3 Node No. = 1 Load Condition = 1
-1.58092e-001 -1.27119e-001 -1.97575e-001
Element No. = 3 Node No. = 2 Load Condition = 1
-1.67164e-001 -1.63406e-001 -2.98624e-001
Element No. = 3 Node No. = 3 Load Condition = 1
-3.24308e-002 -1.29723e-001 -2.91820e-001
Element No. = 3 Node No. = 4 Load Condition = 1
-2.33590e-002 -9.34360e-002 -1.90771e-001
Element No. = 4 Node No. = 1 Load Condition = 1
-6.92007e-001 -2.94617e-001 -2.24102e-001
Element No. = 4 Node No. = 2 Load Condition = 1
-6.01716e-001 6.65455e-002 -7.18784e-001
Element No. = 4 Node No. = 3 Load Condition = 1
5.78599e-002 2.31440e-001 -7.86503e-001
Element No. = 4 Node No. = 4 Load Condition = 1
-3.24308e-002 -1.29723e-001 -2.91820e-001

References and Suggested Readings

1. O.C. Zienkiewicz, R.L. Taylor, The Finite Element Method, vols 1 and 2 (McGraw-Hill, 1989)
2. D.J. Allman, A compatible triangular element including vertex rotations for plane elasticity
analysis. Comput. Struct. 19(1), 1–8 (1984)
3. J.L. Batoz, K.J. Bathe, L.W. Ho, A study of the three-node triangular element. J. Num. Meth.
Engg. 115(12), 1771–1812 (1980)
4. B.M. Irons, Engineering application numerical integration in stiffness method. J. AIAA 14,
2035–2037 (1966)
5. C. Maforano, S. Odorizzi, R. Vitaliani, Shortened quadrature rules for finite elements. Adv. Eng.
Softw. 4(2) (1982)
6. J.E. Akin, Application and Implementation of Finite Element Methods (Academic Press, 1982)
7. M.J. Turner, R.W. Clough, H.C. Martin, L.P. Topp, Stiffness and deflection analysis of complex
structures. J. Aero. Sc. 23(9), 805–823 (1956)
8. G.R. Cowper, Gaussian quadrature formulas for triangles. Int. J. Numer. Meth. Eng. 7, 405–408
(1973)
9. G.R. Cowper, G.M. Lindberg, M.B. Olson, A shallow shell finite element of triangular shape.
Int. J. Solids Struct. 6, 1133–1156 (1970)
Chapter 11
Finite Element Analysis of Plate Bending
Problems

11.1 Introduction

Bending of plates forms another important class of problems. Solutions obtained by

the application of classical theory of plate flexure are limited mainly to simple types
of plates with simple loading and boundary conditions. With the advent of the finite
element method, the plate bending problems have received considerable attention.
As a result of which, a large number of different plate bending element formulations
have been made (11.1). This chapter is limited to the discussion of some of these
elements.
For most of the elements, the starting point is the choice of the displacement
function. However, attempts have been made to directly formulate the shape func-
tions. The determination of the shape functions now becomes more difficult. The
continuity of displacements along the edges now implies continuity of slope. The
elements, which are formulated by satisfying both the deflection and slope continuity
along the edges, are referred to as conforming elements. Elements, which preserve
the continuity of deflection, but violate the condition of slope compatibility is referred
to as ‘non-conforming elements’. These elements, however, do converge.
For conforming elements, the formulation may be quite tedious. Though these
elements gradually converge to the correct solution, they may yield inferior accuracy
in some cases. Hence, non-conforming elements are of much greater practical usage.

11.2 Beam Element

A uniform beam element is considered and is shown in Fig. 11.1. Axial and torsional
deformations have not been considered. At each node at two ends of the member,
there are two degrees of freedom—deflection w and rotation dw/dx. The steps for
deriving the stiffness matrix of the beam element are same as those for plane elasticity
problems. The derivation of the stiffness matrix of the beam element is presented in

© The Author(s) 2022 235

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_11
236 11 Finite Element Analysis of Plate Bending Problems

Fig. 11.1 Beam element P1 –X1 Z P –X

P2 –X2
X

P4 –X4
L

the following. The displacement and the force vectors for the beam element are
{ }
{X }eT = X 1 X 2 X 3 X 4 (11.1)

{ }
{P}eT = P1 P2 P3 P4 (11.2)

11.2.1 Displacement Function

Total degrees of freedom for this element are four. Therefore, four constants are to be
associated with the displacement function. Unlike plane elasticity problems, where
u and v have been independently prescribed, once the deflection is written, the slope
can be obtained by its differentiation. Therefore, all the four unknown coefficients
are reflected in the expression for deflection. The expression for deflection is given
by the following polynomial expression.

{ f } = w = α1 + α2 x + α3 x 2 + α4 x 3 (11.3)

or
⎧ ⎫
⎪
⎪ α1 ⎪
[ ] ⎨ ⎪ ⎬
α
{ f } = 1 x x2 x3 2
⎪
⎪ α ⎪
⎩ 3⎪ ⎭
α4

or

{ f } = [C]{α} (11.4)

Though f indicates a single quantity for this element, the matrix notation of f is
given for the sake of generality.
11.2 Beam Element 237

11.2.2 Displacement Function in Terms of Nodal

Displacements

Differentiating w with respect to x in Eq. (11.3) yields

dw
= α2 + 2α3 + 3α4 x 2 (11.5)
dx
Node i is chosen as the origin.
At node i, x = 0 and at node j, x = L. Substituting the respective values of x in
Eqs. (11.3) and (11.5).
⎫
X1 = (w) ⎪
⎪
( ) x=0
⎪
⎬
X2 = dwd x x=0
3⎪ (11.6)
X3 = (w) ) = α1 + α2 L + α3 L + 2α4 L ⎪
( dw x=L
2
⎪
⎭
X4 = d x x=L = α2 + 2α3 L + 3α4 L

Equation (11.6) in matrix form becomes

⎧ ⎫ ⎡ ⎤⎧ ⎫
⎪
⎪ X1 ⎪ 1 0 0 0 ⎪ α1 ⎪
⎨ ⎪ ⎬ ⎢ ⎪
⎨ ⎪ ⎬
X2 0 1 0 0 ⎥ α
=⎢
⎣1
⎥
3 ⎦⎪
2
(11.7)
⎪
⎪ X3 ⎪
⎪ L L2 L ⎪ α ⎪
⎩ ⎭ ⎩ 3⎪ ⎭
X4 0 1 2L 3L 2 α4

or

{X }e = [A]{α} (11.8)

Inverting Eq. (11.8) becomes

{α} = [A]−1 {X }e (11.9)

Substituting {α} from Eq. (11.9) into Eq. (11.4) yields

{ f } = [C] [A]−1 {X }e
{ f } = w = [N ] {X }e (11.10)

For the beam element,

⎡ ⎤
1 0 0 0
⎢ 0 1 0 0 ⎥
[A]−1 =⎢
⎣ −32 −2 3
⎥
−1 ⎦ (11.11)
L L L2 L
2
L3
1
L2
− L23 L12
238 11 Finite Element Analysis of Plate Bending Problems

[( )( )( 2 )( 2 )]
3x 2 2x 3 2x 2 x3 3x 2x 2 −x x3
[N ] = 1 − 2 + 3 x− + 2 − 3 + 2
L L L L L2 L L L
(11.12)

11.2.3 Strain (Curvature)–Nodal Parameter Relationship

For the beam element, the only ‘strain’, component present is the curvature where
the shear deformation is not considered. The ‘strain’ is then

d 2w
{ε} = − (11.13)
dx2
Substituting [N] from Eq. (11.12) into Eq. (11.10) and then differentiating twice,
it gives Eq. (11.13) as
[( )( )( )( )]
6 12x 4 6x 6x 12 2 6x
[ε] = 2
− 3 − 2 − 2
+ 3 − 2 {X }e (11.14)
L L L L L L L L
or

{ε} = [B] {X }e (11.15)

11.2.4 Stress (Moment)–Strain (Curvature) Relationship

The ‘stress’ referred to in the beam element is the bending moment. The following
is the moment–curvature relationship

d 2w
M = −E I (11.16)
dx2
In conformity with the standard notations used, Eq. (11.16)

{σ } = [D]{ε}
{σ } = [M] and [D] = [E I ] (11.17)

Combining Eqs. (11.15) and (11.17), yields

{σ } = [D] [B] {X }e (11.18)

11.2 Beam Element 239

11.2.5 Derivation of the Element Stiffness Matrix

The element stiffness matrix of the beams is derived here using the principle of virtual
work.
Let a virtual displacement δ{X }e be imposed at the nodes. The work done by the
external forces due to this virtual displacement is given by

δ{X }eT {P}e

i.e. they constitute the product of the individual nodal forces and their corresponding
virtual displacements.
Due to the virtual displacements, the displacements and the strains will undergo
virtual changes and they become (from Eqs. (11.10) and (11.15))

δ{ f } = δw = [N ] δ {X }e (11.19)

δ{ε} = [B] δ {X }e (11.20)

The internal work done per unit length by the stresses (moment) is

δ{ε}T {σ }

Substituting respective values from Eqs. (11.18) and (11.20), the internal work
per unit length = δ{X }eT [B]T [D]{X }e .
Total internal work over the entire element is

(L
δ{X }eT [B]T [D] [B] d x {X }e
0

Equating the external work to the internal work

(L
δ{X }eT {P}e = δ {X }eT [B]T [D] [B] d x {X }e
0

or,

(L
{P}e = [B]T [D] [B] d x {X }e
0

or,
240 11 Finite Element Analysis of Plate Bending Problems

{P}e = [K ]e {X }e (11.21)

where

(L
[k]e = [B]T [D] [B] d x (11.22)
0

[k]e can be explicitly evaluated. After performing the matrix multiplication,

Eq. (11.22) becomes
⎡ 2
⎤
36 − 144x + 144x symmetrical
⎢ L4 L5 L6 ⎥
⎢
( ⎢ 24 − 84x + 72x 2
L 16 − 48x + 36x 2 ⎥
⎥
⎢ L3 4 5 L2 L3 L4 ⎥
[k]e = E I ⎢ L L ⎥d x
⎢ − 36 + 144x − 144x 2 24 + 84x − 72x 2 36 − 144x + 144x 2 ⎥
⎢
0 ⎣ L 4 5 L6 3 4 L5 L4 L5 L6
⎥
L L L ⎦
12 − 60x + 72x 2 8 − 36x + 36x 2 − 12 + 60x − 72x 2 4 − 24x +36x 2
3
L L 4 L5 L2 L3 L4 L3 L4 L5 L2 L3 L4
⎡ ⎤
12 symmetrical
⎢ 2 ⎥
EI ⎢ 6L 4 L ⎥
[k]e = 3 ⎢
⎣ −12
⎥
⎦
(11.23)
L −6 L 12
6L 2 L2 −6 L 4 L 2

11.2.6 Determination of Equivalent Loading on the Beam

If the loads applied are such that they correspond to the nodal displacements, then
the loading vector can be formed directly by intuitive reasoning. But in many cases,
loads act inside the element or the nodal parameters may be such that there may not
exist any corresponding ‘physical quantity’. For such cases consistent loading vector
or equivalent loading at the nodes are to be obtained by the application of variational
principles.
Let {P}e be the equivalent nodal forces. Due to the virtual displacement δ{X }eT ,
the work done by the nodal forces is given by

δ{X }eT {P}e

If {p} is the loading within the element—it may be distributed or it may have a
discrete number of concentrated loads or it may have any other variation within the
element, then the work done by this loading is

(L
δ { f }T { p} d x
0
11.3 Rectangular Plate Bending Element 241

From Eq. (11.19), the work done by the load inside the element is

(L
δ {X }eT [N ]T { p} d x
0

Equating both the work,

(L
δ {X }eT {P}e = {X }eT δ [N ]T { p} d x
0

or

(L
{P}e = [N ]T { p} d x (11.24)
0

Equation (11.24) is the next general expression for determining equivalent nodal
forces. If {p} is uniformly distributed load, i.e. {P} = q, then Eq. (11.24) becomes

(L
{P}e = q [N ]T d x (11.25)
0

For the beam element, substituting the value of [N] from Eq. (11.12) into Eq. (11.25)
and performing the necessary integration, yields
⎧ ⎫
⎪
⎪ q L/2 ⎪
⎪
⎨ ⎬
q L 2 /12
{P}e = (11.26)
⎪
⎪ q L/2 ⎪
⎪
⎩ ⎭
−q L 2 /12

11.3 Rectangular Plate Bending Element

The rectangular elements can be effectively used for plates having rectangular edges.
Rectangular elements can also be employed for irregular plates in conjunction with
other types of elements (e.g. triangular elements). A node of the plate bending element
will have three degrees of freedom—the transverse deflection and two orthogonal
rotations.
The rectangular plate bending element along with their dimensions, coordinate
system and node numbering as shown in Fig. 11.2a. The positive directions of all
242 11 Finite Element Analysis of Plate Bending Problems

Fig. 11.2 Rectangular plate y

bending element
2
4
θy, My
b

θx, Mx
t w, P
1 3 x
a
w, P

the displacements are shown in Fig. 11.2b. The positive directions of {rotations are}
indicated by right-hand screw rule.
{ The displacements
} at node 1 are w1 θx1 θ y1
and the corresponding forces are P1 Mx1 M y1 . Therefore, complete displacement
vectors for this element are
{ }
{X }eT = w1 θ x1 θ y1 w2 θ x2 θ y2 w3 θx3 θ y3 w4 θx4 θ y4 (11.27)

{ }
{P}eT = P1 Mx1 M y1 P2 Mx2 M y2 P3 Mx3 M y3 P4 Mx4 M y4 (11.28)

11.3.1 Displacement Function

There are three degrees of freedom associated with each node. So for a four-noded
rectangle, there are in all twelve degrees of freedom. The polynomial expression
to be chosen should have twelve terms. A complete cubical polynomial (Fig. 11.3)
contains 10 terms of quartic polynomial that are to be eliminated. A suitable function
is given by

{ f } = w = a1 + a2 x + α3 y + α4 x 2 + α5 x y + α6 y 2 + α7 x 3 + α8 x 2 y
+ α9 x y 2 + α10 y 3 + α11 x 3 y + α12 x y 3 (11.29)

or,
⎧ ⎫
⎪
⎪ α1 ⎪⎪
⎨ α2 ⎪
⎪ ⎬
[ ]
{ f } = 1 x y x 2 x y y2 x 3 x 2 y x y2 y3 x 3 y x y3 . (11.30)
⎪
⎪ .. ⎪
⎪
⎪
⎩ ⎪
⎭
α12

or,
11.3 Rectangular Plate Bending Element 243

Fig. 11.3 Parallelogram y η

element 3(–a, b ) 4(a, b)

ξ, x

1
2
(–a,–b) (a,–b)

{ f } = [C] {α} (11.31)

The displacement function of Eq. (11.29) gives the following expression for
rotations.
∂w ( )
θx = − = − α3 + α5 x + 2α6 y + α3 x 2 + 2α9 x y + 3α10 y 2 + α11 x 3 + 3α12 x y 2
∂y
(11.32)

and
∂w
θy = = α2 + 2α4 x + α5 y + 3α7 x 2 + 2α8 x y + α9 y 2 + 3α11 x 2 y + α12 y 3
∂x
(11.33)

11.3.1.1 Checking the Continuity Requirements of the Displacement

Function

Let us verify whether the displacement function chosen satisfies the continuity of the
slopes along the edges. Let it be checked with respect to the edge x = a.

w = α1 + α2 .a + α3 y + α4 .a 2 + α5 ay + α6 y 2 + α7 a 3 + α8 a 2 y
+ α9 ay 2 + α10 y 3 + α11 a 3 y + α12 ay 3
( ) ( )
w = α1 + α2 a + α4 a 2 + α7 a 3 + α3 + α5 a + α8 · a 2 + α11 a 3 y
+ (α6 + α3 a) y 2 + (α10 + α12 a)y 3 (11.34)

The expression given by (11.34) is a cubic polynomial in y. If the flexural action

of the edge in the direction of y is considered (treating the edge as a beam), a
cubic polynomial uniquely defines the end deflections and end slopes. Thus the
compatibility of the deflections amongst the common boundary between the adjacent
elements is ensured.
244 11 Finite Element Analysis of Plate Bending Problems

At x = a, θ y is given by
( )
( ) ∂w ( )
θ y x=a = = α2 + 2α4 a + 3α7 a 2
∂x x=a
( )
+ α5 + 2α8 a + 3α11 a 2 y + α9 y 2 + α12 y 3 (11.35)

This cross-rotation term θ y along the edge is similar to the angle of twist in a
bar subjected to torsion. For this bar, a two-term polynomial uniquely determines
the displacement field. Equation (11.35) represents a cubic function, which does not
uniquely define the cross-rotation values along the edge. θ y is, therefore, discontin-
uous along this edge. Similar conclusions can be arrived at the other edges of the
element.
This element, therefore, does not satisfy all slope continuity requirements and is
termed as ‘non-conforming’. This aspect has been discussed at the end of the chapter.

11.3.2 Displacement Function Expressed in Terms of Nodal

Displacements

The coordinates of nodes 1, 2, 3 and 4 are (0, 0), (0, b), (a, 0) and (a, b) respectively.
Substituting the values of the nodal coordinates in Eqs. (11.29), (11.32) and (11.33)
respectively the following equations result.
Inverting Eq. (11.37)

{α} = [A]−1 {X }e (11.38)

Combining Eqs. (11.31) and (11.38) yields

{ f } = w = [C] [A]−1 {X }e

or

{ f } = w = [N ] {X }e (11.39a)

where

[N ] = [C] [A]−1 (11.39b)

11.3 Rectangular Plate Bending Element 245
⎧ ⎫ ⎡ ⎤ ⎧ ⎫
⎪
⎪ w1 ⎪ ⎪ 1 0 0 0 0 0 0 0 0 0 0 0 ⎪
⎪ α1 ⎪ ⎪
⎪
⎪ ⎪
⎪ ⎢0 ⎪ ⎪
⎪
⎪
⎪ θ ⎪
x1 ⎪
⎪ ⎢ 0 −1 0 0 0 0 0 0 0 0 0 ⎥ ⎥
⎪
⎪
⎪
⎪ α2 ⎪ ⎪
⎪
⎪
⎪
⎪ ⎪ ⎢ ⎥ ⎪ ⎪
⎪
⎪ θ y1 ⎪
⎪
⎪ ⎢0 1 0 0 0 0 0 0 0 0 0 0 ⎥ ⎪
⎪
⎪ α3 ⎪ ⎪
⎪
⎪
⎪ ⎪ ⎢ ⎥ ⎪ ⎪
⎪
⎪ w2 ⎪ ⎪
⎪ ⎢1 0 b 0 0 b2 0 0 0 b3 0 0 ⎥ ⎪
⎪
⎪ α4 ⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎢ ⎥ ⎪
⎪ ⎪
⎪
⎪
⎪ θ ⎪
⎪
⎪
⎢0 0 −1 0 0 −2b 0 0 0 −3b2 0 0 ⎥ ⎪
⎪ α5 ⎪ ⎪
⎪
⎬ ⎢ ⎥ ⎪ ⎪
x2
⎨ ⎢0 ⎨ ⎬
θ y2 1 0 0 b 0 0 0 b2 0 0 b3 ⎥ α6
=⎢
⎢1
⎥
⎪
⎪ w ⎪
3 ⎪ ⎢ a 0 a2 0 0 a3 0 0 0 0 0 ⎥ ⎥ ⎪
⎪ α7 ⎪ ⎪
⎪
⎪ ⎪
⎪ ⎢0 ⎪ ⎪
⎪
⎪
⎪ θ x3 ⎪
⎪
⎪ ⎢ 0 −1 0 −a 0 0 −a 2
0 0 −a 3 0 ⎥ ⎥
⎪
⎪
⎪
⎪ α 8
⎪
⎪
⎪
⎪
⎪
⎪ ⎪
⎪ ⎢ ⎥ ⎪
⎪ ⎪
⎪
⎪
⎪ θ y3 ⎪
⎪ ⎢0 1 0 2a 0 0 3a 2 0 0 0 0 0 ⎥ ⎪
⎪ α 9 ⎪
⎪
⎪
⎪ ⎪
⎪ ⎢ 3 ⎥ ⎪
⎪ ⎪
⎪
⎪
⎪ w ⎪
4 ⎪ ⎢1 a b a2 ab b2 a3 a 2 b ab2 b3 a b ab ⎥
3 ⎪
⎪ α10 ⎪⎪
⎪
⎪ ⎪
⎪ ⎢ ⎥ ⎪
⎪ ⎪
⎪
⎪
⎪ θ ⎪ ⎣0 −1 −a −2b −a 2 −2ab −3b2 −a 3 −3ab2 ⎦ ⎪ α ⎪
⎪ x4 ⎪
⎩ ⎪
⎭
0 0 0 ⎪
⎪
⎩
11 ⎪
⎪
⎭
θ y4 0 1 0 2a b 0 3a 2 2ab b2 0 3a 2 b b3 α12
(11.36)

or

{X }e = [A] {α} (11.37)

11.3.3 Strain–Nodal Parameter Relationship

The ‘strains’ in plate bending problem are the curvatures. The strain matrix is given
by
⎧ 2 ⎫
∂ w
⎪
⎨ − ∂ 2x 2 ⎪
⎬
{ε} = − ∂∂ yw2 (11.40)
⎪
⎩ 2 ∂2w ⎪
⎭
∂x ∂y

By directly differentiating w given in Eq. (11.29) with respect to the quantities

indicated in Eq. (11.40), yields
⎧ ⎫
⎨ −(2 α4 + 6α7 x + 2α8 y + 6α11 x y) ⎬
{ε} = −(2 α6 + 2α9 x + 2α10 y + 6α12 x y) (11.41)
⎩ ⎭
2 α5 + 4 α8 x + 4 α9 y + 6 α11 x 2 + 6 α12 y 2

or

{ε} = [Q] {α} (11.42)

where
246 11 Finite Element Analysis of Plate Bending Problems
⎡ ⎤
0 0 0 −2 0 0 −6x −2y 0 0 −6x y 0
[Q] = ⎣ 0 0 0 0 0 −2 0 0 −2x −6y 0 −6x y ⎦ (11.43)
0 0 0 0 2 0 0 4x 4y 0 6x 2 6y 2

Substituting {α} from Eq. (11.38) into Eq. (11.42), yields

{ε} = [Q] [A]−1 {X }e

or

{ε} = [B] {X }e (11.44)

Equations (11.42) and (11.43) reveal that the displacement function of Eq. (11.29)
satisfies one of the requirements of convergence, as it contains constant strain
(curvature) terms.

11.3.4 Stress (Moment)–Strain (Curvature) Relationship

The moment–curvature relationship for the orthotropic plate has been deduced in
Chap. 6 and they are reproduced below for the orthotropic plate
⎧ ⎫ ⎡ ⎤ ⎧ ∂2w ⎫
⎨ Mx ⎬ D x D1 0 ⎪
⎨ − ∂2x 2 ⎪
⎬
⎣
{σ } = M y = D1 D y 0 ⎦ = − ∂∂ yw2 (11.45)
⎩ ⎭ ⎪
⎩ 2 ∂2w ⎪
⎭
Mx y 0 0 Dx y ∂x ∂y

The orthotropic constants are given by

(E I )x (E I ) y
Dx = , Dy = , D1 = ν y Dx = D y νx and
1 − νx ν y 1 − νx ν y
( √ )
1 − νx ν y √
Dx y = Dx D y
2
For an isotropic plate the constants of Eq. (11.45) will be

Et 3
Dx = D y = D = ( )
12 1 − ν 2

D1 = ν D (11.46)

1−ν
Dx y = D
2
11.3 Rectangular Plate Bending Element 247

Equation (11.45) in compact form becomes

{σ } = [D] {ε} (11.47)

Substituting {ε} from Eq. (11.44) into Eq. (11.47) yields

{σ } = [D] [B] {X }e (11.48)

⎡ ⎤
ZA
⎢ ZB −ZC ⎥
⎢ ⎥
⎢ ⎥
⎢ ZD ZE ZF ⎥
⎢ ⎥
⎢ ZG ⎥
⎢ ZH 0 ZA ⎥
⎢ ⎥
⎢ −Z I ZJ 0 Z B ZC ⎥
⎢ ⎥
⎢ −Z D 0 −Z F ⎥
1 ⎢ ⎥
ZD ZE ZF
⎢ ⎥
[D] [B] = ⎢
ab ⎢
⎥
⎥
(11.49)
⎢ −Z K 0 −Z M 0 0 0 ZA ⎥
⎢ ⎥
⎢ ⎥
⎢ −Z L 0 −Z N 0 0 0 Z B −ZC ⎥
⎢ ⎥
⎢ −Z D ZE 0 ZD 0 0 Z D −Z E −Z F ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 0 −Z K 0 −Z M −Z G Z H 0 ZA ⎥
⎢ ⎥
⎣ 0 0 0 −Z L 0 −Z N Z I ZJ 0 Z B ZC ⎦
−Z D 0 0 ZD ZE 0 ZD 0 −Z F −Z D −Z E Z F

The product [D] [B] has been explicitly evaluated and it is given by Eq. (11.49),
where

p= a
b ZG = 6D1 p
ZA = 6Dx p–1 + 6D1 p ZH = 2MD1 a
ZB = 6Dy p + 6D1 p–1 ZI = 6Dy p
ZC = 4Dy a ZJ = 2Dy a
ZD = 2Dxy ZK = 6Dx p–1
ZE = 2bDxy ZL = 6D1 p–1
ZF = 2aDxy ZM = 2Dx b
ZN = 2D1 b

11.3.5 Derivation of the Element Stiffness Matrix

The element stiffness matrix is derived here by applying the principle of minimum
potential energy. The potential energy of the plate element is given by (11.2)

(a ( b ( )
1 ∂ 2w ∂ 2w ∂ 2w
Φ= − Mx 2 − M y 2 + 2Mx y dx dy
2 ∂x ∂y ∂x ∂y
0 0
248 11 Finite Element Analysis of Plate Bending Problems

(a ( b
− { f }T q d x dy (11.50)
0 0

where q is any discrete loading inside the element. Based on the notations used so
far for the rectangular plate bending element, Eq. (11.50) can be written as

(a ( b (a ( b
1
Φ= {ε} {σ } d x d y −
T
w T q d x dy (11.51)
2
0 0 0 0

Combining Eqs. (11.39a, 11.39b), (11.44), (11.48) and (11.51) yields

(a ( b
1
Φ= {X }eT [B]T [D] [B] {X }e d x d y
2
0 0
(a ( b
− {X }eT [N ]T q d x dy (11.52)
0 0

The principle of minimum potential energy requires

( ]
∂Φ
= {0} (11.53)
∂ {X }e

Therefore, performing the partial differentiation of Eq. (11.52) yields

( ] (a ( b ( |( |
∂Φ a | a |
= [B]T [D] [B] d x d y {X }e − | | [N ]T q d x dy = {0}
∂ {X }e | |
0 0
0 0

or,

[k]e {X }e = {P}e (11.54)

where

(a ( b
[k]e = [B]T [D] [B] d x d y (11.55)
0 0

(a ( b
{P}e = [N ]T q d x dy (11.56)
0 0
11.4 Parallelogram Element of Plate Bending 249

[k]e can be explicitly evaluated, but has not been presented here. The complete
matrix [k]e for the skew element is given in the next section. By putting angle φ = 90°
for the skew element, the element stiffness matrix [k]e for the rectangular element
can be obtained (Table 11.1).
If the uniform load is present in the plate element, i.e. q is constant, then the
element loading matrix is given by
⎧ ⎫
⎪
⎪ 1/4 ⎪
⎪
⎪
⎪ − b/24 ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ a/24 ⎪ ⎪
⎪
⎪ ⎪
⎪
⎪
⎪ 1/4 ⎪ ⎪
⎪
⎪ ⎪
⎪
⎪
⎪ b/24 ⎪⎪
⎪
⎪ ⎪
⎪
⎨ a/24 ⎬
{P}e = q ab (11.57)
⎪
⎪ 1/4 ⎪
⎪
⎪
⎪ ⎪
⎪
⎪ −b/24 ⎪⎪
⎪
⎪
⎪ ⎪
⎪
⎪ −a/24 ⎪⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ 1/4 ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ / ⎪
b/24 ⎪
⎩ ⎭
−a 24

11.4 Parallelogram Element of Plate Bending

Many practical structures have intersecting edges, which are not orthogonal; some of
them may have highly irregular boundaries. Out of them, structures like swept wings
of aircrafts or skew bridges, and some others can be idealized as a plate having
the shape of a parallelogram. Rectangular elements will not be able to fit into their
boundary. If the rectangular elements are generalized into the shape of a quadrilateral,
then the criterion of constant curvature cannot be satisfied. Parallelogram elements
may be effectively used for the analysis of these plates.
A parallelogram element with dimensions and the coordinate system has been
shown in Fig. 11.3. There are three degrees of freedom at each node—the unknown
rotations are{ assumed along
} skew coordinate system. As such the displacements at
node 1 are w1 θξ 1 θη1 . The complete displacement and the force vectors for this
element are
{ }
{X }eT = w1 θξ 1 θη1 w2 θξ 2 θη2 w3 θξ 3 θη3 w4 θξ 4 θη4

{ }
{P}eT = P1 Mξ 1 Mη1 P2 Mξ 2 Mη2 P3 Mξ 3 Mη3 P4 Mξ 4 Mη4 (11.58)

In Fig. 11.3, two coordinates systems have been shown

250 11 Finite Element Analysis of Plate Bending Problems

Table 11.1 Element stiffness matrix of the skew element

11.4 Parallelogram Element of Plate Bending 251
( ] [ ] ( ]
x 1 cos φ ξ
= (11.59)
y 0 sin φ η

Inverting Eq. (11.59)

( ] [ ] ( ]
ξ 1 − cot φ x
= (11.60)
η 0 cosec φ y

11.4.1 Displacement Function

The displacement function for the plate is

{ f }T = w = α1 + α2 ξ + α3 η + α4 ξ 2 + α5 ξ η + α6 η2 + α7 ξ 3 + α8 ξ 2 η
+ α9 ξη2 + α10 η3 + α11 ξ3 η + α12 ξη3 (11.61)

The rotations are given by

∂w ( )
θξ = − = − α3 + α5 ξ + 2α6 η + α8 ξ 2 + 2α9 ξ η + 3 α10 η2 + α11 ξ 3 + 3α12 ξ η2
∂η
(11.62)
∂w
θη = = α2 + 2α4 ξ + α5 η + 3α7 ξ 2 + 2α8 ξ η + α9 η2 + 3α11 ξ 2 η + α12 η3
∂ξ
(11.63)

11.4.2 Displacement Function in Terms of Nodal

Displacements

Following exactly the same procedure by which Eq. (11.36) is obtained with the
exception that the coordinates of the nodes are in terms of skew coordinates system,
the following familiar equations can be obtained.

{α} = [A]−1 {X }e (11.64)

{ f } = w = [N ] {X }e (11.65)
252 11 Finite Element Analysis of Plate Bending Problems

11.4.3 Curvature–Nodal Parameter Relationship

The curvatures in cartesian coordinate system are given by Eq. (11.40). As the
displacement function w is a function of x and y, the curvatures are to be expressed
in terms of skew coordinates, that is, a relationship of curvatures between the two
axis systems is to be obtained.
Performing the necessary differentiation in Eq. (11.60), following results are
obtained.
∂ξ ∂ξ ∂η ∂η
= 1, = − cot φ, = 0, = cosec φ (11.66)
∂x ∂y ∂x ∂y

By chain rule,
⎫
∂w ∂w ∂ξ ∂ w ∂ η⎪
= . + . ⎪
⎬
∂x ∂ξ ∂x ∂η ∂x
(11.67)
∂w ∂w ∂ξ ∂ w ∂ η⎪
⎪
= . + . ⎭
∂y ∂ξ ∂y ∂η ∂y

Substituting the values of the differentiated quantities of Eq. (11.66) into

Eq. (11.67), yields
)
∂w
∂x
= ∂∂ wξ
∂w ∂w ∂w (11.68)
∂y
= − cot φ ∂ξ
+ cosec φ ∂η

Equation (11.68) reveals

∂ ∂
=
∂x ∂ξ

Hence,

∂2 ∂2
= (11.69)
∂ x2 ∂ ξ2

Similarly, second equation of Eq. (11.68) yields in terms of the operator

∂ ∂ ∂
= − cot φ + cosec φ (11.70)
∂y ∂ξ ∂η

Hence,

∂2 ∂2 ∂2 ∂2
= cot 2
φ + cosec 2
φ − 2 cot φ cosec φ (11.71)
∂ y2 ∂ ξ2 ∂ η2 ∂ ξ ∂η
11.4 Parallelogram Element of Plate Bending 253

Similarly,

∂2 ∂2 ∂2
= cot φ + cosec φ (11.72)
∂x ∂y ∂ ξ2 ∂ξ ∂η

Relations (11.69), (11.71) and (11.72) when written in matrix form become
⎧ ⎫ ⎡ ⎤ ⎧ ∂2w ⎫
∂2w
⎪
⎨ − ∂2x 2 ⎪
⎬ 1 0 0 ⎨ − ∂ξ2 2 ⎪
⎪ ⎬
− ∂∂ yw2 = ⎣ cot 2 φ cosec2 φ cot φ cosec φ ⎦ − ∂∂ηw2 (11.73)
⎪
⎩ 2 ∂2w ⎪
⎭ ⎩ ∂2w ⎪
⎪ ⎭
∂x ∂y
2 cot φ 0 cosec φ 2 ∂ξ ∂η

Equation (11.73) in compact form is written as

{ε}x y = [H ] {ε}ξ η (11.74)

By differentiation of Eq. (11.65), the curvatures in skew coordinates can be

expressed as

{ε}ξ η = [B] {X }e (11.75)

Combining Eqs. (11.74) and (11.75) yields

{ε}x y = [H ] [B] {X }e (11.76)

11.4.4 Moment–Curvature Relationship

In cartesian system, moment–curvature relationship is given by

{σ }x y = [D] {ε}x y (11.77)

where
{ }
{σ }Tx y = Mx M y Mx y

In Eq. (11.77), substitution of {εxy } from Eq. (11.76) yields

{σ }x y = [D] [H ] [B] {X }e (11.78)

254 11 Finite Element Analysis of Plate Bending Problems

11.4.5 Element Stiffness Matrix

Total potential energy is given by

¨ ¨
1
Φ= {ε}Tx y {σ }x y d x d y − w T q d x dy (11.79)
2

Substituting {ε}xy , {σ}xy and w from Eq. (11.76), (11.78) and (11.65) into
Eq. (11.79) yields
¨ ¨
1
Φ= {X }eT [B] [H ] [D] [H ] [B] d x d y −
T T
[N ]T q d x dy (11.80)
2

Applying the principle of minimum potential energy, Eq. (11.80) reduces to

[k]e {X }e = {P}e (11.81)

where
¨
[k]e = [B]T [H ]T [D] [H ] [B] d x d y (11.82)
¨
{P}e = [N ]T q d x dy (11.83)

[B] matrix of Eq. (11.82) has been expressed in terms of ξ and η. Therefore, the
integration is to be performed in terms of ξ and η.
It has been shown in Eq. (11.25) that

d x d y = |J | dξ dη (11.84)

where
| | | |
| ∂x ∂y | |
| | | 1 0 ||
|J | = | ∂∂ ξx ∂ξ
|=| = sin φ (11.85)
| ∂η ∂y
∂η
| cot φ sin φ |

The element stiffness matrix of Eq. (11.82) then finally becomes

¨
[k]e = sin φ [B]T [H ]T [D] [H ] [B] dξ dη (11.86)

The explicit value of [k]e is given in Table 11.1.

11.5 Hermitian Polynomial Interpolation 255

11.5 Hermitian Polynomial Interpolation

Hermitian polynomials have been used to write down directly the shape functions for
a conforming plate bending element. Hermitian polynomial interpolation is useful
in cases where higher derivatives are introduced as degrees of freedom of the plate.
Slopes are the derivatives of the deflection function. So, when the formulation is
thought of in terms of directly choosing the shape function (Eq. (11.65)), the function
should be such that if it is one, its derivative/derivatives are zero at a particular
node. The choice of such a function can be conveniently done by using Hermitian
polynomials. Mathematically, a Hermitian polynomial is given as Nrmi (x). Here the
polynomial is of order 2 m + 1 and is such that at x = x i

ds N
]
dxs
= 1 when s = r for r = 0 to m
ds N (11.87)
and dxs
= 0 when s /= r when x = x j

This is illustrated with an example.

A flexural line element having deflection and rotation at its two ends is shown in
Fig. 11.4.
The function is given as

f = N01 w1 + N11 θx1 + N02 w2 + N12 θx2 (11.88)

Now, there exist two conditions per node, i.e. in all four conditions are available for
the construction of N 01 , N 11 , etc. N 01 = 1 when associated with w1 and must be zero
when the evaluation is performed for the remaining degrees of freedom. Therefore,
in the present case, a cubic polynomial can be chosen, because the four constants
associated with it can be determined on the basis of four conditions mentioned above.
The polynomial then is

Ni = α1 + α2 x + α3 x 2 + α4 x 3 (11.89)

Let us look into the procedure for determination of N 11 , i.e. the shape function
corresponding to θx1 term. Substituting necessary conditions, the following four
equations result

Fig. 11.4 Rectangular plate w1 w2

bending element
θx1 θx2
x
L 2
256 11 Finite Element Analysis of Plate Bending Problems
⎫
At x = 0, Ni = 0 gives α1 = 0 ⎪
⎪
⎪
⎪
d Ni ⎪
⎪
At x = 0, = 1 gives α2 = 1 ⎪
⎬
dx (11.90)
At x = L , Ni = 0 gives α2 L + α3 L 2 + α4 L 3 = 0 ⎪
⎪
⎪
⎪
⎪
⎪
= 0 gives α2 + 2α3 L + 3α4 L = 0⎪
d Ni
At x = L , 2 ⎭
dx
By the solution of the four equations of Eq. (11.90), results
x
N11 = x(1−x)2 where ξ = (11.91)
L
Other functions can be similarly obtained.
( )
N01 = 1 − 3ξ 2 + 2ξ 3 , N02 = 3ξ 2 − 2ξ 3 , N12 = x ξ 2 + ξ (11.92)

These Hermitian interpolation functions can be extended into two dimensions.

This is illustrated in the conforming plate bending element discussed in the next
section.

11.6 A Conforming Plate Bending Element

Based on the concept of the Hermitian polynomial introduced, it can be verified that
at node i, the shape function is given by

Ni = [N0i (x) N0i (y), N1i (x) N0i (y), N0i (x) N1i (y), N1i (x) N1i (y)] (11.93)

which correspond to w, ∂∂ wx , ∂∂ wy and ∂∂x ∂wy .

2

Introducing Hermitian polynomials, the continuity of slopes along edges is satis-

fied and hence the element analysed on this basis can be termed as conforming
element.
For the element of Fig. 11.2, the degrees of freedom per node are

∂w ∂w ∂ w2
w, θx = , θy = and θx y =
∂x ∂y ∂x ∂y

There are, therefore, 16 degrees of freedom for this element.

The displacement function for this element as chosen is given below.
⎫
w = N01 (x) N01 (y)w1 + N11 (x) N01 (y) θx1 + N01 (x) N11 (y)θ y1 + N11 (x) N11 (y) θx y1 ⎪
⎪
⎪
⎪
+ N02 (x) N02 (y)w2 + N12 (x) N02 (y) θx2 + N02 (x) N12 (y)θ y2 + N12 (x) N12 (y) θx y2 ⎬
⎪
(11.94)
+ N03 (x) N03 (y)w3 + N13 (x) N03 (y) θx3 + N03 (x) N13 (y)θ y3 + N13 (x) N13 (y) θx y3 ⎪
⎪
⎪
⎭
+ N04 (x) N04 (y)w4 + N14 (x) N04 (y) θx4 + N04 (x) N14 (y)θ y4 + N14 (x) N14 (y) θx y4
11.7 Isoparametric Plate Bending Element 257

Once, the displacement function has been defined, the rest of the derivation follows
similar lines to those already presented.

11.7 Isoparametric Plate Bending Element

Isoparametric bending element can be used for the analysis of both thick and thin
plates. So this element presents an advantage over the other plate bending elements
discussed so far—it can take into account the shear deformation of the plate. The
formulation of this element is more elegant than that of other plate flexure elements,
which incorporate shear. This element has been effectively used for the analysis of
sandwich and cellular plates (11.11, 11.12).
In the theory of these plates, it has been assumed that normal to the middle surface
before deformation remains normal after deformation of the plate. When shearing
deformation is considered this assumption does not remain valid and it is to be
modified. The derivation of the element is based on the suggestion by Mindlin that
normals to the middle plane of the plate before bending remain straight, but not
necessarily normal to the middle surface after bending. Therefore, the rotations θx
and θy of the plate shown in Fig. 11.5 are given by
( ] ( )
∂w
θx ∂x
+ φx
= ∂w (11.95)
θy ∂y
+ φy

where φ x and φ y are average shear rotation over the thickness of the plate.
The displacement of the plate may be fully described by
⎧ ⎫
⎨w⎬
{ f } = θx (11.96)
⎩ ⎭
θy

11.7.1 Displacement Function

A general formulation is presented so that the derivation is valid for all types of
isoparametric elements—linear, quadratic, cubic, etc., only the appropriate shape
function is to be substituted.
The coordinates within an element are
( ] Σn [ ] ( ]
x 10 xr
= Nr (11.97)
y 01 yr
r =1
258 11 Finite Element Analysis of Plate Bending Problems

Y Z
θx
∂w
∂x
φx

Actual Deformation
Assumed Deformation
Normal to Midsurface
After Deformation Mid Plane

Fig. 11.5 Deformation of the plate in XZ-plane

where N r is the shape function as discussed in Sect. 10.3, x r , yr are the nodal coor-
dinates and n is the number of nodes. A quadratic element after mapping is shown
in Fig. 11.6.
Similarly, the displacement function is given by
⎧ ⎫ ⎡ ⎤⎧ ⎫
⎨w⎬ Σ n 1 0 0 ⎨ wr ⎬
θ = Nr ⎣ 0 1 0 ⎦ θxr (11.98)
⎩ x⎭ ⎩ ⎭
θy r =1 001 θ yr

or,

Fig. 11.6 A η
non-dimensional mapped
element

ξ
11.7 Isoparametric Plate Bending Element 259

{ f } = [N ] {X }e (11.99)

11.7.2 Strain–Nodal Displacement Relationship

The generalized strains are

( ( ) ]
∂ θx ∂ θy ∂ θx ∂ θy
{ε}T = − − − + − φx − φ y (11.100)
∂x ∂y ∂y ∂x

Therefore, performing the necessary differentiation with respect to the different

quantities given in Eq. (11.98) yields
⎧ ⎫ ⎡ ⎤
⎪
⎪ − ∂∂ θxx ⎪
⎪ 0 − ∂∂Nxr 0
⎪
⎪ ⎪
⎪
⎪
⎪ −
∂ θ ⎪ ⎢ − ∂ Nr ⎥ ⎧w ⎫
)⎪
y
⎨ (∂ y ⎬ Σ n ⎢ 0 0 ∂y ⎥ ⎨ r⎬
∂θ ⎢ ∂N ∂N ⎥
− ∂∂ θyx + − ∂ xy = ⎢ 0 − ∂ yr − ∂ xr ⎥ θ (11.101)
⎪
⎪ ⎪
⎪ ⎢ ⎥ ⎩ xr ⎭
⎪
⎪ ⎪
⎪ r =1 ⎣ ∂ Nr − N 0 ⎦ θ yr
⎪ −φx
⎪ ⎪
⎪ ∂x r
⎩ ⎭ ∂ Nr
0 − Nr
−φ y ∂y

While deriving Eq. (11.101), the following assumption has been made
⎫
∂w
−φx = − θx ⎪
⎪
⎬
∂x
(11.102)
∂w ⎪
−φ y = − θy ⎪
⎭
∂y

Equation (11.101) written in compact form becomes

Σ
n
{ε} = [B]r {X r }e (11.103)
r =1

or,

{ε} = [B] {X }e (11.104)

The shape function derivatives are to be obtained by chain rules as given in

Sect. 10.3.
260 11 Finite Element Analysis of Plate Bending Problems

11.7.3 Stress–Strain Relationship

The stress components to be considered in this case are

{ }
{σ }T = Mx M y Mx y Vx Vy (11.105)

The stress–strain relationship for an orthotropic plate is

⎧ ⎫ ⎡ ⎤ ⎧ ⎫
⎪ Mx ⎪ ⎪
⎪ − ∂∂ θxx ⎪
⎪
⎪
⎪ ⎪
⎪
Dx D1 0 0 0 ⎪
⎪ ∂ θy ⎪
⎪
⎪
⎪ ⎪
⎪ ⎢D 0⎥ ⎪
⎪ − ⎪
⎨ y ⎬ ⎢ 1
M
⎢
Dy 0 0 ⎥
⎥
⎨ ( ∂y )⎪
⎬
∂ θ
Mx y = ⎢ 0 0 Dx y 0 0⎥ − ∂∂ θyx + ∂ xy (11.106)
⎪
⎪ ⎪ ⎢ ⎥ ⎪ ⎪
⎪ Vx ⎪
⎪ ⎪ ⎣ 0
⎪ 0 0 Sx 0⎦ ⎪
⎪
⎪ −φ
⎪
⎪
⎪
⎪
⎩ V ⎭ ⎪ ⎪
⎪ x ⎪
⎪
0 0 0 0 Sy ⎩ ⎭
y −φ y

where S x and S y are shear rigidities. For an isotropic plate they are

Gt
Sx = S y = (11.107)
1.2
In compact form, Eq. (11.106) becomes

{σ } = [D] {ε} (11.108)

11.7.4 Element Stiffness Matrix

The element stiffness matrix defined in the usual way is given by

| |
( 1 |( 1 |
| |
[k]e = || || [B]T [D] [B] [J ] dε dη (11.109)
| |
−1 −1

The integration is to be carried out numerically by employing Gaussian Quadra-

ture. For this isoparametric plate bending element, it has been found that the best
results of the integration can be obtained by using ‘reduced integration technique’.
This is explained below.
11.7 Isoparametric Plate Bending Element 261

11.7.5 Reduced Integration Technique

It has already been stated that Gaussian integration of n points gives exact values for
the integral of polynomials of degree (2n – 1) in x. In order to evaluate the final values
of element stiffness matrix, elements of [B]T [D] [B] are to be studied in details. The
shear strain components are functions of N r and its derivative whereas flexural strain
components are functions of derivatives of N r only. Therefore, for all the cases of
linear, quadratic and other isoparametric elements, shear strain components are one
order higher than flexural strain components, which introduces an inconsistency in
the analysis that is responsible for the poor behaviour of these elements in shear. This
requires more detailed discussion. Isoparametric quadratic element is taken up first.

11.7.5.1 Isoparametric Quadratic Element

For an isoparametric quadratic element, the shape functions are quadratic in ξ and
η and their derivatives are linear. It can be seen in [B] matrix of Eq. (11.101) that
flexural strain terms are derivatives of shape function and hence vary linearly whereas
shear strain terms are functions of both the shape functions and its derivatives and
have a quadratic variation. Thus, shear strain terms have a higher order variation,
which introduces inconsistency. Thus for the exact evaluation of flexural components
in Eq. (11.109), a 2 × 2 Gaussian quadrature rule is to be employed, whereas an exact
evaluation of the shear contribution is to be done by 3 × 3 Gaussian quadrature. The
results of the shear forces obtained by 3 × 3 Gaussian integration for the thin plate
have been found to be erratic. On the other hand, it has been found that a reduced
integration of 2 × 2 (which also is economic) gives vastly superior results.
It is preferable to have all terms for moments and shears having the same bilinear
variation. Figure 11.7 shows a least square straight line fit to a parabola and it has been
shown that they intersect at Gauss point values. Thus if a 2 × 2 Gaussian integration
is performed, the quadratic function is sampled at points such that the result will be
identical to that obtained by use of ‘smoothed’ (linear) shape functions. So both the
functions are reduced to the same order by the least square approximation. Thus,
quadratic function is replaced by a linear function.

Fig. 11.7 A least square Quadratic Function f(ξ)

straight line fitting the
quadratic function
Least Squares Function f(ξ)

Gauss Points

ξ = –1 ξ= –1/√3 ξ=0 ξ = 1/√3 ξ = +1

262 11 Finite Element Analysis of Plate Bending Problems

The argument presented above is no doubt approximate, but it has been found that
the use of 2 × 2 integration has provided good convergence characteristics.

11.7.5.2 Isoparametric Linear Element

For an isoparametric linear element, shape functions are linear in ξ and η and their
derivatives are constants (11.4). [B] matrix in Eq. (11.101] reveals that moment terms
are constant whereas shear terms are linear functions of ξ and η.
The elements of [B]T [D] [B] in the element stiffness matrix can have a maximum
of second order variation and accordingly a 2 × 2 Gauss integration is to be employed
for exact evaluation. A reduced one-point integration is to be performed for removing
the inherent inconsistency.
‘If an nth order function f(ξ) (where—1 ≤ ξ ≤ 1) is sampled at n points of an
n-point Gauss rule, then the n values of the functions f(ξ) at these points uniquely
define a function of order (n – 1) which is a least square fit of the function f(ξ). The
interpretation of the above statement in the case of a linear function is represented in
Fig. 11.8 (11.4). The original trapezium is substituted by a rectangle—the inclined
side of the trapezium and the horizontal top side of the rectangle intersects at the
centre, i.e. at the Gauss point for one-point integration. The linear terms of these are
thus replaced by the their Gauss point values which are zero in the present case, and
so will yield the exact value of the function at that point. This reduced one-point
integration not only economizes the solution, but also improves the analysis to a
great extent.

11.8 Smoothed Stresses

It has been found that the values of the moments acting at the nodes are reasonably
accurate, whereas the nodes are the worst sampling points for shear. As reduced
integration is equivalent to the use of substitute lower order variation it is suggested
that the values of the stresses be determined at Gauss points and then extrapolated
to get the nodal values. These stresses are referred to as smoothed stresses.

Fig. 11.8 Modification of Modified Linear Linear Function f(ξ)

linear function Function

Gauss Point

ξ= –1 ξ=0 ξ = +1
11.9 Triangular Plate Bending Elements 263

Stress can be smoothed in another way (11.3). In reduced integration technique,

the element stiffness matrix has been manipulated whereas [B] matrix remains
unchanged. If [B] matrix is modified by least square smoothing, the variation of
both the stress and strain field within the element is improved. This is done in the
quadratic element by substituting the quadratic terms in [B] matrix by their least
square linear fit, i.e., ξ2 and η2 are replaced by 1/3. In this way, nodes can be taken as
the sampling points and the values will be identical to those obtained from bilinear
extrapolation of Gauss point values.
Similar smoothing can also be done in the linear element. The linear terms of ξ,
η in [B] matrix are replaced by their Gauss point values, i.e. zero in this case. The
spurious shear effect is thus removed. This reduces isoparametric linear element to
a constant moment, constant shear element.

11.9 Triangular Plate Bending Elements

Considerable attempt has been made to develop triangular plate bending elements. In
the initial stages difficulties have been encountered in the formulation. A three-noded
triangle will have all 9 degrees of freedom. To choose satisfactory displacement
function based on polynomial expansion posed problems. A complete quadratic
polynomial will contain 6 terms and a complete cubic will contain 10 terms. To
maintain geometric isotropy and then the choice of 9 terms, led to serious difficulties.
Two suggested displacement functions are given below.

w = α1 + α2 x + α3 y + α4 x 2 + α5 y 2 + α6 x 3 + α7 x 2 y + α8 x y 2 + α9 y 3 (11.110)

and
( )
w = α1 + α2 x + α3 y + α4 x 2 + α5 x y + α6 y 2 + α7 x 3 + α8 x 2 y + x y 2 + α9 y 3
(11.111)

In Eq. (11.110) omitting the ‘xy’ term resulted in the assumed displacement
function incapable of representing constant twist term within the element. This made
the element far too stiff. Though the formulation based on the displacement function
(11.111) gave reasonably good results, yet it was observed that the matrix [A] relating
the generalized displacements to the nodal displacements was becoming singular for
certain orientation of the element.
Attempts to develop a satisfactory element may fall under one of the following
classes.
(a) The element may be defined with 6 degrees of freedom—three vertical displace-
ments at the three corner nodes and three slopes at the midside nodes. The
264 11 Finite Element Analysis of Plate Bending Problems

displacement function can then be represented by a complete quadratic poly-

nomial. Inter-element displacement continuity conditions however could not be
satisfied.
(b) In order to obtain a displacement function, which can be represented by a
complete cubic polynomial of 10 terms, insert an unknown displacement at
an internal node. Then the effect of the internal node is eliminated by static
condensation. In this case also the inter-elemental continuity conditions cannot
be satisfied.
(c) In order to satisfy inter-elemental continuity conditions, a complete polynomial
requires 21 terms so that the total degrees of freedom is 21.
(d) Satisfactory stiffness matrix having 9 degrees of freedom has been achieved
by using area coordinates. Continuous displacement field has been obtained by
superimposition of proper combination of shape functions.
For further details of triangular plate bending elements, Ref. [1] may be consulted.

11.10 DKT Element

An efficient triangular plate bending element has been proposed. It is known as the
discrete Kirchhoff triangle or DKT element [2, 3]. The element is developed on the
basis of imposition of zero transverse shear strains at specific points. The basic steps
for the derivation of the stiffness matrix are given below.
The initial element with degrees of freedom is shown in Fig. 11.9a. It has 6 nodes
and 12 d.o.f and involves the area coordinates in the formulation. The final 9 d.o.f
element is shown in Fig. 11.9b.
The starting point is a straight-sided element with corner and midside nodes as
shown in Fig. 11.9a. The slopes θx and θy at the plate middle surface are expressed
in terms of nodal rotations as follows:

y
n
∂w
s

3
s θy
i
3 wi
() ∂y i

θxi y
α x
∂w
(c) Positive Angle
6 5 y
x () ∂x i

n
x

1 4 2 1 2
(a) Initial Element (b) Final Element

Fig. 11.9 DKT element

11.10 DKT Element 265

Σ
6 Σ
6
θx = Ni θxi , θ y = Ni θ yi (11.112)
i=1 i=1

where Ni are the shape functions expressed in area coordinates [Eq. (11.50)].
A nine d.o.f element is sought that has nodal d.o.f as shown in Fig. 11.9(b).
Accordingly, 12 d.o.f θ xi and θ yi at nodes 1 through 6 must be expressed in terms of
wi , θ xi and θ yi at only corner nodes.
The lateral deflection is assumed to vary cubically in an edge-tangent coordinate
system. The value of s is equal to zero at one node and equal to the length of the side
at the other node.
The lateral deflection w along the edge, say 13, can be expressed as

w = a0 + a1 s + a2 s 2 + a3 s 3 (11.113)

where a0 , …, a4 are constants.

The rotations at any midside node, say 4 are to be obtained. This can be done by
differentiating w given by Eq. (11.10) after substituting [N] from Eq. (11.12) and it
becomes
( ) ( ) ( )
∂w 3 1 ∂w 3 1 ∂w
=− w1 − + w2 − (11.114)
∂s 4 2L 12 4 ∂ s 1 2 L 12 4 ∂s 2

where L 12 , is the length of the edge 12.

The detailed derivation of it is shown later.
Similarly,
( ) ( ) ( )
∂w 3 1 ∂w 3 1 ∂w
=− w2 − + w3 − (11.115)
∂s 5 2L 23 4 2 ∂s
2 L 23 4 ∂s 3
( ) ( ) ( )
∂w 3 1 ∂w 3 1 ∂w
=− w3 − + w1 − (11.116)
∂s 6 2L 13 4 ∂ s 3 2 L 13 4 ∂s 1

The slopes ∂∂ wy in Eqs. (11.114)–(11.116) can be expressed in terms of ∂w

∂x
and ∂w
∂y
by transforming the coordinates.

11.10.1 Constraint Equations

The following constraint equations are used.

1. Transverse shear strains γ yz and γ zx vanish at corners 1, 2 and 3. Thus,
266 11 Finite Element Analysis of Plate Bending Problems
( ) ( )
∂w ∂w
θxi = , θ yi = for i = 1, 2, 3 (11.117)
∂x i ∂x i

2. Transverse shear strain γsz vanishes at nodes 4, 5 and 6.

Thus
( )
∂w
θsi = , for i = 4, 5, 6 (11.118)
∂s i

(∂ w) (∂ w) ( )
∂w
∂s i
is expressed in terms of ∂x i
and ∂y i
by coordinate transformation.

3. Normal slopes are assumed to vary linearly along the edge. Thus

[( ) ( ) ]⎫
1 ∂w ∂w
θn4 = + ,⎪⎪
⎪
2 ∂n 1 ∂n 2 ⎪ ⎪
[( ) ( ) ]⎪ ⎪
⎬
1 ∂w ∂w
θn5 = + , (11.119)
2 ∂n 2 ∂n 3 ⎪
[( ) ( ) ]⎪ ⎪
⎪
⎪
1 ∂w ∂w ⎪
θn6 = + ,⎪⎭
2 ∂n 3 ∂n 1

Applying the constraint equations mentioned above, 12 d.o.f of the initial element
an expressed in terms of 9 d.o.f at the corner nodes. Mathematically,
( ( ) ( ) ( ) ]T
{ }T ∂w ∂w ∂w
θx1 θ y1 . . . θ y6 = [T ] w1 ... (11.120)
∂x 1 ∂x 1 ∂y 3

where [t] is the transformation matrix of size 12 × 9.

11.10.2 Transformation Matrix

α is the angle between the edge-tangent axis s and Cartesian coordinate axis x as
shown in Fig. 11.9c.
Applying chain rule
⎫
∂w ∂w∂x ∂w ∂y ⎪
= + ⎪
∂s ∂x ∂s ∂y ∂s ⎬
(11.121)
∂w ∂w ∂x ∂ w ∂ y⎪
⎪
= + ⎭
∂n ∂x ∂n ∂y ∂n
or,
11.10 DKT Element 267
)
w,s = cos α w,x + sin α w,y
(11.122)
w,s = sin α w,x + cos α w,y

∂w ∂w
where w,s = ∂s
, w,n = ∂n
etc.
Similarly,
)
w,x = cos α w,s + sin α w,n
(11.123)
w,y = sin α w,s + cos α w,n

Let us proceed to express θx , θy at the midside nodes in terms of the d.o.f of the
corner nodes. Midside node is 4 along the edge 12. For this node 4, applying the
second constraint

θs4 = w,s4 (11.124)

Applying the third constraint

θn4 = 0.5 (w,nl +w,n2 ) (11.125)

From Eq. (11.122), we can write

( ] [ ] ( ]
w,s 1 cos α sin α w,x 1
= (11.126)
w,n 1 − sin α cos α w, y 1

and
( ] [ ] ( ]
w,s 2 cos α sin α w,x 2
= (11.127)
w,n 2 − sin α cos α w, y 2

From Eq. (11.119) and making proper substitutions from Eqs. (11.126) and
(11.127), we get

1[ ( ) ]
θn4 = (w,x1 +w,x2 ) (− sin α) + w, y1 +w, y2 (cos α) (11.128)
2
In order to determine θs4 , a0 , …, a4 in Eq. (11.113) are to be evaluated.
⎫
w ||s=0 = w1 ⎪
⎪
⎬
w |s=L 12 = w2
∂w (11.129)
|
∂ s |s=0
= w,s1 ⎪ ⎪
∂w | ⎭
∂s s=L 12
= w,s2

Substituting the 4 conditions of Eq. (11.129) for the edge 12 into Eq. (11.113)
yields
268 11 Finite Element Analysis of Plate Bending Problems
⎫
a0 = w1 ⎪
⎪
⎪
⎪
a0 + a1 L 12 + a2 L 12 + a3 L 12 = w2 ⎬
2 3
(11.130)
a1 + 2a2 .0 + 3a3 .0 = w, s1 ⎪
⎪
⎪
⎪
⎭
a1 + 2a2 L 12 + 3a3 L 12 = w, s2
2

Solving 4 equations given by Eq. (11.130) yields

⎫
0 = w1 ⎪
⎪
⎪
⎬
a1 = w,sl
a2 = L32 (w2 − w1 ) − w,s2 +2w,s1 (11.131a)
⎪
⎪
L 12
⎪
(w2 − w1 ) ⎭
12
w,sl +w,s2
a3 = L 212
− 2
L 212

Substituting the constants from Eq. (11.131a) into Eq. (11.113) yields

3 1 3 1
w,s4 = − w1 − w,s1 + w2 − w,s2 (11.131b)
2L 12 4 2L 12 4

Substituting w,s1 and w,s2 from Eq. (11.122) into Eq. (11.131a) yields

3 1( )
θs4 = w,s4 = − w1 − w,x1 cos α + w, y1 sin α
2L 12 4
3 1( )
+ w2 − w,x2 cos α + w, y2 sin α (11.132)
2L 12 4

Now,

θx4 = w,x4 cos α − w,n4 sin α

[ ]
3 1( )
= cos α − w1 − w,x1 cos α + w, y1 sin α
2L 12 4
3 1 ( )
+ w1 − w,x2 cos α + w, y2 sin α
2L 12 4
[ ]
1 1( )
− (w,x1 +w,x2 ) (− sin α) + w, y1 +w, y2 (cos α) sin α (11.133)
2 2

Equation (11.133) has been obtained after combining Eqs. (11.123), (11.119) and
(11.125).
Rearranging the terms of Eq. (11.133) yields
( ) ( )
3 1 3
θx4 = − cos α w1 + − cos α w,x1
2
2L 12 2 4
( ) ( )
3 3
+ − sin α cos α w, y1 + cos α w2
4 2L 12
11.10 DKT Element 269
( ) ( )
1 3 3
+ − cos2 α w,x2 + − sin α cos α w, y2 (11.134)
2 4 4

Similarly,
( ) ( )
3 3
θx4 = − sin α w1 + − cos α sin α w,x1
2L 12 4
( ) ( ) ( )
1 3 3 3
+ − sin α w, y1 +
2
sin α w2 + − cos α sin α w2
2 4 2L 12 4
( ) ( )
3 1 3
+ − cos α sin α w,x2 + − sin2 α w, y2 (11.135)
4 2 4

Following the procedure mentioned above, θ x5 , θ y5 , θ x6 , and θ y6 can be expressed

in terms of d.o.f. of the corner nodes.
The transformation matrix is given by Eq. (11.136)
⎡ ⎤
0 1 0 0 0 0 0 0 0
⎢ ⎥
⎢ 0 0 1 0 0 0 0 0 0 ⎥
⎢ ⎥
⎢ 0 0 0 0 1 0 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 0 0 0 1 0 0 0 ⎥
⎢ ⎥
⎢ 0 0 0 0 0 0 0 1 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 0 0 0 0 0 0 1 ⎥
⎢ 2 2 ⎥
⎢ − 3C 1 − 3C − 3SC 3C 1 − 3C − 3SC ⎥
[T ] = ⎢
⎢ 2L 12 2 4 4 2L 12 2 4 4 0 0 0 ⎥
⎥
⎢ 2 2 ⎥
⎢ − 2L3S 3C
− 4 S 1
2 − 4
3S 3S 3SC
− 4 1
2 − 4
3S 0 0 0 ⎥
⎢ 12 2L 12 ⎥
⎢ 2 2 ⎥
⎢ 0 0 0 3C
− 2L 1 − 3C − 3SC 3C 1 − 3C − 3SC ⎥
⎢ 23 2 4 4 2L 23 2 4 4 ⎥
⎢ 2 2 ⎥
⎢ 0 0 0 − 2L3S − 3SC 1 − 3S 3S − 3SC 1 − 3S ⎥
⎢ 4 2 4 2L 23 4 2 4 ⎥
⎢ 23 ⎥
⎢ 3C 1 3C 2
− 3SC 3C 1 − 3C 2
− 3SC ⎥
⎢ 2L
⎣ 31 2 − 4 4 0 0 0 − 2L
31 2 4 4 ⎥
⎦
3S 3SC 1 3S 2 3S 3SC 1 3S 2
2L 31 − 4 2 − 4 0 0 0 2L − 4 2 − 4
31
C = cos α ; S = sin α (11.136)

11.10.3 Element Stiffness Matrix

Equation (11.120) is written in more compact form as

{X θ } = [T ] {X }e (11.137)

Since θx and θy vary quadratically, these can be related to nodal degrees of freedom
as
( ] [ ]
θx N1 0 N2 0 · · · N6 0
= {θx1 · · · · · · θx6 }T
θy 0 N1 0 N2 · · · 0 N6
270 11 Finite Element Analysis of Plate Bending Problems

= [N ] {X θ } (11.138)

where N 1 , N 2 , etc. are defined in Eq. (10.50) and the relation between area coordinate
and the Cartesian coordinate is given by Eq. (10.43). The strain vector is given by
[ ( )]
{ }T ∂θx ∂θ y ∂θ y ∂θx
εx ε y γx y = −z −z −z +
∂x ∂y ∂y ∂y
or
( ]
θx
{ε} = −z[∂] (11.139)
θy

where
⎡ ⎤
∂
∂x
0
⎢ ∂ ⎥
[∂] = ⎣ 0 ∂y ⎦
∂ ∂
∂y ∂x

Combining Eqs. (11.137), (11.138) and (11.139) yields

{ε} = −z[Bθ ] [T ] {X }e (11.140)

where
⎡ ∂N ∂ N2
⎤
∂x
1
0 ∂x
0 · · · ∂∂Nx6 0
⎢ ∂ N1
0 ∂∂Ny2 · · · 0 ∂∂Ny6 ⎥
[Bθ ] = ⎣ 0 ∂y ⎦ (11.141)
∂ N1 ∂ N1 ∂ N2 ∂ N2 ∂ N6 ∂ N6
∂y ∂x ∂y ∂x
· · · ∂y ∂x

The stress–strain relationship for an isotropic plate is given by

⎧ ⎫ ⎡ ⎤⎧ ⎫
⎨ σx ⎬ E ⎣
1ν 0 ⎨ εx ⎬
σ = ν 1 0 ⎦ εy (11.142)
⎩ y ⎭ 1 − ν2 ⎩ ⎭
σx y 0 v 1−ν
2
γx y

or

{σ } = [D] {ε} (11.143)

From Eq. (11.140), increment in strain is given by

∂ { ε } = −z[Bθ ] [T ] ∂ {X }e (11.144)

Strain energy stored in the element becomes

11.11 The Patch Test 271
( ( (
U= ∂ {ε}T {σ } dv (11.145)

Combining Eqs. (11.145), (11.143) and (11.144) yields

( ( (
U= (−z) ∂ {X }eT [T ]T [Bθ ]T [D] (−z) [Bθ ] [T ] {X }e d x d y dz

or
( (
U = δ {X }eT [T ]T [Bθ ]T [D] [Bθ ] [T ] d x d y (11.146)

where
⎡ ⎤
1ν 0
Et 3
[D] = ( ) ⎣ν 1 0 ⎦
12 1 − ν 2
0 0 1−ν
2

It can be shown that the element stiffness matrix is given by

( (
[K]e = [T ]T [Bθ ] [D] [Bθ ] [T ] d x d y (11.147)

11.11 The Patch Test

It has been shown in Sect. 11.3 that the four-noded rectangular plate bending element
is a non-conforming element. Despite its non-conformity, this element is a useful one
and results obtained by using this element do converge to the correct solution. There
are many other non-conforming elements that have been successfully used in the
stress analysis. Non-conformity should not be treated as a numerical disadvantage;
the presence of kinks at the inter-elemental boundaries of the structure makes it more
flexible and thus tends to compensate for whatever over-stiff representation is made
within the individual element (Fig. 11.10).
Are all non-conforming elements acceptable? What is its test of acceptability?
The basic requirement of the element is that it should converge, that is, the results
tend towards the correct solution with the refinement of the mesh. The patch test has
been devised to answer these queries [3]. It is a simple test that can be performed
numerically to check the validity of an element formulation and its programmed
implementation.
In the test, one assembles a small number of elements into a patch which should
have at least one nodal point completely surrounded by elements. The test is now
performed to see whether a patch of elements subjected to a state of constant strain can
272 11 Finite Element Analysis of Plate Bending Problems

(σ)constant

i
x
1 εx dx

Fig. 11.10 Patch test for plane stress element

reproduce the exact constitutive behaviour of the material and will result in correct
stresses when it becomes infinitesimally small. Though there is a requirement that
the patch size be infinitesimal, but for elements where polynomial displacement
functions are used, the size of the element may be assumed to possess any value
[4, 5].
A rigid body displacement of the patch will cause no strain, and if proper stress–
strain relationships are used, no changes in stress will result.
A physical problem is based on differential equations. For many such problems,
exact solution exists. The same problem when attempted by the finite element method
results in a set of algebraic equations of the following form

[K ] {X } = {P} (11.148)

which when solved given an approximation to the differential equation and its
boundary conditions.
The test can be carried out in one of the two ways. A patch of elements is made
and a node i inside the patch is selected which is connected to the elements. It is to
be remembered that inside nodes should not be fixed. The differential equation of
the arbitrary system is solved and the parameters X at all nodes of a patch, which
assembles completely the nodal variable, X i are determined.
In the first test, the exact values of the parameters X substituted into the ith equation
and then the following equation is verified

K i j X j −Pi = 0 (11.149)

In the other test, only the values of X from the exact solution corresponding to
the boundaries of the patch are inserted and X i is then formed as
( )
X i = K ii−1 Pi − K i j X j j /= i (11.150)
11.11 The Patch Test 273

and then compare with the value.

11.11.1 The Patch Test for the Plane Stress Element

A small number of elements are assembled into a patch. At least one node is placed
within the patch (Fig. 11.11), built of four-noded elements. Nodes at the boundary are
loaded by consistent loads appropriate to a state of constant stress. Supports are to be
placed in such a manner that rigid body motions are to be prevented. In this example,
if instead of roller on the left edges, it is considered to be clamped, then it would have
been fatal for the patch test, as the effect of Poisson’s ratio in the y-direction would
have been prevented. By applying the theory of elasticity approach, the stresses at
the nodes are to be computed. Then the patch is to be studied by one of the above
approaches (either Eq. (11.149) or Eq. (11.150)). If the agreement is good, then the
patch is passed. The patch test is then to be repeated for constant σy and constant
τxy , as the test must be repeated for all other constant stress states demanded by the
element.

11.11.2 The Patch Test for Plate Bending Elements

In plate bending problems, the terms corresponding to ‘strains’ are curvatures. There-
fore constant strain terms in patch test are of constant curvatures. For performing
the patch test, the patch is to be subjected either a set of prescribed displacements or
boundary loads that correspond to a set of constant curvature.

y
Po Po
2 2 Po unit/length
1

Po Po

1
Po Po x
2 2
1 1

Fig. 11.11 Patch test for plane stress

274 11 Finite Element Analysis of Plate Bending Problems

11.11.2.1 Prescribed Displacement Approach

Suggestions have been made as to the form of deflection to be assessed. They are as
follows:
( )
C x 2 + x y + y2
w= (11.151)
2
∂w ( y)
=C x+ (11.152)
∂x 2
∂w (x )
=C +y (11.153)
∂y 2

∂ 2w C
= (11.154)
∂ x∂ y 2

where C is assumed to be the value of the deflection that is much smaller than the
thickness, one of the above two approaches can be adopted. For the case of plate
bending with prescribed displacements, they are restated as follows:
1. The values of w, ∂w/∂x and ∂w/∂y at the boundary nodes 1, 2, 3 and 4 as prescribed
(Fig. 11.12). Based on these values, the assembled stiffness equations corre-
sponding to the degrees of freedom at the inner nodes, 5, 6, 7 and 8 are obtained.
Then determine the curvature values of each element and check whether they are
constants.
2. The values of w, ∂w/∂x and ∂w/∂y at all eight nodes are prescribed. These values
impose a state of constant curvature in each element. Then compute total lateral
forces and bending moments at each of the four inner nodes using the assembled
stiffness equations. The patch test is passed if the lateral forces and moments
at each inner node obtained as a summation of contributions from all elements
surrounding the node vanish.

Fig. 11.12 Patch test for y

plate bending
1 2

5 6

8 7

4 3
11.11 The Patch Test 275

z, w
y z, w y
w=0
P

Mo
Mo P
wy =0
x
x
w=0
wx =0 P

P
w=0

(a) Constant curvature (b) Constant twist

Fig. 11.13 Patch test for plate bending

11.11.2.2 Prescribed Boundary Load Approach

The boundary loads to be applied should induce constant curvature. The appropriate
loading conditions for these cases are shown in Fig. 11.13. For both y constant
curvature and constant twist tests, only a quadrant of a plate need to be analysed
taking advantage of the symmetry and antisymmetry conditions.
When the constant curvature test is to be performed, a constant moment M 0 is
applied in the x-direction while M y and M xy are kept zero. Then using moment–
curvature relationship given by Eq. (7.27) yields

∂ 2w ∂ 2w 12ν M0 ∂ 2w
= −ν = − and =0 (11.155)
∂ y2 ∂x2 Eh 3 ∂ x∂ y

Similarly, the constant curvature test can be performed in the y-direction.

For the constant twist test, the centre and the four midside points of the deflected
plate lie in the same plane with no deflection. Attention is focused on a quadrant.
Any of the quadrants is balanced by four corner loads in the same manner as for
the whole plate based on equilibrium. The centre and four midsides of the quadrant
again lie in the same plane, but not parallel to xy-plane. On the further division of
this quadrant into four subquadrants reveals that each subquadrant is under same
type of corner load. Thus, it can be divided into an infinite number of quadrants.
Taking an infinitesimal quadrant, each of the four corner loads is split into two
halves to produce two twisting moments P2 d x and P2 dy. Thus the plate is subjected
to constant twisting moment (Pdx/2)/dx and (Pdy/2)/dy at all places and remains
anticlastic, that is after deflection. Thus

∂ 2w ∂ 2w
for Mx = M y = 0, = =0 (11.156a)
∂x2 ∂ y2
276 11 Finite Element Analysis of Plate Bending Problems

P ∂ 2w 6P (1 + v)
for Mx y = −M yz = , = (11.156b)
2 ∂ x∂ y Eh 3

The patch test can now be performed in one of the two ways mentioned earlier.
The outside boundary of the patch is chosen as rectangular so as to facilitate the
enforcement of constant curvature of the plate. For quadrilateral types of elements,
the mesh sizes should be as irregular as possible.

11.12 Horizontally Curved Isoparametric Beam

The horizontally curved isoparametric beam formulation given below follows from
Ref. [6].

11.12.1 Displacement Function in Terms of Nodal

Parameters

A one-dimensional parabolic isoparametric element with three nodes is chosen

(Fig. 11.14). Shape functions in the three nodes in terms of the natural coordinate ξ
are defined by
]
Ni = 21 ξ0 (1 + ξ0 ) for i = 1 or 3 and ξ0 = ξ ξi
(11.157)
Ni = 1 − ξ 2

Thus at any point on the beam axis

y
3

2
1 ξ = –1 ξ=0 ξ=1
ξ
x
(a) Physical coordinate (b) Isoparametric coordinate

Fig. 11.14 Horizontally curved beam and its isoparametric mapping

11.12 Horizontally Curved Isoparametric Beam 277

Fig. 11.15 Positive s

y
directions of model rotations

θsi

αi

θti
x

⎫
Σ
3
x= Ni x i ⎪
⎪
⎬
i=1
(11.158)
Σ3 ⎪
y= Ni yi ⎪
⎭
i=1

Nodal displacements at the node i are taken as

⎧ ⎫
⎨ wi ⎬
{X }i = θsi (11.159)
⎩ ⎭
θti

where wi the displacement of the node in the vertical z-direction, θ si is the total
rotation of the cross section due to bending and shear strains, and θ ti is the angle of
twist at the node (see Fig. 11.15).
The following relationships are well known in the isoparametric formulation.

dw
θs = +φ (11.160)
ds
where φ is the rotation due to transverse shear.
⎫
Σ
3
w= Ni wi ⎪⎪
⎪
⎪
i=1 ⎪
⎪
Σ
3 ⎬
θs = Ni θsi (11.161)
⎪
⎪
i=1 ⎪
⎪
Σ3 ⎪
θt = Ni θti ⎪
⎭
i=1

The displacement matrix at any point is given by

278 11 Finite Element Analysis of Plate Bending Problems
⎧ ⎫ ⎧ ⎫
⎨w⎬ ⎨ {x}1 ⎬
f = θs = [N ] {X }e = {[n 1 ] [n 2 ] [n 3 ]} {x}2 (11.162)
⎩ ⎭ ⎩ ⎭
θt {x}3

where
⎡ ⎤
N1 0
[n 1 ] = ⎣ N1 ⎦
0 N1

11.12.2 Stress–Strain Relations

For a beam horizontally curved in plan, the bending moment and the torque are given
by
( )
dθt θs
T = GJ − (11.163)
ds R
( )
dθs θt
M = −E I + (11.164)
ds R

where R is the radius of curvature of the beam and EI and GJ are the flexural rigidity
and torsional rigidity respectively. Stress-resultants for the horizontally curved beam
are related as follows:
⎧ ⎫ ⎡ ⎤ ⎧ ( dθs )
θ ⎫
⎨M⎬ EI 0 0 ⎨ ds − Rt ⎬
V = ⎣ 0 G As 0 ⎦ dw
− θs (11.165)
⎩ ⎭ ⎩ dθ ds
θs ⎭
T 0 0 GJ ds
t
− R

where As is the shear area of the plate

or,

{σ } = [D] {ε} (11.166)

11.12 Horizontally Curved Isoparametric Beam 279

11.12.3 Strain–Displacement Relationship

From Eq. (11.148), the strain components are given by

⎧ ( dθs θ
)⎫
⎨ − ds + Rt ⎬
{ε} = dw
− θs (11.167)
⎩ dθ
ds ⎭
ds
t
− θRs

Substituting relations given by Eq. (11.161) into Eq. (11.167) yields

{ε} = [B] {X }e (11.168)

where
⎡ ⎤
Σ
3 Σ
3 0 − ddsNi − NRi
[B] = [B]i = ⎣ d Ni −Ni 0 ⎦ (11.169)
ds
i=1 i=1 0 − NRi ddsNi

Equation (11.169) contains a term

d Ni d Ni dξ 1
= . = Ni' (11.170)
ds dξ ds J

where
/[
Σ( )2 Σ( )2 ]
J= Ni' xi + Ni' yi (11.171)

The curvature in cartesian coordinate is given by

d2 y
1 dx2
=[ ( )2 ]1/2 (11.172)
R
1 + ddyx

However,
Σ3
dy Ni' yi
= Σi=1
3 '
(11.173)
dx i=1 Ni x i

Also
( ) Σ ' Σ '' Σ '' Σ '
d2 y d dy dξ Ni x i Ni yi − Ni x i Ni yi
2
= . = (Σ '' )3 (11.174)
dx dξ dx dx Ni x i
280 11 Finite Element Analysis of Plate Bending Problems

The curvature in Eq. (11.172) can be obtained from Eqs. (11.173) and (11.174).
Combining Eqs. (11.169) and (11.170)
⎡ N'
⎤
0 − Ji − NRi
⎢ ' ⎥
[B]i = ⎣ NJi −Ni 0 ⎦ (11.175)
'
N
0 − NRi Ji

11.12.4 Element Stiffness Matrix

The element stiffness matrix of the order 9 × 9 is given by

(
[k]e = [B]T [D] [B] ds (11.176)

The rotations tangential and normal to the curve have been considered as
unknowns in the local coordinate system. They are required to be transformed in
the global degrees of freedom through the transformation matrix.

11.13 Nonuniform Straight Beam Element

In many practical structures, beams with tapered cross sections are provided. It
is rather common to divide the tapered beam for the analysis into a number of
elements and consider each element as uniform. This stepped beam approach may
not yield accurate results unless the beam is divided into a sufficiently large number
of elements. In the following, the stiffness matrix of the beam element is formulated
by considering the variation of the depth within the element.
In Fig. 11.16, a nonuniform beam has been shown. I is the second moment of
area at node 1, the length of the element is L and the modulus of elasticity is E. The
second moment of area at any distance x from 1 is given by
[ ( x )α ]
I (x) = I0 1 + r (11.177)
L
where r and α are parameters associated with the variation of the geometry along the
length. For example, for the tapered beam element of Fig. 11.16b,

r = (m − 1) and α = 1

The strain energy for the nonuniform beam element is

11.13 Nonuniform Straight Beam Element 281

x α
I(x) = I0[1+r( ) ]
L
z1,w1 z2,w2

M1 , θ 1 x Io mIo
1 M2 , θ 2
2
L

(a) Non-uniform beam element (b) Non-uniform beam

Fig. 11.16 Nonuniform beam

(L ( )
1 ∂ 2w
U= E I (x) dx (11.178)
2 ∂x2
0

The displacement function for a nonuniform beam remains same as that for the
uniform beam (see Eq. (11.10)).
The element stiffness matrix for the nonuniform beam is given by Yang [7].
⎡ 12
⎤
L2
C11
⎢ ⎥
⎢ ⎥
⎢ 6C ⎥
⎢
E I0 ⎢ L 21 4 C 22 ⎥
⎥
[k]e = ⎢ ⎥ (11.179)
L ⎢ 12 ⎥
⎢ − L 2 C11 − L C21 L 2 C11
6 12 ⎥
⎢ ⎥
⎣ ⎦
6
C
L 41
2 C42 − L C41 4 C44
6

where
( )
1 4 4
C11 = 1 + 3r − +
α+1 α+2 α+3
( )
2 7 6
C21 = 1 + 2r − +
α+1 α+2 α+3
( )
4 12 9
C22 = 1+r − +
α+1 α+2 α+3
( )
1 5 6
C41 = 1 + 2r − +
α+1 α+2 α+3
( )
2 9 9
C42 = 1 + 2r − +
α+1 α+2 α+3
( )
2 6 9
C44 = 1+r − +
α+1 α+2 α+3
282 11 Finite Element Analysis of Plate Bending Problems

11.14 Computer Program for Isoparametric Quadratic

Bending Element

The computer programs in C and FORTRAN for the solution of plate bending prob-
lems using isoparametric plate bending element have been given in the attached CD.
The given computer programs consider two load cases—uniform load throughout
the plate and concentrated load at mid-span. It has been assumed in the computer
program that all elements have identical dimensions so that the element stiffness
matrix and the element loading matrix have been computed for one typical element
and then assembled for all the elements to form the overall stiffness and overall
loading matrices.
Gaussian numerical integration has been done for the evaluation of the element
stiffness matrix and the element loading matrix for the uniform load. The overall
stiffness matrix is stored in half band form and simultaneous equations are solved
with the help of subroutine GAUSS. For calculating STIFF, 2-point integration is to
be performed.
For obtaining shear forces at the nodes, smoothed stresses have been used, i.e. ξ2
and η2 values in SFD, SFDE, SFDZ, etc. are reduced to 1/3. This helps in directly
obtaining the nodal shear forces (11.3) and interpolation of the nodal values from
those at the GAUSS points may be avoided. Another elegant feature of the computer
program is that the mesh generation of the rectangular plate is automatically done.
The boundary conditions specified in the computer program are also general. It
can take up 3 support conditions—simple, fixed, and free as also the symmetry of the
plate centre lines. NBON1, NBON2, NBON3, and NBON4 refer to the four edges
in the order 1, 2, 3 and 4 specified in Fig. 11.17. Edge numbers 2 and 3 can have
symmetry depending on the plate and the loading. They will have values 1 if free, 2
if simply supported and 3 if fixed depending upon the edge conditions.
The items of the computer output which will be of interest to the analyst are the
nodal displacements and stress resultants at nodes of interest.
A square plate having all edges simply supported has been analysed on the basis
2 S b2
of a 4 × 4 mesh division. In order to treat it as a thin plate SxDa = yD = 35,000

Fig. 11.17 Edge coding A

1
x

4
B
2

3
y
11.14 Computer Program for Isoparametric Quadratic Bending Element 283

has been assumed. The input data of the plate is given in Table 11.2. The computer
output of the plate is shown in Table 11.3.
Exercise 11

11.1 Based on Table 11.1, calculate central deflection for a clamped square plate
subjected to a concentrated load at the centre. Divide the plate into four
quadrants and do not take advantage of the symmetry.
11.2 Derive the stiffness matrix for the 4-noded quadrilateral plate bending element.
11.3 Obtain a quadrilateral element from four complete triangular elements. State
the procedure for deriving the stiffness matrix of the four-noded quadrilateral
from the assembly of four triangular elements.

Prob. 11.3

11.4 To incorporate the transverse shear effect, five degrees of freedom are assumed,
they are, w, θ x , θ y , γ x and γ y . Given θx = ∂w
∂x
+γx and θ y = ∂w
∂y
+γ y . Consider a
four-noded rectangular element having twenty degrees of freedom, and derive
the element stiffness matrix starting with a displacement function of 20 terms.
11.5 Derive the element stiffness for an isoparametric beam element.
11.6 Extend the derivation of Prob. 11.5 to incorporate the twisting effect of the
bar.
11.7
(a) What would be the effect on the results if two-point integration is
performed on an isoparametric linear plate bending element?

Table 11.2 Input data for plate bending problem

1.0 1.0 90.0 1.0 1.0
4 4 8 3 195 51 2 2 2
2 2 2 2 0 0
1.0 1.0 0.3 0.35 35000.0 35000.0 0.3
6 1.0 -1.0
7 -1.0 -1.0
9 1.0 1.0
10 -1.0 1.0
0 0 0
284 11 Finite Element Analysis of Plate Bending Problems

Table 11.3 Computer output of the plate bending problem

A B PHAI QL QLC
1.00000 1.00000 90.0000 1.00000 1.00000
NX NY NN NFREE NT NBH NQ NQL NELM NENP NLC
4 4 8 3 195 51 2 2 16 24 2
NBON1 NBON2 NBON3 NBON4 NSYMX NSYMY
2 2 2 2 0 0
DX DY D1 DXY SX SY
1.00000 1.00000 .300000 .350000 35000.0 35000.0
NODE NUMBERS
THE ELEMENT NO.1
1 2 3 11 17 16 15 10
THE ELEMENT NO.2
3 4 5 12 19 18 17 11
THE ELEMENT NO.3
5 6 7 13 21 20 19 12
THE ELEMENT NO.4
7 8 9 14 23 22 21 13
THE ELEMENT NO.5
15 16 17 25 31 30 29 24
THE ELEMENT NO.6
17 18 19 26 33 32 31 25
THE ELEMENT NO.7
19 20 21 27 35 34 33 26
THE ELEMENT NO.8
21 22 23 28 37 36 35 27
THE ELEMENT NO.9
29 30 31 39 45 44 43 38
THE ELEMENT NO.10
31 32 33 40 47 46 45 39
THE ELEMENT NO.11
33 34 35 41 49 48 47 40
THE ELEMENT NO.12
35 36 37 42 51 50 49 41
THE ELEMENT NO.13
43 44 45 53 59 58 57 52
THE ELEMENT NO.14
45 46 47 54 61 60 59 53
THE ELEMENT NO.15
47 48 49 55 63 62 61 54
THE ELEMENT NO.16
49 50 51 56 65 64 63 55
THE OVERALL STIFFNESS MATRIX
THE PLOAD MATRIX
THE DISPLACEMENT MATRIX
-.629004E-20 .178275E-19 -.221340E-19 .337624E-19 -.261318E-19 .337624E-19
-.221340E-19 .178275E-19 -.629004E-20 .125736E-19 .115841E-02 .159925E-02
.115841E-02 .125736E-19 -.143785E-19 .116400E-02 .210058E-02 .270585E-02
.290003E-02 .270585E-02 .210058E-02 .116400E-02 -.143785E-19 .281087E-19
.270839E-02 .374553E-02 .270839E-02 .281087E-19 -.240665E-19 .159967E-02
.290129E-02 .374527E-02 .401899E-02 .374527E-02 .290129E-02 .159967E-02
-.240665E-19 .328166E-19 .270549E-02 .374454E-02 .270549E-02 .328166E-19
-.214072E-19 .115510E-02 .209631E-02 .270514E-02 .289913E-02 .270514E-02
.209631E-02 .115510E-02 -.214072E-19 .178260E-19 .115561E-02 .159838E-02
.115561E-02 .178260E-19 -.819565E-20 .181994E-19 -.222819E-19 .340273E-19
-.266436E-19 .340273E-19 -.222819E-19 .181994E-19 -.819565E-20 .398811E-21
-.167851E-20 .191891E-20 .170956E-19 -.128952E-19 .170956E-19 .191889E-20
-.167849E-20 .398811E-21 .828165E-20 .249098E-02 .367225E-02 .249098E-02
.828168E-20 -.127838E-19 .248039E-02 .473497E-02 .639662E-02 .697028E-02
.639662E-02 .473497E-02 .248039E-02 -.127838E-19 .278137E-19 .639180E-02
.993486E-02 .639180E-02 .278138E-19 -.168106E-19 .367146E-02 .696790E-02
.993535E-02 .113114E-01 .993535E-02 .696790E-02 .367146E-02 -.168106E-19
.188887E-19 .639731E-02 .993674E-02 .639731E-02 .188886E-19 .541049E-21
.249726E-02 .474306E-02 .639797E-02 .697199E-02 .639797E-02 .474306E-02
.249726E-02 .541054E-21 -.167564E-20 .249629E-02 .367390E-02 .249629E-02

(continued)
References and Suggested Readings 285

Table 11.3 (continued)

-.167567E-20 .401141E-20 -.238363E-20 .219916E-20 .165933E-19 -.119250E-19
.165933E-19 .219918E-20 -.238364E-20 .401141E-20
CENTRAL DEFLECTION
4.018994436947302E-003 1.131136125263715E-002
THE VALUES OF MOMENTS AND SHEARS
ZI= 1.00000000000000 ETA= -1.00000000000000 ELEMENT NO.= 6 LOAD NO=1
.493820E-01 .497628E-01 -.134819E-03 -3.19236 2.55931
ZI= 1.00000000000000 ETA= -1.00000000000000 ELEMENT NO.= 6 LOAD NO=2
.292060 .291338 .255590E-03 162.483 -161.283
ZI= -1.00000000000000 ETA= -1.00000000000000 ELEMENT NO.= 7 LOAD NO=1
.493820E-01 .497628E-01 .134820E-03 3.19236 2.55931
ZI= -1.00000000000000 ETA= -1.00000000000000 ELEMENT NO.= 7 LOAD NO=2
.292060 .291338 -.255583E-03 -162.483 -161.283
ZI= 1.00000000000000 ETA= 1.00000000000000 ELEMENT NO.= 9 LOAD NO=1
.426605E-01 .373798E-01 .122599E-03 18.9528 -3.59912
ZI= 1.00000000000000 ETA= 1.00000000000000 ELEMENT NO.= 9 LOAD NO=2
.715450E-01 .970611E-01 -.547278E-03 26.0118 9.10107
ZI= -1.00000000000000 ETA= 1.00000000000000 ELEMENT NO.= 10 LOAD NO=1
.396778E-01 .364850E-01 .223782E-03 12.3734 -3.59912
ZI= -1.00000000000000 ETA= 1.00000000000000 ELEMENT NO.= 10 LOAD NO=2
-.695150E-02 .735121E-01 -.739110E-03 209.199 9.10107

(b) Discuss the reduced integration technique with respect to the isopara-
metric cubic plate bending element.

References and Suggested Readings

1. E.B. Becker, G.F. Carey, J.T. Oden, Finite Element Method, vols. 1–5 (Prentice Hall, 1981)
2. J.L. Batoz, An explicit formulation for an efficient triangular plate bending element. Int. J.
Numer. Methods Eng. 18, 1077–1089 (1982)
3. R.D. Cook, D.S. Malkus, M.E. Plesha, Concepts and Applications of the Finite Element
Analysis, 3rd ed. (Wiley, 1989)
4. G.P. Bazeley, Y.K. Cheung, B.M. Irons, O.C. Zienkiewicz, Triangular elements in bending—
conforming and non-conforming solutions, in Proceedings of the Conference on Matrix
Methods of Structural Mechanics, Air Force Institute of Technology, Wright-Patterson Air
Force Base, Ohio, USA (1965)
5. O.C. Zienkiewicz, R.L. Taylor, The Finite Element Method, vol. 1, 4th ed. (McGraw-Hill, 1989)
6. S. Dasgupta, D. Sengupta, Horizontally curved isoparametric beam element with or without
elastic foundation including effect of shear deformation. Comput. Struct. 29(6), 967–973 (1988)
7. T.Y. Yang, Finite Element Structural Analysis (Prentice Hall Inc., 1986)
8. M.M. Hrabok, T.M. Hradey, A review and catalog of plate bending finite elements. Comput.
Struct. 19(3), 479–495 (1984)
9. S. Timoshenko, S. Woinsky-Krieger, Theory of Plates and Shells, 2nd ed. (McGraw-Hill Book
Co., 1959)
10. A.S. Mawaneya, Shear in isoparametric beam and plate bending elements. Proc. Inst. Civ. Eng.
(London) 61(Part 2), 197–204 (1976)
11. M. Mukhopadhyay, D.K. Dinker, Isoparametric linear bending element and one-point integra-
tion. Comput. Struct. 9, 365–369 (1978)
12. E.M. Hinton, A. Razzaque, O.C. Zienkiewicz, J.D. Davies, A simple finite element solution
for plates of homogeneous, sandwich and cellular construction. Proc. Inst. Civ. Eng. (London)
59(Part 2), 43–65 (1975)
Chapter 12
Finite Element Analysis of Shells

12.1 Introduction

A shell is a curved surface in space. Usually it is thin in comparison to its span.

There exists very limited scope for obtaining analytical solution for shell problems.
For shells having arbitrary shape and loading conditions, irregular stiffening and
support conditions, cut-outs and many other aspects of practical design, a closed-
bound solution is practically impossible. As such the finite element method is well
suited more to the structural analysis of shells. The development of the finite element
method for the shell analysis is not without formidable difficulties and pitfalls.
It is seen that there are three distinct approaches to the representation of the shell
structures. They are (1) flat elements, (2) curved elements and (3) three-dimensional
elements. The finite element analysis of these elements will be presented one after
another.

12.2 Flat Shell Element

A curved shell may be considered to consist of flat elements. Only triangular elements
are used. For many practical purposes, finite element solutions obtained on the basis
of flat elements give acceptable results.
A triangular element subjected simultaneously to membrane and bending actions
are shown in Fig. 12.1.
The element is lying in the x ' y' -plane, and the d.o.f corresponding to the membrane
action and the bending action are shown in Fig. 12.1. The displacement vector is given
by
{ }
{X }eT = u 1 u 2 u 3 v1 v2 v3 w1 w2 w3 θx1 θx2 θx3 θ y1 θ y2 θ y3 (12.1)

© The Author(s) 2022 287

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_12
288 12 Finite Element Analysis of Shells

u1 w1 θy1
v1
1 y′ 1 θx1

w2 w3
2 3 2 3
v2 v3
u2 θx2
u3 θy2 θy3 θx3
x′
(a) Membrane action (b) Bending action

Fig. 12.1 Flat triangular element

The stiffness matrix can be formed by combining the membrane stiffness with the
bending stiffness. The membrane stiffness part is same as a CST element. To this,
the triangular plate bending stiffness such as obtained from DKT element may be
added. While this is being done, let it be borne in mind that θ z' component has not
been considered in the displacement vector. It will be convenient in the assembly
stage if θ z is considered. This can be incorporated in the displacement vector and the
stiffness matrix may be modified by incorporating an appropriate number of zeros.
For the flat shell element in local coordinates, the stiffness matrix is given by
⎧ ⎫
⎪
⎪ u1 ⎪ ⎪
⎤⎪⎪ ⎪
⎡ ⎪ u2 ⎪ ⎪
[K m ] [0] [0] ⎪ ⎪
⎪
⎪
⎪
⎪
⎢ 6 × 6 6 × 9 6 × 3 ⎥⎪⎪ .
.. ⎪⎪
⎢ ⎥⎪⎪ ⎪
⎪
⎢ ⎥⎪⎪
⎪ v2 ⎪
⎪
⎪
⎢ [ ] ⎥ ⎪
⎪ w ⎪
⎪
⎢ ⎥⎨ 1 ⎬
⎢ [0] K p [0] ⎥
[K ]e {X }e = ⎢ ⎥ (12.2)
⎢ 9 × 6 9 × 9 9 × 3 ⎥⎪ ⎪
⎢ ⎥⎪⎪ θx1 ⎪ ⎪
⎢ ⎥⎪⎪ ⎪
⎪
⎢ ⎥⎪⎪
⎪
⎪
⎪
⎣ [0] [0] [0] ⎦⎪ θ y3 ⎪⎪
⎪
⎪ θz 1 ⎪⎪
⎪
⎪ ⎪
⎪
3×6 3×9 3×3 ⎪ ⎪ θ
⎪
⎪
⎪
⎪ z2 ⎪
⎪
⎩ ⎭
θz3

Instead of arranging the displacement vector as per Eq. (12.1) it can be arranged
node-wise.
The displacement vector for a typical node i is then given by
{ }
{X i }T = u i vi wi θxi θ yi θzi (12.3)

The element stiffness matrix can then be accordingly cast. In that case, the part
of the stiffness matrix involving the displacement at node i may be written as
12.2 Flat Shell Element 289
⎡[ ] ⎤
K rms 0 0 0 0
⎢2 × 2 0 0 0 0⎥
⎢ [ b] ⎥
[ ' ] ⎢ ⎥
⎢0 0 Kr s 0⎥
Kr s e = ⎢ ⎥ (12.4)
⎢0 0 0⎥
⎢ ⎥
⎣0 0 3×3 0⎦
0 00 0 0 0

Formulating the stiffness matrix in the manner of Eq. (12.4) has some advantages
when it is to be converted into global axes system.

12.2.1 Transformation of the Stiffness Matrix and Assembly

The stiffness matrix in the local coordinates is to be transformed along a common

global system. The direction cosines between local and global axes are to be obtained.
An arbitrary shell divided into triangular elements is shown in Fig. 12.2. The local
axis system is chosen for a typical element. Suppose the local x ' -axis is directed along
the side 12. The coordinates of node 1 and 2 are (x1 , y1 , z 1 ) and (x2 , y2 , z 2 ) with
respect to global axis system. Thus, the vector V 12 in terms of global coordinates are
⎧ ⎫ ⎧ ⎫
⎨ x2 − x1 ⎬ ⎨ x21 ⎬
V12 = y2 − y1 = y21 (12.5)
⎩ ⎭ ⎩ ⎭
z2 − z1 z 21

The direction cosines are given by the components of this vector by its length, i.e.
defining a unit vector

z 3 2 y
y′
z′ x′
1

x
(a) Arbitrary shell (b) A typical triangular element with
local and global coordinates

Fig. 12.2 An arbitrary shell with a different axis system

290 12 Finite Element Analysis of Shells
⎧ ⎫ ⎧ ⎫
⎨ λx ' x ⎬ 1 ⎨ 21 ⎬
x
vx ' = λx ' y = y (12.6)
⎩ ⎭ li j ⎩ 21 ⎭
λx z
' z 21

where
/
li j = 2
x21 + y21
2
+ z 21
2

The direction cosines in the other two directions are also to be obtained. z' -
direction is normal to the plane of the triangle. By forming cross products of two
vectors V 12 and V 13 , we obtain a vector parallel to z' -direction.
⎧ ⎫
⎨ y21 z 31 − z 21 y31 ⎬
Vz ' = V12 × V13 = z 21 x31 − x21 z 31 (12.7)
⎩ ⎭
x21 y31 − y21 x31

Length of this vector is

l z ' = (y21 z 31 − z 21 y31 )2 + (z 21 x31 − z 31 x21 )2 + (x21 y31 − y21 x31 )2

= 2Δ = 2 × area of the triangle

Therefore, unit vector of Vz ' is

⎧ ⎫ ⎧ ⎫
⎨ λz ' x ⎬ 1 ⎨ y21 z 31 − z 21 y31 ⎬
vz ' = λz ' y = z x − x21 z 31 (12.8)
⎩ ⎭ 2Δ ⎩ 21 31 ⎭
λz ' z x21 y31 − y21 x31

Finally, the direction cosines of y' -axis can be obtained in a similar manner to the
direction cosines of a vector normal to both x ' - and z' -directions. It is
⎧ ⎫ ⎧ ⎫
⎨ λy' x ⎬ ⎨ λz ' y λ x ' z − λz ' z λ x ' y ⎬
v y ' = λ y ' y = vz ' × v x ' = λz ' z λ x ' x − λz ' x λ x ' z (12.9)
⎩ ⎭ ⎩ ⎭
λy' z λz ' x λ x ' y − λz ' y λ x ' x

It may be noted that as the cross product of vectors in Eq. (12.7) involves only
unit vectors, so for the determination of direction cosines of z' -axis will not entail
any division by length.
Therefore, the rotational matrix between two sets of axis system is given by
⎧ ⎫
⎨ λx ' x λx ' y λx ' z ⎬
[λ] = λ y ' x λ y ' y λ y ' z (12.10)
⎩ ⎭
λz ' x λz ' y λz ' z
12.2 Flat Shell Element 291

The transformation matrix [T ] for the total shell element can now be built up. The
stiffness matrix of the element in global coordinate can be expressed as
[ ]
[K ]e = [T ]T K ' [T ] (12.11)

If the stiffness matrix is formed in the manner given in Eq. (12.4), then the nodal
transformation matrix is
[ ]
[λ] [0]
[L] = (12.12)
[0] [λ]

and the transformation matrix [T ] will be

⎡ ⎤
[L] [0] [0]
[T ] = ⎣ [0] [L] [0] ⎦ (12.13)
[0] [0] [L]

Typical stiffness submatrix in global coordinates will now become

[ ]
[K r s ]e = [L]T K r' s e [L] (12.14)
[ ]
where K r' s e is given by Eq. (12.4).
It may be noted that because of 0 in some diagonal elements of the stiffness matrix
(Eq. 12.2), the stiffness matrix will be singular if all the elements assembled are
coplanar. One approach of removing singularity is to eliminate elements of stiffness
matrix related to θ z terms, thus reducing the size of element stiffness matrix from
18 × 18 to 15 × 15. In another approach, in order to obviate the difficulty, the
following modification has been suggested. The 3 × 3 on diagonal null matrix shown
in Eq. (12.2) is replaced by the following equation
⎧ ⎫ ⎡ ⎤⎧ ⎫
⎨ M21 ⎬ 1.0 −0.5 −0.5 ⎨ θz1 ⎬
M = α E V ⎣ −0.5 1.0 −0.5 ⎦ θz2 (12.15)
⎩ 22 ⎭ ⎩ ⎭
M23 −0.5 −0.5 1.0 θz3

where E is the elastic modulus, V is the element volume and α is a number such
as 0.3 or less. So, some fictitious stiffness has been introduced corresponding to θ z
terms, but it offers no resistance to the mode θ z1 = θ z2 = θ z3 or to any other rigid
body motion.
292 12 Finite Element Analysis of Shells

12.3 Shell of Revolution

Both bending and membrane forces that occur in shells have been considered for
the shells of revolution. The structure is axisymmetric. In addition, the case of
axisymmetric loading only is dealt with.
A shell is divided into a series of frustum. Three displacements u, v and w in
the middle surface of the shell have been shown in Fig. 12.3. Due to axisymmetric
loading, the v component will be zero, u and w are the tangential and normal directions
respectively.
To simplify the analysis, the frustrum is assumed to be of conical shape as shown
in Fig. 12.4. The local axis system corresponds to u, w, dw ds
. The global axis system
corresponds to u, w, and dw ds
. Their relation at node i is given by
⎧ ⎫ ⎡ ⎤⎧ ⎫
⎨ ui ⎬ cos φ sin φ 0 ⎨ u i ⎬
w = ⎣ − sin φ cos φ 0 ⎦ wi (12.16)
⎩ ( dwi ) ⎭ ⎩ ⎭
ds i 0 0 1 βi
{ }
or {X }i = [λ] X i (12.17)

u is assumed to vary linearly with s and with w as cubic in s. Thus

u = α1 + α2 s (12.18)

Fig. 12.3 Shell of revolution

Ns
w Ms
φ
Nθ
s
Mθ

Fig. 12.4 Axisymmetric ui

shell element
ri βi
wi
i
φ r s
L
12.3 Shell of Revolution 293

w = α3 + α4 s + α5 s 2 + α6 s 3

It can be shown that the displacement function is

⎧ ⎫
⎪
⎪ ui ⎪
⎪
⎡( ) ⎤⎪⎪
⎪
⎪
⎪
⎪
1−s ' 0 0 '
s ( 0 ) ( 0 ⎨ wi
⎪ ⎪
⎬
{f} = ⎣ ( ) ( ) 3 ) ⎦ .
' 2
1 − 3s + 2s ' 3 ' ' 2
L s − 2s + s ' 3 '
0 3s − 2s ' ' 2
−s + s ' 3 L ⎪ . ⎪
0 ⎪
⎪ .) ⎪
⎪
⎪
⎪ ( ⎪
⎪
⎪
⎩ dw ⎪
⎭
ds j
(12.19)

where s ' = s
L
or,
[ ]
{ f } = N ' {X } (12.20)

Replacing the values of {X}i from Eq. (12.17) into Eq. (12.20) yields
⎧{ } ⎫
[ ]⎪ X i⎪
[ ' ] [λ] [0] ⎨ ⎬
{f} = N (12.21)
⎩ {X } ⎪
[0] [λ] ⎪ ⎭
j

or,
[ ] { }
{ f } = N ' [T ] X (12.22)

where [T ] is the transformation matrix.

The strain matrix for the shell is given by
⎧ ⎫ ⎧ ⎫
⎪
⎪ εs ⎪
⎪ ⎪
⎪ du/ds ⎪
⎪
⎨ ⎬ ⎨ ⎬
εθ (w cos φ + u sin φ)/r
{ε} = = (12.23)
⎪
⎪ κ ⎪ ⎪ −d w/ds
2 2 ⎪
⎩ s⎪ ⎭ ⎪ ⎩ sin φ dw ⎪
⎭
κθ − r ds

Combining Eqs. (12.19) and (12.23) yields

⎡
( −1/L ) ( 0 ) ( 0 )
⎢ 1 1 − s ' sin φ 1 − 3s ' 2 + 2s ' 3 cos φ L s ' − 2s ' 2 + s ' 3 cos φ
⎢
{ε} = ⎣ r ( ) r ( ) r
' '
0 1
(L 2 6 − 12s
) φ (
1
L
4 − 6s )
0 6 s ' − s ' 2 sin
rL
−1 + 4s ' − 3s ' 2 sinr φ
294 12 Finite Element Analysis of Shells
⎧ ⎫
⎤ ⎪ ui ⎪
1/L ( 0 ) 0 ⎪
⎪ ⎪
⎪
s' '2 ' 3 cos φ
( '2 ) ⎪
' 3 cos φ ⎥ ⎨ wi
⎪
⎬
sin φ 3s ( − 2s ) L −s ( + s ) ⎥
⎦ ⎪ ...
r r r (12.24)
'
0 1
L 2(
−6 + 12s ) (
1
2 − 6s)' ⎪ ⎪
⎪
⎪ ⎪
⎩ ( dw )
L
0 −6 s ' − s ' 2 sin φ
2s '
− 3s ' 2 sin φ ⎪ ⎪
⎭
rL r ds j

Equation (12.24) can be written as

( )
[[ ' ] [ ' ]] {X }i
{ε} = Bi | B j (12.25)
{X } j

Using the relation given by Eq. (12.17), Eq. (12.25) becomes

({ } )
[[ ' ] [ '] X
{ε} = Bi [λ] | B j [λ] { } i (12.26)
X j
{ }
or {ε} = [B] X (12.27)

The internal stress resultants for the shell element have been shown in Fig. 12.3.
The ‘stress’ vector is given by
⎧ ⎫
⎪
⎪ Ns ⎪
⎪
⎨ ⎬
Nθ
{σ } = (12.28)
⎪
⎪ M ⎪
⎩ s⎪ ⎭
Mθ

The stress–strain relationship for the shell elements is given by

σ = [D]∊ (12.29)

where
⎡ ⎤
1 ν 0 0
Et ⎢ ⎢ ν 1 0 0 ⎥ ⎥
[D] = (12.30)
1 − ν2 ⎣ 0 0 t /12 νt /12 ⎦
2 2

0 0 νt 2 /12 t 2 /12

The stiffness matrix of the shell element is given by

(
[k]e = [B]T [D][B]d A (12.31)
A

Now, d A = 2 πr ds = 2 πr Lds ' (12.32)

12.4 General Shell Finite Element of Triangular Shape 295

Therefore, replacing the value of dA from Eq. (12.32) into Eq. (12.31) yields
( 1
[k]e = [B]T [D][B]2π r L ds ' (12.33)
0

Based on the geometry of Fig. 12.4, it can be seen that

r = ri + s sin φ (12.34)

This value of r from Eq. (12.34) is to be substituted into Eq. (12.33) before the
integration is carried out.

12.4 General Shell Finite Element of Triangular Shape

We have already seen that the disadvantage of the flat shell element lies in the
non-coupling between bending and stretching within each element, which results
in the requirement of a large number of elements of the whole structure to achieve
satisfactory accuracy. A shell element formulation is presented in the following,
which can take care of the general shape [4].
The geometry of an arbitrary shallow shell element is shown in Fig. 12.5. The
shell is defined by the height ζ (ξ, η) above the base plane in which ξ, η are taken as
local coordinates and x, y as global coordinates. The base triangle has been indicated
as 1' 2' 3' . Their base dimensions are a, b and c, and the rotation can be derived without
much difficulty.

Fig. 12.5 Shell element 3

geometry and coordinate w
v
system Co
u
2

1
ζ(ξ, η)
z η

3′
y

c
θ
1′
b
a 2′
ξ
296 12 Finite Element Analysis of Shells

Special displacement functions have been used for ensuring conformity and high
accuracy.

12.4.1 Derivation of the Stiffness Matrix

The generalized displacements corresponding to the membrane action are u, v and

their first derivative at each corner node plus u and v at the element’s centroid. This
gives a total of twenty degrees of freedom corresponding to u and v. The generalized
displacements for w are w and its first and the second derivatives at the corner nodes.
This results in a total of 18 d.o.f. corresponding to w. In all, the element then has 38
degrees of freedom. The displacement vector for the element is given by
( ( ) ( ) ( )( ) ( ) ( ) ( 2 )
∂u ∂u ∂v ∂v ∂w ∂w ∂ w
{X }eT = u1 v1 w1
∂ξ 1 ∂η 1 ∂ξ 1 ∂η 1 ∂ξ 1 ∂η 1 ∂ξ 2 1
( 2 ) ( 2 ) )
∂ w ∂ w
u 2 · · · · · · · · · u 3 · · · · · · · · · u c vc (12.35)
∂η2 1 ∂ξ ∂η 1

The subscripts 1, 2, 3 denote the corners and ‘c’ denotes the centroid of the
element.
The displacement u, v and w are expressed in polynomials as follows:

u = α1 + α2 ξ + α3 η + α4 ξ2 + α5 ξη + α6 η2 + α7 ξ3 + α8 ξ2 η + α9 ξη2 + α10 η3
(12.36a)

v = α11 + α12 ξ + α13 η + α14 ξ2 + α15 ξη + α16 η2 + α17 ξ3 + α18 ξη + α19 ξη2 + α20 η3
(12.36b)
w = α21 + α22 ξ + α23 η + α24 ξ2 + α25 ξη + α26 η2 + α27 ξ3 + α28 ξ2 η

+ α29 ξη2 + α30 η3 + α31 ξ4 + α32 ξ3 η + α33 ξ2 η2 + α34 ξη3 + α35 η4 + α36 ξ5 + α37 ξ3 η2

+ α38 ξ2 η3 + α39 ξη4 + α40 η5 (12.36c)

It may be noted that the displacement function w contains 20 constants whereas

the degrees of freedom associated with w is only 18. Further, for the first time, we
are encountering a situation where the displacement function is not evenly balanced
between ξ and η. There is a special purpose for writing it in this manner. As w in
Eq. (12.36c) does not contain ξ 4 η, the requirement that the slope normal to the edge
12 is cubic automatically satisfied.
40 constants associated with Eqs. (12.36a)–(12.36c) are to be evaluated. 38 equa-
tions can be obtained on the basis of substituting ξ, η values corresponding to each
element of the displacement vector in Eqs. (12.36a)–(12.36c). Two additional equa-
tions are to be sought. They are obtained on the basis of imposition of the conditions
for cubic variation of normal slope along the remaining two edges which are however
somewhat complicated.
12.4 General Shell Finite Element of Triangular Shape 297

Fig. 12.6 Oblique boundary η P

θ
O ξ

In investigating the above conditions, only the fifth-degree terms in Eq. (12.36c)
are to be considered, since the lower order terms satisfy the conditions automatically.
Consider the derivative normal to the line OP in Fig. 12.6. The derivation is given by

∂w ∂w ∂w
= sin θ − cos θ
∂η ∂ξ ∂η
( )
= 5α36 ξ 4 + 3α37 ξ 2 η2 + 2α38 ξ η3 + α39 η4 sin θ
( )
− 2α37 ξ 3 η + 3α38 ξ 2 η2 + 4α39 ξ η3 + 5α40 η4 cos θ (12.37)

But on OP

ξ = s cos θ, η = s sin θ (12.38)

where s is the distance along OP. Substitution of Eq. (12.38) into Eq. (12.37) yields

∂w { ( ) (
= s 4 5α36 cos4 θ sin θ + α37 3 cos2 θ sin3 θ − 2 cos4 θ sin θ + α38 2 cos θ sin4 θ
∂η
) ( ) }
− 3 cos3 θ sin2 θ + α39 sin5 θ − 4 cos2 θ sin3 θ − 5α20 cos θ sin4 θ
(12.39)

Hence the condition that the normal derivative varies cubically along OP or any
parallel line is that the expression in the parenthesis of Eq. (12.39) must vanish.
To apply this condition to the edge 2' 3' of the triangular element of Fig. 12.4, the
appropriate values of cos θ and sin θ must be inserted, that is
a c
cos θ = √ , sin θ = √ (12.40)
a2 + c2 a + c2
2

Substitution of Eq. (12.40) into Eq. (12.39) yields

( ) ( ) ( )
5a 4 c α36 + 3a 2 c3 − 2a 4 c α37 + −2ac4 + 3a 3 c2 α38 + c5 − 4a 2 c3 α39 + 5ac4 α40 = 0 (12.41)

Similarly, the condition for the cubic variation of the normal slope along the edge
13 is
298 12 Finite Element Analysis of Shells

( ) ( ) ( )
5b4 c α36 + 3b2 c3 − 2b4 c α37 + −2bc4 + 3b3 c2 α38 + c5 − 4b2 c3 α39 − 5bc4 α40 = 0 (12.42)

Therefore, with appropriate substitution the following equations result:

⎧ ⎫
⎨ {X }e ⎬
0 [T ] {α}
⎩ ⎭ = (12.43)
0 40 × 40 40 × 1
40 × 1

where [T ], the transformation matrix is given in Table 12.1.

The determinant of the transformation matrix [T ] has the value
( )( )
−64c34 (a + b)31 a 2 + c2 b2 + c2 /729

which is non-zero for all practical purposes. Hence from Eq. (12.43), we get
⎧ ⎫
⎨ {X }e ⎬
{α} = [T ]−1 0 = [T1 ]{X }e (12.44)
⎩ ⎭
0

where [T1 ] is of the size 40 × 38 and contains first 38 columns of [T ]–1 .

Following the shallow shell theory of Novozhilov (12.1), the membrane strains
of the shell are given by
2 ⎫
∂ ζ
εξ ξ = ∂u
∂ξ − ∂ξ 2 w
⎪
⎪
⎪
⎪
⎬
∂v − ∂ 2 ζ w
εηη = ∂η (12.45)
∂η 2 ⎪
⎪
⎪
εξ η = ∂u + ∂v − 2 ∂ 2 ζ w ⎪
⎭
∂η ∂ξ ∂ξ ∂η

εξ ξ , εηη are axial strains in ξ - and η-directions and εξ η is the shear strain.
where
⎡ ⎤
1 −b 0 b2 0 0 −b3 0 0 0
[S1 ] = ⎣ 0 1 0 −2b 0 0 3b2 0 0 0 ⎦
0 0 1 0 −b 0 0 b2 0 0
⎡ ⎤
1 −b 0 b2 0 0 − b3 0 0 0 b4 0 0 0 0 − b5 0 0 0 0
⎢ ⎥
⎢0 1 0 −2b 0 0 3b2 0 0 0 −4b3 0 0 0 0 5b4 0 0 0 0⎥
⎢ ⎥
[ ] ⎢0 0 1 0 −b 0 0 b2 0 0 0 − b3 0 0 0 0 0 0 0 0⎥
S2 = ⎢
⎢0
⎥
⎢ 0 0 2 0 0 −6b 0 0 0 12b2 0 0 0 0 −20b3 0 0 0 0⎥⎥
⎢ ⎥
⎣0 0 0 0 1 0 0 −2b 0 0 0 3b2 0 0 0 0 0 0 0 0⎦
0 0 0 0 0 2 0 0 −2b 0 0 0 2b2 0 0 0 −2b3 0 0 0
⎡ ⎤
1 a 0 a2 0 0 a3 0 0 0
[S3 ] = ⎣ 0 1 0 2a 0 0 3a 2 0 0 0 ⎦
0 0 1 0 a 0 0 a2 0 0
12.4 General Shell Finite Element of Triangular Shape 299

Table 12.1 Transformation

matrix
⎡ ⎤
[S1 ] [0] [0]
⎢ ⎥
⎢ [0] [S1 ] [0] ⎥
⎢ ⎥
⎢ ⎥
⎢ [0] [0] [S2 ] ⎥
⎢ ⎥
⎢ [S3 ] [0] [0] ⎥
⎢ ⎥
⎢ ⎥
⎢ [0] [S3 ] [0] ⎥
⎢ ⎥
⎢ [0] [0] [S4 ] ⎥
⎢ ⎥
[T ] = ⎢ ⎥
⎢ [S5 ] [0] [0] ⎥
⎢ ⎥
⎢ ⎥
⎢ [0] [S5 ] [0] ⎥
⎢ ⎥
⎢ [0] [0] [S6 ] ⎥
⎢ ⎥
⎢ ⎥
⎢ [S7 ] [0] [0] ⎥
⎢ ⎥
⎢ ⎥
⎣ [0] [S7 ] [0] ⎦
[0] 0 [S8 ]

⎡ ⎤
1 a 0 a2 0 0 a3 0 0 0 a4 0 0 0 0 a5 0 0 0 0
⎢0 3a 2 0 0 4a 3 0 5a 4 0⎥
⎢ 1 0 2a 0 0 0 0 0 0 0 0 0 ⎥
⎢ ⎥
⎢0 0 1 0 a 0 0 a2 0 0 0 a3 0 0 00 0 0 0 0⎥
[S4 ] = ⎢ ⎥
⎢0 0 0 2 0 0 6a 0 0 0 12a 2 0 0 0 0 20a 3 0 0 0 0⎥
⎢ ⎥
⎣0 0 0 0 1 0 0 2a 0 0 0 3a 2 0 0 00 0 0 0 0⎦
0 0 0 0 0 2 0 0 2a 0 0 0 2a 2 0 00 2a 3 0 0 0
⎡ ⎤
10c 0 0 c2 0 0 0 c3
[S5 ] = ⎣ 0 1 0 0 c 0 0 0 c2 0 ⎦
001 0 0 2c 0 0 0 3c2
⎡ ⎤
1 0 c 0 0 c2 0 0 0 c3 0 0 0 0 c4 0 0 0 0 c5
⎢ 0 1 0 0 c 0 0 0 c2 0 0 0 0 c3 0 0 0 0 c4 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 1 0 0 2c 0 0 0 3c2 0 0 0 0 4c3 0 0 0 0 5c4 ⎥
[S6 ] = ⎢ ⎥
⎢ 0 0 0 2 0 0 0 2c 0 0 0 0 2c2 0 0 0 0 2c3 0 0 ⎥
⎢ ⎥
⎣ 0 0 0 0 1 0 0 0 2c 0 0 0 0 3c2 0 0 0 0 4c3 0 ⎦
0 0 0 0 0 2 0 0 0 6c 0 0 0 0 12c2 0 0 0 0 20c3
[ ]
a − b c (a − b)2 (a − b)c c2 (a − b)3 (a − b)2 c (a − b)c2 c3
[S7 ] = 1
⎡
3 3 9 9( 9 27) ( 27 ) ( 27 )27 ⎤
[ ] 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 5a 4 c 3a 2 c3 − 2a 4 c −2ac4 − 3a 3 c2 c5 − 4a 2 c3 5ac4
S8 = ⎣ ( ) ( ) ( ) ⎦
4 2 3
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 5b c 3b c − 2b c 4 4 3
−2bc − 3b c 2 c5 − 4b2 c3 5bc4

The bending strains are given by

300 12 Finite Element Analysis of Shells

∂ 2w ∂ 2w ∂ 2w
κξ ξ = − , κ ηη = − , κ ξ η = − (12.46)
∂ξ 2 ∂η2 ∂ξ ∂η

For shallow shells, the area of the shell surface is approximately equal to its projected
area. Therefore, the strain energy expression can be integrated over the base plane.
The strain energy expression for the shell is
( ( ([( )2 ( )2 ( )( ) ( ) ]
Et ∂u ∂v ∂u ∂v 1 ∂u ∂v 2
U= ( ) + + 2ν + (1 − ν) +
2 1−ν 2 ∂ξ ∂η ∂ξ ∂η 2 ∂η ∂ξ
[( ) ( ) ( )( )]
∂2ζ ∂ 2 ζ ∂u ∂2ζ ∂ 2 ζ ∂v ∂2ζ ∂u ∂v
−2 +ν 2 + +ν 2 + (1 − ν) + w
∂ξ 2 ∂η ∂ξ ∂η 2 ∂ξ ∂η ∂ξ ∂η ∂η ∂ξ
⎡( )2 ( )2 ( )2 ⎤
∂2ζ ∂2ζ ∂2ζ ∂2ζ ∂2ζ
+⎣ + + 2ν 2 + 2(1 − ν) ⎦w 2
∂ξ 2 ∂η 2 ∂ξ ∂η 2 ∂ξ ∂η
⎡( )2 ( )2 ( )2 ⎤⎫
t2 ∂2w ∂2w ∂2w ∂2w ∂2w ⎬
+ ⎣ + + 2ν + 2(1 − ν) ⎦ ∂ξ ∂η (12.47)
12 ∂ξ 2 ∂η 2 ∂ξ ∂η
2 2 ∂ξ ∂η ⎭

where the integral is over the base triangle 1' 2' 3' in Fig. 12.4. The function ζ (ξ, η)
has been assumed to be of quadratic form given below:

ζ (ξ, η) = C1 + C2 ξ + C3 η + C4 ξ 2 + C5 ξ η + C6 η2 (12.48)

This implies that the shell element has constant curvatures and this is consistent
with the approximation of the shallow shell theory. Only shape quantities required
in calculating strain energy are the shell curvatures. They may be calculated from
Eq. (12.48) or may be calculated directly. In order to use Eq. (12.48), ξ and η-values
at 6 points (at 1' , 2' , 3' and at mid-edges) are to be specified and then the constants
are to be evaluated.
For obtaining the stiffness matrix of the element, Eqs. (12.47), (12.36a)–(12.36c)
and (12.44) are combined and the necessary integration is carried out and the strain
energy expression reduces to

Et
U= ( ) {α}T [k]e {α} (12.49)
2 1 − ν2

[k]e can be explicitly evaluated. In order to obtain this, Eqs. (12.36a)–(12.36c) are
rewritten as
12.4 General Shell Finite Element of Triangular Shape 301
⎫
Σ
10
⎪
⎪
u= αi ξ η
m i ni
⎪
⎪
⎪
⎪
i=1 ⎪
⎪
⎪
⎪
⎪
⎬
Σ
20
v= αi ξ pi ηqi (12.50)
⎪
⎪
i=11 ⎪
⎪
⎪
⎪
⎪
⎪
Σ
40 ⎪
ri si ⎪
w= αi ξ η ⎪ ⎭
i=21

Based on the notations of Eq. (12.50), it can be shown that

( ) ( )
ki j = m i m j F m i + m j − 2, n i + n j + qi q j F pi + p j , qi + q j − 2
1 [ ( ) ( )]
(1 − ν) n i n j F m i + m j , n i + n j − 2 + pi p j F pi + p j − 2, qi + q j
2
[ ]
1 ( )
+ (1 − ν)n j pi + νm j qi F m j + pi − 1, n j + qi − 1
2
[ ]
1 ( )
+ (1 − ν)n i p j + νm i q j F m i + p j − 1, n i + q j − 1
2
( )
∂2ζ ∂2ζ [ ( ) ( )]
− + ν 2 m i F m i + r j − 1, n i + s j + m j F m j + ri − 1, n j + si
∂ξ 2 ∂η
( )
∂2ζ ∂2ζ [ ( ) ( )]
− + ν 2 qi F pi + r j , qi + s j − 1 + q j F p j + ri , q j + si − 1
∂η 2 ∂ξ

∂2ζ [ ( ) ( )
− (1 − ν) n F m i + r j , n i + s j − 1 + n j F m j + ri , n j + si − 1
∂ξ ∂η i
( ) ( )]
+ pi F pi + r j − 1, qi + s j + p j F p j + ri − 1, q j +si
⎡( )2 ( )2 ( )2 ⎤
∂2ζ ∂2ζ ∂2ζ ∂2ζ ∂2ζ ( )
+ ⎣ + + 2ν 2 + 2(1 − ν) ⎦ F ri + r j , si + s j
∂ξ 2 ∂η2 ∂ξ ∂η2 ∂ξ ∂η

t2 { ( )( ) ( ) ( )( ) ( )
+ r r r − 1 r j − 1 F ri + r j − 4, si + s j + si s j si − 1 s j − 1 F ri + r j , si + s j − 4
12 i j i
[ ( )( ) ( )( )]
+ 2(1 − ν)ri r j si s j + vri s j ri − 1 s j − 1 + νr j si r j − 1 si − 1
( )}
× F ri + r j − 2, si + s j − 2 (12.51)

where
[ ] m!n!
F(m, n) = cn+1 a m+1 − (−b)m+1 (12.52)
(m + n + 2)!

It may be noted that i and j run from 1 to 40. Therefore, mi and ni are defined to
be zero for i > 10, pi and qi are zero for i < 11 and i > 20 and r i and si are zero for i
< 21.
Combining Eqs. (12.44) and (12.49) yields

Et
U= ( ) {X }eT [K]1 {X }e (12.53)
2 1 − ν2
302 12 Finite Element Analysis of Shells

where

[K ]1 = [T1 ]T [k]e [T1 ] (12.54)

[k]e is the stiffness matrix of size 38 × 38 in terms of generalized displacements in

the local axis system ξ, η.
The displacement vectors in the local global axis system are related as follows:
{ }
{X }e = [R] X g e (12.55)
{ }
where X g e is the displacement vector in the global degrees of freedom.
( ( ) ( ) ( ) ( ) ( ) ( )
{ } ∂u ∂u ∂v ∂v ∂w ∂w
Xg e = u1 v1 w1
∂x 1 ∂y 1 ∂x 1 ∂y 1 ∂x 1 ∂y 1
( 2 ) ( 2 ) ( 2 ) )
∂ w ∂ w ∂ w
u 2 . . . . . . . . . ., u c vc (12.56)
∂ x 2 1 ∂ y 2 1 ∂ x∂ y 1

Combining Eqs. (12.53)–(12.55) yields

Et { }T [ ] { }
U= ( ) X g e Kg e X g e (12.57)
2 1−ν 2

[ ]
where K g e = [R]T [T1 ]T [K ]1 [T1 ][R] (12.58)
[ ]
Kg e
is the stiffness matrix in global coordinates of size 38 × 38.

12.4.2 Consistent Load Vector

From the principle of virtual work, the consistent load vector can be calculated. The
applied loads in u-, v- and w-directions are Qu (ξ, η), Qv (ξ, η), Qw (ξ, η) respectively
(Table 12.2). The transformed load vector becomes

{P}e = [R]T [T1 ]T {Q} (12.59)

where the entries in the column vector {Q} are

(( ⎫
Q u ξm i ηni dξdη i < 11 ⎪
⎪
⎪
⎪
⎪
⎬
((
Qi = Q v ξ pi ηqi dξdη 10 < i < 20 (12.60)
⎪
⎪
⎪
⎪
(( ⎪
⎭
Q w ξri ηsi dξdη i > 20
12.4 General Shell Finite Element of Triangular Shape 303

Table 12.2 Rotation matrix [R]

⎡ ⎤
[R1 ]
⎢ ⎥
⎢ [R2 ] ⎥
⎢ ⎥
⎢ ⎥
⎢ [R1 ] ⎥
⎢ ⎥
[R] = ⎢
⎢ [R2 ] ⎥
⎥
⎢ ⎥
⎢ [R1 ] ⎥
⎢ ⎥
⎢ ⎥
⎣ [R2 ] ⎦
[R3 ]
(submatrices not shown in the above matrix are null matrices)

where
⎡ ⎤
cos θ 0 0 sin θ 0 0
⎢ 0 cos2 θ sin θ cos θ sin θ cos θ sin2 θ ⎥
⎢ 0 ⎥
⎢ ⎥
⎢ 0 − sin θ cos θ cos θ2
0 − sin θ sin θ cos θ ⎥
2
[R1 ] = ⎢ ⎥
⎢ − sin θ 0 0 cos θ 0 0 ⎥
⎢ ⎥
⎣ 0 − sin θ cos θ − sin2 θ 0 cos θ
2
sin θ cos θ ⎦
0 sin2 θ − sin θ cos θ 0 − sin θ cos θ cos2 θ
⎡ ⎤
1 0 0 0 0 0
⎢ 0 cos θ sin θ ⎥
⎢ 0 0 0 ⎥
⎢ 0 − sin θ cos θ ⎥
⎢ 0 0 0 ⎥
⎢ ⎥
[R2 ] = ⎢ 0 0 0 cos2 θ 2 sin θ cos θ sin2 θ ⎥
⎢ ⎥
⎢0 0 0 − sin θ cos2 θ − sin2 θ sin θ cos θ ⎥
⎢ ⎥
⎣ cos θ ⎦
0 0 0 sin θ −2 sin θ cos θ cos θ
2 2

[ ]
cos θ sin θ
[R3 ] =
− sin θ cos θ

The above integration is to be carried out over the base triangle 1' 2' 3' of Fig. 12.5.

12.4.3 Condensation of Stiffness Matrix

It is convenient for this element to condense out the centroidal displacements uc and
vc . Since these displacements are inside the element, they will remain unaffected
when the elements are joined together. The matrix equation is partitioned as follows:
304 12 Finite Element Analysis of Shells
⎧ ⎫ ⎧ ⎫
[ ]⎨ {X o }e ⎬ ⎨ {Po }e ⎬
[ ]{ } [K ]o [K ]oc
Kg Xg = − − − = − − −− (12.61)
e e [K ]oc [K ]c ⎩ ⎭ ⎩ ⎭
{X c }e {Pc }e

As given in Sect. 5.4, the final equation of the condensed stiffness matrix becomes

[K ]e {X o }e = {Po }e
(12.62)
36 × 36 36 × 1 36 × 1

where

[K ]e = [K ]o − [K ]oc [K ]−1 T
c [K ]oc (12.63)

{P}c = { Po }e − [K ]oc [K ]−1

c {Pc }e (12.64)

12.5 Isoparametric General Shell Element

A shell element shape can be modelled by a three-dimensional solid element that

has a thickness dimension considerably smaller than its other dimensions. Consider
a typical shell element of Fig. 12.7. The external faces of the element are curved,
while the sections across the thickness are generated by straight lines. Pair of points,
i-top and i-bottom, each with given Cartesian prescribes the shape of the element.

ξ=1 (–1,1) (1,1)

ξ= –1 η

ζ η=1
0
η ξ

ξ
z
ζ= –1 (–1,–1) (1,–1)

η= –1 ξ=1 y
x

Fig. 12.7 Curved shell element

12.5 Isoparametric General Shell Element 305

12.5.1 Geometry of the Shell Element

Let ξ, η be the curvilinear coordinates in the middle plane of the shell and ζ a
linear coordinate in the thickness direction. If we further assume that ξ, η, ζ vary
between −1 and +1 on the respective faces of the element, we can then write a
relationship between the cartesian coordinates of any point of the shell and the
curvilinear coordinates of the form
⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎨x ⎬ Σ 1+ζ⎨ i⎬
x Σ 1−ζ⎨ i⎬
x
y = Ni (ξ, η) yi + Ni (ξ, η) yi (12.65)
⎩ ⎭ 2 ⎩ ⎭ 2 ⎩ ⎭
z z i top z i bottom

where N i (ξ, η) are the shape functions of two-dimensional element.

It is convenient to write the relationship of Eq. (12.65) in a form specified by the
‘vector’ containing the upper and the lower points (i.e., a vector of length equal to
the shell thickness t i ) and the midsurface coordinates.
We can rewrite Eq. (12.65) as
⎧ ⎫ ⎧ ⎫
⎨ x ⎬ Σ ⎨ xi ⎬ Σ ζ
y = Ni yi + Ni V 3i (12.66)
⎩ ⎭ ⎩ ⎭ 2
z z i mid
⎧ ⎫ ⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎨ xi ⎬ ⎨ xi ⎬ ⎨ x j − xk ⎬ ⎨ l3i ⎬
with V 3i = yi − yi = y j − yk = ti m 3i (12.67)
⎩ ⎭ ⎩ ⎭ ⎩ ⎭ ⎩ ⎭
z i top z i bottom z j − zk n 3i
⎧ ⎫ ⎧ ⎫
⎨ l3i ⎬ x − xk ⎬
1 ⎨ j
where m = y j − yk (12.68)
⎩ 3i ⎭ ti ⎩ ⎭
n 3i z j − zk

in which l 3i , m3i and n3i are the direction cosines of the line jk (Fig. 12.8). Therefore
⎧ ⎫
⎨ l3i ⎬
V 3i = ti m 3i (12.69)
⎩ ⎭
n 3i

Equation (12.66) combined with Eqs. (12.67) and (12.68) yields

⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎨ x ⎬ Σ ⎨ xi ⎬ Σ ti ⎨ l3i ⎬
y = Ni yi + Ni ζ m 3i (12.70)
⎩ ⎭ ⎩ ⎭ 2⎩ ⎭
z zi n 3i

It may (be noted)that unlike Eq. (12.66), the suffix mid is eliminated in Eq. (12.70)
and xi = x j + xk /2 and so on.
306 12 Finite Element Analysis of Shells

J=itop
V3i
P ii
ζ=0

K=ibottom

(a)

βi
z
V3i P
V2i ti
wi β(ζ )
2
y V2i
vi ζti
2 ii
V1i
ui αi V1i

(b) (c)

Fig. 12.8 a Typical node i with V 3i , b Orthogonal vectors at node i. Directions of rotational d.o.f
αi and βi translational d.o.f ui, , vi , wi , c displacements at arbitrary point P on V 3i

12.5.2 Displacement Field

The displacement field has now to be specified for the element. As the strains in
the direction normal to the midsurface assumed to be negligible, the displacement
throughout the element will be taken to be uniquely defined by the three cartesian
components of the midsurface node i displacement plus the displacement relative
to the node i created by the rotation of V 3i . The relative displacement components,
shown in Fig. 12.8c, must be resolved into x, y and z components before being added
to the displacement of point P
) ( ( )
ti ti
u p = u i − αi ζ l2i + βi ζ l1i (12.71)
2 2

in which nodal rotations α and β are presumed to be small. Displacements of an

arbitrary point in the element are
⎧ ⎫ ⎧ ⎫
⎨ u ⎬ Σ ⎨ ui ⎬ Σ ( )
ti αi
v = N i v + N i ζ [μi ] (12.72)
⎩ ⎭ ⎩ i⎭ 2 βi
w wi
12.5 Isoparametric General Shell Element 307

u, v and[w are
] the displacements in the directions of the global x-, y- and z-axes. The
matrix μi which involves quantities such as l1i , l 2i, etc. are to be determined. As
an infinity of vector dimensions normal to the given direction can be generated, a
particular scheme has to be devised to ensure a unique definition. One such scheme
is discussed in the following.
Vectors V 1i and V 2i in Fig. 11.8 are perpendicular to each other and to V 3i .
Thus V 1i and V 2i are tangent to the midsurface, but they are not required to have a
particular relation to cartesian coordinate directions. V 1i and V 2i are used to define
the directions of nodal rotational d.o.f α i and β i , which are shared by all elements
that share node i. Set of α i - and β i -directions may differ from node to node.
The Jacobian matrix is defined as
⎡ ∂x ∂y ∂z ⎤
∂ξ ∂ξ ∂ξ
⎢ ∂y ⎥
[J ] = ⎣ ∂∂ηx ∂η
∂z
∂η ⎦ (12.73)
∂x ∂y ∂z
∂ζ ∂ζ ∂ζ

The basis of it is given later.

Equation (12.73) represents the midsurface tangent vector. First row contains the
components of the vector V 1i in the global x-, y- and z-directions
( ) ( ) ( )
∂x ∂y ∂z
V 1i = i+ j+ k (12.74)
∂ξ i ∂ξ i ∂ξ i

Similarly,
( ) ( ) ( )
∂x ∂y ∂z
V 2i = i+ j+ k (12.75)
∂η i ∂η i ∂η i
( )
V 3i = V 1i × V 2i ti (12.76)

Finally,

V 2i = V 3i × V 1i (12.77)

[ V] 1i and V 2i are given by dividing each vector by its

Direction cosines of
magnitude. Therefore, μi of Eq. (12.72) is given by
⎡ ⎤
[ ] [ V 2i V 1i ] −l2i l1i
μi = − |V 2i | |V 1i | = ⎣ −m 2i m 1i ⎦ (12.78)
−n 2i n 1i
| | | |
where |V 2i | and |V 1i | are the magnitudes of vectors V 2i and V 1i respectively.
Combining Eqs. (12.70) and (12.73), we get
308 12 Finite Element Analysis of Shells
⎡Σ( ζ ti
) Σ( )
ζ ti Σ( ) ⎤
xi + l ∂ Ni
2 3i ) ∂ξ
yi + m 3i ∂∂ξNi z i + ζ2ti n 3i ∂∂ξNi
⎢Σ( ζ ti Σ( 2) Ni Σ (
ζ ti
) Ni ⎥
[J ] = ⎣ xi + l ∂ Ni
2 3i ∂η
yi + m 3i ∂∂η z i + ζ2ti n 3i ∂∂η ⎦ (12.79)
Σ ti Σ ti 2 Σ ti
l N
2 3i i
m N
2 3i i
n
2 3i i
N

Displacements of any arbitrary point in the element are given by Eq. (12.72).

12.5.3 Strains Inside the Element

The strains in terms of displacement derivatives are expressed as

⎧ ⎫
⎪
⎪ u ,x ⎪
⎪
⎧ ⎫ ⎡ ⎪
⎪ ⎪
⎤ ⎪
⎪ u ,y ⎪
⎪
⎪
⎪
⎪ εx ⎪ ⎪ 1 0 0 0 0 0 0 0 0 ⎪
⎪ ⎪
⎪
⎪ ⎪
⎪ ⎢0 ⎪ u ,z ⎪
⎪
⎪
⎪
⎪ ε y ⎪⎪
⎪ ⎢ 0 0 0 1 0 0 0 0⎥⎥
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎨ ⎬ ⎢ ⎥ ⎨ v,x ⎪⎬
εz ⎢0 0 0 0 0 0 0 0 1⎥
=⎢ ⎥ v,y (12.80)
⎪ γx y ⎪
⎪ ⎪ ⎢0 1 0 1 0 0 0 0 0⎥ ⎪
⎪ ⎪
⎪
⎪ ⎪
⎪ ⎢ ⎥ ⎪
⎪ v,z ⎪⎪
⎪
⎪
⎪ γ ⎪
⎪ ⎣0 0 0 0 0 1 0 1 0⎦ ⎪
⎪ ⎪
⎪
⎪
⎩
yz ⎪
⎭ ⎪
⎪ w ⎪
⎪
γzx 0 0 1 0 0 0 1 0 0 ⎪
⎪
,x ⎪
⎪
⎪
⎪ w ⎪
⎪
⎪
⎩ ,y ⎪
⎭
w,z

{ε}x yz = [H ][ u ]x yz (12.81)

where [H] is the left matrix on the right-hand side.

All six strain terms have been included in Eq. (12.80) because the shell midsurface
has no particular orientation with respect to the cartesian coordinates xyz.
Again,

∂u ∂u ∂ x ∂u ∂ y ∂u ∂z
= + + (12.82)
∂ξ ∂ x ∂ξ ∂ y ∂ξ ∂z ∂ξ

Equation (12.82) can be written as

⎧ ⎫
∂u ⎪
[ ]⎪
∂u ∂ x ∂ y ∂z ⎨ ∂∂ux ⎬
= ∂y (12.83)
∂ξ ∂ξ ∂ξ ∂ξ ⎪ ⎩ ∂u ⎪ ⎭
∂ξ

Similarly,
⎧ ∂u ⎫ ⎡ ∂x ∂y ∂z
⎤⎧ ⎫ ⎧ ⎫
⎪ ∂u ∂u
⎨ ∂ξ ⎪
⎬ ∂ξ ∂ξ ∂ξ
⎪
⎨ ⎪
⎬ ⎪
⎨ ∂x ⎪
⎬
∂u ⎢ ∂x ∂y ∂z ⎥ ∂∂ux ∂u
∂η =⎣ ∂η ∂η ∂η ⎦ ∂y = [J ] ∂y (12.84)
⎩ ∂u ⎪
⎪ ⎭ ∂x ∂y ∂z
⎪
⎩ ∂u ⎪
⎭ ⎪
⎩ ∂u ⎪
⎭
∂ζ ∂ζ ∂ζ ∂ζ ∂z ∂z
12.5 Isoparametric General Shell Element 309

In the above equation, [J] is the Jacobian matrix. This is already introduced in
Eq. (12.73), which is defined in terms of nodal coordinates in Eq. (12.79).
Therefore,
⎧ ⎫ ⎧ ∂u ⎫
⎪ ∂u
⎨ ∂x ⎪ ⎬ ⎪
⎨ ∂ξ ⎪
⎬
∂u −1 ∂u
∂ = [J ] ∂η (12.85)
⎩ ∂u ⎪
⎪ y
⎭ ⎪
⎩ ∂u ⎪
⎭
∂z ∂ζ

The derivatives of displacements in the global x-, y- and z-coordinates are related
to the local ξ, η, ζ (curvilinear) coordinates as
⎧ ⎫ ⎧ ⎫
⎪
⎪ u ,x ⎪
⎪ ⎪
⎪ u ,ξ ⎪
⎪
⎪
⎪ ⎪ ⎪ ⎪
⎪
⎪ u ,y ⎪
⎪
⎪
⎪
⎪
⎪ u ,η ⎪
⎪
⎪
⎪
⎪u ⎪ ⎪
⎪ ⎡ ⎤ ⎪
⎪ ⎪
⎪
⎪
⎨ ,z ⎬ [J ] −1
0 0 ⎪
⎨ u ,ζ ⎪
⎬
. = ⎣ 0 [J ] −1
0 ⎦ . (12.86)
⎪
⎪ ⎪ ⎪ ⎪
⎪
⎪ . ⎪ ⎪
⎪ 0 0 [J ]−1 ⎪⎪
⎪ . ⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ . ⎪
⎪ ⎪
⎪ . ⎪
⎪
⎪
⎩ ⎪
⎭ ⎪
⎩ ⎪
⎭
w,z w,ζ

or,
[ ]−1
{u}x yz = J {u}ξ ηζ (12.87)

From Eq. (11.72) we get

⎧ ⎫ ⎡ ⎤
⎪
⎪ u ,ξ ⎪
⎪ Ni,ξ 0 0 −ζ ti Ni,ξ l2i /2 ζ ti Ni,ξ l1i /2 ⎧ ⎫
⎪
⎪ ⎪
⎪ ⎢
⎪
⎪ ⎪
u ,η ⎪ ⎢ Ni,η 0 0 −ζ ti Ni,η l2i /2 ζ ti Ni,η l1i /2 ⎥ ⎥⎪⎪ ui ⎪
⎪
⎪
⎪ ⎪
⎪ ⎢ ⎪ ⎪
⎪
⎨ u ,ζ ⎪
⎬ Σ⎢ 0 0 0 −ti Ni l2i /2 −ti Ni l1i /2 ⎥⎨ vi ⎪
⎥ ⎪
⎪ ⎪
⎬
⎢ ⎥
v,ξ = ⎢ 0 Ni,ξ 0 −ζ ti Ni,ξ m 2i /2 −ζ ti Ni,ξ m 1i /2 ⎥ wi (12.88)
⎪
⎪ ⎪
⎪ ⎢ ⎥⎪ ⎪
⎪
⎪ . ⎪
⎪ ⎢ . . . . . ⎥⎪⎪ αi ⎪
⎪
⎪
⎪ ⎪
⎪ ⎢ ⎥⎪⎪
⎩
⎪
⎪
⎪
⎪
⎪ . ⎪ ⎪
⎪
⎣ . . . . . ⎦ βi ⎭
⎩ ⎭
w,ζ 0 0 0 −ti Ni n 2i /2 ti Ni n 1i /2

or,
Σ
{u}ξ ηζ = [G i ]{X i }le (12.89)

Combining Eqs. (12.80), (12.87) and (12.89) yields

310 12 Finite Element Analysis of Shells
⎧ ⎫
⎪ εx ⎪ ⎧ ⎫
⎪
⎪ ⎪
⎪ ⎪ ui ⎪
⎪
⎪ εy ⎪⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎨ ⎬ Σ ⎨ vi ⎪
⎬
εz
= [Bi ] wi (12.90)
⎪ . ⎪
⎪ ⎪ ⎪
⎪ ⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ αi ⎪
⎪
⎪
⎪
⎪ . ⎪
⎪ ⎪
⎩ ⎪
⎪
⎩ ⎪
⎭ βi ⎭
γzx

where
Σ
[Bi ] = [H ][J ]−1 [G i ]
6×9 9×9 9×5 (12.91a)

Σ
and [B] = [Bi ] (12.91b)

The complete strain–displacement matrix [B] is built of as many as 6 × 5 blocks

of [Bi ] as there are nodes in the element.

12.5.4 Stress–Strain Relationship

The stress–strain relationship in cartesian coordinates is given by

{σ }x yz = [D]{ε}x yz (12.92)

The stress–strain relationship in local coordinates is given by

{σ }ξ ηζ = [D]{ε}ξ ηζ (12.93)

or
⎧ ⎫ ⎡ ⎤⎧ ⎫
⎪
⎪ σ1 ⎪⎪ D11 D12 0 0 0 0 ⎪ ε1 ⎪
⎪ ⎪
⎪
⎪ ⎪
⎪ ⎢D ⎪
⎪ε ⎪
⎪
⎪
⎪ σ ⎪
⎪ ⎢ 12 D22 0 0 0 0 ⎥ ⎥⎪⎪
⎪
⎪
⎪
⎨ ⎪ ⎥⎪ ⎪
2 2
⎬ ⎢ ⎨ ⎬
σ3 ⎢ 0 0 0 0 0 0 ⎥ ε3
⎢ ⎥ (12.94)
⎪
⎪ τ12 ⎪
⎪ ⎢ 0 0 0 G 12 0 0 ⎥⎪ γ12 ⎪
⎪
⎪ ⎪
⎪ ⎢ ⎥⎪⎪
⎪
⎪
⎪
⎪
⎪
⎪ τ ⎪ ⎣ 0 0 5G 23 /6 ⎦ ⎪ γ ⎪
⎪ ⎪
⎩
23 ⎪
⎭
0 0 0 ⎪
⎪
⎩
23 ⎪
⎪
⎭
τ31 ξ ηζ 0 0 0 0 0 5G 31 /6 γ31 ξ ηζ

where directions 1 and 2 are tangent to the midsurface and direction 3 is normal to
it. The factor of 5/6 accounts for the parabolic variation of the transverse shear strain
through the thickness. Equation (12.94) has been contrived to make σ 3 = 0.
The relationship of nodal degrees of freedom between the local and the global
axes with the help of Eq. (12.78) is as follows:
12.5 Isoparametric General Shell Element 311
⎧ ⎫
⎧ ⎫ ⎡ ⎤ ⎪ ui ⎪
⎪ ui ⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
1 0 0 0 0 0 ⎪
⎪ vi ⎪ ⎪
⎪
⎪ ⎪ ⎢0 0 ⎥ ⎨ ⎪
⎥ ⎪
⎨ vi ⎪
⎬ ⎢ 1 0 0 0 ⎪ ⎪
⎬
⎢ ⎥ wi
wi = ⎢ 0 0 1 0 0 0 ⎥ (12.95)
⎪
⎪ ⎪
⎪ ⎢ ⎥ ⎪ θxi ⎪
⎪
⎪ αi ⎪
⎪ ⎣0 0 0 −l2i −m 2i −n 2i ⎦ ⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎩β ⎪ ⎭ ⎪
⎪ θ ⎪
⎪
i 0 0 0 l1i m 1i n 1i ⎪
⎩ ⎪
yi
⎭
θzi

or, {X i }le = [T ]{X i }eg (12.96)

12.5.5 Stiffness Matrix of the Shell Element

The strain energy of the shell element is given by

(
U= {ε}Tx yz {σ }x yz d V (12.97)

Using Eqs. (12.90), (12.92) and (12.96), the stiffness matrix of the element in
terms of global displacements is given by

(+1(+1(+1
[ ]T [ ]
[K]e = T [B]T [D][B] T det[J ]dξ dη dζ (12.98)
−1 −1 −1

[ ]
where T is of the size 5n × 6n, [B] 6 × 5n, [D] 6 × 6 and n is the number of
nodes per element. If the material properties are independent of ζ and if small errors
are acceptable, then the thickness direction integration can be done explicitly. In so
doing one discards terms in [J] that depend on ζ under the assumption that these
terms are negligible if the element
[ ] is not sharply curved.
For orthotropic material D is generated in the principal material directions and
[ ]
[D] is to be obtained from D by coordinate transformation as
[ ]
[D] = [Tε ]T D [Tε ] (12.99)

where
⎡ ⎤
l12 m 21 n 21 l1 m 1 m 1n1 n 1 l1
⎢ l2 m 22 n 22 ⎥
⎢ 2 l2 m 2 m 2n2 n 2 l2 ⎥
⎢ 2 ⎥
⎢ l3 m 23 n 23 l3 m 3 m 3n3 n 3 l3 ⎥
{Tε } = ⎢ ⎥ (12.100)
⎢ 2l1l2 2m 1 m 2 2n 1 n 2 l1 m 2 + l2 m 1 m 1 n 2 + m 2 n 1 n 1l2 + n 2 l1 ⎥
⎢ ⎥
⎣ 2l2 l3 2m 2 m 3 2n 2 n 3 l2 m 3 + l3 m 2 m 2 n 3 + m 3 n 2 n 2 l3 + n 3l2 ⎦
2l3l1 2m 3 m 1 2n 3 n 1 l3 m 1 + l3 m 1 m 1 n 3 + m 3 n 1 n 1l3 + n 3l1
312 12 Finite Element Analysis of Shells

where 11 , l 2 , etc. are direction cosines between the two axes system.
This transformation must be carried out at each Gauss point used in generating
[K]e by numerical integration. Direction cosines needed in [Tε ] are the direction
cosines of vectors V 1 , V 2 and V 3 at the Gauss point. In turn these vectors can be
found by shape function interpolation from nodal values
Σ Σ Σ
V1 = Ni V 1i , V 2 = Ni V 2i , V 3 = Ni V 3i (12.101)

where N i are evaluated at the Gauss point in question.

12.6 Vertically Curved Beam Element

A vertically curved circular beam is shown in Fig. 12.9. E, s, I, L are the material
and geometric properties of the beam element. The beam subtends an angle φ at the
centre. The element has two corner nodes. At each node, the unknowns are u, ∂u∂s
,w
and θ.
The displacement functions are
)
u = N1 u 1 + N2 u '1 + N3 u 2 + N4 u '2
(12.102)
w = N1 w1 + N2 θ1 + N3 w2 + N4 θ2

where

Fig. 12.9 Vertically curved z y

circular beam element

t = 3 in
Free edge

3 2
E = 3 × 10 K/in
ν=0
2
g = 0.09 K/ft

CL
R=1
5′

40° Supported by rigid

diaphragm
12.6 Vertically Curved Beam Element 313
⎫
N1 = 1 (− 3ξ 2 + 2ξ 3 ) ⎪
⎪
⎬
N2 = L ξ − 2ξ 2 + ξ 3
(12.103)
N3 = 3ξ(2 − 2ξ 3 ) ⎪
⎪
⎭
N4 = L −ξ + ξ 3

when
s Rθ θ
ξ= = =
L Rφ φ

The strain energy of the beam is given by

( (
EA EI
u= ε ds +2
κ 2 ds (12.104)
2 s L s

where ε and κ are the axial strain and curvature of the middle surface.
)
ε = ∂u∂s
+ wR = u ' + wR
(12.105)
κ = Rs ∂u − ∂∂sw2 = Rs u ' − w ''
2

∂s

Combining Eqs. (12.104) and (12.105)

(L (L
EA ( ' )2 EI ( ' )2
U= u ds + u ds
2 2R 2
0 0
(L (L
EA ' EI
+ u w ds − u ' w '' ds
R R
0 0
(L (L
EA EI ( '' )2
+ w 2 ds + w ds (12.106)
2R 2 2
0 0

The strain energy expression consists of axial, axial-flexural coupling and flexural
behaviours.
Applying Castigliano’s theorem, the element stiffness matrix is given by Yang
and Kim (12.10)

[K ]e = [K ]uu + [K ]uw + [K ]ww (12.107)

⎡6 ⎤
Symmetrical
E A( ) L
α ⎢ 21 2L ⎥
[k]uu = 1+ 2 ⎢
⎣
3 ⎥
⎦ (12.108a)
5 R − L −2 L
6 1 6

2
1
− 6 −2 3
L 1 2L
314 12 Finite Element Analysis of Shells

⎡ ⎤
− 21 Symmetrical
E A⎢
⎢
L
+ α α ⎥
⎥
[k]uw = 10 L 2 (12.108b)
R ⎣ 1
2
L
10
+ α
L
1
2
⎦
α α L α
− α2
2
− 60
L
− L
− L60 − 2 10
+ L
⎡ 13φ 2 ⎤
35
+ 12αL2
Symmetrical
E A⎢
⎢
11Lφ 2
+ 6α L 2 φ2
+ 4α
⎥
⎥
[k]ww = ⎢ 210
9φ 2
2 105
13Lφ 2 13φ 2 ⎥ (12.108c)
L ⎣ 70
+ 12αL 2 4202 2
− 6α
L 35
+ 12α
L2
⎦
2
L φ 11Lφ 2 L 2 φ2
− 420 + L − 140 + 2α − 210 − 6α
13Lφ 6α
L 105
+ 4α
EI
α=
EA

For vertically curved beams where thickness/radius ratio is small, α/R2 ≤ 1 and α/R2
may be neglected for such cases.

12.7 Computer Program for the Finite Element Analysis

of Shallow Shells of General Shape Using Triangular
Element

The computer program in ‘C’ as well as FORTRAN to solve a shallow shell of general
shape by using the triangular element of Sect. 12.4 are given in the attached CD. It
has already been shown that the element is quite sophisticated. The element stiffness
is explicitly evaluated. However, the element is valid for shallow shell formulation
only. The computer program has the provision of catering normal pressure on the
surface as well as the vertical load.
A barrel vault cylindrical as shown in Fig. 12.10 subjected to vertical loading has
been analysed on the basis of a 4 × 4 mesh division of the shell. The input data is
given in Table 12.3. The nodal connectivity and the coordinates are generated by the
computer program. The computer output primarily consists of nodal displacements
and stress resultants, which are given in Table 12.4.
Exercise 12
12.1 Develop a shell element by superimposing the constant strain triangle with non-
compatible bending element having vertex degrees of freedom w, ∂w ∂x
and ∂w∂y
.
12.2 A doubly curved shell is modelled by flat triangular elements such as those
of Eq. (12.2). Is the inter-elemental compatibility maintained of the edge
12.7 Computer Program for the Finite Element Analysis of Shallow Shells … 315

w1 w2
E, A, I, L
s θ2
2
1
u2
u1 θ1 ∂u
∂s
∂u 2

∂s R
1 θ

Fig. 12.10 Barrel vault

Table 12.3 Input data for shell problem

SOLN. OF A DIAPHRAGM SUPPORTED CYLINDRICAL SHELL
CONSIDERING QUARTER SHELL (MESH 2 X 2)
43.2E4 0. 0.25 25. 1.E30 1.E30
2 2 7 1 0
16.06969 25.0
2 1
-0.09
1 1 1 1 0 0 0 1 1 0 1 1 0
2 1 1 0 0 0 0 1 1 0 1 0 0
3 1 1 0 0 0 0 1 1 0 1 0 0
4 1 0 1 0 0 0 0 1 0 0 1 0
7 1 0 1 1 1 0 0 1 1 0 1 0
8 0 0 0 1 1 0 0 0 1 0 1 0
9 0 0 0 1 1 0 0 0 1 0 1 0

displacements?
12.3 An isoparametric curved element is considered for the axisymmetric shell as
shown in Fig. 12.3. Derive the stiffness matrix of the element.

r2–r1

RS
2 ξ
φ

ξ
φ1

1
CL
(a) parent element (b) Curvilinear element
Prob. 12.3.
316 12 Finite Element Analysis of Shells

Table 12.4 Computer output for the shell problem

Shallow Shell FEM using Cowper's Element
emod=432000 pr=0.000000 th=0.250000 rx1=25 ry1=1e+030 rxy1=1e+030
nx=2 ny=2 nbp=7 nlc=1 nl=0
xlen=16.069690 ylen=25.000000
nnod = 9
x Co-ordinates
0.00000e+000 8.03485e+000 1.60697e+001 0.00000e+000 8.03485e+000
1.60697e+001 0.00000e+000 8.03485e+000 1.60697e+001
y co-ordinates
0.00000e+000 0.00000e+000 0.00000e+000 1.25000e+001 1.25000e+001
1.25000e+001 2.50000e+001 2.50000e+001 2.50000e+001
nelm = 8
Nodal connectivity
5 4 1 1 2 5 6 5 2 2 3 6
8 7 4 4 5 8 9 8 5 5 6 9

Iload = 2, Icurve = 1
qo= -9.00000e-002
Node number boundary condition code numbers
1 1 1 1 0 0 0 1 1 0 1 1 0
2 1 1 0 0 0 0 1 1 0 1 0 0
3 1 1 0 0 0 0 1 1 0 1 0 0
4 1 0 1 0 0 0 0 1 0 0 1 0
7 1 0 1 1 1 0 0 1 1 0 1 0
8 0 0 0 1 1 0 0 0 1 0 1 0
9 0 0 0 1 1 0 0 0 1 0 1 0
rx= -4.00000e-002 ry= -1.00000e-030 rxy= 1.00000e-030
nbh = 60
Node Global U, V, W Displacements
1 -2.52771e-025 8.45449e-004 3.76523e-024
2 1.64401e-024 2.31428e-003 -1.03472e-024
3 2.73820e-024 -1.11845e-002 -1.30357e-023
4 -2.01431e-024 7.60297e-004 2.14358e-002
5 1.81474e-003 1.48234e-003 -5.76753e-002
6 4.58028e-002 -7.30827e-003 -2.19612e-001
7 -1.72833e-024 3.84346e-024 2.75567e-002
8 2.62600e-003 3.96076e-024 -7.92146e-002
9 6.33728e-002 -1.14202e-023 -3.06600e-001
Strain Energy = 1.22271e+003
node nx ny nxy mx my mxy
1 -3.692e-020 -6.444e+000 -5.202e-001 -1.129e-022 1.301e-001 -5.839e-022
2 -2.090e-020 -2.590e+000 -5.929e+000 -3.952e-022 1.408e-001 -5.235e-001
3 1.286e-019 -8.413e+000 -8.452e+000 -1.250e-021 7.404e-002 -5.017e-001
4 1.959e+002 1.090e+000 4.296e-001 -1.673e+000 -1.449e-001 -3.743e-022
5 -4.909e+002 9.655e+000 -2.548e+000 -8.242e-001 1.615e-002 -3.450e-001
6 -1.907e+003 -4.774e+001 -4.040e+000 1.385e-001 6.094e-001 -3.551e-001
7 2.429e+002 5.846e+000 -1.511e-019 -1.962e+000 -4.803e-002 3.953e-021
8 -6.901e+002 1.490e+001 2.125e+000 -9.085e-001 2.928e-001 -1.027e-022
9 -2.670e+003 -5.449e+001 -8.700e+000 1.082e-001 9.076e-001 6.127e-023

12.4 Consider a thin-walled cylindrical shell that is symmetrically loaded, but

without axial loads, u-displacement consideration is not necessary. Using
cubic polynomial for w, generate the 4 × 4 element stiffness matrix.
12.5 What difficulties are experienced when flat shell elements meeting at a node
become coplanar? How are the difficulties obviated? Discuss all possibilities.
12.6 If the flat shell element is so constructed as to require only the continuity
of displacements u, v and w at the corner nodes with continuity of the
References and Suggested Readings 317

normal shape being imposed along the element sides, many of the difficulties
encountered with the nodal assembly in global coordinates disappear. How?
12.7 How are you going to extend the analysis of Art. 12.5 to cater axisymmetric
curved thick shell?
12.8 What modifications will you have to perform on the analysis presented in
Art. 12.5 to deal with thick plates?
12.9 Write a computer program in FORTRAN to analyse a shell of general shape
for any arbitrary loading.
12.10 An end of an isoparametric bar element whose geometry is defined by the
position of nodes along its centerline and vectors V 2i and V 3i that span its
rectangular cross-section is shown below.
(a) Write an equation of geometry similar to Eq. (12.69)
(b) Write an equation of displacement analogous to Eq. (12.69)
12.11 (a) How can the transformation [D]' to [D] be made more computationally
efficient?
(b) Rewrite Eq. (12.86) so that [D]' instead of [D] remains in the equation.

References and Suggested Readings

1. V.V. Novozhilov, Theory of Thin Shells (P. Noordhoff, 1959)

2. D.G. Ashwell, R.H. Gallagher (eds.), Finite Elements for Thin Shells and Curved Members
(Wiley, 1976)
3. G. Wempner, Mechanics and finite elements of shells. Appl. Mech. Rev. 42(5) (1989)
4. G.R. Cowper, G.M. Lindberg, M.B. Olson, A shallow finite element of triangular shape. Int.
J. Solids Struct. 6, 1133–1156 (1970)
5. W. Flugge, Stresses in Shells, 2nd edn. (Springer, 1973)
6. H. Kraus, Thin Elastic Shells (Wiley, New York, 1967)
7. S. Dasgupta, D. Sengupta, Horizontally curved isoparametric beam element with or without
elastic foundation including the effect of shear deformation. Comput. Struct. 29(6), 967–973
(1988)
8. G. Sinha, A.H. Sheikh, M. Mukhopadhyay, A new finite element model for the analysis of
arbitrary stiffened shells. Finite Elem. Anal. Des. 12, 241–275 (1992)
9. P.L. Gould, Analysis of Shells and Plates (Springer, 1988)
10. T.Y. Yang, H.W. Kim, Vibration and buckling of shells under initial stress. AIAA J. 11(11),
1525–1531 (1973)
Chapter 13
Semi-analytical and Spline Finite Strip
Method of Analysis of Plate Bending

13.1 Introduction

Many practical structures have regular boundaries, e.g. rectangular and circular
plates. For them, the solution by the finite element analysis may entail a substan-
tial computer time. The finite element analysis of plates requires a two-dimensional
domain to be divided in both directions. As a consequence, a large number of simul-
taneous equations result. For the vibration and stability analyses of plates when the
problem reduces to an eigenvalue problem, this becomes more uneconomical. For
rectangular and circular plates, it was thus felt to develop a method, which is simpler
than the conventional finite element method and which also takes less time for the
computer solution. The finite strip method is a step forward in that direction.
The finite strip method is a simpler version of the finite element method presented
so far. The plate is divided in one direction only, that is, the elements are in the
form of strips. The displacement function is so chosen that it satisfies the boundary
conditions in one direction. The unknowns in the problem are those at the nodal
lines. Thus, a two-dimensional problem is reduced to a one-dimensional one. The
rest of the procedure is the same as the finite element method. It will be seen that the
number of unknowns are significantly less than those needed for the analysis of the
same plate by the finite element method.
The displacement function satisfies the boundary conditions along one direction of
the plate, which converts the method to one that resembles a series solution. Further,
this method is similar to the finite element method. Due to these reasons, the finite
strip method is also termed as a semi-analytical finite element method.
This method has been applied to the analysis of various structures such as plates
with inplane forces, plate bending, folded plates, shells and so on [1]. But this chapter
is limited to the discussion of plate bending problems only.

© The Author(s) 2022 319

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_13
320 13 Semi-analytical and Spline Finite Strip …

13.2 Beam Function

The characteristics function of a beam satisfying different boundary conditions can

be expressed in a general form as

Y = C1 sin(μy/b) + C2 cos (μy/b) + C3 sinh(μy/b) + C4 cosh (μy/b)

(13.1)

where b is the length of the beam.

The coordinate system is shown in Fig. 13.1. Equation (13.1) has been obtained
from the vibration analysis of the beam treating it as a continuous system. Y represents
to some scale deflected shape of the beam corresponding to different nodes. For mth
node, Eq. (13.1) can be written as

Ym = C1 sin(μm y/b) + C2 cos(μm y/b) + C3 sinh (μm y/b) + C4 cosh(μm y/b)

(13.2)

Therefore, the different deflected shapes are controlled by the parameter μ.

If a beam is considered to have both ends simply supported, then the boundary
conditions are as follows:

At y = 0, Y = 0; d 2 Y /dy 2 = 0 (13.3)

At y = bmin Y = 0; d 2 Y /dy 2 = 0 (13.4)

Putting the boundary conditions corresponding to Eq. (13.3) into Eq. (13.1), the
following conditions are obtained.

C2 + C4 = 0 (13.5)

Solving Eq. (13.5), the following values are obtained

C2 − C4 = 0 (13.6)

Substituting the conditions corresponding to Eq. (13.4) into Eq. (13.1), yields
)
C1 sin μ + C3 sinh μ = 0
(13.7)
C1 sin μ + C3 sinh μ = 0

Fig. 13.1 A typical beam y

b

Ym
13.2 Beam Function 321

Equation (13.7) gives

)
C1 sin μ = 0
(13.8)
C3 sinh μ = 0

Second of Eq. (13.8)

C3 = 0 (13.9)

Therefore,

sin μ = 0

or,

μ = rπ (13.10)

r = 0, 1, 2, 3, ……., m, ………..
Therefore, the beam function for a beam simply supported at both ends is
μm y
Ym = sin (13.11)
b
where μm = mπ
m = 1, 2, 3, ………
This is diagrammatically shown in Fig. 13.2
In this way, beam functions for beams having different boundary conditions can
be derived. Other beam functions are given below.
(2) One end simply supported, and the other end clamped
μm y μm y
Ym = sin − αm sinh (13.12)
b b

where μm = 4m4+ 1 π
m = 1, 2, 3, ………

sin μm
αm =
sinh μm

Fig. 13.2 Ym for a simply m=3 m=2

supported beam

m=1
322 13 Semi-analytical and Spline Finite Strip …

(3) Both ends clamped

sin μm y μm y ( μm y μm y )
Ym = − sinh − αm cos − cosh (13.13)
b b b b
2m + 1
μm = π, m = 1, 2, 3, . . . . . . .
2
sin μm − sinh μm
αm =
cos μm − cosh μm

(4) Both ends free

μm y μm y ( μm y μm y )
Ym = sin + sinh − αm cos + cosh (13.14)
b b b b
2m + 1
μm = π, m = 1, 2, 3,
2
sin μm − sinh μm
αm =
cos μm − cosh μm

(5) One end clamped, other end free

μm y μm y ( μm y μm y )
Ym = sin − sinh − αm cos − cosh (13.15)
b b b b

2m − 1
μ1 = 1.875, μ2 = 4.694, μ3 = 7.855 m ≥ 4, μm = π
2
sin μm + sinh μm
αm =
cos μm + cosh μm

(6) One end simply supported and the other end free
μm y μm y
Ym = sin + αm sinh (13.16)
b b
4m − 3
μm = π, m = 2, 3
4
sin μm
αm =
sinh μm
13.2 Beam Function 323

(7) One end clamped and the other end elastically restrained against rotation
μm y μm y ( μm y μm y )
Ym = sin − sinh + αm cos − cosh (13.17)
b b b b
sin μm − sinh μm
where αm = cosh μm − cos μm
.
Different values of μm for various values of β = C R D/b where C R is the restraining
coefficient, D is the flexural srigidity of the beam, are given in Table 13.1.

(8) One end simply supported and the other end elastically restrained against
rotation
μm y μm y
Ym = sin − αm sinh (13.18)
b b
sin μm
where αm = sinh μm
.
Different values of μm for different values of β = C R D/b are given in Table 13.2.

(9) Both edges possessing same degree of elastic restraint against rotation
μm y μm y
Ym = sin − sinh
b b

Table 13.1 Values of μm for one edge clamped and the other edge elastically restrained against
rotation
β m
1 2 3 4 5 6
0.01 4.686 7.782 10.898 14.015 17.134 20.254
0.05 4.547 7.585 10.661 13.751 16.850 19.958
0.10 4.431 7.450 10.522 13.615 16.720 19.834
1.00 4.042 7.134 10.257 13.388 16.523 19.660
1000.00 3.927 7.069 10.211 13.352 16.494 19.635

Table 13.2 Values of μm for one edge simply supported and the other edge elastically restrained
against rotation
β m
1 2 3 4 5 6
0.01 3.890 7.004 10.119 13.256 16.354 19.474
0.05 3.770 6.820 9.891 12.977 16.075 19.180
0.10 3.665 6.688 9.752 12.840 15.943 19.055
0.50 3.367 6.418 9.520 12.640 15.768 18.300
1.00 3.274 6.356 9.475 12.605 15.739 18.876
1000.00 3.142 6.284 9.425 12.567 15.708 18.850
324 13 Semi-analytical and Spline Finite Strip …

Table 13.3 Values of μm for both edges elastically restrained against rotation (the coefficient of
restraint being same at both edges)
β m
1 2 3 4 5 6
0.01 4.642 7.711 10.802 13.895 16.991 20.089
0.05 4.374 7.329 10.339 13.375 16.431 19.592
0.10 4.156 7.069 10.066 13.106 16.172 19.257
0.50 3.577 6.547 9.613 12.712 15.827 18.950
1.00 3.399 6.428 9.525 12.643 15.770 18.902
10.00 3.173 6.399 9.436 12.575 15.715 18.855

( μm y μm y μm y )
+ αm cos − 2μm β sinh − cosh (13.19)
b b b

where αm = cosh μm −sincosμmμ−m +

sinh μm
2β μm sinh μm
.
Different values of μm for different degrees of coefficient of restraint are given in
Table 13.3.
The beam functions for the first 6 cases have been obtained from Ref. [1] and the
next 3 from Ref. [2]. The beam functions are orthogonal and possess the following
properties:

⎫
(b ⎪
⎪
⎪
Ym Yn dy = 0 ⎪
⎪
⎪
⎪
⎬
0 (13.20)
(b ⎪
⎪
⎪
⎪
Ymiv Yn dy = 0 ⎪
⎪
⎪
⎭
0

where m /= n.
( b ''
However, 0 Ym Yn dy = 0 for opposite edges simply supported.

13.3 Model of the Plate

The plate is divided into a number of strips in either x or y direction. Figure 13.3
shows the plate divided in the x-direction. Each strip may have its individual elastic
properties. A typical strip is shown in Fig. 13.4. At each nodal line i and j, there are
two unknowns corresponding to each harmonic number (i.e. corresponding to each
value of m).
13.4 The Displacement Function 325

b d

Fig. 13.3 Plate divided into strips

i j

Fig. 13.4 A typical strip

13.4 The Displacement Function

A displacement function of the form given below is chosen

γ
Σ
w = f m (x) Ym (13.21)
m =1

where f m (x) contains the terms of that of the beam in x direction obtained from the
finite element analysis of beam [Eq. (11.10)].

f m (x) = [N ] {X m }e (13.22)
326 13 Semi-analytical and Spline Finite Strip …

Therefore, Eq. (13.21) becomes

Σ
r
w = [N ]m {X m }e (13.23)
m =1

where
[( )( )( )( )]
3x 2 2x 3 2x 2 x3 3x 2 2x 3 −x 2 x3
[N ]m = 1− 2
+ 3 x −
d
+ 2 2
− 3
d
+ 2 Ym (13.24)
d d d d d d

and
⎧ ⎫
⎪
⎪ Wim ⎪
⎪
⎨ ⎬
θim
{X m }e = Ym (13.25)
⎪
⎪ W ⎪
⎪
⎩ jm ⎭
θ jm

The total displacements along the sides i and j are

⎧ ⎫ ⎧ ⎫
⎪
⎪ Wi ⎪ ⎪ ⎪
⎪ Wim ⎪
⎪
⎨ ⎬ Σ r ⎨ ⎬
θi θim
= Ym (13.26)
⎪
⎪ W ⎪
⎪ ⎪ W ⎪
⎩ ⎭ m =1 ⎪
⎩ jm ⎪
⎭
j
θj θ jm

13.5 Curvature-Nodal Parameter Relationship

The curvatures of the plate strip are given by

⎧ ⎫
∂2w
⎪
⎨ − ∂2x 2 ⎪
⎬
{ε } = − ∂∂ yw2 (13.27)
⎪
⎩ 2 ∂2w ⎪
⎭
∂x ∂y

Performing the necessary differentiation of Eq. (13.23) as required in Eq. (13.27)

Σ
r
{ε} = [B]m {X m }e = [B] {X }e (13.28)
m =1

or,
13.7 Strip Stiffness Matrix 327
⎧ ⎫
⎪
⎪ [X ]1e ⎪
⎪
⎪
⎨ [X ]2e ⎪
⎬
{ ε } = [ [B]1 [B]2 ....... [B]r ] .. (13.29)
⎪
⎪ . ⎪
⎪
⎪
⎩ ⎪
⎭
[X ]r e

where
⎡( 6 )
12x Y
(
4
)
6x
(
12x − 6 Y
) (
2
) ⎤
2 − m d) − d 2 (Ym m − 6x2 Ym
⎢ d( d3 d3 d2 ) (d d ) ( ) ⎥
⎢ 2 3 2 3 3x 2 − 2x 3 Y '' − x 3 − x 2 Y '' ⎥
[B]m = ⎢ − 1− 3x − 2x3 Ym'' − x − 2x x
d + d 2 Ym −
''
m m ⎥
⎢ ( d2 )d ( ) ( d 2 3
)d (
2
d ) d ⎥
⎣ 2 2 2 3 2 ⎦
− 12x
2 + 12x3 Ym' 2 − 8xd −
6x Ym' − 12x2 − 12x Ym' 6x 4x
− d Ym'
d d d2 d d3 d2
(13.30)

13.6 Moment—Curvature Relationship

For the rectangular orthotropic plate

⎧ ⎫ ⎡ ⎤ ⎧ ∂2w ⎫
⎨ Mx ⎬ D x D1 0 ⎪ − ∂x2
⎨ ⎪
⎬
= ⎣ D1 D y 0 ⎦ − ∂∂ yw2
2
{σ } = My
⎩ ⎭ ⎪
⎩ 2 ∂2w ⎪
⎭
Mx y 0 0 Dx y ∂x ∂y

or,

{σ } = [D]{ε} (13.31)

Combining eqns. (13.28) and (13.31), yields

{σ } = [D][B]{X }e (13.32)

or,

Σ
r
{σ } = [D] [B]m {X }me (13.33)
m =1

13.7 Strip Stiffness Matrix

By the use of the principle of virtual work or other variational principles, it can be
shown that the strip stiffness matrix is given by
328 13 Semi-analytical and Spline Finite Strip …
( b ( d
[K ]e = [B]T [D] [B] d x d y (13.34)
0 0
( b ( d
= [ [B]1 [B]2 · · · [B]r ]T [D] [ [B]1 [B]2 · · · [B]r ] d x d y (13.35)
0 0
⎡ ⎤
[B]1T [D] [B]1 [B]1T[D] [B]2 · · · [B]1T [D] [B]r
( b ( d ⎢ [B]T [D] [B]1 [B]2T[D] [B]2 · · · [B]2T [D] [B]r ⎥
⎢ 2 ⎥
= ⎢ .. .. .. ⎥ dx dy
0 0 ⎣ . . . ⎦
[B]r [D] [B]1 [B]r [D] [B]2 · · · [B]r [D] [B]r
T T T

or,
⎡ ⎤
[K ]11e [K ]12e · · · [K ]1r e
⎢ [K ]21e [K ]22e · · · [K ]2r e ⎥
⎢ ⎥
[K ]e = ⎢ . .. .. ⎥ (13.36)
⎣ .. . . ⎦
[K ]r 1e [K ]r 2e · · · [K ]rr e
(b (d
where [K ]mne = 0 0 [B]me T
[D] [B]ne d x d y
For the evaluation of [K]11e [K]12e etc. in Eq. (13.36), the following integral values
are to be obtained
( b ( b ( b ( b
Ym Yn dy, Ym'' Yn'' dy, Ym' Yn' dy and Yn dy
0 0 0 0

The first two integrals are zero for all the beam functions for m = n. However,
the last two integrals simply supported ends. However, even in these cases, these are
assumed to be zero with the result that

[K ]mn = 0 when m /= n (13.37)

Equation (13.36) can then be written as

⎡ ⎤
[K ]11e
⎢ [K ]22e ⎥
⎢ ⎥
⎢ .. ⎥
[K ]e = ⎢
⎢ . ⎥
⎥ (13.38)
⎢ .. ⎥
⎣ . [K ]r e ⎦

The advantages and justification of the assumption of Eq. (13.37) will be discussed
after the loading matrix.
13.8 Loading Matrix 329

13.8 Loading Matrix

All loads must be resolved into the beam function series in the y direction similar to
the displacement function. A load of intensity 1/unit length in the z direction on a
line parallel to y axis is expressed as

q = q1 Y1 + q2 Y2 + . . . . + qr Yr (13.39)

or

Σ
r
q = qm Ym (13.40)
m =1

Multiplying both sides of Eq. (13.39) by Y m and integrating with respect to y from
σ to b and noting the orthogonality property given by Eq. (13.20).
(b
q Ym dy
qm = (0b (13.41)
0 q Ym2 dy

For a uniformly distributed load from σ to b, Eq. (13.41) becomes

(b
q Ym dy
qm = (b
0
(13.42)
0 Ym2 dy

For a concentrated load P at y = c,

P Ym (c)
qm = ( b (13.43)
2
0 Ym dy

The loading matrix for the plate strip is given by (Eq. 11.56),
( b ( b
{P}e = [N ]T q d x dy
0 0
⎧ ⎫
( ( ⎪
⎪ [N ]1T ⎪
⎪ (13.44)
b b ⎨ ⎬
[N ]2T
{P}e = (q1 Y1 + q2 Y2 + qr Yr ) d x d y
⎪
⎪ ⎪
⎪
0 0
⎩ ⎭
[N ]rT
(b
Using the orthogonality relation 0 Ym Yn dy = 0 for m /= n, Eq. (13.44) can
be simplified as
330 13 Semi-analytical and Spline Finite Strip …
⎧ ⎫
⎪
⎪ [N ]1T q1 Y1 ⎪
⎪
( b ( b ⎪
⎨ [N ]T q2 Y2 ⎪⎬
2
{P}e = . dx dy (13.45)
0 ⎪⎪ .. ⎪
⎪
0
⎪
⎩ ⎪
⎭
[N ]rT qr Yr
⎧ ⎫
⎪ {P}1e ⎪
⎪
⎪ ⎪
⎪
⎪
⎪ {P} ⎪ ⎪
⎨ . 2e ⎪
⎪ ⎬
{P}e = .
. (13.46)
⎪
⎪ .. ⎪ ⎪
⎪
⎪ ⎪
⎪
⎪ . ⎪ ⎪
⎪
⎩ ⎭
{P}r e

For simply supported strip subjected to uniformly distributed load q, the following
equation results
⎧ ⎫
( ⎪
⎪ b/2 ⎪⎪
b ⎨ 2 ⎬
b /12
{P}me = q Ym dy (13.47)
⎪
⎪ b/2 ⎪⎪
0
⎩ 2 ⎭
−b /12

13.9 Force Displacement Relationship

Using the well-known force–displacement relation, Eq. (13.36) can be connected to

Eq. (13.46) by the following relation.
⎡ ⎤
[K ]11e [0] ⎧ ⎫ ⎧ ⎫
⎢ .. .. ⎥ ⎪ {X }1e ⎪ ⎪ {P}1e ⎪
⎢ . [K ]22e . ⎥ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎢ ⎥ ⎪
⎪ {X } ⎪
⎪ ⎪
⎪ {P} ⎪
⎪
⎢ .. .. .. ⎥ ⎪
⎨ . 2e ⎪
⎬ ⎪
⎨ . 2e ⎪
⎬
⎢ . . . ⎥ .. ..
⎢ ⎥ = (13.48)
⎢ .. .. .. ⎥
. ⎪ ⎪ ⎪ ⎪
⎢ . . ⎥ ⎪
⎪ .. ⎪
⎪ ⎪
⎪ .. ⎪
⎪
⎢ ⎥ ⎪
⎪ . ⎪
⎪ ⎪
⎪ . ⎪
⎪
⎢ .. .. .. ⎥ ⎪
⎩ ⎪
⎭ ⎪
⎩ ⎪
⎭
⎣ . . . ⎦ {X }r e {P}r e
[0] [K ]rr e

Equation (13.48) is in uncoupled form. Therefore, the solution can be separately

sought for each harmonic term. Thus for harmonic term m, the corresponding eqn. is

[K ]mme {X }me = {P}me (13.49)

13.10 Spline Finite Strip Method of Analysis of Plate Bending 331

''
(b
Table 13.4 Values of integrals Cmn = 0 Ym'' Yn d x (b = 1)
(a) One edge clamped, other edge simply supported, Eq. (13.12)
n m=1 2 3 4 5
1 −5.7572 −2.1424 1.9001 −1.6426 1.4292
2 −2.1424 −21.448 −3.9096 3.8223 −3.583
3 1.9001 −3.9096 −47.091 −5.5583 −5.6446
4 −1.6424 3.8223 −5.5583 −82.459 −7.2169
5 1.4292 −3.583 5.6446 −7.2169 −127.77
(b) Both edges clamped Eq. (13.13)
n m=1 2 3 4 5
1 −12.699 0.23614 9.7041 −0.1129 7.6117
2 0.23614 −45.983 −0.0243 17.133 −0.0115
3 9.7041 −0.0243 −98.911 −0.0019 −24.349
4 0.1129 17.133 −0.0019 −171.59 −0.0136
5 7.6117 −0.0115 24.349 −0.0136 −264.37

Equation (13.49) will give the values of displacements corresponding to mth

harmonic term. Thus when all displacements for r harmonic terms are obtained,
the final displacement will be the value evaluated after summing up as given by
Eq. (13.26)
(b (b
The advantage of assuming 0 Ym Yn dy and 0 Ym'' Yn dy to be zero for m /= n
is now revealed. The greatest advantage lies in uncoupling the Eq. (13.48). There-
fore, the solution can be sought for individual harmonic term and this results in
considerable saving in computation time.
(b
These assumptions are however to be justified. The values of 0 Ym' Yn dy for m
/= n for two functions given by Eqs. (13.12) and (13.13) are given in Table (13.4).
The most dominant terms are when m = n and this may be the reason why results
obtained by ignoring these integrals for m /= n have yielded exceedingly good results.

13.10 Spline Finite Strip Method of Analysis of Plate

Bending

The semi-analytical finite strip method presented so far is indeed an elegant method
for obtaining economic solution, specifically for regular shaped plates. It has been
extensively used for various types of regular plate problems. But the method suffers
from a number of drawbacks. It fails for plates having mixed boundary conditions,
internal cutouts, continuous span, and discrete supports at strip ends. As the method
uses the characteristic beam function in one direction which is continuously differ-
entiable, it imposes limitations on the plate problems involving an abrupt change
332 13 Semi-analytical and Spline Finite Strip …

of properties, concentrated and patch loads, internal supports etc., since the second
or the third derivatives should really be continuous. To overcome the difficulties
mentioned above and to retain the advantages of the finite strip method, the spline
finite strip method is developed [2–4].

13.10.1 The Spline Function

Spline is originally the name of a small flexible wooden strip employed by

draughtsmen as a tool for drawing a continuous smooth curve segment by segment. It
has now been used as a mathematical tool for the solution of engineering problems.
A variety of spline functions are available. The spline function adopted here to
represent displacement is the B3 -spline of equal section length. This is also known
as the cubic B3 -spline function. It is given by (see Fig. 13.5).

Σ
m +1
y = αi φi (13.50)
i =−1

in which each local B3 spline φ i has non-zero values over four successive sections
with section knot x = x i as the centre, is defined as
⎧ ⎫
⎪
⎪ 0, x < xi−2 ⎪
⎪
⎪
⎪( )3 ⎪
⎪
⎪
⎪ x − xi − 2 , xi − 2 ≤ x ≤ xi−1 ⎪
⎪
⎪
⎪ ( ) ( )2 ( )3 ⎪
⎪
⎪
⎪ ⎪
⎪
⎪ 2
⎨ h + 3h x − xi−1 + 3h x − xi−1 − 3 x − xi−1 , xi−1 ≤ x ≤ xi ⎪
⎬
φi = 1/6h 3 (13.51)
⎪
⎪ ( ) ( )2 ( )3 ⎪
⎪
⎪
⎪ h 3 + 3h 2 xi+1 − x + 3h xi−1 − x − 3 xi−1 − x , xi ≤ x ≤ xi + 1 ⎪
⎪
⎪
⎪ ( ) ⎪
⎪
⎪
⎪ 3 ⎪
⎪
⎪
⎪ xi + 2 − x , xi+1 ≤ x ≤ xi+2 ⎪
⎪
⎪
⎩ ⎪
⎭
0, xi+2 < x

The use of B3 -spline offers certain distinct advantages when compared with the
conventional finite element method and the semi-analytical finite strip method.
(1) It is computationally more efficient than the finite element method.

ϕi
ϕ–1 ϕ0 ϕ1 ϕ2 . . . ϕm–2 ϕm–1 ϕm ϕm+1

xi– 2 xi–1 xi xi+1 xi+2

h h h h h h h h h h h h h

m section
(a) (b)

Fig. 13.5 a Typical B3 spline; b basis of B3 -spline expression

13.10 Spline Finite Strip Method of Analysis of Plate Bending 333

Table 13.5 Amended scheme for boundary local splines

Boundary condition Amended local splines
φ −1 φ −0 φ1
Free end y(x 0 ) /= 0 φ −1 φ −0 − φ1 −
'
y (x 0 ) /= 0 4φ −1 1
φ0 + φ−1
2
Simply y(x 0 ) /= 0 Eliminated φ −0 − φ1 −
'
supported y (x 0 ) /= 0 4φ −1 1
φ0 + φ−1
2
end
Clamped y(x 0 ) = 0 Eliminated Eliminated φ1 −
end 1
φ0 + φ−1
2

Sliding y(x 0 ) /= c Eliminated φ0 ϕ1 −

clamped 1
ϕ0 + ϕ−1
2
end

When using B3 splines as displacement functions, continuity is ensured up to the

second order (C2 -continuity). However, to achieve the same continuity conditions for
the conventional finite element, it is necessary to have three times as many unknowns
at the nodes.
(2) It is more flexible than the semi-analytical finite strip method in the boundary
conditions treatment.
Only several local splines around the boundary point need to be amended to fit any
specified boundary conditions.
The B3 -spline representations [φ] can be easily managed to adapt to various
prescribed. boundary condition. Due to the localization of the B3 -splines, only three
boundary local splines have to be amended, i.e.,
[ ]
[Φ] = φ −1 , φ 0 , φ 1 , φ2 , φ3 , φ4 ............... , φm−3 , φ m − 1 , φ m , φ m + 1
(13.52)

in which the φ i ’s are the standard local splines and the φ i ’s are the amended local
splines. The amended scheme is given in Table 13.5. It may be mentioned that there
are other possibilities for the amended scheme. The reader may check the amended
spline by substituting φi ’s from Eq. (13.51).

13.10.2 Displacement Functions

Let us consider the formulation of rectangular plates only. A typical plate strip is
shown in Fig. 13.6, which is rectangular. The width of the strip is b and length is a,
which is divided into m sections, a knot is placed on each section.
334 13 Semi-analytical and Spline Finite Strip …

Section-knots
b

y
d θ
Typical strip

t z(w)
x
h h h h h
m sections

Fig. 13.6 Typical strip

Let [φ 1 ] and [φ 3 ] be B3 -spline representations for the displacements w of the nodal

lines i and j respectively (m-sections) and [φ 2 ] and [φ 4 ] be B3 -spline representations
for the rotations θ of nodal lines i and j, respectively (m sections), {wi } and {wj }
be the displacement parameter vectors [wi − 1 , wi0 , wi1 , wi2 , ….., wim-1 , wim , wim+1 ]T
and [wj−1 , wj0 , wj1 , wj2 , …….., wjm-1 , wjm , wjm+1 ]T corresponding to φ 1 and φ 3 ,
respectively, and {θ i } and {θ j } be the rotation parameter vectors [θ i-1 , θ i0 , θ i1 , θ i2 , ….,
θ im −1 , θ im , θ im+1 ]T and [θ j-1 , θ j0 , θ j1 , θ j2 , …….., θ jm −1 , θ jm , θ jm+1 ]T corresponding to
φ 2 and φ 4 respectively. Then, with a cubic polynomial interpolation in the x-direction,
the displacement function for the strip will be.
⎡ ⎤ ⎧ ⎫
[φ1 ] 0 ⎪
⎪ {wi } ⎪
⎪
⎢ [φ2 ] ⎥ ⎨ {θi } ⎬
{ f } = w = [N1 , N2 , N3 , N4 ] ⎢
⎣
⎥ { }
⎦ ⎪ wj ⎪
[φ3 ] ⎪
⎩{ }⎪ ⎭
0 [φ4 ] θj
(13.53)

where
( ) ( ) ( )
N1 = 1 − 3x 2 + 2x 3 , N2 = x 1 − 2x + x 2 , N3 = 3x 2 − 2x 3 , N4 = x x 2 − x
(13.54)

and x = x/d. In concise/matrix form,

{
f } = [N ][φ]{δ} (13.55)

13.10.3 Strain–Displacement Relationship

The strain–displacement relationship is

13.10 Spline Finite Strip Method of Analysis of Plate Bending 335
⎧ ⎫ ⎡ [ '' ] ⎤ ⎡ ⎤
∂2w
⎪
⎨ − ∂2x 2 ⎪
⎬ − N 0 0 [[Φ] ]
{ε} = − ∂∂ yw2 = ⎣ 0 −[N ] 0 ⎦ ⎣ Φ'' ⎦ {δ} (13.56)
⎪
⎩ 2 ∂2w ⎪
⎭ [ ]
∂x ∂y
0 2 N' Φ'

where,
[ ]
[ '] ∂ 1( ) ( ) 1( )
N = [N ] = − 6x + 6x 2 , 1 − 4x + 3x 2 , 6x − 6x 2 , (3x − 2x) , (13.57)
∂x d d
[ ]
[ '' ] ∂2 −6 −2 6 −2
N = [N ] = (1 − 2x), (2 − 3x), 2 (1 − 2x), (1 − 3x) (13.58)
∂x2 d2 d d d

13.10.4 Stiffness Matrix

The stiffness matrix [K] for an orthotropic plate strip may be worked out by hand
and expressed in an explicit form as follows,
( (
[ ]T [ ]T ]
[K ] = [B] [D] [B] d (ar ea) =
T
[Φ]T , Φ'' , Φ'
⎡ [ ] ⎤ ⎡ ⎤
T
− N '' 0 0 Dx D1 0
⎢ ⎥ ⎣ D1 D y
× ⎣ − [N ]T 0 ⎦
[ ' ]T ⎦
0 0
0 0 2 N 0 0 Dx y
⎡ [ '' ] ⎤ ⎡ ⎤
− N 0 0 [[Φ]]
× ⎣ 0 − [N ] 0 ⎦ ⎣ Φ'' ⎦ d (ar ea)
[ ] [ ']
0 0 2 N' Φ
⎡ ⎤ ⎡ ⎤
( b[ ] [C1 ] [C3 ] 0 [Φ]
[ ] [ ] [ ]
= [Φ]T , Φ'' , Φ'
T T
⎣ [C2 ] [C4 ] 0 ⎦ ⎣ [Φ''] ⎦ dy
0
0 0 [C5 ] Φ'
( b
{ [ ] [ ]T
= [Φ]T [C1 ] [Φ] + [Φ]T [C3 ] Φ'' + Φ'' [C2 ] [Φ]
{[ ]
0
[ ] [ ]T [ ]}
+ Φ'' [C4 ] Φ'' + Φ' [C5 ] Φ' dy
T
(13.62)

where,
336 13 Semi-analytical and Spline Finite Strip …

( (k) (l)
Table 13.6 Values of cross-segmental integrations of two standard B3 -splines, φi φ j dy

φ (1)
Sij Section Section Section Section
φ (2) L1 L2 L3 L4
Section
L4 1 60 129 20
Section
L3 60 933 1188 129
Section × 1
140 × h
36
L2 129 1188 933 60
Section
L1 20 129 60 1

( d [ ]T [ ]
[C1 ] = N '' Dx N '' d x
0
( d [ ]T
[C2 ] = N '' D1 [N ]d x
0
( d [ ]
[C3 ] = [N ]T D1 N '' d x (13.63)
0
( d
[C4 ] = [N ]T D y [N ] d x
0
( d[ ]T [ ]
[C5 ] = N' Dx y N ' d x
0

where Dxy , Dy etc. are the usual orthotropic plate constants.

The integrations in Eq. (13.63) can be worked out explicitly since only
simple polynomials are involved while the integrations for the coupling matrices
( b (k) (l)
0 φi φ dy in Eq. (13.62) may be reduced to summation operations of the inte-
( b j(k) (l)
grals 0 φi φ j dy over a typical segment. Such segmental integrals have also been
tabulated explicitly, which are given in Tables 13.6, 13.7, 13.8, 13.9 and 13.10.
The explicit value of the stiffness matrix is given in Table 13.11 [3].

13.10.5 The Loading Matrix

The loading matrix of the strip can be obtained by combining Eq. (11.56) with
Eq. (13.55) and is given by
13.11 Computer Program for the Spline Finite … 337

Table 13.7 Values of cross-segmental integrations of standard B3 -spline and its first derivative,
( ' dy
φ(1) φ(2)

φ (1)
S ij Section Section Section Section
φ (2) L1 L2 L3 L4
Section
L4 −1 −38 −71 −10
Section
L3 −18 −183 −150 −9
Section × 1
20 × 1
36
L2 9 150 183 18
Section
L1 10 71 38 1

Table 13.8 Values of cross-segmental integrations of first derivatives of standard B3 -splines,

( ' ' dy
φ(1) φ(2)
'
φ(1)
Sij Section Section Section Section
'
φ(2) L1 L2 L3 L4
Section
L4 −3 −36 21 18
Section
L3 −36 −87 102 21
Section × 1
10 × 1
36h
L2 21 102 −87 −36
Section
L1 18 21 −36 −3

( b ( a
{P}e = [φ]T [N ]T q d x dy (13.64)
0 0

13.11 Computer Program for the Spline Finite Strip

Method of Analysis of Plates in Bending

The computer programs in C and FORTRAN for the solution of the plate bending
problem by the spline finite strip method are given in the attached CD. The steps
338 13 Semi-analytical and Spline Finite Strip …

Table 13.9 Values of cross-segmental integrations of standard B3 -spline and its second derivative,
( '' dy
φ(1) φ(2)

φ (1)
Sij Section Section Section Section
''
φ(2) L1 L2 L3 L4
Section
L4 3 66 99 12
Section
L3 6 −33 −132 −21
Section × 1
10 × 1
36h
L2 −21 −132 −33 6
Section
L1 12 99 66 3

Table 13.10 Values of cross-segmental integrations of the second derivatives of standard B3 -spline,
( '' ''
ϕ(1) ϕ(2) dy
''
φ(1)
S ij Section Section Section Section
''
φ(2) L1 L2 L3 L4
Section
L4 6 0 −18 12
Section
L3 0 −18 36 −18
Section × 1
36h 3
L2 −18 36 −18 0
Section
L1 12 −18 0 6

have been explained with comment statements. The stiffness matrix of a strip has
been evaluated explicitly.
A simply supported square plate under uniform load has been solved with the help
of the computer program. Only a quarter of a plate has been analyzed and symmetry
conditions have been imposed at the symmetric edges. Input of the problem has been
shown in Table 13.12 while the output has been presented in Table 13.13.

( a ( a ( a
I1i j = tφiT φ j dy, I2i j = tφinT φ j dy, I3i j = tφiT φ ''j dy,
0 0 0
13.11 Computer Program for the Spline Finite … 339

Table 13.11 Bending stiffness matrix of a rectangular strip

Table 13.12 Input data for the spline problem

( a ( a
I4i j = tφi''T φ ''j dy, I5i j = tφi'T φ 'j dy
0 0

Exercise 13
13.1 Derive explicitly the stiffness matrix of a plate having all edges clamped, for
harmonic number 1. Consider the plate as isotropic. The following are the
values of different integrals.
( b ( b ( b
am = Ym2 dy, cm = Ym'' Ym dy, Sm = Ym dy
0 0 0
340 13 Semi-analytical and Spline Finite Strip …

Table 13.13 Computer output for plate-bending by spline finite strip method
xl = 0.500 yl = 0.500 zl = 0.010 emod = 10920000.000
pr = 0.300 udl = 1.000
nelm =8 nint =8 nbdl =2 nbdr =3 nbdu =2 nbdd =3 npl =0 nsp =1
xsp = 5.00000e-001 ysp = 5.00000e-001
OVERALL STIFFNESS MATRIX
OVERALL LOADING MATRIX
THE DISPLACEMENT MATRIX
-1.76404e-034 9.27327e-022 3.73065e-021 6.29547e-021 8.12018e-021
9.40646e-021 1.03057e-020 1.09018e-020 9.72081e-021 1.13534e-020
3.13306e-035 1.36833e-019 1.45069e-003 2.89110e-003 5.56194e-003
7.92384e-003 9.91410e-003 1.14934e-002 1.26364e-002 1.33277e-002
2.02228e-002 4.19290e-019 1.25896e-020 8.92288e-005 1.78135e-004
3.43359e-004 4.89642e-004 6.13011e-004 7.10950e-004 7.81850e-004
8.24736e-004 1.25145e-003 4.42778e-020 9.52250e-020 1.38857e-003
2.77567e-003 5.36569e-003 7.66402e-003 9.60543e-003 1.11480e-002
1.22653e-002 1.29414e-002 1.96383e-002 4.16529e-019 2.05274e-020
1.72203e-004 3.44033e-004 6.64902e-004 9.49829e-004 1.19060e-003
1.38198e-003 1.52062e-003 1.60452e-003 2.43486e-003 8.48871e-020
6.16039e-020 1.25823e-003 2.51605e-003 4.88794e-003 7.00952e-003
8.81118e-003 1.02477e-002 1.12906e-002 1.19224e-002 1.80952e-002
3.63007e-019 2.59240e-020 2.45684e-004 4.90979e-004 9.50828e-004
1.36053e-003 1.70758e-003 1.98386e-003 2.18424e-003 2.30557e-003
3.49897e-003 1.19746e-019 4.32815e-020 1.08798e-003 2.17578e-003
4.23952e-003 6.09899e-003 7.68745e-003 8.95949e-003 9.88558e-003
1.04477e-002 1.58599e-002 3.06195e-019 2.97446e-020 3.07675e-004
6.14954e-004 1.19262e-003 1.70877e-003 2.14698e-003 2.49642e-003
2.75013e-003 2.90387e-003 4.40728e-003 1.48558e-019 3.06854e-020
8.92399e-004 1.78471e-003 3.48379e-003 5.02284e-003 6.34425e-003
7.40672e-003 8.18251e-003 8.65424e-003 1.31396e-002 2.46962e-019
3.24228e-020 3.56897e-004 7.13393e-004 1.38489e-003 1.98619e-003
2.49764e-003 2.90607e-003 3.20290e-003 3.38288e-003 5.13461e-003
1.71115e-019 2.10871e-020 6.80567e-004 1.36108e-003 2.65980e-003
3.84042e-003 4.85790e-003 5.67868e-003 6.27946e-003 6.64537e-003
1.00909e-002 1.86154e-019 3.41996e-020 3.92535e-004 7.84667e-004
1.52422e-003 2.18745e-003 2.75236e-003 3.20395e-003 3.53241e-003
3.73166e-003 5.66424e-003 1.87291e-019 1.32361e-020 4.58553e-004
9.17080e-004 1.79333e-003 2.59170e-003 3.28149e-003 3.83920e-003
4.24817e-003 4.49756e-003 6.83011e-003 1.24451e-019 3.52158e-020
4.14094e-004 8.27783e-004 1.60855e-003 2.30935e-003 2.90674e-003
3.38462e-003 3.73236e-003 3.94337e-003 5.98577e-003 1.97016e-019
6.38704e-021 2.30630e-004 4.61247e-004 9.02281e-004 1.30461e-003
1.65273e-003 1.93457e-003 2.14146e-003 2.26772e-003 3.44401e-003
6.23087e-020 3.55466e-020 4.21308e-004 8.42211e-004 1.63677e-003
2.35017e-003 2.95845e-003 3.44515e-003 3.79937e-003 4.01434e-003
6.09354e-003 2.00260e-019 -6.21528e-034 2.14787e-019 4.01747e-019
6.08060e-019 8.51322e-019 1.05241e-018 1.20987e-018 1.32280e-018
1.43782e-018 1.41336e-018 4.39539e-035
Values of deflection w and moments Mx, My
0.00406236 0.0479514 0.0479506

( b ( b ( b
em = Ym'2 dy, fm = Ym Ym' dy, dm = Ym' Ym'' dy
0 0 0
( b
gm = Ymn2 dy
0
References and Suggested Readings 341

13.2 Determine the loading matrix corresponding to Eq. (13.41) for a load varying
linearly at y = 0 to q0 at y = b.
13.3 Based on the derivation of Prob. 13.1, determine the deflection for a square
plate dividing it into two equal strips for a half plate and subjected to hydrostatic
loading for the harmonic number 1.
13.4 Check, the orthogonality relationship for functions (13.14) and (13.18).
13.5 Deflection and bending moment at a point for a particular plate for different
harmonic numbers are given below.

m 1 2 3 4 5 6 7
C0 0.00523 0 0.00032 0 0.00012 0 0.00001
Mn −0.11751 0 −0.0142 0 −0.00439 0 −0.0018

Find out the deflection and moment at that point

13.6 Derive the stiffness matrix of a skew strip.

References and Suggested Readings

1. Y.K Cheung, Finite Strip Method in Structural Analysis (Pergamon Press, 1976).
2. Y.K Cheung, S.C. Fan, C.Q. Wu, Spline finite strip in structural analysis, Proceedings of the
International Conference on Finite Element, Shaighai, China, 1982, pp. 704–709.
3. S.C. Fan, Y.K. Cheung, Flexural vibration of rectangular plates with complex support conditions.
J. Sound Vib. 93(1), 181–194 (1984)
4. W.Y. Li, Y.K. Cheung, L.G. Tham, Spline finite strip analysis of general plates. J. Eng. Mech.
Div. Proc. ASCE 112, 43–54 (1986)
Chapter 14
Dynamic and Instability Analyses
by the Finite Element Method

14.1 Introduction

The finite element method which has been presented so far for the static analysis can
be extended without much difficulty to the dynamic and the elastic instability analysis
of structures. The procedure is similar. The mass matrix for dynamic analysis and
the geometric matrix for instability analysis is to be determined. For more details
than presented in this chapter, the reader can refer to relevant publications [1–10].

14.2 Dynamic Analysis

By the application of the finite element technique, the continuous structure reduces
to a system having multiple degrees of freedom. Dynamic analysis is of two types—
free vibration analysis and response of the structure to forced vibrations; damping
may or may not be considered depending on its importance. In forced vibrations,
the external force that acts is a function of time. Depending on the nature of time
variation of this force, the problem can be categorized into four types: harmonic,
periodic, transient and random. The present discussion on vibration is limited to free
vibration analysis only.

14.2.1 Torsional Vibration of Shafts

A shaft element of Fig. 14.1 has 2 nodes at its two ends. The unknown displacements
at each end are the angles of twist φ 1 and φ 2 . The displacement function is given by

f = a1 + a2 x (14.1)

© The Author(s) 2022 343

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_14
344 14 Dynamic and Instability Analyses by the Finite Element Method

Fig. 14.1 Prismatic element

1 2
T1,φ1 T2,φ2
L

or,
( )
[ ] α1
f =φ= 1x (14.2)
α2

The nodal values are

( ) [ ]( )
φ1 10 α1
=
φ2 1L α2

or,
( ) [ ]−1 ( )
α1 10 φ1
= (14.3)
α2 1L φ2

Combining Eqs. (14.2) and (14.3) yields

[ ]−1 ( )
10 φ1
f = φ = [1 x ]
1L φ2

or,

f = φ = [N ] {X }e (14.4)

where
[( x) x]
[N ] = 1− (14.5)
L L
General derivation for the static analysis still holds good. But two additional
forces are created. The mass of the vibrating structure induces an additional force.
The inertia force is replaced by its static equivalent which is

− I φ̈ in the present case

where I is the mass moment of inertia per unit length. From Eq. (14.4), the following
equation can be written as
{ }
φ̈ = [N ] Ẍ e (14.6)
14.2 Dynamic Analysis 345

Equivalent nodal forces corresponding to the inertia force due to the mass of the
structure can be written from Eq. (11.24) as

(L
{ }
{P}m
e =− [N ]T I [N ] d x Ẍ e (14.7)
o

or,
{ }
{P}m
e = −[M]e Ẍ e (14.8)

where

(L
[M]e = [N ]T I [N ] d x (14.9)
o

Substituting the values of [N] from Eq. (14.8) into Eq. (14.9) yields
[ ]
I L 2 −1
[M]e = (14.10)
6 −1 2

[M]e is referred to as the mass matrix. If [M]e is computed according to Eq. (14.9),
it is referred to as a consistent mass matrix.
In addition to the inertia of the mass, the other additional force that may be
considered in the vibration analysis is the damping force. Similar to the derivation
presented for the mass matrix, the damping matrix of the element can be shown to
be
(
[C]e = [N ]T μ [N ] d x (14.11)

where μ is some numerical value of the viscous damping. Generation of the damping
matrix creates difficulties, mainly because of the damping mechanism and the
damping levels in the structures. As such it will not be discussed here.
Now, if all the elements are assembled, the final equations will be
{ } { }
[M] Ẍ + [C] Ẋ + [K] {X } = {P} (14.12)

If free undamped vibration is considered, then Eq. (14.12) becomes

{ }
[M] Ẍ + [K] {X } = {0} (14.13)

Assuming,
346 14 Dynamic and Instability Analyses by the Finite Element Method

i pt
{X } = a e {ψ} (14.14)

So,
{ }
Ẍ = − ap 2 ei pt {ψ} (14.15)

Combining Eqs. (14.13–14.15) results

− p 2 [M] {ψ} + [K ] {ψ} = {0}

or,

[K ]−1 [M]{ψ} = ω2 [I ]{ψ} (14.16)

where ω2 = 1/p2 and [I] is an identity matrix.

Equation (14.16) can be written as
[ ]
D {ψ} = ω2 [I ] {ψ} (14.17)

where
[ ]
D = [K]−1 [M] (14.18)

Equation (14.17) is a typical eigenvalue problem. There are various techniques

available for their solution (14.1). A non-trivial solution of Eq. (14.17) will be
|[ ] |
det | D − ω2 [I ]| = 0 (14.19)

14.2.2 An Example

If a uniform shaft having one end fixed, the other end free is divided into two equal
elements and the node numbers are as shown in Fig. 14.2, then the complete mass
matrix becomes
[ ]
IL 21
[M] =
12 1 4

and the complete stiffness matrix becomes

[ ]
2G J 1 −1
[K] =
L −1 2
14.2 Dynamic Analysis 347

T2,φ2
I,G,J
T1,φ1

L/2 L/2

Fig. 14.2 Example of fixed-free shaft

Therefore,
[ ] [ ] [ ]
[ ] L 21 IL 21 I L2 56
D = =
2G J 11 12 14 24G J 3 5

2
Equation (14.19) then becomes after putting d = 24G
IL
J
.
Let
| |
| 5d − ω2 6d ||
| =0
| 3d 5d − ω2 |

or,
( )2
5d − ω2 = 18d 2

or,

5d−ω2 = ±4.2426d

or,

ω2 = 0.7574d, 9.2426d

Therefore, two values of the natural frequencies squared are

GJ GJ
p 2 = 2.597 and 31.687
I L2 I L2

14.2.3 Flexural Vibration of Beams

The consistent mass matrix is given by

(
[M] = [N ]T ρ A [N ] d x (14.20)
348 14 Dynamic and Instability Analyses by the Finite Element Method

Fig. 14.3 Lumped mass of a M1 M1 M1

beam 1 2 3

a a a a

where ρ is the mass density of the material and A is the cross-sectional area of
the uniform beam. Substituting [N] from Eq. (14.20) and performing the necessary
integration, Eq. (14.20) becomes
⎡ ⎤
156
ρ AL ⎢ 2
⎢ 22L 4L symmetrical
⎥
⎥
[M]e = ⎣ ⎦ (14.21)
420 54 13L 156
−13L −3L 2
−22L 4L 2

Formation of the mass matrix in another form is also popular. The uniform beam
having cross-sectional area A has been divided into four equal parts and the masses are
lumped at the 5 stations—two end stations being supported, have not been numbered
(Fig. 14.3). The masses between the stations have been lumped at the stations. Thus,
M 1 = ρAa. The mass matrix in this case is
⎡ ⎤
M1 0 0
[M]e = ⎣ 0 M1 0 ⎦ (14.22)
0 0 M1

The elegance of Eq. (14.22) lies in the fact that [M]e is a diagonal matrix, and this
leads to computational advantages.
For flexural vibrations of bar elements, it has been found that the consistent mass
matrix formulation improves the results significantly. However, there is not much
to gain from the consistent mass matrix formulation of torsional or longitudinal
vibration problems.

14.2.4 In-Plane Vibration of Plates

The general expression for consistent mass matrix is

(
[M]e = [N ]T ρ [N ] dv (14.23)

ρ is the mass density of the material of the structure.

For a plate of constant thickness t subjected to in-plane forces, Eq. (14.23)
becomes
14.2 Dynamic Analysis 349
(
[M]e = ρt [N ]T [I ] [N ] d A (14.24)

For a constant strain triangle, substituting [N] from Eq. (14.10) into Eq. (14.24)
performing necessary integration yields
⎡ ⎤
2
⎢0 ⎥
⎢ 2 symmetrical ⎥
tΔ ⎢
⎢1 0 2
⎥
⎥
[M]e = ρ ⎢ ⎥ (14.25)
12 ⎢ 0 1 0 2 ···⎥
⎢ ⎥
⎣1 0 1 0 2 ⎦
0 1 0 1 0 2

14.2.5 Flexural Vibration of Plates

For the non-conforming plate bending element of Sect. 11.3, the shape function is
related as follows:

[N ] = [C] [A]−1 (14.26)

Therefore, the consistent mass matrix is

(
[M]e = ρt [N ]T [I ] [N ] d A (14.27)
A

Substituting [N] from Eq. (14.26) into Eq. (14.27) yields

(
( −1 )T
[M]e = ρt [A] [C]T [C] [A]−1 d A (14.28)
A

Equation (14.28) has been explicitly evaluated and the result is given below
350 14 Dynamic and Instability Analyses by the Finite Element Method
⎡ ⎤
3454
⎢ ⎥
⎢ −461 80 ⎥
⎢ ⎥
⎢ 461 −63 80 symmetrical ⎥
⎢ ⎥
⎢ ⎥
⎢ 1226 −274 199 3454 ⎥
⎢ ⎥
⎢ 274
⎢ −60 42 461 80 λ = ρabt
6300
⎥
⎥
⎢ ⎥
⎢ 199 −42 40 461 63 80 ⎥
[M]e = λ ⎢
⎢ 1226
⎥
⎥
(14.29)
⎢ −199 274 394 116 116 3454 ⎥
⎢ −199 −42 −116 −30 −28 −461 ⎥
⎢ 40 80 ⎥
⎢ ⎥
⎢ −274 42 −60 −116 −28 −30 −461 63 80 ⎥
⎢ ⎥
⎢ 394 −116 −274 −199 3454 ⎥
⎢ 116 1226 199 274 1226 ⎥
⎢ ⎥
⎣ 116 −30 28 199 40 42 274 −60 −42 461 80 ⎦
−116 28 −30 −274 −42 −60 −199 42 40 −461 −63 80

14.2.5.1 Vibration of Isoparametric Bending Element

The displacements of the plate are given by three quantities w, θ x , θ y . The nodal
accelerations are
{ }T { }
f¨r = ẅr θ̈xr θ̈ yr (14.30)

The inertia forces are produced due to the accelerations given by Eq. (14.30).
The lateral inertia forces—ρt ẅ result from transverse acceleration ẅ and ρ is the
mass density of the plate.( The rotational
) ) components θ̈x and θ̈ y produce
acceleration
(
rotary inertia couples— ρt 3 /12 θ̈x and— ρt 3 /12 θ̈ y respectively where t is the
thickness of the plate.
From Eq. (14.24), the work done by nodal forces is
(
{P}e = − [N ]T {q} d x (14.31)

where {q} are the inertia forces. They are given by

⎧ ⎫ ⎡ ⎤⎧ ⎫
⎪
⎨ ρt3ẅ ⎪ ⎬ ρt 0 0 ⎨ ẅ ⎬
⎢ ⎥
{q} = ρt12 θ̈x = ⎣ 0 ρt12 0 ⎦ θ̈x
3

⎪
⎩ ρt 3 θ̈ ⎪
⎭ ⎩ ⎭
0 0 ρt12
3

12 y
θ̈ y

or,
{ }
{q} = [a] f¨ (14.32)

Again from Eq. (11.99), the following relation is known

{ f } = [N ] {X }e (14.33)

Differentiating Eq. (14.33) twice with respect to time yields

14.3 Elastic Instability Analysis 351
{ } { }
f¨ = [N ] Ẍ e (14.34)

Combining Eqs. (14.32), (14.34) and (14.31)

(
{P}e = [N ]T [a] [N ] d A (14.35)
A

From the vibration analysis of simply supported plate, it has been found that there
was not much to gain by using a consistent mass formulation of Eq. (14.35) over
lumped mass matrix.

14.3 Elastic Instability Analysis

The analysis presented here determines the intensity of the critical load of the elastic
structure when bucking takes place. This load is important in design, as it indicates
initiation of buckling. The treatment presented here is similar to that of the dynamic
analysis. Just as the additional matrix required for the dynamic analysis over conven-
tional static analysis is the mass matrix and for the elastic instability analysis, it is the
geometric stiffness matrix. The computer programs for static analysis require only
some modifications to be applied to both dynamic analysis and elastic instability
analysis of structures.

14.3.1 Column Instability Analysis

A uniform prismatic element is shown in Fig. 14.4. The member can deform by axial
and flexural actions. Based on the assumptions made for flexural members, the strain
is related to the displacement as follows:
( ) ( )2
du d 2w 1 dw
εx = −z + (14.36)
dx dx2 2 dx

Fig. 14.4 Prismatic element X1 X3

X2
P P X
X4
L

Z
352 14 Dynamic and Instability Analyses by the Finite Element Method

(a) Undeformed stage (b) Deformed stage (c)

Fig. 14.5 Movement of the ends due to P

where u and w are the displacements along x- and z-axes. The first term of Eq. (14.36)
is the axial strain component, the second term is the flexural strain component. The
third term indicates the coupling of the flexural and axial components and the basis
of this term is explained below.
At the instant of buckling, two ends of the column move by, say, δx (Fig. 14.5c).
If the length of the curve between x and x + dx in the displaced position is ds, then
[ ( )2 ]1/2
√ dw
ds = dx2 + dw 2 = 1+ dx (14.37)
dx

For small deflections and at the instant of buckling, dw

dx
is small compared to unity.
Therefore, neglecting higher order terms on expansion of Eq. (14.37) yields
[ ( )2 ]
1 dw
ds = 1 + dx
2 dx

or,
( )2
1 dw
ds − d x = dx (14.38)
2 dx

The total strain energy of the element is given by

(
1
U= E εx2 dv (14.39)
2
v

Substituting εx from Eq. (14.36) into Eq. (14.39) yields

( ( [( )2 ( )2 ( )4 ( )( )
E du d 2w 1 dw du d 2w
U= + z2 + − 2z
2 dx dx2 4 dx dx dx2
L A
( )2 ( )2 ( )( )2 ]
d w 2
dw du dw
−z + d A dx (14.40)
dx2 dx dx dx

Let us note the following geometric properties

14.3 Elastic Instability Analysis 353
( ( (
d A = A, zd A = 0 and z2d A = I (14.41)
A A A

On replacing the sectional properties into Eq. (14.40) results in

( [ ( )2 ( )2 ( )( )2 ( )4 ]
1 du d 2w du dw EA dw
U= EA + EI + EA + dx
2 dx dx2 dx dx 4 dx
L
(14.42)

Buckling represents flexural actions, therefore, the first term of Eq. (14.42) which
indicates the strain energy due to pure axial deformation will not be considered. The
last term of Eq. (14.42) is a high-order term and it is also disregarded. Further, it may
be noted that
du
P = EA (14.43)
dx
where tensile value of P is taken as positive. Equation (14.42) then becomes
( [ ( )2 ( )2 ]
1 d 2w dw
U= EI +P dx (14.44)
2 dx2 dx
L

Referring to beam element given in Sect. 11.2, Eq. (14.10) gives

w = [N ] {X }e (14.45)
[( 2
)( )( 2 )( 3 )]
2x 3 2x 2 x3 2x 3 x2
where [N ] = 1 − 3x L 2 + L 3 x − L
+ L 2
3x
L 2 − L 3
x
L 2 − L
It has been already shown in Sect. 14.2.5, that the first part of Eq. (14.44) is given
by
( ( )2 (
1 d 2w 1
EI dx = {X }eT [B]T [D] [B] d x {X }e (14.46)
2 dx2 2
L L

Differentiating w with respect to x in Eq. (14.45) gives

dw
= [H ] {X }e (14.47)
dx
where
[( )( )( )( 2 )]
6x 6x 2 4x 3x 2 6x 2 6x 3x 2x
[H ] = − 2
+ 3 1− + 2 − 3 + 2 −
L L L L L L L2 L
(14.48)
354 14 Dynamic and Instability Analyses by the Finite Element Method

Therefore,
( )2
dw
= {X }eT [H ]T [H ] {X }e (14.49)
dx

Then, the second term of Eq. (14.44) becomes

( ( )2 (
1 dw P
P dx = {X }eT [H ]T [H ] d x {X }e (14.50)
2 dx 2
L L

Equation (14.44) can now be written as (assuming P to be compressive)

(L (L
1
U = {X }eT [B] [D] [B] d x − P
T
[H ]T [H ] d x {X }e
2
o o

which on minimization yields (potential of the applied flexural load has been
considered as zero)

[K ]e {X }e −[G]e {X }e = 0 (14.51)

where

[G]e is called the geometric stiffness matrix of the element,

[K]e is given by Eq. (11.23),
[G]e is given by

⎡ 36 ⎤
L
symmetrical
P ⎢⎢3 4L ⎥
⎥
[G]e =
30 ⎣ − 36
L
−3 36
L
⎦
3 −L −3 4L

When the stiffness of all such elements is combined, the final equation becomes

[K ] {X } = [G] {X } (14.52)

Equation (14.52) is an eigenvalue problem and can be compared with Eq. (14.17)
for free vibration analysis. The solution of Eq. (14.52) will yield n values of P
provided [K] and [G] are of size (n × n). The lowest value of P will correspond to
the buckling load.
14.3 Elastic Instability Analysis 355

Fig. 14.6 Example 1 3 5

2 4 6

L/2 L/2

14.3.1.1 An Illustrated Example of a Column

The critical load of a uniform column having both ends fixed is to be found by
dividing the column into two equal elements (Fig. 14.6).
The displacement labels have been indicated in Fig. 14.6. Stiffness matrix of each
element is known. In the assembled form, the stiffness matrix and geometric stiffness
matrix of the entire structure are given by
⎡ ⎤
12
⎢ 3L L 2 ⎥
⎢ symmetrical ⎥
⎢
8E I ⎢ −12 −3 24 ⎥
⎥
[K ]e = 3 ⎢ 2 ⎥
L ⎢ 3L L2 0 2L 2 ⎥
⎢ ⎥
⎣ 0 0 −12 −3 12 ⎦
L2
0 0 3L 2 −3L L 2

⎡ ⎤
36
⎢ 3L L 2 ⎥
⎢ 2 symmetrical ⎥
⎢
P ⎢ −36 − 2 72
3L ⎥
⎥
[G]e = ⎢ 3L 2 ⎥
15L ⎢ 2 − L4 − 3L 2L 2 ⎥
⎢ 2 ⎥
⎣ 0 0 −36 − 3L 22
36 ⎦
0 0 3L2
− L
2
− 3L
2
2L

Displacements corresponding to 1, 2, 3 and 4 are restrained. Therefore, Eq. (14.52)

becomes
[ ]( ) [ ]( )
8E I 24 0 X3 P 72 0 X3
= (14.a)
L3 0 2L 2 X4 15L 0 2L 2 X4

P L2
Putting 15E I
= λ, Eq. (14a) can be written as
[ ] ( ) ( )
192 − 72λ 0 X3 0
= (14.b)
0 2L 2 (8 − λ) X4 0

So, a non-trivial solution of Eq. (14b) is

356 14 Dynamic and Instability Analyses by the Finite Element Method

2L 2 (192−72λ) (8−λ) = 0

The least value of λ is given by

8
λ=
3
or,

8 × 15E I 40E I
Pcr = =
3L 2 L2

The exact solution to this problem is Pcr = 39.44 EL 2I

14.3.2 Plate Instability Analysis

Strains can be shown to be related to the displacements as given by

( )2 ⎫
∂u
− z ∂∂ xw2 + 21 ∂w ⎪
2
εx = ⎪
⎬
∂x ( ∂x )
2
ε y = ∂v − z ∂2w
+ 1 ∂w (14.53)
∂y ∂ y2 2 ∂y ⎪
⎪
γx y = ∂u + ∂∂vx − 2z ∂∂x∂wy + ∂w
2
∂w ⎭
∂y ∂x ∂y

The strain energy of an isotropic plate is given by

[ )2 ]
( ( ∂2w ∂2w
)2
∂2w ∂2w
(
∂2w
U= D
2 ∂x2
+ ∂ y2
+2 .
∂ x 2 ∂ y2
+ 2 (1 − ν) ∂ x∂ y
dA
(
A
( )2 ( ( )2
+ 21 N x ∂w
∂x
d A + Ny ∂w
∂y
dA (14.54)
( A ( ) ( ∂w ) A
+ N x y ∂w
∂x ∂y
dA
A

where N x and N y are the uniform tensile forces per unit length along x- and y-axes
and N xy the in-plane shearing forces per unit length respectively. While deriving the
expression for U, assumptions similar to that in the previous section have been made.
The term within the third bracket is that due to flexural actions. This bending strain
energy part is given
(
1
Uf = {X }eT [B]T [D] [B] d A {X }e (14.55)
2
A

with the usual notations adopted in the finite element plate bending chapter.
The strain energy part for the last three terms of Eq. (14.54) are derived as follows.
14.3 Elastic Instability Analysis 357

Introducing
[ ]
Nx Nx y
[Q] = (14.56)
Nx y N y

and
( )
∂w
[L] = ∂x (14.57)
∂w
∂y

The last three terms of the strain energy expression can be written as
(
1
U= [L]T [Q] [L] d x (14.58)
2
A

The displacement function of a plate element is given by

w = [N ] {X }e (14.59)

Differentiating Eq. (14.57) according to Eq. (14.59) yields

{L} = [H ] {X }e (14.60)

Substituting {L} from Eq. (14.60) into Eq. (14.59) yields

(
1
Ui = {X }eT [H ]T [Q] [H ] d A {X }e (14.61)
2
A

( )
Therefore, substituting U f and Ui from Eqs. (14.55) and (14.61) into
Eq. (14.54) yields
(
U = {X }eT 21 [B]T [D] [B] d A {X }e
(A (14.62)
+ 21 {X }eT [H ]T [Q] [H ] d A {X }e
A

which on minimization (potential of the applied flexural load has been considered as
zero) yields

[K ]e {X }e + [G]e {X }e = {0} (14.63)

where
[K]e plate stiffness matrix for flexure
358 14 Dynamic and Instability Analyses by the Finite Element Method

and
[G]e geometric stiffness matrix

(
[G]e = [H ]T [Q] [H ] d A (14.64)
A

Equation (14.62) is an eigenvalue problem. On solution, the critical value of axial

loads may be determined.
For rectangular element where a 12-term polynomial is assumed as the displace-
ment function given by Eq. (14.29), the explicit value of [G]e is given below for the
case when N x only is present and other in-plane forces are zero.
⎡ ⎤
224
⎢ −84 ⎥
⎢ 1104 ⎥
⎢ ⎥
⎢ −56 0 448 ⎥
⎢ ⎥
⎢ 0 0 0 96 ⎥
⎢ ⎥
⎢ 0 78 0 −36 48 ⎥
⎢ ⎥
Nx ⎢⎢ 56 −21 −14 0 0 112 ⎥
⎥
[G]e =
630 ⎢
⎢ 0 −39 0 18 −24 0 24 ⎥
⎥
⎢ −21 −42 ⎥
⎢ 204 0 78 0 0 552 ⎥
⎢ ⎥
⎢ −14 0 112 0 0 −28 0 0 224 ⎥
⎢ ⎥
⎢ 0 −552 −84 0 −39 0 39 −102 −21 552 ⎥
⎢ ⎥
⎣ 0 0 0 −48 18 0 −18 −39 0 0 48 ⎦
0 −102 −21 −39 0 0 0 −276 −42 102 39 276
(14.65)

Exercise 14
14.1 Consider an element of a shaft, subjected to torsion. The element has three
nodes including one midside node. Find the consistent mass matrix and the
stiffness matrix for the element.
14.2 Dividing the shaft of Prob. 14.1 into two elements, find the natural frequencies
of the shaft. Consider both ends fixed.
14.3 Determine the natural frequencies of a bar under axial vibration. The bar,
having one end fixed, other end free, is to be divided into two elements. The
element will have two nodes, one at each of two ends.
14.4 Formulate the consistent mass matrix of an isoparametric linear beam
element.
14.5 Determine the natural frequencies of a uniform beam, one end fixed and the
other end free treating the entire beam as one element and using Eq. (14.21).
14.6 Repeat Problem 14.4 using an isoparametric quadratic beam element.
14.3 Elastic Instability Analysis 359

14.7 Calculate explicitly the consistent mass matrix of a plate vibrating in its own
plane using an isoparametric linear element.
14.8 Formulate a consistent mass matrix of a plate vibrating in its own plane using
a triangular element having three corner nodes and three midside nodes.
14.9 For a uniform column having one end fixed and the other end pinned,
determine the critical load by dividing it into two equal elements.
14.10 Determine the critical load P for the frame. The frame is to be considered as
three elements formed by three members.

P P

I=∞

I I L

Prob. 14.10

14.11 Determine the critical load P of the frame.

I P

2I 2L

Prob. 14.11

14.12 Determine the critical load by finite element analysis of torsional buckling
of thin sections. The member is to be considered not free to warp given E, G,
J and β where β is the warping constant.
14.13 Extend the analysis of Prob. 14.12 to the case of flexural-torsional buckling.
360 14 Dynamic and Instability Analyses by the Finite Element Method

14.14 Using 12-term polynomial expression of Eq. (14.29) for rectangular plate
bending, Determine the critical value of N x for a square plate having all
edges clamped. Divide the plate into four equal parts.

References and Suggested Readings

1. K.J. Bathe, Finite Element Procedures (Prentice Hall, 1996)

2. G. Bergan et al. (eds.), Finite Elements in Nonlinear Mechanics (Tapir Publishers, Norway,
1976)
3. A. Chajes, Principles of Structural Stability (Prentice Hall, 1974)
4. R.W. Clough, J. Penzien, Dynamics of Structures (McGraw-Hill, 1975)
5. R.H. Gallagher, Finite Element Analysis Fundamentals (Prentice Hall, 1975)
6. R. Glowinsky, E.Y. Rodin, O.C. Zienkiewicz, Energy Methods in Finite Element Analysis (John
Wiley and Sons, 1979)
7. S. Levy, J.P.D. Wilkinson, The Component Element Method in Dynamics (McGraw-Hill, 1976)
8. M. Mukhopadhyay, Vibrations, Dynamics and Structural Systems, 2nd edn. (Oxford and IBH
Publishing Co., 2000)
9. R. Szilard, Theory and Analysis of Plates: Classical and Numerical Methods (Prentice Hall,
1974)
10. G.B. Warburton, The Dynamical Behaviour of Structures (Pergamon Press, 1976)
Chapter 15
The Finite Difference Method
for the Analysis of Beams and Plates

15.1 Introduction

Another versatile numerical method that exists for the analysis of structural
mechanics problems is the finite difference technique. The method is most general
and can be applied to the solution of the problem of any field for which a differential
equation is available. The method is suitable for the computer and can be easily
programmed.
The differential equation is the starting point of the method. The continuum is
divided in the form of a mesh and the unknowns in the problems are those at the
nodes. The derivatives of the equation are expressed in the finite difference form.
The differential equation, split in this discrete form, is applied at each node. This
results in a set of simultaneous equations [1–3].
In the finite element method, the final differential equation is surpassed, whereas
it is the differential equation to start within the finite difference method. The approx-
imation involved in the finite element method is physical in nature, as the actual
continuum is replaced by finite elements. The element formulation is mathemati-
cally exact. The finite difference technique involves the exact representation of the
continuum in terms of the differential equation and on this actual physical system,
the mathematical model is approximated.
However, the finite difference method has several disadvantages. The derivatives
of the approximated solution introduce inaccuracies. The imposition of boundary
conditions has always been a problem with the finite difference method, especially
when the boundaries are irregular and complex. The method cannot be formulated
with non-uniform and nonrectangular meshes.

© The Author(s) 2022 361

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_15
362 15 The Finite Difference Method for the Analysis of Beams and Plates

15.2 Finite Difference Representation of Derivatives

A variable w as a function of x is shown in Fig. 15.1. The function is divided into

equally spaced interval of h. Let i be any station. Stations to its right are indicated as
i + 1, i + 2,… and the corresponding values of the function are wi+1 , wi+2 … Stations
to the left of ‘i’ are designated as i − 1, i – 2… and the corresponding values of the
function are wi-1 , wi-2 …

15.2.1 First Derivative

The first derivative is given by

( )
dw 1
= (wi+1 − wi−1 ) (15.1)
dx i 2h

Equation (15.1) suggests that the slope of the curve between stations i – 1 and i
+ 1 has been considered to be constant.

15.2.2 Second Derivative

The second derivative can be approximated by

( ) [( ) ( ) ]
d 2w 1 dw dw
= − (15.2)
dx2 i h dx i+1/2 dx i−1/2

It can be shown that

Fig. 15.1 A function w

wi–2 wi–1 wi wi+1 wi+2

i –1 i + 1
2 2
i–2 i–1 I i+1 i+2 x
h h h h
15.2 Finite Difference Representation of Derivatives 363
( ) ⎫
(wi+1 − wi )⎪
dw 1
= ⎪
⎪
dx h ⎬
i+1/2
( ) (15.3)
dw ⎪
= (wi − wi−1 )⎪
1
⎪
⎭
dx i−1/2 h

Combining Eqs. (15.2) and (15.3) yields

( )
d 2w 1
= (wi+1 − 2wi + wi−1 ) (15.4)
dx2 i h2

Bending moment is proportional to the second derivative.

15.2.3 Third Derivative

The shear force is a function of the third derivative and it can be approximated as
( ) [( ) ( ) ]
d 3w 1 d 2w d 2w
= − (15.5)
dx3 i 2h dx2 i+1 dx2 i−1

Substituting the values of second derivatives from Eq. (15.4) into Eq. (15.5), it
can be shown that
( 3 )
d w 1 [ ]
= wi+2 − 2w i+1 + 2wi−1 − wi−2 (15.6)
dx3 i 2h 3

15.2.4 Fourth Derivative

Likewise, the fourth derivative can be shown to be

( )
d 4w 1[ ]
= wi+2 − 4wi+1 + 6wi − 4wi−1 + wi−2 (15.7)
dx4 h 4

In all the above finite difference expressions, a central difference scheme has
been adopted where it may be noted that the points are located symmetrically about
the point at which the derivatives are of interest to us. Though there are two other
schemes viz. the backward differences and the forward differences, it is the central
difference scheme that is adopted for the solution of structural mechanics problems.
The above four derivatives are shown schematically in Fig. 15.2.
364 15 The Finite Difference Method for the Analysis of Beams and Plates

Fig. 15.2 Schematic

representation of derivatives

15.3 Errors in the Finite Difference Expressions

By applying Taylor’s expansion of the functions of w about x, gives

h ' h2 h3
w(xi+1 ) = w(xi ) + w (xi ) + w'' (xi ) + w''' (xi ) + · · · (15.8)
1! 2! 3!

h ' h2 h3
w(xi−1 ) = w(xi ) − w (xi ) + w'' (xi ) − w''' (xi ) + · · · (15.9)
1! 2! 3!
Substituting for w(x i ) in Eq. (15.9) from Eq. (15.8), yields

2h 3 '''
w(xi+1 ) − w(xi−1 ) = 2hw' (xi ) + w (xi ) + · · · (15.10)
3!
Arranging different terms of Eq. (15.10), gives

1 [ ] h2 h4
w' (xi ) = w(xi+1 ) − w(xi−1 ) − w''' (xi ) − wv (xi ) + · · · (15.11)
2h 3! 5!
Therefore, the error in the first derivative I given by

h 2 ''' h4
φ1 = − w (xi ) − wv (xi ) − · · · (15.12)
3! 5!
Similarly, it can be shown that error in second, third and fourth derivatives is given
by

2h 2 iv 2h 4 vi
φ2 = − w (xi ) − w (xi ) − · · · (15.13)
4! 6!
15.4 Equivalent Concentrated Load 365

h2 v
φ3 = − w (xi ) − · · · (15.14)
4

h 2 vi
φ4 = − w (xi ) − · · · (15.15)
6

15.4 Equivalent Concentrated Load

For the analysis of beams by the finite difference method, the concentrated loadings
at the nodes are required for the analysis.
Three successive nodes are assumed to be simply supported and the support reac-
tions are determined as shown in Fig. 15.3. They are then reversed to indicate the
equivalent concentrated loading. For trapezoidal loading and two-degree parabolic
loading, the equivalent concentrated loading has been shown in Fig. 15.4.

Fig. 15.3 Equivalent concentrated load

Fig. 15.4 Equivalent concentrated load for some loading cases

366 15 The Finite Difference Method for the Analysis of Beams and Plates

15.5 Boundary Conditions for Beam Bending

The boundary conditions for three cases—simple support, fixed support and free
end have been considered here. The node ‘i’ of Fig. 15.5 has been considered as the
boundary of the member. The nodes i + 1 and i + 2 in Fig. 15.6 are fictitious nodes
outside the member.

15.5.1 Simple Support

The boundary conditions are

)
wi = 0
(15.16)
Mi = 0

Second of Eq. (15.16) can be expressed in terms of displacements as

wi−1 − 2wi + wi+1 = 0 (15.17)

15.5.2 Fixed End

The boundary conditions are

)
w i =
( dw )0 (15.18)
dx i
=0

Fig. 15.5 Boundary node i

Fig. 15.6 A simply supported beam

15.6 A Statically Determinate Static Problem 367

The second of Eq. (15.18) when expressed in the finite difference form becomes

wi+1 − wi−1 = 0 (15.19)

15.5.3 Free End

The boundary conditions at the free end of a uniform beam are

)
Mi = 0
(15.20)
Vi = 0

When expressed in the finite difference form, Eq. (15.20) becomes

)
wi−1 − 2wi + wi+1 = 0
(15.21)
−wi−2 + 2wi−1 − 2wi+1 + wi+2 = 0

For non-uniform beam, second condition corresponding to Eq. (15.20) leads to

considerable difficulty.

15.6 A Statically Determinate Static Problem

That ‘universal’ problem of a simply supported prismatic beam under uniform load
is solved here by the finite difference technique. The beam is divided into four equal
parts each of length l = L/4. The nodes are numbered from 0 to 4 as shown in Fig. 15.6.
The followings are the differential equations associated with a beam problem.

d 2w
EI = −M (15.22)
dx2

d2 M
= −q (15.23)
dx2
and,
( )
d2 d 2w
EI 2 = q (15.24)
dx2 dx

For a uniform beam, Eq. (15.24) becomes

368 15 The Finite Difference Method for the Analysis of Beams and Plates

d 4w
EI =q (15.25)
dx4
Though it is always preferable to use the more general equation (Eq. 15.22), but
for this statically determinate problem, bending moments can be easily determined.
Hence the displacements at the nodes will be obtained first by the use of Eq. (15.22).
Equation (15.22) when expressed in the finite difference form becomes

1 Mi
(wi−1 − 2wi + wi+1 ) = − (15.26)
h2 EI
The bending moments at nodes 1, 2, and 3 are given by

M1 = M3 = 3q L 2 /32 and M2 = q L 2 /8

Applying Eq. (15.26), first at node 1, i.e., i = 1 it becomes

M1 h 2
w0 − 2w1 + w2 = − (15.27)
EI
Substituting the value M 1 and h into the Eq. (15.27) and noting that w0 = 0,
Eq. (15.27) becomes

3 q L4
−2w1 + w2 = − (15.28)
512 E I
Applying Eq. (15.26) to nodes 2 and 3, yields

1 q L4
w1 − 2w2 + w3 = − (15.29)
128 E I

3 q L4
w2 − 2w3 = − (15.30)
512 E I
Equations (15.28–15.30) when written in matrix form becomes
⎡ ⎤⎧ ⎫ ⎧ ⎫
−2 1 0 ⎨ w1 ⎬ 4 ⎨3⎬
⎣ 1 −2 1 ⎦ w2 = − q L
4 (15.31)
⎩ ⎭ 512E I ⎩ ⎭
0 1 −2 w3 3

On solution of Eq. (15.31), yields

⎧ ⎫ ⎧ ⎫
⎨ w1 ⎬ q L4 ⎨ ⎬
20
w2 = 28 (15.32)
⎩ ⎭ 2048E I ⎩ ⎭
w3 20
15.6 A Statically Determinate Static Problem 369

4
Mid-span deflection is given by w2 = 512E7q L
I
which is about 5% in error when
compared to the exact value.
The same problem is solved on the basis of Eq. (15.25). Equation (15.25) when
expressed in the finite difference form becomes

EI
(wi+2 − 4wi+1 + 6wi − 4wi−1 + wi−2 ) = pi (15.33)
h4
It may reveal from Eq. (15.33) that the right hand side depends only on the
intensity of loading at the node and is independent of the manner in which it is
applied. Therefore, in order to substitute the nodal loading, Eq. (15.33) is modified
as
EI
(wi+2 − 4wi+1 + 6wi − 4wi−1 + wi−2 ) = pi h = Pi (15.34)
h3
where Pi is the equivalent load at the node.
Applying boundary conditions of Eqs. (15.16) and (15.17) gives
⎤
w0 = 0
w−1 + w1 = 0 ⎥
⎥
⎦ (15.35)
w4 = 0
w3 + w5 = 0

Applying Eq. (15.34) to nodes 1, 2 and 3 and substituting the appropriate boundary
conditions of (15.35), the following equations result.
⎡ ⎤⎧ ⎫ ⎧ ⎫
5 −4 0 ⎨ w1 ⎬ 4 ⎨1⎬
⎣ −4 6 −4 ⎦ w2 = − q L 1 (15.36)
⎩ ⎭ 256E I ⎩ ⎭
1 −4 5 w3 1

On solution of Eq. (15.36), yields

⎧ ⎫ ⎧ ⎫
⎨ w1 ⎬ q L4 ⎨ ⎬
5
w2 = 7 (15.37)
⎩ ⎭ 512E I ⎩ ⎭
w3 5

Once the deflections have been obtained, both the shearing force and the bending
moment can now be determined.
Applying Eq. (15.26) to node 2 yields

EI
M2 = − (w1 − 2w2 + w3 )
h2
16E I q L4 q L2
=− 2 × × (−4) =
L 512E I 8
370 15 The Finite Difference Method for the Analysis of Beams and Plates

Similarly,

16E I q L4 3
M1 = 2
(0 − 2 × 5 + 7) = q L2
L 512E I 32
The shear force equation for a uniform beam is

d3 y
V = −E I (15.38)
dx3
When expressed in the finite difference form, Eq. (15.38) becomes

EI
Vi = (wi+2 − 2wi+1 + 2wi−1 − wi ) (15.39)
2h 3
Applying Eq. (15.39) to node 1 gives

64E I q L 4 qL
V1 = − (5 − 2 × 7 + 5 + 0) =
L 3 512E I 4
Similarly,

64E I q L 4
V2 = − (2 × 5 − 2 × 5) = 0
L 3 512E I

15.7 A Statically Indeterminate Static Problem

The analysis of an indeterminate structural problem by the finite difference technique

is not different from that of a determinate problem.
For the beam of Fig. 15.7 which is divided into 4 equal parts and the node numbers
indicated, the boundary conditions at node 0 are

w0 = 0 and w−1 + w1 = 0 (15.40)

Boundary conditions at node 4 are

w4 = 0 and w3 −w5 = 0 (15.41)

The couple applied at 0 is to be replaced by two equal and opposite forces as

shown in Fig. 15.7b.
Applying Eq. (15.34) at nodes 1, 2 and 3 and applying appropriate boundary
conditions of (15.40) and (15.41), results in the following equations.
15.8 Free Vibration of Beams 371

Fig. 15.7 A statically indeterminate problem

⎡ ⎤⎧ ⎫ ⎧ ⎫
5 −4 1 ⎨ w1 ⎬ ⎨1⎬
EI ⎣ ⎦ w2 = 4
−4 6 −4 0 (15.42)
h3 ⎩ ⎭ L⎩ ⎭
1 −4 7 w3 0

On solution of Eq. (15.42) yields

⎧ ⎫ ⎧ ⎫
⎨ w1 ⎬ L2 ⎨
26 ⎬
w = 24 (15.43)
⎩ 2 ⎭ 704E I ⎩ ⎭
w3 −10

Slope at 0 is given by
( )
dw 1
= (w1 − w−1 ) (15.44)
dx 0 2h

Combining Eq. (15.40) and (15.41) yields

( )
dw 4 26L 2 L
= ×2× = (15.45)
dx 0 2L 704E I 6.77E I

The exact value of the rotation at the end is L/(4EI). The error is significant in this
and this is primarily due to replacing the couple by two equal and opposite forces.
With more divisions, the error will reduce.

15.8 Free Vibration of Beams

The beam of Fig. 15.7 is to be analysed for free vibration, i.e., the natural frequencies
and mode shapes are to be determined.
The differential equation of free vibration is given by
372 15 The Finite Difference Method for the Analysis of Beams and Plates
( )
d2 d 2w
E I 2 = p 2 mw (15.46)
dx2 dx

where m is the mass per unit length.

For a uniform beam, Eq. (15.46) becomes,

d 4w
EI = p 2 mw (15.47)
dx4
When expressed in finite difference form, Eq. (15.47) becomes

EI
(wi+2 − 4wi+1 + 6wi − 4wi−1 + wi−2 ) = p 2 mwi (15.48)
h4

Introducing λ = p 2 mh 4 /E I , Eq. (15.48) becomes

(wi+2 − 4wi+1 + 6wi − 4wi−1 + wi−2 ) = λwi (15.49)

The boundary condition at nodes 0 and 4 are given by Eqs. (15.40) and (15.41)
The beam is divided into 4 equal parts so that h = L/4. Equation (15.49) as is
applied to nodes 1, 2 and 3 respectively and the appropriate boundary conditions are
incorporated and the followings are the resulting equations.
⎡ ⎤⎧ ⎫ ⎡ ⎤⎧ ⎫
5 −4 1 ⎨ w1 ⎬ 1 0 0 ⎨ w1 ⎬
⎣ −4 6 −4 ⎦ w2 = λ⎣ 0 1 0 ⎦ w2 (15.50)
⎩ ⎭ ⎩ ⎭
1 −4 7 w3 001 w3

Equation (15.50) is an eigenvalue problem. For non-trivial solution,

| |
| 5 − λ −4 1 ||
|
| −4 6 − λ −4 | = 0 (15.51)
| |
| 1 −4 7 − λ |

The determinant of Eq. (15.51) is

(5 − λ) [(6 − λ) (7 − λ) − 16] + 4[−4(7 − λ) + 4 ] + 1[16 − 6 + λ] = 0

(15.52)

Equation (15.52) is a cubic equation where the lowest root is

λ1 = 0.72

Therefore,
15.9 Buckling of Columns 373

Fig. 15.8 Mode shape in the

first mode

p2 h 4 m
= 0.72
EI
or,
/
EI
p1 = 13.6
m L4
/
The exact value of p1 = 15.9 mELI4 .
The other roots will give other natural frequencies.
Assume w1 = 1 and substituting the value of λ1 = 0.72 into second and third
equations of Eq. (15.50), yields

5.28w2 − 4w3 = 4
(15.53)
−4w2 + 6.28w3 = −1

On solution, w2 = 1.231 and w3 = 0.625 w3 = 0.625

Therefore, the mode shape in first mode is as Fig. 15.8.

15.9 Buckling of Columns

The differential equation of axially loaded column is given by

d 2w
EI + pw = 0 (15.54)
dx2
When expressed in the finite difference form, Eq. (15.54) becomes

EI
(wi+1 − 2wi + wi−1 ) + pwi = 0
h2
or,

wi+1 − (2 − λ)wi + wi−1 = 0 (15.55)

2
where λ = phEI
Consider a column having both ends hinged. The column is uniform and is divided
into four equal parts (Fig. 15.9). Applying Eq. (15.55) to nodes 1, 2 and 3 and
374 15 The Finite Difference Method for the Analysis of Beams and Plates

Fig. 15.9 Hinged—hinged –1

column

substituting the appropriate boundary conditions, results in

⎡ ⎤⎧ ⎫ ⎧ ⎫
−2 + λ 1 0 ⎨ w1 ⎬ ⎨ 0 ⎬
⎣ 1 −2 + λ 1 ⎦ w2 = 0 (15.56)
⎩ ⎭ ⎩ ⎭
0 1 −2 + λ w3 0

Non-trivial solution of Eq. (15.56) is

(−2 + λ) [(−2 + λ)2 − 1] − 1(−2 + λ) = 0

or, λ2 − 4λ + 2 =√0
or, λ = −2 ± 2
Minimum value of λ = 0.586

ph 2 EI
= 0.586 or, Pcr = 9.376 2
EI L
The answer differs from the exact value by 5%.
15.10 Finite Difference Representation of the Plate Equation 375

15.10 Finite Difference Representation of the Plate

Equation

The plate equation is given by

∂ 4w ∂ 4w ∂ 4w p
+2 2 2 + 4 = (15.57)
∂x 4 ∂x ∂y ∂y D

The plate is divided into equally paced square mesh having Δx = Δy = h

(Fig. 15.10). The finite difference expressions are written with respect to the node (i,
j). ‘i’ are numbered in the x-direction and ‘j’s are in the y-direction. Therefore
( )
∂ 4w 1 || |
= wi+2, j − 4wi+1, j + 6wi, j − 4wi−1, j + wi−2, j | (15.58)
∂x4 i, j h 4

( )
∂ 4w 1 || |
= wi, j+2 − 4wi, j+1 + 6wi, j − 4wi, j−1 + wi, j−2 | (15.59)
∂ y4 i, j h 4

( ) | ( )|
∂4w | ∂2 ∂2w |
| |
=| 2 |
∂ x 2 ∂ y2 | ∂y ∂x2 |
i, j i, j
|( ) ( ) ( ) |
| |
1 || ∂ 2 w ∂2w ∂2w |
|
= 2| − 2 + |
h | ∂x2 ∂x2 ∂x2
i, j+1 i, j i, j−1 |
1 [( ) ( )
= wi+1, j+1 − 2wi, j+1 + wi−1, j+1 − 2 wi+1, j − 2wi, j + wi−1, j
h4
( )]
+ wi+1, j−1 − 2wi, j−1 + wi−1, j−1

or,

Δy

Δy i–2, j+2 i–1, j+2 i, j+2 i+1, j+2 i+2, j+2

Δy i–2, j+1 i–1, j+1 ii, j+1 i+1, j+1 i+2, j+1
Δy i–2, j i–1, j I, j i+1, j i+2, j
Δy i–2, j–1 i–1, j–1 I, j–1 i+1, j–1 i+2, j–1
Δy i–2, j–2 i–1, j–2 I, j–2 i+1, j–2 i+2, j–2

Δx Δx Δx Δx Δx Δx

Fig. 15.10 Plate with mesh division

376 15 The Finite Difference Method for the Analysis of Beams and Plates
( )
∂4w =
∂ x 2 ∂ y 2 i, j
1 [w − 2wi, j+1 + wi−1, j+1 − 2wi+1, j + 4wi, j − 2wi−1, j + wi+1, j−1 − 2wi, j−1 + wi−1, j−1
]
h 4 i+1, j+1
(15.60)

The finite difference expressions for the bending moments are given by
[ ]
∂ 2w ∂ 2w
(Mx )i, j = −D +v 2
∂x 2 ∂y i, j

D [( ) ( )]
= − 2 wi+1, j − 2wi, j + wi−1, j + ν wi, j+1 − 2wi, j + wi, j−1 (15.61)
h
D [( ) ( )]
=− wi, j+1 − 2wi, j + wi, j−1 + ν wi+1, j − 2wi, j + wi−1, j (15.62)
h2
and
[ ]
( ) ∂ 2w
Mx y = (1 − ν) D
i, j ∂ x∂ y i, j
(1 − v)D [( )]
= wi+1, j+1 − wi+1, j−1 − wi−1, j+1 + wi−1, j−1 (15.63)
4h 2
The finite difference expressions for shearing forces are
( )
∂ 3w ∂ 3w
(Vx )i, j = −D +
∂x3 ∂ x∂ y 2 i, j
(
= −D wi+2, j − 2wi+1, j + 2wi−1, j − wi−2, j + wi+1, j+1 − 2wi+1, j
)
+wi+1, j−1 − wi−1, j+1 + 2wi−1, j − wi−1, j−1 (15.64)

and
( )
( ) ∂ 3w ∂ 3w
Vy i, j = −D +
∂ y3 ∂ y∂ x 2 i, j
(
= −D wi, j+2 − 2wi, j+1 + 2wi, j−1 − wi, j−2 + wi+1, j+1 − 2wi, j+1
)
+wi−1, j+1 − wi+1, j−1 + 2wi, j−1 − wi−1, j−1 (15.65)

By substituting the finite difference expressions of the different terms of

Eq. (15.57), the final equation for a node (i, j) is shown in Fig. 15.11 and is denoted
as Eq. (15.66). For the sake of illustrating the method, an example is presented.
15.10 Finite Difference Representation of the Plate Equation 377

Fig. 15.11 Finite difference 1

equation at node (i, j)

2 -8 2

1 -8 20 -8
4
1 =h ( Dq )
I, j
i, j
h
2 -8 2
h

15.10.1 A Plate Example

A simply supported uniform square plate of Fig. 15.12 is divided into a 4 × 4 mesh
size. It is uniformly loaded with a load having intensity q0 . The different nodes are
designated in Fig. 15.12 taking into account the symmetry of the problem. The simply
supported end conditions gives

w2 = −w−2 and w3 = −w−3 (15.66)

If the plate dimension is a, then applying Eq. (15.66) at points 1, 2 and 3, gives
⎡ ⎤⎧ ⎫ ⎧ ⎫
20 −32 8 ⎨ w1 ⎬ 4 ⎨1⎬
⎣ −8 24 −16 ⎦ w2 = q0 a 1 (15.67)
⎩ ⎭ 256D ⎩ ⎭
2 −16 20 w3 1

Solution of the Eq. (15.68) gives

Fig. 15.12 The plate 3′ 2′ 3′

example

3′ 3′
3 2 3
2′ 2′ a
1
3′ 3′
3 2 3

3′ 2′ 3′
a
378 15 The Finite Difference Method for the Analysis of Beams and Plates
⎧ ⎫ ⎧ ⎫
⎨ w1 ⎬ q a 4 ⎨ 0.00403 ⎬
0
w = 0.00293 (15.68)
⎩ 2⎭ D ⎩ ⎭
w3 0.00214

Once the deflections at the nodal points are obtained, the bending moment and
the shear force at any nodal point can be determined. For example, if M x is required
at node 1 of Fig. (15.12), then applying Eq. (15.61) at node 1 and assuming v = 0.3,

16D
Mx1 = − (w2 − 2w1 + w2 ) + 0.3(w2 − 2w1 + w2 )
a2
16D
= − 2 × 2 × 1.3 × (w2 − w1 )
a
= 0.0458q0 a 2

Exercise 15

15.1 For a uniform square plate under bending and as per mesh division shown,
determine the deflection at nodal points.

Prob. 15.1

15.2 Determine the buckling load for the columns.

15.10 Finite Difference Representation of the Plate Equation 379

Prob. 15.2

15.3 Determine the natural frequencies of a uniform beam having both ends free.
Divide the beam into 4 equal divisions.
15.4 Determine the buckling load of a square uniform plate dividing it into a mesh
3 × 3 size. The plate is simply supported and an advantage may be taken
from the symmetry of the mode of buckling.
15.5 For a uniform rectangular plate having rectangular mesh division where
Δx /= Δy, write finite difference expression for the plate equation.
15.6 For a uniform square plate having two opposite edges simply supported and
the other two opposite edges free, determine the fundamental frequency on
the basis of dividing the plate into a 4 × 4 grid.
15.7 For a uniform rectangular plate of sides a and b having a = 2b, determine
the nodal deflection by dividing the plate into 4 × 4 mesh divisions for a
concentrated load P applied at mid-span.
15.8 Determine the critical load of a column having one end fixed and the other
end free. The column is of uniform cross-section. The critical load is to
be determined by applying the finite difference method to the higher order
equation given below

P ''
wiv + w =0
EI
15.9 For the uniform plate shown, determine the maximum bending moment at
mid-span.
380 15 The Finite Difference Method for the Analysis of Beams and Plates

Prob. 15.9

15.10 Determine the critical load for the column.

Prob. 15.10

References and Suggested Readings

1. D.D. McCracken, W.S. Dorn, Numerical Methods and FORTRAN Programming (John Wiley &
Sons, 1966)
2. F. Schied, Numerical analysis (McGraw-Hill Book Co., Schaum Series, 1968)
3. P.C. Wang, Numerical Methods in Structural Mechanics (John Wiley & Sons, 1980)
Chapter 16
Adaptive Finite Element Analysis

16.1 Introduction

The choice of optimum mesh size to obtain results within desired accuracy has been
haunting the engineers since the inception of the finite element method. In order to
achieve this, one of the popular approaches has been to study the convergence of
results by making the mesh finer and finer. In such cases, experience of the analysts
based on previous computation rules of previous elements’ shapes and sizes plays
a very significant role. But it necessitates many runs of the computer program. The
approach is more time-consuming and expensive as such. It is highly desirable to
evolve an approach by which the refinement on mesh size becomes automatic, that
is, accuracy to be achieved without any user intervention. In order to reach this goal,
the error estimation of the solution is necessary. Attempts have been continuing to
develop simple and reliable error estimates.
Thus, the approaches of arriving at accurate results obtained from the finite
element method are as follows:
1. Experience-based discretization
2. Adaptive finite element technique.

16.2 The Adaptive Finite Element Technique

The earlier practice based on the experience of the analysts is indeed an expensive
procedure. An automatic mesh generator has thus become very much desirable. In the
following, the adaptive finite element method is discussed. Various topics involved
in this are:
(a) superconvergent patch recovery technique, (b) error estimation, (c) error
estimator, (d) refinement strategy of meshes and (e) adaptive mesh generator.

© The Author(s) 2022 381

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_16
382 16 Adaptive Finite Element Analysis

16.3 Superconvergent Patch Recovery Technique

The obvious choice of calculating the derivatives (stresses) from the finite element
approximation is to directly differentiate the finite element solution at salient points.
For example, stresses or stress resultants are obtained by differentiating the displace-
ment solution and using the appropriate constitutive relations. Direct differentiation
will result in a lower order derivative that will be discontinuous. Points where accurate
stresses are to be determined, thus, have inferior values, especially at the boundary
of the elements and at the interelemental nodes. The superconvergent patch recovery
technique (SPR) has been developed to evaluate the stresses accurately [1, 2].
The directly computed consistent stresses are
{ h} { }
σ = [D][H ] u h (16.1)

where
[D] is the elasticity matrix,
[H] is the differential operator, which defines strain as
{ }
{ε} = [H ] u h (16.2)
{ h}
u the finite element approximation is given by
{ h}
u = [N ]{u} (16.3)

where
[N ] is the shape function matrix
{u} is the nodal displacement matrix { }
It has already been indicated that stresses σ h obtained from Eq. (16.1) are less
accurate as they do not possess interelemental continuity and have a low accuracy
at the nodes and at the element boundaries. The objective of the recovery of finite
element derivatives is to find nodal parameters {σ ∗ } such that smoothed continuous
stress field {σ∗ } is defined by the basis function [N] and nodal parameters {σ ∗ } as

{σ∗ } = [N ]{σ ∗ } (16.4)

In the SPR technique, the domain is divided into finite element meshes, which
will result into many overlapping node patches. A node is connected to a number
of elements, which is referred to as a patch (Figs. 16.1 and 16.2). In this recovery
process, the nodal superconvergent stresses {σ ∗ } belong to a polynomial expansion
p
σ∗ of the same complete polynomial order p as that present in basis functions, which
p
are assumed valid over a node patch. Thus σ∗ within a node patch is given by

σ∗p = {P}{a} (16.5)

16.3 Superconvergent Patch Recovery Technique 383

Vertex
node

Fig. 16.1 Superconvergent points in a triangular element

Fig. 16.2 Superconvergent

points in a rectangular
element

where
{P} contains appropriate polynomial terms.
and{a} is a set of unknown parameters.
For one-dimensional problem of order p, {P} and {a} are expressed as
{ }
{P} = 1 x x 2 x 3 . . . x p (16.6)

{ }
{a} = a1 a2 a3 . . . a p+1 (16.7)

For two or three-dimensional problems, complete polynomial terms for the appro-
priate element order can be written. Therefore, two dimensions and linear expansions
we have
{ }
{P} = 1 x y (16.8)

and for quadratic expansion

{ }
{P} = 1 x y x 2 x y y 2 (16.9)
384 16 Adaptive Finite Element Analysis

In case of quadrilaterals in which terms additional to the complete polynomial

occur, results indicate small improvement by using identical terms to those occurring
in the shape function matrix [N]. Thus, for a bilinear quadrilateral, it is
{ }
{P} = 1 x y x y (16.10)

p
Thus, σ∗ in a node patch for a quadratic triangular element can be written as
⎧ ⎫
⎪
⎪ a1 ⎪⎪
⎪
⎪ ⎪
⎪
⎪
⎪ a ⎪
⎪
⎪
⎨
2 ⎪
⎬
{ } a
σ∗ = 1 x y x x y y
p 2 2 3
(16.11)
⎪
⎪ a4 ⎪⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎩
a 5 ⎪
⎭
a6

Now, the unknown coefficients {a} are to be determined. For this, a discrete least
square fit is to be employed. The fit is made over a set of superconvergent sampling
points existing in the chosen patch in order to minimize the function

Σ
n
[ h ]2
F(a) = σ (xi , yi ) − {P(xi , yi )}{a} (16.12)
i=1

where (xi , yi ) are the coordinates of a group of total number of sampling points n.
However, it is important to recognize that such a suggestion is based on numerical
experiments [1, 2] rather than vigorous mathematical analysis. The midside nodes
of a triangular element and Gauss points of a quadratic element are superconvergent
points for superconvergent stress extraction.
Total number of sampling points in each element is

n = mk (16.13)

The implication of the minimization condition of F(a) is

Σ
n Σ
n
{ }
{P(xi , yi )} {P(xi , yi )}{a} =
T
{P(xi , yi )}T σ h (xi , yi ) (16.14)
i=1 i=1

From Eq. (16.14), {a} can be determined as

{a} = [A]−1 {b} (16.15)

where
16.4 Example of Verification of SPR 385

Σ
h
[A] = {P(xi , yi )}T {P(xi , yi )} (16.16)
i=1

Σ
n
{ }
{b} = {P(xi , yi )}T σ h (xi , yi ) (16.17)
i=1

Number of equations in each patch is modest and recovery is performed only for
each vertex node. The procedure, therefore, is quite inexpensive. We may note that
p
precisely the same matrix [A] in the solution of each component of σ∗ and a single
inversion of [A] is necessary.
Depending on the values of the parameter {a}, the recovered nodal values of {σ ∗ }
p
are calculated by inserting appropriate coordinates into the expression for σ∗ . The
entire procedure can be repeated for all components of the stress or the formulation
can be adjusted so that the unknown coefficients {a} for each component of stress
can be found in a single run. For accurate evaluation of nodal superconvergent stress
σ ∗ , following procedure is recommended [1, 2].
• In each patch, only the nodes inside the patch are to be considered for interpolation
using patch Eq. (16.11). Boundary nodes of a patch should not be interpolated
from Eq. (16.11).
• The patches will frequently overlap for internal midside nodes. This suggests that
such nodal values are frequently evaluated from more than one patch. For such
nodes, average values of the recovered stresses are to be considered.
• In case where the vertex node is at the boundary of the domain, boundary patches
should not be considered. Instead, the recovery of the boundary nodes should be
determined from exterior patches that include them.
The matrix [A] turns out to be singular in certain category of problems or numer-
ically is very close to be singular. As such some special techniques such as singular
value decomposition technique (SVA) may be applied for the inversion of [A] matrix
[3].

16.4 Example of Verification of SPR

The problem to be modelled here is a portion of an infinite plate with a central circular
hole subjected to unidirectional tensile load. The geometry of the problem is shown
in Fig. 16.3a. Plane strain conditions have been assumed along with a Poisson’s ratio
of 0.3 and Young’s modulus of 1000. The portion of the plate is analyzed, shown in
Fig. 16.3b. The convergence of the stress components has been observed at point 1
(Fig. 16.3b). The boundary conditions prescribed are such that on edges AB and ED,
symmetry conditions are imposed and on edges BC and DC, the plate is loaded by
tractions given by the analytical solution
386 16 Adaptive Finite Element Analysis

Fig. 16.3 Infinite plate with

a central circular hole

B C
A r
θ
R=a D
E

(a)

B C

E = 1000
(1.5, 45°) 4
v = 0.3
Thickness = 1.0
A
a = 1.0 1
Plane strain conditions

D
O E

1 3

(b)

2( ) ⎫
σx = 1 − ar 2 23 cos 2θ + cos 4θ + 23 ar 4 cos 4θ ⎪
4
⎬
2( ) 4
σ y = − ar 2 21 cos 2θ − cos 4θ − 23 ar 4 cos 4θ (16.18)
2( ) 4 ⎪
⎭
σx y = − ar 2 21 sin 2θ + sin 4θ + 23 ar 4 sin θ

Figure 16.4 shows semiuniform meshes that are considered for convergent study.
Figure 16.5 shows the convergence of stress components at point 1. It may be observed
that the stresses determined by the SPR technique are more accurate than the finite
element solutions.

16.5 Error Estimation

The finite element method is a numerical method. As such the solution obtained from
it is an approximate one. The differences between the exact and approximate solutions
are termed as discretization error or simply error. Thus the error in displacement is
16.5 Error Estimation 387

(a) (b)
Mesh 1 Mesh 2
Elements = 32; Nodes = 81 Elements = 128; Nodes = 289

(c)
Mesh 3
Elements = 288; Nodes = 625

Fig. 16.4 Semiuniform meshes used for accuracy observations

given by
{ }
{eu } = {u} − u h (16.19)

where {u}is the exact solution and {u h } is the finite element approximation.
Similarly, the error in stresses is given by
{ }
{eσ } = {σ } − σ h (16.20)

An analyst faces two important queries.

1. What is the procedure to determine reliability of the FEM solution?
388 16 Adaptive Finite Element Analysis

Fig. 16.5 Convergence of

stress components point 1

2. How to determine the desired level of accuracy by efficiently controlling the

error?
In order to seek an answer to both the above queries, the estimation of the error
is made, which aids in adaptive computations. It is nothing but an iterative process
in which refinement of approximations is continually made so that the numerical
solution tends to near the ‘exact’ solution.
The error estimates are divided into two categories: (1) a priori error estimator
and (2) posteriori error estimator. A priori error estimator provides only a qualitative
description on the rate of convergence of the finite element solution. This error
estimator is rather complex to use and difficult to implement.
A posteriori error estimator is used only after a particular FEM solution is obtained.
This estimation can be divided into two categories. The computation residuals from
16.6 ZZ Error Estimator 389

the equilibrium equations and subsequent determination of interelemental traction

jumps along element boundaries fall in the first category. In the second category,
stress smoothing technique is employed [4]. This error estimator is effective in many
classes of application [5–9].
The procedures for adaptive finite element analysis are indicated as follows:
1. h-adaptivity—meshes are refined and/or coarsened to improve the finite element
solutions.
2. p-adaptivity—by keeping mesh size unchanged, the order of the polynomial is
varied for all elements to improve the finite element solution.
3. h–p adaptivity—both h and p types of refinement are adopted simultaneously for
improving the finite element solution.
In order to use p-adaptivity for the problem, a new code is required to be developed.
But the procedure is rather costly, but it gives a very accurate solution. However, h–p
adaptivity will prove to be not only costly but complicated as well. h-adaptivity is
most widely used in practice. We shall limit ourselves to h-adaptivity only.
In general, an automatic adaptive finite element modelling environment consists
of the following.
1. An automatic adaptive mesh generator capable of producing the desired mesh
configuration.
2. The finite element solver for class of problems under consideration.
3. A posteriori error estimator used for determining the discretization error in the
requested norms.
4. Mesh indicators specify where and how to improve the mesh to obtain the desired
level of accuracy.
Figure 16.6 shows a typical self-explanatory h-adaptive finite element scheme.

16.6 ZZ Error Estimator

The ZZ error estimator that is quantified on the basis of energy norm is an integral
scalar quantity and is defined for elasticity problems as [1–10]
(
( { })T ( { })
||ees ||2e = {σ∗ } − σ h [D]−1 {σ∗ ) − σ h d A (16.21)
A

where the integration is to be carried out over an elemental area. {σ∗ } is the smoothed
stress field from a recovery procedure (for example, nodal averaging or supercon-
vergent patch recovery). In Eq. (16.21), ||ees || e is the estimated discretization error
of an element in energy norm.
The underlying principle of ZZ-error estimator is to use improved stress field that
may be derived from post-processing of the finite element stress field rather than
390 16 Adaptive Finite Element Analysis

Fig. 16.6 A h-adaptive finite Input Data

element scheme ● Geometry, boundary conditions
● Material properties
● Loading conditions
● Element size in initial mesh

Automatic Mesh Generation

Finite Element Analysis

Error Estimate
● Construct improved stress field
● Calculate element errors
● Calculate global error

Remeshing Strategies
● Calculate permissible errors
● Calculate new elements sizes

Is error smaller than

prescribed value?
NO

YES

Stop

the exact field that is unknown. In the same manner, the square of the energy of an
element is defined as
(
{ h }T { }
||U ||2e = σ [D]−1 σ h d A (16.22)
A

Summation of the squared values of ||ees ||e and ||U ||2e over all elements gives rise
to squared energy norms for the whole domain

Σ
n
||U ||2 = ||U ||2e (16.23)
e=1
16.7 ZZ—Refinement Framework 391

where n is the total number of elements and the square of the energy norm is

Σ
n
||ees ||e = ||ees ||2e (16.24)
e=1

The estimated relative percentage of error is defined as

||ees ||
μes = √ × 100 (16.25)
||U ||2 + ||ees ||2

μes is widely accepted as the measure of accuracy of the computed solution.

Thus, the adaptively refined mesh should satisfy the condition

μes < μ (16.26)

where μ is the maximum permissible relative percentage of error or the target error
set by the user.
The quantities ||ees ||2e and ||U ||2 can be calculated by numerical integration such
as Gaussian quadrature. Equations (16.21) and (16.22) require three-point Gaussian
integration scheme evaluation.

16.7 ZZ—Refinement Framework

It is basically a h-refinement strategy and is dependent on the h-convergent charac-

teristics of the finite element method [9]. With trial functions in the displacement
formulation of degree p, the error in stresses of the order of O(h) p or O(N D F)− p/2
(where NDF is the number of degrees of freedom) and is assumed to be equally appli-
cable to the energy norm of the error. However, this is true if singularity exists in the
domain. It can be shown that, in general, the rate of convergence of the estimated
error for problems with singularity is

O(N D F) − [min (λ, p)]/2 (16.27)

where λ is a number associated with the intensity of singularity. For elasticity

problems, it ranges from 0.5 for a nearly closed crack to 0.7 for 90° corner [11]
We defined optimized mesh as one in which the error in each element (in the limit)
is constant. It is assumed in the optimal mesh that the energy norm error ||ees || is
equal in all elements. Thus, if the permissible error in the whole domain is
/
μ ||U ||2 + ||ees ||2 (16.28)
392 16 Adaptive Finite Element Analysis

then, in accordance with the definition of the optimum mesh, the requirement that
the permissible error in any element should be
/
||U ||2 + ||ees ||2
em = μ (16.29)
n

As mentioned above, the following two equations can be written as

||ees ||e ∼ p
= c · h old (16.30)

em = c · h new
p
(16.31)

Dividing Eq. (16.30) by Eq. (16.31), we get

p ( )p
em h new h new
= p = (16.32)
||ees ||e h old h old

Equation (16.32) can be written as

( )1/ p
h new em
= (16.33)
h old ||ees ||e

||ecs ||e
which can be written in the final form of the introducing ψe = em
and βe = 1
p

h old
h new = β
(16.34)
ψe e

Three conditions are possible depending on the ψe of the element

• ψe > 1, the element should be refined
• ψe = 1, no change in the element size is required
• ψe < 1, the element size may be derefined, if desired for computational efficiency.
In Eq. (16.34), h old is the size of an element in current mesh, and h new is the new
element size.

16.8 Adaptive Mesh Generation

Considerable input data are to be prepared for carrying out the finite element analysis;
out of which the data required as an input for mesh generation are quite large. The
manual preparation of these data is tedious, time consuming and error prone. As
such it is highly desirable to have a computerized or automatic generation of finite
16.8 Adaptive Mesh Generation 393

element mesh. This will require a minimum amount of data to be prepared for mesh
generation. The mesh generation in successive cycle is then to be carried out without
any user intervention. Attempts are being made to develop a highly effective and
robust mesh generator, which will be versatile and computationally efficient. Thus
the labour of creating a mesh manually is avoided. Very broadly automatic mesh
generation methods are grouped into two categories [4].
(a) Methods that are based on mapping the domain or various subdomains into
which it is divided into simple forms in which direct structured mesh forms can be
avoided (Figs. 16.7 and 16.8) and (b) methods that are based on joining a set of
predetermined nodes to construct a mesh or to satisfy prescribed element shape and
size requirements.

Fig. 16.7 Structured mesh

Fig. 16.8 Unstructured

mesh
394 16 Adaptive Finite Element Analysis

16.8.1 Mesh Generation Based on Mapping

In the first category are included methods such as isoparametric mapping concept
[12] and those based on transfinite interpolation [13]. When mapping techniques are
used in mesh generation, the nodal domain is to be subdivided into regular grids to
induce a mesh. However, the manual process in subdivision of a domain is tedious
and sometimes it is really difficult in case of a domain with complex geometry or
simple geometry with complex mesh density. As the element sizes are controlled
by the subdivision of simple subregions, more subregions are needed to generate
the mesh that can accommodate changes in the desired element sizes from region to
region. This has further restricted their use by the requirement of optimal mesh in
adaptive analysis. For this reason, we shall not discuss this any further.
In the second category, many variants are possible. But as the mesh is constructed,
it can, in general, achieve a prescribed density. Most of the techniques in the second
category currently in use may be classified into the following categories:
• Delaunay triangulation method
• Domain decomposition (Quadtree) method
• Advancing front technique.

16.8.2 Delaunay Triangulation Method

The meshing techniques that utilize Delaunay’s criterion are quite popular. The crite-
rion states that the circumcircle of the element contains no points [14]. It generates
triangular mesh a given set of nodes.
The boundary of the geometry is first meshed so as to generate a given set of
nodes. The boundary nodes are triangulated according to Delaumay criterion. Nodes
are then inserted incrementally into the mesh, redefining the triangle as a new node
is inserted to maintain the Delaunay criterion. The method that is chosen for defining
the locations of the interior nodes, distinguishes one Delauny algorithm from the
other. An important property of the Delaunay triangulation is that it generates as
close as possible to the equilateral triangle from a given set of nodes. Large number
of researchers have utilized Delaunay criterion for triangulation [15–17].

16.8.3 Domain Decomposition Method (Quadtree)

Given a geometrical shape for the finite element analysis, there are several methods
that may be used to produce the necessary nodal points and subsequent element mesh.
Domain decomposition method is one such that is useful for generating necessary
nodal points. In two-dimensional cases, domain decomposition method takes the
form of a quadtree structure in which the corner point or the central point of each
16.8 Adaptive Mesh Generation 395

cell may be chosen as nodal points. Quadtree method is developed and implemented
by Shephard and his coworkers [18, 19].
Domain decomposition technique has the advantage of being easily created and
accessed due to tree-like structure. However, the spacing of the resulting meshes does
not transit smoothly from one location to the other, which is an essential drawback.
Thus, the shape of the mesh is not essentially good. Another disadvantage of the
scheme is that it is difficult to fit the tree structure to the boundary of the arbitrary
domain.

16.8.4 Advancing Front Technique

Advancing front technique (AFT) or moving front method is a technique in which the
adaptive refinement of the grid is accomplished by means of a remeshing procedure,
which is based on the information provided by the computer solution [7, 20, 21].
This technique starts and completes the meshing in a systematic, ordered and highly
programmable manner. Unlike Delaunay’s triangulation, AFT does not depend on
any geometrical criterion, but on logical steps to mesh a given complicated domain.
Logical steps to mesh a given complicated domain. The AFT was developed by
Peraire et al. [19]. The AFT is composed of four distinct phases of meshing a domain
in a desired way. They are.
• Generation of the background mesh
• Geometrical representation of the domain
• Generation of boundary nodes
• Generation of elements.

16.8.4.1 Structured and Unstructured Mesh

Structured and unstructured quadrilateral meshes are shown in Figs. 16.7 and 16.8.
From the figure, it can be seen that the bisection is made to regular mesh lines
running to the boundary to retain the four-neighbour structure. This strategy is simple
to implement. However, it clearly refines many more elements than necessary. The
customary way of avoiding the excess of refinement is to introduce irregular nodes
when the edges of a refined element meet the middle of a coarser one (Fig. 16.8).
The mesh is no longer structured.

16.8.4.2 The Background Mesh

Our discussion is limited to triangular elements only. The background mesh is simply
a structured or unstructured mesh (Figs. 16.9 and 16.10). The first step in mesh
generation is to construct by hand coarse background mesh of triangular elements,
which completely covers the solution domain of interest as shown in Fig. 16.9. The
396 16 Adaptive Finite Element Analysis

background mesh may have any arbitrary orientation and need not align with the
boundaries of the domain. For the generation of triangular elements, it is convenient
to define a node spacing function δ. It represents the side of an equilateral triangle
that is yet to be generated. At each node on the background mesh, the nodal value δ
is required to be specified at the beginning of the mesh generation. The background
mesh is used to provide a piecewise linear spatial distribution of the parameter δ over
the mesh to be generated.
For uniformly distributed elements in a domain, very simple shaped background
grids are sufficient (Fig. 16.9). Relatively complicated background grids are required
for possible variation of the sizes of the elements in the domain (Fig. 16.10). During
the mesh generation process, local values of δ will be obtained by linear interpolation
of the triangles of the background grid between specified values. In course of the
adaptive analysis, the current mesh becomes background mesh for the next mesh
generation, thus allowing for a more flexible definition of spatial variation of the
mesh generation parameter δ. Referring to Eq. (16.34), h new is defined for an element.
During adaptive mesh refinement, the δ-value of a node can be computed as
Σn
h enew
δ= e=1
(16.35)
m

Fig. 16.9 A simple

background mesh

Fig. 16.10 A complex

background mesh
16.8 Adaptive Mesh Generation 397

Fig. 16.11 a Boundary End points Boundary nodes

segments, end points, loops
and meshing, b Conversion
of domain into polygon 1
object upon joining the
boundary nodes 2

7
3 6

8
4
5
(a) (b)

where m is a number of elements connected to a vertex node or simply a node.

16.8.4.3 Geometrical Representation of the Domain

In advancing front technique, the boundary of the solution domain is represented

by the union of closed loops of curved segments. There is only one closed loop for
simple connected regions whereas for multi-connected regions, there will be as many
internal loops as the number of openings inside the domain. Each loop is composed
of convenient number of segments and each segment will have definite starting and
ending points. End-points are always the intersection points of a segment with its
adjacent segments.
Thus as shown in Fig. 16.11, the domain has two closed loops: one, the outer
boundary and the other, the inner hole. The outer loop has segments, and the inner
loop has two curvilinear segments, 7 and 8. However, the maximum number of
segments to define a loop is arbitrary. The segments of the interior are defined in a
clockwise fashion. This means that as the boundary curve is traversed, the region to
be triangulated lies always to the left as shown (Fig. 16.11).
The curvilinear boundary segments can be approximated by spline functions. The
spline function is represented by control points that belong to boundary segments.
The number and position of control points selected by the user in each segment should
be done in such a way that the spline curves can accurately describe the boundary
segment of the loop and consequently the entire boundary of the domain. The control
points should include the end points of a segment (Fig. 16.12).

16.8.4.4 Generation of Boundary Nodes

After defining each segment of the boundary with the help of spline functions, next
step is to generate the nodes on the boundary of the domain of interest. The first step
398 16 Adaptive Finite Element Analysis

Fig. 16.12 Curve with

spline interpolation

1 2 3 4 5 6 7

li
s

towards the generation of boundary nodes is to place the points of intersection of

the segments. Additional nodes of the boundary segments of the entire domain have
to be placed before beginning the process of generation of triangles. Each boundary
segment is then considered in turn and nodal points are generated in boundary
segments with spacing of points being determined by interpolated values of δ over
the background mesh. Thus, the boundary domain is discretized into a polygon with
sides connected to each other at a boundary node (Fig. 16.11) by generating nodes
on the spline curves which represent boundary segments.
Suppose that the boundary of the domain consists of M segments. For a segment
of typical length li (i = 1, 2, 3, . . . , M), uniformly distributed sample points are
first placed along the curve with spline interpolation (Fig. 16.12). A piecewise linear
interpolation of nodal spacing values on the midpoint are lengths (obtained by joining
two successive sample points) for all arc lengths are to be obtained (Fig. 16.13). Then
(l 1
the trapezoidal rule may be applied for evaluating the integral, Ai = 0 i δ(s) dl where
δ(s) is termed as node spacing function, s being the arc length of the segment i.
Let Hi be the number of sides and Hi+1 is the number of boundary nodes of the
chosen segment, then Hi is chosen to be the nearest integer to

Fig. 16.13 Nodal spacing

value at midside

B
M
A
16.8 Adaptive Mesh Generation 399

1
1 δ6
1 δ5
1 δ4
δ2
1 1
δ1 δ3

1 3 4 5

S0 K=1 S1 n
1 2 3 4 5 6
S S
SK

Fig. 16.14 Curves approximated as straight lines

(li
1
Ai = dl (16.36)
δ(s)
0

note that Hi is an integer.

Once the number of nodes is obtained, their positions are to be determined
(Fig. 16.14). The position of the node n k (k = 0, 1, 2, . . . Hi ) to be generated by
the boundary curve is defined by SK and t is defined by the equation given below.
( SK
Hi 1
K = dl (16.37)
Ai 0 δ(s)

Similar procedure is adopted for other segments.

Equation (16.37) can be rewritten as
( sK
K Ai 1
= dl (16.38)
Hi 0 δ(s)

With K = 1, the distance of the first node from the 0th node is to be determined.
By solving Eq. (16.36), the area of each trapezoid is to be stored. Adding these
areas from s = 0, it will indicate the location of the node at a particular trapezoid.
The distance of the node for that trapezoid is to be determined. Let the node fall in
fourth trapezoid (Fig. 16.14). The equation of the edge of the trapezoid is given by
(Fig. 16.14)

1
= mS + c (16.39)
δ(s)
400 16 Adaptive Finite Element Analysis

assuming a straight line variation considered at the fourth station. The boundary
conditions are:
1 1
S = 0, = (16.40a)
δ(s) δ(4)
1 1
and S=s = (16.40b)
δ(3) δ(s)

Substituting the conditions of Eqns. (16.40a) and (16.40b) into Eq. (16.39), m and
c can be evaluated.
Therefore,
[ ]
Ai 1 1
K = Area of the trapezoid [(1) + (2) + (3)] + + /2 × Δ
Hi δ(4) δ(Δ)
(16.41)

where Δ = SK − S4
Likewise, values of other node positions in a segment are obtained.

16.8.4.5 Generation of Triangular Elements

Before the generation starts, a generation front has to be set up. At the start of the
process, the front consists of a sequence of straight line segments, which connect
consecutive boundary nodes. During the generation process, any straight line segment
that is available to form an element side is termed active whereas any segment that
is not active is removed from the front. Thus while the domain boundary will always
remain the same, the generation front will change continuously and has to be updated
whenever a new element is found. The updating process is illustrated in Fig. 16.15.
The element generation steps are as follows [20, 22].
1. An active side (with nodes A and B) in the front is chosen as a base.
2. Determine the node spacing value δ M at the midpoint M of AB by interpolating
about the background mesh (Fig. 16.13).
Determine δ1 according to
⎧
⎨ 0.55AB i f δ M < 0.55AB
δ1 = δ M i f 0.55 AB < δ M < 2 AB
⎩
2 AB i f δ M > 2 AB

The above inequalities have been suggested by Peraire et al. [19] to ensure geomet-
rical compatibility and also to ensure that element with excessive distortion is not
generated.
3. Construct a point c, δ1 from M (Fig. 16.13)
16.8 Adaptive Mesh Generation 401

3 2 1

4
5
17 15
6
16 18
14
7 19
8
9 10 12 13
11

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 ⊗ 19 20
2 3 4 5 6 7 8 9 10 11 12 13 14 15 1 17 18 19 ⊗ 20 16

3 2 1

4
5
17 15
6
16 18
14
7 20 19
8
9 13

Fig. 16.15 Updation of front during triangle generation process

4. Determine all the active nodes (nodes still available to be a part of the new
element), which lie within the circle with centre at C and radius n1AB . However,
there is no unique choice for n1 (n1 = 5 has been chosen in [19]). These nodes are
ordered according to their distance from C with the first node in the list closest
to C and are denoted by N1 , N2 . . . , Ne (N 1 being the first node).
5. If AN1 > 1.5δ1 and B N1 > 1.5δ1 , then place C at the head of the list.
6. The required connecting point N j is then taken to be the first node in the list that
is such that the interior of the triangular ABN j does not contain any other node
N i in the list (excluding C) and such that the line MN j does not intersect any
existing side of the front. The coordinates are transferred back to the original
space and the front is updated as indicated previously.
7. Triangle generation process ceases when the number of active sides of the front
reduces to zero.
402 16 Adaptive Finite Element Analysis

For comparison purpose, the effectivity index of error estimates is presented. This
index is defined as
||ees ||
θ= (16.42)
||e||exact

16.8.4.6 Example of h-Adaptive Finite Element Analysis

The test problem that is considered concerns the plane stress analysis of a square plate
with a square cutout loaded by uniform tension on opposite outer edges as shown in
Fig. 16.16. Derefinement schemes have not been implemented. Due to the symmetry,
one quarter of the plate is modelled as shown in Fig. 16.16b. This example has been
used extensively in the literature as a benchmark problem for adaptive refinement
technique. The exact value of ||U ||2 is 0.311329399.
Table 16.1 summarizes the h-adaptive finite element analysis of the domain for a
target error of 3%. Figure 16.17 shows h-adaptive meshing sequence.

(a)

E = 1.0E5
50 v = 0.3
Thickness = 1.0
Plane Stress Conditions
P=1.0
50

50 50

(b)

Fig. 16.16 L-shaped domain

References and Suggested Readings 403

Table 16.1 Summary of h-adaptive finite element analysis

Target error 3.0%
Mesh Nelm Node ||ees ||2 μes % θ
1 18 49 9.87365e–3 17.8 0.895
2 138 307 1.68943e–3 4.48 1.64
3 285 620 1.76370e–4 2.04 1.168

Fig. 16.17 h-adaptive match

sequence

(a) (b)
Mesh 1: Starting Mesh Mesh 2

(c)
Mesh 3

References and Suggested Readings

1. O.C. Zienkiewicz, J.Z. Zhu, The superconvergent patch recovery and a posteriori error estimate,
part I: the recovery technique. IJNME 33, 1331–1364 (1992)
2. O.C. Zienkiewicz, J.Z. Zhu, The superconvergent patch recovery and a posteriori error estimate,
part II: error estimates and adaptivity. IJNME 33, 1365–1382 (1992)
3. W.H. Press, S.A. Teukdsky, W.I. Velteshink, B.P. Flannery, Numerical recipes in FORTRAN:
the art of scientific computing (Cambridge University Press, New Delhi, 1992)
4. O.C. Zienkiewicz, J.Z. Zhu, Adaptivity and mesh generation. IJNME 32, 183–210 (1991)
5. O.C. Zienkiewicz, Y.C. Liu, G.C. Huang, Error estimation and adaptivity in flow formulation
of forming problems. IJNME 25, 23–42 (1988)
6. M. Ainsworth, J.Z. Zhu, A.W. Craig, O.C. Zienkiewicz, Analysis in Zienkiewicz—Zhu error
estimator in the finite element method. IJNME 28, 2161–2174 (1989)
404 16 Adaptive Finite Element Analysis

7. O.C. Zienkiewicz, G.C. Huang, Y.C. Liu, Adaptive FEM computation of forming process—
application to porous and nonporous materials. IJNME 30, 1527–1553 (1990)
8. O.C. Zienkiewicz, J.Z. Zhu, N.G. Gong, Effective and practical adaptive analysis proceedings
for the finite element method. IJNME 28, 879–891 (1989)
9. O.C. Zienkiewicz, J.Z. Zhu, Error esimations and adaptive refinement for plate bending
problem. IJNME 28, 2839–2883 (1989)
10. O.C. Zienkiewicz, J.Z. Zhu, A simple error estimator and adaptive procedures for practical
engineering analysis. IJNME 24, 337–351 (1987)
11. O.C. Zienkiewicz, R.L Taylor, The Finite Element Method, v.1 and v.2 (McGraw-Hill
International Edition, 1988)
12. O.C. Zienkiewicz, D.V. Phillips, An automatic mesh generation scheme for the plane and
curved surfaces by isoparametric coordinates. IJNME 3, 519–528 (1971)
13. W.J. Gordon, C.A. Hall, Construction of curvilinear coordinate system and application to mesh
generation. IJNME 7, 461–477 (1973)
14. B.N. Delaunay, Sur la sphere vide Izvesia Academia Nauk. SSSR, VII Seria 72, 793–800
(1934)
15. S.H. Lo, Delaunay triangulation of nonconvex planar domains. IJNME 28, 2695–2707 (1989)
16. H. Borouchaki, P.I. George, Optimal Delaunay point insertion. IJNME 39, 3407–3437 (1997)
17. R.W. Lewis, Y. Zheng, A.S. Usmani, Aspects of adaptive mesh generation based on domain
decomposition and Delaunay triangulation. Finite Elem. Anal. Des. 20, 47–70 (1995)
18. M.S. Shephard, Approaches to automatic generation and control of finite element meshes.
Appl. Mech. Rev. Trans. ASME 41, 169–188 (1988)
19. M.S. Shephard, in Automatic Generation of Finite Element Models Solving Large Size Problems
in Mechanics, ed. by M. Papradrakis (John Wiley, New York, 1993), pp. 431–460
20. J. Peraire, M. Vabdati, K. Morgan, O.C. Zienkiewicz, Adaptive meshing for possible flow
computations. J. Comput. Phys. 72, 449–466 (1987)
21. S.H. Lo, A new mesh generation scheme for arbitrary planar domains. IJNME 21, 1403–1425
(1985)
22. K.S.R.K. Murthy, An investigation on h-adaptive finite element analysis of planar mixed
mode crack problems. Doctoral dissertation, Department of Ocean Engineering and Naval
Architecture, Indian Institute of Technology, Kharagpur, 2003
Chapter 17
Geometrical Nonlinear Finite Element
Analysis

Abstract All the chapters of the book have so far dealt with problems that are linear.

17.1 Introduction

All the chapters of the book have so far dealt with problems that are linear.
Nonlinearity in structural mechanics may be classified under the following two heads:
(a) Material nonlinearity
(b) Geometrical nonlinearity.
Material nonlinearity arises in problems where the stress is not linearly propor-
tional to strain. This results due to the presence of plasticity, creep or other complex
relations. For this category of problems, small strains and small displacements are
considered, which signifies that the geometry of the elements remains basically
unchanged during the loading process.
Geometric nonlinearity arises from the consideration of the nonlinear strain–
displacement relationship. It encompasses large strains and large displacements. For
example, in the linear analysis of a plate, the middle surface is assumed to remain
unstretched. If the membrane action is considered in the plate flexure, it will result in
the nonlinear strain–displacement relationship. In many cases, large displacements
may occur without causing large strains. The elastic postbuckling behaviour of the
structure comes under this category.
Nonlinear problems may be encountered that involve both the material nonlin-
earity and geometric nonlinearity. It will, thus, involve nonlinear constitutive
relationship as well as nonlinear strain–displacement relation.

17.2 Nonlinear Equation Solving Procedures

A set of nonlinear equations in the finite element analysis can be written as

{ }
{ ψ }{X } = [ K S {X } ]{ X } − P = { 0 } (17.1)
© The Author(s) 2022 405
M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_17
406 17 Geometrical Nonlinear Finite Element Analysis

A few basic techniques exist for obtaining the solution of nonlinear equations.
Fundamentals of some of them are discussed here.

17.2.1 Direct Iteration Method

Figure 17.1 illustrates the direct iteration method of one variable. For a multivariable
problem, the same concept can be extended. In this direct iteration procedure, one
starts with an initial value {X } = {X }0 .
The first approximation of {X } is thus
[ ]−1 { }
{ X }1 = K 0s P (17.2)
[ ]
where K S0 = [ K { X } ]
Next, [K S {X }] is modified according to {X }1 . The second approximation of {X }
is
[ ]−1 { }
{ X }2 = K S1 P (17.3)

The process is repeated and after nth iteration, it is

{ }
{ X }n = ( [K ]n−1
s )−1 P (17.4)

The iteration is continued till the values of the two successive iterations are quite
close. In order to list the closeness, various norms and convergence criteria are used.
One such is given below.

Fig. 17.1 Direct iteration P

technique P = K.X

0 1 2 3 4
X X X X X X
17.2 Nonlinear Equation Solving Procedures 407

{ e } = { X }n − { X }n−1 (17.5)

| e | = max ei (maximum value)

or
/
|e| = {e}T {e} (17.6)

Convergence is attained when

|e| = ≤ α

where α is the tolerance of error.

While using direct iteration procedure, one may have to be careful on the possi-
bility of the divergence of the results. One of the disadvantages of the method is that
the direct iteration procedure is to be used for every load level and the results of a
particular load level cannot be used for next load level.

17.2.2 Newton–Raphson Method

In order to solve Eq. (17.1), suppose an approximate solution of the following form
has been reached.

{X } = {X }n (17.7)

The function {ψ} ( {X }) is expressed in terms of Talyor’s series and higher order
terms are ignored to get the improved solution as
( )
( ) ( ) d{ψ}
{ψ} { X }n+1 = { ψ } { X }n + { ΔX }n = 0 (17.8)
d{ X } n

where

{X }n+1 = {X }n + {ΔX }n (17.9)

In the above equation

d{ψ}
= [K ({X })]T (17.10)
d{X }

[K ({X })]T represents the tangent stiffness matrix.

Improved values of {X }n+1 are obtained from Eq. (17.9) by calculating
408 17 Geometrical Nonlinear Finite Element Analysis

Load

KS.X

P
1
KT
n KS = Secant stiffness
P –KS.X
KT = Tangent stiffness
n
X = Displacement after nth. iteration
n+1
X = Displacement after (n+1)th iteration
ΔX = Incremental displacement
P = Load level
KS n n
P = KS.X
1
displacement
n n+1
X X
n
ΔX

Fig. 17.2 Newton–Raphson method

( )−1 ( )
{ΔX }n+1 = [K ]nT {P} − [K ]nS {X }n (17.11)

Figure 17.2 illustrates the Newton–Raphson method for a single variable. For
every incremental solution, a new set of linearized equation is required to be solved.

17.2.3 Modified Newton–Raphson Method

In Newton-Raphson method, Eq. (17.11) is to be solved at each iteration stage that is

time consuming. So the method is modified to reduce the computational effort. An
approximation given below is introduced

[K ]nT = [K ]0 (17.12)

The modified algorithm then is

( )−1 ( )
[K ]0 {P} − [K ]nS {X }n = {ΔX }n+1 (17.13)

The solution procedure thus obtained will save the cost of solution but the rate
of convergence will now be slow. Figure 17.3 gives an illustration of the procedure.
For certain problems, if the process is found to be slow, then [K ]T may be updated
after some iterations.
17.2 Nonlinear Equation Solving Procedures 409

Fig. 17.3 Modified P

Newton–Raphson method P=K.X

0 1 2 3 4
X X X X X X
0 displacement
ΔX 3
ΔX
1 2
ΔX ΔX

17.2.4 Incremental Techniques

The solution of nonlinear equations can be carried out by treating the nonlinearity
as piecewise linear. In this technique, the load is divided into a number of small
increments that may be equal or unequal. One proceeds from one load level to its
next increment, and the solution for this increment is assumed to be linear. Solution
of displacements is thus obtained for each increment of load. In order to obtain the
total displacement at any stage of loading, all these displacement increments are to
be added.
Let the initial load and displacement given by {P}0 and {X }0 for an undeformed
body, are null vectors. If the total load is divided into M increments, then

Σ
M
{P} = {P}0 + {ΔP}i (17.14)
i=1

After applying nth increment, the load is

Σ
n
{P}n = {P}0 + {ΔP}i (17.15)
i=1

Similarly, the displacements after nth increment are

410 17 Geometrical Nonlinear Finite Element Analysis

Fig. 17.4 Incental technique P

Incremental solution
n
P 1
n–1
KT
n–1
P
Exact
solution

n
ΔX

n–1 n
X X X

Σ
n
{X }n = {X }0 + {ΔX }i (17.16)
i=1

The increments of displacements are to be evaluated by considering the stiffness

matrix at the end of the previous increment
[ ( )]i−1
K {X }i−1 T {ΔX }i = {ΔP}i for i = 1, 2, 3, . . . , M (17.17)

.
It is to be noted that the tangent stiffness matrix is used in the formulation of
Eq. (17.17). The incremental technique is explained in Fig. 17.4

17.3 Formulation of the Geometric Nonlinear Problem

There are two approaches used for the formulation of the geometric nonlinear
problem—the Eulerian approach and the Lagrangian approach. In the former, the
fsormulation is done on the basis of changed geometry at each stage. In the
Lagrangian method, the displacements are referred to the original configuration.
The treatment here is limited to the Lagrangian approach only. In this context, two
different notations are quite common. Mallet and Marcal [1, 2] have formulated the
nonlinear problem on the basis of N-notation. The notations of Zienkiewicz and his
coworkers [3] are referred as B-notation. It can be shown that final matrices obtained
by both notations are identical.
17.3 Formulation of the Geometric Nonlinear Problem 411

17.3.1 Equilibrium Equations

The equilibrium equations may be obtained by application of virtual principles as

(
{ψ} = [B]T { σ }d V − { P }e = { 0 } (17.18)
v

where { ψ } denotes the sum of external and internal generalised forces and [B] is
obtained from the following relationship.

d{ ε } = [ B] d{ X }e (17.19)

Again matrix [B] may be expressed as

[B] = [B]0 + [B({X })] L (17.20)

where [B] 0 is the linear part and is same as that derived in the earlier chapters. [B] L
is the nonlinear part and is dependent on displacements.
If the strains are reasonably small, we can still write the elastic relationship

{σ } = { D}{ε} (17.21)

where
( )
1
{ ε } = [ B ]0 + [ B ] L { X }e (17.22)
2

It will be explained later how this is obtained.

With the help of Eqns. (17.20)–(17.22), Eq. (17.18) becomes
( ( )
1
([ B ]0 + [ B ] L )T [ D ] [ B ]0 + [ B ] L d V { X }e − { P }e = {0}
2
v

or,

[K ] S {X }e = {P}e = {0} (17.23)

where [K ] S is the secant stiffness matrix and is given by

( ( )
1
[K ] S = ([B]0 + [B] L )T [D] [B]0 + [B] L d V (17.24)
2
V

Equation (17.24) when expanded becomes

412 17 Geometrical Nonlinear Finite Element Analysis
( (
1
[K ] S = [B]0T [D][B]0 d V + [B]0T [D][B] L d V
2
V V
( ( (17.25)
1
+ [B]TL [D] [B]0 d V + [B]TL [D] [B] L d V
2
V V

Equation (17.25) can be written as

[K ] S = [K ]0 + [K ] L S (17.26)

where, [K ] 0 = linear part of the stiffness matrix

(
= [B]0T [D] [B]0 d V
V

and [K ] L S = nonlinear part of the secant stiffness matrix

( ( (
1 1
= [B]0T [D][B] L d V + [B ]TL [D ] [B ]0 d V + [B ]TL [D ] [B ] L d V
2 2
v V V
(17.27)

17.3.2 Incremental Equilibrium Equation

The tangent stiffness matrix given by Eq. (17.11) may be obtained from the incre-
mental equilibrium equations. Taking appropriate variation of Eq. (17.18) with
respect to {X }, this may be obtained as
( (
d{ψ} = d[B]T {σ }d V + [B]T d{σ }d V − d{P}e = [K ]T d{X }e − d{P}e
V V
(17.28)

Substituting Eqs. (17.20) and (17.22) into Eq. (17.28), it yields to

(
d([B]0 + [B] L )T { σ }d V
V
(
+ ([B]0 + [B] L )T {D}([B]0 + [B] L ) d V d{X }e + d{P}e = {0}
V

or,
17.4 Large Deflection Analysis of Plates in b-notation 413

[K ]σ d{X }e + ([K ]0 + [K ] L T ) d{X }e = d{P}e (17.29)

where
(
[K ]0 = [B]0T [D] [B]0 d V (17.30a)
v
( ( (
[K ] L T = [B]oT [D] [B] L d V + [B]TL [D] [B]0 d V + [B]TL [D] [B] L d V
v v v
(17.30b)

and,
(
[K ]σ d{X }e = d[B]TL {σ } d V (17.30c)
v

It will be seen that [K ]σ is a symmetric matrix dependent on the stress level. This
matrix is referred to as the initial stress matrix or the geometric matrix.
Thus, the tangent stiffness matrix consists of three parts. They are

[K ]T = [K ]0 + [K ] L T + [K ]σ (17.31)

17.4 Large Deflection Analysis of Plates in b-notation

Figure 17.5 indicates the general pattern of variation of deflection at a point in the
plate with the load. It may be noted that the non-linear deflection in plate is always
less than linear deflection.
In such cases, lateral displacements are coupled. Taking the middle plane of the
plate as the reference plane for the analysis, strain matrix can be given by
⎧ ∂u ⎫ ⎧ ( )2 ⎫
1 ∂w
⎪ ⎪ ⎪
⎪ 2( ∂ x ) ⎪
⎪
⎪
⎪ ∂x ⎪
⎪ ⎪
⎪ 2 ⎪⎪
⎪
⎪ ∂v ⎪
⎪ ⎪
⎪ ∂w ⎪
⎪
⎪
⎪ ∂y ⎪
⎪ ⎪
⎪
1
⎪
⎪
⎪
⎪ ∂u ∂v ⎪
⎪ ⎪
⎪ (
2
)
∂(y ) ⎪
⎪
⎨ ∂y + ∂x
⎪ ⎪
⎬ ⎪
⎨ ∂w ∂w ⎪ ⎬
∂x ∂y
{ε} = −−− + (17.32)
⎪
⎪ ⎪
⎪ ⎪
⎪ −−− ⎪
⎪
⎪ − ∂∂ xw2 ⎪ ⎪ ⎪
2
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ∂ w
2 ⎪
⎪ ⎪
⎪ 0 ⎪
⎪
⎪ − ∂ y22
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ 0
⎪
⎪
⎪
⎩ 2∂ w ⎭ ⎪
⎩ ⎪
⎭
− ∂x ∂y 0

or,
414 17 Geometrical Nonlinear Finite Element Analysis

Fig. 17.5 Load–deflection P

curve of a plate
Large deflection theory

Small deflection theory

( )
{ε p }
{ε} = { } (17.33)
εb

Similarly,{ε}, {σ } and [ D ] can be split into two parts: one is related to inplane
action and other is related to the bending action.
Therefore, the stress resultant vector is
( )
{ σ p}
{σ } = { b } (17.34)
σ

where
⎧ ⎫ ⎧ ⎫
{ } ⎨ Nx ⎬ { } ⎨ Mx ⎬
σ p
= Ny , σ b
= My (17.35)
⎩ ⎭ ⎩ ⎭
Nx y Mx y

And the rigidity matrix is

[ ]
[ D p ] [ [ 0] ]
[ D] = (17.36)
[0] Db

For the isotropic late problem, the submatrices in the above equation are
⎡ ⎤
[ ] 1v 0
Et ⎣
D p
= v1 0 ⎦ (17.37a)
1 − v2
0 0 1−v
2
17.4 Large Deflection Analysis of Plates in b-notation 415
⎡ ⎤
[ ] 1ν 0
Et 3
D b
= ( )⎣ν 1 0 ⎦ (17.37b)
12 1 − ν 2
0 0 1−ν
2

From Eqns. (17.32) and (17.33), one obtains

{ } { } { }
εp = εp 0 + εp L (17.38)

where,
⎧ ⎫
∂u
⎪
⎨ ⎪
⎬
{ p} ∂x
∂v
ε 0= ∂y (17.39a)
⎪
⎩ ∂u + ∂v ⎪ ⎭
∂y ∂x
⎧ ( )2 ⎫
⎪
⎪
1 ∂w ⎪
{ p} ⎨ 2( ∂ x ) ⎪⎬ 2
ε L= 1 ∂w (17.39b)
⎪
⎪
2 ∂y ⎪
⎩ ∂w ∂w ⎪
∂x
· ∂y ⎭

Equation (17.30) can be written as

⎡ ⎤
∂w
0 ( )
{ } 1⎢ ∂x
∂w ⎥
∂w
∂x
1
ε p
L
= ⎣ 0 ∂y ⎦ ∂w = [A]{θ } (17.40)
2 ∂w ∂w ∂y 2
∂y ∂x

where,
⎡ ⎤
∂w
∂x
0
⎢ ∂w ⎥
[A] = ⎣ 0 ∂y ⎦ (17.41)
∂w ∂w
∂y ∂x
( )
∂w
{θ } = ∂x (17.42)
∂w
∂y

The deflection w of the plate can be expressed as

w = [N ]b {X }e (17.43)

where [N ]b is the shape function matrix for w.

Equation (17.42) can thus be written with the help of Eq. (17.43) as:
( ) [ ]
∂ ∂
∂x
[N ]b {X }e ∂x
[N ]b
{θ } = ∂ = ∂ {X }e = [G]{X }e (17.44)
∂y
[N ]b {X }e ∂y
[N ]b
416 17 Geometrical Nonlinear Finite Element Analysis

[ ]
∂
∂x
[N ]b
[G] = ∂ (17.45)
∂y
[N ]b

Combining Eqns. (17.40) and (17.44) yields

{ } 1
εp L
= [ A][ G]{ X }e (17.46)
2
From eqns. (17.33) and (17.38), the strain vector can be written as
( ) ( )
{ p}
{ε} = { ε b }0 +
{ ε p }L
= { ε }0 + { ε } L (17.47)
ε 0 {0 }

It has been shown a number of times in the previous chapters that

{ ε }0 = [ B ]0 { X }e (17.48)

Further,
(1 )
[ A ][ G ] 1
{ ε }L = 2 { X }e = [ B ] L { X }e (17.49)
{0 } 2

Therefore,
( )
1
{ ε } = [ B ]0 + [ B ] L { X }e (17.50)
2

This is directly used in Eq. (17.22).

The variation of the strain can be written as
( )
d{ ε p } L
d{ ε } = d{ ε0 } + d{ ε } L = [ B ]0 d{ X }e + (17.51)
{0}

Taking the variation of Eq. (17.40) yields

( )
{ } 1 1 1
d ε p
L
= d [ A ] { θ } = d[ A ]{ θ } + [ A ]d{ θ } (17.52)
2 2 2

We may note that

⎡ ( ∂w ) ⎤
d ∂x (0 ) ( )
⎢ ⎥
d[A]{Y } =⎢ 0 d ∂w ⎥ y1
⎣ ( ) ( ∂ y ) ⎦ y2
d ∂w
∂y
d ∂w
∂x
17.4 Large Deflection Analysis of Plates in b-notation 417
⎡ ⎤ ( )
y1 0 ∂w
=⎣ 0 y2 ⎦ d ∂w
∂x (17.53)
∂y
y2 y1

It gives

d[ A ] { θ } = [ A ] d{ θ } (17.54)

Combining Eqns. (17.52) and (17.54) yields

{ }
d ∈ p L = [ A ] d{ θ } = [ A ] [ G ] d{ X }e (17.55)

Hence, (17.56)
p
[ B ]L = [ A ] [ G ] (17.56)

and,
[ ]
[ A] [G]
[ B ]L = (17.57)
[ 0]

Combining Eqns. (17.51), (17.55), (17.56) and (17.57) yields

d{ ∈ } = ([ B ]0 + [ B] L ) d{ X }e (17.58)

With [B]0 and [B] L , the stiffness matrix for the plate element is given by
Eq. (17.25). It should be noted that the volume integral will be replaced by area
integral, as the contribution across the thickness direction is taken into account in
the rigidity matrix (Eq. (17.37).
Similarly, the stiffness matrix can be obtained from Eq. (17.29). The derivation
of the geometric matrix [K ]σ is presented below.
Equation (17.30), is reproduced.
(
[K ]σ d{X }e = d[B]TL {σ }dv
v

Using Eqs. (17.34), (17.35), (17.41) and (17.57), the above equation becomes

( [ ]⎧ ⎫
∂w ∂w ⎨ N x ⎬
d ∂x
0 d ∂y
[K ]σ d{X }e = [G]T N dA
0 d ∂w
∂y
d ∂w
∂x
⎩ y ⎭
A Nx y
( ( )
∂w ∂w
T d ∂ x Nx + d ∂ y Nx y
= [G] dA
d ∂w N + d ∂w
∂y y
N
∂x xy
A
418 17 Geometrical Nonlinear Finite Element Analysis

( [ ]( )
Nx Nx y d ∂w
∂x
= [ G] T
dA
Nx y N y d ∂w
∂y
A
( [ ] ( ∂w )
Nx Nx y ∂x
= [G]T d ∂w dA (17.59)
Nx y Ny ∂y
A

Using Eq. (17.42), Eq. (17.59) becomes

(
[K ]σ d{X }e = [ G]T [ S]d{ θ }d A (17.60)
A

where
[ ]
Nx Nx y
[S] = (17.61)
Nx y N y

Replacing { θ } from Eq. (17.44) with Eq. (17.60), yields

(
[K ]σ d{ X }e = [ G]T [ S][ G] d A d{ X }e (17.62)
A

Therefore,
(
[ K ]σ = [ G]T [ S][ G] d A (17.63)
A

17.5 Large Deflection Analysis of Plates in n-notation

It can be shown that [1, 2] the equilibrium equations can be written as

( )
1 1
[N0 ] + [N1 ] + [N2 ] {X }e = {P}e (17.64)
2 3

where,
(
[N0 ] = [B]0T [D] [B]0 d A (17.65a)
A
17.5 Large Deflection Analysis of Plates in n-notation 419
(
( T )
[N1 ] = [B]0 [D] [B] L + [B]TL [D] [B]0 + [G]T [S]0 [G] d A (17.65b)
A
(
( T )
[N2 ] = [B] L [D] [B] L + [G]T [S] L [G] d A (17.65c)
A

The matrices [S]0 and [S] L are defined later.

In the above equations, [N0 ] is independent of displacement {X }e , and [N1 ] is
linearly dependent on {X }e and [N2 ] is quadratically dependent upon {X }e . With
respect to B-notation, the secant stiffness matrix Eq. (17.25) results in an unsym-
metric matrix while the secant stiffness matrix given by Eq. (17.64) in N-notation
results in a symmetric matrix. It can be shown that Eqs. (17.25) and (17.64) are
identical.
Substituting the value from Eqs. (17.65a)–(17.65c) into Eq. (17.64), yields
( (
1
[K ]s = [B]0T [D][B]0 d A + [B]0T [D][B] L d A
2
A A
( (
1 1
+ [B]TL [D][B]0 d A + [B]TL [D][B] L d A
2 3
A A
( (
1 1
+ [G] [S]0 [G]d A +
T
[G]T [S]0 [G]d A (17.66)
2 3
A A

Explicitly,

[S] = [S]0 + [S] L (17.67)

where [S] is given by Eq. (17.61). [S] 0 corresponds to the linear part of the strain
{ ε p }0 (see Eq. (17.39a)) and [S] L corresponds to nonlinear part of strain { ε p } L
(Eq. 17.39b).
Thus,
[ ] [ ]
N x0 N x0y N xL N xLy
[S ] 0 = , [S] L = (17.68)
N x0y N y0 N xLy N yL

where
⎧ ⎫ ⎧ ⎫
⎪
⎨ Nx ⎪
o
⎬ [ ⎪
⎨ N xL ⎪
⎬ [ ]{ }
] { p}
N y0 = D p ε 0 , N yL = D p ε p L (17.69)
⎪
⎩ N0 ⎪ ⎭ ⎪
⎩ NL ⎪ ⎭
xy xy
420 17 Geometrical Nonlinear Finite Element Analysis

Now, taking the fifth integral of Eq. (17.66) it may be processed with the help of
Eq. (17.44) as
( (
[G]T [S]0 [G] d A { X }e = [G]T [S]0 {θ } d A (17.70)
A A

Using Eq. (17.42) and (17.68), the equation becomes

( ( [ ]( )
∂w
N x0 N x0y ∂x
[G] [S]0 [G]d A{X }e =
T
[G] T
∂w dA
N x0y N y0 ∂y
A A
( ( )
N x0 ∂w
∂x
+ N x0y ∂w
∂y
= [G] T
dA
N x0y ∂w
∂x
+ N y0 ∂w
∂y
A
⎧ ⎫
[ ] No ⎪
( ∂w ∂w ⎪ ⎨ x ⎬
0 ∂y
= [G]T ∂ x ∂w ∂w N y0 d A (17.71)
0 ∂y ∂x ⎪ ⎩ N0 ⎪ ⎭
A xy

With the help of Eqs. (17.35), (17.37a), (17.39) and (17.41), the above equation
becomes
( (
[ ]{ }
[G] [S]0 [G] d A {X }e = [G]T [A]T D p ε p 0 d V
T

A V
( [ ]( p )
[ ] [D p ] [0 ] { }
= ([ A][G ]) [ 0]
T [ ] { ε b }0 d A (17.72)
[ 0] D b
ε 0
A

Combining Eqs. (17.31), (17.47), (17.48), (17.57) and (17.72) yields

( (
[ G]T [ S] 0 [ G] d V { X }e = [B ]TL [ D] [ B] 0 d A { X }e
V A

Following the steps as above, the last integral of Eq. (17.66) may be processed as
(
[G]T [S] L [G]d A{X }e
A
(
[ ]{ }
= [G]T [A]T D p ε p L d A
A
( [ ]( p )
( ) [D p ] [0] {ε } L
= ([A][G])T [0] [ b] dA
[0] D {0}
A
17.5 Large Deflection Analysis of Plates in n-notation 421
(
1
= [B]TL [D][B] L d A{X }e
2
A
(17.73)

Substituting the appropriate terms from Eqs. (17.72) and (17.73) into the secant
stiffness matrix expressions given by Eq. (17.60), yields
(
[K ] S {X }e = [B]0T [D][B]0 d V {X }e
V
(
1
+ [B]0T [D][B] L d V {X }e
2
v
(
1
+ [B]TL [D][B]0 d V {X }e
2
v
(
1
+ [B]TL [D][B] L d V {X }e
3
v
(
1
+ [B]TL [D][B]0 d V {X }e
2
v
(
1
+ [B]TL [D][B] L d V {X }e
6
v

or
(
[K ] S {X }e = [B]0T [D][B]0 d V {X }e
v
(
1
+ [B]0T [D][B] L d V {X }e
2
v
(
+ [B]TL [D][B]0 d V {X }e
v
(
1
+ [B]TL [D][B] L d V {X }e (17.74)
2
v

Equation (17.74) is identical to Eq. (17.25), the secant stiffness matrix derived in
B-notation.
422 17 Geometrical Nonlinear Finite Element Analysis

17.6 Example of a Pin-Jointed bar

Figure 17.6 presents a pin-jointed bar. The derivation is presented for various matrices
required for carrying out large deflection analysis of this bar [4]. The derivation has
been carried out in the local x–y coordinate system. Nodal displacements are (u1 , v1 )
and (u2 , v2 ).
The displacement within an element may be expressed in the terms of nodal
displacements as
⎧ ⎫
]⎪
⎪ u1 ⎪
( ) [ ⎨ ⎪ ⎬
u N1 0 N2 0 v1
= (17.75)
v 0 N1 0 N2 ⎪⎪ u ⎪
⎩ 2⎪ ⎭
v2

where
1 1 x
N1 = (1 − ξ ) , N2 = (1 + ξ )and ξ = (17.76)
2 2 L
The nodal displacement matrix is
{ }
{ X }eT = u 1 v1 u 2 v2 (17.77)

From the above equation, we get

du 1 du dv 1 dv
= . , = . , dv = a.L .dξ (17.78)
dx L dξ dx L dξ

where a is the cross-sectional area of the bar.

Fig. 17.6 A pin-jointed bar x

y′
2
v

2
u
L

1
ξ=
L
y

0
ξ=
1
v

1
–
ξ=

u1

x′
17.6 Example of a Pin-Jointed bar 423

For this problem, the strain is

[( )2 ( )2 ]
du 1 du dv
εx = + + (17.79)
dx 2 dx dx

or,

{ ε } = { ε }0 + { ε } L (17.80a)

where,
[( )2 ( )2 ]
du 1 du dv
{ ε }0 = , { ε }L = + (17.80b)
dx 2 dx dx

The constitutive relation for the bar is

{σ } = [ D] {ε } (17.81a)

or,

{{ S }0 { S } L } = [ D ]{{ ε }0 { ε } L } (17.81b)

where,

[D] = E (17.81c)

From Eqs. (17.75), (17.76) and (17.78), we get

du d N1 d N2 1 u2 − u1
= u1 + u2 = (−u 1 + u 2 ) = (17.82)
dx dx dx 2L 2L
Therefore,

du u
{ ε }0 = = (17.83)
dx l0

where u = u 2 − u 1 , l0 = 2L
Similarly,

dv v
= (17.84)
dx l0

where v = v2 − v1
Thus, the nonlinear strain is
424 17 Geometrical Nonlinear Finite Element Analysis
[( ) ( ) ]
1 u 2 v 2
{ ε }L = (17.85)
2 l0 l0

Combining Eqns. (17.81), (17.83) and (17.85), yields

( [( ) ( )2 ])
u 1 u 2 v
{ { S }0 { S } L } = E + (17.86)
l0 2 l0 l0

where,
( ) [( ) ( )2 ]
u E u 2 v
{ S }0 = E and { S } L = + (17.87)
l0 2 l0 l0

Now, the nonlinear strain given by Eq. (17.80) may be expressed as

( ∂u )
1 [ ∂u ∂v
]
{ ε }L = ∂x
∂v = [ A ]{ θ }
2 ∂x ∂x
∂x

Thus
[ ∂u ∂v
]
[A] = ∂x ∂x
(17.88)

And {θ } may be expressed in terms of nodal displacement vector {X }e with the

help of Eq. (17.75) as
( ∂u ) ( ∂[N ]
)
[θ ] = ∂x
∂v = ∂x
∂[N ] { X }e = [ G ]{ X }e (17.89)
∂x ∂x

where
[ d N1 d N2
]
0 0
[G] = dx
d N1
dx
d N2
0 dx
0 dx

From the Eq. (17.82), the linear strain matrix may be written as
[ ∂N ∂ N2
] [ ]
[B]0 = ∂x
1
0 ∂x
0 = − l10 0 1
l0
0 (17.90)

Now, the nonlinear strain matrix may be obtained with the help of Eqs. (17.76),
(17.83), (17.84), (17.88) and (17.79) as
[ ]
[ ] ∂ N1 0 ∂ N2 0
[B] L = ∂∂ux ∂v ∂x ∂x
∂x 0 ∂∂Nx1 0 ∂∂Nx2
17.7 Computer Program for Geometrically Nonlinear … 425
[ ]
[ ] −1 0 1 0
v
= lu0 l0 l0
l0 0 − l10 0 l10
1
= 2 [−u − v u v] (17.91)
l0

Substituting Eqns. (17.81c) and (17.90) with Eq. (17.30a) gives

(
[K ]0 = [B]0T [D] [B]o d V
V
⎤⎡
−1
(1
1⎢ ⎥ { }
⎢ 0 ⎥ E 1 −1 0 1 0 dξ
=a L ⎣ ⎦
l0 1 l0
−1
0
⎡ ⎤
1 0 −1 0
Ea ⎢⎢ 0 0 0 0⎥
⎥
= (17.92)
⎣
l0 −1 0 1 0 ⎦
0 0 0 0

In a similar manner, the other matrices required for the formation of secant and
tangent stiffness matrices can be derived.

17.7 Computer Program for Geometrically Nonlinear

Analysis of Plates

The computer program in C and FORTRAN for the solution of geometrically

nonlinear problem of bare plates is given in the attached CD. Isoparametric quadratic
plate bending element presented in Sect. 14.7 has been used for obtaining the solution.
Notations used in the computer program are explained at the beginning. Also
indicated are various steps explained with ‘comment’ statements. Though the present
program enables the analysis of rectangular or parallelogrammic plates, yet with a
little more modification, the computer program can be used for the large deflection
analysis of plates of general shapes.
426 17 Geometrical Nonlinear Finite Element Analysis

The main features of the program are:

1. Automatic mesh generation.
2. Generation of linear element stiffness matrices.
3. Generation of overall linear matrix in single array according to skyline storage
scheme.
4. Generation of element loading matrix.
5. Generation of overall loading matrix.
6. Fitting of boundary conditions of the structure.
7. Calculations of linear nodal displacements.
8. Generation of nonlinear element stiffness matrices—both tangent stiffness
matrix and secant stiffness matrix.
9. Formation of overall nonlinear tangent stiffness matrix and total stiffness matrix
(linear plus nonlinear tangent stiffness matrix).
10. Newton–Raphson iteration technique for solving large deflection problem.
11. Calculation of stress resultants.
A major elegance in the computer program lies in its optimum utilisation of
memory allocations, as it crunches the matrices by removing zeroes. It further has
the provision of taking advantage of the symmetry of the plate. The nonlinear secant
stiffness matrix in the computer program is written in B-notations, so it is unsym-
metric. As such it may be noted that the nonlinear load calculation in the Newton–
Raphson technique is performed at the element level and the overall load needed in
the interaction is achieved from the summation of these element loads. A tolerance
limit is specified for the interaction process.
All stiffness matrices are calculated by using Gauss’s quadrature technique
of numerical integration. Solutions to calculations are obtained by using Gauss’s
solution procedure.
The preparation of input data is very simple. The input data for the plate of
Fig. 17.7 are given in Table 17.1. The computer output for that has been given in
Table 17.2. The plate has all the edges clamped.
Results obtained by using the computer program developed here are presented in
Table 17.3 and compared them with those obtained by Levy[5]. Results have been
presented in the following non-dimensional form.
Central deflection: w = wt Extreme fibre stress: Tcs = σEta2 .
2

where a is the plate dimension and t is the plate thickness.

Table 17.3 presents the results for different load factors. A mesh of 8 × 8 for the
whole plate has been chosen and a tolerance of 1% (α = 0.01) has been considered
in the analysis. Results show good agreement for displacement and stresses.
17.7 Computer Program for Geometrically Nonlinear … 427

a = 300 in, t = 3 in
6
v = 0.316, E = 30 × 10 Psi,
2
q = 0.3 lb/in

Fig. 17.7 A square plate

Table 17.1 Input data for the nonlinear plate problem

300.0 300.0 90.0 0.3
4 4 8 5 2 1 1 12
2 2 2 2
3.0e08 0.316 3.0
2 1.0 1.0
3 -1.0 1.0
5 -1.0 -1.0
9 -1.0 1.0
6 1.0 -1.0
7 -1.0 -1.0
10 1.0 1.0
11 -1.0 1.0
8 1.0 -1.0
12 1.0 1.0
14 1.0 -1.0
15 -1.0 -1.0
163
3.0 3.0
1
1
1 0.09
38.3 63.4 95.0 134.9 184.0 245.0 318.0 402.0 17.79
428 17 Geometrical Nonlinear Finite Element Analysis

Table 17.2 Computer output for the nonlinear plate problem

a b phai ql
300.00 300.00 90.000 .30000
nx ny nn nfree nq nlc npt nspt
4 4 8 5 2 1 1 12
nbon1 nbon2 nbon3 nbon4
2 2 2 2
emodp prp thick
.30000E+09 .31600 3.0000
exl eyl= 75.000000 75.000000
nds(1) 163
points of displacement are 163
nt= 325nbh= 85ncen= 163ntot= 24055nfree4=
10
th1 th2 gr thick
3.0000 3.0000 .00000 3.0000
element stiffness matrix generation starts
ind= 1
nonl= 1
nll, toll = 1 9.000000E-02
alf are 38.300000
the present alf is 38.300000
err= 12.494700
err= 1.045809E-01
err= 9.173726E-02
err= 1.070604E-01
err= 8.132938E-02
displacement are 3.930287E-02
element no ...zi eta,displaceemnt
2 1.0000 1.0000 .20819E-14
420.540600 1330.825000 -46.452770 -17904.030000
-36708.720000 -2439.932000 -418.881700 266438.100000
element no ...zi eta,displaceemnt
3 -1.0000 1.0000 .20819E-14
420.540600 1330.825000 -46.452780 -8049.217000
-33594.600000 -2439.932000 863.809900 264767.800000
element no ...zi eta,displaceemnt
5 -1.0000 -1.0000 .82664E-15
-71.389440 -22.559060 -3.273324 -6561.609000
-2609.120000 -237.560300 1302.911000 1771.461000
element no ...zi eta,displaceemnt
9 -1.0000 1.0000 .82664E-15
-71.389440 -22.559060 -3.273324 -6570.386000
-2636.894000 -237.560300 4903.771000 -17.545690
element no ...zi eta,displaceemnt
6 1.0000 -1.0000 .39303E-01
-43.610440 -199.119700 131.009800 17650.350000
15256.610000 1340.942000 -1655.323000 -272952.800000
element no ...zi eta,displaceemnt
7 -1.0000 -1.0000 .39303E-01
177.987600 -129.094700 127.976000 -4758.072000
8175.547000 1968.741000 271.321500 -264918.400000
element no ...zi eta,displaceemnt
10 1.0000 1.0000 .39303E-01
4.460321 -46.997020 -94.725400 10161.910000
-8440.975000 -2513.179000 967.573800 240480.200000

(continued)
References and Suggested Reading 429

Table 17.2 (continued)

element no ...zi eta,displaceemnt
11 -1.0000 1.0000 .39303E-01
226.058400 23.028000 -97.759250 -12246.510000
-15522.040000 -1885.380000 -3357.863000 234889.800000
element no ...zi eta,displaceemnt
8 1.0000 -1.0000 .27253E-14
-147.375800 -46.570740 -10.109550 -12447.280000
-8275.946000 689.889200 2081.333000 -12821.710000
element no ...zi eta,displaceemnt
12 1.0000 1.0000 .27253E-14
-147.375700 -46.570740 -10.109550 -10883.950000
-3328.700000 689.889000 -2985.272000 18481.820000
element no ...zi eta,displaceemnt
14 1.0000 -1.0000 .12262E-14
218.778900 692.338400 32.382250 -13037.360000
-26479.450000 2429.966000 -1044.802000 -87092.560000
element no ...zi eta,displaceemnt
15 -1.0000 -1.0000 .12262E-14
218.778900 692.338400 32.382250 -4290.706000
-23715.510000 2429.966000 1849.126000 -89187.210000

Table 17.3 Comparison of displacements and stresses for all edges clamped plate
Serial Load factor Central deflection Central stresses Edge stresses
number Exact (17.5) Present Exact (17.5) Present Exact (17.5) Present
1 17.79 0.2370 0.7346 2.60 2.63 5.48 5.34
2 38.30 0.4710 0.4660 5.20 5.48 11.52 11.26
3 63.40 0.6950 0.6862 8.00 7.28 18.03 15.29
4 95.00 0.9120 0.8987 11.10 11.13 25.32 25.07
5 134.90 1.1210 1.1025 13.30 13.87 33.50 33.19
6 184.00 1.3230 1.2979 15.90 16.85 42.40 42.05
7 245.00 1.5210 1.4890 19.20 19.27 52.80 51.88
8 318.00 1.7140 1.6750 21.90 22.04 63.90 62.45
9 402.00 1.9020 1.8526 25.10 24.82 75.80 73.44

References and Suggested Reading

1. R. Mallet, P. Marcal, Finite element analysis of nonlinear structures. J. Struct. Div. Proc. ASCE
94, 2081–2105 (1968)
2. S. Rajasekaran, D.W. Murray, On incremental finite element matrices. PJ. Struct. Div. Proc.
ASCE 99, 2423–2438 (1973)
3. O.C. Zienkiewicz, R.L. Taylor, The Finite Element Method, vol. I and II, 4th edn (McGraw-Hill,
1989)
4. P.C. Wang, Numerical Methods in Structural Mechanics (John Wiley and Sons, 1980)
5. S. Levy, Square plate with clamped edges under normal pressure producing large deflections, in
NASA Technical Note (1942), p. 847
Chapter 18
Finite Element Method of Analysis
of Stiffened Plates

18.1 Introduction

We have seen in the previous chapters that the finite element method has proved to
be the most versatile amongst the different numerical methods. One of the important
categories of structures is the stiffened plate, which has considerable application in
civil, aerospace and ship and various other categories of structures. The stiffened
plate is assumed to consist of two parts: the plate and the stiffener. If the stiffener
centroid is coincident with the plate middle surface, then no inplane stresses will be
developed due to stiffener bending (Fig. 18.1). If the centroid of the stiffener and the
plate middle surface does not coincide, the inplane stress will be developed in the
plate due to stiffener bending (Fig. 18.2). The term eccentric stiffened plates will be
used to designate this structure. The finite element analysis for the first category of
structures will involve only the bending displacements as the unknowns, while the
latter category of structures will involve both inplane and bending displacements as
unknowns.
This stiffener is usually treated as a beam element. In the earlier formulations, the
mesh division of the complete structure is made in such a way that the stiffener is
placed along the nodal line [1, 2]. Later investigations have been conducted to model
the stiffener in such a way that it can be placed anywhere within the plane element
and not necessarily on the nodal lines [3, 4]. It introduces considerably flexibility in
the analysis. The mesh division can be made according to the requirements of the
stress output and resolutions sought and not according to stiffener locations.

18.2 Modeling the Plate and the Stiffener

In the case of concentric stiffened plates, both the plate and the stiffener undergo
bending deformation only. The stiffened plate for such cases is to be analyzed as a

© The Author(s) 2022 431

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_18
432 18 Finite Element Method of Analysis of Stiffened Plates

(a) Rib Stiffened (b) Cellular

(c) Corrugated

Fig. 18.1 Different types of stiffened plates

t
t

t
(a) Concentric (b) Eccentric

Fig. 18.2 Concentrically and eccentrically stiffened plates

plate-bending problem. Only bending and torsional displacements are to be consid-

ered for the analysis of the beam. However, the analysis presented in the following
is that of eccentrically stiffened plate from which the concentrically stiffened plate
can be formulated by eliminating inplane stress part of the former.

18.3 Rectangular Stiffened Plate Bending Element

It is convenient to consider the plate middle surface as the reference axis. Though the
load acts normal to the middle plate of the plate, the plate as such will be subjected
to inplane and bending deformations when the stiffener is placed eccentric to it. In
small deflection theory, the inplane and the plate-bending displacement functions
can be assumed independently.
For structures having regular shaped boundaries, rectangular elements can be
effectively employed in the analysis. A typical rectangular element having sides a
and b is shown in Fig. 18.3. The thickness of the element is t. Therefore, there are
18.3 Rectangular Stiffened Plate Bending Element 433

Fig. 18.3 A rectangular y

element
2 4 v u
y x

b
z

1 x
a 3 w

(a) (b )

20 degrees of freedom for the element. The displacement vector is given by

{
u 1 v1 u 2 v2 u 3 v3 u 4 v4 w1 θx1 θ y1}
{X }eT = (18.1)
w2 θx2 θ y2 w3 θx3 θ y3 w4 θx4 θ y4

It may be noted that inplane displacements at the four nodes are numbered first
and then the bending displacements with the usual notations. The plate stress part
of the four nodal rectangular plate elements has been discussed in Sect. 9.7 and the
rectangular plate-bending element in Sect. 11.3. As such let us delve straight into
the stiffener formulation.

18.3.1 Stiffness Matrix of the Stiffener Element

The stiffener is considered as eccentric to and integral with the plate. As such it is
assumed to be placed along the nodal line parallel to the x axis. Due to the requirement
of the conformity of displacements between the plate and the stiffener, the following
displacement functions for the stiffener are assumed.
⎧ '⎫
⎪
⎪ α ⎪
⎪ 1' ⎪
⎪ ⎪
⎪
⎪
⎪ α ⎪
⎪ 2' ⎪
⎪ ⎪
⎧ ⎫ ⎡ ⎤⎪ α ⎪ ⎪
⎨u ⎬ 1 x00 0 0 00 ⎪ ⎨ 3' ⎪
⎪ ⎪
⎬
α
w = ⎣0 0 1 x x x 0 0⎦
2 3 4 (18.2)
⎩ ⎭ ⎪ α5' ⎪
φ 000 0 0 1x ⎪ ⎪
⎪
⎪
⎪
α6' ⎪
0
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
' ⎪
⎪
⎪ α ⎪
⎪ 7⎪
⎩ '⎪ ⎭
α8

or,

{ f s } = [C]{α} (18.3)
434 18 Finite Element Method of Analysis of Stiffened Plates

Fig. 18.4 A stiffener 1 Plane reference

element surface
y

L
z

w, yi, Tsi
Ui, Nsi
w, xi, Msi
wi, Vsi

where φ is the angle twist.

The stiffener displacements are shown in Fig. 18.4 along with u and w, φ the angle
of twist has been assumed as an independent displacement.( Figure)18.4 indicates
( ∂w) ∂w
four generalized displacements u, w, θ y = ∂ x and θx = ∂ y = φ . Substituting
the nodal values in Eq. (18.3).

⎧ '⎫ ⎡ ⎤ ⎧ ⎫
⎪
⎪ α ⎪ 1 0 0 0 0 0 0 0 ⎪ u1 ⎪
⎪ 1' ⎪
⎪ ⎪ ⎢ 1
⎪
⎪
⎪
⎪ w1 ⎪
⎪
⎪
⎪ α ⎪ − 0 0 0 1
0 0 0⎥⎥ ⎪
⎪
⎪
⎪ 2' ⎪
⎪ ⎪ ⎢
⎪
L
⎢ 0 1
L
⎥
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ α ⎪
⎪ ⎢ 0 0 0 0 0 0⎥ ⎪ ⎪ w, ⎪
⎪
⎨ '⎪
⎪ 3
⎬ ⎢ ⎥ ⎪⎨
x1 ⎪
⎬
α4 ⎢ 0 0 1 0 0 0 0 0⎥ w, y1
=⎢ ⎥= (18.4)
⎪ α5' ⎪
⎪ ⎪ ⎢ 0 − L32 − L22 0 0 L32 − L1 0⎥ ⎪ u2 ⎪
⎪
⎪ ⎪ ⎢ ⎥ ⎪⎪ ⎪
⎪
⎪
⎪
⎪ α6' ⎪
⎪
⎪
⎪
⎢ 0 L22 L12
⎢ 0 0 − L22 1
2 0 ⎥
⎥ ⎪
⎪
⎪
⎪ w 2
⎪
⎪
⎪
⎪
⎪
⎪α ⎪ ⎪
⎪ ⎣ 0 0
L
⎪ ⎪
⎪
⎪
'
7⎪ 0 1 0 0 0 0⎦ ⎪ ⎪
⎪ w, ⎪
x2 ⎪
⎪
⎩ '⎭ ⎩ ⎭
α8 0 0 0 − L1 0 0 0 L 1
w, y2

or,

{α} = [Q]{X s }e (18.5)

where the subscript s refers to the stiffener.

Combining Eqs. (18.3) and (18.5), we get

{ f s } = [Ns ]{X s }e (18.6)

18.3 Rectangular Stiffened Plate Bending Element 435

where [N s ] = [C] [Q] whose explicit values are given below by Eq. (18.7)
⎡ ⎤
1− x
l
⎢ 1 − 3x2 + 2x3
2 3
⎥
⎢ l L ⎥
⎢ x − 2x 2 + x 3 ⎥
⎢ L2 ⎥
⎢ L
⎥
⎢ 1 − x
⎥
[Ns ]T = ⎢ L ⎥ (18.7)
⎢ x
⎥
⎢ 3x 2 L ⎥
⎢ 2x 3 ⎥
⎢ L 2 2 − L 33 ⎥
⎣ − x + x2 ⎦
L L
x
L

The strain components for the stiffener are

⎧ ∂u ⎫
⎨ ∂x ⎬
{εs } = − ∂∂ xw2
2
(18.8)
⎩ ∂φ ⎭
∂x

Using Eqs. (18.3), (18.8) can be written as

⎡ ⎤
01000 0 0 0
{εs } = ⎣ 0 0 0 0 0 −2 −6x 0 ⎦[Q]{X s }e (18.9)
00000 0 0 1

or,

{εs } = [Bs ]{X s }e (18.10)

The ‘stress–strain’ relation of the stiffener is given by

⎧ ⎫ ⎡ ⎤⎧ ∂ u ⎫
⎨ Ns ⎬ E A x E Sx 0 ⎨ ∂x ⎬
Ms = ⎣ E Sx E Ix 0 ⎦ − ∂∂ xw2
2
(18.11)
⎩ ⎭ ⎩ ∂φ ⎭
Ts 0 0 G Jx ∂x

or,

{σs } = [Ds ]{εs } (18.12)

Here

Ax the cross-sectional area of the x directional stiffener

S x first moment of the area of the x directional stiffener with respect to the middle
surface of the plate
I x the moment of inertia of the x directional stiffener with respect to the reference
surface
436 18 Finite Element Method of Analysis of Stiffened Plates

J x the polar moment of inertia of the x directional stiffener

Combining Eqs. (18.10), and (18.12), we get

{σs } = [Ds ][Bs ]{X s }e (18.13)

It can be shown that the stiffness matrix of the stiffener is given by

(L
[K s ]e = [Bs ]T [Ds ][Bs ] d x (18.14)
0

[K s ]e has been evaluated explicitly and is given by Eq. (20.15)

⎡ ⎤⎧ ⎫
E Ax
0 − ELSx 0 − ELAx 0 E Sx
0 ⎪
⎪ u1 ⎪ ⎪
⎥⎪⎪ w1 ⎪
L L
⎢ ⎪
0 − L3 ⎥⎪ ⎪
2E I x 6E I x 12E I x 6E I x
⎢ 0 0 0 ⎪ ⎪
⎥⎪ ⎪
L3 L2 L2
⎢ − E Sx 6E Ix 4E I x E Sx
− 6E I x 2E I x ⎪
⎪ ⎪
⎪
⎢ 2 0 2 0 ⎥⎪⎪ w, x1 ⎪
⎪
⎢ L L L L L L
⎥⎨ ⎬
⎢ 0 0 0 G Jx
0 0 0 − GLJx ⎥ w, y1
[K s ]e = ⎢0 E Ax L ⎥
⎢ − L 0 E Sx
0 E Ax
0 − ELSx 0 ⎥⎪ u ⎪
⎢ L L ⎥⎪⎪ 2 ⎪ ⎪
⎢ ⎥⎪⎪ w2 ⎪ ⎪
⎢ E0S −6ELI 3 −2ELI 2 − L2
12E I x 6E I x 12E I x 6E I x
0 0 L3
0 ⎥⎪⎪
⎪
⎪
⎪
⎪
⎣ x x x
0 − ELSx − 6EL 2Ix 4ELIx 0 ⎦⎪⎪ w, ⎪
⎪
L L 2 L ⎪
⎩
x2 ⎪
⎭
0 0 0 − GLJx 0 0 0 G Jx
L
w, y2
(18.15)

The stiffness matrix of the y directional stiffener can be similarly obtained.

18.4 Isoparametric Stiffened Plate Bending Element

One of the most popular elements for the analysis of plate structures is the isopara-
metric element. Like the previous section, let us treat the plate part and stiffener part
separately. We put forward the formulation of the stiffener in a more general manner
than done previously. The restriction that the stiffener should be placed on the nodal
line is removed and it may to now be placed anywhere within the plate element and
need not be parallel to any edge of the plate.

18.4.1 Stiffness Matrix of Arbitrarily-Oriented Eccentric

Stiffener

An arbitrarily oriented stiffener placed within the plate element and its isoparametric
mapping is shown in Fig. 18.5.
18.4 Isoparametric Stiffened Plate Bending Element 437

y′ y

x′

φ
x

3 y′ η
y
2
1 x′
4
G
G
8 ξ
5
6
7
x
(a) Global coordinate (b) Isoparametric coordinate

Fig. 18.5 Arbitrary oriented stiffener within the plate element

The displacement field of the stiffener is given by

⎧ ⎫ ⎧ ' ( )⎫
⎨ U ⎬ ⎨ u − z θ x ' ( x ') ⎬
{ f } = V = v' − z θ y' x ' (18.16)
⎩ ⎭ ⎩ ( )⎭
W w + c' θ y ' x '

where c' is the distance of a point on the stiffener cross-section measured from y axis
(Fig. 18.6). x ' –y' are the coordinate system tangential to the Gauss points.
The strain components are given by

Fig. 18.6 Cross-section of C′ d1

the stiffener
y′

d2

z
438 18 Finite Element Method of Analysis of Stiffened Plates
⎧ ⎫
⎪ ∂ u' '
⎪
⎨ ∂ x'
− z ∂∂θxx' ⎬
'
{εx ' } = ∂w ' ∂ θy
+ c ∂ x ' − θx ' (18.17)
⎪
⎩
∂ x' ' ⎪
⎭
−z ∂∂θxy'

The generalized strain components of the stiffener in the x ' –y' coordinate system
are given by
⎧ ⎫
⎪ ∂ u' ⎪
⎪
⎪ ∂ x' ⎪
⎪
⎨ − ∂∂ θxx'' ⎬
{ε x ' } = ∂ wx ' (18.18)
⎪
⎪ − θx ' ⎪
⎪
⎪
⎩
∂ x'
∂θ ' ⎪
⎭
− ∂ xy'

The relationship between {εx ' } and {ε x ' } is given by

{εx ' } = [Hs ]{ε x ' } (18.19)

where
⎡ ⎤
1z0 0
[Hs ] = ⎣ 0 0 1 −c' ⎦ (18.20)
000 z

The xy-coordinate axis system is related to x' y' -axis system is as follows
)
x = x ' cos φ − y ' sin φ
(18.21)
y = x ' sin φ + y ' cos φ

The relationship between local displacement in the x ' –y' axis system and global
displacement in the x–y coordinate is
⎧ ⎫
⎧ '⎫ ⎡ ⎤⎪ u ⎪
⎨ ⎬
u cos φ sin φ 0 0 ⎨ ⎪
⎪ ⎬
⎣ ⎦ v
θx ' = 0 0 cos φ sin φ (18.22)
⎩ ⎭ ⎪ θx ⎪
θy' 0 0 − sin φ cos φ ⎪⎩ ⎪ ⎭
θy

Using the relations given in Eqs. (18.21) and (18.22), Eq. (18.18) can be expressed
in terms of x–y coordinate system
18.4 Isoparametric Stiffened Plate Bending Element 439
⎧ ( ) ⎫
⎪
⎪ ∂u
cos2 φ + 21 ∂∂ uy + ∂∂ xv sin 2φ + ∂∂ vy sin2 φ ⎪ ⎪
⎪
⎪ ∂ x ( ) ⎪
⎪
⎪
⎪ ⎪
⎪
⎨ − ∂ θx cos2 φ − ∂ θx
+
∂ θy
sin 2 φ −
∂ θy
sin 2
φ ⎬
∂x 2 ∂y ∂(x ) ∂ y
{ε x ' } = (∂ w ) (18.23)
⎪
⎪ − θx cos φ + ∂∂ wy − θ y sin φ ⎪
⎪
⎪
⎪ ( ∂x ) ⎪
⎪
⎪
⎪ ⎪
⎩ − 1 − ∂ θx + ∂ θ y sin 2φ + ∂ θx sin2 φ − ∂ θ y cos2 φ ⎪ ⎭
2 ∂x ∂y ∂y ∂x

or,

{ε x ' } = [T ]{ε x } (18.24)

where [T ] is the transformation matrix and is given by

⎡ ⎤
cos2 φ 0 0 0
⎢ sin2 φ 0 0 0 ⎥
⎢ ⎥
⎢ 1 sin 2φ ⎥
⎢2 0 0 0 ⎥
⎢ cos φ
2
0 − 2 sin 2 φ ⎥
1
⎢ 0 ⎥
⎢ ⎥
[T ]T = ⎢ 0 sin2 φ 0 1
sin 2φ ⎥ (18.25)
⎢ 2 ⎥
⎢ 0 1
sin 2φ 0 cos2 φ ⎥
⎢ 2 ⎥
⎢ 0 1
sin 2φ 0 − sin2 φ ⎥
⎢ 2 ⎥
⎣ 0 0 cos φ 0 ⎦
0 0 sin φ 0

and {ε x } is the generalized strain vector in the x–y coordinates and is given by
( ( )
∂u ∂v ∂u ∂v ∂ θx ∂ θy ∂ θy
{ε x } =
T
+ − − −
∂x∂y ∂y ∂x ∂x ∂y ∂x
( )( ))
∂ θx ∂ w ∂w
− − θx − θy (18.26)
∂y ∂x ∂y

Using the shape function relation of the type given by Eqs. (14.98), (18.26) can
be written as

Σ
n
{ε x } = [Bsr ]{X r }e = [Bs ]{X }e (18.27)
r =1

where,
⎡ ∂ Nr ⎤
0 0∂x
0 0
⎢ 0 0 0 − ∂ Nr 0 ⎥
[Bsr ] = ⎢ ∂x
⎣ 0 0 ∂ Nr −Nr 0 ⎦
⎥ (18.28)
∂x
0 0 0 0 ∂∂Nxr
440 18 Finite Element Method of Analysis of Stiffened Plates

The stress–strain relationship for the stiffener is given by

{σx ' } = [Ds ]{εx ' } (18.29)

where,
{ }
{σx ' }T = σx ' s τx ' y ' s τzx ' s (18.30)

⎡ ⎤
E 0 0
[Ds ] = ⎣ 0 G 0 ⎦ (18.31)
0 0 G

Combing Eqs. (18.19) and (18.24), Eq. (18.29) can be written as

{σx ' } = [Ds ][Hs ][T ][Bs ]{X }e (18.32)

Using the principle of virtual work, the element stiffness matrix of the stiffener is
expressed as
(
[K s ]e = [Bs ]T [T ]T [Ds ][T ][Bs ] d x ' (18.33)

where
⎡ ⎤
E
(b/2 (d2
[ ] ⎢ E z E z 2 sym ⎥ '
Ds = ⎢ ⎥ dy dz (18.34)
⎣ 0 0 G ⎦
−b/2 −d1 /2 '
( '2 )
0 0 Gc G c +z 2

On integration, Eq. (18.34) reduces to

⎡ ⎤
E As
[ ] ⎢ E Sx ' s E Ix ' s sym ⎥
Ds = ⎢
⎣ 0
⎥ (18.35)
0 Sis 0 ⎦
0 0 0 G Js

where

As cross-sectional area of the stiffener

Ix' s second moment of the stiffener cross-sectional area about the reference axis
S is shear rigidity of the stiffener
Jx' s polar moment of inertia of the stiffener cross-sectional area
Sx ' s first moment of the stiffener cross-sectional area about the reference axis
References and Suggested Readings 441
˜( )
y '2˜+ z 2 dy ' dz is interpreted as the polar moment of inertia of the cross-
section. G y ' dy ' dz is zero for sections symmetrical about z axis and lying in
the x–y plane. However, for sections, which are not symmetrical about z axis, this
quantity has been taken to be zero. The element stiffness matrix of the stiffener given
by Eq. (18.32) is evaluated numerically.
The main elegance of the formulation of the stiffener presented above reveals that
the stiffener can be placed anywhere within the plate element. As such the mesh
division is no longer dependent on the stiffener location. A careful study of the terms
involved in the determination of [K s ]e for the stiffener will immediately reveal that
the contribution of the stiffener is reflected at all eight nodes of the plate element in
which it is placed and that there is a mutual coupling at all these nodes. No doubt
the model gives a considerable advantage to the user.

References and Suggested Readings

1. T.S. Koko, M.D. Olson, Nonlinear analysis of stiffened plates using superelements. Int. J. Numer.
Eng. 31, 319–413 (1991)
2. C.S. Rossow, A.K. Ibrahimkhail, Constrained method of analysis of stiffened plates. Comput.
Struct. 8, 51–60 (1978)
3. M. Mukhopadhyay, S.K. Satsangi, Isoparametric stiffened plate-bending element for the analysis
of ships’ structures. Trans. Roy. Inst Nav. Archit. 126, 144–151 (1984)
4. D. Venugopal Rao, A.H. Sheikh, M. Mukhopadhyay, Finite element large deflection analysis of
stiffened plates. 47(6), 987–993 (1993)
Chapter 19
Selected Topics

A few selected topics of special interest have been put forward in this chapter.

19.1 Rayleigh–Ritz Method

In many problems of continuum mechanics, a functional can be formed that describes

the problem. In Rayleigh–Ritz method, approximate solutions of the variables of
the functional are assumed. The classical solution of the differential equation is thus
avoided. Substitution of approximate solutions results in a set of algebraic equations.
Incorporation of more terms in assumed solutions will yield more accurate result
(19.1).
In Rayleigh–Ritz method, an approximate solution of variable of the following
form is assumed
Σ
u= βi Fi (19.1)
i

where β i is the generalized coordinate whose value is to be determined.

One of the major requirements is that the function, F i = F i (x) must be admissible,
that is, it must satisfy compatibility conditions and essential boundary conditions.
A functional of the following type is assumed

(x2
( )
πp = I u, u ,x , x d x (19.2)
x1

where u ,x = ∂∂ux with boundary conditions u (x1 ) = u (x2 ) = 0.

The principle of stationary potential energy is one of the many variational prin-
ciples. It states that the functional π p is also stationary with respect to β 1 , β 2 , ……,

© The Author(s) 2022 443

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0_19
444 19 Selected Topics

β 8 . This will result in equations containing β i coefficients.

∂π p
= 0, i = 1, 2, . . . . . . . . . . . . , n (19.3)
∂βi

Equations of the type of Eq. (19.3) are linear in terms of β i where π p is a quadratic
function of u and u,x .

19.2 An Example

Determine the central deflection of the beam of Fig. 19.1

An approximate solution for the deflection, that is, w displacement is assumed

πx 3π x
w = β1 sin + β2 sin (19.4)
L L
Equation 19.4 satisfies the boundary conditions of the problem, that w = 0, and M
= 0 at x = 0 and at x = L. Only two terms have been considered here. Incorporation of
more terms would result in more number of equations. Equation 19.4 is a symmetric.
The functional is given by

(L ( )2
EI d 2w
πp = d x − Pwc (19.5)
2 dx2
o

In Eq. (19.5), the first term corresponds to the internal energy of the beam and the
second to the potential of loads. wc is the central deflection.
Substituting w from Eq. (19.4) into Eq. (19.5), yields

(L ( ) π (2 ( )2 )2
EI πx 3π 3π x
πp = β1 sin + β2 sin dx
2 L L L L
o

Fig. 19.1 Uniform beam P

under a central point load L/2

x, u

z, w
19.3 Rayleigh–Ritz Finite Element Method 445
{ )π ( ( )}
3π
− P β1 sin + β2 sin (19.6)
2 2

After carrying out the necessary operations and applying orthogonality relation-
ship
( ( )4 )
EI ) π (4 L 3π L
πp = β12 + β22 − P(α1 − α3 ) (19.7)
2 L 2 L 2

Applying the principle of stationary potential energy given by Eq. (19.3), yields
{ ( ) } ⎫
2β1 πL L2 − P = 0 ⎬
∂π p 4
∂β1
= EI
2 { ( )4 L } (19.8)
+ P = 0⎭
∂π p
∂β2
= EI
2
2β2 3πL 2

Solution of Eq. (19.8) gives

( ) ( )4
2P L 3 2P L3 1
β1 = , β2 = − (19.9)
EI π 4 EI π4 3

Substituting β 1 , β 2 from Eq. (19.9) into Eq. (19.4), yields

( )
2P L 3 πx 1 3π x
w= sin − 4 sin (19.10)
E I π4 L 3 L

If n terms instead of two terms are considered, then for n → ∞, this solution for
central deflection is given by
( )
2P L 3 1 1
wc = 1+ + 4 + ········· (19.11)
E I π4 34 5

Equation (19.11) is an infinite series. On evaluation, the value of wc coincides

with the exact result

P L3
wc = (19.12)
48 E I

19.3 Rayleigh–Ritz Finite Element Method

The finite element method is nothing but Rayleigh–Ritz method. Instead of using
approximating function for the whole field, functions are assumed in the piecewise
fashion from the nodal degrees of freedom of the field. As shown in the previous
446 19 Selected Topics

Fig. 19.2 Axially loaded bar q = cx

1 2 3 4
(a) Axially loaded bar

x
L

(b) Typical element

section, the functional for the structural problem will consist of the sum of the internal
strain energy and the potential of the vector. The variational principle of the type of
Eq. (19.3) is to be applied. The method is explained with the help of an example.
An axially loaded bar divided into three elements is shown in Fig. 19.2. A typical
element of length L is shown. Axial strain is given by

du
εx = (19.13)
dx
Using interpolation functions, Eq. (19.13) becomes

{εx } = [B]{X }e (19.14)

where
d [ ]
[B] = [N ] = − L1 1
L
(19.15)
dx
The strain energy of the element is

(L (l
1 1
U= E εx2 A d x = εxT · E A · εx d x
2 2
0 0

or,

1
U= {X }eT [k]e {X }e (19.16)
2
where

(L
[k]e = [B]T · AE · [B]d x (19.17)
0
19.4 Weighted Residual Methods 447

The potential of the load is given by

(L (L
Ω=− u σx A d x = − uT · q d x (19.18)
0 0

For the bar having axial force,

q
σx = (19.19)
A
Therefore,

Ω = −{X }eT {P}e (19.20)

where {P}e is the consistent load vector as

(L
{P}e = [N ]T q d x (19.21)
o

Applying Eq. (19.3), that is,

∂π p
= {0} (19.22)
∂ {X }e

It will give

[k]e {X }e = {P}e

19.4 Weighted Residual Methods

For certain physical problems, the functional may not be obtained. This problem
arises when the differential equation of the problem contains certain derivatives of
the odd order. Examples may be cited of fluid mechanics problems where for some
types of flow, only the differential equations and boundary conditions are available.
These types of problems can still be solved by the finite element method that is
applied by means of a weighted residual method. Like the Rayleigh–Ritz method,
the differential equations of physical problem form a part of the integral expressions,
which is used by the weighted residual method. Functional and residual formulations
are both known as weak forms of stating governing differential equations (‘Weak’
form is because the conditions are enforced in an average or integral sense). The
448 19 Selected Topics

differential equations themselves are mentioned as strong form (it is ‘strong’ because
the conditions are enforced at every point).
The governing differential equations and natural or nonessential or Neumann
boundary conditions of a physical problem are written as

Du − f = 0 in domain V (19.23)

Bu − g = 0 on the boundary S of V (19.24)

where, D, B = differential operators

u = dependent variable, for example, displacement
x = independent variable, for example, coordinate
f , g = functions of x or constants or zero.
For beam bending, Eq. (19.23) becomes

E I w,x x x x = q (19.25)

d4
Thus, D = EI , u = w and f = q
dx4
Equation (19.24) when applied to beam bending results in the following two
equations

E I w,x x − M B = 0
(19.26)
E I w,x x x − VB = 0

where M B , V B are bending moment and shear force respectively at the end of the
beam. If u is the exact solution of the differential (19.23), u is assumed to be an
approximate solution. The difference between the exact and the approximate solution
can be expressed in the form of residuals, that is,

R D = Du − f (interior residual) (19.27)

R B = Bu − g (boundary residual) (19.28)

If u happens to be close to u, the residual will be small. A variety of methods known

as weighted residual techniques exist for solving partial differential equations. The
basic objectives of the weighted residual methods are to formulate a scheme by
which small residuals will be obtained. This results in a set of algebraic equations.
Equations of the following type making weighted average of the residuals as zero
are obtained.
19.5 Galerkin Method 449
(
Wi R D dv = 0 (19.29)
v

where W i represents a set of weighted functions.

The accuracy of the solution depends to a great extent on the choice of the approx-
imating function u (also known as trial function) and the number of terms included
in it.

19.5 Galerkin Method

Galerkin method is one of the weighted residual methods. In this, the weighting
function is assumed same as the approximating function. Further, nonessential
boundary conditions are introduced by using boundary integral RB in combination
with integration by parts. The method is explained with an example.
The equilibrium equation of a prismatic beam is given by

d 4w
EI = P (19.30)
dx4
Galerkin equation then becomes

(l { }
d 4w
EI − P δw d x = 0 (19.31)
dx4
o

Equation (19.31) is integrated by parts twice that yields

(L ( )( ) (L [ ]x=L
d2w d 2 δw d 2 w dδw d3w
EI dx − p δw d x − EI . − EI δw = 0 (19.32)
dx2 dx2 dx2 dx dx3
0 o x=o

The boundary conditions for the simply supported beam (Fig. 19.1) are

d 2w
at x = 0, L M = E I =0 (19.33a)
dx2

w = 0 (19.33b)

The above boundary conditions when substituted in Eq. (19.32) indicate

dδw
= 0, δw = 0 (19.34)
dx
Thus Eq. (19.32) reduces to
450 19 Selected Topics

(L ( )( ) (L
d 2w d 2 δw
EI dx = p δw d x (19.35)
dx2 dx2
0 0

Boundary conditions in w, dw
dx
, are called essential and are satisfied exactly.
Let us formulate the beam problem when both ends are considered free, which is
a more general case. The boundary conditions then will be

d 2w d 3w
EI = M, − E I = V (19.36)
dx2 dx3

where M and V are the applied bending moment and shear force respectively.
The boundary residuals are given by
{ }
d 2w
R 'B = EI −M (19.37a)
dx2
{ }
d 3w
R ''B = − E I − V (19.37b)
dx3

Instead of Eq. (19.31), Galerkin’s equation then becomes

(L ( ) [( ) ( ) ]x=L
d4w d3w d2w d δx
EI 4
− p δv d x = EI 3
+ V δw − E I 2
− M
dx
(19.38)
dx dx dx
0 x=0

It may be noted that boundary error in moment is weighted with rotations and
shear with displacements.
Integrating Eq. (19.38) by parts twice results

(L ( )( ) (L [ ]x=L
d 2w d 2 δw d δv
EI dx = p δw d x + V δv + M (19.39)
dx2 dx2 dx
0 0 x=0

It may further be noted that Eq. (19.39) is second order whereas the differential
equation of equilibrium is fourth order.

19.5.1 An Example of Galerkin Method

An example is presented for the determination of critical load of a plate in in-plane

shear by Galerkin method. A square plate having all edges simply supported is
subjected to uniform shearing forces. N xy applied along all four edges as shown in
Fig. 19.3.
19.5 Galerkin Method 451

Fig. 19.3 Simply supported a

plate in pure shear
x

Nxy a

The deflection function of the plate satisfying the boundary conditions is expressed
as
πx πy 2π x 2π y
w = C1 sin sin + C2 sin sin (19.40)
a a a a
The equilibrium equation of the plate is

∂ 4w ∂ 4w ∂ 4w Nx y ∂ 2 w
+ 2 + + 2 = 0 (19.41)
∂x4 ∂ x 2∂ y2 ∂ y2 D ∂x ∂y

The weighting functions are

∂w πx πy
W1 = = sin sin (19.42a)
∂C1 a a
∂w 2π x 2π y
W2 = = sin sin (19.42b)
∂C2 a a

The Galerkin residual equation is

(a (a ( )
∂w 4 ∂w 4 ∂w 4
Wi + 2 + dx dy = 0 (19.43)
∂x4 ∂ x 2∂ y2 ∂ y4
0 0

Substituting appropriate values from Eqs. (19.40), (19.42a) and (19.42b) into
Eq. (19.43) yields

(a (a [
4C1 π 4 2 πx 2 πy 64C2 π 4 2π x 2π y πx πy
4
sin sin + 4
sin sin sin sin
a a a a a a a a
0 0
452 19 Selected Topics
(
2N x y C1 π 2 πx πx πy πy
+ cos sin cos sin
D a2 a a a a
)]
4C2 π 2
2π x πx 2π y πy
+ cos sin cos sin dx dy = 0 (19.44a)
a2 a a a a
(a (a [
4C1 π 4 πx 2π x πy 2π y 64C2 π 4 2π x 2π y
4
sin sin sin sin + sin2 sin2
a a a a a a4 a a
0 0
(
2N x y C1 π 2 πx 2π x πy 2π y
+ 2
cos sin cos sin
D a a a a a
]
4C2 π 2
2π x 2π x 2π y 2π y
+ cos sin cos sin dx dy = 0 (19.44b)
a2 a a a a

The following definite integrals are to be noted

(a
mπ x a
sin2 dx = (19.45a)
a 2
0

(a
mπ x nπ y
sin sin d x = 0 for m /= n (19.45b)
a a
0

(a
mπ x nπ x
cos sin d x = 0 for m /= n (19.45c)
a a
0

(a
2π x πx 2a
cos sin dx = − (19.45d)
a a 3π
0

(a
2π x πx 4a
sin cos dx = (19.45e)
a a 3π
0

Using the above relations, Eqs. (19.44a) and (19.44b) reduce to

( ) ) ( ⎫
4C1 π 4 a 2
+ Dx y 4Ca22π − 2a = 0⎬
2N 2 2

a4 2
( ) ) 2(
3π
(19.46)
64C2 π 4 a 2
+ Dx y Ca1 π2 4a = 0⎭
2N 2

a4 2 3π

or,
)
π4 32N
a2
C1 + 9Dx y C2 = 0
32N x y 4 (19.47)
9D
C1 + 16π
a2
C2 = 0
19.6 Galerkin Finite Element Method 453

Nontrivial solution of homogeneous Eq. (19.47) is given by

| 4 |
| π 32N x y |
| a2 |
| 32Nx y 9D |= 0 (19.48)
| 9D 16π 4
a2
|

The determinant yields the critical load of inplane shear and is given by

( ) π2
Nx y cr
= 11.1 D (19.49)
a2

19.6 Galerkin Finite Element Method

The differential equation of a uniform bar under axial load is

E Aσ,x x + q = 0 (19.50)

For an axially loaded bar

σx = E εx = E u x (19.51)

Combining Eqs. (19.50) and (19.51) yields

E A uxx + q = 0 (19.52)

At the free end of the bar, F = 0. At the other end where an axial force is applied,
the non-essential boundary condition is given by

E Au ,x − F = 0 (19.53)

Essential boundary conditions deal with the prescribed values of u.

Let the bar be divided into ‘n’ elements with each element is of length L For each
element, the displacement function is given by

u = [N ] {X }e (19.54a)

where
[ ]
[N ] = L − x x
{ L L } (19.54b)
{ X }eT = u 1 u 2

Weights W i in Galerkin method are

454 19 Selected Topics

∂u
Wi = = Ni (19.55)
∂u i
}
N1 = L −L
x
(19.56)
N2 = Lx

Galerkin’s residual equation becomes

nl (L
Σ ( )
Ni E A u ,x x + q d x = 0 (19.57)
j=1 0

Equation (19.54) is to be repeated for all shape functions.

As the total number of shape functions is equal to the total number of degrees of
freedom, total number of Galerkin residual equations will be equal to the total d.o.f.
Integration Eq. (19.47) for the ith element yields

(L (L
[ ]L
Ni E A u ,x x d x = Ni E A u ,x o
− Ni,x E A u ,x d x (19.58)
0 0

Using nonessential boundary condition of Eq. (19.43) in Eq. (19.48), Eq. (19.47)
becomes

nl (L
Σ Σ
( ) nl
− Ni,x E A u ,x + Ni q d x + [Ni Fi ] oL = 0 (19.59)
j=1 0 j=1

From Eqs. (19.54a) and (19.54b) gives

{ }
[ ] u1
u ,x = − 1 1
L L
= [B] {X }e (19.60)
u2

Combining Eqs. (19.59) and (19.60) gives

nl (L
Σ nl (L
Σ Σ
nl
[ ]L
[B] [E A] [B] d x {X }e =
T
[N ]T q d x + [N ]T F 0
j=1 0 j=1 0 j=1
(19.61)

Except for the last term, Eq. (19.61) reduces to the standard formula

[K ]{X } = {P} (19.62)

The last term of Eq. (19.51) needs some explanation.

19.7 Torsional Stiffness of Prismatic Beam Element 455

L L
Fj—1 Fj—1 Fj Fj

a b b c
el (j — 1) el (j)

Fig. 19.4 Bar elements with nodal loads

The shape function matrix for a typical element can be written at its ends x = 0
and x = L as
[ ] [ ]
[N ] 0 = 10 and [N ] L = 01

For two adjacent elements (j–1) and j as shown in Fig. 19.4, last summation in
Eq. (19.6) produces the terms
{ }
1
− F j−1
0
node a
{ }
0
node b +( F j−1
1 (19.64)
{ }
1
− Fj
0
{ }
0
node c + Fj
1
( )
On assembly, the resultant axial force at node b is F j−1 − F j . This is identified
with the externally applied load.

19.7 Torsional Stiffness of Prismatic Beam Element

Total torsion acting on a thin-walled beam is split into two parts.

T (x) = Tx (x) + Tω (x) (19.65)

where T x is the St. Venant torsion and T ω is the warping moment.

Tx = G J θ' (x) (19.66)

Tw = −E Iw θ''' (x) (19.67)

456 19 Selected Topics

E and G are Young’s modulus and shear modulus of elasticity respectively. J is

St. Venant’s torsional constant, I w is warping torsional stiffness for determination of
which one may refer to Refs. [2, 3].
Kawai [4] has developed a suitable beam element to take care of torsion. He
represented angle of twist in the following form for a two-noded beam element

θ (x) = α1 + α2 x + α3 x 2 + α4 x 3 (19.68)

or,
⎧ ⎫
⎪
⎪ α1 ⎪
[ ]⎨ ⎪ ⎬
α
θ (x) = 1 x x 2 x 3 2
(19.69)
⎪
⎪ α ⎪
⎩ 3⎪ ⎭
α4

Equation (19.69) can be represented as

θ = [Q]{α} (19.70)

where
[ ]
[Q] = 1 x x2 x3

{ α }T = { α1 α2 α3 α4 } (19.71)
( )
Each node of the beam element has two degrees of freedom θ and θ ' = dθ
dx
. The
nodal displacement vector is
{ }
{X }eT = θ1 θ1' θ2 θ2' (19.72)

The corresponding force vector is written as

{ }
{ P }eT = Tx1 Tω1 Tx2 Tω2 (19.73)

The angle of twist can be expressed in terms of nodal displacements following

the steps mentioned in previous chapters

θ = [Q][A]−1 {X }e (19.74)

where
19.7 Torsional Stiffness of Prismatic Beam Element 457

⎡ ⎤
1 0 0 0
⎢ 0 1 0 0 ⎥
[A]−1 = ⎢
⎣ − 32 − 2
⎥ (19.75)
L L
3
L2
− L1 ⎦
2
L3
1
L2
− L23 1
L2

Based on Eq. (19.74),

θ = [B]1 {X }e (19.76)

where
[ ]
−6 ( ) 6( )
[B]1 = ξ − ξ 2 (1 − 4ξ + 3ξ 2 ) ξ − ξ 2 (− 2ξ + 3ξ 2 )
L L
(19.77)

with ξ = Lx
Similarly,

θ ''' = [B]2 { X }e (19.78)

where
6 [ ]
[B]2 = 2
L
1− 2
L
1 (19.79)
L2

From Eq. (19.66) and substituting θ ' from Eq. (19.76) in it

Tx = G J [B]1 {X }e (19.80)

Similarly, from Eqs. (19.67) and (19.78), we get

Tω = −E Iω [B]2 {X }e (19.81)

Equations (19.80) and (19.81) reveal that St. Venant’s torsion T x varies parabol-
ically within the element and the warping torsion T ω is constant along the
element.
The element stiffness matrix consists of two parts

[k]e = [k] j + [k]ω (19.82)

where [k]j is the St. Venant’s torsional stiffness matrix and [k]ω is the warping
torsional stiffness matrix.
458 19 Selected Topics

(L
[k ] j = [ B ] 1T G J [ B ] 1 d x (19.83)
0

(L
[ k ]ω = [ B ] 1T (− E Iω ) [ B ] 2 d x (19.84)
o

Explicit evaluation of Eqs. (19.83) and (19.84), yields

⎡ 6 ⎤
5
symmetrical
GJ ⎢ L 2L 2
⎥
[k] j = ⎢ 10 15 ⎥ (19.85)
L ⎣− 6 L 6 ⎦
5 10 2 5
−L 2L 2
L
10
− L30 10 15

and
⎡ ⎤
12 symmetrical
E Iω ⎢
⎢ 6L 4L
2 ⎥
⎥
[k] ω = ⎣ ⎦ (19.86)
L3 − 12 − 6L 12
6L 2L 2 − 6L 4L 2

19.8 Torsion of Noncircular Sections

The differential equation in terms of stresses for a bar subjected to torsion is [5, 6]

∂Tx z ∂Tyz
− = −2G θ (19.87)
∂y ∂x

where θ is the angle of twist per unit length.

The stresses are expressed in terms of stress functions as

∂ψ ∂ψ
τx z = G θ and τ yz = − G θ (19.88)
∂y ∂x

whereψ is the stress function.

Combining Eqs. (19.87) and (19.88) results

∇ 2 ψ = −2 (19.89)

ψ = 0at the boundary.

Torque carried by the section is given by
19.8 Torsion of Noncircular Sections 459

L
1

Fig. 19.5 Torsion element

( (
T = 2Gθ ψ dA (19.90)
A

The functional of the boundary value problem is

( ( [( )2 ( )2 ]
G θ2 dψ dψ
πp = + − 4ψ dx dy (19.91)
2 dx dy
A

Let us assume that the section consists of triangular elements. A typical element
is shown in Fig. 19.5. ψ for the element is expressed as

ψ = N1 ψ1 + N2 ψ2 + N3 ψ3 (19.92)

Substituting from Eq. (19.92) into Eq. (19.91), yields

⎡⎧ ⎫2 ⎧ ⎫2 ⎧ ⎫⎤
( ( ⎨ 3 ⎬ ⎨Σ ⎬ ⎨Σ ⎬
⎢ Σ ∂ Ni
3 3
G θ2 ∂ Ni ⎥
πp =
2
⎣
⎩ ∂x
ψi
⎭
+
⎩ ∂y
ψi
⎭
− 4
⎩
Ni ψi ⎦ dx dy
⎭
(19.93)
A L=1 L=1 L=1

Applying the principle of stationary potential

( ( [ (Σ 3
)
∂π p G θ2 ∂ Ni ∂Nj
= 2 ψi
∂Ψ j 2 A L=1
∂ x ∂x
( 3 ) ]
Σ ∂Nj ∂Nj
+2 ψi − 4N j d x d y = 0
i=1
∂ x ∂y
j = 1, 2, 3 (19.94)

We can write Eq. (19.94) in matrix notations as

460 19 Selected Topics
( ( ([ ] )
∂ Ni ∂ N j ∂ Ni ∂ N j { }
. + . { ψi } − 2 N j dx dy = 0 (19.95)
A ∂x ∂x ∂y ∂y

Equation (19.95) can be written as

[k]e {ψ} = {P}e (19.96)

where
( ( ( )
∂ Ni ∂ N j ∂ Ni ∂ N j
ki j = . + . dx dy (19.97)
A ∂ x ∂ x ∂y ∂y
( (
{P}e = 2 [ N ] dx dy (19.98)
A

The explicit evaluation of element stiffness matrix is identical to that presented

in Sect. 9.2.
The values of shape function mentioned in Eq. (19.92) are given in terms of nodal
coordinates as
⎫
N1 = a1 + b2Δ1 x + c1 y
⎬
N2 = a2 + b2Δ2 x + c2 y
(19.99)
a3 + b3 x + c3 y ⎭
N3 = 2Δ

where

a1 = x2 y3 − y2 x3 b1 = y2 − y3 c1 = x3 − x2
a2 = x3 y1 − y3 x1 b2 = y3 − y1 c1 = x1 − x3
a3 = x1 y2 − y1 x2 b3 = y1 − y2 c3 = x2 − x1
⎡ ⎤
1 x1 y1
2Δ = det ⎣ 1 x2 y2 ⎦
1 x3 y3
= 2 × area of the triangle 123

Therefore, k ij of Eq. (19.97) is given by

1 ( )
ki j = bi b j + ci c j (19.100)
4Δ
The stiffness matrix becomes
⎡ ⎤
b b + c1 c1 b1 b2 + c1 c2 b1 b3 + c1 c3
1 ⎣ 1 1
[k]e = b2 b1 + c2 c1 b2 b2 + c2 c2 b2 b3 + c2 c3 ⎦ (19.101)
4Δ
b3 b1 + c3 c1 b3 b2 + c3 c2 b3 b3 + c3 c3
19.9 Axi-symmetrical Element 461

Similarly, it can be shown that

2Δ { }
{ P }eT = 111 (19.102)
3

19.9 Axi-symmetrical Element

Many practical problems can be encountered involving axi-symmetric solids (such

as solids of revolution) subjected to axially symmetric loading. Due to the symmetry
of the problem, the stress components are independent of angular coordinate θ.
The mathematical problem is similar to that of plane stress and plane strain. By
virtue of symmetry, the state of strain and therefore the complete state of stress is fully
defined by two components of displacements in any plane along its axis of symmetry.
For all stress and strain components, all derivatives with respect to θ will vanish. As
such v, γ rθ , γ θ z , τ rθ and τ θ z are zero (Fig. 19.6). The radial and axial coordinates are
denoted as r and z with u and w being the corresponding displacement. The nonzero
stress components then are σ r , σ θ , σ z and τ rz .
In plane stress or plain strain components, as either the stress or strain in the
direction normal to the plane is zero, the strain energy of the plate is associated with
three strain components. A major difference in the treatment creeps in the case of
axi-symmetric solids. Any displacement in the radial direction will induce in strain
in the circumferential direction. The circumferential stress in not zero, the work done
by it in the strain energy expression should be considered.
The strain–displacement relations for the non-zero strain then are

∂u u ∂w ∂u ∂w
εr = , εθ = , εz = , γr z = + (19.103)
∂r r ∂z ∂z ∂r

The constitutive relation for the isotropic case is given by

y
τyz
τxz

y x
x

z
(a) (b)

Fig. 19.6 Bar subjected to torsion

462 19 Selected Topics
⎧ ⎫ ⎡ ⎤ ⎧ ⎫
⎪
⎪ σr ⎪ 1−ν ν ν 0 ⎪ εr ⎪
⎨ ⎪ ⎬ ⎢
⎪
⎨ ⎪ ⎬
σz E ⎢ 1−ν ν 0 ⎥ ⎥ εz
= ⎣
⎪
⎪ σ ⎪ (1 + ν) (1 − 2ν) 1−ν 0 ⎦ ⎪ ε ⎪
⎩ θ⎪ ⎭ 1 − 2ν
⎪
⎩ θ ⎪ ⎭
τr z symmetrical 2
γr z
(19.104)

The axi-symmetric ring element has triangular cross-section with nodes i, j and
m (Fig. 19.7). Each node has two displacements u and w. The algebra involved in
the determination of shape function is identical to the CST element.
The displacement field is therefore given by
{ }
u [ ]
{f} = = I N1 I N2 I N3 {X }e (19.105)
w

where
ai + bi r + ci z
Ni = (19.106)
2Δ
{ }
{ X }eT = u 1 w1 u 2 w2 u 3 w3

ai , bi etc. are explained in Chap. 9.

The strain–displacement relationship is given by
⎧ ⎫ ⎧ ⎫ ⎡ ⎤
⎪ ∂u ⎪
⎪
⎪ εr ⎪ ⎪ ⎪
⎪ ∂r ⎪
⎪ b1 0 b2 0 b3 0
⎨ ⎬ ⎨ ⎬
1 ⎢ ⎥
εz u N1 N2
⎢ r 0 r 0 r 0⎥
N3
{ε} = = r = ⎣ (19.107)
⎪ εθ ⎪
⎪ ⎪ ⎪
⎪
∂w
⎪
⎪ 2Δ 0 c1 0 c2 0 c3 ⎦
⎩ ⎭ ⎪ ∂z
⎩ ∂u + ∂w ⎪
⎭
γr z ∂z ∂r
c1 b1 c2 b2 c3 b3

or,

{ε} = [B]{X }e (19.108)

Fig. 19.7 Axi-symmetric z

solid

j
I
19.10 Three-Dimensional Elements 463

It may be noted in Eq. (19.107) that strains are no longer constant within the
element as in CST, but are functions of r and z.
Strain-strain relationship is given in Eq. (19.104)

{σ } = [D]{ε} (19.109)

In the derivation of the stiffness matrix, the volume integral has to be taken over
the whole ring of material.

d V = 2πr d A = 2πr dr dz (19.110)

The stiffness matrix of axi-symmetric element then becomes

( (
[k]e = 2π [B]T [D] [B] r dr dz (19.110)

The evaluation of the element stiffness matrix requires the integration to be carried
out. Numerical integration may be performed.
Zienkiewicz [7, 8] suggested a simple procedure to evaluate r and z at the
centroidal point, that is,
ri + r j + r m }
r = 3
zi + z j + zm (19.111)
z = 3

The above values of r and z are substituted into [B] matrix.

19.10 Three-Dimensional Elements

Shape functions are given below for the solid elements of serendipity family.

19.10.1 Linear Element (8 Nodes) (Fig. 19.8a)

1
Nr = (1 + ξo ) (1 + ηo ) (1 + ζo ) (19.112)
8

ξ0 = ξ ξr , η0 = ηηr , ζ0 = ζ ζr
464 19 Selected Topics

19.10.2 Quadratic Element (20 Nodes) (Fig. 19.8b)

Corner nodes
1
Nr = (1 + ξo ) (1 + ηo ) (1 + ζo ) (ξo + ηo + ζo − 2) (19.113a)
8
Typical mid-side node

ξr = ±1, ηr = 0, ζr = ±1

1( )
Nr = 1 − η2 (1 + ξo ) (1 + ζo ) (19.113b)
4

19.10.3 Cubic Element (32 Nodes) (Fig. 19.8c)

Corner node

ζ=1
ζ
η
ξ
ξ=1

η = –1
(a) 8-noded (b) 20-noded

(c) 32-noded

Fig. 19.8 Isoparametric brick element

19.10 Three-Dimensional Elements 465

1 [ ( ) ]
Nr = (1 + ξo ) (1 + ηo ) (1 + ζo ) 9 ξ 2 + η2 + ζ 2 − 19 (19.114a)
64
1
ξr = ±1, ηr = ± , ζr = ±1
3
9 ( )
Nr = 1 − η2 (1 + 9ηo ) (1 + ξo ) (1 + ζo ) (19.114b)
64

19.10.4 A16 Noded Solid (Fig. 19.9a)

Corner nodes
1
Nr = (1 + ξo ) (1 + ηo ) (1 + ζo ) (ξo + ηo − 1) (19.115a)
8
Typical mid-side nodes (2, 6, 10, 14)

1 ( )
Nr = 1 − ξ 2 (1 + ηo ) (1 + ζo ) (19.115b)
4
Typical mid-side node (4, 8, 12, 16)

1 ( )
Nr = 1 − η2 (1 + ξo ) (1 + ζo ) (19.115c)
4

ζ ζ
15 14 13 22 21 20 19
η η
23
16 18
12 24
17
9 10 13
11 14 15 16
ξ ξ
6 8
7 5 7
10 9
8 4 12 11 6
5
1 2 3 1 2 3 4
(a) 16 - noded solid (b) 24 - noded solid

Fig. 19.9 Solid element

466 19 Selected Topics

19.10.5 A24 Noded Solid (Fig. 19.9b)

Corner nodes
1 [ ( )]
Nr = (1 + ξo ) (1 + ηo ) − 10 + 9 ξ 2 + η2 (1 + ζo ) (19.116a)
64
Mid-side nodes (5, 6, 11, 12, 17,18, 23, 24)

9 ( )
Nr = (1 + ξo ) 1 − η2 (1 + 9ηo ) (1 + ζo ) (19.116b)
64
Mid-side nodes (2, 3, 8, 9, 14, 15, 20, 21)

9 ( )
Nr = (1 + ηo ) 1 − ξ 2 (1 + 9ζo ) (1 + ζo ) (19.116c)
64

References and Suggested Readings

1. R.D. Cook, D.S. Malkus, M.E. Plesha, Concepts and Applications of Finite Element Analysis,
3rd edn. (Wiley, 1989)
2. V. Fidyosev, Strength of Materials (Mir Publishers, Moscow, Russia, 1968)
3. L.R. Herrman, Elastic torsional analysis of irregular shapes. J. Eng. Mech. Div. ASCE, EM6
11–19 (1965)
4. T. Kawai, The application to the finite element method to ship structures. Comput. Struct. 3,
1175–1194 (1973)
5. S. Timoshenko, J.N. Goodier, Theory of Elasticity, 2nd edn. (McGraw-Hill, 1951)
6. I.H. Shames, C.L. Dym, Energy and Finite Element Methods in Structural Mechanics (McGraw-
Hill Book Co., 1985)
7. O.C. Zienkiewicz, R. Lewis, E. Stagg (eds.), Numerical Methods in Offshore Engineering (Wiley,
1977)
8. O.C. Zienkiewicz, R.L. Taylor, The Finite Element Method, vol. 1, 4th edn. (McGraw-Hill, 1989)
Appendix A
Fixed-End Forces

Table A.1 Values of fixed-end forces of prismatic beams for various types of load conditions
Loadings Fixed-end forces
Pab2 Pa 2 b
P1 = L2
, P2 = L2
a P b  
ab2 a2 b
P1 P2 P3 = P b
L + L3
− L3
P4 =
 
a2 b 2
P La + L3
− ab
L3

L
P3 P4
wL 2
P1 = P2 =
w/unit length 12
wL
P1 P2 P3 = P4 = 2

L
P3 P4
wC
 
P1 = 12L 2
C 2 (L − 3a) + 12a 2 b
c/2 c/2  
wC
w/unit length P2 = 12L 2
C 2 (L − 3b) + 12ab2
P1 P2 P3 = wCb
+ P1 −P 2 wCa
P4 = 12L 2 +
P2 −P1
12L 2 L L

a b

L
P3 P4
(continued)

© The Author(s) 2022 467

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0
468 Appendix A: Fixed-End Forces

Table A.1 (continued)

Loadings Fixed-end forces
 
P1 = − ML0 b 2 − 3b
L
a b  
P1 P2 P2 = ML0 a 2 − 3a
L

P3 = −P4 = 6 ML03ab
Mo
L
P3 P4
P1 = − TL0 b
a b
P1 T0 P2 P2 = − TL0 a

  a 2
wL 2 wa 2 a
P1 = − 10 − 10 + 3
12 60 L L
2
wb2 b b
w − +3
60 L L

P1 a b P2   a 2
wb2 wa 2 a
P2 = − 5 −3
L 12 60 L L
2
wa 2 b b
− 10 − 10 + 3
60 L L

wL 2
P1 = P2 =
Parabolic 15

P1 L P2
Appendix B

All the source files are given in the attached CD. The computer programs are arranged
under two Folders. The folder FORTRAN contains all the computer programs written
in FORTRAN and those under the folder C are in C. The followings are the list of
FORTRAN files included:
1. DIR_TR.F: Direct stiffness method for the analysis of plane truss.

2. DIR_FR.F: Direct stiffness method for the analysis of plane frame.

3. DIR_GR.F: Direct stiffness method for the analysis of grid.

4. SUB_ST.F: Substructure analysis of plane truss.

5.CST_PS.F: Plane stress analysis using constant strain triangle.

6. ISO_PS.F: Plane stress analysis using isoparametric element.

7. ISO_PL_BEND.F: Plate bending analysis using isoparametric bending element.

8. SHELL.F: Analysis of shells using shallow shell triangular element.

9. SPLINE.F: Spline finite strip analysis of plate bending.

10. NONLN_FEA.F: Geometric non-linear finite element analysis of plates.

The following computer programs are given in C under the folder C:

1. dir_tr.c: Direct stiffness method for the analysis of plane truss.
2. dir_fr.c: Direct stiffness method for the analysis of plane frame.

3. dir_gr.c: Direct stiffness method for the analysis of grid.

4. sub_st.c: Substructure analysis of plane truss.

© The Author(s) 2022 469

M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0
470 Appendix B

5. cst_ps.c: Plane stress analysis using constant strain triangle.

6. iso_ps.c: Plane stress analysis using isoparametric element.

7. iso_pl_bend.c: Plate bending analysis using isoparametric bending element.

8. shell.c: Analysis of shells using shallow shell triangular element.

9. spline.c: Spline finite strip analysis of plate bending.

10. nonln_fea.c: Geometric non-linear finite element analysis of plates.

Index

A Delaunay triangulation, 394

Adaptive finite element, 381–389, 391, 392, Displacements, determination, 18
394–397, 399, 400, 402, 403 DKT element, 264
Adaptive mesh generation, 389, 392 Domain decomposition, 394
Advancing front technique, 395, 397 Dynamic problems, 343
Area coordinate, 264, 265
Assembly of elements, 76, 96
Axisymmetric element, 461 E
Eccentric stiffener, 432
Eigenvalues, 319, 346, 354
B Equilibrium conditions, 4, 154, 159
Background mesh, 395 Equilibrium matrix, 136
Banded matrix, 113 Error estimation, 386
Basic functions Error in finite difference method, 364
Beam function, 320–329, 331
Body forces, 149, 154
Boundary conditions, 161–164, 320, F
331–333, 361, 366, 369, 370 Finite difference method, 362–373,
Buckling, 351–354, 359 375–377
Finite element, definitions, 167–169
Finite strip, 319–331
C Flat shell, 287, 295
Castigliano’s theorems, 16–19, 182 Free vibration, 343, 345
Characteristics functions, 320 Frontal solution, 113
Compatibility conditions, 5
Complementary energy, 16, 17
Concentric stiffener, 432 G
Consistent load, 302 Galerkin finite element, 453
Consistent mass matrix, 345, 347, 348 Galerkin method, 449
Constant strain triangle, 177 Gauss quadrature, 207, 208, 222
Continuum, 167, 168, 170–172, 174 Geometric stiffness matrix, 413, 417
Convergence criteria, 184, 186, 189, 198,
204, 206, 222
H
h-adaptivity, 389, 413, 417
D Half band width, 95, 106
Damping matrix, 345 Hermitian polynomial, 255, 256
© The Author(s) 2022 471
M. Mukhopadhyay and A. Sheikh, Matrix and Finite Element Analyses of Structures,
https://doi.org/10.1007/978-3-031-08724-0
472 Index

Higher order element, 200, 211–213 Plate bending, 156–161

Horizontally curved beam, 276, 277, 279, Potential energy, 15, 16
280 Prestrain, 52
Hybrid element, 169 Principle of contragredience, 134, 136
Principle of virtual work, 13–15, 239, 327,
440
I
Interpolation function, 197, 198
Isoparametric element, 197–199, 201–208, R
210–223, 232 Rayleigh-Ritz finite element, 445
Isoparametric general shell element, Rayleigh-Ritz method, 443
304–312 Rectangular element, plane stress, 187–191
Rectangular element, plate bending,
241–243, 245–247, 249
J Reduced integration, 260–262
Jacobian, 202, 204, 206, 217, 307, 309 Rotational restraints, elastic, 323, 324
Rotation matrix, 78, 88, 303

K
Kinematic indeterminacy, 10 S
Semianalytical methods, 319, 331
Settlement of supports, 51–54
L
Shape functions, 198, 200, 201, 205
Lack of fit, 52
Shear deformation, 59, 109, 257, 260
Lagrangian interpolation, 210
Shells, 287–314
Linear triangle, plane stress
Shells of revolution, 292–294
Loads, 3, 4
Skew element, 249
Lumped mass matrix, 348, 351
Skew support, 95
Smoothed stress, 261
Space frame, 9, 10
M
Spline finite strip, 331–336, 338
Mesh, 167, 168, 174
Statical indeterminacy, 6, 7, 9–11
Stiffened plate, 431–441
N Stiffener element, 433–436
Natural coordinates, 211 Structural analysis, objective, 1
Newton-Cotes formula, 207 Substructure techniques, 113–125, 127–130
Non-linear problems, 405–418 Superconvergent patch recovery, 382
Numerical integration, 204–206, 216, 217, Surface force, 149
222

T
O Three-dimensional element, 463
Orthogonality relationship, 324, 329, 445 Transfinite interpolation, 394
Orthotropic plates, 159, 222, 246, 260 Transformation matrix, 79, 81, 84–86, 88,
93, 96, 103, 106, 291, 293, 298, 299
Triangular element, bending, 263
P Triangular element, plane stress, 218–221
p-adaptivity, 389
Pascal triangle, 185
Patch test, 271 V
Plane strain, 153, 154, 177, 183 Vertically curved beam, 312, 314
Plane stress, 152, 154, 177, 178, 180, 181, Vibration, 170, 173
183, 185, 186, 188, 190, 192 Virtual work, 13–15
Index 473

W Z
zz-error estimator, 389
Weighted residual method, 447 zz-refinement, 391